ILSB Report 206 hjb/ILSB/240929

ILSB Report / ILSB-Arbeitsbericht 206

(supersedes CDL–FMD Report 3–1998)

A SHORT INTRODUCTION
TO BASIC ASPECTS
OF CONTINUUM MICROMECHANICS

Helmut J. Böhm

Institute of Lightweight Design and Structural Biomechanics (ILSB)

TU Wien, Vienna, Austria

updated: September 29, 2024

©Helmut J. Böhm, 1998, 2024
Contents

Notes on this Document iv

1 Introduction 1
1.1 Inhomogeneous Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Homogenization and Localization . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Volume Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Overall Behavior, Material Symmetries . . . . . . . . . . . . . . . . . . . . 7
1.5 Major Modeling Strategies in Continuum Micromechanics of Materials . . 9
1.6 Model Verification and Validation . . . . . . . . . . . . . . . . . . . . . . . 14

2 Mean-Field Methods 16
2.1 General Relations between Mean Fields in Thermoelastic Two-Phase Materials 16
2.2 Eshelby Tensor and Dilute Matrix–Inclusion Composites . . . . . . . . . . 20
2.3 Mean-Field Methods for Two-Phase Thermoelastic Composites with Aligned
Reinforcements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.1 Effective-Field Approaches . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.2 Effective-Medium Approaches . . . . . . . . . . . . . . . . . . . . . 30
2.4 General Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.5 Other Analytical Estimates for Elastic Composites with Aligned Reinforce-
ments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.6 Mean-Field Methods for Composites with Nonaligned Reinforcements . . . 36
2.7 Mean-Field Methods for Non-Ellipsoidal Reinforcements . . . . . . . . . . 41
2.8 Mean-Field Methods for Multi-Phase Thermoelastic Composites . . . . . . 43
2.8.1 Mean-Field Methods for Multi-Phase Thermoelastic Composites with
Aligned Reinforcements . . . . . . . . . . . . . . . . . . . . . . . . 43
2.8.2 Analytical Models for Composites Reinforced by Coated Inhomo-
geneities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.9 Mean-Field Models for Nonlinear and Inelastic Composites . . . . . . . . . 49
2.9.1 Viscoelastic Composites . . . . . . . . . . . . . . . . . . . . . . . . 50
2.9.2 (Thermo-)Elastoplastic Composites . . . . . . . . . . . . . . . . . . 51
2.10 Mean-Field Methods for Conduction and Diffusion Problems . . . . . . . . 55

3 Bounding Methods 61
3.1 Classical Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2 Improved Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.3 Bounds on Nonlinear Mechanical Behavior . . . . . . . . . . . . . . . . . . 66
3.4 Bounds on Conduction and Diffusion-Like Properties . . . . . . . . . . . . 67

ii
4 Some Comparisons of Mean-Field Estimates and Bounds 68
4.1 Comparisons of Mean-Field and Bounding Predictions for Effective Thermo-
elastic Moduli of Two-Phase Composites . . . . . . . . . . . . . . . . . . . 68
4.2 Comparisons of Mean-Field and Bounding Predictions for Effective Con-
ductivities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 Comparisons of Mean-Field and Bounding Predictions for Effective Elastic
Moduli of Multi-Phase Composites . . . . . . . . . . . . . . . . . . . . . . 77

5 General Remarks on Modeling Approaches Based on Discrete Micro-

structures 83
5.1 Microgeometries and Volume Elements . . . . . . . . . . . . . . . . . . . . 83
5.2 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.3 Numerical Engineering Methods . . . . . . . . . . . . . . . . . . . . . . . . 93
5.4 Evaluation of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

6 Periodic Microfield Approaches 104

6.1 Basic Concepts of Periodic Homogenization . . . . . . . . . . . . . . . . . 104
6.2 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.3 Application of Loads and Evaluation of Fields . . . . . . . . . . . . . . . . 114
6.4 Periodic Models for Composites Reinforced by Continuous Fibers . . . . . 118
6.5 Periodic Models for Composites Reinforced by Short Fibers . . . . . . . . . 125
6.6 Periodic Models for Particle-Reinforced Composites . . . . . . . . . . . . . 130
6.7 Periodic Models for Porous and Cellular Materials . . . . . . . . . . . . . . 137
6.8 Periodic Models for Some Other Inhomogeneous Materials . . . . . . . . . 142
6.9 Periodic Models Models for Diffusion-Type Problems . . . . . . . . . . . . 144

7 Windowing Approaches 147

8 Embedding Approaches and Related Models 152

9 Hierarchical and Multi-Scale Models 158

10 Closing Remarks 162

Bibliography 164

iii
Notes on this Document

The present document is an expanded and updated version of the CDL–FMD report 3–
1998, “A Short Introduction to Basic Aspects of Continuum Micromechanics”, which, in
turn, is based on lecture notes prepared for the European Advanced Summer Schools
Frontiers of Computational Micromechanics in Industrial and Engineering Materials held
in Galway, Ireland, in July 1998 and in August 2000. Related lecture notes were used for
graduate courses in a Summer School held at Ameland, the Netherlands, in October 2000
and during the COMMAS Summer School held in September 2002 in Stuttgart, Germany.
All of the above documents are related to the micromechanics section of the lecture notes
for the course “Composite Engineering” (317.003) offered regularly at TU Wien.

The course notes “A Short Introduction to Continuum Micromechanics” (Böhm, 2004)

for the CISM Course on Mechanics of Microstructured Materials held in July 2003 in
Udine, Italy, may be regarded as a compact version of a precursor to the present report
that employs a somewhat different notation. The course notes “Analytical and Numerical
Methods for Modeling the Thermomechanical and Thermophysical Behavior of Microstruc-
tured Materials” (Böhm et al., 2009) are also related to the present report, but emphasize
different aspects of continuum micromechanics, among them the thermal conduction be-
havior of inhomogeneous materials and the modeling of cellular materials.

The ILSB Report 206 is being updated periodically to reflect current developments in
continuum micromechanics as seen by the author; its most recent version can be accessed
via https://www.ilsb.tuwien.ac.at/links/downloads/ilsbrep206.pdf. Due to the
dynamic nature of the document it is suggested that in citing it the time stamp should be
given. A somewhat dated version of this report (August 12, 2015) can be accessed via the
DOI: 10.13140/RG.2.1.3025.7127.

A type of Voigt/Nye notation is used for the mechanical variables in chapters 1 to 3.

Tensors of order 4, such as elasticity, compliance, concentration and Eshelby tensors, are
written as 6 × 6 (quasi-)matrices, and stress- as well as strain-like tensors of order 2 as
6-(quasi-)vectors. These 6-vectors are connected to index notation by the relations
       
σ1 σ11 ε1 ε11
σ2  σ22  ε2  ε22 
       
σ3  σ33  ε3  ε33 
σ=   
σ4  ⇔ σ12  ε=   
ε4  ⇔ γ12  ,
       
σ5  σ13  ε5  γ13 
σ6 σ23 ε6 γ23

iv
where γij = 2εij are the shear angles1 (“engineering shear strains”). Tensors of order 4 are
denoted by bold upper case letters, stress- and strain-like tensors of order 2 by bold lower
case Greek letters, and 3-vectors by bold lower case letters. Conductivity-like tensors of
order 2 are treated as 3×3 matrices and denoted by calligraphic upper case letters. All
other variables are taken to be scalars.

In using the present notation it is assumed that the 4th-order tensors show orthotropic
or higher symmetry and that the material axes are aligned with the coordinate system
where appropriate. Details of the ordering of the components of ε and σ will not impact
formulae. It is worth noting, however, that with a notation like the present one some coeffi-
cients of Eshelby tensors differ from their equivalents in index notation, compare Pedersen
(1983).

The tensor product between two tensors of order 2 as well as the dyadic product between
two vectors are denoted by the symbol “⊗”, where [η ⊗ ζ]ijkl = ηij ζkl , and [a ⊗ b]ij = ai bj ,
respectively. The contraction between a tensor of order 2 and a 3-vector is denoted by
the symbol “∗”, where [ζ ∗ n]i = ζij nj . Other products are defined implicitly by the types
of quasi-matrices and quasi-vectors involved. A superscript T denotes the transpose of a
tensor, matrix or vector.

Constituents (phases) are denoted by superscripts, with (p) standing for a general phase,
(m)
for a matrix, (i) for inhomogeneities of general shape, and (f) for fibers. Axial and trans-
verse properties of transversally isotropic materials are marked by subscripts A and T ,
respectively, and effective (or apparent) properties are denoted by a superscript asterisk ∗ .

The present use of phase volume fractions as microstructural parameters is valid for the
microgeometries typically found in composite, porous and polycrystalline materials. Note
that for penny shaped cracks a crack density parameter (O’Connell and Budiansky, 1974)
is the proper choice.

1
This notation allows strain energy densities to be obtained as the scalar product U = 21 σ T ε.

v
Chapter 1

Introduction

In the present report some basic issues of and some of the modeling strategies used for
studying static and quasistatic problems in continuum micromechanics of materials are
discussed. The main emphasis is put on application related (or “engineering”) aspects,
and neither a comprehensive theoretical treatment nor a review of the pertinent literature
are attempted. For more formal treatments of many of the concepts and methods discussed
in the present work see, e.g., Mura (1987), Aboudi (1991), Nemat-Nasser and Hori (1993),
Suquet (1997), Markov (2000), Bornert et al. (2001), Torquato (2002), Milton (2002), Qu
and Cherkaoui (2006), Buryachenko (2007), Dvorak (2013), Kachanov and Sevostianov
(2013), Kachanov and Sevostianov (2018) as well as Buryachenko (2022a); Yvonnet (2019)
is specialized to numerical continuum micromechanics. Shorter overviews of continuum mi-
cromechanics were given, e.g., by Hashin (1983), Zaoui (2002) and Kanouté et al. (2009).
Discussions of the history of the development of the field can be found in Markov (2000)
and Zaoui (2002).

Due to the author’s research interests, more room is given to the thermomechanical
behavior of two-phase materials showing a matrix–inclusion topology (“solid dispersions”)
rather than to materials with other phase topologies or phase geometries.

1.1 Inhomogeneous Materials

Many industrial and engineering materials as well as the majority of biological materials
are inhomogeneous, i.e., they consist of dissimilar constituents (“phases”) that are dis-
tinguishable at some (small) length scale(s). Each constituent shows different material
properties and/or material orientations and may itself be inhomogeneous at some smaller
length scale(s). Inhomogeneous materials (also referred to as microstructured, hetero-
geneous or complex materials) play important roles in materials science and technology.
Well-known examples of such materials are composites, concrete, polycrystalline materials,
porous and cellular materials, functionally graded materials, wood, and bone.

The behavior of inhomogeneous materials is determined, on the one hand, by the rel-
evant material properties of the constituents and, on the other hand, by their volume
fractions, geometries and topologies (the “phase arrangement”). Obviously, the availabil-
ity of information on these counts determines the accuracy of any model or theoretical
description. The behavior of inhomogeneous materials can be studied at a number of

1
length scales ranging from sub-atomic scales, which are dominated by quantum effects, to
scales for which continuum descriptions are best suited. The present report concentrates
on continuum models for heterogeneous materials, the pertinent research field being cus-
tomarily referred to as continuum micromechanics of materials. The use of continuum
models puts a lower limit on the length scales that can be covered in a relatively carefree
way with the methods discussed here, which typically may be taken as being of the order
of 1 µm (Forest et al., 2002).

As will be discussed in Section 1.2, an important aim of bridging length scales for in-
homogeneous materials lies in deducing their overall (“effective” or “apparent”) behavior2
(e.g., macroscopic stiffness, thermal expansion and strength properties, heat conduction
and related transport properties, electrical and magnetic properties, electromechanical
properties, etc.) from the corresponding material behavior of the constituents (as well as
that of the interfaces between them) and from the geometrical arrangement of the phases.
Such scale transitions from lower (finer) to higher (coarser) length scales aim at achiev-
ing a marked reduction in the number of degrees of freedom describing the system. The
continuum methods discussed in this report are most suitable for (but not restricted to)
handling scale transitions from length scales in the low micrometer range to macroscopic
samples, components or structures with sizes of millimeters to meters.

In what follows, the main focus will be put on describing the thermomechanical be-
havior of inhomogeneous two-phase materials. Most of the discussed modeling approaches
can, however, be extended to multi-phase materials (albeit with restrictions), and there
is a large body of literature applying analogous or related continuum methods to other
physical properties of inhomogeneous materials, compare, e.g., Hashin (1983), Torquato
(2002) as well as Milton (2002), and see Sections 2.10 and 6.9 of the present report.

The most basic classification criterion for inhomogeneous materials is aimed at the mi-
croscopic phase topology. In matrix–inclusion arrangements (“solid suspensions”, as found
in particulate and fibrous materials, such as “typical” composite materials, in porous ma-
terials, and in closed-cell foams3 ) only the matrix shows a connected topology and the
constituents play clearly distinct roles. In interpenetrating (interwoven, co-continuous,
skeletal) phase arrangements (as found, e.g., in open-cell foams, in certain composites, or
in many functionally graded materials) and in typical polycrystals (“granular materials”),
in contrast, the phases cannot be readily distinguished topologically.

Obviously, an important parameter in continuum micromechanics is the level of inho-

mogeneity of the constituents’ behavior, which is often described by a phase contrast. For
example, a scalar elastic contrast of a two-phase composite may be defined in terms of the
the Young’s moduli of matrix and inhomogeneities (or reinforcements), E (i) and E (m) , as
2
The designator “effective material properties” is typically reserved for describing the macroscopic
responses of bulk materials, whereas the term “apparent material properties” is used for the properties of
finite-sized samples (Huet, 1990), for which boundary effects play a role.
3
With respect to their thermomechanical behavior, porous and cellular materials can usually be treated
as inhomogeneous materials in which one constituent shows vanishing stiffness, thermal expansion, con-
ductivity, etc., compare Section 6.7.

2
 E (i)
cel = . (1.1)
E (m)
Analogous scalar contrasts may be defined in terms of other material parameters; for a
more sophisticated elastic contrast tensor see, e.g., Sevostianov and Kachanov (2014).

Length Scales
In the present context the lowest length scale described by a given micromechanical model
is termed the microscale, of characteristic length ℓ, the largest one the macroscale, of char-
acteristic length L, and intermediate ones are called mesoscales4 . The fields describing the
behavior of an inhomogeneous material, i.e., in mechanics the stresses σ(x), strains ε(x)
and displacements u(x), are split into contributions corresponding to the different length
scales, which are referred to as micro-, macro- and mesofields, respectively. The phase
geometries on the meso- and microscales are denoted as meso- and microgeometries.

Micromechanical models are based on the assumption that the length scales in a given
inhomogeneous material are well separated, i.e., ℓ ≪ L. This is understood to imply that
for each micro–macro pair of scales, on the one hand, the fluctuating contributions to the
fields at the smaller length scale (“fast variables”) influence the behavior at the larger
length scale only via their volume averages. On the other hand, gradients of the fields as
well as compositional gradients at the larger length scale (“slow variables”) are taken to
be insignificant at the smaller length scale, where these fields appear to be locally constant
and can be described in terms of uniform “applied fields” or “far fields”. Formally, this
splitting of the strain and stress fields into slow and fast contributions can be written as

ε(x) = hεi + ε′ (x) and σ(x) = hσi + σ ′ (x) , (1.2)

where hεi and hσi are the macroscopic, volume averaged fields, whereas ε′ (x) and σ ′ (x)
stand for the microscopic fluctuations.

Unless specifically stated otherwise, in the present report the above conditions on the
slow and fast variables are assumed to be met. If this is not the case to a sufficient degree
(e.g., in the cases of insufficient separation of the length scales, of the presence of marked
compositional or load gradients, of regions in the vicinity of free surfaces of inhomogeneous
materials, or of macroscopic interfaces adjoined by at least one inhomogeneous material),
embedding schemes, compare Chapter 8, or special analysis methods must be applied. The
latter may take the form of second-order homogenization schemes that explicitly account
for deformation gradients on the microscale (Kouznetsova et al., 2004) and result in non-
local homogenized behavior, see, e.g., Feyel (2003).

Length scales below a given microscale are not necessarily amenable to continuum
mechanical descriptions; for an overview of pertinent methods see, e.g., Raabe (1998).
4
The nomenclature with respect to “micro”, “meso” and “macro” is far from universal, the naming
of the length scales of inhomogeneous materials being notoriously inconsistent even in literature dealing
specifically with continuum micromechanics. In a more neutral way, the smaller length scale may be
referred to as the “fine-grained” and the larger one as the “coarse-grained” length scale.

3
1.2 Homogenization and Localization
The two central aims of continuum micromechanics can be stated to be, on the one hand,
the bridging of length scales and, on the other hand, studying the structure–property re-
lationships of inhomogeneous materials.

The first of the above issues, the bridging of length scales, involves two main tasks.
On the one hand, the behavior at some larger length scale (the macroscale) must be es-
timated or bounded by using information from a smaller length scale (the microscale),
i.e., homogenization problems must be solved. The most important applications of ho-
mogenization are materials characterization, i.e., simulating the overall material response
under simple loading conditions such as uniaxial tensile tests, and constitutive modeling,
where the responses to general loads, load paths and loading sequences must be described.
Homogenization (also referred to as “upscaling” or “coarse graining”) may be interpreted
as describing the behavior of a material that is inhomogeneous at some lower length scale
in terms of a (fictitious) energetically equivalent, homogeneous reference material, which is
sometimes referred to as the homogeneous equivalent medium (HEM). Since homogeniza-
tion links the phase arrangement at the microscale to the macroscopic behavior, it will,
in general, provide microstructure–property relationships. On the other hand, the local
responses at the smaller length scale may be deduced from the loading conditions (and,
where appropriate, from the load histories) on the larger length scale. This task, which
corresponds to “zooming in” on the local fields in an inhomogeneous material, is referred
to as localization, downscaling or fine graining. In either case the main inputs are the
geometrical arrangement and the material behaviors of the constituents at the microscale.
In many continuum micromechanical methods, homogenization is less demanding than lo-
calization because the local fields tend to show a much more marked dependence on details
of the local geometry of the constituents.

For a volume element Ωs of an inhomogeneous material that is sufficiently large, contains

no significant gradients of composition, and shows no significant variations in the applied
loads, homogenization relations take the form of volume averages of some variable f (x),
Z
1
hf i = f (x) dΩ . (1.3)
Ωs Ωs

Accordingly, the homogenization relations for the stress and strain tensors can be expressed
as
Z Z
1 1
hεi = ε(x) dΩ = [u(x) ⊗ nΓ (x) + nΓ (x) ⊗ u(x)] dΓ
Ωs Ωs 2Ωs Γs
Z Z
1 1
hσi = σ(x) dΩ = t(x) ⊗ x dΓ , (1.4)
Ωs Ωs Ωs Γs

where Γs stands for surface of the volume element Ωs , u(x) is the displacement vector,
t(x) = σ(x) ∗ nΓ (x) is the surface traction vector, and nΓ (x) is the surface normal vec-
tor. Equations (1.4) are known as the average strain and average stress theorems, and the
surface integral formulation for ε given above pertains to the small strain regime and to
continuous displacements. Under the latter conditions the mean strains and stresses in a

4
control volume, hεi and hσi, are fully determined by the surface displacements and trac-
tions. If the displacements show discontinuities, e.g., due to imperfect interfaces between
the constituents or due to (micro) cracks, correction terms involving the displacement
jumps across imperfect interfaces or cracks must be introduced, compare Nemat-Nasser
and Hori (1993). In the absence of body forces the microstresses σ(x) are self-equilibrated
(but in general not zero). In the above form, eqn. (1.4) applies to linear elastic behavior,
but it can be modified to cover the thermoelastic regime and extended into the nonlinear
range, e.g., to elastoplastic materials described by secant or incremental plasticity models,
compare Section 2.9. For a discussion of averaging techniques and related results for finite
deformation regimes see, e.g., Nemat-Nasser (1999).

The microscopic strain and stress fields, ε(x) and σ(x), in a given volume element Ωs
are formally linked to the corresponding macroscopic responses, hεi and hσi, by localization
(or projection) relations of the type

ε(x) = A(x)hεi and σ(x) = B(x)hσi . (1.5)

A(x) and B(x) are known as mechanical strain and stress concentration tensors (or in-
fluence functions (Hill, 1963), influence function tensors, interaction tensors), respectively.
When these are known the localization problem is solved.

Equations (1.2) and (1.4) imply that the volume averages of fluctuations vanish for
sufficiently large integration volumes,
Z Z
′ 1 ′ ′ 1
hε i = ε (x) dΩ = o and hσ i = σ ′ (x) dΩ = o . (1.6)
Ωs Ωs Ωs Ωs

Similarly, surface integrals over the microscopic fluctuations of the field variables tend to
zero for appropriate volume elements5 .

For suitable volume elements of inhomogeneous materials that show sufficient separa-
tion between the length scales and for suitable boundary conditions the relation
Z
1 T 1 1
he
σ eεi = e T (x) e
σ σ iT he
ε(x) dΩ = he εi (1.7)
2 2Ω 2
Ω

must hold for general statically admissible stress fields σ e and kinematically admissible
strain fields e
ε, compare Hill (1967) and Mandel (1972). This equation is known as Hill’s
macrohomogeneity condition, the Hill–Mandel condition or the energy equivalence con-
dition, compare Nemat-Nasser (1999), Bornert (2001) and Zaoui (2001). Equation (1.7)
states that the strain energy density of the microfields equals the strain energy density of
the macrofields, making the microscopic and macroscopic descriptions energetically equiva-
lent. Interestingly, the fluctuations of the microfields do not contribute to the macroscopic
strain energy, i.e.,
T
hσ ′ ε′ i = 0 . (1.8)
5
Whereas volume and surface integrals over products of slow and fast variables vanish, integrals over
2
products of fluctuating variables (“correlations”), e.g., hε′ i, remain finite in general.

5
The Hill–Mandel condition forms the basis of the interpretation of homogenization pro-
cedures in the thermoelastic regime6 in terms of a homogeneous comparison material (or
“reference medium”) that is energetically equivalent to a given inhomogeneous material.

1.3 Volume Elements

The second central issue of continuum micromechanics, studying the structure–property
relationships of inhomogeneous materials, obviously requires suitable descriptions of their
structure at the appropriate length scale, i.e., within the present context, their microge-
ometries.

The microgeometries of real inhomogeneous materials are at least to some extent ran-
dom and, in the majority of cases of practical relevance, their detailed phase arrangements
are highly complex. As a consequence, exact expressions for A(x), B(x), ε(x), σ(x), etc., in
general cannot be given with reasonable effort, and approximations have to be introduced.
Typically, these approximations are based on the ergodic hypothesis, i.e., the heterogeneous
material is assumed to be statistically homogeneous. This implies that sufficiently large
volume elements selected at random positions within a sample have statistically equivalent
phase arrangements and give rise to the same averaged material properties7 . As mentioned
above, such material properties are referred to as the overall or effective material properties
of the inhomogeneous material.

Ideally, the homogenization volume should be chosen to be a proper representative

volume element (RVE) as defined by Hill (1963), which essentially is a subvolume of Ωs
of sufficient size to contain all information necessary for describing the behavior of the
composite. Representative volume elements have been defined, on the one hand, by re-
quiring them to be statistically representative of a given microgeometry in terms of purely
geometry-based descriptors, the resulting “geometrical RVEs” being independent of the
physical property to be studied. On the other hand, their definition can be based on the
requirement that the overall responses with respect to some given physical behavior do
not depend on the actual position of the RVE nor on the boundary conditions applied to
it8 . The size of such “physical RVEs” depends not only on the microgeometry, but also
on the physical property considered. Bornert (2001) argued that, for given microgeome-
tries, geometrically representative volumes are always smaller than mechanically (and thus
6
For a discussion of the Hill–Mandel condition for finite deformations and related issues see, e.g.,
Khisaeva and Ostoja-Starzewski (2006).
7
Some inhomogeneous materials are not statistically homogeneous by design, e.g., functionally graded
materials in the direction(s) of the gradient(s), and, consequently, may require nonstandard treatment.
For such materials it is not possible to define effective material properties in the sense of eqn. (1.10).
Deviations from statistical homogeneity may also be introduced into inhomogeneous materials as side
effects of manufacturing processes.
8
These two approaches emphasize different aspects of the original definition of Hill (1963), which requires
the RVE of an inhomogeneous material to be (a) “entirely typical of the whole mixture on average” and to
contain (b) “a sufficient number of inclusions for the apparent properties to be independent of the surface
values of traction and displacement, so long as these values are macroscopically uniform”.
Alternatively, less stringent definitions were proposed, e.g., regarding an RVE as the “smallest possible
structure that could sufficiently capture the macroscopic response” of an inhomogeneous material (Firooz
et al., 2019).

6
physically) representative ones. For alternative definitions of RVEs see, e.g., Bouhfid et al.
(2019).

Within either definition, an RVE must be sufficiently large to allow a meaningful sam-
pling of the microscopic fields and sufficiently small for the influence of macroscopic gradi-
ents to be negligible. In addition, it must be smaller than typical samples or components9 .
For further discussion of microgeometries and homogenization volumes see Section 5.1. For
methods involving the analysis of discrete volume elements, the latter must, in addition,
be sufficiently small for an analysis of the microfields to be feasible.

(p)
The fields in a constituent can be split into phase averages and fluctuations by
analogy to eqn. (1.2) as
′ ′
ε(p) (x) = hεi(p) + ε(p) (x) and σ (p) (x) = hσi(p) + σ (p) (x) . (1.9)
′ ′
The ε(p) (x) and σ (p) (x) are referred to as intra-phase fluctuations, whereas differences be-
tween the hεi(p) and hσi(p) describe the inter-phase heterogeneity of the fields. In the case
of materials with matrix–inclusion topology the variations of the fields between the individ-
ual particles or fibers making up the inhomogeneity phase, the inter-particle or inter-fiber
fluctuations, and the field gradients within individual inhomogeneities, the intra-particle
or intra-fiber fluctuations, may also be of interest.

Needless to say, volume elements should be chosen to be as simple possible in order to

limit the modeling effort, but their level of complexity must be sufficient for covering the
aspects of their behavior targeted by a given study, compare, e.g., the examples given by
Forest et al. (2002).

1.4 Overall Behavior, Material Symmetries

The homogenized strain and stress fields of an elastic inhomogeneous material as obtained
by eqn. (1.4), hεi and hσi, can be linked by effective elastic tensors E∗ and C∗ as

hσi = E∗ hεi and hεi = C∗ hσi , (1.10)

which may be viewed as the elasticity tensor (or “tensor of elastic moduli”) and compliance
tensor, respectively, of an appropriate, energetically equivalent homogeneous material, with
C∗ = E∗ −1 . Using eqns. (1.4) and (1.5) these effective elastic tensors can be obtained from
the local elastic tensors, E(x) and C(x), and the concentration tensors, A(x) and B(x),
by volume averaging
Z
∗ 1
E = E(x)A(x)dΩ
Ωs Ωs
Z
∗ 1
C = C(x)B(x)dΩ , (1.11)
Ωs Ωs
9
This requirement was symbolically denoted as MICRO≪MESO≪MACRO by Hashin (1983), where
MICRO and MACRO have their “usual” meanings and MESO stands for the length scale of the homog-
enization volume. As noted by Nemat-Nasser (1999) it is the dimension relative to the microstructure
relevant for a given problem that is important for the size of an RVE.

7
Other effective properties of inhomogeneous materials, e.g., tensors describing their ther-
mophysical behavior, can be evaluated in an analogous way.

The resulting homogenized behavior of many multi-phase materials can be idealized as

being statistically isotropic or quasi-isotropic (e.g., for composites reinforced with spher-
ical particles, randomly oriented particles of general shape or randomly oriented fibers,
many polycrystals, many porous and cellular materials, random mixtures of two phases)
or statistically transversally isotropic (e.g., for composites reinforced with aligned fibers or
platelets, composites reinforced with nonaligned reinforcements showing a planar random
or other axisymmetric orientation distribution function, etc.), compare (Hashin, 1983). Of
course, lower material symmetries of the homogenized response may also be found, e.g.,
in textured polycrystals or in composites containing reinforcements with orientation dis-
tributions of low symmetry, compare Allen and Lee (1990).

Statistically isotropic multi-phase materials show the same overall behavior in all direc-
tions, and their effective elasticity tensors and thermal expansion tensors take the forms


∗ ∗ ∗
  ∗
E11 E12 E12 0 0 0 α
E12∗ ∗ ∗  α∗ 
 ∗ E11 E12 0 0 0   ∗
E12 E12 E11 0
∗ ∗
0 0  α 
∗ 
E =  α =
∗  (1.12)
0 0 0 E ∗
0 0  0
 44   
 0 0 0 0 E44∗
0  0
∗ 1 ∗ ∗
0 0 0 0 0 E44 = 2 (E11 − E12 ) 0

in Voigt/Nye notation. Two independent parameters are sufficient for describing isotropic
∗ 2
overall linear elastic behavior (e.g., the effective Young’s modulus E ∗ = E11
∗
−2E12 ∗
/(E11 +
∗ ∗ ∗ ∗ ∗
E12 ), the effective Poisson’s ratio ν = E12 /(E11 + E12 ), the effective shear modulus
G∗ = E44 ∗
= E ∗ /2(1+ν ∗ ), the effective bulk modulus K ∗ = (E11
∗ ∗
+2E12 )/3 = (E ∗ /3(1−2ν ∗ ),
or the effective Lamé constants) and one is required for the effective thermal expansion
behavior in the linear range (the effective coefficient of thermal expansion α∗ = α11 ∗
).

The effective elasticity and thermal expansion tensors for statistically transversally
isotropic materials have the structure
 ∗ ∗ ∗
  ∗
E11 E12 E12 0 0 0 αA
E12∗ ∗ ∗   ∗
 ∗ E22 E23 0 0 0   αT 
E12 E23 ∗ ∗
E22 0 0 0   αT ∗
∗
E =   α = 
∗  , (1.13)
0 0 0 E ∗
0 0  0
 44   
 0 0 0 0 E44 ∗
0  0
∗ 1 ∗ ∗
0 0 0 0 0 E66 = 2 (E22 − E23 ) 0

where 1 is the axial direction and 2–3 is the transverse plane of isotropy. Generally, the
thermoelastic behavior of transversally isotropic materials is described by five indepen-
dent elastic constants and two independent coefficients of thermal expansion. Appropriate
elasticity parameters in this context are, e.g., the axial and transverse effective Young’s
2E12
∗2 ∗ E ∗2 +E ∗ E ∗2 −2E ∗ E ∗2
E11
moduli, EA∗ = E11∗
− E ∗ +E ∗ ∗
∗ and ET = E22 −
23
E11
22 12
∗ E ∗ −E ∗2
23 12
, the axial and transverse
22 23 22 12
∗ ∗ ∗ ∗
effective shear moduli, GA = E44 and GT = E66 , the axial and transverse effective Poisson’s

8
 E∗ E ∗ E ∗ −E ∗2
ratios, νA∗ =ν12 = E ∗ +E
12
∗ and νT∗ =ν23 = E11 23
∗2 , as well as the effective transverse (plane
12
11 E22 −E12
∗ ∗
22 23
strain) bulk modulus KT = (E22 +E23 )/2 = EA /2[(1−νT∗ )(EA∗ /ET∗ )−2νA∗ 2 ]. The transverse
∗ ∗ ∗ ∗

(“in-plane”) properties are related via G∗T = ET∗ /2(1 + νT∗ ), but there is no general linkage
between the axial properties EA∗ , G∗A and νA∗ beyond the above definition of KT∗ . For the
special case of materials consisting of aligned phases that are continuous in the 1-direction,
however, the Hill (1964) connections,

(f)  
(f) (m) 4(νA − ν (m) )2 ξ 1 1−ξ
EA∗ = ξEA + (1 − ξ)E + (f) (m) (f)
+ (m) − ∗
(1/KT − 1/KT )2 KT KT KT
(f)  
(f) νA − ν (m) ξ 1−ξ 1
νA∗ = ξνA + (1 − ξ)ν (m) + (f) (m) (f)
+ (m) − ∗ , (1.14)
1/KT − 1/KT KT KT KT

allow the effective moduli EA∗ and νA∗ to be expressed by KT∗ , some constituent properties,
and the fiber volume fraction ξ = Ω(f) /Ω; see also Milton (2002). Analogous relations hold
for unidirectionally reinforced composites of tetragonal macroscopic symmetry (Berggren
et al., 2003). Both an axial and a transverse effective coefficient of thermal expansion,
∗ ∗ ∗ ∗
αA = α11 and αT = α22 , are required for transversally isotropic materials.

The overall material symmetries of inhomogeneous materials and their effect on var-
ious physical properties can be treated in full analogy to the symmetries of crystals as
discussed, e.g., by Nye (1957). Accordingly, deviations of predicted elastic tensors from
macroscopically isotropic elastic symmetry can be assessed via a Zener (1948) anisotropy
∗ ∗ ∗
ratio, Z = 2E66 /(E22 − E23 ), or other anisotropy parameters, see, e.g., Kanit et al. (2006).

The influence of the overall symmetry of the phase arrangement on the overall me-
chanical behavior of inhomogeneous materials can be marked10 , especially in nonlinear
regimes. Accordingly, it is good practice to aim at approximating the symmetry of the
actual material as closely as possible in any modeling effort.

1.5 Major Modeling Strategies in Continuum Micro-

mechanics of Materials
All micromechanical methods described in the present report can be used to carry out ma-
terials characterization, i.e., simulating the overall material response under simple loading
conditions such as uniaxial tensile tests. Many homogenization procedures can also be em-
ployed directly to provide micromechanically based constitutive material models at higher
length scales. This implies that they allow evaluating the full homogenized stress and
strain tensors for any pertinent loading condition and for any pertinent loading history11 .
10
Overall properties described by tensors or lower rank, e.g., thermal expansion and thermal conduction,
are less sensitive to material symmetry effects than are mechanical responses, compare Nye (1957).
11
The overall thermomechanical behavior of homogenized materials is often richer than that of the
constituents, i.e., the effects of the interaction of the constituents in many cases cannot be satisfactorily
described by simply adapting material parameters without changing the functional relationships in the
constitutive laws of the constituents. For example, a composite consisting of a matrix that follows J2
plasticity and elastic reinforcements shows some pressure dependence in its macroscopic plastic behavior,

9
This task is obviously much more demanding than materials characterization. Compared
to semi-empirical constitutive laws, as proposed, e.g., by Davis (1996), micromechanically
based constitutive models have both a clear physical basis and an inherent capability for
“zooming in” on the local phase stresses and strains by using localization procedures.

Evaluating the local responses of the constituents (in the ideal case, at any material
point) for a given macroscopic state of a sample or structure, i.e., localization, is of special
interest for identifying local deformation mechanisms and for studying as well as assessing
local strength relevant behavior, such as the onset and progress of plastic yielding or of
damage, which, of course, can have major repercussions on the macroscopic behavior. For
valid descriptions of local strength-relevant responses details of the microgeometry tend to
be of major importance and may, in fact, determine the macroscopic response, an extreme
case being the mechanical strength of brittle inhomogeneous materials.

Because for realistic phase distributions an exact analysis of the spatial variations of the
microfields in large volume elements tends to be beyond present capabilities12 suitable ap-
proximations must be introduced. For convenience, the majority of the resulting modeling
approaches may be treated as falling into two groups. The first of these comprises methods
that describe interactions, e.g., between phases or between individual reinforcements, in a
collective way, its main representatives being

• Mean-Field Approaches (MFAs) and related methods (see Chapter 2): Highly ideal-
ized microgeometries are used (compare, e.g., fig 2.1) and the microfields within each
constituent are approximated by their phase averages hεi(p) and hσi(p) , i.e., phase-
wise uniform stress and strain fields are employed. The phase geometry enters these
models, sometimes implicitly, via statistical descriptors, such as volume fractions,
macroscopic symmetry, phase topology, reinforcement aspect ratios, etc. In MFAs
the localization relations take the form

hεi(p) = Ā(p) hεi

hσi(p) = B̄(p) hσi (1.15)

and the homogenization relations can be written as

Z X
(p) 1
hεi = (p) ε(x) dΩ with hεi = V (p) hεi(p)
Ω (p) p
ZΩ X
1
hσi(p) = (p) σ(x) dΩ with hσi = V (p) hσi(p) , (1.16)
Ω Ω (p) p

where (p) denotes a given P phase of the material, Ω(p) is the volume occupied by
(p) (p)
this phase, and V =Ω / k Ω(k) =Ω(p) /Ωs is the volume fraction of the phase. In
contrast to eqn. (1.5) the phase concentration tensors Ā and B̄ used in MFAs are
and the macroscopic flow behavior of inhomogeneous materials can lose normality even though that of
each of the constituents is associated (Li and Ostoja-Starzewski, 2006). Also, two dissimilar constituents
following Maxwell-type linear viscoelastic behavior do not necessarily give rise to a macroscopic Maxwell
behavior (Barello and Lévesque, 2008).
12
Exact predictions of the effective properties would require an infinite set of correlation functions for
statistically characterizing the inhomogeneous microstructure, compare Torquato et al. (1999).

10
 not functions of the spatial coordinates13 .
Mean-field approaches tend to be formulated (and provide estimates for effective
properties) in terms of the phase concentration tensors, they pose low computational
requirements, and they have been highly successful in describing the thermoelastic
response of inhomogeneous materials. Their use in modeling nonlinear composites
continues to be a subject of active research. Their most important representatives
are effective-field and effective-medium approximations.

• Bounding Methods (see Chapter 3): Variational principles are used to obtain upper
and (in many cases) lower bounds on the overall elastic tensors, elastic moduli, secant
moduli, and other physical properties of inhomogeneous materials the microgeome-
tries of which are described by statistical parameters. Many analytical bounds are
obtained on the basis of phase-wise constant stress polarization fields, making them
closely related to MFAs. Bounds — in addition to their intrinsic value — are tools of
vital importance in assessing other models of inhomogeneous materials. Furthermore,
in most cases one of the bounds provides useful estimates for the physical property
under consideration, even if the bounds are rather slack (Torquato, 1991).

Because they do not explicitly account for pair-wise or multi-particle interactions mean-
field approaches have sometimes been referred to as “non-interacting approximations” in
the literature. This designator, however, is best limited to the dilute regime in the sense
of Section 2.2, collective interactions being explicitly incorporated into mean-field models
for non-dilute volume fractions. MFA and bounding methods implicitly postulate the ex-
istence of a representative volume element.

The second group of approximations are based on studying discrete microgeometries,

for which they aim at evaluating the microfields at high resolution, thus fully accounting
for the interactions between phases within the “simulation box”. It includes the following
groups of models, compare the sketches in fig. 1.1.

• Periodic Microfield Approaches (PMAs), often referred to as periodic homogenization

schemes and sometimes as unit cell methods, see Chapter 6. In such models the inho-
mogeneous material is approximated by an infinitely extended model material with
a periodic phase arrangement. The resulting periodic microfields are usually evalu-
ated by analyzing a repeating volume element (which may describe microgeometries
ranging from rather simplistic to highly complex ones) via analytical or numerical
methods. Such approaches are often used for performing materials characterization
of inhomogeneous materials in the nonlinear range, but they can also be employed as
micromechanically based constitutive models. The high resolution of the microfields
provided by PMAs can be very useful in studying the initiation of damage at the
microscale. However, because they inherently give rise to periodic configurations of
damage and patterns of cracks, PMAs typically are not a good choice for investi-
gating phenomena such as the interaction of the microgeometry with macroscopic
cracks.
Periodic microfield approaches can give detailed information on the local stress and
13
Surface integral formulations analogous to eqn. (1.4) may be used to evaluate consistent expressions
for hεi(p) for void-like and hσi(p) for rigid inhomogeneities embedded in a matrix.

11
 strain fields within a given unit cell, but they tend to be relatively expensive compu-
tationally. Among the methods in the present group they are the only ones that do
not intrinsically give rise to boundary layers in the microfields.

• Windowing Approaches (see Chapter 7): Subregions (“windows”) — usually, but

not necessarily, of rectangular or hexahedral shape — are randomly chosen from
a given phase arrangement and subjected to boundary conditions that guarantee
energy equivalence between the micro- and macroscales. Accordingly, windowing
methods describe the behavior of individual inhomogeneous samples rather than of
inhomogeneous materials and typically give rise to apparent rather than effective
macroscopic responses. For the special cases of macrohomogeneous stress and strain
boundary conditions, respectively, lower and upper estimates for and bounds on the
overall behavior of the inhomogeneous material can be obtained. In addition, mixed
homogeneous boundary conditions can be applied in order to generate estimates.

• Embedded Cell or Embedding Approaches (ECAs; see Chapter 8): The inhomoge-
neous material is approximated by a model consisting of a “core” containing a discrete
phase arrangement that is embedded within some outer region showing smeared-out
material behavior; far-field loads are applied to this outer region. The material prop-
erties of the embedding layer may be described by some macroscopic constitutive law,
they can be determined self-consistently or quasi-self-consistently from the behavior
of the core, or the embedding region may take the form of a coarse description and/or
discretization of the phase arrangement. ECAs can be used for materials character-
ization, and they are usually the best choice for studying regions of special interest
in inhomogeneous materials, such as the surroundings of tips of macroscopic cracks.
Like PMAs, embedded cell approaches can resolve local stress and strain fields in the
core region at high detail, but tend to be computationally expensive.

• Other homogenization approaches employing discrete microgeometries, such as sub-

modeling techniques (on which some remarks are given in Chapter 8), or the statistics-
based non-periodic homogenization scheme of Li and Cui (2005).

Because the above group of methods explicitly study mesodomains as defined by Hashin
(1983) they are sometimes referred to as “Mesoscale Approaches”. Their ability of highly
resolving the microfields underlies the designation “Full Field Models” and their reliance
on numerical methods led to the name “Direct Numerical Simulations”. Figure 1.1 shows
a sketch of a volume element as well as PMA, ECA and windowing approaches applied to it.

Some further descriptions that have been applied to studying the macroscopic thermo-
mechanical behavior of inhomogeneous materials, such as isostrain and isostress models
(the former being known as the “rules of mixture”) and the Halpin–Tsai equations14 are
14
For certain moduli of composites reinforced by particles or by continuous aligned fibers the Halpin–
Tsai equations can be obtained from the estimates of Kerner (1956) and Hermans (1967), respectively.
Following the procedures outlined in Halpin and Kardos (1976) these models, together with Hill’s con-
nections, eqns. (1.14), and some minor approximations, yield sets of specific “Halpin–Tsai-parameters”
which, however, do not appear to have been used widely in practice. In most cases, instead, the Halpin–
Tsai equations were applied in a semi-empirical way to other moduli or to other composite geometries, or
were used with other sets of parameters of various provenience.

12
 PHASE ARRANGEMENT PERIODIC APPROXIMATION,
UNIT CELL

EMBEDDED CONFIGURATION WINDOW

Figure 1.1: Schematic sketch of a random matrix–inclusion microstructure and of the volume ele-
ments used by a periodic microfield method (which employs a slightly different periodic “model”
microstructure), an embedding scheme and a windowing approach to studying this inhomoge-
neous material.

not discussed here because, with the exception of special cases, their connection to ac-
tual microgeometries is not very strong, limiting their predictive capabilities. For brevity,
a number of micromechanical models with solid physical basis, such as expressions for
self-similar composite sphere assemblages (CSA, Hashin (1962)) and composite cylinder
assemblages (CCA, Hashin and Rosen (1964)), are not covered within the present discus-
sion, either.

For studying materials that are inhomogeneous at a number of (sufficiently widely

spread) length scales (e.g., materials in which well defined clusters of inhomogeneities are
present), hierarchical procedures that use homogenization at more than one level are a
natural extension of the above concepts. Such multi-scale or sequential homogenization
models are the subject of a short discussion in Chapter 9.

A final group of models used for studying inhomogeneous materials is the direct nu-
merical simulation of microstructured structures or samples. Until fairly recently, com-
putational requirements restricted such modeling approaches to configurations exceeding
the size of the microscale by less than, say, one or two orders of magnitude, see, e.g.,

13
Papka and Kyriakides (1994), Silberschmidt and Werner (2001), Luxner et al. (2005) or
Tekoğlu et al. (2011). However, improvements in computing power have allowed extending
this type of model to larger structures, compare, e.g., Bishop et al. (2015). Even though
the geometries considered may appear similar to the ones used in windowing approaches,
compare Chapter 7, models of the above type by design aim at describing the behavior
of small structures rather than that of materials. As a consequence, they are typically
subjected to boundary conditions and load cases that are pertinent to structural (among
them bending or indentation loads), but not to material behavior. The effects of these
boundary conditions as well as the sample’s size in terms of the characteristic length of
the inhomogeneities, ℓ, tend to play considerable roles in the mechanical behavior of such
inhomogeneous structures, which is typically evaluated in terms of force vs. displacement
rather than stress vs. strain curves. Arguably, such structural models are not part of the
“core tool set” of micromechanics, because they do not involve scale transitions.

1.6 Model Verification and Validation

Micromechanical approaches aim at generating predictive models for the behavior of in-
homogeneous materials. Obviously, model verification, i.e., monitoring a model’s correct
setup and implementation, as well as validation, i.e., assessing the accuracy of a model’s
representation of the targeted material behavior, play important roles in this context. Of
course, both model verification and validation in continuum micromechanics require keep-
ing tabs on the consistency and plausibility of modeling assumptions and results throughout
a given study.

For verification, it is often possible to compare predictions obtained with different, un-
related micromechanical methods that pertain to a given geometrical configuration and
employ the same constitutive models as well as material parameters. In such a setting,
finding good agreement over sets of markedly different material parameters is a strong
indicator (but definitely not a guarantee) that the models involved successfully capture
the physics of the problem.

Bounding methods, see Chapter 3, play an especially important role in verification

studies: When it comes to studying the linear elastic (or conduction) responses of inho-
mogeneous materials, in many cases appropriate bounds are available, compare Section 3.
Analytical or numerical models that violate the Hashin–Shtrikman bounds pertaining to
the macroscopic symmetry and microscopic geometrical configuration of the material in the
linear regime by more than trivial differences (which may be due to, e.g., roundoff errors),
must be viewed as at least potentially flawed15 . In such verification assessments the value
of an estimate relative to the pertinent bounds may provide additional information: it is
well known that stiff inhomogeneities in a compliant matrix give rise to effective moduli
closer to the lower and compliant inhomogeneities in a stiff matrix to estimates closer to the
upper bound (Torquato, 1991), with analogous relationships holding for conduction prob-
lems. It is good practice to assess linear versions of models intended for studying nonlinear
behavior against the appropriate bounds — this essentially provides a check if a given
15
If data from measurements do not comply with the pertinent Hashin–Shtrikman bounds the most
common reason is inaccurate data for the constituent behavior or the volume fractions.

14
model is capable of producing a physically valid partitioning of the stress (flux) and strain
(gradient) fields between the target material’s phases in a fairly well understood regime.
Issues cropping up with linear behavior tend not to disappear, but rather to become more
marked in the more complex nonlinear setting. For generality of the results it typically
makes sense to carry out verification analyses for parameter combinations that “stress” the
methods involved, which in the context of continuum micromechanics often means phase
contrasts that differ clearly from unity and high inhomogeneity volume fractions.

In the case of modeling efforts making use of discretizing numerical engineering methods
for evaluating position dependent fields in a volume element, convergence studies involving
successively finer discretizations are an important part of verification. In this context it is
worth noting that convergence in terms of the macroscopic responses does not guarantee
convergence in terms of the microscopic fields, and, especially, of the latters’ minimum
and maximum values. For full field simulations convergence studies in terms of the size of
volume elements may also be of considerable interest.

The validation of micromechanical models against experimental data is obviously highly

important, but may be tricky in practice due to the considerable number of potential
sources of discrepancies — on the modeling side they include the representativeness of the
volume elements used, the quality of the material models employed for the constituents,
the accuracy of the material parameters prescribed to these constitutive models as well as
errors in the implementation or application of the model. Obtaining perfect or near-perfect
agreement between measurements and predictions without “tuning” input parameters in
general is not the most probable of outcomes, especially when damage and failure modeling
are involved. Accordingly, even highly successful validation against some set of experimen-
tal data does not remove the need for model verification as defined above. Specifically, a
model giving predictions that are close to experimental results but fall outside the pertinent
bounds in the linear range cannot be treated as verified — after all, even good agreement
with measurements may be due to the canceling of different errors arising from the above
issues. In a similar vein, obtaining responses close to experimental results from a model
that takes major liberties in terms of the microgeometry constitutes neither verification
nor proper validation.

Obtaining reliable values for material parameters pertinent to the microscale tends to
be a major challenge in micromechanics, especially when a model involves damage. A
fairly common way of dealing with this problem is carrying out parameter identification
via inverse procedures built around the micromechanical model and available experimental
results. When following such a strategy, using the same set of experimental data for
both parameter identification and model verification must be avoided — in such a setting
“overspecialization” of the parameters to that data set may occur, which makes model and
parameters unsuitable for generalization to other situations. In a common scenario of this
type the nonlinear macroscopic behavior of a composite is to be studied and experimental
data are limited to results from uniaxial tensile tests. In such a case even a perfect match
of predictions with the above experimental results is far from guaranteeing that load cases
involving other macroscopic triaxialities, e.g., shear, or complex stress trajectories will be
described reasonably well by the model.

15
Chapter 2

Mean-Field Methods

In this chapter mean-field methods are discussed mainly in the context of two-phase ma-
terials. Special emphasis is put on effective-field methods of the Mori–Tanaka type, which
may be viewed as the simplest mean-field approaches to modeling inhomogeneous materi-
als that encompass the full physical range of phase volume fractions16 . Unless specifically
stated otherwise, the material behavior of both reinforcements and matrix is taken to be
linear (thermo)elastic, both strains and temperature changes being assumed to be small.
Perfect bonding between the constituents is assumed in all cases. There is an extensive
body of literature covering relevant mean-field approaches, so that the following treatment
is far from complete.

2.1 General Relations between Mean Fields in Thermo-

elastic Two-Phase Materials
Throughout this report additive decomposition of strains is used. For example, for the
case of thermoelastoplastic material behavior the total strain tensor can be written in this
approximation as
ε = εel + εpl + εth , (2.1)
where εel , εpl and εth denote the elastic, plastic and thermal strains, respectively. The strain
and stress tensors may be split into volumetric/hydrostatic and deviatoric contributions

ε = εvol + εdev = Ovol ε + Odev ε

σ = σ hyd + σ dev = Ovol σ + Odev σ , (2.2)

Ovol and Odev being volumetric and deviatoric projection tensors.

For thermoelastic inhomogeneous materials, the macroscopic stress–strain relations can

be written in the form

hσi = E∗ hεi + λ∗ ∆T
hεi = C∗ hσi + α∗ ∆T . (2.3)
16
Because Eshelby and Mori–Tanaka methods are specifically suited for matrix–inclusion-type micro-
topologies, the expression “composite” is often used in the present chapter instead of the more general
designation “inhomogeneous material”.

16
Here the expression α∗ ∆T corresponds to the macroscopic thermal strain tensor, λ∗ =
−E∗ α∗ is the macroscopic specific thermal stress tensor (i.e., the overall stress response
of a fully constrained material to a purely thermal unit load, also known as the tensor of
thermal stress coefficients), and ∆T stands for the (spatially homogeneous) temperature
difference with respect to some (stress-free) reference temperature. The constituents, here
a matrix (m) and inhomogeneities (i) , are also assumed to behave thermoelastically, so that
hσi(m) = E(m) hεi(m) + λ(m) ∆T hσi(i) = E(i) hεi(i) + λ(i) ∆T
hεi(m) = C(m) hσi(m) + α(m) ∆T hεi(i) = C(i) hσi(i) + α(i) ∆T , (2.4)
where the relations λ(m) = −E(m) α(m) and λ(i) = −E(i) α(i) hold.

From the definition of phase averaging, eqn. (1.16), relations between the phase aver-
aged fields in the form
hεi = ξhεi(i) + (1 − ξ)hεi(m) = εa
hσi = ξhσi(i) + (1 − ξ)hσi(m) = σ a (2.5)
follow immediately, where ξ = V (i) = Ω(i) /Ωs stands for the volume fraction of the rein-
forcements and 1 − ξ = V (m) = Ω(m) /Ωs for the volume fraction of the matrix. εa and
σ a denote far-field (“applied”) homogeneous stress and strain tensors, respectively, with
εa = C∗ σ a . Perfect interfaces between the phases are assumed in expressing the macro-
scopic strain of the composite as the weighted sum of the phase averaged strains.

The phase averaged strains and stresses can be related to the overall strains and stresses
by the phase averaged strain and stress concentration (or localization) tensors Ā, β̄, B̄, and
κ̄ (Hill, 1963), respectively, which are defined for thermoelastic inhomogeneous materials
by the expressions
(m) (i)
hεi(m) = Ā(m) hεi + β̄ ∆T hεi(i) = Ā(i) hεi + β̄ ∆T
hσi(m) = B̄(m) hσi + κ̄(m) ∆T hσi(i) = B̄(i) hσi + κ̄(i) ∆T , (2.6)
see eqn. (1.15) for the purely elastic case. Ā and B̄ are referred to as the mechanical or
(elastic) phase stress and strain concentration tensors, respectively, and β̄ as well as κ̄ are
the corresponding thermal concentration tensors. In contrast to the elastic tensors E and
C, the concentration tensors Ā and B̄ are not necessarily symmetric.

By using eqns. (2.5) and (2.6), the strain and stress concentration tensors can be shown
to fulfill the relations
(i) (m)
ξ Ā(i) + (1 − ξ)Ā(m) = I ξ β̄ + (1 − ξ)β̄ =o
ξ B̄(i) + (1 − ξ)B̄(m) = I (i)
ξ κ̄ + (1 − ξ)κ̄ (m)
=o , (2.7)
where I stands for the symmetric rank 4 identity tensor and o for the rank 2 null tensor.

The effective elasticity and compliance tensors of a two-phase composite can be obtained
from the properties of the constituents and from the mechanical concentration tensors as
E∗ = ξE(i) Ā(i) + (1 − ξ)E(m) Ā(m)
= E(m) + ξ[E(i) − E(m) ]Ā(i) = E(i) + (1 − ξ)[E(m) − E(i) ]Ā(m) (2.8)

17
 C∗ = ξC(i) B̄(i) + (1 − ξ)C(m) B̄(m)
= C(m) + ξ[C(i) − C(m) ]B̄(i) = C(i) + (1 − ξ)[C(m) − C(i) ]B̄(m) , (2.9)

compare eqn. (1.11).

The tensor of effective thermal expansion coefficients, α∗ , and the specific thermal
stress tensor, λ∗ , can be related to the thermoelastic phase behavior and the thermal
concentration tensors as

α∗ = ξ[C(i) κ̄(i) + α(i) ] + (1 − ξ)[C(m) κ̄(m) + α(m) ]

= ξα(i) + (1 − ξ)α(m) + (1 − ξ)[C(m) − C(i) ]κ̄(m)
= ξα(i) + (1 − ξ)α(m) + ξ[C(i) − C(m) ]κ̄(i) . (2.10)

(i) (m)
λ∗ = ξ[E(i) β̄ + λ(i) ] + (1 − ξ)[E(m) β̄ + λ(m) ]
(m)
= ξλ(i) + (1 − ξ)λ(m) + (1 − ξ)[E(m) − E(i) ]β̄
(i)
= ξλ(i) + (1 − ξ)λ(m) + ξ[E(i) − E(m) ]β̄ . (2.11)

The above expressions can be derived by inserting eqns. (2.4) and (2.6) into eqns. (2.5)
and comparing with eqns. (2.3). Alternatively, the effective thermal expansion coefficient
and specific thermal stress coefficient tensors of multi-phase materials can be obtained as
X
α∗ = ξ (p) (B̄(p) )T α(p)
(p)
X
λ ∗
= ξ (p) (Ā(p) )T λ(p) , (2.12)
(p)

compare (Mandel, 1965; Levin, 1967), where the sums run over all phases (p) . The above
expression for α∗ being known as the Mandel–Levin formula. If the effective compliance
tensor of a two-phase material is known, eqn. (2.7) can be inserted into eqn. (2.12) to give
the overall coefficients of thermal expansion as

α∗ = (C∗ − C(m) )(C(i) − C(m) )−1 α(i) − (C∗ − C(i) )(C(i) − C(m) )−1 α(m) . (2.13)

The mechanical stress and strain concentration tensors for a given phase are linked by
expressions of the type

Ā(p) = C(p) B̄(p) E∗ and B̄(p) = E(p) Ā(p) C∗ , (2.14)

from which E∗ and C∗ may be eliminated via eqn. (2.79). For two-phase materials the
concentration tensors can be evaluated from the effective and the phase-level elastic tensors,
e.g.,

(1 − ξ)Ā(m) = (E(m) − E(i) )−1 (E∗ − E(i) )

ξ B̄(i) = (C(i) − C(m) )−1 (C∗ − C(m) ) . (2.15)

18
on the basis of eqns. (2.7) and (2.7). Furthermore, by invoking the principle of virtual
work Benveniste and Dvorak (1990) developed relations that the thermal strain concentra-
(p)
tion tensors, β̄ , to the mechanical strain concentration tensors, Ā(p) , and the thermal
stress concentration tensors, κ̄(p) , to the mechanical stress concentration tensors, B̄(p) ,
respectively, as
(m)
β̄ = [I − Ā(m) ][E(i) − E(m) ]−1 [λ(m) − λ(i) ]
(i)
β̄ = [I − Ā(i) ][E(m) − E(i) ]−1 [λ(i) − λ(m) ]
κ̄(m) = [I − B̄(m) ][C(i) − C(m) ]−1 [α(m) − α(i) ]
κ̄(i) = [I − B̄(i) ][C(m) − C(i) ]−1 [α(i) − α(m) ] . (2.16)

From eqns. (2.9) to (2.16) it is evident that the knowledge of one elastic phase concentration
tensor is sufficient for describing the full thermoelastic behavior of two-phase inhomoge-
neous materials within the mean-field framework. It is worth mentioning that a fair number
of additional relations between phase averaged tensors have been reported in the literature.

An additional set of useful concentration tensors are the partial strain and stress con-
centration tensors, T̄(p,q) and W̄(p,q) , defined by

ε(p) = T̄(p,q) ε(q)

σ (p) = W̄(p,q) σ (q) , (2.17)

which connect the fields in phases (p) and (q) , compare, e.g., Dvorak et al. (1992). For
two-phase composites with matrix–inclusion topology the partial concentration tensors
T̄(i,m) and W̄(i,m) are linked to the “standard” phase averaged concentration tensors by the
relations
 −1
Ā(m) = ξ T̄(i,m) + (1 − ξ)I
 −1
B̄(m) = ξ W̄(i,m) + (1 − ξ)I
 −1
Ā(i) = T̄(i,m) ξ T̄(i,m) + (1 − ξ)I
 −1
B̄(i) = W̄(i,m) ξ W̄(i,m) + (1 − ξ)I ; (2.18)

The inverse of the partial inhomogeneity stress concentration tensor [W̄(i,m) ]−1 = W̄m,i
was referred to as the “bridging tensor” by Huang (2000).

Equations (2.3) to (2.18) do not account for temperature dependence of the thermoe-
lastic moduli. For a mean-field framework capable of handling temperature dependent
moduli for finite temperature excursions and small strains see, e.g., Boussaa (2011).

Alternatively to using concentration tensors most analytical micromechanical mod-

els may be formulated in terms of contribution tensors, which were introduced by Horii
and Nemat-Nasser (1985), elaborated by Kachanov et al. (1994) for porous materials, ex-
tended to composite materials by Sevostianov and Kachanov (1999) as well as Eroshkin
and Tsukrov (2005) and further modified by Sevostianov and Kachanov (2011). Contribu-
tion tensors are defined via the differences between the effective compliance and elasticity

19
 (i)
tensors caused by the presence of inhomogeneities and the corresponding elastic tensors
of the reinforcement-free matrix, (m) , as
1 ∗
H(i) = (C − C(m) )
ξ
1 ∗
N(i) = (E − E(m) ) . (2.19)
ξ

The compliance contribution tensors H(i) and elasticity contribution tensor N(i) can be
shown to be connected to the phase averaged strain and stress concentration tensors, Ā(i)
and B̄(i) , by the relations

H(i) = (C(i) − C(m) )B̄(i)

N(i) = (E(i) − E(m) )Ā(i) (2.20)

and they are linked to each other via the expressions

−1
H(i) = −C(m) N(i) E(m) + ξN(i)

−1
N(i) = −E(m) H(i) C(m) + ξH(i) (2.21)

In contrast to concentration tensors the contribution tensors H(i) and N(i) are symmetric,
as is evident from eqns. (2.19).

2.2 Eshelby Tensor and Dilute Matrix–Inclusion Com-

posites
Eshelby’s Eigenstrain Problem
A large percentage of the mean-field descriptions used in continuum micromechanics of
materials have been based on the work of Eshelby, who initially studied the stress and
strain distributions in homogeneous media that contain a subregion (“inclusion”) that
spontaneously changes its shape and/or size (“undergoes a transformation”) so that it
no longer fits into its previous space in the “parent medium”. Eshelby’s results show
that if an elastic homogeneous ellipsoidal inclusion (i.e., an inclusion consisting of the
same material as the matrix) in an infinite matrix is subjected to a homogeneous strain εt
(called the “stress-free strain”, “unconstrained strain”, “polarization strain”, “eigenstrain”,
or “transformation strain”), the stress and strain states in the constrained inclusion are
uniform17 , i.e., σ (i) = hσi(i) and ε(i) = hεi(i) . The uniform strain in the constrained
inclusion (the “constrained strain”), ε(i) = εc , is related to the stress-free strain εt by the
expression (Eshelby, 1957)
εc = Sεt , (2.22)
where S is referred to as the (interior-point) Eshelby tensor. For eqn. (2.22) to hold, εt
may be any kind of eigenstrain that is uniform over the inclusion (e.g., a thermal strain or
17
This “Eshelby property” or “Eshelby uniformity” is limited to inclusions and inhomogeneities of ellip-
soidal shape (Lubarda and Markenscoff, 1998; Kang and Milton, 2008). However, hyperboloidal domains
will (Franciosi, 2020) and certain non-dilute periodic arrangements of non-ellipsoidal inclusions can (Liu
et al., 2007) also give rise to homogeneous fields.

20
a strain due to some phase transformation involving no changes in the elastic constants of
the inclusion).

The Eshelby tensor S depends only on the material properties of the parent medium
(in many cases the matrix) and on the specific shape of the ellipsoidal inclusions. Eshelby
tensors for inclusions of general shape can be obtained as
Z
(i,0)
S (x) = b (0) (x − x′ ) E(0) dΩ′ = P(i,0) (x)E(0)
G , (2.23)
Ω(i)

compare, e.g., Kachanov and Sevostianov (2018). Here G b (0) and E(0) stand for the modi-
fied Green’s tensor and the elasticity tensor of the parent medium (0) , respectively. P(i,0)
is known as the Hill tensor, mean polarization factor tensor or shape tensor. Here, the
superscripts in S(i,0) (x) and P(i,0) (x) explicitly denote evaluation with respect to the parent
medium (0) and the shape of inclusion (i) . The so-called first Eshelby problem consists in
finding solutions for the above integral.

Closed form expressions for the Eshelby tensor of spheroidal inclusions in an isotropic
matrix are available as functions of the aspect ratio (eccentricity) a, see, e.g., Pedersen
(1983), Tandon and Weng (1984), Mura (1987) or Clyne and Withers (1993), the formu-
lae resulting for continuous fibers of circular cross-section (a → ∞), spherical inclusions
(a = 1), and thin circular disks or layers (a → 0) being rather simple. Analogous expres-
sions for the Eshelby tensor have also been reported for spheroidal inclusions in a matrix
of transversally isotropic material symmetry (Withers, 1989; Parnell and Calvo-Jurado,
2015), provided the material axes of the matrix are aligned with the orientations of non-
spherical inclusions. Expressions for the Eshelby tensors of general ellipsoidal inclusions
in an isotropic matrix in terms of incomplete elliptic integrals (which, in general, must
be evaluated numerically) can be found, e.g., in Mura (1987). Instead of evaluating the
Eshelby tensor for a given configuration, the Hill tensor, P(i,m) = SC(m) (where S denotes
S(i,m) ), may be solved for instead, compare, e.g., Masson (2008) or Barthélémy (2020).
For detailed discussion of Eshelby and Hill tensors pertaining to ellipsoidal inclusions see
Parnell (2016); information on position-dependent and averaged Eshelby tensors for poly-
hedral inhomogeneities embedded in an isotropic matrix is provided, e.g., by Kachanov
and Sevostianov (2018).

For materials of low elastic symmetry the modified Green’s function tensor G b (0) as
required in eqn. (2.23) is not available. However, the problem can be transformed into
a surface integral in Fourier space (Mura, 1987), where the modified Green’s tensors are
known explicitly for all material symmetries. Approximations to the Eshelby tensors of
ellipsoids embedded in matrices of general elastic symmetry can then be obtained by using
numerical quadrature for the back transformation, compare Gavazzi and Lagoudas (1990).

Eshelby’s Inhomogeneity Problem

For mean-field descriptions of dilute matrix–inclusion composites, the main interest lies
on the stress and strain fields in inhomogeneous inclusions (“inhomogeneities”) that are
embedded in a matrix. The case where the inhomogeneities do not interact with each other
can be handled on the basis of Eshelby’s theory for homogeneous inclusions, eqn. (2.22),

21
by introducing the concept of an equivalent homogeneous inclusion. This approach in-
volves replacing an actual perfectly bonded inhomogeneity, which has different material
properties than the matrix and which is subjected to a given unconstrained eigenstrain εt ,
with a (fictitious) “equivalent” homogeneous inclusion on which a (fictitious) “equivalent”
eigenstrain ετ is made to act. This equivalent eigenstrain is chosen in such a way that the
inhomogeneous inclusion and the equivalent homogeneous inclusion attain the same stress
state σ (i) as well as the same constrained strain εc (Eshelby, 1957; Withers et al., 1989).
When σ (i) is expressed in terms of the elastic strain in the inhomogeneity or inclusion, this
condition translates into the equality

σ (i) = E(i) [εc − εt ] = E(m) [εc − ετ ] . (2.24)

Here εc − εt and εc − ετ are the elastic strains in the inhomogeneous inclusion and the
equivalent homogeneous inclusion, respectively. Obviously, in the general case the stress-
free strains will be different for the equivalent inclusion and the real inhomogeneity, εt 6= ετ .
Plugging the result of applying eqn. (2.22) to the equivalent eigenstrain, εc = Sετ , into
eqn. (2.24) leads to the relationship

σ (i) = E(i) [Sετ − εt ] = E(m) [S − I]ετ , (2.25)

which can be rearranged to obtain the equivalent eigenstrain as a function of the known
stress-free eigenstrain εt of the inhomogeneous inclusion as

ετ = [(E(i) − E(m) )S + E(m) ]−1 E(i) εt . (2.26)

This, in turn, allows the stress in the inhomogeneity, σ (i) , to be expressed explicitly as

σ (i) = E(m) (S − I)[(E(i) − E(m) )S + E(m) ]−1 E(i) εt . (2.27)

The concept of the equivalent homogeneous inclusion can be extended to the “inho-
mogeneity problem”, where a uniform mechanical strain εa or stress σ a is applied to a
system consisting of a perfectly bonded inhomogeneous elastic inclusion in an infinite ma-
trix. Here, the strain in the inhomogeneity, ε(i) , is a superposition of the applied strain
and of a term εc that accounts for the constraint effects of the surrounding matrix. Well-
known expressions for dilute, non-interacting elastic concentration tensors based on the
above considerations were proposed by Hill (1965b) and elaborated by Benveniste (1987).
For deriving them, the conditions of equal stresses and strains in the actual inhomogeneity
(elasticity tensor E(i) ) and the equivalent inclusion (elasticity tensor E(m) ) under an applied
far-field strain εa take the form

σ (i) = E(i) [εa + εc ] = E(m) [εa + εc − ετ ] (2.28)

and
ε(i) = εa + εc = εa + Sετ , (2.29)
respectively, where eqn. (2.22) is used to describe the constrained strain of the equivalent
homogeneous inclusion. On the basis of these relationships the strain in the inhomogeneity
can be expressed as
ε(i) = [I + SC(m) (E(i) − E(m) )]−1 εa . (2.30)

22
Because the strain in the inhomogeneity is homogeneous, ε(i) = hεi(i) , the strain concen-
tration tensor for dilute inhomogeneities follows directly from eqn. (2.30) as
(i)
Ādil = [I + SC(m) (E(i) − E(m) )]−1 . (2.31)

By using hεi(i) = C(i) hσi(i) as well as εa ≅ C(m) σ a , the dilute stress concentration tensor
for the inhomogeneities is found from eqn. (2.30) as
(i)
B̄dil = E(i) [I + SC(m) (E(i) − E(m) )]−1 C(m)
= [I + E(m) (I − S)(C(i) − C(m) )]−1 . (2.32)

For the dilute case the “standard” and partial inhomogeneity concentration tensors coin-
cide, i.e.,
(i,m) (i) (i,m) (i)
T̄dil = Ādil and W̄dil = B̄dil , (2.33)
(p)
so that the dilute partial concentration tensors for an inhomogeneity embedded in a
medium (q) result as
(p,q)
T̄dil = [I + S(p,q) C(q) (E(p) − E(q) )]−1 = [I + P(p,q) (E(p) − E(q) )]−1
(p,q)
W̄dil = [I + E(q) (I − S(p,q) )(C(p) − C(q) )]−1 . (2.34)

The dilute elasticity and compliance contribution tensors can be obtained from eqns. (2.31)
and (2.32) via eqns. (2.19) as
(i) (i)  −1
Ndil = (E(i) − E(m) )Ādil = (E(i) − E(m) )−1 + SC(m)
(i) (i)  −1
Hdil = (C(i) − C(m) )B̄dil = (C(i) − C(m) )−1 + E(m) (I − S) . (2.35)

It is worth noting that in eqns. (2.24) to (2.35) the Eshelby tensor always refers to the
equivalent homogeneous inclusion and is independent of the material symmetry and elastic
properties of the actual inhomogeneities.

Dilute thermal concentration tensors can be generated from eqns. (2.31) and (2.32)
by using eqns. (2.7) and (2.16). Alternative, but fully equivalent expressions for dilute
mechanical and thermal inhomogeneity concentration tensors were given, e.g., by Mura
(1987), Wakashima et al. (1988) and Clyne and Withers (1993). All of the above relations
were derived under the assumption that the inhomogeneities are dilutely dispersed in the
matrix and thus do not “feel” any effects of their neighbors (i.e., they are loaded by the
unperturbed applied stress σ a or applied strain εa , the so-called dilute or non-interacting
case). Accordingly, the resulting inhomogeneity concentration tensors are independent of
the reinforcement volume fraction ξ. The corresponding matrix concentration tensors can
be obtained via eqns. (2.7) and, accordingly, depend on ξ.

Following, e.g., Dederichs and Zeller (1973) or Kachanov and Sevostianov (2018) the
strain field in an inhomogeneity of general shape can be given as
Z
(i) (0) (i)
ε (x) = hεi + (E − E ) (m) b (0) (x − x′ ) ε(x′ ) dΩ′
G . (2.36)
V (i)

23
Solutions of eqn. (2.23), i.e., Hill or Eshelby tensors, in general hold for eqn. (2.36) only
if ε is constant within the inhomogeneity, a condition that is fulfilled solely for ellipsoids.
The integral equation (2.36) is a Lippmann–Schwinger equation and describes the so-called
second Eshelby problem18 .

The stress and strain fields outside an inhomogeneity or a transformed homogeneous

inclusion in an infinite matrix are not uniform on the microscale19 (Eshelby, 1959). Within
the framework of mean-field approaches, which aim to link the average fields in matrix
and inhomogeneities with the overall response of inhomogeneous materials, however, it is
only the average matrix stresses and strains that are of interest. For dilute composites,
such expressions follow directly by combining eqns. (2.31) and (2.32) with eqn. (2.7), with
(m) (m)
the dilute matrix concentration tensors fulfilling Ādil → I and B̄dil → I. Expressions for
the overall elastic and thermal expansion tensors can be obtained in a straightforward way
(i) (i)
from the dilute concentration tensors Ādil and B̄dil by using eqns. (2.8) to (2.16), typical
results being
(i) (i)
E∗NI = E(m) + ξ(E(i) − E(m) )Ādil = E(m) + ξNdil
(i) (i)
C∗NI = C(m) + ξ(C(i) − C(m) )B̄dil = C(m) + ξHdil . (2.37)

Such models are often referred to as Non-Interacting Approximations (NIA). It must be

kept in mind that all mean-field expressions of this type are strictly valid only for vanish-
ingly small inhomogeneity volume fractions20 .

2.3 Mean-Field Methods for Two-Phase Thermoelas-

tic Composites with Aligned Reinforcements
Models for the overall thermoelastic behavior of composites with reinforcement volume
fractions of more than a few percent must explicitly account for interactions between inho-
mogeneities, i.e., for the effects of all surrounding reinforcements on the stress and strain
fields experienced by a given fiber or particle. Within the mean-field framework such in-
teraction effects as well as the concomitant perturbations of the stress and strain fields
in the matrix are typically accounted for in a collective way via approximations that are
phase-wise constant. Beyond such “background” effects, at non-trivial volume fractions
interactions between individual reinforcements give rise to, on the one hand, inhomoge-
neous stress and strain fields within each inhomogeneity as well as in the matrix. Such
18
Eshelby tensors in the strict sense are solutions to the integral, eqn. (2.23). Tensors obtained by solving
the integral equation (2.36) and then using eqn. (2.30) to extract S coincide with the proper Eshelby tensor
only if the inhomogeneity is of ellipsoidal shape.
It may be noted that Eshelby (1957) did not proceed via eqns. (2.23) and (2.36).
19
The fields outside a single inclusion can be described via the exterior-point Eshelby tensor, see, e.g.,
Mura (1987), Ju and Sun (1999), Meng et al. (2012) or Jin et al. (2016). From the (constant) interior-
point fields and the (position dependent) exterior-point fields the stress and strain jumps at the interface
between inclusion and matrix can be evaluated.
20
With growing inhomogeneity volume fraction the predictions of dilute approximations deviate increas-
ingly from solutions accounting for interactions between inhomogeneities and phase arrangement effects
become more and more pronounced. The detailed behavior depends on the elastic contrast between the
phases; as a rule of thumb, reasonably dependable results from non-interacting models cannot be expected
for ξ / 0.1.

24
intra-particle, intra-fiber and intra-matrix fluctuations cannot be resolved by mean-field
approaches. On the other hand, the interactions cause the levels of the average stresses
and strains in individual inhomogeneities to differ, i.e., inter-particle and inter-fiber fluc-
tuations are present. These are not resolved by most mean-field-type methods, either; for
an exception see, e.g., El Mouden and Molinari (2000).

Two main strategies are available for handling non-dilute reinforcement volume frac-
tions within mean-field methods in continuum micromechanics: Either non-interacting
inhomogeneities are subjected to suitably modified stress or strain fields (rather than to
the far fields) or they are embedded into a suitably selected, fictitious medium (rather than
into the matrix). This way the Eshelby-based apparatus for non-interacting reinforcements
discussed in section 2.2 can be directly leveraged into the non-dilute regime. The above two
strategies are commonly referred to as effective-field and as effective-medium approaches,
respectively. Figure 2.1 schematically compares the material and loading configurations
underlying non-interacting (NIA or “dilute Eshelby”) models, an effective-field scheme and
two effective-medium approaches.

i i i i
m
m m eff eff
σa <σ> (m)
σa σa
NIA MTM CSCS, DS GSCS
Figure 2.1: Schematic comparison of the Eshelby method in the non-interacting approximation
(NIA), Mori–Tanaka approaches (MTM), differential schemes (DS) and classical (CSCS) as well
as generalized self-consistent (GSCS) schemes.

2.3.1 Effective-Field Approaches

One way of introducing collective interactions between inhomogeneities into mean-field
models consists of approximating the stresses acting on an inhomogeneity, which may be
viewed as perturbation stresses caused by the presence of other inhomogeneities (“image
stresses”, “background stresses”) superimposed on the applied far-field stress, by a suit-
able uniform effective stress, σ E . On the basis of analogous considerations the far-field
strain may be replaced by an effective strain εE . Following this concept of effective fields,
which was introduced by Mossotti (1850), the the phase averaged strain and stress in an
inhomogeneity are described as
(i) (i) (i)
hεi(i) = Ādil εE = Ādil Aeff hεi
(i) (i) (i)
hσi(i) = B̄dil σ E = B̄dil Beff hσi (2.38)

(Sevostianov and Kachanov, 2014). Here the inhomogeneity effective strain and stress
(i) (i)
concentration tensors, Aeff and Beff link the effective and the macroscopic fields, providing
corrections for collective interactions. What remains, of course, is the task of finding
suitable expressions for the effective fields, εE and/or σ E .

25
Mori–Tanaka-Type Estimates
The idea of choosing the matrix fields as the effective fields goes back to Brown and Stobbs
(1971) as well as Mori and Tanaka (1973); such effective-field approaches are generically
referred to as Mori–Tanaka methods. Benveniste (1987) based his treatment of Mori–
Tanaka approaches on the relationships
(i,m) (i)
hεi(i) = T̄dil hεi(m) = Ādil hεi(m)
(i,m) (i)
hσi(i) = W̄dil hεi(m) = B̄dil hσi(m) (2.39)

or, equivalently,
(i)
Ā(i) = Ādil Ā(m)
(i)
B̄(i) = B̄dil B̄(m) , (2.40)

which literally transcribe the above assumptions, compare also Fig. 2.1. Consistent ex-
pressions for hεi(m) and/or hσi(m) can be obtained by inserting eqn. (2.39) into eqn. (2.5),
leading to the relations
(i)
hεi(m) = [(1 − ξ)I + ξ Ādil ]−1 hεi
(i)
hσi(m) = [(1 − ξ)I + ξ B̄dil ]−1 hσi . (2.41)

These, in turn, allow the Mori–Tanaka strain and stress concentration tensors for matrix
and inhomogeneities to be written in terms of the dilute concentration tensors as21
(m) (i) (i) (i) (i)
ĀMT = [(1 − ξ)I + ξ Ādil ]−1 ĀMT = Ādil [(1 − ξ)I + ξ Ādil ]−1
(m) (i) (i) (i) (i)
B̄MT = [(1 − ξ)I + ξ B̄dil ]−1 B̄MT = B̄dil [(1 − ξ)I + ξ B̄dil ]−1 (2.42)

(Benveniste, 1987). Equations (2.42) may be evaluated with any strain and stress concen-
(i) (i)
tration tensors Ādil and B̄dil pertaining to dilute inhomogeneities embedded in a matrix. If,
for example, the equivalent inclusion expressions, eqns. (2.31) and (2.32), are employed, the
Mori–Tanaka matrix strain and stress concentration tensors for the non-dilute composite
take the form
(m)
ĀMT = {(1 − ξ)I + ξ[I + SC(m) (E(i) − E(m) )]−1 }−1
(m)
B̄MT = {(1 − ξ)I + ξE(i) [I + SC(m) (E(i) − E(m) )]−1 C(m) }−1 , (2.43)

compare Benveniste (1987), Benveniste and Dvorak (1990) or Benveniste et al. (1991).
The Mori–Tanaka approximations to the effective elastic tensors are obtained by combin-
ing eqns. (2.42) and/or (2.43) with eqns. (2.8) and (2.9). By inserting eqns. (2.42) into
eqns. (2.8) the Mori–Tanaka estimates for the effective elastic tensors are obtained as
(i) (i)
E∗MT = [ξE(i) Ādil + (1 − ξ)E(m) ][ξ Ādil + I]−1
(i) (i)
C∗MT = [ξC(i) B̄dil + (1 − ξ)C(m) ][ξ B̄dil + I]−1 . (2.44)

A number of authors gave different but essentially equivalent Mori–Tanaka-type ex-

pressions for the phase concentration tensors and effective thermoelastic tensors of inho-
mogeneous materials, among them Pedersen (1983), Wakashima et al. (1988), Taya et al.
21
Equations (2.18) and (2.42) are identical because they essentially describe the same physics.

26
(1991), Pedersen and Withers (1992) as well as Clyne and Withers (1993). Alternatively, a
Mori–Tanaka method can be formulated to directly give the macroscopic elasticity tensor
as

−1
E∗T = E(m) I − ξ[(E(i) − E(m) ) S − ξ(S − I) + E(m) ]−1 [E(i) − E(m) ] (2.45)

(Tandon and Weng, 1984). Because eqn. (2.45) does not explicitly use the compliance
tensor of the inhomogeneities, C(i) , it can be modified in a straightforward way to describe
the macroscopic stiffness of porous materials by setting E(i) → 0, producing the relationship
 ξ −1
EMT,por = E(m) I + (I − S)−1 . (2.46)
1−ξ
which, however, should not be used for void volume fractions that are much in excess of,
say, ξ = 0.2522 .
As is evident from their derivation, Mori–Tanaka-type theories at all volume fractions
describe composites consisting of randomly positioned, aligned, ellipsoidal inhomogeneities
embedded in a matrix, i.e., inhomogeneous materials with a distinct matrix–inclusion mi-
crotopology. More specifically, it was shown by Ponte Castañeda and Willis (1995) that
Mori–Tanaka methods are a special case of Hashin–Shtrikman variational estimates, com-
pare Section 2.4, in which the spatial arrangement of the inhomogeneities follows an aligned
ellipsoidal distribution that is characterized by the same aspect ratio as the shape of the
inhomogeneities themselves (Hu and Weng, 2000), compare fig. 2.2.

Figure 2.2: Sketch of ellipsoidal inhomogeneities in an aligned ellipsoidally distributed spatial

arrangement as used implicitly in Mori–Tanaka-type models (a=2.0).

For many two-phase composites Mori–Tanaka estimates coincide with one of the Hashin–
Shtrikman-type bounds as formulated in Hashin and Shtrikman (1961) and Hashin and
22
Standard Mori–Tanaka theories use the small strain assumption, the shapes of inhomogeneities being
described by ellipsoids that maintain their aspect ratios throughout the deformation history. In porous or
cellular materials with high void volume fractions the linear elastic range tends to be very limited, followed
by bending and buckling of cell walls or struts (Gibson and Ashby, 1988), which implies changes to the
shapes of the voids. Such effects cannot be captured by Mori–Tanaka models, which consequently may
markedly overestimate the effective stiffness of such materials. Within the above constraints Mori–Tanaka
methods can account for some effects of fluids contained in the pores, compare, e.g., Kitazono et al. (2003).
It is worth noting that the predictions of different mean field models for the the ratio EA /ET , and thus
for the macroscopic elastic anisotropy, of porous materials tend to vary considerably.

27
Shtrikman (1963)23 . For composites reinforced by spherical or aligned spheroidal reinforce-
ments that are stiffer than the matrix, MT predictions for the overall elastic moduli are,
accordingly, on the low side (see the comparisons in Section 4.1 as well as tables 6.1 to 6.3).
For materials containing compliant reinforcements in a stiffer matrix, in contrast, Mori–
Tanaka models provide upper estimates. In situations of high elastic contrast Mori–Tanaka
approaches tend to considerably under- or overestimate the effective elastic properties. For
discussions of further issues with respect to the range of validity of Mori–Tanaka theories
for elastic inhomogeneous two-phase materials see Christensen et al. (1992).

Mori–Tanaka-type theories can be implemented into computer programs in a straight-

forward way: Because they are explicit algorithms, all that is required are matrix addi-
tions, multiplications, and inversions plus expressions for the Eshelby tensor. Despite their
limitations, Mori–Tanaka models provide useful accuracy for the elastic contrasts pertain-
ing to most practically relevant composites. This combination of features makes them
important tools for evaluating the stiffness and thermal expansion properties of inhomo-
geneous materials that show a matrix–inclusion topology with aligned inhomogeneities.
“Extended” Mori–Tanaka models for nonaligned reinforcements are discussed in Section
2.6, multi-phase Mori–Tanaka schemes in Section 2.8 and Mori–Tanaka-type approaches
to describing thermoelastoplastic materials in Section 2.9,

Maxwell Schemes
The first “micromechanical” model to appear in the literature, the scheme of Maxwell
(1873) for evaluating the effective conductivities of inhomogeneous materials, can be inter-
preted in terms of a mean-field method in elasticity, see Torquato (2002) and Sevostianov
and Giraud (2013). It is based on studying a single region of inhomogeneous material
that contains a sufficient number of inhomogeneities to be treated as an RVE and that is
embedded in a much larger matrix region, compare fig. 2.3. This arrangement is described,
on the one hand, by a sub-model incorporating the above set of inhomogeneities (“inhomo-
geneous region”, left side of fig. 2.3), which are treated as non-interacting via eqns. (2.31)
or (2.32), and, on the other hand, by a sub-model in which the region consists of an (un-
known) uniform effective material (“effective inhomogeneity”, right side of fig. 2.3), the
overall elastic responses of the two configurations being required to be equal. Analogous
approaches were used, e.g., by Giordano (2003) and, independently of Maxwell’s ideas, by
Shen and Yi (2001).

Using eqns. (2.8) and (2.9), summing up over the contributions of the individual inho-
mogeneities (q) in the arrangement on the left, and employing dilute Eshelby expressions
for both configurations leads to the relations
X (q) (∗,m)
E(m) + ξ (q) (E(q) − E(m) )Ādil = E(m) + ξE (E∗ − E(m) )T̄dil
(q)6=(m)
X (q) (∗,m)
C(m) + ξ (q) (C(q) − C(m) )B̄dil = C(m) + ξE (C∗ − C(m) )W̄dil , (2.47)
(q)6=(m)

23
It should be noted, though, that Mori–Tanaka estimates correspond to Hashin–Shtrikman-type bounds
only if the constituents of the composite are elastically well-ordered, compare Section 3.1.

28
 Figure 2.3: Sketch of the two configurations underlying Maxwell (1873) schemes.

where ξE and ξ (q) are the volume fractions of the inhomogeneous region and of the indi-
vidual inhomogeneities, respectively. The superscript (∗,m) indicates that the dilute partial
concentration tensors, see eqn. (2.34), pertain to a region of effective material embedded in
the matrix. If the shapes and material behaviors of all inhomogeneities (q) are
P identical they
(q)
can be viewed as forming a reinforcement phase of volume fraction ξ = (q)6=(m) ξ /ξE .
The two-phase Maxwell estimates for the effective elastic tensors, E∗MX and C∗MX , can then
be extracted from eqn. (2.47) as
h i−1
E∗MX = E(m) + ξ (E(i) − E(m) )−1 + (S(i,m) − ξSE )C(m)
h
i−1
C∗MX = C(m) + ξ (C(i) − C(m) )−1 + E(m) (1 − ξ)I − S(i,m) + ξSE . (2.48)

Here SE is an Eshelby tensor pertaining to the shape of the inhomogeneous region and
S(i,m) = S is the “standard” Eshelby tensor depending on the shape of the inhomogeneities.
Approaches in which the behavior of the inhomogeneous region is described by some in-
teracting model rather than the dilute Eshelby one have been referred to as generalized
Maxwell schemes, see, e.g., Kanaun (2016). For recent discussions of Maxwell models and
their extensions see Sevostianov et al. (2019) and Buryachenko (2022b).

Equations (2.48) can be used to obtain expressions for the inhomogeneity fields that are
of the same form as eqns. (2.38), making Maxwell schemes proper effective-field models.
If the inhomogeneous region is chosen to have the same shape as the (aligned) inhomo-
geneities, so that SE = S(i,m) , the two-phase Maxwell scheme, eqn. (2.48), coincides with
the Mori–Tanaka expressions, eqn. (2.44).

Hashin–Shtrikman Estimates of Ponte and Willis

Ponte Castañeda and Willis (1995) developed a variationally-based effective-field scheme
for inhomogeneous materials that consist of ellipsoidal arrangements of ellipsoidal inho-
mogeneities embedded in a matrix. In such materials the spatial correlations of the in-
homogeneity arrangement can be described by an Eshelby tensor Sd and the shapes of
the inhomogeneities via the Eshelby tensor S(i,m) (which corresponds to S as used in the
preceding sections), both of which are evaluated using the matrix as the embedding ma-
terial. The corresponding phase arrangements can be visualized by interpreting fig. 2.2 as
consisting of identical, aligned, non-overlapping “safety ellipsoids” (exclusion volumes de-
scribing the ellipsoidal arrangement of reinforcements) that contain (aligned or nonaligned)

29
ellipsoidal inhomogeneities, the aspect ratios of the safety ellipsoids. Using the concept of
Hashin–Shtrikman estimates, compare section 2.4, the inhomogeneity strain concentration
tensors of such microgeometries can be expressed as
(i)  −1
ĀPW = I + (S(i,m) − ξSd C(m) (E(i) − E(m) )
(i)  (i)
= Ādil I − ξSd C(m) (E(i) − E(m) )Ādil ]−1 , (2.49)

which fits the effective-field mold. The macroscopic elasticity tensor follows as
 −1
E∗PW = E(m) + ξ(E(i) − E(m) ) I + (S(i,m) − ξSd )C(m) (E(i) − E(m) )
 (i)
−1 −1
= E(m) + ξ(E(i) − E(m) )Ādil − Sd C(m) (2.50)

via eqn. (2.8). This model is known as the Hashin–Shtrikman estimates of Ponte Castañeda
and Willis (1995). It is strictly valid as long as the inhomogeneities do not penetrate their
respective safety ellipsoids. When the safety ellipsoids and the inhomogeneities are aligned
and identical, Sd = S(i,m) , eqn. (2.50) recovers Mori–Tanaka expressions, which shows that
Mori–Tanaka methods are a special case of Hashin–Shtrikman estimates.
Analogous expressions to eqn. (2.50) can be obtained with two-phase Maxwell models,
compare eqns. (2.48), and the Interaction Direct Derivative (IDD) approach of Du and
Zheng (2002). A further related model is the double inclusion model of Hori and Nemat-
Nasser (1993), the connections of which to the Ponte–Willis estimates and a number of
other micromechanical schemes were discussed by Hu and Weng (2000). All these models
provide more flexibility in terms of the arrangement of inhomogeneities than do Mori–
Tanaka methods.

Advanced Effective-Field Methods

Equations (2.38) obviously allow incorporating more complex approximations to the effec-
tive fields than covered up to this point, see, e.g., Kanaun and Levin (1994), Rodin and
Weng (2014) or Saadat et al. (2015). For composites showing a matrix–inclusion topology,
in the most general case the effective fields may be evaluated from regions containing a
number of interacting inhomogeneities (a “cloud of inhomogeneities”), which gives rise to
the Multi-Particle Effective-Field Method (MEFM) discussed in depth by Buryachenko
(2007). Such models are formulated in terms of integral equations, are very powerful and
subject to few restrictions, but are rather complex. They subsume the simpler effective-field
models such as Mori–Tanaka methods as special cases.

2.3.2 Effective-Medium Approaches

The second group of mean-field estimates for the overall thermomechanical moduli of inho-
mogeneous materials are effective-medium approaches, in which an inhomogeneity or some
phase arrangement (often referred to as a kernel) is embedded in the effective material, the
properties of which are not known a priori. Figure 2.1 sketches two models of this type,
the classical self-consistent and generalized self-consistent schemes24 .
24
Within the classification of micromechanical methods given in Section 1.5, self-consistent mean-field
schemes may also be viewed as analytical embedding approaches.

30
Classical Self-Consistent Estimates
If the kernel consists of just an inhomogeneity, classical (or two-phase) self-consistent
schemes (CSCS) are obtained, see, e.g., Hill (1965b). They are based on treating both
phases, (p) and (q) with volume fractions ξ (p) and ξ (q) , as inhomogeneities embedded in
an a priori unknown, homogeneous reference medium 0 and requiring the mean-field ap-
proximations to the stress and strain perturbations to vanish, which translates into the
conditions
(p,0) (q,0)
ξ (p) (E(p) − E0 )T̄dil + ξ (q) (E(q) − E0 )T̄dil = 0
(p) (p) 0 (p,0) (q) (q) 0 (q,0)
ξ (C −C )W̄dil +ξ (C −C )W̄dil = 0 , (2.51)
Accordingly — in contrast to effective-field approximations — no distinction is made be-
tween contiguous (matrix) and non-contiguous (inhomogeneity) phases. Identifying the
reference medium with the effective material and applying eqns. (2.8) the relationships
(p,∗)
E∗SC = E(q) + ξ (p) [E(p) − E(q) ] T̄dil
= E(q) + ξ (p) [E(p) − E(q) ] [I + S(p,∗) C∗SC (E(p) − E∗SC )]−1
(p,∗)
C∗SC = C(q) + ξ (p) [C(p) − C(q) ] W̄dil
= C(q) + ξ (p) [C(p) − C(q) ] [I + E∗SC (I − S(p,∗) )(C(p) − C∗SC )]−1 (2.52)
are obtained, which do not obviously show the symmetry in terms of the phases evident in
(p,∗) (p,∗)
eqns. (2.51). In eqns. (2.52) S(p,∗) , T̄dil and W̄dil stand for the Eshelby tensor and dilute
partial concentration tensors, compare eqn. (2.34), of an inhomogeneity (p) embedded in
the (unknown) effective material. Equations (2.52) provide systems of implicit nonlinear
equations for the unknown elastic tensors, E∗SC and C∗SC , describing the behavior of the
effective medium. These systems can be solved by self-consistent iterative schemes of the
type
ESC,n+1 = E(q) + ξ (p) [E(p) − E(q) ][I + Sn(p,∗) Cn (E(p) − En )]−1
 −1
CSC,n+1 = ESC,n+1 . (2.53)
(p,∗)
The Eshelby tensor Sn in eqn. (2.53) pertains to an inhomogeneity embedded in the
n-th iteration for the effective medium; it must be recomputed for each iteration25 .

The predictions of the CSCS for two-phase materials differ clearly from the ones ob-
tained with Mori–Tanaka methods, tending to be close to the Hashin–Shtrikman lower
bounds (see Section 3) if the volume fraction of the more compliant phase is high and
close to the upper bounds it it is low (compare figs. 4.1 to 4.9). Generally, two-phase
self-consistent schemes are well suited to describing the overall properties of two-phase
materials that do not show a matrix–inclusion microtopology for some or all of the vol-
ume fractions of interest26 and are, accordingly, not the method of choice for describing
25
For aligned spheroidal but non-spherical, isotropic inhomogeneities in an isotropic matrix the effective
medium shows transversally isotropic behavior, which must be accounted for in evaluating the Eshelby
(p,∗)
tensor Sn .
26
Classical self-consistent schemes have been shown to correspond to perfectly disordered materials
(Kröner, 1978) or self-similar hierarchical materials (Torquato, 2002). Because all phases in expressions
such as eqn. (2.86) are treated on an equal footing, the CSCS is sometimes referred to as a “symmetric”
scheme.

31
“standard” composite materials. The two-phase CSCS may, however, be used for study-
ing Functionally Graded Materials (FGMs), which tend to show an interpenetrating phase
topology for volume fractions around ξ (p) = ξ (q) = 12 , with phase (q) acting acting as the
matrix for ξ (p) → 0 and (p) for ξ (p) → 1. Multi-phase versions of the CSCS, see, e.g.,
eqn. (2.86), are important methods for modeling materials with grain-like microstructures,
such as polycrystals. For porous materials classical self-consistent schemes predict a break-
down of the stiffness due to percolation of the pores at ξ (p) = 12 for spherical voids and at
ξ (p) = 31 for aligned cylindrical voids (Torquato, 2002).

Because self-consistent schemes are by definition implicit methods, their computational

requirements are in general higher than those of Mori–Tanaka-type approaches. Like
effective-field models, they have formed the basis for describing the behavior of nonlin-
ear inhomogeneous materials.

Generalized Self-Consistent Estimates

Generalized self-consistent schemes (GSCS) are effective-medium theories that employ a
kernel made up of an inhomogeneity surrounded by an appropriate volume of matrix ma-
terial, compare fig. 2.1. When such configurations are embedded in a matrix and subjected
to mechanical or thermal loads the fields in them in general are not homogeneous, even in
the dilute case. However, the general relationships presented in Section 2.1 continue to hold.

In contrast to the other methods covered here, the GSCS proposed by Christensen and
Lo (1979), see also Christensen and Lo (1986), provides results for individual effective mod-
uli rather than elastic tensors. The results were obtained from the differential equations
describing the elastic responses of spherical or cylindrical three-phase regions, respectively,
under appropriate boundary and loading conditions, plus an energy equivalence closure.
This leads to cubic equations for the effective shear modulus G∗ in the case of spherical
particles or the transverse shear modulus G∗T in the case of aligned, continuous fibers. The
original scheme pertains to isotropic inhomogeneities, but predictions for composites re-
inforced by continuous, aligned, transversally isotropic fibers can be obtained as a special
case of models for composites reinforced by (multi-) coated fibers developed by Hervé and
Zaoui (1995), compare Section 2.8.2. The results for the bulk modulus, the axial Young’s
and shear moduli, the transverse bulk modulus as well as the coefficients of thermal expan-
sion coincide with the ones obtained from Mori–Tanaka approaches as well as composite
sphere or composite cylinder assemblages.

GSCS models of the type sketched in Fig. 2.1 pertain to materials with matrix–inclusion
microtopologies. Compared to Mori–Tanaka and Maxwell models the three-phase GSCS
provides improved predictions for the effective shear modulus or the effective transverse
shear modulus of typical composites. In three-dimensional cases it is available for spherical
or aligned, continuous reinforcements; semi-analytical solutions were also proposed for
randomly oriented spheroidal inhomogeneities (Riccardi and Montheillet, 1999). In 2D
solutions were reported for aligned elliptical inhomogeneities (Huang and Hu, 1995).

32
Differential Schemes
A further important type of effective-medium mean-field approach are differential schemes
(McLaughlin, 1977; Norris, 1985), which may be viewed as involving repeated cycles of
adding small concentrations of inhomogeneities to a material and then homogenizing. Fol-
lowing Hashin (1988) the resulting estimates for the overall elastic tensors can be described
by the coupled systems of differential equations
dE∗D 1 (i,∗)
= [E(i) − E∗D ] T̄dil
dξ 1−ξ
dC∗D 1 (i,∗)
= [C(i) − C∗D ] W̄dil (2.54)
dξ 1−ξ

with the initial conditions E∗D =E(m) and C∗D =C(m) , respectively, at ξ=0. By analogy to
(i,∗) (i,∗)
eqn. (2.52) W̄dil and T̄dil depend on the current approximations to the effective response,
E∗D and C∗D . Equations (2.54) can be conveniently integrated with standard numerical al-
gorithms for initial value problems, e.g., Runge–Kutta methods.

Differential schemes clearly pertain to matrix–inclusion microtopologies and have been

argued to describe polydisperse distributions of the sizes of the inhomogeneities27 . Models
of this group tend to show excellent agreement with numerical models at elevated inhomo-
geneity volume fractions. Nevertheless, probably due to their potential association with
specific microgeometries that are not necessarily typical of “classical composites” (in which
reinforcement size distributions usually are rather sharp28 ) and their higher mathematical
complexity (as compared, e.g., to Mori–Tanaka models), Differential Schemes have seen
surprisingly limited use in modeling the thermo-mechanical behavior of actual composite
materials.

2.4 General Considerations

An important concept in analytical micromechanics is that of the homogeneous reference
material (or comparison medium) (0) , which may be a constituent of a given composite or
some suitable fictitious material. It allows the stress field in an inhomogeneous material
to be expressed as
σ(x) = E(x) ε(x) = E0 ε(x) + τ 0 (x) , (2.55)
where E0 is the elasticity tensor of the comparison medium and τ 0 (x) = (E(x) − E0 ) ε(x)
is known as the stress polarization, phase averaging of which leads to

hτ 0 i(p) = (E(p) − E0 )hεi(p) = hσi(p) − E0 hεi(p) , (2.56)

27
The interpretation of differential schemes of involving the repeated addition of infinitesimal volume
fractions of inhomogeneities of increasingly larger size, followed each time by homogenization, is due to
Roscoe (1973). However, Hashin (1988) pointed out that the requirements of infinitesimal volume fraction
and growing size of the inhomogeneities may lead to contradictions. Avellaneda (1987) showed that
differential schemes can be realized by hierarchical laminates.
28
It is worth noting that, with the exception of very specific inhomogeneity shapes in two-dimensional
conduction (Torquato and Hyun, 2001), none of the mean-field methods discussed in Section 2.3 corre-
sponds to a realization in the form of a monodisperse matrix–inclusion microgeometry.

33
the phase averaged stress polarizations. Equations (2.55) can be used to develop integral
representations of micromechanics, and eqns. (2.56) form the basis of generalized mean-field
approaches and of many of the bounding methods discussed in Chapter 3.

Hashin–Shtrikman Tensor and Hashin–Shtrikman Estimates

Bornert (2001) pointed out that the use of phase-wise constant stress polarizations in
micromechanical models may be interpreted as emphasizing the effects of inter-phase het-
erogeneity rather than intra-phase fluctuations. This closely corresponds to the concepts
underlying mean-field models as used in the present chapter.

Equation (2.56) can be used as the starting point of variational procedures that pro-
vide strain concentration tensors in dependence on the elasticity tensor of a homogeneous
reference material as
(m)  −1
ĀHS = [L(i,0) + E(m) ]−1 ξ(L(i,0) + E(i) )−1 + (1 − ξ)(L(i,0) + E(m) )−1
(i)  −1
ĀHS = [L(i,0) + E(i) ]−1 ξ(L(i,0) + E(i) )−1 + (1 − ξ)(L(i,0) + E(m) )−1 . (2.57)

The tensor L(i,0) is referred to as the overall constraint tensor (Hill, 1965b) or Hill’s influence
tensor and is defined as

L(i,0) = E(0) [(S(i,0) )−1 − I] = [P(i,0) ]−1 − E(0) . (2.58)

Equations (2.57) lead to estimates for the overall elasticity tensors of two-phase materials
of the form
 −1
E∗HS = E(m) + ξ(E(i) − E(m) ) I + (1 − ξ)(L(i,0) + E(m) )−1 (E(i) − E(m) )
h i−1
= ξ(L(i,0) + E(i) )−1 + (1 − ξ)(L(i,0) + E(m) )−1 − L(i,0) , (2.59)

Alternative expressions involving Hill tensors can be given as

 
E∗HS = (1 − ξ)E(m) [I + P(m,0) (E(m) − E(0) )]−1 + ξE(i) [I + P(i,0) (E(i) − E(0) )]−1 ×
 −1
(1 − ξ)[I + P(m,0) (E(m) − E(0) )]−1 + ξ[I + P(i,0) (E(i) − E(0) )]−1
 (m,0) (i,0)  (m,0) (i,0) −1
= (1 − ξ)E(m) T̄dil + ξE(i) T̄dil (1 − ξ)T̄dil + ξ T̄dil , (2.60)

compare, e.g., Walpole (1966) or Willis (1977). Relations like eqns. (2.59) and (2.60),
which provide approximations of effective elastic tensors in terms of a reference medium,
have been called Hashin–Shtrikman tensors by Bornert (2001).

If this homogeneous reference material is chosen such that E(0) − E(p) is negative or
positive semi-definite for all phases (p) , Hashin–Shtrikman tensors give rise to lower and
upper Hashin–Shtrikman-type bounds, respectively, compare Section 3.1. Reference media
falling outside this range lead to estimates rather than bounds, which are often referred
to as Hashin–Shtrikman estimates for the special case of E(0) = E(m) . These encompass
Mori–Tanaka methods and, if ellipsoidal arrangements of inhomogeneities are explicitly ac-
counted for, the Ponte–Willis estimates, see Section 2.3.1. Classical self-consistent schemes
can be obtained by using the effective material as the comparison medium.

34
 Furthermore, Hashin–Shtrikman estimates can be extended by considering geometries
consisting of multi-phase “representative morphological patters” (also called “motifs”)
rather than uniform inhomogeneities embedded in a matrix, see the discussion by Bornert
(2001). For example, if the motif takes the form of a particle or fiber surrounded by a
matrix layer of appropriate thickness, the Generalized Self-Consistent scheme may be re-
covered. Adding further layers to such a motif leads to models for composites reinforced
by coated or multi-coated particles or fibers, compare Section 2.8.2.

Integral Equations
Very general analytical models for the elastic behavior of inhomogeneous materials in terms
of their microstructure and of the properties of their constituents can be given in the form
of integral equations. In the case of composites having a matrix–inclusion topology the
present understanding is that the most general descriptions of this type, which require the
lowest number of underlying hypotheses, are “General Integral Equations”, compare, e.g.,
Buryachenko (2015). All mean-field models discussed in Sections 2.3 and 2.8, including
the MEFM, can be obtained as approximations to the General Integral Equations, which
are discussed in detail in Buryachenko (2022a).

2.5 Other Analytical Estimates for Elastic Composites

with Aligned Reinforcements
For convenience, in this section some analytical methods are discussed that are of practical
importance but are not or only partly based on the mean-field assumptions. Nevertheless,
the relations between effective elastic tensors and phase averaged fields given in Section
2.1 apply to them, which allows, e.g., concentration tensors to be extracted.

Interpolative Schemes
Interpolative schemes obtain estimates for the effective elastic tensors by interpolating
between lower and upper estimates or lower and upper bounds. The simplest model of
this type are the Voigt–Reuss-Hill (VRH) estimates (Hill, 1952), which are given by the
arithmetic mean of the lower and upper Hill bounds, E∗H− and E∗H+ , defined by eqn. (3.1).
A more complex interpolative scheme was proposed by Lielens et al. (1998), in which the
(i,m)
(non-dilute) partial strain concentration tensor T̄L is obtained by interpolating between
dilute partial strain concentration tensors related to the Hashin–Shtrikman bounds as
h

−1 i−1
(i,m)
T̄L = (1 − f (ξ)) I + S(i,m) (C(m) E(i) − I) + f (ξ) I + S(m,i) (C(i) E(m) − I) (2.61)
.
(i,m)
From T̄L the strain concentration tensors can be evaluated via eqns. (2.18) and the
effective elasticity tensors via eqns. (2.85). The interpolation factors f (ξ) are chosen as
ξ + ξ2
f (ξ) = . (2.62)
2
This approach is capable of providing useful estimates for some composites, see, e.g.,
Ghossein and Lévesque (2012), but has a weaker physical background than “standard”
mean-field methods. A related scheme based on the Hill bounds, eqns. (3.1) was proposed
by Perdahcıoğlu and Geijselaers (2011).

35
Three-Point Estimates
The analytical estimates discussed in Section 2.3 are essentially based on two-point statis-
tics of the phase geometry, compare fig. 3.1, making them sensitive to reinforcement volume
fraction and aspect ratio. Models making use of 3-point (or higher) statistical descriptors
can resolve further details of the microgeometry, e.g., the relative sizes of reinforcements.

Torquato (1997, 1998a) developed third-order estimates for the effective elastic moduli
of two-phase composites on the basis of weak-contrast expansions that incorporate three-
point statistics via a pair of three-point microstructural parameters, η(ξ) and ζ(ξ), which
are also used in the three-point bounds discussed in Section 3.2. For convenience, these
models are referred to as “three-point estimates” in the following. The three-point pa-
rameters are functions of the inhomogeneity volume fraction and are available as as fits or
in tabulated form for a number of configurations involving spherical particles or aligned,
continuous, cylindrical fibers; for additional information see Section 3.2. Like Generalized
Self-Consistent schemes, three-point estimates provide results in terms of effective elastic
moduli rather than elasticity tensors.

The three-point estimates lie between the corresponding three-point bounds, and tend
to give excellent agreement with with numerical results obtained with periodic homogeniza-
tion using multi-inhomogeneity volume elements, especially at moderate elastic contrasts
and inhomogeneity volume fractions; differential schemes may do slightly better at ele-
vated volume fractions. In assessing the results of these models it should be borne in
mind, though, that they do not account for the effects of flaws, which are typically present
in actual materials.

2.6 Mean-Field Methods for Composites with Non-

aligned Reinforcements
The macroscopic symmetry of composites reinforced by nonaligned short fibers in many
important cases is isotropic (e.g., for random fiber orientations) or transversally isotropic
(for fiber arrangements with axisymmetric orientation distributions, e.g., fibers with planar
random orientation). However, processing conditions can give rise to a wide range of fiber
orientation distributions and thus to lower overall symmetries, compare, e.g., Allen and
Lee (1990).

Reinforcements with General Orientation Distributions

Descriptions of the microgeometries of nonaligned and hybrid matrix–inclusion compos-
ites typically make use of the orientation distribution functions (ODFs) and/or aspect
ratio or length distribution functions (LDFs) of the reinforcements, both of which can be
determined experimentally. Beyond this, additional, qualitative information may be re-
quired on the phase geometry to be modeled, as is the case for materials reinforced by
short fibers: At elevated fiber volume fractions local domains of (nearly) aligned fibers are
typically observed in short fiber-reinforced composites, which give rise to a “grain-type”
mesostructure, referred to as “aggregate systems” by Eduljee and McCullough (1993). For
modeling composites of this type two-step homogenization schemes that involve meso-level

36
averages at the “grain” level, see, e.g., Camacho et al. (1990) or Pierard et al. (2004), may
be appropriate. At low to moderate fiber volume fractions, in contrast, the orientations of
neighboring fibers are essentially independent within the geometrical constraints of non-
penetration. Such “dispersed systems” (Eduljee and McCullough, 1993) can be modeled
via single-step mean-field schemes involving configurational averaging procedures, which
may encompass aspect ratio averaging in addition to orientational averaging.

Studying the responses of nonaligned composites with dispersed geometries typically

involves orientational averaging of tensor valued variables, which, in general, can be done
by direct numerical integration, see, e.g., Pettermann et al. (1997), or on the basis of
expansions of the ODF in terms of generalized spherical harmonics (Viglin (1961) expan-
sions), compare, e.g., Advani and Tucker (1987). The latter approach can be formulated
in a number of ways, e.g., via texture coefficients or texture matrices, compare Siegmund
et al. (2004) or Hessman et al. (2021). For a discussion of a number of issues relevant to
configurational averaging see Eduljee and McCullough (1993).

In order to obtain mean-field methods for composites reinforced by nonaligned inho-

mogeneities one may consider tensors X(i) in suitably oriented local coordinate systems
described by the Euler angles ϕ, ψ, θ. By transforming the X(i) into the global coordi-
nate system the tensors X(i)∠ (ϕ, ψ, θ) are obtained, which form the basis for generating
orientational averages
Z 2π Z 2π Z π
(i)
X = X(i)∠ (ϕ, ψ, θ) ρ(ϕ, ψ, θ) dϕ dψ dθ . (2.63)
0 0 0

Here ρ(ϕ, ψ, θ) denotes the orientation distribution function of the reinforcements, which
is assumed to be normalized such that
Z 2π Z 2π Z π
1 = ρ(ϕ, ψ, θ) dϕ dψ dθ = 1 .
0 0 0

(i)
Appropriately transforming the dilute inhomogeneity concentration tensors, Ādil and
(i)
B̄dil , which are discussed in Section 2.2, provides rotated dilute concentration tensors, de-
(i)∠ (i)∠
noted as Ādil and B̄dil , to which the orientational averaging procedure can be applied in
(i) (i)
order to arrive at orientation averaged dilute concentration tensors, Ādil and B̄dil ,
respectively. The latter concentration tensors can be interpreted as describing an “equiv-
alent phase of nonaligned, non-interacting inhomogeneities”. The core statement of the
Mori–Tanaka approach, eqn. (2.39), can then be written in the form
(i) (i) (m) (i)
hσi(i) = B̄dil hσi(m) = B̄dil BMT hσi = BMT hσi , (2.64)

which allows evaluating the stress concentration tensors for the orientation averaged case,
(m) (i)
BMT and BMT , as
(m) (i)
BMT = [(1 − ξ)I + ξ B̄dil ]−1
(i) (i) (i)
BMT = B̄dil [(1 − ξ)I + ξ B̄dil ]−1 (2.65)

37
by analogy to eqns. (2.42) and (2.43). Equivalent expressions can be given for the strain
concentration tensors. On the basis of eqns. (2.83), (2.8) and (2.9) the effective macroscopic
elasticity tensors take the form (Benveniste, 1987)
(i) (i)
E∗MT = E(m) + ξ (E(i) − E(m) )Ādil [(1 − ξ)I + ξ Ādil ]−1
(i) (i)
C∗MT = C(m) + ξ (C(i) − C(m) )B̄dil [(1 − ξ)I + ξ B̄dil ]−1 . (2.66)
If both phases are isotropic, eqns. (2.66) can be simplified to give
(i) (i)
E∗MT = E(m) + ξ(E(i) − E(m) ) Ādil [(1 − ξ)I + ξ Ādil ]−1
(i) (i)
C∗MT = C(m) + ξ(C(i) − C(m) ) B̄dil [(1 − ξ)I + ξ B̄dil ]−1 . (2.67)
For this case analytical expressions for E∗MT in terms of two order parameters were given
by Giordano (2005).

Orientational averaging can also be used with the effective-medium approaches dis-
cussed in Section 2.3.2. For the classical self-consistent scheme, the equivalent of eqn. (2.53)
takes the form
(p,∗)
ESC,n+1 = E(m) + ξ (p) (E(p) − E(q) ) T̄dil,n
 −1
CSC,n+1 = ESC,n+1 (2.68)
and for the differential scheme the expression
dE∗D 1 (i,∗)
= (E(i) − E∗D ) T̄dil (2.69)
dξ 1−ξ
can be obtained from eqn. (2.54). In contrast to the Mori–Tanaka model, which is explicit,
(i,∗)
multiple evaluations of orientation averages of the type (E(i) − E∗ ) T̄dil are required
for both of the above methods, making them more expensive. It is worth noting that the
analytical evaluation of the Eshelby tensors is only supported if the effective medium is
isotropic, which implies spatially random fiber orientations, or transversally isotropic at
all steps29 .

The stress hσi(i) evaluated from eqn. (2.64) is an average over all inhomogeneities,
irrespective of their orientations, and, accordingly, provides rather limited information.
The average stresses in inhomogeneities of a given orientation (ϕ, ψ, θ), which may be of
higher practical interest, can be obtained as
(i)∠ (m)
hσi(i)∠ = B̄dil BMT hσi , (2.70)
compare (Duschlbauer et al., 2003b). Results obtained with the above relation are in good
agreement with numerical predictions for moderate reinforcement volume fractions and
elastic contrasts as well as spheroidal fibers of moderate aspect ratios, compare fig. 2.4.

29
The expressions of Clyne and Withers (1993) for Eshelby tensors of spheroids embedded in transver-
sally isotropic matrices hold only for the case that the major axis of the former is aligned with the axis
of symmetry of the latter, a condition that is not necessarily fulfilled for nontrivial ODFs. The same
restriction also holds for Mori–Tanaka-type methods, such as eqns. (2.65) and (2.66). It can, of course, be
circumvented by numerical evaluation of the Eshelby or Hill tensors.

38
 4

σ1 [MPa] 2

1
σ1 - MTM
0 σax - MTM
σ1 - UC
-1
0 10 20 30 40 50 60 70 80 90
Θ[°]
Figure 2.4: Dependence of the averaged stresses in individual fibers on the fiber orientation angle
Θ predicted for a uniaxially loaded SiC/Al MMC reinforced by randomly oriented spheroidal
fibers of aspect ratio a = 5 (Duschlbauer et al., 2003b). Results for the maximum principal stress
σ1 and the axial stress σax obtained with an “extended” Mori–Tanaka scheme are shown as solid
and dashed lines, respectively. Unit cell results are presented in terms of the mean values (solid
circles) and standard deviations (error bars) of the maximum principal stress within individual
fibers. Θ describes the angle between a given fiber and the loading (axial) direction.

“Extended” Mori–Tanaka methods for modeling the elastic behavior of microstruc-

tures that contain nonaligned inhomogeneities were developed by a number of authors
(Benveniste, 1990; Dunn and Ledbetter, 1997; Pettermann et al., 1997; Mlekusch, 1999).
These models differ mainly in the algorithms employed for orientational or configurational
averaging30 . In addition to modeling composites reinforced by nonaligned fibers or parti-
cles, methods of this type have also been used for studying fabric-reinforced composites,
the orientation of the fiber tows being described by appropriate “equivalent orientations”
(Gommers et al., 1998) and for approximating the behavior of composites reinforced by heli-
coidally twisted fibers (Shi et al., 2004). In cases involving finite numbers of fibers of known
orientations, the integrals in eqn. (2.63) degenerate into sums and the model becomes a
multi-phase Mori–Tanaka method analogous to eqn. (2.83), compare, e.g., Duschlbauer
et al. (2011). Extended Mori–Tanaka methods are subject to similar limitations as other
multi-phase Mori–Tanaka approaches, and may lead to non-symmetric “effective stiffness
tensors” as mentioned in Section 2.8.

Mean-field approaches based on orientational averaging and Mori–Tanaka methods have

also been employed in studies of the nonlinear behavior of nonaligned composites. Secant
plasticity schemes of the above type were used for describing ductile matrix materials
(Bhattacharyya and Weng, 1994; Dunn and Ledbetter, 1997) and incremental approaches
30
Alternative modified Mori–Tanaka models for nonaligned composites that use spatial averaging of the
Eshelby tensors, see, e.g., Johannesson and Pedersen (1998), have a weaker mechanical basis.

39
to modeling composites consisting of an elastoplastic matrix reinforced by nonaligned or
random short fibers were proposed, e.g., by Lee and Simunovic (2000), where debonding
between reinforcements and matrix is also accounted for.

Another option for studying composites with nonaligned reinforcements is provided

by the Hashin–Shtrikman estimates of Ponte Castañeda and Willis (1995). As discussed
in Section 2.3.1 this method is based on “ellipsoid-in-ellipsoid” phase arrangements, with
different Eshelby tensors pertaining to the ellipsoids describing the two-point correlations of
the phase arrangement and to those describing the shapes of the inhomogeneities. Equation
(2.85) can be directly extended to nonaligned reinforcements by converting the sum into
an orientation average to give
h i−1
(i)∠
E∗PW = E(m) + ξ (i) (E(i) − E(m) ) Ādil − Sd C(m) (2.71)

As mentioned before, such Hashin–Shtrikman estimates are rigorous within the effective-
field setting, provided none of the (aligned) “safety ellipsoids” described by Sd overlap
and none of the (nonaligned) inhomogeneities penetrate outside the associated safety el-
lipsoids. These requirements result in a limited range of strict applicability in terms of
inhomogeneity volume fractions, which can be rather small for markedly prolate or oblate
reinforcements that show a considerable degree of misalignment.

A number of analytical descriptions for short fiber-reinforced composites apply mean-

field models within a two-step strategy (i.e., a hierarchical approach as discussed in Chap-
ter 9). Laminate analogy approaches, see, e.g., Fu and Lauke (1998), and Pseudo-Grain
models, compare, e.g., (Pierard et al., 2004), approximate nonaligned reinforcement ar-
rangements by a stack of layers or a number of two-phase regions (“grains”), respectively,
each of which pertains to one fiber orientation and/or one reinforcement aspect ratio or
material. The effective elastic tensors are then obtained by a second homogenization step
over the two-phase layers or grains. For laminate analogy approaches this is of the Voigt
(strain coupling) type; such models can be useful especially for describing composites with
planar random fiber orientations (Huang, 2001). Pseudo-Grain models are more flexible in
terms of the homogenization method used for the second step and they have formed the
basis for descriptions of the nonlinear and damage behavior of composites reinforced by
nonaligned short fibers (Doghri and Tinel, 2005; Kammoun et al., 2011).

In addition to the above models a number of other analytical methods have been pro-
posed for studying composites with nonaligned reinforcements. Most of them are based on
the assumption that the contribution of a given fiber to the overall stiffness and strength
depends solely on its orientation with respect to the applied load and on its length, in-
teractions between neighboring fibers being neglected. The paper physics approach (Cox,
1952) and the Fukuda–Kawata theory (Fukuda and Kawata, 1974) are based on sum-
ming up stiffness contributions of fibers crossing an arbitrary normal section on the basis
of fiber orientation and length distribution functions, compare also Jayaraman and Ko-
rtschot (1996). Such theories use shear lag models (Cox, 1952; Fukuda and Chou, 1982) or
modifications thereof (Fukuda and Kawata, 1974) for describing the behavior of a single
fiber embedded in matrix material, the results being typically given as modified rules of
mixtures with fiber direction and fiber length corrections.

40
Randomly Oriented Reinforcements
For the special case of randomly oriented (“uniform random”) fibers or platelets of a
given aspect ratio, orientation averaged dilute partial strain concentration tensors, known
as Wu tensors (Wu, 1966), may be used. Wu tensors can be inserted into Mori–Tanaka
methods or classical self-consistent schemes (Berryman, 1980) to describe composites with
randomly oriented phases of matrix–inclusion or certain interpenetrating topologies; in
the former case eqn. (2.67) is recovered. Tandon and Weng (1986) proposed an alternative
Mori–Tanaka based model for composites reinforced by randomly oriented inhomogeneities.
A further mean-field method for such materials, the Kuster and Toksöz (1974) model,
essentially is a dilute description applicable to matrix–inclusion topologies and tends to
give non-physical results at high reinforcement volume fractions. For discussions on the
relationships between some of the above approaches see, e.g., Berryman and Berge (1996)
or Hu and Weng (2000). It is worth noting that, due to the overall isotropic behavior
of composites reinforced by randomly oriented fibers or platelets, their elastic response
must comply with the Hashin and Shtrikman (1963) bounds for macroscopically isotropic
materials, compare Chapter 3 and see also table 6.2 in Section 6.5.

2.7 Mean-Field Methods for Non-Ellipsoidal Reinfor-

cements
When non-ellipsoidal inhomogeneities are subjected to a homogeneous eigenstrain or a far-
field load, the resulting stress and strain fields in them are, in general, inhomogeneous. As
a consequence, the interior-point Eshelby tensors and dilute inhomogeneity concentration
tensors depend on the position within such inhomogeneities. Analytical solutions (per-
taining to the first Eshelby problem) are available only for certain inhomogeneity shapes,
see, e.g., Mura (1987) and Onaka (2001). For some further shapes volume averaged Es-
helby tensors have been reported. They have been proposed for use within approximate
mean-field models for composites reinforced by non-ellipsoidal inhomogeneities, see, e.g.,
Zheng et al. (2006) and Hashemi et al. (2009). Depending on how they were evaluated,
such tensors pertain to either the first or the second Eshelby problem, compare Section
2.2, but not to both of them.

An elegant approach to handling non-ellipsoidal inhomogeneities within a mean-field

framework is provided by the contribution tensor formalism, compare eqns. (2.19) to (2.21).
Using eqns. (2.19) contribution tensors pertaining to dilute inhomogeneities of general
(i) (i)
shape, Hdil and Ndil , can be extracted from any solution for the effective compliance or
elasticity tensors, E∗dil and C∗dil , obtained for suitable dilute matrix–inhomogeneity config-
urations. In the case of non-ellipsoidal reinforcement shapes or highly anisotropic matrix
behavior evaluating E∗dil or C∗dil typically involves numerical methods. For ellipsoidal in-
homogeneities in isotropic or transversally isotropic matrices dilute contribution tensors
can be generated with much smaller effort from solutions based on Eshelby tensors and
equivalents, e.g., via eqn. (2.20). For dilute contribution tensors eqns. (2.21) simplify to
(i) (i) (i) (i)
Hdil = −C(m) Ndil C(m) and Ndil = −E(m) Hdil E(m) . (2.72)

The dilute contribution tensors directly give rise to non-interacting solutions analogous

41
to the ones discussed at the end of Section 2.2. They also form the basis for effective-field
and effective-medium schemes that are suitable for handling non-dilute inhomogeneity
volume fractions and are fully equivalent to the models presented in Section 2.3, where
descriptions employing concentration tensors are used. Within a contribution tensor-based
two-phase mean-field framework, the Mori–Tanaka method takes the form
(i) (i)  (i) −1
HMT = Hdil (1 − ξ)I + ξ(C(i) − C(m) )−1 Hdil
 (i) −1 −1
= (1 − ξ)Hdil + ξ(C(i) − C(m) )−1
(i) (i)  (i) −1
NMT = Ndil (1 − ξ)I + ξ(E(i) − E(m) )−1 Ndil
 (i) −1 −1
= (1 − ξ)Ndil + ξ(E(i) − E(m) )−1 , (2.73)

the classical self-consistent scheme becomes

(p) (p,∗)
HSC = ξ (p) (C(p) − C(q) )(C(p) − C∗SC )−1 Hdil
(p) (p,∗)
NSC = ξ (p) (E(p) − E(q) )(E(p) − E∗SC )−1 Ndil , (2.74)

the differential scheme is given by

dC∗D 1 (i,∗)
= Hdil
dξ (1 − ξ)
dE∗D 1 (i,∗)
= Ndil , (2.75)
dξ (1 − ξ)

and the Maxwell scheme results as

h i−1
(i) (i) (m)
HMX = [Hdil ]−1 − ξE
(I − SE )
h i−1
(i) (i) −1 (m)
NMX = [Ndil ] − ξSE C , (2.76)

(i,∗)
compare Eroshkin and Tsukrov (2005) and Kachanov and Sevostianov (2018). The Hdil
(i,∗)
and Ndil appearing in eqns. (2.74) and (2.75) are defined as
(i,∗) (i,∗) (i,∗) (i,∗)
Hdil = (C(i) − C∗n )W̄dil and Ndil = (E(i) − E∗n )T̄dil (2.77)

where E∗n and C∗n stand for the n-th iterates of E∗SC and C∗SC for the effective medium. For
in-depth discussions of contribution-tensor based mean-field models see Sevostianov (2014)
or Kachanov and Sevostianov (2018).
(i) (i)
When expressions for Hdil and Ndil are evaluated from numerical models via eqn. (2.19),
both ξ and terms of the type E∗ − E(m) may become very small, so that considerable care
is required due to the potential sensitivity of the results to roundoff and discretization
errors31 , especially in the case of non-convex inhomogeneities, compare, e.g., Sevostianov
et al. (2008). In such models either macrohomogeneous or periodicity boundary conditions
31
Because extracting hεi(i) or hσi(i) is not subject to this issue, when using numerical models it may be
(i) (i)
preferable to evaluate Hdil and Ndil via the phase-averaged dilute concentration tensors plus eqns. (2.20)
rather than directly via eqns. (2.19).

42
may be employed, the wide matrix region surrounding the inhomogeneity making the re-
sults rather insensitive to the boundary conditions actually used.

For non-ellipsoidal inhomogeneities the combination of numerically evaluated dilute

contribution tensors with mean-field methods according to eqns. (2.73) or (2.76) gives useful
approximations to the more accurate results obtained by the computationally much more
expensive full field models discussed in Chapters 5 to 8, see, e.g., Trofimov et al. (2017).
Such modeling strategies are, however, not very efficient when incorporating implicit mi-
cromechanical schemes, e.g., eqns. (2.74) or (2.75), which require multiple evaluations of
the contribution tensors, as indicated by eqns. (2.77).

2.8 Mean-Field Methods for Multi-Phase Thermoe-

lastic Composites
Multi-phase materials can show a wide range of microgeometries. The present section
concentrates on two important groups of composites with matrix–inclusion microtopology.
One of them consists of a contiguous matrix phase that is reinforced by aligned, randomly
positioned, uniform inhomogeneities showing different shapes and/or material behavior. In
the other group all reinforcements are identical, aligned and consist of a core surrounded
by a coating, making them non-uniform. The methods discussed in this context can be
extended to handling many of the more complex configurations.

2.8.1 Mean-Field Methods for Multi-Phase Thermoelastic Com-

posites with Aligned Reinforcements
The majority of the general relations between mean-fields in thermoelastic two-phase ma-
terials given in Section 2.1 can be directly extended to multi-phase materials consisting of
N phases (p) . Specifically, the equivalents of eqn. (2.7) take the form
X X (p)
ξ (p) Ā(p) = I ξ (p) β̄ =o
(p) (p)
X X
ξ (p) B̄(p) = I ξ (p) κ̄(p) = o , (2.78)
(p) (p)

the effective elastic tensors can be evaluated as

X X
E∗ = ξ (p) E(p) Ā(p) C∗ = ξ (p) C(p) B̄(p) (2.79)
(p) (p)

by analogy to eqns. (2.8) and (2.9), and eqns. (2.14) become

43
 hX i−1 hX i−1
(p) (p) (p) (q) (q) (q) (p) (p) (p) (q) (q) (q)
Ā = C B̄ ξ C B̄ and B̄ = E Ā ξ E Ā ,
(q) (q)
(2.80)
respectively. Equations (2.12), (2.14) and (2.17) hold irrespective of the number or topol-
ogy of the phases. There seem to be no multi-phase equivalents of eqns. (2.15), however,
which imposes considerable restrictions on doing localization or evaluating effective CTEs
with multi-phase mean-field methods.

The effective properties of multi-phase composites with matrix–inclusion topology can

be obtained by summing up the contribution tensors, H(i) or N(i) , pertaining to the dilute
or non-dilute micromechanical model to be used,
X X
C∗ = C(m) + ξ (i) H(i) and E∗ = E(m) + ξ (i) N(i) . (2.81)
(i)6=(m) (i)6=(m)

Approaches based on these relations often lead to compact expressions for the elastic ten-
sors of multi-phase composites, compare Kachanov and Sevostianov (2018).

Formally, multi-phase versions of the methods discussed in Sections 2.3.1 and 2.3.2 can
be obtained in a straightforward way. However, for the “simpler” effective-field methods
the range of application of the resulting expressions tends to be rather limited. No major
issues have been reported, in contrast, for more general effective-field approaches such as
the multi-particle effective-field method or for effective-medium models like eqns. (2.86)
and (2.87).

Multi-Phase Mori–Tanaka Estimates

For multi-phase materials consisting of a matrix (m) into which a number of aligned inhomo-
geneity phases (i) (understood to consist of inhomogeneities of identical material behavior
and shape) are embedded, Mori–Tanaka phase concentration tensors may be obtained as
(m)  P (i) (i) −1 (i) (i) 
X (i) −1
ĀMT = ξ (m) I + ξ Ādil ĀMT = Ādil ξ (m) I + ξ (i) Ādil
(i)6=(m) (i)6=(m)
(m)  P (i) −1 (i) (i) 
X (i) −1
B̄MT = ξ (m) I + ξ (i) B̄dil B̄MT = Ādil ξ (m) I + ξ (i) B̄dil (2.82)
(i)6=(m) (j)6=(m)

and the corresponding effective elastic tensors become

hX ihX i−1
∗ (p) (p) (p) (p) (p)
EMT = ξ E Ādil ξ Ādil
(p) (p)
h X ih X i−1
(i) (i)
= E(m) + ξ (i) (E(i) − E(m) )Ādil ξ (m) I + ξ (i) Ādil
(i)6=(m) (i)6=(m)
hX ihX i−1
(p) (p)
C∗MT = ξ (p) C(p) B̄dil ξ (p) B̄dil
(p) (p)
h X ih X i−1
(m) (i) (i) (m) (i) (m) (i) (i)
= C + ξ (C −C )B̄dil ξ I+ ξ B̄dil (2.83)
(i)6=(m) (i)6=(m)

44
 (p)
in direct equivalence to the two-phase expressions, eqns. (2.44); here is understood to
encompass all phases.

The multi-phase Mori–Tanaka methods defined by eqns. (2.82) and/or (2.83) are known
to give rise to non-symmetric effective “elastic tensors” (and thus to unphysical results) in
many practically relevant situations, compare Benveniste et al. (1991) or Ferrari (1991).
Such behavior crops up, e.g., when studying microstructures involving aligned spheroidal
inhomogeneities that show both different material behaviors and different aspect ratios; it
may also appear for composites reinforced by nonaligned inhomogeneities. It is a conse-
quence of assumptions on the linkage between phase arrangement and inhomogeneity shape
implicitly incorporated into Mori–Tanaka methods, compare fig. 2.2 and the remarks on
Ponte–Willis estimates in Section 2.3.1. Recent discussions of these issues were provided
by, e.g., Sevostianov and Kachanov (2014), Rodin and Weng (2014) and Jiménez Segura
et al. (2023).

The above limitations may be dealt with by directly symmetrizing the Mori–Tanaka
(i) (i)
elasticity tensors, eqns. (2.83), or by symmetrizing the tensors Aeff or Beff defined by
eqns. (2.38). The latter approach was followed by Sevostianov and Kachanov (2014) and
by Jiménez Segura et al. (2023) within the contribution tensor and the concentration
tensor frameworks, respectively. It is interesting to note that when using such fixes the
original Mori–Tanaka assumption of equating the effective fields with the matrix fields,
viz., εE = hεi(m) and σ E = hσi(m) , compare eqns. (2.38) and (2.39), is no longer fulfilled
once symmetrization cuts in32 .

Multi-Phase Maxwell Schemes

A Maxwell estimate for the macroscopic elasticity tensor of a multi-phase composite of
matrix–inhomogeneity topology can be given in the form
" #−1
 X −1
E∗MX = E(m) + ξ (i) [(E(i) − E(m) )−1 + S(i,m) C(m) ]−1 − SE C(m) , (2.84)
(i)6=(m)

which is a generalization of eqns. (2.48). For the special case that all inhomogeneity phases
are aligned and have identical shapes (but different material properties), their aspect ratio
is a natural choice for the effective region, so that SE = S(i,m) . Under these conditions the
predictions of Mori–Tanaka methods and Maxwell schemes coincide.

For general cases, i.e., when the inhomogeneity phases are either not aligned or show
different aspect ratios, the proper choice of the shape of the inhomogeneous region, compare
fig. 2.3, and thus of SE , tends to be a nontrivial issue. Sevostianov and Kachanov (2014)
conjectured that the pertinent Hill tensor PE = SE C(m) can be Papproximated as a weighted
(i) (i,m)
sum of the Hill tensors of the inhomogeneity phases, PE ≈ ξ P , but Buryachenko
(2022b) gave counter-examples and concluded that the Maxwell scheme tends to break
down for complex microgeometries.
32
Even seemingly minor deviations from symmetry in E∗ and C∗ tend to compromise the validity of
concentration tensors evaluated via eqns. (2.82). This issue, which is most marked for hybrid composites
containing reinforcements of widely different shape, is not resolved by symmetrization.

45
Multi-Phase Ponte–Willis Estimates
The effective elasticity tensor of the Hashin–Shtrikman scheme of (Ponte Castañeda and
Willis, 1995) for composites reinforced by multiple aligned inhomogeneity phases can be
written as
!−1 !
X (i,m)
X (i,m)
E∗PW = E(m) + I − ξ (i) Ndil Sd C(m) ξ (i) Ndil , (2.85)
(i)6=(m) (i)6=(m)

(i,m)
where Ndil stands for the dilute elasticity contribution tensor defined in eqn. (2.35).
In eqn. (2.85) all phases follow the same ellipsoidal arrangement statistics described by
Pd = Sd C(m) .

Multi-Phase Classical Self-Consistent Scheme

The multi-phase version of the classical self-consistent estimates for the effective elasticity
tensor takes the form
X  −1
E∗SC,n+1 = E(p) I + Sn(p,∗) CSC,n [E(p) − ESC,n ]
(p)

C∗SC,n+1 = (E∗SC,n+1 )−1 (2.86)

(p,∗)
in analogy to the two-phase case, eqn. (2.53). The Eshelby tensor Sn pertains to an
(p)
ellipsoidal inhomogeneity with a shape characteristic of phase that is embedded in the
effective medium. This model is not primarily aimed at classical composites with matrix–
inclusion topology, but rather at materials with grain-like microgeometries. It may require
under-relaxation to achieve convergence.

Multi-Phase Differential Scheme

The two-phase Differential Scheme, eqns. (2.54), can be extended to composites containing
multiple inhomogeneity phases to give
dE∗D 1 X (i,∗)
I
= I
η (i) [E(i) − E∗D ] T̄dil (2.87)
dξ 1−ξ
(i)6=(m)
P
where ξ I = (i)6=(m) ξ
(i)
and the partial volume fractions η (i) are defined in analogy to
eqn. (2.92). Like its two-phase counterpart, eqn. (2.87) can be integrated up numerically,
e.g., with Runge–Kutta algorithms. In eqn. (2.87) the η (i) are implicitly assumed to be
increased proportionally, but different integration paths in phase volume space can be en-
forced by introducing suitable parameterizations; in general, these lead to different results
for identical target volume fractions, compare Norris (1985).

Multi-Phase Hashin–Shtrikman Tensors

Multi-phase versions of eqns. (2.59) and (2.60) take the forms
X −1
E∗HS = ξ (p) (L(i,0) + E(p) )−1 − L(i,0) , (2.88)
(p)

compare Bornert (2001), and

46
 X  X −1
(p) (p) (p,0) (p) (p,0)
E∗HS = ξ E T̄dil ξ T̄dil , (2.89)
(p) (p)

see, e.g., Walpole (1966), respectively. When composites of matrix–inclusion topology

are to be studied, both of the above expressions are limited to inhomogeneity phases of
identical shape, which is described by L(i,0) in eqn. (2.88).

Two-Step and Multi-Step Models for Multi-Phase Composites

Two-step and multi-step mean-field models have been devised with the main aim of cir-
cumventing the limitations of the “direct” mean-field models discussed above.

In two-step methods, an N -phase composite with matrix–inclusion topology is treated

as consisting of N − 1 two-phase regions, each of which consists of a single inhomogeneity
phase embedded in the matrix. In the first step, each of these “pseudo-grains” is homog-
enized by a standard two-phase model. An N − 1-phase homogenization method is then
applied in the second step to estimate the effective tensors. For example, the first step
of such a pseudo-grain model may be consist of two-phase Mori–Tanaka models such as
eqns. (2.44), all of which use an inhomogeneity volume fraction of ξ I , compare eqn. (2.92),
and the second one of a multi-phase classical self-consistent scheme, eqn. (2.86), with grain
volume fractions of η (i) = ξ (i) /ξ I . Results obtained with such a PGR/MT–SC model are
presented in section 4.3. Conceptually similar pseudo-grain approaches have been widely
used for homogenizing composites with nonaligned reinforcements, compare section 2.6.

Multi-step models (also referred to as iterative or sequential approaches) handle N -

phase composites via a sequence of N − 1 two-phase homogenization procedures such that
the elasticity tensors resulting from step n − 1 serve as the embedding material (“matrix”)
of step n, the actual matrix being used for this purpose in step 1. In order to recover
the prescribed phase volume fractions of the inhomogeneity phases, ξ (i) , the inhomogeneity
(i)
volume fraction for step n, ξn , must be chosen as

ξ (n)
ξn(i) = P(N−1) . (2.90)
(q)
(q)=(n) ξ

Most applications of such multi-step methods have employed the Mori–Tanaka method for
all steps, resulting in Sequential Mori–Tanaka or Multi-Step Mori–Tanaka models, see, e.g.,
Yang et al. (2007). In order to fulfill the homogenization conditions, the size of inhomo-
geneities should be increasing with growing number (n) of the phases. The predictions of
such multi-step models for the effective tensors in general depend on the chosen sequence
of the phases in the homogenization scheme.

Other Analytical Estimates for Multi-Phase Composites

The most general analytical models for multi-phase composites, such as Multi-Particle
Effective-Field Methods, see, e.g., (Buryachenko, 2022b), are based on integral equation
formulations, compare (Buryachenko, 2022b), and tend to be too complex to be described
by the notation used in the present report.

47
2.8.2 Analytical Models for Composites Reinforced by Coated
Inhomogeneities
Composites reinforced by coated particles or fibers (sometimes called “core–shell” inhomo-
geneities) show a special type of three-phase, or, in the case of multiple coatings, multi-
phase microgeometry. The modeling of such materials has been the focus of considerable
research interest, especially due to its relevance to nanocomposites. The most widely
used analytical approaches have been versions of the Double Inclusion Method (Hori and
Nemat-Nasser, 1993) and self-consistent schemes based on exact solutions for dilute, coated
or multi-coated spherical inhomogeneities, see Hervé and Zaoui (1990), Hervé and Zaoui
(1993) and Bonfoh et al. (2012), as well as cylindrical continuous fibers, see Hervé and
Zaoui (1995). The following discussion will concentrate on the case of single, uniform in-
terphases, i.e., three-phase configurations.

Arguably, the most flexible models for handling composites reinforced by coated inho-
mogeneities are two-step schemes, which describe the coated particles or fibers via equiva-
lent uniform inhomogeneities and were referred to as the “replacement method” by (Hashin,
1972). Models of this type can also be applied to composites with imperfect interfaces33 ,
compare, e.g., Duan et al. (2022), and graded interphases (Sevostianov and Kachanov,
2007a). Within the two-step framework eqns. (2.78) and (2.79) give rise to the relations
(c) (l)
E∗ = ξ (m) E(m) Ā(m) + ξ (c) E(c) Ādil + ξ (l) E(l) Ādil
= ξ (m) E(m) Ā(m) + ξ I EIeqv ĀIeqv
ĀIeqv = η (c) Ā(c) + η (l) Ā(l) , (2.91)

at the level of the composite. Here the definitions

I (c) (l) (c) ξ (c) (l) ξ (l)

ξ =ξ +ξ η = I η = I (2.92)
ξ ξ

are used, with (c) denoting the inhomogeneity core, (l) the coating layer and I the equivalent
inhomogeneity. At the level of the coated inhomogeneity the expressions
(c) (l)
ĀIdil,eqv = η (c) Ādil + η (l) Ādil
  −1
(c) (l)
EIeqv = η (c) E(c) Ādil + η (l) E(l) Ādil ĀIdil,eqv (2.93)

(l) (c)
allow evaluating ĀIdil,eqv as well as EIeqv whenever Ādil and Ādil are known, whereas the
relations

ĀIdil,eqv = [I + SI C(m) (EIeqv − E(m) ]−1

1
Ā(c) = (c) (E(c) − E(l) )−1 (EIeqv − E(l) )−1
η
1
Ā(l) = (c) (E(l) − E(c) )−1 (EIeqv − E(c) )−1 (2.94)
η
33
In the present context a coating of finite thickness is referred to as an interphase, contrasting with a
zero-thickness interface between two constituents where fields may be discontinuous.

48
may be used to extract the dilute concentration tensors if EIeqv is available. Together,
eqns. (2.91) to (2.94) split up the three-phase problem into two independent two-phase
ones: the “equivalent homogeneous fiber sub-model” provides expressions for ĀIdil,eqv and
EIeqv , whereas the “composite level-level sub-model” leverages this data into obtaining re-
sults on ĀI and E∗ for non-dilute volume fractions ξ I . For the latter step essentially any
two-phase mean-field model can be used, and the two-step model as a whole supports
mean-field localization for all constituents, stress concentration tensors being extracted
via eqn. (2.14). For further details see Böhm (2019) and Böhm (2023). Multiple coatings
can be handled by analogy and result in a sequence of two-phase micromechanical problems.

The equivalent homogeneous fiber sub-model can be tackled, on the hand, by obtaining
(c) (l)
dilute phase-averaged concentration tensors Ādil and Ādil from the non-uniform local fields
predicted by the exact solutions for coated spherical particles, see Hervé and Zaoui (1990),
and for continuous, coated cylinders, see Wang and Huang (2016), Wang et al. (2016)
and Chatzigeorgiou and Meraghni (2019). On the other hand, the elasticity tensor of
the equivalent uniform inhomogeneity can be approximated via a number of mean-field
methods, among them Mori–Tanaka-type schemes of the form
h i−1
(i) (l) (c) (c) (l) −1 (l) (c,l) (l)
Eeqv = E + η (E − E ) + η S C , (2.95)

where the Eshelby tensor S(c,l) describes the inhomogeneity core embedded in the inter-
phase material34 .

The results of a number of analytical models for composites reinforced by simply coated
spheres or cylinders were compared to numerical predictions for some material combina-
tions by Böhm (2019) and Böhm (2023), respectively. The combination of the dilute
concentration tensors from the exact solutions with the three-point estimates of Torquato
(1997, 1998a) or the Differential Scheme at the composite level was consistently found to
give excellent agreement. Interestingly, two-step schemes that combine eqn. (2.95) with
the Mori–Tanaka method at the composite level, so called MTMT-models (Friebel et al.,
2006), also tracked the numerical predictions well, making them (as well as the more com-
plicated Generalized Self-Consistent Scheme based on a Reformulated Double Inclusion
Model proposed by Dinzart et al. (2016), which yields identical results) good candidates
for handling more general ellipsoidal reinforcement shapes.

2.9 Mean-Field Models for Nonlinear and Inelastic

Composites
Since the late 1970s considerable effort has been directed at modeling the mechanical
behavior of inhomogeneous materials in which one or more constituents show nonlinear
elastic, viscoelastic, elastoplastic or viscoelastoplastic responses. The main motivation of
such studies has been the need to describe the time dependent, creep and relaxation be-
havior of polymer matrix composites and the responses of composites with metallic phases.
34
Equation (2.95) corresponds to a single, simplified step of the General Explicit Eshelby-Based Es-
timator (GEEE) of Ghazavizadeh et al. (2019). The GEEE uses Ponte–Willis estimates for modeling
non-homothetic, multi-coated inhomogeneities.

49
Mean-field methods have been developed for and successfully adapted to studying many
aspects of the above problems.

For nonlinear composites the instantaneous stiffness operators are not phase-wise uni-
form even if each constituent is homogeneous, so that Eshelby’s results and mean-field
models cannot be extended directly from linear elasticity to nonlinear behavior. To deal
with this problem, “linear comparison composites” may be defined, which approximate the
actual, nonlinear materials’ responses for a given state. Accordingly, the nonlinear problem
is reduced to a sequence of linear ones by suitable linearization schemes. In continuum mi-
cromechanics, the most important approaches of this type are secant (Tandon and Weng,
1988) and incremental (Hill, 1965a) methods. In addition, tangent concepts (Molinari
et al., 1987), affine (“thermoelasticity-like”) formulations (Masson et al., 2000; Brenner
et al., 2001) and incremental secant algorithms (Wu et al., 2013) have been reported.

2.9.1 Viscoelastic Composites

Viscoelastic materials show hereditary behavior, i.e., their response at a given time depends
on their previous load history. Important issues in the mechanical behavior of viscoelastic
composites are, on the one hand, quasi-static responses such as relaxation and creep, which
can be described via relaxation function and creep compliance tensors. The dynamic be-
havior under periodic excitations, on the other hand, can be studied via complex modulus
tensors.

Correspondence principles are available (Hashin, 1965, 1970) that directly relate the
analysis of linear viscoelastic composites to that of linear elastic composites of identical
phase geometry for both of the above sets of problems, see also Schapery (1974) and Hashin
(1983).

For the quasi-static case the correspondence principle requires formulating the viscoelas-
tic problem in the Laplace–Carson transformed domain, where the transformed relaxation
function and creep compliance tensors are equivalent to the elasticity and compliance ten-
sors, respectively, in elastic micromechanics. On this basis replacement schemes can be de-
fined (Hashin, 1972) that allow to obtain Laplace–Carson transformed macroscopic moduli,
modulus tensors and phase averaged microfields of viscoelastic materials from mean-field
results, such as the ones discussed in Sections 2.3 to 2.5. The back transformation from the
Laplace–Carson to the time domain, however, typically is not straightforward, and closed
form solutions are not available in most cases. Accordingly, approximations must be in-
troduced or numerical methods must be used, compare, e.g., Schapery (1962) or Lévesque
et al. (2007). The correspondence principle for periodic excitations uses transforms to
Fourier space and the resulting replacement scheme (Hashin, 1972) directly generates ef-
fective complex modulus tensors from the effective elastic tensors obtained for a given
microgeometry35 .

35
The correspondence principles can be applied, on the one hand, to analytical expressions, such as the
CSA and CCA models (Hashin, 1972) or Mori–Tanaka and self-consistent schemes (Pichler and Lackner,
2009). On the other hand, numerical discrete microfield approaches of the type discussed in Chapters 6 to
7 can be adapted to incorporate them, see, e.g., Yi et al. (1998) or Brinson and Lin (1998).

50
 For in-depth discussions of and alternative concepts for mean-field models of linear
and nonlinear viscoelastic composites see, e.g., Paquin et al. (1999), Brenner and Masson
(2005), Lévesque et al. (2007), as well as Lahellec and Suquet (2007).

2.9.2 (Thermo-)Elastoplastic Composites

The following discussion is restricted to mean-field models based on continuum plasticity36 .
Nearly all work reported on such models for elastoplastic or viscoelastoplastic inhomoge-
neous materials has relied on secant, incremental, tangent, or affine linearization strategies;
for an overview see, e.g., Ponte Castañeda and Suquet (1998).

The main difficulties in applying mean-field methods to composites with elastoplastic

constituents lie in the path dependence of plastic behavior and in the often strong intra-
phase fluctuations of the microstress and microstrain fields in elastoplastic inhomogeneous
materials. Accordingly, each material point in an elastoplastic phase tends to follow a dif-
ferent trajectory in stress space, so that even a two-phase elastoplastic composite effectively
behaves as a multi-phase material and phase averages over reinforcements and matrix are
less useful descriptors than in the linear elastic regime. As a consequence, in mean-field
models of elastoplastic composites choices have to be made with respect to the linearization
procedure, the linear homogenization model, and the phase-wise equivalent stresses and
equivalent strains to be used in evaluating the elastoplastic constituent material behavior
(Zaoui, 2001).

Secant Methods for Elastoplastic Composites

Secant plasticity concepts in continuum micromechanics, see, e.g., Tandon and Weng (1988)
or Dunn and Ledbetter (1997), are based on the deformation theory of plasticity, in which
the elastoplastic behavior under radial, monotonic loading is approximated by nonlinear
elastic models.

In the simplest case of an isotropic elastoplastic phase that is described by J2 plasticity,

(p) (p)
the secant “elasticity” and “compliance” tensors, Esec and Csec , take the form
1 1
E(p)
sec = 3K
(p) E
Ovol + 2G(p) E
sec Odev and C(p)
sec = (p)
OC
vol + OC
(p) dev
, (2.96)
3K 2Gsec
respectively. Here OE and OC are the volumetric and deviatoric “partitioning tensors” for
(p)
elasticities and compliances37 , respectively. The secant shear modulus Gsec is equal to the
elastic shear modulus of the matrix, G(p) , in the elastic range. In the post-yield regime it
can be obtained from the current phase averages of the equivalent stress hσeqv i(p) and the
equivalent plastic strain hεeqv,pl i(p) as
G(p) hσeqv i(p)
G(p)
sec = (2.97)
hσeqv i(p) + 3G(p) hεeqv,pl i(p)
36
In a separate type of model phase averaged stress fields obtained by mean-field methods have been
used in dislocation-based descriptions of elastoplastic matrix behavior, see, e.g., Taya and Mori (1987).
37
In index notation the OE and OC are identical and correspond to the volumetric and deviatoric
projection tensors O defined in eqn. (2.2). If an engineering notation based on shear angles is used,
however, there are differences in the “shear terms” of OE C
dev , Odev and Odev .

51
on the basis of an additive strain decomposition, compare eqn. (2.1). The equivalent plastic
strain hεeqv,pl i(p) must be evaluated from hσeqv i(p) via an appropriate hardening law. The
bulk moduli K (p) are not affected by yielding due to the J2 assumption.

Expressions for the macroscopic secant tensors E∗sec and C∗sec can be obtained from the
phase secant tensors, eqn. (2.96), by mean-field relationships equivalent to eqns. (2.8) and
(2.9). For the case of elastic inhomogeneities embedded in an elastoplastic matrix they can
be given as

E∗sec = E(m)
sec + ξ(E
(i)
− E(m) (m)
sec )Āsec
C∗sec = C(m)
sec + ξ(C
(i)
− C(m) (m)
sec )B̄sec . (2.98)
(m) (m)
The non-dilute secant strain and stress concentration tensors, Āsec and B̄sec , can be gener-
(m) (m)
ated from dilute secant concentration tensors, Ādil,sec and B̄dil,sec , via a suitable mean-field
theory, e.g., a Mori–Tanaka method or a self-consistent scheme, compare Section 2.3. The
dilute secant concentration tensors and the Eshelby tensors used in them must be evalu-
ated using the current secant tensors of the constituents. Iterative procedures are required
for obtaining solutions corresponding to prescribed macroscopic strain or stress states.

In first order (“classical”) methods, the phase average of the equivalent stress required
in eqn. (2.97) is approximated from the phase averaged stress tensor,
h3 i 21
(p)
hσeqv i ≈ hσiTdev hσidev . (2.99)
2
This neglects contributions due to the local stress fluctuations and, accordingly, tends to
underestimate hσeqv i(p) , leading to errors in the estimates for the macroscopic elastoplastic
response38 . Clear improvements in this respect can be obtained by second order approxi-
mations that evaluate the phase averaged equivalent stress in terms of approximations to
(p)
the second order moments of stress, σ ⊗ σ , (Suquet, 1995; Buryachenko, 1996; Hu,
1997; Pierard et al., 2007), on the basis of energy considerations (Qiu and Weng, 1992), or
from a “current stress norm” (Ju and Sun, 2001).

Alternatively, secant theories for composites with nonlinear constituents can be ob-
tained from variational principles (Ponte Castañeda, 1991) or they can be formulated in
terms of potentials (Bornert and Suquet, 2001), which allows for a concise mathematical
presentation. Because secant models treat elastoplastic composites as nonlinearly elastic
materials they are limited to strictly monotonic loading and to radial (or approximately
radial) trajectories of the phase averaged stresses of the constituents in stress space39 ,
38
Because the square of the deviatoric stresses is required for evaluating hσeqv i(p) , the fluctuations give
rise to non-vanishing contributions.
As an extreme case, using eqn. (2.99) for evaluating the equivalent stress leads to predictions that materials
with spherical reinforcements will not yield under macroscopically hydrostatic loads or unconstrained
thermal expansion. This is in contradiction to other results.
39
The condition of radial loading paths in stress space at the constituent level is generally violated, at
least to some extent, in the phases of elastoplastic inhomogeneous materials, even for macroscopic loading
paths that are perfectly radial (Pettermann, 1997). This behavior is due to changes in the accommodation
of the phase stresses and strains in inhomogeneous materials upon yielding of a constituent and (to a much
lesser extent) in the strain hardening regime.

52
which precludes their use as micromechanically based constitutive models or as lower scale
models in multi-scale analysis.

“Modified secant models” (Ponte Castañeda and Suquet, 1998) that use second order
approximations for hσeqv i(p) have been found to be highly suitable for materials charac-
terization of elastoplastic composites, where they have shown excellent agreement with
predictions from multi-particle unit cell models (Segurado et al., 2002a) and experiments.
Modified secant models have also proved quite flexible. For example, a method of this type
was adapted to incorporate a nonlocal plasticity model for the matrix (Hu et al., 2005) in
order to study particle size effects on the macroscopic yield behavior of MMCs.

Incremental Methods for Elastoplastic Composites

Incremental mean-field methods can be formulated on the basis of strain and stress rate
tensors for elastoplastic phases (p) , dhεi(p) and dhσi(p) , which can be expressed as
(p) (p)
dhεi(p) = Āt dhεi + β̄ t dT
(p) (p)
dhσi(p) = B̄t dhσi + κ̄t dT . (2.100)

by analogy to eqn. (2.6). Here dhεi stands for the macroscopic strain rate tensor, dhσi for
(p) (p)
the macroscopic stress rate tensor, and dT for a homogeneous temperature rate. Āt , β̄ t ,
(p) (p)
B̄t , and κ̄t are instantaneous phase averaged strain and stress concentration tensors,
respectively. For elastic inhomogeneities embedded in an elastoplastic matrix40 , the overall
instantaneous “tangent stiffness” tensor of the elastoplastic two-phase composite can be
written in terms of the phase properties and the instantaneous concentration tensors as
(m) (m)
E∗t = E(i) + (1 − ξ)[Et − E(i) ]Āt
(m) (m)
= [C(i) + (1 − ξ)[Ct − C(i) ]B̄t ]−1 . (2.101)

Expressions of this type are closely related to eqns. (2.8) and (2.9); many of the general
relations given in Section 2.1 have equivalents in the incremental mean-field framework.

Using the Mori–Tanaka formalism of Benveniste (1987), the instantaneous matrix con-
centration tensors can be written as
(m) (m) (m)
Āt = {(1 − ξ)I + ξ[I + St Ct (E(i) − Et )]−1 }−1
(m) (m) (m) (m)
B̄t = {(1 − ξ)I + ξE(i) [I + St Ct (E(i) − Et )]−1 Ct }−1 , (2.102)

in direct analogy to eqn. (2.43). Expressions for the instantaneous thermal concentration
tensors and instantaneous coefficients of thermal expansion can also be derived by anal-
ogy to the corresponding thermoelastic relations, e.g., eqns. (2.11) and (2.16). Equations
(2.102) employ the instantaneous Eshelby tensor St , which depends on the current state of
(m)
the (elastoplastic) matrix material, Et . Because the latter tensor typically shows a low
symmetry, St must in general be evaluated numerically.

40
Analogous expressions can be derived for elastoplastic inhomogeneities in an elastic matrix or, in
general, for composites containing any required number of elastoplastic phases.

53
 By replacing rates such as dhεi(p) and dhσi(p) with the corresponding finite increments,
∆hεi(p) and ∆hσi(p) , respectively, formulations of eqns. (2.100) to (2.102) can be obtained
that are suitable for implementation as micromechanically based constitutive models at
the integration point level within Finite Element codes. The resulting incremental Mori–
Tanaka (IMT) methods make no assumptions on the overall yield surface and the overall
flow potential, the effective material behavior being entirely determined by the incremental
mean-field equations and the constitutive behavior of the phases. As a consequence, map-
ping of the stresses onto the yield surface cannot be handled at the level of the homogenized
material and radial return mapping must be applied to the matrix at the microscale in-
stead. This, in turn, implies that the constitutive equations describing the overall behavior
cannot be integrated directly (as is the case for homogeneous elastoplastic materials), and
iterative algorithms are required. For example, Pettermann (1997) used an implicit Euler
scheme in an implementation of an incremental Mori–Tanaka method as a user supplied
material routine (UMAT) for the Finite Element code ABAQUS (Simulia, Pawtucket, RI).
Extended versions of such algorithms can also handle thermal expansion effects41 and tem-
perature dependent material parameters.

Incremental mean-field models of the type discussed above tend to overestimate the
macroscopic strain hardening in the post-yield regime to such an extent that their prac-
tical applicability is rather limited, especially for matrix dominated deformation modes,
compare, e.g., the discussions by Gilormini (1995) and Suquet (1997). Later developments
have involved the use of tangent operators that reflect the symmetry of the elastoplastic
phase, e.g., “isotropized” operators42 for statistically isotropic materials such as particle-
reinforced composites (Bornert, 2001) together with algorithmic modifications (Doghri and
Ouaar, 2003; Doghri and Friebel, 2005). These improvements have succeeded in markedly
reducing the overprediction of the strain hardening behavior by incremental mean-field
models such as IMTs for particle-reinforced composites, making them attractive candidates
for use at the lower length scale in hierarchical and multi-scale models of ductile matrix
composites. A pragmatic extension of spectral isotropization schemes to fiber-reinforced
materials was introduced by Selmi et al. (2011) and further developments aimed at cir-
cumventing isotropization were proposed by Brassart et al. (2012), Lahellec and Suquet
(2013) and Wu et al. (2013). Modified IMTs were also successfully extended to the large
strain regime (Huber et al., 2007). An alternative approach to handling the above prob-
lems, based on transforming the elastic Eshelby tensor, was proposed by Peng et al. (2016).

Incremental mean-field models have typically used eqn. (2.99) for evaluating the phase
averaged equivalent stresses, which, accordingly, tend to be underestimated. The improved
estimators for hσeqv i(p) used in modified secant models are not suitable for incremental
methods, for which an empirical correction was proposed by Delannay et al. (2007); in
addition an approach due to Berbenni (2021) is applicable. Furthermore, because they
assume elastoplastic phases to yield as a whole once the phase averaged equivalent stress
41
Elastoplastic inhomogeneous materials such as metal matrix composites typically show a hysteretic
thermal expansion response, i.e., the “coefficients of thermal expansion” are not material properties in the
strict sense. This dependence of the thermal expansion behavior on the instantaneous mechanical response
requires special treatment within the IMT framework, compare Pettermann (1997).
42
For a discussion of a number of issues pertaining to the use of “isotropic” versus “anisotropic” tangent
operators for macroscopically isotropic, elastoplastic materials see, e.g., Chaboche and Kanouté (2003).

54
exceeds the yield stress of the elastoplastic constituent, mean-field approaches predict sharp
transitions from elastic to plastic states instead of the actual, gradual progress of yielded
regions at the microscale.

Other mean-field schemes can also be combined with secant or incremental approaches
to obtain descriptions for elastoplastic inhomogeneous materials, the most important ap-
plication having been the use of classical self-consistent schemes for describing the elasto-
plastic behavior of polycrystalline materials, see, e.g., Hill (1965a), Hutchinson (1970) and
Berveiller and Zaoui (1981). In the case of incremental methods what has to be done is
replacing elastic, Eshelby and concentration tensors with the corresponding instantaneous
tensors for each elastoplastic phase in expressions such as eqns. (2.52), (2.86), (2.50) or
(2.85). The weaknesses and strengths of such procedures are closely related to those dis-
cussed above for Mori–Tanaka-based methods. In addition, mean-field approaches have
been employed for obtaining estimates of the nonlinear response of inhomogeneous ma-
terials due to microscopic damage or to combinations of damage and plasticity, see, e.g.,
Tohgo and Chou (1996) and Guo et al. (1997).

As an alternative to directly extending mean-field theories into secant or incremental

plasticity, they can also be combined with the Transformation Field Analysis of Dvorak
(1992) in order to obtain descriptions of the overall behavior of inhomogeneous materi-
als in the plastic range, see, e.g., Plankensteiner (2000). Such approaches may markedly
overestimate the strain hardening of elastoplastic composites because they use elastic ac-
commodation of microstresses and strains throughout the loading history. Chaboche et al.
(2001) gave a discussion of modifications aimed at improving this behavior of the Trans-
formation Field Analysis see Chaboche et al. (2001).

For a comparison of the predictions of a number of analytical modeling approaches for

elastoplastic fiber-reinforced composites see, e.g., Wang and Huang (2018).

2.10 Mean-Field Methods for Conduction and Diffu-

sion Problems
The mathematical descriptions of, on the one hand, steady-state thermoelasticity and, on
the other hand, thermal conduction as well as other diffusion-type problems for heteroge-
neous materials are based on Poisson equations, share many common features, and can be
attacked using similar techniques. Table 2.1 lists the principal variables of these two sets
of problems, emphasizing the analogies between them. A number of other steady-state
diffusion phenomena are mathematically equivalent to heat conduction (Hashin, 1983),
among them electrical conduction and the diffusion of moisture. The differential equa-
tions describing antiplane shear in elastic solids, Darcy creep flow in porous media, and
equilibrium properties such as overall dielectric constants and magnetic permeabilities are
also of the Poisson type and thus equivalent to diffusive transport problems. For further
discussions see Torquato (2002).

An important difference between elasticity and conduction or diffusion problems con-

cerns the orders of the tensors involved, which is lower in the latter case. The displacements

55
 Table 2.1: Principal variables in steady state elasticity and heat conduction problems.

physical problem elasticity thermal conduction

field variable displacement field temperature field
(potential) u [m] T [K]
generalized gradient strain field temperature gradient field
tensor (intensity) ε [] d [Km−1 ]
generalized flux stress field heat flux field
tensor σ [Pa] q [Wm−2 ]
generalized property elasticity thermal conductivity
tensor E [Pa] K [Wm−1 K−1 ]
compliance thermal resistivity
C [Pa−1 ] R [mKW−1 ]

u are vectors whereas the temperatures T are scalars, stresses σ and strains ε are tensors
of order 2, whereas the heat fluxes q and thermal gradients d are (physical) vectors, and
the elasticity tensor E as well as its inverse, the compliance tensor C = E−1 , are of or-
der 4, whereas the conductivity tensor K and its inverse, the resistivity tensor R = K−1 ,
are of order 2. The differences in the orders of the tensors directly affect the number of
parameters required for describing the pertinent material property tensors as well as their
symmetry properties (Nye, 1957). For example, cubic geometrical symmetry gives rise
to macroscopic cubic symmetry in elasticity (with three independent elastic moduli) but
isotropic behavior in thermal conduction.

The phase averages of the temperature gradient d and the heat flux q mentioned in
table 2.1 are defined as
Z Z
1 1
hdi = (p) d(x) dΩ = (p) T (x) n(x) dΓ
Ω Ω(p) Ω Γs
Z Z
1 1
hqi = (p) q(x) dΩ = (p) x q(x) dΓ (2.103)
Ω Ω(p) Ω Γs

in analogy to eqns. (1.4) and (1.16). The Hill–Mandel condition in thermal conduction
becomes Z
hq di = qT (x) d(x) dΩ = hqiT hdi ,
T
(2.104)
Ω

in direct analogy to eqn. (1.7).

Dilute Inhomogeneities
In conduction and diffusion-type problems the effects of dilute inclusions can be described
via the rank 2 depolarization tensor (also referred to as “shape tensor” or “diffusion Es-
helby tensor”), S. This tensor, introduced by Fricke (1924), is directly analogous to the
“mechanical” Eshelby tensor, S, discussed in Section 2.2. In the case of spheroidal inho-
mogeneities embedded in an isotropic matrix the depolarization tensor depends only on
the formers’ aspect ratio, see, e.g., Hatta and Taya (1986) and Torquato (2002), whereas

56
for matrices of lower symmetry their material behavior also comes into play, see, e.g., Gi-
raud et al. (2007) for transversally isotropic matrices. For diffusion problems the Eshelby
property takes the form of constant fluxes and constant gradients within dilute ellipsoidal
inhomogeneities subjected to uniform far-field thermal loads. Dilute gradient and flux
(i) (i)
concentration tensors, Adil and Bdil , can be obtained by analogy to eqns. (2.31) and (2.32)
as
(i)
Adil = [I + SR(m) (K(i) − K(m) )]−1
(i)
Bdil = [I + K(m) (I − S)(R(i) − R(m) )]−1 , (2.105)
providing the basis for non-interacting approximations applicable to dilute inhomogeneity
volume fractions. I stands for the rank 2 identity tensor.

Equivalents to many of the general relations for elasticity problems discussed in Section
2.1 hold for diffusion-type problems, typical examples being the expressions for the effective
conductivity and resistivity tensors in terms of the gradient and flux concentration tensors,
K∗ = ξK(i) Ā(i) + (1 − ξ)K(m) Ā(m)
R∗ = ξR(i) B̄ (i) + (1 − ξ)R(m) B̄ (m) , (2.106)
which are directly related to eqns. (2.8) and (2.9), and the linkages between different
concentration tensors,
Ā(p) = R(p) B̄ (p) K∗ and B̄ (p) = K(p) Ā(p) R∗ , (2.107)
which correspond to eqns. (2.14).

Effective-Field Approaches
Analogous methods to the effective-field and effective-medium approaches introduced in
Sections 2.3, 2.5 and 2.6 have been developed for the conduction and/or diffusion behavior
of non-dilute inhomogeneous materials, see, e.g., Hatta and Taya (1986), Miloh and Ben-
veniste (1988), Dunn and Taya (1993), Chen (1997) as well as Torquato (2002). In the
case of Mori–Tanaka methods the phase concentration tensors can be evaluated from their
dilute equivalents, eqns. (2.105), as
(m) (i) (i) (i) (i)
ĀMT = [(1 − ξ)I + ξAdil ]−1 ĀMT = Adil [(1 − ξ)I + ξAdil ]−1
(m) (i) (i) (i) (i)
B̄MT = [(1 − ξ)I + ξBdil ]−1 B̄MT = Bdil [(1 − ξ)I + ξBdil ]−1 (2.108)
in direct analogy to eqns. (2.41) and (2.42).

The mean-field interpretation of the Maxwell scheme in thermal conduction leads to

the relations
h i−1
∗
KMX = K(m) + ξ(K(i) − K(m) ) I + (Si − ξSE )R(m) (K(i) − K(m) ) , (2.109)

with the shape tensor SE accounting for the shape of the inhomogeneous region according
to fig. 2.3 and Si for the shape of the inhomogeneities, compare eqn. (2.48). The Hashin–
Shtrikman estimates of Ponte Castañeda and Willis (1995) can be written as
 (i)
−1 −1
∗
KPW = K(m) + ξ(K(i) − K(m) )Adil − Sd R(m) , (2.110)

57
where Sd serves for describing the ellipsoidal arrangement of inhomogeneities by analogy
to eqn. (2.50). This scheme was extended to nonaligned inhomogeneities by Duan et al.
(2006).

Effective-Medium Approaches
The classical self-consistent scheme for thermal conduction, which is often referred to as
the symmetrical Bruggemann (1935) method, can be denoted in the form

Kn+1 = K(m) + ξ[K(i) − K(m) ][I + Sn(p,∗) Rn (K(i) − Kn )]−1

 −1
Rn+1 = Kn+1 , (2.111)

which is clearly equivalent to eqn. (2.53). The differential scheme for the effective conduc-
tivity tensor can be written as
∗
dKD 1 (i,∗)
= [K(i) − KD
∗
] Adil . (2.112)
dξ 1−ξ
For a number of cases analytical solutions are available, see, e.g., Phan-Thien and Pham
(2000). Otherwise, the above initial value problem can be solved numerically in analogy
to eqn. (2.54).

Nonaligned Reinforcements
For conduction problems the equivalents of “extended” Mori–Tanaka models for nonaligned
reinforcements, compare, e.g., eqns. (2.66), take the form
(i) (i)
∗
KMT = K(m) + ξ (K(i) − K(m) )Adil [(1 − ξ)I + ξ Adil ]−1
(i) (i)
R∗MT = R(m) + ξ (R(i) − R(m) )Bdil [(1 − ξ)I + ξ Bdil ]−1 . (2.113)

The corresponding non-dilute phase flux concentration tensors are

(i)∠
B̄ (m) MT
= [ξ (m) I + ξ (i) hhBdil ii]−1
(i)∠ (i)∠
B̄MT (ϕ, ψ, θ) = Bdil B̄ (m) MT
, (2.114)

compare eqns. (2.63) to (2.70), the gradient concentration tensors being obtained by anal-
ogy.

Similarly, the other methods and relationships discussed in Section 2.6 have analogies
in modeling conduction or diffusion in inhomogeneous materials.

It is worth noting that, due to the lower ranks of the tensors involved in diffusion-type
problems, the latter tend to be easier to handle than mechanical ones (for materials that
show orthotropic or higher symmetry the tensors K, R, Ā, B̄, S, H and N are diagonal
tensors). Extensive discussions of diffusion-type problems in inhomogeneous media can be
found, e.g., in Markov (2000), Torquato (2002) and Milton (2002). Analogous mean-field
descriptions can also be devised for a number of coupled problems, such as the electrome-
chanical behavior of inhomogeneous materials with at least one piezoelectric constituent,
see, e.g., Huang and Yu (1994).

58
Non-Ellipsoidal Inhomogeneities
Diffusive transport in composites containing inhomogeneities of general shape can be de-
scribed in terms of resistivity and conductivity contribution tensors,
1 1
H = (R∗ − R(m) ) and N = (K∗ − K(m) ) , (2.115)
ξ ξ

by direct analogy to the contribution tensor formalism in elasticity, eqns. (2.19) to (2.76).
Dilute contribution tensors can again be obtained from standard mean-field expressions in
the case of ellipsoidal inhomogeneities and from numerical models for other reinforcement
shapes; in the latter case the linearly independent thermal load cases are required. Links
between, on the one hand, resistivity and conductivity contribution tensors and, on the
other hand, flux and gradient concentration tensors are provided by direct equivalents to
eqns. (2.20).

Multi-Phase Composites
The diffusion and conduction behavior of composites reinforced by multiple inhomogeneity
phases can be handled in direct analogy their elastic responses, compare section 2.8. Since
the tensors involved in the models are of rank 2 rather than rank 4, however, difficulties such
as the prediction of non-symmetric effective tensors by Mori–Tanaka methods in elasticity,
do not occur in conduction.

Interphases and Interfacial Conductances

The effects of finite interfacial conductances or of interphases, i.e., coatings of finite thick-
ness, on the effective conduction behavior of materials reinforced by ellipsoidal fibers or
particles can be studied by approaches based on equivalent uniform inhomogeneities in
analogy to Section 2.8.2. Exact expressions for the dilute phase concentration tensors in
the core inhomogeneity and the coating were given by Kerner (1956) for spheres and by
Hervé-Luanco and Joannès (2016) for infinitely long cylinders. These allow a two-step
treatment in analogy to Section 2.8.2.

If position-dependent interfacial properties or non-ellipsoidal inhomogeneities combined

with interfacial effects are involved, dilute inhomogeneity concentration tensors that are
averaged over the inhomogeneities can be employed (Duschlbauer, 2004). Appropriate
(i,r) (i,r)
averaged “replacement” dilute inhomogeneity concentration tensors, Ādil and B̄dil , can
be evaluated from high-resolution numerical models involving a single inhomogeneity. In
conductivity, three linearly independent load cases must be evaluated for a given dilute
configuration, e.g., via the Finite Element method. Using eqn. (5.8), the corresponding
average gradients and fluxes in the phases can then be extracted, and the replacement
concentration tensors evaluated.
(i,r) (i,r)
In order to achieve a workable formulation, in addition to evaluating Ādil and/or B̄dil ,
the “physical” conductivity and resistivity tensors of the inhomogeneities, K(i) and R(i) ,
must be substituted by suitable “replacement” conductivity and resistivity tensors that
account for the presence of the interface and are defined as

59
 1 (i,r)
K(i,r) = K(m) + ∗
(Kdil − K(m) )[Ādil ]−1
ξdil
1 (i,r)
R(i,r) = R(m) + (R∗ − R(m) )[B̄dil ]−1 , (2.116)
ξdil dil
∗
where Kdil and R∗dil are the (numerically evaluated) effective conductivity and resistivity
tensors of the dilute single-inhomogeneity configurations and ξdil is the corresponding re-
inforcement volume fraction. These replacement conduction tensors ensure consistency
within the mean-field framework by enforcing the equivalents of eqns. (2.9) to be fulfilled.
The replacement tensors can then be inserted in lieu of the “standard” concentration and
conduction tensors into the mean-field expressions given above in order to obtain estimates
for composites reinforced by non-ellipsoidal inhomogeneities (Duschlbauer, 2004; Nogales
and Böhm, 2008). Analogous replacement tensor schemes can also be set up for the elastic
behavior.

Mean-field models employing replacement tensors following eqn. (2.116), are not suit-
able for studying non-ellipsoidal inhomogeneities with volume fractions closely approaching
unity because in such configurations interfaces between inhomogeneities and matrix cease
to make physical sense.

In the case of finite interfacial conductances the “equivalent inhomogeneity” or “re-

placement inhomogeneity” must account for the effects of interfacial temperature jumps,
see the discussion by Duschlbauer (2004). Replacement inhomogeneity formalisms can eval-
uate the effects of such interfacial barrier resistances on the macroscopic conductivity of
composites for more complex geometries than “standard methods”, such as the well-known
model of Hasselman and Johnson (1987), an example being polyhedral reinforcements hav-
ing inhomogeneously distributed interfacial conductances (Nogales and Böhm, 2008).

The above groups of methods — in agreement with numerical models for composites
with imperfect interfaces (compare Section 6.9) and the pertinent bounds (Torquato and
Rintoul, 1995) — predict that finite interfacial conductances can give rise to marked size ef-
fects in the macroscopic conductivities. Since interfacial resistivities can hardly be avoided,
below some critical size even highly conductive reinforcements accordingly fail to improve
the overall conductivity of matrix–inclusion composites of given matrix behavior.

Cross-Property Relationships
Finally, it is worth mentioning that a number of authors have studied cross-property rela-
tions, compare, e.g., Gibiansky and Torquato (1996), and cross-property bounds, see, e.g.,
Sevostianov and Kachanov (2002), that link the elastic and conduction behaviors of given
microgeometries.

60
Chapter 3

Bounding Methods

Whereas mean-field methods, unit cell approaches and embedding strategies can typically
be used for both homogenization and localization tasks, bounding methods are limited to
homogenization. The emphasis in this chapter again is put on materials consisting of two
perfectly bonded constituents.

Rigorous bounds on the overall elastic properties of inhomogeneous materials are typi-
cally obtained from appropriate variational (minimum energy) principles. Only outlines of
bounding methods are given here; formal treatments were provided, e.g., by Nemat-Nasser
and Hori (1993), Ponte Castañeda and Suquet (1998), Markov (2000), Bornert (2001),
Gross and Seelig (2001), Torquato (2002), Milton (2002) and Parnell and Calvo-Jurado
(2015).

3.1 Classical Bounds

Hill Bounds
Classical expressions for the minimum potential energy and the minimum complementary
energy in combination with uniform stress and strain trial functions lead to the simplest
variational bounding expressions, the upper bounds of Voigt (1889) and the lower bounds
of Reuss (1929). In their tensorial form (Hill, 1952) they are known as the Hill bounds (or
Voigt–Reuss–Hill bounds) and can be written as
hX i−1 X
E∗H− = V (p) C(p) ≤E≤ V (p) E(p) = E∗H+ . (3.1)
(p) (p)

These bounds, while universal and very simple, do not contain any information on the
microgeometry of an inhomogeneous material beyond the phase volume fractions, so that
the macroscopic elastic symmetry of the bounding expressions, eqns. (3.1), depends solely
on the elastic symmetries of the constituents. Hill bounds are too slack for most practical
purposes43 . but, in contrast to the Hashin–Shtrikman and higher-order bounds, they also
hold for volume elements that are too small to be proper RVEs.
43
The bounds on the Young’s moduli obtained from eqn. (3.1) are equivalent to Voigt and Reuss ex-
pressions in terms of the corresponding phase moduli only if the constituents have Poisson’s ratios that
give rise to equal Poisson contractions. Due to the homogeneous stress and strain assumptions used for
obtaining the Hill bounds, the phase strain and stress concentration tensors corresponding to them are
(p) (p)
ĀV = I and B̄R = I, respectively.

61
Hashin–Shtrikman-Type Bounds
Considerably tighter bounds on the macroscopic elastic responses of inhomogeneous mate-
rials can be obtained from a variational formulation due to Hashin and Shtrikman (1961)
which is based on phase-wise uniform stress polarization tensors, eqn. (2.56). For macro-
scopically isotropic composites containing spherical inhomogeneities these bounds were
originally stated in terms of the effective bulk modulus K ∗ and the effective shear mod-
ulus44 G∗ . Analogous expressions for (macroscopically transversally isotropic) composites
reinforced by aligned continuous fibers were given by Hashin (1972) and an extension to
composites containing transversally isotropic aligned continuous fibers can be found in
Hashin (1983). For the original Hashin–Shtrikman bounds the isotropic constituents were
assumed to be elastically “well-ordered”, with (K (i) − K (m) )(G(i) − G(m) ) > 0.

Walpole (1966), on the one hand, removed the restriction to well-ordered constituent
properties by proposing generalized reference materials and, on the other hand, provided
tensorial expressions for Hashin–Shtrikman-type bounds in the form
hX ihX i−1
∗ (p) (p) (p,−) (p) (p,−)
EHSW− = ξ E T̄dil ξ T̄dil
(p) (p)
hX ihX i−1
(p,+) (p,+)
E∗HSW+ = ξ (p) E(p) T̄dil ξ (p) T̄dil . (3.2)
(p) (p)

Equations (3.2) are closely related to the Hashin–Shtrikman tensor, eqn. (2.89). The par-
(p,−) (p,+)
tial strain concentration tensors, T̄dil and T̄dil , which pertain to a phase (p) embedded
in the lower or the upper reference material, described by elasticity tensors E− and E+ ,
respectively, can be evaluated from eqn. (2.31) or an equivalent. Willis (1977) extended
the Hashin–Shtrikman-type bounds to elastically anisotropic macroscopic behavior due to
anisotropic constituent behavior and/or aligned ellipsoidal reinforcements.

In order to obtain proper bounds, the lower and upper comparison media must be
at least as compliant or as stiff, respectively, as any of the constituents, which may be
written as E− ≤ E(p) and E+ ≥ E(p) , compare Bornert (2001). For “typical” two phase-
composites, which consist of stiff reinforcements in a compliant matrix, well-orderedness
implies E− = E(m) and E+ = E(i) . The two-phase version of eqns. (3.2),
 (m,−) (i,−)  (m,−) (i,−) −1
E∗HS− = (1 − ξ)E(m) T̄dil + ξE(i) T̄dil (1 − ξ)T̄dil + ξ T̄dil
 (m,+) (i,+)  (m,+) (i,+) −1
E∗HS+ = (1 − ξ)E(m) T̄dil + ξE(i) T̄dil (1 − ξ)T̄dil + ξ T̄dil , (3.3)
44
Whereas engineers tend to describe elastically isotropic material behavior by Young’s moduli and
Poisson’s ratios (which can be measured from uniaxial tensile experiments in a relatively straightforward
way), many homogenization expressions are best formulated in the bulk and shear moduli, which are
directly linked to the hydrostatic and deviatoric responses, compare eqn. (2.96). For obtaining bounds on
the effective Young’s modulus E ∗ from the results on K ∗ and G∗ see Hashin (1983), and for bounding
expressions on the Poisson’s ratios ν ∗ see Zimmerman (1992). The latter procedure can be extended
to macroscopically transversely isotropic materials, where, however, it tends to give rise to rather slack
bounds on some of the engineering moduli, especially on the transverse Poisson’s ratios.

62
can be transformed for this case into
h ih i−1
(i,m) (i,m)
E∗HS− = (1 − ξ)E(m) + ξE(i) T̄dil (1 − ξ)I + ξ T̄dil
h i−1
= E(m) + ξ (E(i) − E(m) )−1 + (1 − ξ)S(i,m) C(m)
h ih i−1
(m,i) (m,i)
E∗HS+ = (1 − ξ)E(m) T̄dil + ξE(i) (1 − ξ)T̄dil + ξI
h i−1
= E(i) + (1 − ξ) (E(m) − E(i) )−1 + ξS(m,i) C(i) . (3.4)

The first of the above equations is equivalent to the Mori–Tanaka expression, eqn. (2.44),
and the second one is a “color inverted” version of it, i.e., it corresponds to a microstruc-
ture of inhomogeneity volume fraction ξ in which the materials of matrix and inhomogene-
ity have been interchanged. Thus, for two-phase materials with well-ordered constituent
elasticities the Hashin–Shtrikman bounds can be evaluated via Mori–Tanaka estimates,
compare Weng (1990) or Gross and Seelig (2001). In the case of multi-phase composites
Mori–Tanaka models correspond to the upper or lower Hashin–Shtrikman-type bound if
the matrix is the most compliant or the stiffest constituent. However, in general the above
“shortcut” cannot be used for multi-phase materials.

For composites with matrix–inclusion topology eqns. (3.2) can only be used if all in-
(p,−) (i,−)
homogeneity phases have identical shapes, so that all pertinent T̄dil = T̄dil and all
(p,+) (i,+) (m,−) (m,+)
T̄dil = T̄dil are identical; the T̄dil and T̄dil are then also evaluated for the same
aspect ratio, as is the case for eqns. (3.3) and (3.4). Equations (3.2) to (3.4) deliver valid
bounds whenever the comparison media fulfill the conditions E− ≤ E(p) and E+ ≥ E(p) for
all constituents (p) . Choosing the largest possible E− and the smallest possible E+ leads
to the tightest, optimal bounds. Using reference media of vanishing or infinite stiffness
recovers the lower and upper Hill bounds, respectively.

Explicit constructions for optimal comparison media pertinent to the two-phase case
were given by Walpole (1966) for materials in which constituents and macroscopic behavior
are isotropic. Parnell and Calvo-Jurado (2015) provide an in-depth discussion of the evalu-
ation of Hashin–Shtrikman-type bounds for composites consisting of aligned, transversally
isotropic reinforcements in a transversally isotropic matrix.

For multi-phase materials in which one of the phases is the most compliant and another
the stiffest one, the choice of E− and E+ is obvious. In all other situations generalized,
“synthetic” reference media must be provided. Useful tensors E− and E+ for general
multi-phase materials can typically be constructed from the minima and maxima of the
eigenvalues of the phase elasticity tensors45 E(p) (Parnell and Calvo-Jurado, 2015). How-
ever, reference media generated this way may be non-optimal.

If at least one inhomogeneity phase shows vanishing stiffness, i.e., for porous materials,
the above considerations lead to a trivial lower bound. Similarly, an infinite upper bound
45
If all phases are isotropic this implies constructing the comparison material from the minima and
maxima of the shear and bulk moduli G(p) and K (p) . If some or all of the phases are transversally
isotropic, E− and E+ can be generated from the minima and maxima of the five phase-level Hill moduli
(p) (p) (p) (p) (p) (p) (p) (p)
k (p) = KT , l(p) = 2KT νA , m(p) = GT , n(p) = EA + 4KT (νA )2 and p(p) = GA .

63
results if at least one of the constituents is rigid. Composites reinforced by randomly
oriented, isotropic fibers or platelets follow the Hashin–Shtrikman bounds for macroscop-
ically isotropic materials. A discussion of Hashin–Shtrikman-type bounds on the elastic
responses of composites with more general fiber orientation distributions can be found in
Eduljee and McCullough (1993).

Hashin–Shtrikman-type variational formulations can also be employed for generating

bounds for more general phase arrangements. Evaluating the stress polarizations for “com-
posite regions” consisting of inhomogeneities embedded in a matrix gives rise to Hervé–
Stolz–Zaoui bounds (Hervé et al., 1991). When complex phase patterns are considered
(Bornert, 1996; Bornert et al., 1996) numerical methods must be used for evaluating the
polarization fields. If exact solutions are available for the concentration tensors of non-
uniform inhomogeneities, as is the case for coated spherical and cylindrical reinforcements,
compare Section 2.8.2, they can be directly inserted into eqns. (3.3) or (3.4) to provide
bounds.

Hashin–Shtrikman-type bounds can also be derived for simple periodic phase arrange-
ments, see, e.g., Nemat-Nasser and Hori (1993). Among the bounding methods for such
phase arrangements are those of Bisegna and Luciano (1996), which uses approximate
variational principles evaluated from periodic unit cells via Finite Element models, and of
Teply and Dvorak (1988), which evaluates bounds for the elastoplastic behavior of fiber-
reinforced composites with periodic hexagonal phase arrangements.

Hashin–Shtrikman-type bounds are sharp, i.e., they are the tightest bounds that can
be given for the geometrical information used, viz., volume fraction and overall symmetry,
corresponding to two-point correlations. Bounds of this group tend to be rather slack,
however, i.e., the lower and upper bounds are relatively far apart, especially for elevated
phase contrasts.

3.2 Improved Bounds

A considerable number of statistical descriptors have been used for characterizing the phase
arrangements of inhomogeneous materials, see, e.g., Torquato (2002). An important group
of them are n-point probability functions, which are obtained by randomly placing sets
on n points into the microstructure and recording which phases they end up in. For one-
point correlations this procedure provides the phase volume fractions. When two-point
probabilities are evaluated for two-phase composites, on the one hand, the probabilities of
finding both points in the matrix or in the inhomogeneities, Smm (r) and Sii (r), respectively,
or in different phases, Smi (r), can be obtained, compare fig. 3.1. On the other hand, the
vector r between the two points provides information on the anisotropy of the arrangement
and on the dependence of the Spq on the distance between the points. The Hill bounds are
closely related to one-point and Hashin–Shtrikman bounds to two-point probabilities.

Three-point probability functions Spqr (r1 , r2 ) can provide additional information, e.g.,
on inhomogeneity shapes, sizes and clustering. Most work on three-point correlations has
involved isotropic configurations in two or three dimensions, giving rise to three-point

64
 r2 sm
siii sim
r1
smm r

r si
si

r2 r2
smmm smim
r1 r1
sii

sm r1 siim r
2

Figure 3.1: Sketch of sampling for one-point (red), two-point (blue) and three-point (green)
probability functions in a two-dimensional matrix–inclusion composite; adapted from Gillman
et al. (2015).

probabilities such as Siii (r1 , r2 ), where ri stands for the length of the vector ri . They form
the basis of three-point bounds, which use more complex trial functions than Hashin–
Shtrikman bounds and, accordingly, require additional statistical information on the phase
arrangement. The resulting “improved bounds” are significantly tighter than the two-point
Hashin–Shtrikman-type expressions.

Three-point bounds for statistically isotropic two-phase materials were developed by

Beran and Molyneux (1966), Milton (1981) as well as Phan-Thien and Milton (1983) and
can be formulated in such a way that the information on the phase arrangement statis-
tics is contained in two three-point microstructural parameters, η(ξ) and ζ(ξ), which take
the form of multiple integrals over the three-point correlation functions, see, e.g., Torquato
(2002). These correlation functions and the resulting parameters ζ(ξ) and η(ξ) can in prin-
ciple be obtained for any given two- or three-dimensional microstructure that is statistically
homogeneous, but in practice their evaluation may be demanding and require sophisticated
numerical algorithms, see Gillman et al. (2015). Analytical expressions or tabulated data
for ζ(ξ) and η(ξ) in terms of the reinforcement volume fraction ξ are available for a number
of generic microgeometries of practical importance, among them statistically homogeneous
isotropic materials containing identical, bidisperse and polydisperse impenetrable (“hard”)
spheres (that describe matrix–inclusion composites) as well as monodisperse overlapping
spheres (“Boolean models” that can describe many interwoven phase arrangements), and
statistically homogeneous transversally isotropic materials reinforced by impenetrable or
overlapping aligned cylinders. References to a number of expressions for η and ζ applicable
to some two-phase composites can be found in Section 4.1, where results from mean-field
and bounding approaches are compared. Recently, three-point parameters were published

65
for randomly positioned particles having the shapes of Platonic polyhedra (Gillman et al.,
2015). Three-point bounds for multi-phase composites were discussed by Genin and Bir-
man (2009). For reviews of higher-order bounds for elastic (as well as other) properties
of inhomogeneous materials see, e.g., Quintanilla (1999), Torquato (1991) and Torquato
(2002).

Improved bounds can provide highly useful information for low and moderate phase
contrasts (as typically found in technical composites), but even they become rather slack
for elevated phase contrasts and inhomogeneity volume fractions exceeding, say, ξ (i) = 0.3.
It is worth noting that, due to the additional geometrical information available, n-point
bounds are always nested within (n − 1)-point bounds. This implies that, for a given
inhomogeneous material, the three-point bounds are tighter than Hashin–Shtrikman-type
bounds which, in turn, are nested within the Hill bounds.

3.3 Bounds on Nonlinear Mechanical Behavior

Equivalents to the Hill bounds in elasticity, eqn. (3.1), were introduced for nonlinear inho-
mogeneous materials by Bishop and Hill (1951). For polycrystals the nonlinear equivalents
to Voigt and Reuss expressions are usually referred to as Taylor (1938) and Sachs (1928)
bounds, respectively.

In analogy to mean-field estimates for nonlinear material behavior, nonlinear bounds

are typically obtained by evaluating sequences of linear bounds. Such bounds typically
describe responses to loads that are radial in stress space and usually pertain to uniaxial
tensile tests. Talbot and Willis (1985) extended the Hashin–Shtrikman variational princi-
ples to obtain one-sided bounds (i.e., upper or lower bounds, depending on the combination
of constituents) on the nonlinear mechanical behavior of inhomogeneous materials.

An important development took place with the derivation of a variational principle by

Ponte Castañeda (1992), which allows upper bounds on specific stress-strain responses of
elastoplastic inhomogeneous materials to be generated on the basis of upper bounds on the
elastic tensors46 . It uses a sequence of inhomogeneous reference materials, the properties
of which have to be obtained by optimization procedures for each strain level. Essentially,
the variational principle guarantees the best choice for the comparison material at a given
load. The Ponte Castañeda bounds are closely related to mean-field approaches using im-
proved secant plasticity methods, compare Section 2.9. For higher-order bounds on the
nonlinear response of inhomogeneous materials see, e.g., Talbot and Willis (1998).

The study of bounds — like the development of improved estimates — for the overall
nonlinear mechanical behavior of inhomogeneous materials has been an active field of
research during the past decades, see the reviews by Suquet (1997), Ponte Castañeda and
Suquet (1998) and Willis (2000).
46
The Ponte Castañeda bounds are rigorous for nonlinear elastic inhomogeneous materials and, on the
basis of deformation theory, are excellent approximations for materials with at least one elastoplastic con-
stituent. Applying the Ponte Castañeda variational procedure to elastic lower bounds does not necessarily
lead to a lower bound for the inelastic behavior.

66
3.4 Bounds on Conduction and Diffusion-Like Prop-
erties
Because many bounding approaches are closely related to mean-field models it is not sur-
prising that all of the bounding methods for the macroscopic elastic responses of inho-
mogeneous materials discussed in Sections 3.1 and 3.2 have direct equivalents in terms of
diffusion properties by analogy to Section 2.10. The equivalents in thermal and electrical
conduction to the elastic Hill bounds, eqn. (3.1), are known as the Wiener (1912) bounds
and take the form
hX i−1 X
(p) (p)
V R ≤K≤ V (p) K(p) . (3.5)
(p) (p)

Hashin–Shtrikman bounds for diffusive transport were developed concurrently with the
bounds for elasticity (Hashin and Shtrikman, 1962) and can be expressed in analogy to
eqns. (3.3) as
h i−1
∗
KHS− = K(m) + ξ (K(i) − K(m) )−1 + (1 − ξ)S (i,m) R(m)
h i−1
∗ (i) (m) (i) −1 (m,i) (i)
KHS+ = K + (1 − ξ) (K − K ) + ξS R (3.6)

for the case that the inclusions are more conductive than the matrix. Willis bounds on
effective conductivities can be given in direct equivalence to eqn. (3.2), compare, e.g.,
Calvo-Jurado and Parnell (2017).

Three-point bounds for diffusion properties were proposed, e.g., by Milton (1981). They
require only one of the two statistical parameters used in bounding the elastic behavior,
viz., ζ(ξ).

67
Chapter 4

Some Comparisons of Mean-Field

Estimates and Bounds

4.1 Comparisons of Mean-Field and Bounding Pre-

dictions for Effective Thermoelastic Moduli of Two-
Phase Composites
In order to show some of the basic features of the predictions that can be obtained by
different mean-field (and related) approaches and by bounding methods for the thermo-
mechanical responses of inhomogeneous thermoelastic materials, in this section selected
results on the overall elastic moduli and coefficients of thermal expansion are presented as
functions of the reinforcement volume fraction ξ. The comparisons are based on E-glass
particles or fibers embedded in an epoxy matrix, the pertinent material parameters being
listed in table 4.1. The elastic contrast of this pair of constituents is cel ≈ 21 and the
thermal expansion contrast takes a value of approximately 0.14.

Table 4.1: Constituent material parameters of the epoxy matrix and the E-glass reinforcements
used for generating figs. 4.1 to 4.9.

E[GPa] ν[ ] α[1/K]
matrix 3.5 0.35 3.6×10−5
reinforcements 74.0 0.2 4.9×10−6

Figures 4.1 and 4.2 show predictions for the overall Young’s and shear moduli of a
particle-reinforced two-phase composite using the above constituent parameters. The Hill
bounds can be seen to be very slack. A macroscopically isotropic two-phase composite be-
ing studied, the Mori–Tanaka estimates (MTM) coincide with the lower Hashin–Shtrikman
bounds (H/S LB), compare Section 3.1. The classical self-consistent scheme (CSCS) shows
a typical behavior in that it is close to one Hashin–Shtrikman bound at low volume frac-
tions, approaches the other at high volume fractions, and displays a transitional behavior
in the form of a sigmoid curve in-between.

68
 EFFECTIVE YOUNGs MODULUS [GPa]
80.0
DS
CSCS
3PE (mono/h)
GSCS
3PUB (mono/h)
3PLB (mono/h)
H/S UB
H/S LB; MTM
40.0

Hill bounds
0.0

0.0 0.2 0.4 0.6 0.8 1.0

PARTICLE VOLUME FRACTION [ ]
Figure 4.1: Bounds and estimates for the effective Young’s moduli of glass/epoxy particle-
reinforced composites as functions of the particle volume fraction.

The three-point bounds (3PLB and 3PUB) shown in figs. 4.1 pertain to impenetrable
spherical particles of equal size and use expressions for the statistical parameters η and
ζ listed by Torquato (2002), which are available for reinforcement volume fractions up to
ξ = 0.647 . As expected, these improved bounds are significantly tighter than the Hashin–
Shtrikman bounds. The three-point estimates (3PE) of Torquato (1998a), which were
EFFECTIVE SHEAR MODULUS [GPa]

DS
30.0

CSCS
3PE (mono/h)
GSCS
3PUB (mono/h)
3PLB (mono/h)
20.0

H/S UB
H/S LB; MTM
Hill bounds
10.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

PARTICLE VOLUME FRACTION [ ]
Figure 4.2: Bounds and estimates for the effective shear moduli of glass/epoxy particle-reinforced
composites as functions of the particle volume fraction.
47
This value approaches the maximum volume fraction achievable in random packings of identical
spheres, which is ξ ≈ 0.64 (Weisstein, 2000).

69
evaluated with the same choice of η and ζ, fall between the three-point bounds. For the
composite considered here the results from the generalized self-consistent scheme (GSCS),
which predicts a slightly stiffer behavior than the Mori–Tanaka method, are very close
to the lower three-point bounds even though the GSCS is not associated with monodis-
perse particle microgeometries. The predictions of the differential scheme (DS) pertain to
composites with polydisperse reinforcements and can be seen to be stiffer than either the
three-point estimates for identical spheres or the GSCS results.

Alternatively, the elastic moduli of macroscopically isotropic composites can be visual-

ized by plotting the shear modulus over the bulk modulus following Berryman and Milton
(1988). In fig. 4.3 this format is used to compare the Hashin–Shtrikman (outermost, solid
box) and three-point bounds for impenetrable, identical spheres (inner, dashed box) with a
number of estimates for a glass–epoxy composite having a particle volume fraction of ξ (p) =
0.4. The typical behavior of composites consisting of a compliant matrix reinforced by stiff
particles is shown: the three-point bounds and all estimates pertinent to matrix–inclusion
materials cluster in the corner corresponding to low bulk and shear moduli, whereas the
classical self-consistent scheme predicts clearly different responses appropriate for materials
having other phase-level topologies.

9
EFFECTIVE SHEAR MODULUS [GPa]

7 8 9 10 11 12 13 14
EFFECTIVE BULK MODULUS [GPa]
HSB MTM 3PE CSCS
3PB GSCS DS

Figure 4.3: Bounds and estimates for the effective bulk and shear moduli of a glass/epoxy particle-
reinforced composite of particle volume fraction ξ (p) = 0.4.

Predictions for the macroscopic coefficients of thermal expansion of this macroscopically

isotropic inhomogeneous material are presented in fig. 4.4. Levin’s formula, eqn.(2.12), was
combined with the Hashin–Shtrikman bounds, the three-point bounds, the generalized self-
consistent estimates and the three-point estimates for the effective bulk modulus to obtain
the corresponding bounds and estimates for the CTE. In the case of the classical self-
consistent and differential schemes eqns. (2.10) and (2.16) were used. Being based on the
same estimates for the effective bulk modulus, the results presented for the GSCS and the
Mori–Tanaka-scheme coincide with the upper Levin/Hashin–Shtrikman bounds.

70
 40.0
DS
CSCS

EFFECTIVE CTE [1/K x 10 -6 ]

3PE (mono/h)
GSCS
3PUB (mono/h)
3PLB (mono/h)
H/S UB; MTM
H/S LB
20.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

PARTICLE VOLUME FRACTION [ ]
Figure 4.4: Bounds and estimates for the effective CTEs of glass/epoxy particle-reinforced com-
posites as functions of the particle volume fraction.

Applying the constituent data given in table table 4.1 to an epoxy matrix reinforced by
continuous aligned glass fibers gives rise to transversally isotropic macroscopic behavior.
Pertinent results are presented in figs. 4.5 to 4.9 for the overall transverse Young’s moduli48 ,
EFFECTIVE YOUNGs MODULUS [GPa]
80.0

DS
CSCS
3PE (mono/h)
GSCS
3PB LB (mono/h)
3PB LB (mono/h)
H/S UB
H/S LB; MTM
40.0

Hill bounds
0.0

0.0 0.2 0.4 0.6 0.8 1.0

FIBER VOLUME FRACTION [ ]
Figure 4.5: Bounds and estimates for the effective transverse Young’s moduli of glass/epoxy
fiber-reinforced composites as functions of the fiber volume fraction.

48
The estimates and bounds for the axial Young’s moduli are indistinguishable from each other (and
(i)
from the rule-of-mixture result, EA ∗
= ξEA + (1 − ξ)E (m) , as well as the Hill upper bound given in fig. 4.5)
for the scaling used in fig. 4.5 and are, accordingly, not shown.

71
 EFFECTIVE SHEAR MODULUS [GPa]
DS

30.0
CSCS
3PE (mono/h)
GSCS
3PUB (mono/h)
20.0 3PLB (mono/h)
H/S UB
H/S LB; MTM
Hill bounds
10.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

FIBER VOLUME FRACTION [ ]
Figure 4.6: Bounds and estimates for the effective axial shear moduli of glass/epoxy fiber-
reinforced composites as functions of the fiber volume fraction.

the overall axial and transverse shear moduli, as well as the overall axial and transverse
coefficients of thermal expansion. The results for the three-point bounds shown are based
on the formalisms of Silnutzer (1972) and Milton (1981), correspond to a microgeometry
of aligned impenetrable circular cylindrical fibers of equal diameter, and use statistical
parameters evaluated by Torquato and Lado (1992) for fiber volume fractions ξ / 0.7.
The coefficients of thermal expansion were evaluated on the basis of the relations of Rosen
and Hashin (1970).
EFFECTIVE SHEAR MODULUS [GPa]

DS
30.0

CSCS
3PE (mono/h)
GSCS
3PUB (mono/h)
3PLB (mono/h)
20.0

H/S UB
H/S LB; MTM
Hill bounds
10.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

FIBER VOLUME FRACTION [ ]
Figure 4.7: Bounds and estimates for the effective transverse shear moduli of glass/epoxy fiber-
reinforced composites as functions of the fiber volume fraction.

72
 40.0
DS
CSCS
EFFECTIVE CTE [1/K x 10 -6 ] 3PE (mono/h)
3PUB (mono/h)
3PLB (mono/h)
H/S UB
H/S LB; MTM,GSCS
20.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

FIBER VOLUME FRACTION [ ]
Figure 4.8: Estimates and bounds for the effective axial CTEs of glass/epoxy fiber-reinforced
composites as functions of the fiber volume fraction.

Generally, a qualitatively similar behavior to the particle-reinforced case can be ob-

served. It is noteworthy that the overall transverse CTEs in fig. 4.9 at low fiber volume
fractions exceeds the CTEs of both constituents. Such behavior is typical for continuously
reinforced composites and is caused by the marked axial constraint enforced by the fibers.
As expected for continuously reinforced materials, there is little variation between the pre-
40.0

DS
CSCS
EFFECTIVE CTE [1/K x 10 -6 ]

3PE (mono/h)
3PUB (mono/h)
3PLB (mono/h)
H/S UB; MTM,GSCS
H/S LB
20.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

FIBER VOLUME FRACTION [ ]
Figure 4.9: Estimates and bounds for the effective transverse CTEs of glass/epoxy fiber-reinforced
composites as functions of the fiber volume fraction.

73
dictions of the different models for the axial thermal expansion behavior, fig. 4.8. The
Mori–Tanaka estimates correspond to the upper bound for the transverse CTE in analogy
to the macroscopically homogeneous case. For the axial CTE, however, the Mori–Tanaka
results agree with the lower bound, which is a consequence of the axial constraint intro-
duced by the fibers.

In figs. 4.1 to 4.7 the classical self-consistent scheme is not in good agreement with
the three-point bounds shown, because the latter explicitly correspond to matrix–inclusion
topologies. Considerably better agreement with the CSCS can be obtained by using three-
point parameters of the overlapping sphere or cylinder type (which can also describe cases
where both phases percolate, but are not as symmetrical with respect to the constituents as
the CSCS). From a practical point of view it is worth noting that despite their sophistication
higher order estimates (and improved bounds) may give overly optimistic predictions for
the overall moduli because they describe ideal composites, whereas in actual “two-phase”
materials it is practically impossible to avoid flaws such as porosity.

Obviously, for materials containing aligned, ellipsoidal inhomogeneities of general as-

pect ratios fewer analytical models are available than for the special cases of spherical
or unidirectional cylindrical reinforcements covered by figs. 4.1 to 4.9, viz., Mori–Tanaka
methods, classical self-consistent schemes and differential models plus approaches requiring
additional microgeometrical descriptors, such as the Ponte–Willis estimates and Maxwell
schemes. Interestingly, these models tend to predict different degrees of macroscopic
anisotropy, as described, e.g., by the ratio between axial and transverse effective Young’s
moduli, EA∗ /ET∗ , for aligned microgeometries. This modeling issue is most marked for ma-
terials containing distinctly oblate or prolate, non-overlapping pores at elevated volume
fractions, compare Fig. 4.10, is less pronounced for compliant inhomogeneities embedded
in a stiff matrix and tends to be rather limited for “classical composites” (stiff reinforce-
ments in a more compliant matrix).
2.0
EA /E T [ ]

CSCS, a=10.0
CSCS, a=0.1
DS, a=10.0
1.0

DS, a=0.1
MTM, a=10.0
MTM, a=0.1
0.0

0.0 0.2 0.4 0.6

PORE VOLUME FRACTION [ ]
Figure 4.10: Comparison of mean field estimates for the macroscopic anisotropy parameter
EA∗ /E ∗ obtained as function of the inhomogeneity volume fraction for an isotropic matrix con-
T
taining ellipsoidal pores of aspect ratios a = 0.1 or a = 10.0 by three mean-field models.

74
 Before closing this section it is worth mentioning that more complex responses may be
obtained when at least one of the constituents is transversally isotropic and one of the re-
sulting two elastic or thermal expansion contrasts exceeds unity whereas the other is smaller
(such situations can occur, e.g., in metals reinforced by carbon fibers). Also, it is worth
noting that the differences between the predictions of different mean-field models tend to
be more pronounced when these algorithms are used to describe nonlinear responses, e.g.,
elastoplastic behavior. Furthermore, the angular dependences of stiffnesses and CTEs in
fiber-reinforced materials can show a fairly rich behavior, compare, e.g., Pettermann et al.
(1997).

4.2 Comparisons of Mean-Field and Bounding Pre-

dictions for Effective Conductivities
Predictions for the effective conductivities of composites as functions of the phase volume
fractions are qualitatively similar to the corresponding data for the elastic moduli, as can be
seen in figs. 4.11 and 4.12, which pertain to a polyetherimide matrix reinforced by aligned
short graphite fibers of aspect ratio a = 10. The plots use material parameters given by
Harte and Mc Namara (2006), which are listed in table 4.2. The fibers show transversally
isotropic conductivity, the axial conductivity contrast taking a value of approximately 37,
whereas the transverse conductivity contrast is an order of magnitude smaller.

Table 4.2: Constituent material parameters of polyetherimide matrix and the T-300 graphite
fibers used for generating figs. 4.11 to 4.12.

kA [Wm−1 K−1 ] kT [Wm−1 K−1 ]

matrix 0.22 0.22
reinforcements 8.40 0.84

Because for the composite described by figs. 4.11 and 4.12 the matrix conductivity is
smaller than either the axial or the transverse fiber conductivities, the Mori–Tanaka es-
timates for the effective conductivity coincide with the lower Hashin–Shtrikman bounds.
Even though the selected fiber aspect ratio is rather moderate, the macroscopic conduction
behavior approaches that of a continuously reinforced composite, with the upper Wiener
and Hashin–Shtrikman bounds being nearly identical. As in the elastic case the classical
self-consistent scheme, which does not describe matrix–inclusion topologies for all volume
fractions, is close to one Hashin–Shtrikman bound at low fiber volume fractions and close
to the other at high ones. The predictions of the differential scheme differ rather strongly
from the lower Hashin–Shtrikman bounds, a behavior that tends to be especially marked in
materials with elevated conductivity contrasts (which may reach very high values, indeed,
in the case of electrical conduction).

75
 EFFECTIVE CONDUCTIVITY [W/mK]
DS

8.0
CSCS
H/S UB
H/S LB; MTM
Wiener bounds
4.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

FIBER VOLUME FRACTION [ ]
Figure 4.11: Bounds and estimates for the effective axial conductivity of composites consisting of
aligned short graphite fibers (a = 10) in a polyetherimide matrix as functions of the fiber volume
fraction.
EFFECTIVE CONDUCTIVITY [W/mK]
0.8
0.4

DS
CSCS
H/S UB
H/S LB; MTM
Wiener bounds
0.0

0.0 0.2 0.4 0.6 0.8 1.0

FIBER VOLUME FRACTION [ ]
Figure 4.12: Bounds and estimates for the effective transverse conductivity of composites con-
sisting of aligned short graphite fibers (a = 10) in a polyetherimide matrix as functions of the
fiber volume fraction.

76
4.3 Comparisons of Mean-Field and Bounding Pre-
dictions for Effective Elastic Moduli of Multi-Phase
Composites
For the sake of brevity only three-phase composites with matrix–inclusion microtopology
are considered here. The first four figures explore fictitious composites made up of a set
of three isotropic phases, the normalized elastic parameters of which are listed in table
4.3. One of the inhomogeneity phases, denoted as (i1) , is stiffer than the matrix, whereas
inhomogeneity phase (i2) is more compliant. The elastic contrast between phases (i1) and
(i2)
takes a value of 25.

Table 4.3: Constituent material parameters of fictitious three-phase composites used in generating
figs. 4.13 to 4.16.

E[ ] ν[ ]
(m)
matrix 1.0 0.30
(i1)
reinforcements 5.0 0.10
(i2)
reinforcements 0.2 0.40

Whereas for plots of the effective responses of two-phase composites one of the phase
volume fractions tends to be a natural choice for the independent variable, things may
be less clear-cut for multi-phase composites. Most of the following diagrams show the
behavior
P of an (i)effective modulus with respect to the total inhomogeneity volume fraction,
I
ξ = (i)6=(m) ξ = 1 − ξ (m) , compare also eqns. (2.92).

Figure 4.13 presents the normalized, effective elastic moduli of composites reinforced
by randomly positioned spherical particles of phases (i1) and (i2) embedded in the matrix
(m)
, the volume fraction ξ (i1) being chosen as 9 ξ (i2) . Such materials show macroscopically
isotropic elastic behavior. Bounding results are provided by the Hill bounds, eqn. (3.1)
and the Hashin–Sthrikman–Walpole bounds (HSW), eqns. (3.2), for which the lower and
upper comparison media correspond to the two inhomogeneity phases. In addition, es-
timates obtained with the Mori–Tanaka Method (MTM), the classical Self-Consistent
Scheme (CSCS), the Differential Scheme (DS) and a pseudo-grain model (PGR/MT-SC)
are given. For the DS the inhomogeneity volume fractions of the inhomogeneity phases
are increased proportionally. The PGR/MT-SC scheme describes individual pseudo-grains
by a Mori–Tanaka method and homogenizes the grain-level elastic tensors by a Classical
Self-Consistent Scheme, compare section 2.8.1. For particle volume fractions ξ I ' 0.5 the
various predictions differ considerably. Since since ξ I = 1 refers to a two-phase inhomoge-
neous material, the plots are qualitatively different from two-phase results such as fig. 4.1.
The MTM elasticity tensors remain symmetric and, because the matrix is not one of the
comparison media, the MTM results do not coincide with one of the HSW bounds.

Figure 4.14 shows the effective normalized Young’s moduli of composites reinforced by
spherical particles consisting of a core of material (i1) concentrically coated by a homoge-
neous layer of phase (i2) , i.e., to the case of stiff core particles surrounded by a compliant

77
 EFFECTIVE YOUNGs MODULUS
PGR/MT-SC
DS

4.0
CSCS
MTM, Maxwell
HSW bounds
Hill bounds
2.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

TOTAL INHOMOGENEITY VOLUME FRACTION
Figure 4.13: Bounds and estimates for the normalized effective Young’s modulus of a fictitious
three-phase particle reinforced composite with constituent properties following table 4.3, evalu-
ated as functions of the total particle volume fraction ξ I = ξ (i1) + ξ (i2) , where ξ (i1) = 9 ξ (i2) .

interphase. Predictions obtained by combining the exact results for the equivalent homoge-
neous particles proposed by Hervé and Zaoui (1990) with the two-phase Hashin–Shtrikman
bounds (CE-H/S), the two phase Mori–Tanaka method (CE-MTM) and the two-phase Dif-
ferential Scheme (CE-DS) are given. The results marked as GEEE were obtained with the
General Explicit Eshelby-Based Estimator of Ghazavizadeh et al. (2019); they also pertain
EFFECTIVE YOUNGs MODULUS

C-GEEE
CE-DS
4.0

CE-H/S UB
CE-H/S LB, CE-MTM
HSW bounds
Hill bounds
2.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

TOTAL INHOMOGENEITY VOLUME FRACTION
Figure 4.14: Bounds and estimates for the normalized effective Young’s modulus of a fictitious
composite reinforced by particles made up of a core of phase (i1) coated by phase (i2) , evaluated
as functions of the total inhomogeneity volume fraction ξ I = ξ (i1) + ξ (i2) , where ξ (i1) = 9 ξ (i2) .
Constituent properties are given in table 4.3.

78
to the MTMT-approaches mentioned in section 2.8.2. The CE-H/S bounds, which account
for the fact that phase (i2) acts as a coating for spherical cores of phase (i1) , making the
geometry “more deterministic”, can be seen to be markedly tighter than the HSW bounds,
which pertain to the case where spheres of the two inhomogeneity phases are randomly
distributed in the matrix. The GEEE predictions, which make use of an approximate
rather than exact model for the effective inhomogeneities, can be seen to fall slightly out-
side the the CE-H/S bounds. Interchanging the roles of phases (i1) and (i2) in the coated
inhomogeneities leads to considerably different results.

Predictions for the normalized axial shear modulus of three-phase composites rein-
forced by aligned, continuous, randomly arranged fibers consisting of materials (i1) and (i2)
are presented in fig. 4.15. The resulting composite shows transversally isotropic macro-
scopic elastic symmetry. All parameters except the shapes of the reinforcements are kept
as in the previous two figures. Most aspects of the behavior predicted for individual macro-
scopic moduli are similar to fig. 4.13, with the various estimates differing considerably for
ξ I ' 0.5. The Hashin–Shtrikman-type bounds can be seen to be clearly tighter than the
Hill bounds. As in the particle reinforced case the Mori–Tanaka predictions coincide with
Maxwell results obtained when setting SE = S(i,m) in eqn. (2.84). As in the analyses un-
derlying Fig. 4.13 no difficulties were encountered in the application of any of the methods
considered.
EFFECTIVE AXIAL SHEAR MODULUS

PGR/MT-SC
DS
CSCS
MTM, Maxwell
4.0

HSW bounds
Hill bounds
2.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

TOTAL INHOMOGENEITY VOLUME FRACTION
Figure 4.15: Bounds and estimates for the normalized effective axial shear modulus of a fictitious
three-phase composite reinforced by aligned, continuous fibers with properties as given in table
4.3, evaluated as functions of the total fiber volume fraction ξ I = ξ (i1) + ξ (i2) , where ξ (i1) = 9 ξ (i2) .

The micromechanical models cannot be used in such a carefree way for the configura-
tion covered by Fig. 4.16, which shows bounds and estimates for the normalized effective
transverse shear modulus of a fictitious three-phase hybrid composite in which the inhomo-
geneities of phase (i1) are aligned, prolate spheroids of aspect ratio a(i1) = 10, whereas phase
(i2)
consists of aligned, oblate spheroids of aspect ratio a(i2) = 0.1. The elastic properties of

79
 EFFECTIVE TRV. SHEAR MODULUS
PGR/MT-SC

3.0
DS
CSCS
Maxwell
MTM/S
Hill bounds
2.0
1.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

TOTAL INHOMOGENEITY VOLUME FRACTION
Figure 4.16: Bounds and estimates for the normalized effective transverse shear modulus of
a fictitious three-phase hybrid composite reinforced by aligned spheroidal inhomogeneities of
aspect ratios a(i1) = 10 and a(i2) = 0.1, evaluated as functions of the total fiber volume fraction
ξ I = ξ (i1) + ξ (i2) , where ξ (i1) = 9 ξ (i2) . Constituent properties are given in table 4.3.

the constituents again follow table 4.3 and the volume fractions of the fiber phases vary pro-
portionally with ξ (i1) = 9 ξ (i2) . For these configurations Hashin–Shtrikman–Willis bounds
as described by eqns. (3.2) cannot be evaluated and the Mori–Tanaka method gives rise to
non-symmetrical elasticity tensors, so that a symmetrized Mori–Tanaka scheme (MTM/S)
as proposed by Sevostianov and Kachanov (2014) is used instead. Nevertheless, the predic-
tions obtained with the different estimates agree fairly well and the Maxwell method using
the conjecture of Sevostianov and Kachanov (2014) for the shape of the inhomogeneous
region gives reasonable predictions for this problem.

Table 4.4: Constituent material parameters of fictitious three-phase composite used in generating
fig. 4.17.

EA [ ] ET [ ] GA [ ] νA [ ] νT [ ]
(m)
matrix 1.0 1.0 3.846 0.30 0.30
(i1)
reinforcements 10.0 3.0 2.5 0.10 0.40
(i2)
reinforcements 2.0 5.0 2.0 0.35 0.15

Table 4.4 introduces a further set of constituent properties which describe two transver-
sally isotropic fiber phases embedded in an isotropic matrix. The material parameters
are chosen such that the matrix becomes the lower comparison medium required for the
Hashin–Shtrikman–Willis bounds, whereas a synthetic upper comparison medium must
be constructed, e.g., from the maxima of the Hill moduli of the phases. Figure 4.17
presents predictions for the normalized, macroscopic transverse bulk modulus of a com-
posite reinforced by aligned, ellipsoidal, short fibers of aspect ratio a(i1) = a(i2) = 5 with
the above phase behavior. The volume fractions of both fiber phases are assumed to be

80
 EFFECTIVE TRV. BULK MODULUS
PGR/MT-SC
DS

6.0
CSCS
HSW upper
HSW lower, MTM
Hill bounds
4.0
2.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

TOTAL INHOMOGENEITY VOLUME FRACTION
Figure 4.17: Bounds and estimates for the normalized effective transverse bulk modulus of a
fictitious three-phase composite reinforced by aligned short fibers of aspect ratios a(i1) = a(i2) = 5,
evaluated as functions of the total fiber volume fraction ξ I = ξ (i1) + ξ (i2) , where ξ (i1) = ξ (i2) . The
material properties of matrix and fibers follow Table 4.4.

identical. Because the lower comparison medium corresponds to the matrix, the lower
Hashin–Sthrikman–Willis bounds coincide with the Mori–Tanaka predictions, which are
slightly more compliant than the Maxwell results (not shown). For this case the Hashin–
Sthrikman–Willis bounds are markedly tighter than the Hill ones and the different esti-
mates agree reasonably well.

The final example, Fig. 4.18, targets the transverse Young’s moduli of composites con-
sisting of unidirectional, continuous glass fibers embedded in an epoxy matrix with 5%
of porosity, i.e., ξ (m) = 19 ξ (i2) , where the spherical pores form phase (i2) . The material
properties of matrix and fibers follow Table 4.1 and the same scaling is used as for Fig. 4.5,
which pertains to the same fibers embedded in a pore-free matrix, in order to allow direct
comparisons.

As in the case of Fig. 4.16 for this composite the Hashin–Shtrikman–Willis bounds
cannot be evaluated and the standard Mori–Tanaka approach results in non-symmetrical
elasticity tensors, so that a symmetrized Mori–Tanaka scheme (MT/S) must be used; its
results differ slightly from those of a Maxwell scheme using the conjecture of Sevostianov
and Kachanov (2014) (not shown). Furthermore, the presence of a pore phase of vanishing
stiffness implies that lower bounds become trivially zero, so that only the upper Hill bounds
are given. No solutions from a Differential Scheme are shown because at ξ (i) = ξ (i1) = 0
a porosity of 5% is present in the matrix, so that no suitable starting configuration is
available for a DS with proportionally increasing inhomogeneity volume fractions (this
issue can be resolved, however, by suitably modifying the model). Figure 4.18, however,
shows results from a sequential Mori–Tanaka model (seqMTM), compare section 2.8.1, in
which the porous matrix is homogenized in a first step and the fibers are embedded in the
resulting medium in a second step; this procedure is compatible with the assumption that

81
 EFFECT. TRV. YOUNGs MODULUS [GPa]
80.0
seqMTM
PGR/MT-SC
CSCS
MTM/S
Hill U.B.
40.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0

FIBER VOLUME FRACTION [ ]

Figure 4.18: Bounds and estimates for the normalized effective transverse Young’s modulus of
unidirectionally, continuously reinforced glass/epoxy composites with a porous matrix, evaluated
as functions of the fiber volume fraction ξ (i) = ξ (i1) . The material properties of matrix and fibers
follow Table 4.1 and the pore volume fraction is proportional to the matrix volume fraction with
ξ (v) = ξ (i2) = 0.05 ξ (m) .

the voids have a much smaller diameter than the fibers. As in Fig. 4.5 the estimates —
with the exception of the classical self-consistent scheme — give rather similar predictions,
the influence of the matrix porosity on the overall behavior being rather limited.

82
Chapter 5

General Remarks on Modeling

Approaches Based on Discrete
Microstructures

In what follows, micromechanical approaches based on discrete microstructures are un-

derstood to encompass the periodic homogenization, embedding and windowing methods
discussed in Chapters 6 to 8 and sketched in fig. 1.1. Broadly speaking, these “full field
models” trade off potential restrictions to the generality of the phase arrangements against
the capability of using fine-grained geometrical descriptions and of resolving details of the
stress and strain fields at length scales smaller than the inhomogeneities49 . The most
important applications of such methods are homogenizing the behavior of inhomogeneous
materials and evaluating the microscopic stress and strain fields in relevant microgeometries
at high resolution. The latter information, on the one hand, is important when the local
fields fluctuate strongly and, consequently, much information is lost by volume averaging.
On the other hand, it is important for understanding the damage and failure behavior of
inhomogeneous materials, which in many cases depends on details of their microgeometry.

The most common work flow in models involving discrete microstructures consists of
first obtaining appropriate phase arrangements, see Section 5.1, discretizing the resulting
volume elements and preparing them for application of a numerical engineering method,
see Sections 5.2 and 5.3, solving for the microfields, and, finally, postprocessing the results,
see Section 5.4.

5.1 Microgeometries and Volume Elements

The heterogeneous volume elements (“microgeometries”) used in full field models range
from highly idealized periodic geometries (“simple periodic arrays”), such as cubic arrays
of spheres in a matrix, to large phase arrangements that aim at closely approximating
the geometrical complexity and/or arrangement statistics of actual inhomogeneous mate-
rials (“microstructural volume elements”). There are two main strategies for generating
heterogeneous volume elements for modeling, viz., using synthetic microstructures, usu-
49
Such methods by construction aim at resolving inter-particle and intra-particle interactions at the
maximum level attainable within continuum mechanics.

83
ally obtained from computer-based simulations, or employing phase arrangements that
are directly based on microgeometries obtained by experimental methods. The following
considerations concentrate on microgeometries with matrix–inclusion topology.

Synthetic Phase Arrangements

Synthetically generated volume elements that go beyond simple periodic arrays may be
classified into two groups, the first of which makes use of generic arrangements of a num-
ber of randomly positioned and, where appropriate, randomly oriented, sized, or shaped
reinforcements. The second group aims at constructing phase arrangements that have
phase distribution statistics identical to those of the target material. A further possible
approach to synthesizing realistic volume elements by computer algorithms, which consists
of modeling relevant production processes, has seen little use to date for composites due to
the complexity of the task. Synthetic volume elements have found wide use for studying
composites with matrix–inclusion and interpenetrating phase arrangements, polycrystals
as well as cellular materials and a considerable amount of relevant work has been done in
the context of modeling the behavior of concrete-like materials. For a recent review on
generating three-dimensional inhomogeneous volume elements see Bargmann et al. (2018).

Synthesizing generic “multi-inhomogeneity” model geometries with random arrange-

ments of reinforcements in many cases has involved random sequential addition meth-
ods50 . The reinforcement volume fractions that can be reached by RSA methods tend
to be moderate due to jamming (“geometrical frustration”), however, their representa-
tiveness for actual phase arrangements generated by mixing processes is open to question
(Stroeven et al., 2004), and they may give rise to biased phase arrangements (Pontefisso
et al., 2016). Major improvements can be obtained by using RSA geometries as “start-
ing configurations” for random perturbation, “hard core shaking” or “migration models”,
compare, e.g., Buryachenko et al. (2003) or Schneider (2017). Such algorithms apply small
random displacements to each inhomogeneity to obtain improved and, if the size of the
volume element is reduced in the process, more tightly packed arrangements of reinforce-
ments. Alternatively, simply periodic arrays or arrangements generated with sophisticated
packing algorithms, see, e.g., Maggi et al. (2008), were used as starting configurations,
whereas RSA methods were combined with stirring models (Zangenberg and Brøndsted,
2013) or with optimization procedures (Pathan et al., 2017) to reach elevated inhomogene-
ity volume fractions. RSA based approaches have been used to generate (approximately)
statistically homogeneous distributions of particulate or fibrous reinforcements, see, e.g.,
Gusev (1997), Böhm and Han (2001), Lusti et al. (2002), Duschlbauer et al. (2006) or
Rasool and Böhm (2012), as well as idealized clustered microgeometries (Segurado et al.,
50
In random sequential addition (RSA) algorithms (also known as random sequential insertion or random
sequential adsorption models, and sometimes referred to as “static simulations”), positions and, where ap-
plicable, orientations for new reinforcements are created by random processes, a candidate inhomogeneity
being accepted if it does not “collide” with any of the existing ones and rejected otherwise. In contrast to
random packing, in generating microgeometrical models a “collision” often involves the violation of some
minimum distance rather than actual touching of (or overlapping with) neighboring inhomogeneities.
Whereas collision checking is straightforward with circular or spherical inhomogeneities, algorithms such
as the separating axis method (Schneider and Eberly, 2003) may be required for polyhedral shapes.
If periodic phase arrangements are generated, up to 8 periodic copies must be maintained for any inho-
mogeneity that is intersected by one or more faces of the volume element.

84
2003), and they can be adapted for obtaining volume elements that follow prescribed fiber
orientation distributions, see, e.g., Schneider (2017).

Matrix–inclusion configurations of elevated inhomogeneity volume fraction can also be

obtained by “random particle dropping” (Chiaia et al., 1997) or gravitational methods
(Khalevitsky and Konovalov, 2019), discrete-element-based models (Ismail et al., 2016),
particle expansion methods (Zhang et al., 2019), or molecular-dynamics-like algorithms
(Lubachevsky et al., 1991), also known as collective rearrangement models, see, e.g., Ghos-
sein and Lévesque (2013), as well as Discrete Element-based models, compare Majewski
et al. (2017). Furthermore, arrangements of particles may be generated via hydrodynamic
interactions in a pseudo-fluid (Elliott and Windle, 2000). For a discussion of methods for
generating geometrically complex volume elements containing fibrous inhomogeneities that
are compliant in bending, see, e.g., Altendorf and Jeulin (2011).

Additional approaches to generating multi-inhomogeneity synthetic volume elements

with matrix–inclusion topology have been based on “eroding” (or shrinking) the individ-
ual cells of suitable Voronoi tesselations (Fritzen and Böhlke, 2011) of the whole volume
or on first filling the unit cell with tetrahedra which are then assigned to the required
phases by a suitable algorithm (Galli et al., 2008), randomness being introduced via the
underlying tesselation in both cases. Both methods result in arrangements of randomly
positioned and oriented inhomogeneities of irregular polygonal or polyhedral shape, the
first mentioned one being applicable to a wide range of inhomogeneity volume fractions.

The most important alternative strategy for generating volume elements by simula-
tion aims at obtaining “statistically reconstructed” phase arrangements, which are not
identical to any given sample, but rather are statistically equivalent to the material to
be modeled; the resulting phase arrangements have been called “statistically equivalent
virtual microstructures”. Statistical reconstruction typically gives rise to optimization
problems in which some starting configuration (e.g., a periodic phase arrangement of ap-
propriate volume fraction) is modified such that suitable statistical and/or stereological
descriptors of the phase distributions approach the chosen target descriptor(s) as closely
as possible. Procedures for reconstructing matrix–inclusion and more general microge-
ometries have been reported that employ simulated annealing procedures (Rintoul and
Torquato, 1997; Torquato, 1998b; Bochenek and Pyrz, 2004), genetic algorithms (Zeman
and Šejnoha, 2001), and other minimization methods (Roberts and Garboczi, 1999). For
in-depth discussions of the underlying issues, such as statistical descriptors for the phase
arrangements of inhomogeneous materials, see, e.g., Torquato (2002) or Zeman (2003);
questions of the uniqueness of reconstructed two-phase microgeometries were discussed
by Jiao et al. (2007). Alternative approaches to statistical reconstruction were reported,
e.g., by Vaughan and McCarthy (2010) or de Francqueville et al. (2019). Most studies on
statistical reconstruction have been based on a number of statistical descriptors of phase
arrangements, among them n-point, nearest neighbor and radial distribution functions as
well as correlation functions, i.e., they follow the concept of geometrical RVEs discussed
in Section 1.3. Reconstructed microstructures considering both statistical geometrical de-
scriptors and model responses (and thus combine both flavors of RVEs) have been termed
“statistically equivalent representative volume elements” (SERVE); for in-depth discus-
sions see the overview of Ghosh et al. (2023).

85
 Microstructures generated by statistical reconstruction are especially attractive because
the phase arrangements in at least some actual composites were found to be not completely
spatially random (Trias, 2005). For volume elements that contain considerable numbers
of inhomogeneities or other microstructural features, statistically reconstructed phase ar-
rangements are more specific to a given target material than are generic random microge-
ometries. Because at present computational requirements limit the sizes of volume elements
that can be handled routinely by numerical engineering methods, this advantage has been
of limited practical impact in the context of continuum micromechanics. This issue can
be addressed by the concept of “statistically similar volume elements”, in which volume
elements of smaller size than RVEs are optimized to approach statistical descriptors of the
target material as closely as possible, see, e.g., Balzani et al. (2010) and Scheunemann et al.
(2015). A recent development is the generation of volume elements via Neural Networks
trained on actual microstructures, see, e.g., Henkes and Wessels (2022).

Synthetically generated matrix–inhomogeneity microgeometries have tended to employ

idealized reinforcement shapes, equiaxed particles embedded in a matrix, for example,
being often represented by spheres, and fibers by cylinders or prolate spheroids51 of appro-
priate aspect ratio. However, recent work has also targeted polyhedral inhomogeneities,
see, e.g., Nogales and Böhm (2008), Rasool and Böhm (2012), Zhang et al. (2014), Sheng
et al. (2016) or Böhm and Rasool (2016), as well as reinforcements of other, non-ellipsoidal
shapes, see, e.g., Peng et al. (2020) or Majewski et al. (2022).

Real Structure Phase Arrangements

Instead of generating phase arrangements by computer algorithms, volume elements may
be chosen to follow as closely as possible the actual microgeometry in some appropriate
subvolume of the material to be modeled, obtained from metallographic sections (Fisch-
meister and Karlsson, 1977), serial sections (Terada and Kikuchi, 1996; Li et al., 1999),
tomographic data (Hollister et al., 1994; Kenesei et al., 2004; Chawla and Chawla, 2006;
Buffière et al., 2008), etc. The resulting volume elements are often called “real microstruc-
ture” models.

The generation of real-structure models from pixel (digital images) or voxel (tomo-
graphic) data describing the geometries of inhomogeneous materials has been the focus of
considerable research efforts. Such work involves the steps of selecting from experimental
data sets appropriate volume elements for analysis (“registration”) and of identifying the
regions occupied by the different constituents by thresholding of the grey values of the pix-
els or voxels (“segmentation”). At this stage pixel or voxel models (compare Section 5.3)
can be generated directly from the segmented data set or contouring procedures may be
used for obtaining “smooth” phase domains, the latter operation typically being more man-
51
For uniform boundary conditions it can be shown that the overall elastic behavior of matrix–inclusion
type composites can be bounded by approximating the actual shape of particles by inner and outer en-
velopes of “smooth” shape, e.g., inscribed and circumscribed ellipsoids. This is known as the Hill modi-
fication theorem (Hill’s comparison theorem, auxiliary theorem or strengthening theorem), compare Hill
(1963) and Huet et al. (1990). Approximations of actual inhomogeneity shapes by ellipsoids typically work
considerably better for convex than for non-convex particle shapes (Kachanov and Sevostianov, 2005).

86
power intensive52 . Alternatively, irregular particle shapes in real-structure arrangements
have been approximated by ellipsoids of appropriate size, shape and orientation, compare
Li et al. (1999). Computed tomography has proven especially useful for determining the
microgeometries of inhomogeneous materials with constituents that differ considerably in
X-ray absorption, e.g., porous and cellular materials.

Real microstructure models provide accurate descriptions of actual phase arrangements,

which, however, may depend to a considerable extent on details of the underlying experi-
ments, e.g., the resolution of the digital images, as well as of the registration, segmentation
and, where applicable, smoothing procedures. In general, the resulting volume elements
obviously are non-periodic, which restricts their use with the periodic homogenization tech-
niques discussed in Chapter 6, but makes them well suited to windowing and embedding
approaches, compare Chapters 7 and 8. For a discussion of three-dimensional sampling
of general microstructures and the categorization of a variety of types of resulting volume
elements see Echlin et al. (2014).

Sizes of Volume Elements

When small volume elements are used in full field simulations, the predicted macroscopic
responses tend to show a marked dependence on the size of the volume elements, see, e.g.,
Iorga et al. (2008). Such behavior may be due, on the one hand, to the boundary conditions
used (e.g., in the case of multi-inhomogeneity, simple periodic arrays using non-periodicity
BCs) or, on the other hand, to the “insufficient geometrical information” contained in the
volume elements. The latter issue immediately raises the question of the size of volume el-
ement required for adequately capturing the macroscopic physical behavior of the material
to be studied53 and thus for evaluating effective rather than apparent responses. On the one
hand, considerations of computational cost obviously put a premium on using the smallest
viable volume element but, on the other hand, suitably resolving microgeometry effects
militates for using microgeometries that are or approach representative volume elements
(compare Section 1.3). These conflicting demands led to the concepts of “minimum RVEs”
by Ren and Zheng (2004), of “statistical RVEs” by Trias et al. (2006), of P-SERVEs by
Swaminathan et al. (2006) and of statistical volume elements (SVEs) by Ostoja-Starzewski
(2006), which are less stringent than that of “proper RVEs”.

When the philosophy of geometrical RVEs is followed, estimates for suitable sizes of
volume elements can be obtained on the basis of descriptors of the microgeometry alone,
the physical property to be modeled playing no role. In such a context the adequacy of
the size of a volume element may be assessed, e.g., on the basis of experimentally obtained
correlation lengths (Bulsara et al., 1999) or covariances (Jeulin, 2001) of the phase ar-
rangement or by posing requirements such as having at least two statistically independent
inhomogeneities in the volume element (Zeman and Šejnoha, 2001). Such approaches can
52
For a review of work directed at automatically generating high quality structured meshes from voxel
data sets, see, e.g., Young et al. (2008).
53
Traditionally, the quality of RVEs has been studied in terms of macroscopic responses (and thus of
phase averages). Assessing higher statistical moments of phase-level microfields in addition to the first one
appears feasible but may lead to a further tightening of requirements on RVEs. At present, no pertinent,
systematic studies seem to be available.

87
be extended to anisotropic microgeometries (Wang et al., 2019) and they have proven suc-
cessful for micromechanical models of the elastic behavior of inhomogeneous materials.

The concept of physical RVEs as stated in Section 1.3, in contrast, implies that the
suitability of a given size of volume element for micromechanical modeling depends on
the physical property to be studied. Assessing the representativeness of a volume element
in this sense can, in principle, be carried out by windowing analysis, compare Chapter 7,
with identical predictions of the macroscopic behavior under macroscopically homogeneous
stress and strain boundary conditions, respectively, being indicative of physical RVEs in
the strict sense. In practice, this criterion has proved difficult to fulfill in a rigorous way,
and, consequently, the suitability of volume elements for a given task is typically assessed
by specifying suitable thresholds or by checking other criteria for representativeness. The
most important pertinent approach is based on studying the convergence of estimates for
the macroscopic behavior with growing size of the volume elements, compare, e.g., Savvas
et al. (2016). Furthermore, the compliance of the predicted effective responses with tight
bounds (e.g., three-point bounds, see Section 3.2) and/or the appropriate macroscopic ma-
terial symmetry may be checked for. It must be kept in mind, though, that fulfilling the
latter criteria — while going a considerable way towards ensuring useful modeling results
— is not sufficient for establishing proper RVEs. For comparisons of different geometry
and microfield based criteria for choosing the size of model geometries see, e.g., Trias et al.
(2006) and Moussaddy et al. (2013).

Assessments based solely on the geometry or on the overall elastic behavior predict
that relatively small volume elements can give fairly accurate results. Zeman (2003) re-
ported that the transversally elastic behavior of composites reinforced by continuous fibers
can be satisfactorily described by unit cells containing reconstructed arrangements of 10
to 20 fibers. “Pragmatic” definitions, in which a given volume element must fulfill some
given criterion to a specified accuracy to be accepted as a physical RVE, have been used
to a considerable extent. Using a nonlocal Hashin–Shtrikman model Drugan and Willis
(1996) found that, for statistically isotropic composites consisting of a matrix reinforced
by spherical particles, volume elements with sizes of some two and five particle diameters
are sufficient for obtaining errors of less than 5% and less than 1%, respectively, in terms
of the macroscopic elastic stiffness. Interestingly, these sizes of volume elements come out
as being independent of the particle volume fraction54 , a prediction that is not shared by
alternative models for assessing RVE sizes (Pensée and He, 2007; Xu and Chen, 2009).
Considerably larger volume elements are required for obtaining a given level of accuracy
for aligned ellipsoidal inhomogeneities, where the RVE size depends on the reinforcement
volume fraction (Monetto and Drugan, 2009), and different estimates for the RVE size
result when modeling elastic and thermal conduction behavior (Kanit et al., 2003).

For cases involving nonlinear or inelastic constituent behavior, a number of numerical

studies (Zohdi and Wriggers, 2001; Jiang et al., 2001; Böhm and Han, 2001) have indicated
that larger volume elements than in the linear case tend to be required for satisfactorily
approximating the overall symmetries and for obtaining good agreement between the re-
sponses of different phase arrangements designed to be statistically equivalent. Later stud-
54
For a given level of uncertainty, this corresponds to the number of identical inhomogeneities within
the volume element scaling with their volume fraction, ξ.

88
ies (Cugnoni and Galli, 2010; Galli et al., 2012; Zhang et al., 2014) have confirmed the need
for relatively large volume elements in order to approach representativeness in elastoplastic
composites, especially at elevated strains. They report a clear dependence of the RVE size
on the inhomogeneity volume fraction and on the macroscopic strain for particle-reinforced
ductile matrix composites. The main reason for this behavior lies in the marked inhomo-
geneity of the microscopic strain fields that is typically present in nonlinear composites.
For example, there tend to be contiguous zones of concentrated plastic strain, which evolve
to become considerably larger than individual reinforcements, thus effectively introducing
a new length scale into the problem55 . In the case of path dependent material behaviors
assessments of the suitability of a given volume element, strictly speaking, pertain only
to the load paths actually considered. Eliminating the dependence of the homogenized
response on the size of the volume element can be argued, in fact, to become impossible
in the presence of strain localization (Gitman et al., 2007), the statistical homogeneity
of the whole sample or component being lost. Large volume elements are also required
if material behavior depends strongly on “rare features” of the microgeometry, see, e.g.,
Przybyla and McDowell (2010), and for configurations that contain markedly non-equiaxed
phase regions, e.g., randomly oriented fibers of elevated aspect ratio. Comparisons between
equivalent volume element sizes for various macroscopic behaviors of two-dimensional in-
homogeneous materials were reported by Ostoja-Starzewski et al. (2007).

Modeling work in most cases has been based on volume elements that are known to be
of insufficient size to be proper RVEs, the main reasons for using them being limits in the
size of models that can be handled and difficulties in providing suitable RVEs for actual
materials. Such “sub-RVE” volume elements, which may be periodic or non-periodic, are
best referred to as statistical volume elements (SVEs, Ostoja-Starzewski (2006)), testing
volume elements (TVEs, Diebels et al. (2005)) or windows, compare Chapter 7 — simply
calling them RVEs is not good practice. When a number of SVEs of comparable volume
and pertaining to a given inhomogeneous material are available, they may be viewed as
being different realizations of the phase arrangement statistics describing that material. In
such cases ensemble averaging over the results obtained from sets of volume elements can
be used to obtain improved estimates for the effective material properties, compare, e.g.,
Kanit et al. (2003) and Stroeven et al. (2004). The number of different volume elements
required for a given level of accuracy of the ensemble averages decreases as their size in-
creases (Khisaeva and Ostoja-Starzewski, 2006). As to using a lower number of larger vs. a
higher number of smaller volume elements, Harper et al. (2012b) reported that the former
option is more efficient56 . For macroscopically isotropic materials the anisotropic contri-
butions to the ensemble averaged results were reported to be markedly reduced compared
55
For elastoplastic matrices, weaker strain hardening tends to cause more inhomogeneous microstrains,
which leads to a requirement for volume elements with higher numbers of inhomogeneities. Also, in
the hardening regime phase arrangements of high reinforcement volume fraction may be associated with
smaller RVEs than configurations of lower V.F., the inhomogeneities interfering with the formation of
zones of concentrated plastic strain (Cugnoni and Galli, 2010). Extremely large volumes were reported
to be necessary when one of the phases shows softening, e.g., due to damage (Zohdi and Wriggers, 2001;
Swaminathan and Ghosh, 2006; Gitman, 2006). For elastic polycrystals, the anisotropy and shape of the
grains may also influence the required size of volume elements (Ren and Zheng, 2004).
56
This result is subject to the caveat that in practice volume elements exceeding a certain size, determined
by limitations of the available hardware and the spatial resolution of the discretization used, cannot be
handled with comparable efficiency as a larger number of smaller ones.

89
to the predictions of the individual SVEs (El Houdaigui et al., 2007). However, Galli et al.
(2012) found that ensemble averaging over non-periodic statistical volume elements in the
elastoplastic range may be compromised by boundary layer effects.

For linear properties Kanit et al. (2003) proposed confidence intervals for assessing
the quality of the ensemble averaged results obtained from sets of SVEs. In this context
an error measure can be defined in terms of the standard deviation S(Y ) of some given
modulus or tensor element, Y , as

1.96 S(Y )
err(Y ) = √ , (5.1)
nSVE
where nSVE is the number of statistically equivalent SVEs used in ensemble averaging. A
related approach for nonlinear behavior was proposed by Pelissou et al. (2009). A further
refinement consists in weighting the contributions of the SVEs according to statistical pa-
rameters (Qidwai et al., 2012). The statistical evaluation of very large ensembles of SVEs
and the corresponding sizes of RVEs are discussed in Dirrenberger et al. (2014).

The size of the volume element(s) used in modeling affects not only the macroscopic
responses, but also the predicted microfields. Accordingly, some authors have explored the
question of what size of volume element is required for studying the local fields at some
specific location. This may be done by studying series of volume elements of different size
that pertain to a given microstructure, see, e.g., Ozturk et al. (2016).

As to the dimensionality of volume elements, for composites reinforced by aligned, con-

tinuous fibers two-dimensional VEs can be used for studying many (but not all) load cases
in linear and nonlinear regimes, compare Section 6.4. For composites containing particles,
platelets and/or short fibers, however, three-dimensional volume elements are in general
necessary, see, e.g., Iung and Grange (1995), Böhm and Han (2001) and Weidt and Figiel
(2014) as well as Sections 6.5 and 6.6.

The above considerations pertain to synthetic as well as real-structure phase arrange-

ments and to two-phase as well as multi-phase materials. Rather large volume elements
may be necessary if one or more of the inhomogeneity phases have low volume fractions or
if different inhomogeneity phases differ markedly in their volume fractions.

5.2 Boundary Conditions

As the size of volume elements describing inhomogeneous materials is increased, the re-
sulting predictions for the apparent overall behavior approach that of the bulk behavior
more and more closely until representativeness is achieved; from then on further increases
in model size do not lead to further changes in the results and the effective behavior is
obtained. The rate of convergence towards proper effective behavior depends on the bound-
ary conditions applied to the volume element.

90
 Fulfilling the surface integral version of the Hill–Mandel condition, eqn. (1.7), which
can be written in the form57
Z
 T  
t(x) − hσi ∗ nΓ (x) u(x) − hεi ∗ x dΓ = 0 , (5.2)
Γ

see Hill (1967) and Hazanov (1998), for inhomogeneous volume elements of finite size, can
be achieved by four types of boundary conditions, three of which are based on uniform
B.C.s (Hazanov and Amieur, 1995; Ostoja-Starzewski, 2006).

First, the traction term in eqn. (5.2) can be made to vanish over the whole boundary by
specifying appropriate Neumann (“natural”) boundary conditions for the tractions t(x).
These are obtained by prescribing a macroscopically homogeneous stress tensor σ a on all
faces of the volume element,

t(x) = σ a ∗ nΓ (x) ∀x ∈ Γ , (5.3)

leading to statically uniform boundary conditions (SUBC, or uniform Neumann BC, UNBC).

Second, the right hand term in eqn. (5.2) can be enforced to be zero by imposing a
given macroscopically homogeneous strain tensor εa on all boundary surfaces,

u(x) = εa ∗ x ∀x ∈ Γ , (5.4)

resulting in kinematically uniform boundary conditions (KUBC, uniform Dirichlet BC,

UDBC, or linear displacement BC, LDBC), which, in FE parlance, are a type of essential
BC. Because eqns. (5.3) and (5.4) impose homogeneous stress or strain fields on the bound-
ary of the volume element, they are known as macrohomogeneous boundary conditions.
Some authors also refer to them as Hashin boundary conditions.

Third, mixed uniform boundary conditions (MUBC) may be specified, in which the
scalar product under the integral is made to vanish separately for each face Γk that is part
of the surface of the volume element,
T  
[t(x) − hσi ∗ nΓ (x) u(x) − hεi ∗ x dΓ = 0 ∀x ∈ Γk . (5.5)

This involves appropriate combinations of traction and strain components that are uni-
form over a given face of the volume element rather than specifying a macroscopically
homogeneous field. Mixed uniform boundary conditions that fulfill eqns. (5.2) and (5.5)
must be orthogonal in their fluctuating contributions (Hazanov and Amieur, 1995). The
symmetry boundary conditions discussed in Section 6.2 are a type of MUBC and specific
sets of MUBC useful for window-type microgeometries are presented in Chapter 7.

Finally, the displacement and traction fields may be decomposed into slow and fast
contributions in analogy to eqn. (1.2),

u(x) = hεi ∗ x + u′ (x) and t(x) = hσi ∗ nΓ (x) + t′ (x) . (5.6)

57
Equation (5.2) essentially states that energy equivalence is achieved when local fluctuations do not
contribute to the elastic strain energy density.

91
Inserting these expressions into the Hill–Mandel criterion, eqn. (5.2), directly leads to the
condition Z
T
t′ (x) u′ (x) dΓ = 0 (5.7)
Γ
on the boundary fluctuations of displacements and tractions, which is fulfilled by periodic
phase arrangements. By definition, such geometries can be fully described by a single pe-
riodic volume element, often called a unit cell, the surface of which must consist of pairs
of parallel surface elements, compare Section 6.2. For homologous points on such pairs of
faces the stress and strain tensors must be identical, which implies that the displacement
fluctuation vectors u′ (x) at the two faces are identical, too, whereas the traction fluctua-
tion vectors t′ (x) have equal absolute values but opposite orientations, compare fig. 5.1.
As a consequence, contributions to eqn. (5.7) from each pair of homologous faces cancel
out and the Hill condition is fulfilled for the periodic unit cell itself as well as for larger
periodic assemblages of unit cells. A more detailed discussion of periodic unit cells and the
associated periodicity B.C. is given in Section 6.2. Incidentally, if the shape of a volume
element is such that for each surface point there exists an “antipodic” point having the
opposite normal vector (as is the case, e.g., for spheres), eqn. (5.7) can be formally enforced
in a point-wise way even for non-periodic volume elements (Glüge et al., 2012), giving rise
to so-called “antipodic periodicity” boundary conditions. For a further discussion of mi-
croscopic boundary conditions see, e.g., Nguyen et al. (2017).

nW W E nE

S
Figure 5.1: Symbolic sketch of antiperiodic traction fluctuations at faces of a periodic VE under
macroscopic uniaxial tensile loading.

Macrohomogeneous boundary conditions following eqns. (5.3) and (5.4) can be shown
to give rise to lower and upper estimates, respectively, for the overall elastic tensor of a
given mesoscopic volume element (Nemat-Nasser and Hori, 1993). Results obtained with
mixed uniform and periodicity boundary conditions must lie between the corresponding
lower and upper estimates (Hazanov and Huet, 1994) and show a faster convergence to-
wards the effective response with increasing size of volume elements. There is, however,
no guarantee that periodic boundary conditions give the best estimates among the above
four types of boundary conditions (Terada et al., 2000).

KUBC-based predictions for the elastic macroscopic behavior of fiber-reinforced com-

posites can be improved via an approach developed by Ghosh and Kubair (2016). It

92
modifies u(x) in eqn. (5.4) by an additional, position dependent displacement contribution
that is evaluated by a modified Eshelby formalism incorporating two-point statistical in-
formation on the microstructure outside the volume element.

SUBC and KUBC do not pose restrictions on the phase geometry or on the shape of
volume elements, but some MUBC are subject to certain restrictions on the latter count,
compare Section 6.2. Periodic homogenization is best suited to volume elements describing
periodic phase geometries (which restricts admissible shapes), to which periodicity bound-
ary conditions or, in special cases, symmetry and antisymmetry boundary conditions can
be directly applied, see Chapter 6. Antipodic boundary conditions require volume elements
of specific shapes, but are not restricted to periodic geometries.

In addition to the above four types of boundary condition, the Hill–Mandel criterion
can also be fulfilled by configurations in which the volume element proper is surrounded
by a layer of homogeneous material that is given the effective behavior of the core. Elas-
tic tensors obtained by applying macrohomogeneous or periodicity boundary conditions,
eqns. (5.3) to (5.7), to the outer surface of the embedding layer can be shown to fall within
the upper and lower estimates generated by applying the macrohomogeneous B.C. directly
to the volume element, see Temizer et al. (2013). Such configurations form the basis of
self-consistent embedding schemes, compare Chapter 8. Furthermore, weak formulations
of periodicity and macrohomogeneous boundary conditions have been developed, see, e.g.,
Mesarovic and Padbidri (2005) or Larsson et al. (2011).

Among the above types of models and boundary conditions only fully implemented
periodicity B.C. do not lead to boundary layers in the predicted microscopic stress and
strain fields58 , which is a clear advantage since boundary perturbations may lead to spurious
behavior when nonlinear constitutive response of the constituents is considered. Periodic
homogenization also is unique in that volume elements consisting of multiple copies of a
given unit cell give rise to the same prediction as the unit cell itself (unless wave phenomena
or buckling are involved).

5.3 Numerical Engineering Methods

The majority of continuum micromechanical studies of discrete microstructures have em-
ployed standard numerical engineering methods for resolving the microfields. Work re-
ported in the literature has involved Finite Difference and Finite Volume algorithms, com-
pare Adams and Doner (1967), Bansal and Pindera (2006), Cavalcante et al. (2012) or
Chen et al. (2016), spring lattice models, compare Ostoja-Starzewski (2002), the Bound-
ary Element Method (BEM), compare Achenbach and Zhu (1989), Liu et al. (2005b) or
Bai et al. (2015), as well as the Finite Element Method (FEM) and its derivatives such as
Extended Finite Element Methods, mesh-free, particle and isogeometric methods, compare
Sukumar et al. (2001), Dang and Sankar (2007) and Missoum-Benziane et al. (2007), or
58
Perturbations caused by the application of uniform boundary conditions to inhomogeneous surfaces or
by the local incompatibility between homogenized and inhomogeneous regions depend on the local material
contrasts and do not necessarily become smaller with increasing volume element size. For a discussion of
correcting such effects see Fergoug et al. (2022).

93
FE-based discrete dislocation models (Cleveringa et al., 1997). In general, spring lattice
models tend to show some advantages in handling traction boundary conditions and in
modeling the progress of microcracks due to local brittle failure. Boundary element meth-
ods typically are at their best in studying geometrically complex linear elastic problems.

In addition to the above methods, techniques using Fast Fourier Transforms (FFT),
compare Moulinec and Suquet (1994), and Discrete Fourier Transforms (DFT), compare
Müller (1996), have found considerable use in continuum micromechanics, especially in
periodic homogenization. Typically, the starting point for such iterative algorithms is a
Lippmann–Schwinger equation, compare eqn. (2.36). The convolution integral in this equa-
tion is solved in Fourier space (where it reduces to a tensor contraction) and the strains are
updated in physical space within each iteration, transforms and back transforms between
the two spaces being handled by fast numerical algorithms. Such spectral methods are
directly applicable to analyzing periodic volume elements, where they tend to be highly
efficient, compare, e.g., Michel et al. (1999) or Ghossein and Lévesque (2012), for both
linear and nonlinear phase behavior. Further developments of these approaches have al-
lowed the handling of infinite phase contrasts (Brisard and Dormieux, 2010) and nonlinear
problems (Schneider, 2021). Homogenization via FFT at present is a highly active research
field and FFT-based schemes are becoming the methods of choice for handling very large,
periodic, voxel-type models. DFT methods have found use in studying the evolution of
microstructures (Dreyer et al., 1999).

A further approach, the Transformation Field Analysis of Dvorak (1992), allows the
prediction of the nonlinear responses of inhomogeneous materials based on either mean-
field descriptions (compare the remarks in Section 2.9) or on full field approximations.
High computational efficiency is claimed for “classical” TFA models employing piecewise
uniform transformation fields (Dvorak et al., 1994), and the extension to FE2 -type models
has been reported (Marfia and Sacco, 2018). A development of the TFA, the Nonuniform
Transformation Field Analysis (Michel and Suquet, 2004), expands the inelastic strains into
a number of non-uniform, incompressible and orthogonal “plastic flow modes” to achieve
efficient descriptions of the elastoplastic behavior of inhomogeneous materials.

Furthermore, integral equations describing the displacement, stress and strain fields in
inhomogeneous materials may be solved by numerical integration schemes, compare, e.g.,
Lee et al. (2011a) or Tashkinov (2021). For a number of additional numerical approaches
that have been mainly used for periodic homogenization see Section 6.1.

Obviously, when numerical engineering methods are used in continuum micromechanics,

the characteristic length of the discretization (“mesh size”) must be chosen considerably
smaller than the microscale of the considered problem in order to obtain spatially well
resolved results.

At present, due to its flexibility in handling complex geometries, the FEM is the most
popular numerical scheme for evaluating full field models. In nonlinear regimes its capa-
bility of supporting a wide range of constitutive descriptions for the constituents and the

94
interfaces between them are especially appreciated59 . A further asset of the FEM in the
context of continuum micromechanics is its ability to handle discontinuities in the stress
and strain components (which typically occur at interfaces between different constituents)
in a natural way via appropriately placed element boundaries. Finally, Finite Element
codes are widely available and many of them provide for linear constraint equations for
coupling multiple degrees of freedom, which are required for periodic homogenization, com-
pare Section 6.2.

Applications of (more or less) standard Finite Element methods to micromechanical

studies may be classed into five main groups, compare fig. 5.2, all of which involve specific
trade-offs in terms of complexity of the models, required meshing effort, computational
efficiency and aspects of accuracy.

a) b) c) d) e)

Figure 5.2: Sketch of FEM approaches used in micromechanics: a) discretization by standard

elements, b) specialized hybrid elements, c) pixel/voxel discretization, d) “multi-phase elements”,
e) use of XFEM

In most published work in FE-based micromechanics the phase arrangements are dis-
cretized by an often high number of “standard” continuum elements, the mesh being de-
signed in such a way that element boundaries (and, where appropriate, special interface
elements) are positioned at all interfaces between constituents. Such “classical meshing ap-
proaches” typically use unstructured meshes and have the advantage that in principle any
microgeometry can be discretized at the required level of accuracy, making them the most
flexible way of using the FEM in continuum micromechanics. The resulting meshes are
suitable for analysis with many readily available FE packages. However, the actual model-
ing of complex phase configurations in many cases requires sophisticated and/or specialized
59
Constitutive models for constituents used in FEM-based micromechanics have included a wide range
of elastoplastic, viscoelastic, viscoelastoplastic and continuum damage mechanics descriptions as well as
crystal plasticity models, see, e.g., McHugh et al. (1993), and nonlocal models, see, e.g., Bassani et al.
(2001). In addition, the FEM has supported a range of modeling options for interfaces between phases.

95
preprocessors for generating the mesh, a task that may be work intensive and has been
found to be difficult to automatize60 , especially in the case of periodic phase arrangements.
In addition, the resulting stiffness matrices may show unfavorable conditioning due to sub-
optimal element shapes. Classical meshing approaches are capable of highly resolving the
microfields at local “hot spots” (e.g., between closely neighboring reinforcements), but the
resulting mesh refinements can lead to very large models, indeed. Nevertheless, its built-in
capabilities for providing mesh refinement where it is needed and for using standard mesh
refinement procedures are major strengths of this discretization strategy.

Alternatively, finite elements that are less restricted in their shapes may be employed
for meshing of multi-phase arrangements. On the one hand, the application of the Virtual
Element Method, which provides general-purpose elements of general polygonal or polyhe-
dral shapes, to computational micromechanics has been reported (Lo Cascio et al., 2020).
On the other hand, special-purpose hybrid finite elements have been developed, which are
formulated to model the deformation, stress, and strain fields in inhomogeneous regions
consisting of a single inhomogeneity or void together with the surrounding matrix on the
basis of some appropriate analytical theory (Accorsi, 1988). The most highly developed
approach of the latter type at present is the Voronoi Finite Element Method (Ghosh et al.,
1996), in which the mesh for the hybrid elements is obtained from Voronoi tessellations
based on the positions of the reinforcements. This way the meshing effort can be kept
rather low and large planar multi-inhomogeneity arrangements can be analyzed using a
limited number of (albeit rather complex) elements. Numerous phase-level constitutive
and damage models have been implemented into the method and extensions to coated
inhomogeneities (Zhang and Guo, 2021) as well as to three dimensional problems (Ghosh
and Moorthy, 2004) were reported. Such approaches are specifically tailored (and limited)
to inhomogeneous materials with matrix–inclusion topologies, and good accuracy as well
as significant gains in efficiency have been claimed for them.

Especially when the phase arrangements to be studied are based on digital images of
actual microgeometries, a third approach to discretizing microgeometries is of interest. It
involves using a mesh consisting of regular, rectangular or hexahedral, elements of fixed
size having the same resolution as the digital data, each element being assigned to one of
the constituents by operations such as thresholding of the grey values of the corresponding
pixel or voxel. The use of structured meshes in such “digital image based” (DIB) models
has the advantage of allowing an automatic, relatively straightforward model generation
from appropriate experimental data (metallographic sections, tomographic scans) and of
avoiding ambiguities in smoothing the digital data (which are generally present if “stan-
dard”, unstructured FE meshes are employed for discretizing pixel- or voxel-based data of
this type). However, modeling approaches that directly discretize pixel or voxel geome-
tries obviously lead to ragged phase boundaries, which may give rise to some oscillatory
60
Most preprocessors for Finite Element analysis are not geared towards discretizing matrix–inclusion
topologies with thin matrix bridges between closely neighboring inhomogeneities (finely resolving the mi-
crofields in such regions has limited influence on the elastic and small-strain plastic macroscopic responses,
but tends to be important in evaluating the microfields and may be critical for modeling damage initia-
tion). Other major sources of practical difficulties are intersections between phase and cell boundaries at
very acute or obtuse angles and the generation of the periodic meshes at the surfaces of volume elements
required for PMAs.

96
behavior of the solutions (Niebur et al., 1999), can cause high local stress maxima (Terada
et al., 1997), may degrade accuracy, and negate detailed modeling of interfacial effects.
Because the resolution of the microgeometry is determined by the chosen voxel size, mesh
size control tends to be a difficult issue with DIB models, with global mesh refinement
by supersampling tending to be an inefficient approach. Some of these limitations can,
however, be addressed by local smoothing algorithms, compare Boyd and Müller (2006).
Despite the above concerns, pixel- or voxel-based models have been claimed not to cause
unacceptably large errors in the predicted macroscopic behavior even for relatively coarse
discretizations, at least in the linear elastic range (Guldberg et al., 1998), where they have
found wide use. In nonlinear regimes, however, local stress concentrations at ragged inter-
faces may give rise to considerable errors which can be attenuated by averaging procedures,
compare Fang et al. (2016).

When volume elements with matrix–inclusion topology are studied by voxel-based

methods, care may be required to ensure that the “matrix bridges” between closely neigh-
boring inhomogeneities are properly resolved. This issue may also apply to FFT-based
micromechanical methods.

A fourth, related approach also uses structured FE meshes, but assigns phase prop-
erties at the integration point level of standard elements (“multi-phase elements”), see,
e.g., Schmauder et al. (1996) or Quilici and Cailletaud (1999). Essentially, this amounts to
trading off the ragged boundaries caused by the voxel mesh against smeared-out (and typ-
ically degraded) microfields within any element that contains a phase boundary, standard
FE shape functions being of limited suitability for handling stress or strain discontinuities
within elements. With respect to the element stiffnesses the latter concern can be coun-
tered by overintegrating elements containing phase boundaries, which leads to improved
approximations of integrals involving non-smooth displacements by numerical quadrature,
see Zohdi and Wriggers (2001). The resulting stress and strain distributions, however, re-
main smeared-out approximations in elements that contain phase boundaries. Multi-phase
element models are, accordingly, also limited in their capabilities for detailed modeling of
interfaces. An alternative technique giving rise to a similar pattern of strengths and limita-
tions consists in prescribing suitably modified material properties to all elements containing
a phase boundary (Toulemonde et al., 2008). In a recent development, “composite voxel”
techniques (Kabel et al., 2017; Mareau and Robert, 2017) have been introduced within the
framework of FFT-based homogenization, stiffnesses in both linear and nonlinear regimes
being approximated in a consistent way, e.g., via laminate models.

A further computational strategy uses static condensation to remove the degrees of free-
dom of the interior nodes of the inhomogeneities (or, in the case of coated reinforcements, of
inhomogeneities plus interphases) from the stiffness matrix pertaining to a heterogeneous
volume (Liu et al., 2005a), whereas standard meshing is applied to the matrix. This may
considerably reduce the computational requirements of linear multi-inhomogeneity models.

In addition, there are some FE-based micromechanical methods that do not fall within
the groups shown in fig. 5.2. Among them are “embedded element” (“embedded mesh”,
“embedded reinforcement”) or “domain superposition” techniques, which aim at reducing
modeling and computational costs by avoiding conformal meshing of individual reinforce-

97
ments. Such approaches typically use a structured and often relatively coarse mesh of
continuum elements (“host mesh”) for the matrix. This is combined with truss or beam
elements, see, e.g., Harper et al. (2012a), solid elements, compare, e.g., Jiang et al. (2008),
or shell elements, see, e.g., Matveeva et al. (2014), that represent the inhomogeneities;
the nodal points used in discretizing the reinforcements in general do not coincide with
those of the host mesh. The elements describing the reinforcement are superposed over
the host mesh, the different parts of the model being tied together, e.g., via appropriate
extended FE formulations (Radtke et al., 2011) or by constraint equations 61 . Embedded
reinforcement modeling strategies imply the presence of excess volume in the models (“vol-
ume redundancy”), giving rise to a tendency towards overestimating the overall stiffness,
especially at non-dilute volume fractions, which, however, can be compensated for in many
cases, compare Hoffmann (2012). Such approaches can markedly reduce the meshing and
computational effort in FE-based micromechanical analysis, allowing very complex fiber–
matrix configurations to be handled62 . The resolution in terms of local displacement, strain
and stress fields is closely tied to the pertinent capabilities of the host mesh, tending to limit
the local accuracy achievable with such models; this can be counteracted by enriching the
shape functions of appropriate matrix elements and/or by adaptive mesh refinement, com-
pare Goudarzi and Simone (2019). A different FE-based approach to studying composites
reinforced by long fibers that are neither straight nor aligned can be based on modeling
the latter by “strings” of volume elements and filling the matrix space in-between by a
suitable volume mesh (Fliegener et al., 2014). Again, local resolution of the microfields is
traded off against the capability of handling complex fiber reinforcement geometries.

The idea of reducing the considerable effort implied in classical meshing approaches
has also led to applying a number additional of Finite Element modeling techniques to
continuum micromechanics that make use of structured “base meshes” which are suitably
modified for handling inhomogeneous phase arrangements63 . Strategies of this type have
been based on the Extended Finite Element method (XFEM) or the Generalized Finite
Element Method (GFEM), phase boundaries that pass through individual elements be-
ing handled via appropriately enriched, non-smooth shape functions, see, e.g., Moës et al.
(2003), Legrain et al. (2011), Soghrati et al. (2012) and Dunant et al. (2013). The resulting
models can closely follow the shapes of phase boundaries, compare the sketch in fig. 5.2e,
and are subject to few restrictions in terms of microgeometries that can be handled. Al-
ternatively, automated mesh refinement techniques may be applied to regular, voxel-type,
base meshes (Fangye et al., 2020).

A relatively recent development for studying microgeometries involving large numbers

of inhomogeneities (hundreds to thousands) by Finite Element methods involves finite el-
ement programs specially geared towards solving micromechanical problems. Such codes
may make use, e.g., of matrix-free iterative solvers such as Conjugate Gradient (CG)
methods, analytical solutions for the microfields (such as constant strain approximations
61
Embedded element techniques were originally developed for modeling rebars in reinforced concrete
structures.
62
Reinforcements approximated by rigid, one-dimensional objects are referred to as rigid line inclusions;
they show some interesting properties, compare Wang et al. (1985).
63
It is not clear, however, if or to what extent structured and voxel-type meshes introduce a systematic
bias into the orientation of regions of elevated strain in nonlinear regimes at elevated loads.

98
corresponding to the upper Hill bounds, eqn. (3.1)), being employed as starting solutions
to speed up convergence64 . For micromechanical studies involving such solvers see, e.g.,
Gusev (1997), Zohdi and Wriggers (2001) or Arbenz et al. (2008). An alternative approach
in this context are multi-grid solvers, compare, e.g., Gu et al. (2016). Very fast solvers have
opened the possibility of handling very large volume elements as well as inverse problems,
e.g., for finding optimal particle shapes for given load cases and damage modes (Zohdi,
2003). Further increases in efficiency can be achieved for very large voxel-based linear DIB
models by exploiting the fact that, in the case of linear analysis, all elements pertaining to
a given phase have identical element stiffness matrices.

A further, relatively recent numerical approach that is related to FE methods are re-
duced basis (or reduced order) homogenization schemes, see, e.g., Fritzen and Kunc (2018),
which use globally supported trial and weighting functions to obtain much smaller alge-
braic systems. Appropriately choosing such functions and performing efficient numerical
integration on them are important issues for such reduced order methods. The Nonuni-
form Transformation Analysis mentioned earlier was among the first methods of this type.
Another recent strategy for generating computationally less demanding models makes use
of soft computing methods such as neural networks (Kim et al., 2021). Like reduced or-
der schemes, such approaches require considerable up-front effort which typically involves
“classical” numerically-based models.

Discrete microstructure approaches employing numerical engineering methods are best

suited to studying phase arrangements in which the characteristic lengths of the important
geometrical features do not differ excessively. If this is not the case, e.g., for randomly
oriented inhomogeneities of high aspect ratio, very large volume elements may be required,
meshing may become onerous, and the numerical effort for solving the models may be-
come very high, especially for three-dimensional configurations. For methods making use
of structured meshes, such as FFT algorithms and FE-based voxel models, the chosen
mesh or voxel size intrinsically limits the achievable geometrical resolution. In the case of
approaches that employ unstructured meshes and support local mesh refinement, such as
standard FE methods in the sense of figure 5.2, practical limits on resolution tend to be
imposed by the size and conditioning of the resulting systems of algebraic equations. For
example, discrete microstructure models of composites reinforced by polydispersely sized
particles the diameters of which differ by more than, say, an order of magnitude tend to
become very unwieldy; similar difficulties arise when modeling inhomogeneities with thin
coatings. Another issue requiring special precautions is the handling of extremely closely
spaced or touching reinforcements via numerically-based full field models65 , compare, e.g.,
Gusev (2016); such geometrical features may be of considerable importance to the damage
behavior of composites, see, e.g., (Grufman and Ellyin, 2008).

64
Essentially, in such a scheme the initial guess gives a reasonable estimate of “long wavelength” contri-
butions to the solution, and the CG iterations take care of “short wavelength” variations.
65
Strict two-phase configurations may, in fact, constitute a considerable idealization for the regions close
to touching (“percolating”) or nearly touching inhomogeneities. On the one hand, production processes
may give rise to preferential accumulation of pores and impurities at such positions in actual materials.
On the other hand, it is not clear to what extent point contact between neighboring inhomogeneities is a
physically reasonable model.

99
 Under conditions of macroscopic softening, e.g., due to damage or localization, dis-
cretizing methods are liable to producing mesh-dependent results. Such tendencies may
be counteracted by appropriate regularization procedures, among them enrichment with
higher-order gradients, see e.g., Geers et al. (2001a), nonlocal averaging of the rate of an
appropriate internal variable, compare, e.g., Jirásek and Rolshoven (2003), the use of time
dependent formulations involving rate effects, see, e.g., Needleman (1987), or local averag-
ing, compare, e.g., Fang et al. (2016).

Special Finite Element formulations (and dedicated software) have been required in
many cases for asymptotic homogenization models, compare section 6.3, and for other
schemes that concurrently solve the macroscopic and microscopic problems, compare Ur-
bański (1999).

5.4 Evaluation of Results

When linear elastic or thermoelastic inhomogeneous materials are studied, the aim of ho-
mogenization in full field approaches consists in evaluating the effective elasticity tensors,
E∗ , and thermal expansion tensors, α∗ , or (some of) the pertinent moduli. In homoge-
nization studies involving nonlinear, path-dependent constituent behavior, e.g., elastoplas-
ticity, thermoelastoplasticity, viscoelastoplasticity or damage no solutions of a generality
comparable to that of the above tensors exist. Here, typically the evolution of appropriate
macroscopic variables is followed along some specified load path, e.g., in the form of effec-
tive stress–strain relations. Most of these tasks involve evaluating the macroscopic stresses
and/or strains from the responses of a given volume element to some load case. This may
be done by approximate volume integration, compare eqns. (5.8), or via the approaches
discussed in Section 6.3.

In linear elastic homogenization it is typically preferable to solve for the macroscopic

elastic tensors, which automatically provide full information on the macroscopic elastic
symmetry of the volume element, rather than reconstructing them from results on individ-
ual moduli66 . For three-dimensional configurations, six linearly independent mechanical
load cases must be applied to the volume element for evaluating macroscopic elastic ten-
sors, and a homogeneous temperature change is required for obtaining the macroscopic
thermal expansion tensor. The effective elastic tensors of inhomogeneous materials with
orthotropic or higher symmetry show similar structures to those of isotropic symmetry,
see eqns. (1.12) and (1.13), i.e., in Voigt/Nye notation all tensor elements are zero with
the exception of the upper left submatrix and the diagonal of the lower right submatrix.
In contrast, volume elements obtained from real structures or generated with stochastics-
based models are typically too small to be proper RVEs and, accordingly, give rise to
homogenized tensors that show small contributions from lower elastic symmetries, typi-
cally making them triclinc (Karimi et al., 2020). In addition, minor perturbations due to
roundoff errors may be present. Evidently, volume elements aimed at describing statisti-
66
Ensemble averaging over a number of SVEs also is arguably best done in terms of the elastic tensors.
Furthermore, if load-controlled periodic homogenization is used, elastic tensors can be obtained at moderate
cost because solutions for the required linearly independent load vectors can be found cheaply once the
system matrix has been factorized.

100
cally isotropic, transversally isotropic or orthotropic material behavior must be expected
to actually return lower macroscopic material symmetries.

Such deviations from an “elastic target symmetry”, which tend to decrease with grow-
ing size (and number, if ensemble averaging is used) of SVEs used (Harper et al., 2012b),
can be mitigated to some extent by finding the elastic tensor of the required symmetry
that is closest to the “raw” elastic tensor obtained from homogenization and averaging.
Pahr and Böhm (2008) proposed a simplified approach in which this raw elastic tensor ob-
tained from an SVE of the type shown in fig. 6.4 (which aims at describing a statistically
isotropic composite) is first projected onto an orthotropic base67 . Closest isotropic tensors
can then be obtained as the isotropic term of a generalized spherical harmonics expan-
sion, using a procedure developed by He and Curnier (1997). Alternatively, expressions
resulting from the minimization of an appropriately defined distance between an elasticity
tensor of prescribed symmetry and a given anisotropic elasticity tensor, as discussed by
Norris (2006), Moakher and Norris (2006) as well as Bucataru and Slawinski (2009), may
be used for finding the closest isotropic, cubic or transversally isotropic elasticity tensors68 .
Such procedures have been successfully applied to evaluating effective moduli pertain-
ing to macroscopically isotropic elastic behavior from ensemble averaged sets of SVEs by
Rasool and Böhm (2012), giving excellent agreement with the three-point estimates of
Torquato (1997, 1998a) for randomly positioned, non-interpenetrating, identical spherical
reinforcements. Alternative ways for extracting elasticity tensors of the required macro-
scopic symmetry consist, on the one hand, of rotational averaging for obtaining isotropic
or transversally isotropic effective elastic behavior and, on the other hand, of projection to
appropriate tensor bases. For studies involving isotropic averaging in such a context see,
e.g., Cook and Young (1985) or Gusev (2016), and for transversally isotropic averaging
compare, e.g., Aboudi (1987). Even though such methods can be highly useful in SVE-
based homogenization, they are of an inherently ad-hoc character.

In addition to studying the macroscopic responses of volume elements (i.e., homogeniza-

tion), the behavior of the fluctuating displacements, stresses and strains as well as fields
derived from them (and thus localization) are of considerable interest. In micromechan-
ical applications numerical engineering methods directly evaluate the above microscopic
fields at considerable spatial resolution. In the case of displacement-based FE methods
using isoparametric elements, the microscopic stresses and strains are primarily evaluated
at the integration point level of the individual elements, nodal values of these variables
being typically obtained by extrapolation and averaging procedures. Some care, however,
is necessary in using such data, because the extreme values of microscopic stresses and
strains may depend markedly on details of the microgeometry, of the discretization used
and, in the case of nodal data, of the evaluation procedure. Furthermore, idiosyncrasies
67
Actually, in finding the orthotropic elasticity tensor that is closest to a raw elastic tensor the orienta-
tions of the former’s principal axes should also be considered. In raw elastic tensors obtained by periodic
homogenization from reasonably large volume elements typically only relatively small deviations from the
target symmetry are present, whereas results from windowing methods, compare Chapter 7, using small
volume elements may lead to more pronounced deviations from the expected macroscopic elastic symmetry.
68
Applying such procedures to transversally isotropic elastic tensors obtained from micromechanical
modeling may, however, run into difficulties because the algorithms are not informed about microgeomet-
rical parameters such as the aspect ratios of the inhomogeneities, which may effectively constrain physically
admissible anisotropies.

101
of the numerical methods may come into play, one case being the reduced accuracy of
stress and strain fields obtained by displacement-based FE methods at interfaces between
constituents and at the models’ surfaces compared to the interior of regions occupied by a
given phase.

For evaluating volume averages of stress- and strain-like variables from numerical pre-
dictions for microscopic fields in a small strain setting, it is typically good practice to use
direct volume integration on the basis of eqn. (1.16)69 . Many FE codes provide the data
necessary for approximating volume averaging by approximate numerical quadrature,
Z N
1 1X i

hf i = f (z)dΩ ≈ fl Ωl . (5.8)
Ω Ω Ω l=1
Here fl and Ωl are the function value and the integration weight (in terms of the volume
of the integration point), respectively, associated with the l-th integration point within a
given integration volume Ω that contains Ni integration points. By analogy, the standard
deviation of function f (z) over volume Ω can be evaluated as70
v
u Ni
u1 X
S(f ) ≈ t (fl − hf i)2 Ωl . (5.9)
Ω l=1

The standard deviations may, however, be less robust with respect to details of the model
than are the phase averages. If the volumes associated with integration points are not
available a simplified version of eqn. (5.8),

N
1X e

hf i ≈ fn Ωn , (5.10)
Ω n=1
may be used, where Ne is the number of pertinent elements, fn is the averaged or the cen-
troidal value of function f (z) in element n, and Ωn is the volume of the element. Equation
(5.10) typically tends to be less accurate compared to eqn. (5.8).

When phase averages are to be generated of variables that are nonlinear functions of
the stress and strain components (e.g., equivalent stresses, equivalent strains, stress triax-
ialities), only direct volume averaging of these variables may be used, because evaluating
nonlinear variables from the averaged components may lead to unacceptable inaccuracies,
69
It is of practical interest that volume averaged and phase averaged microfields obtained from full
field analysis must fulfill all relations given in Section 2.1; this can be conveniently used to check the
consistency of a given model by inserting appropriate results into eqns. (2.5). Note, however, that in the
finite deformation regime appropriate stress and strain measures must be used for this purpose. Whereas
the volume averaged nominal stress can generally be obtained via eqns. (1.3) and (5.8), this does not hold
for the Cauchy stress (Hill, 1972; Nemat-Nasser, 1999). In fact, volume averaging of stress and strain
tensors in the finite strain regime is not necessarily possible, compare, e.g., Kouznetsova et al. (2002). If
eqn. (5.8) or equivalents are to be used for such a purpose, special care as well as knowledge of the stress
and strain measures actually used by the FE code are required.
70
Procedures analogous to eqns. (5.8) and (5.9) may also be used for evaluating other volume integrals,
e.g., in computing Weibull-type fracture probabilities for reinforcement particles or fibers (Antretter, 1998;
Böhm et al., 2002).

102
compare Section 2.9. Provided sufficiently fine meshes are employed, eqn. (5.8) typically
returns fairly accurate results on phase averages in the small strain regime and useful ap-
proximations in other cases. For most types of finite element eqn. (5.10) is less accurate
than eqn. (5.8).

Besides generating overall and phase averages, microscopic variables can also be eval-
uated in terms of averages and standard deviations in individual inhomogeneities, which
supports assessing inter- and intra-inhomogeneity fluctuations of the stress and strain fields,
compare the fiber level averages and standard deviations of the maximum principal stress
displayed in fig. 2.4.

In addition, distribution functions of the microscopic variables, see, e.g., Bornert et al.
(1994), Böhm and Rammerstorfer (1995) or Böhm and Han (2001)), can be extracted
from full field models for the whole composite, for a given phase or for individual inhomo-
geneities. Such “stress spectra” can help in visualizing the local loading environments the
constitutive models of the phases are subjected to. The predictions for the tails of the dis-
tributions, however, tend to show at least some dependence on the mesh and discretization
used, so that care is typically required in interpreting them.

103
Chapter 6

Periodic Microfield Approaches

Periodic Microfield Approaches (PMAs, periodic homogenization) aim at approximating

the macroscopic and microscopic behavior of inhomogeneous materials by studying model
materials that have periodic microstructures.

6.1 Basic Concepts of Periodic Homogenization

Periodic microfield approaches analyze the behavior of infinite (one, two- or three-dimensio-
nal) periodic phase arrangements under the action of far-field mechanical loads or uniform
temperature fields71 , for which the Hill–Mandel criterion can be fulfilled as discussed in
Section 5.2. The most common approach to studying the stress and strain fields in such pe-
riodic configurations is based on describing the microgeometry by a periodically repeating
unit cell72 (RUC), to which the investigations may be limited without loss of information
or generality, at least for static analysis73 .

71
Standard PMAs cannot handle macroscopic gradients in mechanical loads, temperature or composition
in any direction in which periodicity of geometry and fields is prescribed. Such gradients or boundaries
can be studied, however, in directions where periodicity is not prescribed, a typical case being layer-type
models that are non-periodic in one direction and periodic in the other(s), see, e.g., Wittig and Allen
(1994), Reiter et al. (1997) and Weissenbek et al. (1997).
72
In the present report the designation “unit cell” is used for any volume element that can generate a
periodic microgeometry. Accordingly, a unit cell may comprise a base unit (or part of it) of some simple
periodic arrangement, a collective of such base units, or a phase arrangement of arbitrary geometrical
complexity (multi-fiber or multi-particle unit cell) that shows translational periodicity; in the limiting case
a unit cell may thus be a proper representative volume element. Accordingly, the discussion of volume
element sizes in Section 5.1 is fully pertinent to unit cell models.
73
Periodic phase arrangements are not very well suited to modeling the transient behaviors of structural
composite materials. In dynamic settings, RUC behave like phononic crystals, i.e., as metamaterials
exhibiting phononic band structures and dispersion relations. Studying such behavior, however, is a
research field in its own right, compare, e.g., Hussein et al. (2014), in which periodic volume elements play
an important role, see, e.g., Suzuki and Yu (1998).
Unit cells are typically limited to describing wavelengths smaller than or equal to their relevant dimension,
which is a direct consequence of the boundary conditions required for obtaining periodicity. By analogy,
in stability analysis of inhomogeneous materials RUCs can directly resolve only buckling modes of specific
wavelengths, so that considerable care is required in using them, compare Vonach (2001) or Pahr and
Rammerstorfer (2006). However, bifurcation modes with wavelengths exceeding the size of the unit cell
can be handled via periodic linear models by using the Bloch theorem (Gong et al., 2005).

104
 The literature on periodic homogenization of inhomogeneous materials is fairly exten-
sive, and well developed mathematical theories are available on scale transitions in periodic
structures and materials, compare Michel et al. (2001). A wide variety of unit cells have
been employed in such studies, ranging from geometries that describe simple periodic arrays
(“lattices”) of inhomogeneities to complex periodic phase arrangements, such as volume
elements containing considerable numbers of statistically arranged inhomogeneities. For
some simple periodic phase arrangements and for linear material behavior it has proven
possible to find analytical solutions based on series expansions making direct use of the pe-
riodicity (Sangani and Lu, 1987; Wang et al., 2000; Cohen, 2004; Drago and Pindera, 2008).

Even though most PMA studies in the literature have used standard numerical engineer-
ing methods as discussed in Section 5.3, some other numerical schemes have been proposed
that are specialized to periodic phase arrangements. One of them, known as the Method of
Cells (Aboudi, 1989, 1991), in its basic form discretizes unit cells that correspond to square
arrangements of square fibers into four subcells, within each of which displacements are
approximated by low-order polynomials. Traction and displacement continuity conditions
at the faces of the subcells are imposed in an average sense and analytical and/or semi-
analytical approximations to the deformation fields are obtained in the elastic and inelastic
ranges. Because they use highly idealized microgeometries, provide only limited informa-
tion on the microscopic stress and strain fields, and have limited capabilities for handling
axial shear, the resulting models pose relatively low computational requirements and can
provide constitutive descriptions for analyzing structures made of continuously reinforced
composites, see, e.g., Arenburg and Reddy (1991). Developments of the algorithm led to
the Generalized Method of Cells (Aboudi, 1996), which is more flexible geometrically and
allows finer discretizations of unit cells for fiber- and particle-reinforced composites, rein-
forcement and matrix being essentially split into a number of “subregions” of rectangular
or hexahedral shape. For some comparisons with microfields obtained by Finite Element
based unit cells see, e.g., Iyer et al. (2000) or Pahr and Arnold (2002). Later, higher order
displacement interpolants were brought in to obtain the High-Fidelity Generalized Method
of Cells (Aboudi, 2004), which provides an alternative to Finite Element algorithms for
“pixel element” micromechanical models (compare Section 5.3). This method was also
extended to general quadrilateral cell shapes (Haj-Ali and Aboudi, 2010).

A further group of solution strategies for PMAs reported in the literature (Axelsen and
Pyrz, 1995; Fond et al., 2001; Schjødt-Thomsen and Pyrz, 2004) use numerically evaluated
equivalent inclusion approaches that account for interacting inhomogeneities as provided,
e.g., by the work of Moschovidis and Mura (1975). Alternatively, multipole expansion
methods can handle complex periodic matrix–inclusion microgeometries (Sangani and Mo,
1997; Kushch et al., 2008; Kushch, 2013). Furthermore, the elastic fields in periodic inho-
mogeneous materials can be evaluated numerically via variational methods for determining
stress-free strain fields (Wang et al., 2002).

As mentioned in Section 1.5, periodic homogenization cannot be used freely for mod-
eling damage in and failure of inhomogeneous materials, all relevant geometrical features,
among them crack patterns, being per definition periodic. Consequently, PMAs cannot
describe behavior involving macroscopic localization of damage, failure in periodic models
rather being akin to some kind of fragmentation. This, however, does not rule out the use

105
of PMAs for modeling smeared-out damage in parts of larger models, e.g., within domain-
splitting schemes as used by Raghavan and Ghosh (2004).

For developing periodic microfield approaches the strain and stress fields are typically
decomposed into constant macroscopic strain and stress contributions (“slow variables”),
hεi and hσi, plus periodically varying microscopic fluctuations (“fast variables”), ε′ (z)
and σ ′ (z), compare Section 5.2. Furthermore, it is convenient to introduce a “microscopic
coordinate” z that is scaled such that it can resolve the local features of the unit cell.

An idealized depiction of periodic microfields is presented in fig. 6.1, which shows

the variations of the strains εs (z) = hεs i + ε′s (z) and of the corresponding displacements
us (z) = hεs i z + u′s (z) along some line z in a hypothetical one-dimensional periodic two-
phase material consisting of constituents A and B, the length of the unit cell (“unit of
periodicity”) being cz . The volume averaged strain, hεs i, is linked to the displacement
increment per unit cell, ∆us , by the relations

hεs i = ∆us /cz and us (z + cz ) = us (z) + hεs i cz . (6.1)

The periodicity of the strains and the quasi-periodic, cumulative nature of the displace-
ments are evident.

ε, u
<εs>
εs
us <εs> z

∆us

cz

A B A B A B A
z
Figure 6.1: Schematic depiction of the variation of the strains εs (z) and the displacements us (z)
along a generic “one-dimensional periodic composite” of constituents A and B with unit of peri-
odicity cz . Symmetry points of εs (z) and us (z) are indicated by small circles.

6.2 Boundary Conditions

Unit cells together with the boundary conditions (B.C.s) prescribed on them must gener-
ate valid tilings of the undeformed geometry as well as for all deformed states pertinent

106
to a given micromechanical problem. Accordingly, gaps and overlaps between neighboring
volume elements as well as unphysical constraints on their deformations must not be al-
lowed, i.e., they must be geometrically compatible. In order to achieve this, the boundary
conditions for the unit cells must be specified in such a way that all deformation modes
appropriate for the load cases to be studied can be attained. The major types of boundary
conditions used in periodic microfield analysis are periodicity, symmetry, and antisymme-
try B.C.s74 . In PMA models one of these three types of boundary conditions must be used
for any subset of any exterior boundary (unless it is a free boundary) of the volume ele-
ment, irrespective of the numerical method employed for solving the equilibrium equations.

Generally, for any given periodic phase arrangement unit cells are non-unique, the range
of possible shapes being especially wide when point or mirror symmetries are present in
the microgeometry (as tends to be the case for regular lattices). As an example, fig. 6.2
depicts a (two-dimensional) periodic hexagonal array of circular inhomogeneities (e.g.,
fibers oriented normally to the plane) and some of the unit cells that can be used to study
aspects of the behavior of this phase arrangement. There are considerable differences in
the sizes and capabilities of the unit cells shown.

D
B
A
E
C
G

F
H I J

1 periodic boundary symmetry center

symmetry boundary (pivot point)
point symmetry boundary

Figure 6.2: Periodic hexagonal array of circular inhomogeneities in a matrix and 10 unit cells or
“reduced cells” that can be used to describe the mechanical responses of this arrangement under
loads acting parallel to the coordinate axes.

Periodicity Boundary Conditions

The most general boundary conditions for volume elements in periodic homogenization
are periodicity (periodic fluctuation, “toroidal”, “cyclic”) B.C.s, which can handle any
physically valid deformation state of the unit cell and, consequently, of the inhomogeneous
74
For more formal treatments of boundary conditions for unit cells than given here see, e.g., Anthoine
(1995) or Michel et al. (1999).

107
material to be modeled. Periodicity boundary conditions make use of translatoric symme-
tries of a given geometry; in fig. 6.2 cells A to E belong to this group.

In order to describe an N -dimensional phase arrangement with translatoric periodicity,

a suitable unit cell and a set of N linearly independent periodicity vectors pn are required.
These periodicity vectors are neither unique nor do they have to be orthogonal. For any
given periodic microgeometry the minimum volume of pertinent unit cells is well defined,
but such “minimum unit cells” can take a wide range of shapes, some examples being
shown for a simple two-dimensional case in fig. 6.3. The surface of any unit cell to be used
with periodicity boundary conditions must consist of at least N pairs of faces (or pairs of
parts of faces) Γk , and the surface elements making up a given pair, k − and k + , must be
identical but shifted relative to each other by “shift vectors” ck . Each
P shift vector, in turn,
must be a linear combination of the periodicity vectors, i.e., ck = l ml pl , where the mkl
k

are integer numbers. In fig. 6.3 matching pairs of faces (or, in the case of some cells, parts
of faces) Γk are marked by being drawn in identical color and line style. Obviously, faces
of unit cells may be curved, compare, e.g., Garnich and Karami (2004).

p2
p2

p’1 p1

Figure 6.3: Eight different but equivalent periodic minimum-size unit cells for a two-dimensional
periodic matrix–inclusion medium with two (slightly) non-orthogonal periodicity vectors p1 and
p2 (p′1 and p2 form an alternative pair of periodicity vectors). Paired faces (or parts of faces) Γk
are marked by identical line styles as well as colors.

The selection of the shape of unit cells involves trade-offs: On the one hand, unit cells
of simple shape facilitate the application of periodicity boundary conditions at least to
some extent. On the other hand, low-angle intersections between phase boundaries and
cell faces as well as phase boundaries that closely approach cell faces often make such
cells difficult to mesh for FE analysis. The latter issue can be alleviated or avoided by
choosing suitably shaped and positioned unit cells. The process of defining such cells can
be automated, e.g., by using modified Voronoi-type algorithms, which, however, leads to
volume elements of irregular shape such as the leftmost unit cell in fig. 6.3. In practice,
computer generated periodic volume elements have typically been set up such that the peri-
odicity vectors are orthogonal, which facilitates generating unit cells that are rectangles or
right hexahedra and supports straightforward naming schemes of faces, edges, and vertices.

108
 In the following a nomenclature is used in which the faces of two-dimensional quadri-
lateral unit cells are denoted as N, S, E and W (for North, South, East, and West which
are used as in topographical maps), vertices being named according to the adjoining cell
faces, compare figs. 6.5 to 6.7. The faces of three-dimensional cells of hexahedral shape
are, by analogy, referred to as N, S, E, W, B and T (the latter standing for bottom and
top), and edges as well as vertices are referred to via the adjoining faces (e.g., SE or SWB),
see fig. 6.4.

NWT
WT NT

SWT NET

NE
SW

SWB
NEB

SB EB

SEB

Figure 6.4: Cube-shaped periodic unit cell containing 15 randomly positioned spherical particles
of equal size at a volume fraction of ξ (i) =0.15. Designators of the six faces (East, West, North,
South, Top, Bottom) and of the vertices are given (Pahr and Böhm, 2008).

On the basis of the above discussion of unit cell geometries, eqn. (6.1) can be expanded
into the expression
u(z + ck ) = u(z) + hεi ∗ ck (6.2)
for multi-dimensional cases, ck being an appropriate shift vector and z a local coordinate on
a given face. The unit cells tile the computational space by translation, so that neighboring
cells (and, consequently, the “opposite” faces of a given cell) must fit into each other like
the pieces of a jigsaw puzzle in both undeformed and deformed states. For each pair of
surface elements, Γk , eqn. (6.2) allows expressing periodicity boundary conditions for the
mechanical problem in the small strain regime as

∆uk = uk+ − uk− = u(sk + ck ) − u(sk ) = hεi ∗ ck , (6.3)

where uk− and uk+ are the displacements at pairs of homologous nodes positioned at sk and
sk + ck , respectively, on the surface elements Γk− and Γk+ (which may, e.g., correspond to
faces N and S in figs. 6.4 and 6.5). The vector linking such pairs of nodes in a deformed state
is ĉk = ck + ∆uk . The macroscopic strain hεi in eqn. (6.3) is prescribed in displacement
controlled analysis and is a macroscopic response in load controlled analysis. Equation
(6.3) allows expressing the displacement vector uk+ as

uk+ = uk− + uM + − uM − , (6.4)

109
where uM + and uM − are the displacements of an appropriate pair of homologous “master
nodes”, which may (but do not have to) be chosen among the vertices of the periodic
volume element. These control nodes carry the information on ĉk . Such constraint con-
ditions enforce a seamless fit between neighboring unit cells for all possible deformed states.

For the special case of initially rectangular two-dimensional unit cells, such as the one
shown in fig. 6.5, eqns. (6.4) lead to the expressions

uN (s̃1 ) = uS (s̃1 ) + uNW and uE (s̃2 ) = uW (s̃2 ) + uSE , (6.5)

where the vertex SW (“M − ” in the above scheme) is chosen to be fixed and s̃k are local
“face coordinates” that are used to identify homologous nodes on pairs of faces. Note that
eqns. (6.5) directly imply that

uNE = uNW + uSE .

For numerical analysis the two faces making up a homologous pair Γk are best discretized
in a compatible way, so that the nodal points on them are positioned at identical values of
the “face coordinates” s̃k , ensuring that they are separated by a shift vector ck . Equations
(6.3) then become sets of linear constraints each of which links three nodal displacement
DOFs75 . Comparing eqns. (6.3) and (6.5) shows that the displacements of the “master
nodes”, SE and NW, contain the information on the macroscopic strain tensor hεi. In
addition, the displacements of the master nodes and of faces S and W fully control the
displacements of the “slave faces” N and E.

111111
000000
uNW NW
0000000000000000
1111111111111111
000000
111111
NE N
0000000000000000
1111111111111111
0000000000000000
1111111111111111
u NE

0000000000000000
1111111111111111
^
0000000000000000
1111111111111111
W c 1

~
s2 0000000000000000
1111111111111111
0000000000000000
1111111111111111
c 1
E

SW1111111111111111
0000000000000000
0000000000000000
1111111111111111
u SE

0000000000000000
1111111111111111
SE
~
s1 S
Figure 6.5: Sketch of periodicity boundary conditions as used with an initially rectangular two-
dimensional unit cell.

Conditions analogous to eqn. (6.5) can be specified for any periodic, space-filling and
regular two-dimensional cell that has an even number of sides (e.g., squares, rectangles, or
75
In principle, all variables (i.e., for mechanical analysis the displacements, strains and stresses) must
be linked by appropriate periodicity conditions (note that, in contrast to stresses, boundary traction
vectors are antiperiodic). When a displacement based FE code is used such conditions can be specified
explicitly only for the displacement components (including, where appropriate, rotational DOFs), the
implied natural traction B.C. giving rise to antiperiodic tractions in such a setting (Li, 2012). Usually,
however, the periodicity of nodal tractions is fulfilled only approximately. It is also worth noting that
the averaging procedures typically employed for evaluating nodal stresses do not account for a periodicity
linkage in typical FE implementations.

110
hexagons) or three-dimensional cell that has an even number of faces (e.g., cubes, hexa-
hedra, rhombic dodecahedra, or regular tetrakaidecahedra) arranged such that they form
homologous pairs. The scheme can also be extended to unit cells of less regular shape pro-
vided their faces can be suitably paired off, compare Cruz and Patera (1995), Estrin et al.
(1999) or Xia et al. (2003) as well as figs. 6.2 and 6.3. Periodicity B.C.s may be viewed as the
least restrictive option for multi-inhomogeneity unit cell models using phase arrangements
generated by statistics-based algorithms (see, however, Mesarovic and Padbidri (2005)).
Compared to other discrete microstructure approaches, periodic homogenization typically
shows the fastest convergence in terms of sizes of volume elements, see, e.g., El Houdaigui
et al. (2007). For further discussions of periodicity B.C. see, e.g., Garoz et al. (2019).

For layer-type models, which can be generated by specifying free surfaces at appro-
priate pairs of boundaries, macroscopic rotational degrees of freedom, i.e., macroscopic
bending and twisting, can be studied by appropriately modifying the standard periodicity
B.C.s, eqns. (6.2) to (6.5), for the other faces. When such “structure-type” macroscopic
rotational degrees of freedom are present, out-of-plane shear loads must be accompanied
by appropriate bending moments to achieve stress equilibrium, compare, e.g., Urbański
(1999). Models of this type allow using periodic homogenization for evaluating homog-
enized “laminate stiffnesses” of inhomogeneous plates and shells for use within classical
lamination theory (Jones, 1999), for details see, e.g., Pahr and Rammerstorfer (2006).

The above treatment can be carried over into the large strain regime. Periodicity
boundary conditions that are conceptual developments of eqn. (6.5) can be devised for
cases where standard conditions for homogenization are not met, gradient (nonlocal) the-
ories are employed on the macroscale, and higher-order stresses as well as strain gradients
figure in coupling the length scales (Geers et al., 2001b). The resulting deformation pat-
terns of the cells, however, no longer follow geometrical compatibility as discussed above.

In practice FE-based studies of volume elements subjected to periodicity boundary

conditions can be somewhat expensive and some effort may be required for providing com-
patible discretizations on periodic pairs of faces. Accordingly, there has been research
interest in relaxing the compatibility requirements, see, e.g., Wippler et al. (2011) and in
implementing periodicity B.C. for non-compatible meshes via constraint equations, com-
pare, e.g., Nguyen et al. (2012) or Wang et al. (2017). Such approximations typically
give rise to some perturbations of the microfields at the cell boundaries in analogy to the
boundary layers shown by windowing and embedding models, compare Section 5.2. For
reasonably large volume elements the effects of such perturbations on the predicted ho-
mogenized behavior in linear and weakly nonlinear regimes tends to be minor, however.
For strongly nonlinear local behaviors, such as phase-level damage, however, spurious local
fields due to boundary perturbations may markedly influence the predicted macroscopic
responses, even for volume elements that closely approach representativity for linear prop-
erties.

Finally, it is worth noting that high-quality discretizations of periodic volume elements

can be obtained with reasonable effort by employing preprocessors that support the au-
tomatic generation of compatible meshes on homologous periodic faces, such as Netgen
(https://ngsolve.org/).

111
Symmetry Boundary Conditions
For rectangular and hexahedral volume elements in which the faces of the unit cell coincide
with symmetry planes of the phase arrangement and for which this property is retained
for all deformed states that are to be studied, periodicity B.C.s simplify to symmetry (or
mirror) boundary conditions over a subset of the periodic volume, called a “reduced cell”
in the following. Following the nomenclature of fig. 6.6 these B.C.s take the form

uE (s̃2 ) = uSE vN (s̃1 ) = vNW uW (s̃2 ) = 0 vS (s̃1 ) = 0 , (6.6)

where u and v stand for the displacement components in 1- and 2-direction, respectively.
Equation (6.6) places constraints on the normal displacement components at the cells’
surfaces, but leaves the tangential displacements free, thus enforcing the condition that
pairs of opposite faces must stay parallel throughout the deformation history. Accordingly,
symmetry boundary conditions do not allow fluctuations of the local fields in directions
normal to a given face of a unit cell or reduced cell76 . Symmetry B.C.s pose no requirements
with respect to compatibility of phase arrangements or meshes at different faces, but they
may lead to undesirable shapes of inhomogeneities intersected by a cell face when their
mirror images are taken into account.

111
000
000
111
u
u
000
111
NE
NNW

NW
000 NE
111
000000000000
111111111111
000000000000
111111111111
000000000000 E
111111111111
W 111111111111
000000000000
~ 000000000000
111111111111
s2
000000000000
111111111111
SW 111111111111
000000000000
u
SE SE

S
~
s1
Figure 6.6: Sketch of symmetry boundary conditions as used with a rectangular two-dimensional
reduced cell.

Symmetry boundary conditions are fairly easy to use and tend to give rise to small
reduced cells for simply periodic phase arrangements. However, the load cases that can
be handled are limited to uniform thermal loads, mechanical loads that act in directions
normal to one or more pairs of faces, and combinations of the above77 . In fig. 6.2 cell F
uses boundary conditions of this type. Symmetry boundary conditions are typically highly
useful for describing relatively simple microgeometries but may impose marked limitations
on the geometries of complex phase arrangements.
76
The stress and strain fluctuations ε′ (z) and σ ′ (z) must obviously be symmetric with respect to sym-
metry planes, whereas displacements accumulate across symmetry planes, compare fig. 6.1.
77
Symmetry boundary conditions are compatible with uniaxial stress, uniaxial strain, extensional shear
(obtained, e.g., in the two-dimensional case by applying normal stresses σa to the vertical and −σa to
the horizontal faces), hydrostatic and (hygro)thermal load cases. Accordingly, symmetry B.C.s are often
sufficient for materials characterization. However, the normal and shear stresses evaluated as above do
not pertain to the same material coordinate system, since extensional shear corresponds to pure shear in
a coordinate system rotated by 45◦ .

112
Antisymmetry Boundary Conditions
Antisymmetry (point symmetry, central reflection) boundary conditions require the pres-
ence of centers of point symmetry (“pivot points”) and are, accordingly, even more limited
in terms of the microgeometries that they can handle. In contrast to symmetry boundary
conditions, however, unit cells employing antisymmetry B.C. on all faces are not subject
to restrictions in terms of the load cases that can be handled. Among the volume elements
shown in fig. 6.2, cells G and H use point symmetry B.C.s on all faces and can handle any
in-plane deformation78 . Alternatively, antisymmetry B.C.s can be combined with symme-
try B.C.s to obtain very small reduced cells that are restricted to loads acting normal to
the symmetry faces, compare cells I and J in fig. 6.2. Figure 6.7 shows such a reduced cell,
the antisymmetry boundary conditions being applied on face E where a pivot point P is
present. For this configuration the boundary conditions

uU (s̃P ) + uL (−s̃P ) = 2uP

vN (s̃1 ) = vNW = 2vP vS (s̃1 ) = 0 uW (s̃2 ) = 0 , (6.7)

must be fulfilled, where uU (s̃P ) and uL (−s̃P ) are the displacement vectors of pairs of ho-
mologous nodes U and L that are positioned symmetrically with respect to the pivot point
P. As indicated in fig. 6.7 the local coordinate system s̃ is defined on face E and centered
0110
0
1
uNW
N 1111
0000
u NE
00000000000000
11111111111111
U
NW
0000
1111
00000000000000
11111111111111
U E

W
00000000000000
11111111111111
00000000000000
11111111111111
~
s p

s 11111111111111
00000000000000
u
~ P
P

2
00000000000000
11111111111111
00000000000000
11111111111111
u
11111
00000
L
L
SW
00000000000000
11111111111111
~ S SE
s1
Figure 6.7: Sketch of a quadrilateral two-dimensional reduced cell that combines antisymmetry
boundary conditions on face E with symmetry boundary conditions on faces N, S and W.

on P. The undeformed geometry of such a face also must be antisymmetric with respect to
the pivot point P, and the phase arrangements as well as the discretizations on both halves
of face E must be compatible. Three-dimensional reduced cells employing combinations of
symmetry and point symmetry B.C.s can, e.g., be used to advantage for studying cubic
arrays of particles, see Weissenbek et al. (1994), or woven composites, compare Li and Zou
(2011).

78
Unit cells and reduced cells using antisymmetry boundary conditions may have odd numbers of faces.
Triangular volume elements similar to cell H in fig. 6.2 were used, e.g., by Teply and Dvorak (1988) to study
the transverse mechanical behavior of hexagonal arrays of fibers. Rectangular cells with point symmetries
on each boundary were introduced by Marketz and Fischer (1994) for perturbed square arrangements of
inhomogeneities. The periodic cells D and E also show point symmetry on their faces.

113
 For a reduced cell of the type shown in fig. 6.7, either the displacements of the master
nodes SE and NW, uSE and uNW , or the displacements of the pivot P, uP , can be used for
evaluating the macroscopic strain of the corresponding periodic model material, compare
eqn. (6.17).

The favorable convergence properties of periodicity B.C., compare Section 5.2, have led
to proposals to apply them to non-periodic volume elements, too, see, e.g., Schneider et al.
(2017). This strategy, however, may effectively give rise to unrealistic phase geometries
in the boundary regions, leading to spurious perturbations in stresses and strains. Never-
theless, it has been argued that, for sufficiently large volume elements valid results on the
macroscopic behavior may be obtained79 , see, e.g., Terada et al. (2000). These issues were
discussed in depth by Schneider et al. (2022), who, for synthetic phase arrangements, come
out in favor of an alternative strategy, viz., properly “periodizing” volume elements. For
linear analysis of non-periodic real-structure volume elements, windowing methods using
appropriate mixed boundary conditions, compare Chapter 7, are the preferable choice.

Finally, it is worth noting that all of the above boundary conditions, eqns. (6.4), (6.5),
(6.6) and (6.7) as well as their three-dimensional equivalents, can be handled by any FE
code that provides linear multipoint constraints between degrees of freedom and thus allows
the linking of three or more DOFs by linear equations.

6.3 Application of Loads and Evaluation of Fields

Once suitable volume elements have been defined and appropriate boundary conditions
applied, the appropriate loads in the form of uniform macroscopic stresses, uniform macro-
scopic strains and/or homogeneous temperature excursions must be applied to link the
microscopic and macroscopic fields. Whereas loading by uniform temperature increments
does not pose major difficulties, applying far-field stresses or strains is not necessarily
straightforward. For example, the variations of the stresses along the faces of a unit cell in
general are not known a priori, so that it is not possible to prescribe the boundary tractions
via distributed loads. There are two major approaches to implementing the micro–macro
linkage, both of which allow handling arbitrary stress or strain controlled load cases in
linear and nonlinear regimes.

Asymptotic Homogenization
The most versatile and elegant strategy for linking the macroscopic and microscopic fields in
periodic microfield models is based on a mathematical framework referred to as asymptotic
homogenization, asymptotic expansion homogenization, or homogenization theory, see,
e.g., Suquet (1987) or Auriault (2002). It is based on explicitly introducing macroscopic
and microscopic coordinates, y and z, respectively, into the formulation of the problem.
79
In the same vein, non-periodic, rectangular or cube-shaped volume elements may simply be subjected
to symmetry B.C. (or may be mirrored with respect to two or three of their faces and then subjected to
periodicity B.C). Again, such operations may lead to undesirable reinforcement shapes, changes in the
phase arrangement statistics and perturbations of the microfields.

114
Microscopic and physical coordinates are linked by the expression

zi = xi /ǫ (6.8)

where ǫ = ℓ/L ≪ 1 is a scaling parameter. L and ℓ stand for the characteristic lengths
of the macro- and microscales80 . The displacement field in the unit cell can then be
represented by an asymptotic expansion of the type
(0) (1) (2)
ui (y, z, ǫ) = ui (y) + ǫ ui (y, z) + ǫ2 ui (y, z) + H.O.T. , (6.9)
(0) (1)
where the ui are the effective or macroscopic displacements and ui stands for the peri-
odically varying displacement fluctuations due to the microstructure81 .

Using the chain rule, i.e.,

∂ ∂ 1 ∂
f (y(x), z(x), ǫ) → f+ f , (6.10)
∂x ∂y ǫ ∂z
in the small strain regime the strains can be related to the displacements as
   
1 n ∂ (0) ∂ (0) ∂ (1) ∂ (1) o
εij (y, z, ǫ) = u + u + u + u
2 ∂yj i ∂yi j ∂zj i ∂zi j
   
ǫ n ∂ (1) ∂ (1) ∂ (2) ∂ (2) o
+ u + u + u + u + H.O.T.
2 ∂yj i ∂yi j ∂zj i ∂zi j
(1) (2)
= εij (y, z) + ǫ εij (y, z) + H.O.T. , (6.11)
(0) (0)
where terms of the type εij = 1ǫ ∂z∂ j ui are deleted due to the underlying assumption that
the variations of slow variables are negligible at the microscale. By analogy the stresses
can be expanded into the expression
(1) (2)
σij (y, z, ǫ) = σij (y, z) + ǫ σij (y, z) + H.O.T. . (6.12)

Using the two-scale assumption and, as a consequence, eqn. (6.10), the equilibrium equa-
tions take the form
 ∂ 1 ∂ 
+ σij (y, z, ǫ) + fi (y) = 0 , (6.13)
∂yj ǫ ∂zj
the fi being macroscopic body forces. By inserting eqn. (6.12) into this expression and sort-
ing the resulting terms by order of ǫ a hierarchical system of partial differential equations
is obtained, the first two of which are
∂ (1)
σ =0 (order ǫ−1 )
∂zj ij
∂ (1) ∂ (2)
σij + σ + fi = 0 (order ǫ0 ) . (6.14)
∂yj ∂zj ij
80
Equation (6.8) may be viewed as “stretching” the microscale so it becomes comparable to the
macroscale, f (x) → f (y, y/ǫ) = f (y, z).
81
The nomenclature used in eqns. (6.8) to (6.15) follows typical usage in asymptotic homogenization.
It is more general than but can be directly compared to the one used in eqns. (1.2) to (6.7), where no
macroscopic coordinates y are employed.

115
The first of these equations gives rise to a boundary value problem at the unit cell level that
is referred to as the “micro equation”. By making a specific ansatz for the strains at the
microlevel and by volume averaging over the second equation in the system (6.14), which
is known as the “macro equation”, for elastic problems the microscopic and macroscopic
∗
fields can be linked such that the homogenized elasticity tensor, Eijkl , is obtained as
Z h i
∗ 1 ∂
Eijkl = Eijkl (z) Iklmn + χkmn (z) dΩ . (6.15)
ΩUC ΩUC ∂zl

Here ΩUC is the volume of the unit cell, Eijkl (z) is the microscopic elasticity tensor, which
depends on the constituent present at position z, Iijkl is the 4th-order unit tensor, and the
“characteristic function” χkmn (z), a tensor of order 3, describes the deformation modes of
the unit cell82 and, accordingly, relates the micro- and macrofields. Analogous expressions
can be derived for the tangent modulus tensors used in elastoplastic analysis, compare
Ghosh et al. (1996).

The above relations can be used as the basis of Finite Element algorithms that solve for
the characteristic function χijk (z), a task that in most cases has required special analysis
codes. For detailed discussions of asymptotic homogenization methods within the frame-
work of FEM-based micromechanics see, e.g., Hollister et al. (1991), Ghosh et al. (1996),
Hassani and Hinton (1999), Chung et al. (2001) or Kanouté et al. (2009). Asymptotic
homogenization procedures for elastic composites using commercial FE packages were pro-
posed by a number of authors, among them Banks-Sills and Leiderman (1999), Barroqueiro
et al. (2016), Colera and Kim (2019) and Christoff et al. (2020).

Asymptotic homogenization supports the direct coupling of FE models on the macro-

and microscales, compare, e.g., Ghosh et al. (1996) or Terada et al. (2003), an approach
that has been used in a number of multi-scale studies (compare Chapter 9); concurrent
coupling is sometimes referred to as the FE2 method (Feyel, 2003). A treatment of homog-
enization in the vicinity of macroscopic boundaries can be found in Schrefler et al. (1997).
Asymptotic homogenization schemes are suitably for handling finite strains and they have
also been employed for problems combining higher-order stresses and strain gradients with
nonlocal behavior on the macroscale (Kouznetsova et al., 2004)83 , an approach referred to
as higher order homogenization. For recent reviews of asymptotic homogenization see, e.g.,
Kalamkarov et al. (2009) and Charalambakis (2010).

An alternative unit-cell based asymptotic homogenization scheme solves for a displace-

ment-like “fluctuation function” rather than for the characteristic function appearing in
eqn. (6.15). This Variational Asymptotical Method for Unit Cell Homogenization (VA-
MUCH) (Yu and Tang, 2007) also is suitable for implementation within a Finite Element
framework. Developments of this approach are discussed, e.g., by Tang and Yu (2011).
82
Note that, even though eqns. (6.14) and (6.15) are derived from an explicit two-scale formulation,
neither of them contains the scale parameter ǫ, see the discussion by Chung et al. (2001).
83
Such methods are especially useful for problems in which the length scales are not well separated; in
them the “unit cells” do not necessarily remain periodic during the deformation process.

116
Method of Macroscopic Degrees of Freedom
When asymptotic homogenization is not used, it is good practice to apply far-field stresses
(in the case of load controlled analysis) or strains (in the case of displacement control) to
a given unit cell via concentrated nodal forces or prescribed displacements, respectively, at
the master nodes and/or pivot points discussed in Section 6.2. This approach was termed
the “method of macroscopic degrees of freedom” by Michel et al. (1999).

For load controlled analysis, the nodal forces to be applied to the master nodes can
be evaluated from the macroscopic stress σ a via the divergence theorem, see Smit et al.
(1998). For the configuration shown in fig. 6.5 the concentrated forces acting on the master
nodes SE and NW of a two-dimensional volume element, fSE and fNW , can be shown to be
given by the surface integrals
Z Z
a
fSE = t (s) dΓ and fNW = ta (s) dΓ . (6.16)
ΓE ΓN
a a
Here t (s) = σ ∗ nΓ (s) stands for the surface traction vector corresponding to the homo-
geneous macroscopic (applied, far-field) stress field84 at some given point s on the cell’s
surface ΓUC , and nΓ (s) is the local normal vector of the appropriate face. Equation (6.16)
can be generalized to require that each master node is loaded by a force corresponding to
the surface integral of the surface traction vectors over the face slaved to it via an equiva-
lent of eqns. (6.16). Analogous procedures hold for three-dimensional cases, and symmetry
as well as antisymmetry boundary conditions as described by eqns. (6.6) and (6.7), respec-
tively, can be handled by eqn. (6.16).

For applying far-field strains to periodic volume elements in displacement controlled

analysis, the displacements to be prescribed to the master nodes must be obtained from
the macroscopic strains via appropriate strain–displacement relations. For example, using
the notation of eqns. (6.5), the displacement vectors to be prescribed to the master nodes
NW and SE of the unit cell shown in fig. 6.5 can be evaluated from eqn. (6.3) as
uNW = εa ∗ cW and uSE = εa ∗ cS (6.17)
for an applied strain εa and linear strain–displacement relations85 . For suitably chosen
unit cells, the shift vectors ck are equal to the cell’s side lengths in the undeformed state
and they are, accordingly, referred to as cW and cS in eqn. (6.17). In many cases displace-
ment controlled unit cell models are somewhat easier to handle than load controlled ones86 .

84
Note that the ta (s) are not identical to the actual local values of the tractions at the cell boundaries,
t(s), which contain contributions due to the local field fluctuations. However, the ta (s) are equal to the t(s)
over a given cell face in an integral sense. For geometrically nonlinear analysis eqn. (6.16) must be applied
to the current configuration. Because the far-field stress σ a is constant it can be factored outRof the surface
integrals, so that the force applied to a master node M, fM , can be expressed as fM = σ a ∗ ΓS nΓ (z) dΓ.
85
Equations (6.17) can be directly extended to three-dimensional configurations. For handling finite
strains and geometrical nonlinearities within such a framework see, e.g., Huber et al. (2007) or Barulich
et al. (2018).
86
Even though the loads acting on the master nodes obtained from eqn. (6.16) are equilibrated, in stress
controlled analysis solid body rotations may be induced through small numerical errors. Suppressing
these solid body rotations by deactivating degrees of freedom beyond the ones required for enforcing
periodicity may give rise to incompatibilities between the additional DOFs, the “periodicity DOFs” and
the concentrated loads acting on the master nodes.

117
 The overall stress and strain tensors pertaining to a given volume element, which in
many cases are required for describing the macroscopic behavior, can be evaluated by vol-
ume averaging or via the equivalent surface integrals given in eqn. (1.4). In a finite strain
setting consistent macroscopic strain tensors can be extracted from the volume averages
of the deformation gradients. In practice, it is often fairly straightforward to approximate
volume integrals by numerical integration schemes such as eqns. (5.8) of (5.10), whereas
comparably convenient approximations are not available for surface integrals within an FE
framework.

In the case of rectangular or hexahedral unit cells or of reduced cells that are aligned
with the coordinate axes, averaged engineering stress and strain components can, of course,
be evaluated by simply dividing the applied or reaction forces at the master nodes by the
appropriate surface areas and by dividing the displacements of the master nodes by the ap-
propriate cell lengths, respectively. The displacements of and concentrated forces acting on
the master nodes can also be used for evaluating the macroscopic stresses and strains from
skewed unit cells having non-rectangular periodicity vectors, compare, e.g., Pahr (2003),
where, in addition, some further aspects of unit cells, master nodes, and the method of
macroscopic degrees of freedom are discussed. In general, some care is required in apply-
ing as well as evaluating macroscopic stresses and strains if the volume elements are not
quadrilaterals or hexahedra.

An alternative extracting the effective elastic moduli from the volume averages of the
macroscopic stress and strain tensors consists in basing their evaluation on the average
of the elastic energy, which can have beneficial effects on the numerical accuracy of the
predicted elastic properties (Schneider, 2022).

In order to obtain three-dimensional homogenized elastic tensors with the method of

macroscopic degrees of freedom six suitable, linearly independent load cases must be solved
for in the most general case.

6.4 Periodic Models for Composites Reinforced by

Continuous Fibers
A wide range of periodic models have been reported for the most important groups of
continuously reinforced composites, viz., unidirectionally reinforced, angle-ply and cross-
ply materials, as well as fabric-reinforced composites.

Composites Reinforced by Unidirectional Continuous Fibers

Composites reinforced by continuous, aligned fibers typically show a statistically transver-
sally isotropic overall behavior and can be studied well with periodic homogenization.
Materials characterization with the exception of the overall axial shear behavior can be
carried out with two-dimensional unit cell or reduced cell models employing generalized
plane strain elements that use a global degree of freedom for describing the axial defor-

118
 CH3 RH2 CH1 PH0

CS8 CS7 MS5 PS0

Figure 6.8: Eight simple periodic fiber arrangements of fiber volume fraction ξ=0.475 for modeling
continuously fiber-reinforced composites (Böhm and Rammerstorfer, 1995).

mation of the whole model87 . For handling the overall axial shear response (required, e.g.,
for establishing macroscopic elasticity tensors) special generalized plane strain elements
(Adams and Crane, 1984; Sørensen, 1992) or three-dimensional models with appropriate
periodicity boundary conditions, see, e.g., Pettermann and Suresh (2000), are required88 .
Three-dimensional modeling is required for materials characterization of composites rein-
forced by continuous aligned fibers when the effects of fiber misalignment, of fiber waviness
(Garnich and Karami, 2004), of pores in the matrix, of failed fibers89 (Mahishi, 1986) or
of local damage to matrix or interface are to be studied. Unit cell models used for general
nonlinear constitutive modeling of continuously reinforced composites must be fully three-
dimensional and employ periodicity boundary conditions.

The most basic generalized plane strain models of continuously reinforced composites
make use of simple periodic fiber arrangements as shown in fig. 6.8, all of which can be de-
scribed by rather small reduced cells using symmetry and/or antisymmetry B.Cs, compare
fig. 6.2. The simplest among these phase geometries are the periodic hexagonal (PH0) and
periodic square (PS0) arrays, the use of which goes back to the 1960s (Adams and Doner,
1967). Models with hexagonal symmetry (PH0,CH1,RH2,CH3) give rise to transversally
87
Generalized plane strain elements suitable for such analysis are implemented in a number of commercial
FE codes. Because the axial stiffness of composites reinforced by continuous aligned fibers can usually
be satisfactorily described by Voigt-type (“rule of mixture”) models, PMA studies of such materials have
tended to concentrate on the transverse behavior. Note that plane strain models do not properly account
for the axial constraints in such materials.
88
For linear elastic material behavior there is the additional option of describing out-of-plane shear
behavior via the formal analogy between antiplane shear and diffusion (e.g., conduction) problems.
89
For investigating the axial failure behavior of continuously reinforced MMCs, statistical methods
concentrating on fiber fragmentation within spring lattice models, see, e.g., Zhou and Curtin (1995),
have been used successfully.

119
isotropic thermoelastic overall behavior90 , compare, e.g., Ptashnyk and Seguin (2016),
whereas the other fiber arrangements shown in fig. 6.8 have tetragonal (PS0,CS7,CS8) or
orthotropic (MS5) overall symmetry. In the elastoplastic and damage regimes the macro-
scopic symmetries of hexagonal and square fiber arrangements are degraded, compare
fig. 6.9, hexagonal arrangements ceasing to be strictly transversally isotropic with respect
to yield and ultimate strength. This behavior is due to the lower symmetry of the distri-
butions of stresses and strains in the matrix (which in the nonlinear regime depend on the
load history a given material point has undergone). Such effects become more pronounced
for matrices with decreasing strain hardening of the matrix. In many cases simple periodic
microgeometries do not provide satisfactory descriptions of fiber-reinforced materials, most
of which show at least some randomness in the fiber positions. Much improved models can
be obtained by periodic multi-fiber unit cells that employ quasi-random fiber positions.
Such models can either use symmetry B.C.s, compare the cell shown in fig. 6.10, which
is based on the work of Nakamura and Suresh (1993), or periodicity B.C.s (Moulinec and
Suquet, 1997; Gusev et al., 2000; Zeman and Šejnoha, 2007). With growing computer
power multi-fiber unit cells have become a standard tool for periodic homogenization of
continuously reinforced composites.
6.00E+01

PS/00
5.00E+01

Fiber DN

PH
4.00E+01
APPLIED STRESS [MPa]

PS/45
3.00E+01

Matrix
2.00E+01
1.00E+01

DN/00
PS0/45
PS0/00
PH0/90
PH0/00
0.00E+00

Al99.9 MATRIX
ALTEX FIBER

0.00E+00 5.00E-04 1.00E-03 1.50E-03 2.00E-03 2.50E-03 3.00E-03 3.50E-03 4.00E-03

STRAIN []

Figure 6.9: Transverse elastoplastic response of a unidirectional continuously reinforced AL-

TEX/Al MMC (ξ=0.453, elastic material parameters as in table 6.1, matrix with linear harden-
(m) (m)
ing, yield stress σy =28.2 MPa and hardening modulus Eh =637.5 MPa) to transverse uniaxial
loading as predicted by PMA models PH0, PS0 and DN.

In table 6.1 thermoelastic moduli of an aligned continuously reinforced ALTEX/Al

MMC as predicted by bounding methods, MFAs and unit cell methods using arrange-
ments PH0, CH1 and PS0 (compare fig. 6.8) as well as DN (see fig. 6.10) are listed.
90
Like the macroscopic elastic responses, the phase averages of the local stress and strain fields of hexag-
onal arrangements of aligned fibers are independent of the orientations of transverse load contributions.
The higher statistical moments and the distributions functions of these fields at the phase level, however,
tend to show such a dependence. Consequently, the average equivalent stress and strain in the matrix vary
to some extent with the loading direction.

120
Table 6.1: Overall thermoelastic moduli of a unidirectional continuously reinforced ALTEX/Al
MMC (ξ=0.453 nominal) as predicted by the Hashin–Shtrikman (HSB) and three-point (3PB)
bounds, by the Mori–Tanaka method (MTM), the generalized self-consistent scheme (GSCS), the
differential scheme (DS) and Torquato’s three-point estimates (3PE), as well as by PMA analysis
using periodic arrangements shown in fig. 6.8 (PH0, CH1, PS0) and the multi-fiber cell displayed
in fig. 6.10 (DN). For arrangement PS0 responses in the 0◦ and 45◦ , and for arrangement DN
responses in the 0◦ and 90◦ directions are listed.

EA∗ ET∗ νA∗ νT∗ ∗

αA αT∗

[GPa] [GPa] [] [] [K ×10−6 ]

−1
[K ×10−6 ]
−1

fibers 180.0 180.0 0.20 0.20 6.0 6.0

matrix 67.2 67.2 0.35 0.35 23.0 23.0
HSB/lo 118.8 103.1 0.276 0.277 11.84 15.77
HSB/hi 119.3 107.1 0.279 0.394 12.47 16.46
3PB/lo 118.8 103.8 0.278 0.326 11.85 16.31
3PB/hi 118.9 104.5 0.279 0.347 11.98 16.45
MTM 118.8 103.1 0.279 0.342 11.84 16.46
GSCS 118.8 103.9 0.279 0.337 11.84 16.46
DS 118.8 103.9 0.278 0.339 11.94 16.35
3PE 118.8 103.9 0.279 0.338 11.89 16.40
PH0 118.8 103.7 0.279 0.340 11.84 16.46
CH1 118.7 103.9 0.279 0.338 11.90 16.42
PS0/00 118.8 107.6 0.279 0.314 11.85 16.45
PS0/45 118.8 99.9 0.279 0.363 11.85 16.45
DN/00 118.8 104.8 0.278 0.334 11.90 16.31
DN/90 118.8 104.6 0.278 0.333 11.90 16.46

For this material combination all PMA results (even for arrangement PS0, which is not
transversally isotropic) fall within the Hashin–Shtrikman bounds91 , but the predictions for
the square arrangement show clear in-plane anisotropy and do not fulfill the three-point
bounds (which in this case pertain to aligned, identical, non-overlapping cylindrical fibers,
compare Section 4.1). The results for the (sub-RVE) multi-fiber arrangement indicate some
minor deviation from transversally isotropic macroscopic behavior.

The fiber arrangements shown in figs. 6.8 and 6.10 give nearly identical results for the
overall thermoelastoplastic behavior of continuously reinforced composites under axial me-
chanical loading, and the predicted overall axial and transverse responses under thermal
loading are also very similar. The overall behavior under transverse mechanical loading,
however, depends markedly on the phase arrangement, see fig. 6.9. In addition, for fiber
arrangements of tetragonal or lower symmetry (e.g., PS0, MS5, CS7 and CS8 in fig. 6.8)
the predicted transverse stress-strain responses typically vary strongly with the loading di-
rection in the transverse plane. The behavior of the hexagonal arrangements is sandwiched
91
The constituents’ material properties underlying table 6.1 show a low elastic contrast of cel ≈ 3,
making them relatively insensitive to perturbations of macroscopic symmetry. In general the elastic moduli
obtained from square-type arrangements may violate the Hashin–Shtrikman bounds and usually lie outside
the three-point bounds.

121
 EPS.EFF.PLASTIC

0.015
0.012
0.009
0.006
0.003

Figure 6.10: Microscopic distributions of the accumulated equivalent plastic strain in the matrix
of a transversally loaded unidirectional continuously reinforced ALTEX/Al MMC (ξ=0.453) as
predicted by a multi-fiber cell (arrangement DN) using symmetry BC.

between the stiff (0◦ ) and the compliant (45◦ ) responses of periodic square arrangements
in both the elastic and elastoplastic ranges. Multi-fiber cells with statistical fiber positions
(such as the one shown in fig. 6.10) tend to show noticeably stronger macroscopic strain
hardening compared to periodic hexagonal arrangements of the same fiber volume fraction,
compare Moulinec and Suquet (1997), and to approach transversal isotropy. It is worth
noting that the macroscopic yield surfaces of uniaxially reinforced MMCs in general are not
of the Hill (1948) type, but can be better described by a bimodal description (Dvorak and
Bahei-el Din, 1987). Analogous macroscopic behavior has been reported for viscoelastic
composites (Li et al., 2015).

The distributions of microstresses and microstrains in fibers and matrix typically de-
pend strongly on the fiber arrangement in the transverse plane, especially under thermal
and transverse mechanical loading. In the plastic regime, the microscopic distributions of
equivalent and hydrostatic stresses, as expected, show large-scale patterns when simple pe-
riodic fiber arrangements are used. Such patterns tend to be absent in predictions obtained
with volume elements containing randomly positioned fibers, even though the distributions
of equivalent stresses, plastic strains and stress triaxialities typically are markedly inho-
mogeneous92 , see, e.g., fig. 6.10. As a consequence of the inhomogeneity of the microfields,
92
The corresponding distribution functions, phase averages, and higher statistical moments are also
significantly influenced by the fiber arrangement, compare Böhm and Rammerstorfer (1995).

122
there tend to be strong constrained plasticity effects in continuously reinforced MMCs and
the onset of damage in the matrix, of fracture of the fibers, and of interfacial decohesion
at the fiber–matrix interfaces show a clear dependence on the fiber arrangement.

In general, models based on periodic multi-fiber volume elements are clearly superior to
ones making use of simple periodic arrangements of fibers, making the former the preferred
option in computational periodic homogenization of continuously reinforced, unidirectional
composites.

Cross-Ply and Angle-Ply Composites

Another group of composite materials reinforced by continuous fibers that can be studied
to advantage by unit cell methods are laminates consisting of plies the thickness of which
is not much greater than the fiber diameter. The left side of fig. 6.11 depicts a unit cell for
a cross-ply laminate with double layers of fibers which is suitable for use with periodicity
and symmetry boundary conditions. In the the center a minimum reduced cell for cross
ply laminates with one fiber layer per ply is shown, which requires the use of symmetry
boundary conditions. A unit cell for studying angle-ply laminates with a general ply angle
β via periodicity boundary conditions is displayed on the right side of fig. 6.11. Extending
models of the above type to multi-fiber cells with randomly arranged fiber positions appears
feasible.

0o ply
90 o plies

Figure 6.11: Unit cell for a double layer cross-ply laminate (left), minimum reduced cell for a
cross ply-laminate (center) and unit cell for an angle-ply laminate with ply angle β (right).

Unit cells with symmetry boundary conditions were used for studying the thermome-
chanical behavior of cross ply laminates, e.g., by Lerch et al. (1991) as well as Ismar and
Schröter (2000), and volume elements with periodicity BC, e.g., by Soni et al. (2014). For
unit cell studies of angle ply laminates see, e.g., Xia et al. (2003) and Abolfathi et al.
(2008).

Fabric Reinforced Composites

Periodic homogenization has played an important role in modeling the behavior of fabric-
reinforced composites, i.e., materials containing woven, braided, or knitted reinforcements,
see, e.g., Ivanov and Lomov (2020). In such “textile composites” the reinforcing phase

123
takes the form of textile-like structures consisting of interlacing bundles of continuous
fibers (tows). Unit cell and reduced cell models for such materials are typically based on
modeling fiber bundles as a “mesophase” with smeared-out material properties, which, in
turn, are obtained from analyzing unidirectionally continuously reinforced composites93 .
Free surface boundary conditions are specified for the out-of-plane faces of the cell, and
a number of cells may be stacked on top of each other in order to account for in-plane
offsets and constraint effects between the layers (Byström et al., 2000). Symmetry bound-
ary conditions can be specified for the in-plane faces, giving reasonably small unit cells as
shown in fig. 6.12, and combinations of periodicity, symmetry and antisymmetry bound-
ary conditions may be used to achieve very small volume elements, see, e.g., (Tang and
Whitcomb, 2003). Such models, however, are restricted to handling in-plane normal and
thermal loads. By applying extended periodicity boundary conditions to larger unit cells
macroscopic rotational degrees of freedom can be introduced to allow studying all defor-
mation modes, including the macroscopic warping and twisting of the laminae.

Tows
Matrix Composite

Figure 6.12: Schematic of a reduced cell using symmetry boundary conditions for modeling a
plain weave lamina. The tow region (left), the matrix region (center) and the “assembled” unit
cell (right) are shown.

There is a wide range of weaves as well as knitting and braiding architectures that can
be modeled with PMAs. Unit cells for woven, knitted and braided composites tend to
be fairly complex geometrically, may be difficult to mesh for Finite Element analysis and
may pose considerable computational requirements, especially when nonlinear behavior is
to be studied. Over the past thirty years a large number of studies have been published
on unit cell modeling of fabric-reinforced composites, see, e.g., Cox and Flanagan (1997),
Huang and Ramakrishna (2000), Tang and Whitcomb (2003) or Lomov et al. (2006), and
on software for generating appropriate unit cells (Robitaille et al., 2003; Sherburn, 2007).
Also, Gager and Pettermann (2012) proposed shell element-based unit cells for textile
composites that can significantly reduce the modeling effort.
93
This modeling strategy obviously is a type of multiscale modeling as discussed in Chapter 9.

124
6.5 Periodic Models for Composites Reinforced by
Short Fibers
The overall symmetry of short-fiber-reinforced composites in many cases is isotropic (for
random fiber orientations) or transversally isotropic (for aligned fibers, planar random
fibers and other fiber arrangements with axisymmetric orientation distributions). How-
ever, processing conditions can give rise to a wide range of fiber orientation distributions
and, consequently, lower overall symmetries (Allen and Lee, 1990). The thermoelastic and
thermoelastoplastic behavior of aligned short-fiber-reinforced composites has been success-
fully estimated by Mori–Tanaka methods94 , which can also be extended to nonaligned
fibers and reinforcements showing an aspect ratio distribution, compare Section 2.6. Such
mean-field approaches are, however, limited in resolving details of fiber arrangements, es-
pecially for inelastic material behavior. At present the most powerful tools for studying
the influence of fiber shapes and orientations, of clustering effects, of the interaction of
fibers of different sizes, and of local stress and strain fields between neighboring fibers are
periodic microfield methods. Platelet-reinforced composites can be described by analogy
to short-fiber-reinforced materials.

Composites Reinforced by Aligned Short Fibers

In contrast to continuously reinforced composites, the phase arrangements of discontinu-
ously reinforced materials are inherently three-dimensional. The simplest three-dimensional
PMA models of aligned short-fiber-reinforced composites have used periodic square ar-
rangements of non-staggered or staggered aligned cylindrical fibers95 , see, e.g., Levy and
Papazian (1991) and compare fig. 6.13. Such geometries are relatively simple to set up
and do not pose major computational requirements, but are rather restrictive in terms
of fiber arrangements and load conditions that can be handled. By using larger volume
elements supporting periodicity boundary conditions the full thermomechanical behavior
of the fiber arrangements can be studied. Analogous microgeometries based on periodic
hexagonal arrangements of non-staggered (Järvstråt, 1992) or staggered (Tucker and Liang,
1999) fibers as well as ellipsoidal cells aimed at describing composite “ellipsoid assembly”
microgeometries (Järvstråt, 1993) were also proposed.

For many materials characterization studies, a more economical alternative to the above
three-dimensional unit cells takes the form of axisymmetric models describing the axial
behavior of non-staggered or staggered arrays of aligned cylindrical short fibers in an ap-
proximate way. The basic idea behind these models is replacing unit cells for square or
hexagonal arrangements by circular composite cylinders of equivalent cross sectional area
(and volume fraction) as sketched in Fig. 6.14. The resulting axisymmetric cells are not
proper unit cells, because they overlap and are not space filling. In addition, their as-
sociation with three-dimensional phase arrangements is somewhat tenuous — note that
they do not show the same transverse fiber spacing as the “equivalent” periodic arrange-
ments in Fig. 6.14 — and they are severely limited in the type of loading conditions they
94
For a comparison between unit cell and analytical predictions for the overall elastic properties of
short-fiber-reinforced composites see, e.g., Tucker and Liang (1999).
95
Such square arrangements give rise to tetragonal overall symmetry, and, consequently, the transverse
overall properties are direction dependent.

125
Figure 6.13: Three-dimensional cells for modeling non-staggered (left) and staggered (right)
square arrangements of aligned short fibers. The shaded reduced cells, which follow Levy and
Papazian (1991), require symmetry B.C., cells outlined in bold are suitable for periodicity B.C.

can handle96 . However, they have the advantage of significantly reduced computational
requirements, which has made them especially suitable for qualitative studies of highly
nonlinear behavior such as damage.

Symmetry boundary conditions are used for the top and bottom faces of the cells shown
in Fig. 6.14. The circumferential surfaces must be chosen to enforce identical cross sectional
areas along the axial direction for any aggregate of cells. In the case of non-staggered fibers
this can be easily done by specifying symmetry-type boundary conditions, eqns. (6.6), for
the outer surfaces.

For staggered arrangements a pair of cells with different fiber positions is considered, the
total cross sectional area of which is required to be independent of the axial coordinate97 .
By choosing the two cells making up the pair such that they are antisymmetric with respect
to a pivot point P the behavior of the staggered arrangement can be described by a single
cell with an antisymmetric outer (E-) face, U and L being nodes on this face that are
positioned symmetrically with respect to the pivot. Using the nomenclature of fig. 6.15
96
Axisymmetric cells are limited to load cases in which the deformed shape of the cell remains axisym-
metric. These include uniaxial stress and strain, volumetric stress, “in plane hydrostatic stress” (where
a homogeneous radial stress is prescribed) and isochoric strain (obtained, in the small strain regime, by
applying constant normal displacements at the “upper” and “outer” faces of, vN = 2uE , respectively).
97
As originally proposed by Tvergaard (1990) the nonlinear displacement boundary conditions in
eqn. (6.18) were combined with antisymmetry traction B.C.s for use with a hybrid FE formulation.

126
Figure 6.14: Periodic arrays of aligned non-staggered (top) and staggered (bottom) short fibers
and corresponding axisymmetric cells (left: sections in transverse plane; right: sections parallel
to fibers).

and the notation of eqns. (6.5) to (6.7), this leads to nonlinear relations for the radial
displacements u and linear constraints for the axial displacements v at the outer surface,

(rU + uU )2 + (rL + uL )2 = 2(rP + uP )2 and vU + vL = 2vP , (6.18)

respectively, where r is the radius of the undeformed cell. Nonlinear constraints such
as eqn. (6.18), however, tend to be cumbersome to use and are not widely available in
FE codes98 . For typical small-strain problems the boundary conditions for the radial
displacements at the outer surfaces of the cylinders, eqn. (6.18), can be linearized without
major loss in accuracy to give sets of linear constraint equations

uU + uL = 2uP and vU + vL = 2vP , (6.19)

which can be seen to be formally identical to the antisymmetry boundary conditions de-
scribed by eqn. (6.7).

Axisymmetric cell models of the types shown in fig. 6.14 were the workhorses of PMA
studies of short-fiber-reinforced composites in the 1990s, see, e.g., Povirk et al. (1992)
or Tvergaard (1994). Typically, descriptions using staggered arrangements allow a wider
range of microgeometries to be covered, compare Böhm et al. (1993) or Tvergaard (2003),
and give more realistic descriptions of actual composites. Both staggered and non-staggered
axisymmetric models can be extended to studying a considerable range of arrangements
98
For an example of the use of the nonlinear B.C.s described in eqn. (6.18) with a commercial FE code
see Ishikawa et al. (2000), where cells of truncated cone shape are employed to describe bcc arrangements
of particles.

127
 N NE
N NE
~s
NW U NW U

E
P E
W W P
z z
L −s
~
SW SW L
S S SE
r SE r
undeformed deformed
Figure 6.15: Axisymmetric cell for staggered arrangement of short fibers: undeformed and de-
formed shapes as used in eqn. (6.18).

incorporating aligned fibers of different size and/or aspect ratio by coupling two or more
different cells via the condition of keeping the cross sectional area of the aggregate inde-
pendent of the axial coordinate (Böhm et al., 1993).

Axisymmetric cells and simple three-dimensional unit cells have been used success-
fully for studying the nonlinear thermomechanical behavior of aligned short-fiber-reinforced
MMCs, e.g., with respect to the their stress–strain responses, to the pseudo-Bauschinger
effect, and to thermal residual stresses. They have provided valuable insight into causes
and effects of matrix, interface and fiber damage. Over the past years, however, multi-
fiber unit cells have become the standard tool for studying composites reinforced by aligned
short fibers.

Composites Reinforced by Nonaligned Short Fibers

Among the first unit cell studies of materials reinforced by nonaligned discontinuous fibers
was work based on three-dimensional models of composites reinforced by alternatingly
tilted misaligned fibers (Sørensen et al., 1995) and plane-stress models describing planar
random short fibers (Courage and Schreurs, 1992). Multi-fiber volume elements of com-
posites containing nonaligned fibers have started coming into their own in the 2000s. It
is worth noting that the situation with respect to mean-field models for composites rein-
forced by short fibers is not fully satisfactory for the elastic range, compare Section 2.6,
and MFAs have found limited application for work on inelastic thermomechanical behavior.

At present the most powerful continuum modeling strategy for nonaligned short-fiber-
reinforced composites consists of using three-dimensional multi-fiber volume elements in
which the fibers are randomly positioned and oriented such that the required ODF is ful-
filled to a suitable level of approximation. If the geometries are periodic, they are obviously
suitable for periodic homogenization. Among early reports for such multi-fiber unit cells
were, e.g., Lusti et al. (2002) and Böhm et al. (2002) for spatially random fiber orientations
and Duschlbauer et al. (2006) and Iorga et al. (2008) for planar random fiber orientations.
Models of this type can also account for the distributions of fiber sizes and/or aspect
ratios. Such modeling approaches tend to pose considerable challenges in generating ap-

128
propriate fiber arrangements at non-dilute volume fractions due to geometrical frustration,
relatively large volume elements being required for handling periodic microgeometries with
fiber aspect ratio in excess of, say, 5. The meshing of the resulting phase arrangements for
use with the FEM may also be difficult, compare Shephard et al. (1995), and analyzing
the mechanical responses of the resulting cells may require considerable computing power,
especially for nonlinear constituent behavior. The first studies of this type were, accord-
ingly, restricted to the linear elastic range, where the BEM has been found to answer well,
see, e.g., Banerjee and Henry (1992). Commercial micromechanics codes such as Digimat
(Hexagon AB, Stockholm, Sweden) are now available that provide extensive capabilities
for setting up and solving periodic models with considerable numbers of nonaligned short
fibers.

As a simple example of a multi-fiber unit cell for a composite reinforced by nonaligned

short fibers, fig. 6.16 shows a cube-shaped cell that contains 15 randomly oriented cylindri-
cal fibers of aspect ratio a = 5, supports periodicity boundary conditions, was generated
by random sequential addition, and is discretized by tetrahedral elements. The phase ar-
rangement is set up in such a way that spheroidal fibers of the same aspect ratio, volume
fractions, center points, and orientations can also be used in order to allow studying fiber
shape effects (Böhm et al., 2002).

Figure 6.16: Periodic volume element for a composite reinforced by randomly oriented short fibers
(Böhm et al., 2002). The nominal fiber volume fraction is ξ = 0.15 and the 15 cylindrical fibers
in the cell have an aspect ratio of a = 5.

Table 6.2 lists predictions for the overall elastic response of a composite reinforced
by randomly positioned and oriented fibers obtained by analytical estimates, the Hashin–
Shtrikman bounds (Hashin and Shtrikman, 1963) and results obtained with the above type
of unit cell, for which data are shown pertaining to cylindrical and spheroidal fibers. In-
terestingly, noticeable differences were found between the predictions obtained for given

129
Table 6.2: Overall elastic properties of a SiC/Al2618 MMC reinforced by randomly oriented
fibers (a = 5, ξ=0.15) as predicted by the Hashin–Shtrikman bounds (HSB), by the extended
Mori–Tanaka method (MTM) of Wei and Edwards (1999), by the classical self-consistent scheme
(CSCS) of Berryman (1980), by the Kuster and Toksöz (1974) model (KTM) and by multi-fiber
unit cells of the type shown in fig. 6.16 (MFUC), which contain 15 spheroidal or cylindrical fibers
(Böhm et al., 2002).

E∗ ν∗
[GPa] []
fibers 450.0 0.17
matrix 70.0 0.30
HSB/lo 87.6 0.246
HSB/hi 106.1 0.305
MTM 89.8 0.285
CSCS 91.2 0.284
KTM 90.3 0.285
MFUC/sph 89.4 0.285
MFUC/cyl 90.0 0.284

values of the reinforcement volume fraction and aspect ratio, even when fibers of the above
two shapes occupy the same positions and show the same orientations: the spheroidal fibers
generally give rise to more compliant responses, especially in inelastic regimes. Useful re-
sults were obtained despite the low number of fibers in the unit cell and good agreement
with analytical descriptions was achieved in terms of the orientation dependence of the
stresses in individual fibers, compare fig. 2.4. Whereas the phase arrangement shown in
fig. 6.16 does not contain a sufficient number of fibers for approaching representativeness
even in the elastic range, recent studies using much larger volume elements have markedly
improved this situation, see, e.g., Tian et al. (2015).

Three-dimensional models based on volume elements that contain multiple fibers in

general form the basis of the most powerful and versatile numerical models of aligned or
nonaligned short-fiber-reinforced composites.

6.6 Periodic Models for Particle-Reinforced Compos-

ites
Particle-reinforced composite materials often show rather irregular particle shapes, and
anisotropy in the microgeometry as well as in the overall responses may be introduced
by processing effects such as extrusion textures. Nevertheless, for materials reinforced
by statistically uniformly distributed, (approximately) equiaxed particles isotropic macro-
scopic behavior is a useful approximation that has been applied in many modeling studies.
The thermoelastic behavior of macroscopically isotropic composites has been successfully
described by mean-field methods that approximate the particles as spheres, see Section
2.3. Extensions of the mean-field solutions into the nonlinear range are available, compare
Section 2.9, but they are subject to some limitations in predicting the overall thermome-

130
chanical response in the post-yield regime. In addition, mean-field models are limited in
accounting for many particle shape, clustering, and size distribution effects and cannot
resolve local fluctuations of the stress and strain fields. As a consequence, the past 30
years have witnessed marked interest in periodic homogenization models for studying the
thermomechanical behavior of particle-reinforced composites.

One issue in applying periodic microfield approaches to the modeling of particle-reinfor-

ced materials with macroscopically isotropic behavior is that there exists no simple peri-
odic three-dimensional phase arrangement that shows matrix–inclusion topology and is
inherently elastically isotropic99 . Together with the wide variation in microgeometries
and particle shapes in actual materials, this gives rise to non-trivial tradeoffs between
keeping computational requirements reasonably low (favoring simple particle shapes and
two-dimensional or simple three-dimensional microgeometries) and obtaining sufficiently
realistic models for a given purpose (best fulfilled by three-dimensional volume elements
containing considerable numbers of randomly arranged particles of complex shape). In
many respects periodic models of particle-reinforced composites are subject to similar con-
straints and use analogous approaches as work on short-fiber-reinforced composites.

Many three-dimensional unit cell studies of generic microgeometries for particle-reinfor-

ced composites have been based on simple cubic (sc), face centered cubic (fcc) or body
centered cubic (bcc) arrays of spherical, cylindrical or cube-shaped particles, compare,
e.g., Hom and McMeeking (1991). By invoking the symmetries of these arrays and using
symmetry as well as antisymmetry boundary conditions, rather small “reduced” cells can be
obtained for materials characterization100 , compare Weissenbek et al. (1994) and fig. 6.17.
In addition, work employing hexagonal or tetrakaidecahedral arrays of particles (Rodin,
1993) and Voronoi cells for cubic arrays of particles (Li and Wongsto, 2004) was reported.

Figure 6.17: Examples of reduced cells for particle-reinforced composites using cubic arrays of
inhomogeneities: s.c. arrangement of cubes, b.c.c. arrangement of cylinders and f.c.c. arrangement
of spheres (Weissenbek et al., 1994).
99
Even though cells having the shape of pentagonal dodecahedra and icosahedra have the appropriate
symmetry properties (Christensen, 1987), they are not space filling.
100
Among the reduced cells shown in fig. 6.17 the one on the left uses symmetry boundary conditions and
is restricted in the load cases that can be handled. The other two cells employ antisymmetry B.C.s and
can be used more freely. Unit cells suitable for periodicity boundary conditions are considerably larger,
but allow unrestricted modeling of the full thermomechanical response of cubic phase arrangements.

131
 Effective elastic tensors obtained by periodic homogenization materials containing cubic
arrays of particles show cubic symmetry101 . As a consequence, the corresponding moduli
do not necessarily fulfill the Hashin–Shtrikman bounds for macroscopically isotropic ma-
terials, and often lie outside the three-point bounds, compare table 6.3. Among the cubic
arrays of particles, sc ones are easiest to handle by unit cell models, but their mechanical
behavior shows a more pronounced anisotropy than do bcc and fcc arrangements.

Axisymmetric cells describing staggered or non-staggered arrangements of particles can

be used for materials characterization of particle-reinforced composites in full analogy with
short-fiber-reinforced materials, compare figs. 6.14 and 6.15. By appropriate choice of the
dimensions of the axisymmetric cells sc, bcc and fcc arrays of particles can be approximated
to some extent, and a range of axisymmetric particle shapes can be studied. Such models
were a mainstay of numerical modeling of materials containing particulate inhomogeneities,
see, e.g., Bao et al. (1991) or LLorca (1996). A related type of model are spherical cells
(Guild and Kinloch, 1995).

Due to their low computational requirements and the relative ease of incorporating
irregular particle shapes, planar unit cell models of particle-reinforced materials have also
been used to a considerable extent. Typically, plane stress models (which actually describe
thin “reinforced sheets” or the stress states at the surface of inhomogeneous bodies) show
a more compliant and plane strain models (which describe reinforcing by aligned continu-
ous fibers rather than by particles) show a stiffer overall response than three-dimensional
descriptions, compare table 6.3. With respect to the overall behavior, plane stress analy-
sis may be preferable to plane strain analysis, compare (Weissenbek, 1994), but no two-
dimensional model gives satisfactory results in terms of the predicted microstress and
microstrain distributions102 (Böhm and Han, 2001). Axisymmetric cell models typically
provide better results for the behavior of particle-reinforced composites than do planar ones.

During the past thirty years studies based on increasingly complex, three-dimensional
phase arrangements have assumed a prominent role in the literature. Gusev (1997) used
Finite Element methods in combination with unit cells containing up to 64 statistically
positioned particles to describe the overall behavior of elastic particle-reinforced compos-
ites. Hexahedral unit cells containing up to 10 particles in a perturbed cubic configuration
(Watt et al., 1996) as well as cube shaped cells incorporating at least 15 spherical parti-
101
Cubic elastic symmetry is a special case of orthotropy with equal responses in all principal directions.
It gives rise to direction dependent Young’s and shear moduli and is described by three independent
parameters. For a discussion of the elastic anisotropy of cubic materials see, e.g., Cazzani and Rovati
(2003). The conduction and diffusion behavior of materials with cubic symmetry is isotropic.
102
One issue giving rise to difficulties in describing macroscopically isotropic phase arrangements with
two-dimensional models is the latters’ lack of a proper hydrostatic load case. Its closest planar equivalent,
equi-biaxial loading, is controlled by the plane strain bulk modulus, KT , or the plane stress bulk modulus,
respectively, both of which differ from the “three-dimensional” bulk modulus K.
In the plastic regime, plane stress configurations tend to give much higher equivalent plastic strains in
the matrix and weaker overall hardening than do plane strain and generalized plane strain models using
the same planar phase geometry. The regions of concentrated strains that underlie this behavior depend
strongly on the geometrical constraints, compare, e.g., Iung and Grange (1995), Gänser et al. (1998), Böhm
et al. (1999) or Shen and Lissenden (2002). For planar matrix–inclusion configurations under transverse
loading the equivalent plastic strains tend to concentrate in bands oriented at 45◦ to the loading directions,
compare fig. 6.10, whereas the patterns are qualitatively different in three-dimensional geometries.

132
cles in quasi-random arrangements (Böhm et al., 1999; Böhm and Han, 2001), compare
fig. 6.18, or clusters of particles (Segurado et al., 2003; Lee et al., 2011b) were used in
Finite Element based studies of elastoplastic particle-reinforced MMCs and related ma-
terials. Three-dimensional simulations involving high numbers of particles have been re-
ported, e.g., for investigating the elastic behavior of composites (Michel et al., 1999), for
studying brittle matrix composites that develop damage (Zohdi and Wriggers, 2001), and
for rubber-reinforced polymers (Fond et al., 2001).

Figure 6.18: Unit cell for a particle-reinforced MMC (ξ=0.2) containing 20 spherical particles in
a quasi-random arrangement that is suitable for using periodicity B.C. (Böhm and Han, 2001).

Table 6.3 provides comparisons between the predictions for the effective Young’s moduli,
Poisson numbers and coefficients of thermal expansion predicted by a number of bounding
methods, mean-field approaches and periodic homogenization models. The data pertain
to a particle-reinforced SiC/Al MMC (elastic contrast cel ≈ 5.5) that is subjected to
macroscopic uniaxial stress; the particle volume fraction is chosen as ξ = 0.2. The loading
directions for the cubic arrays are identified by Miller indices, with <ijk> denoting a
family of equivalent directions, results being provided for the <100> (edges), <110> (face
diagonals) and <111> (space diagonals) orientations103 . Despite the moderate elastic
contrast and particle volume fraction considered, fewer than half of the predictions for
the Young’s moduli obtained from the cubic arrays fall within the three-point bounds
pertaining to randomly positioned, identical spheres and some of them even violate the
lower Hashin–Shtrikman bounds. The axisymmetric cells (axi/sc, axi/bcc and axi/fcc)
and the planar multi-particle models (2D/PST using plane stress and 2D/PSE using plane
strain kinematics) do even worse. The two results for three-dimensional multi-particle
volume elements (3D/sr pertains to the arrangement shown in fig. 6.18 loaded in an edge
103
This choice covers the extremal values of the Young’s moduli of the cubic arrays, which are known to
be associated with loading in the <100> and <111> directions (Nye, 1957).

133
Table 6.3: Overall thermoelastic properties of a particle-reinforced SiC/Al MMC (spherical par-
ticles, ξ=0.2) predicted by the Hashin–Shtrikman (HSB) and three-point (3PB) bounds, by the
Mori–Tanaka method (MTM), the generalized self-consistent scheme (GSCS), the differential
scheme (DS), and Torquato’s three-point estimates (3PE), as well as by unit cell analysis using
three-dimensional cubic arrays, axisymmetric cells approximating sc, bcc and fcc geometries, two-
dimensional multi-particle models based on plane stress (2D/PST) and plane strain (2D/PSE)
kinematics, as well as three-dimensional models containing 20 randomly positioned spherical par-
ticles (3D/sr and 3D/eac).

E∗ E ∗ <100> E ∗ <110> E ∗ <111> ν∗ αA∗

[GPa] [GPa] [GPa] [GPa] [] [K ×10−6 ]

−1

particles 429.0 — — — 0.17 4.3

matrix 67.2 — — — 0.35 23.0
HSB/lo 91.2 — — — 0.285 16.8
HSB/hi 115.4 — — — 0.340 18.5
3PB/lo 91.7 — — — 0.323 18.4
3PB/hi 94.2 — — — 0.328 18.5
MTM 91.2 — — — 0.328 18.5
GSCS 91.9 — — — 0.327 18.5
DS 92.7 — — — 0.326 18.4
3PE 92.2 — — — 0.327 18.5
sc — 96.5 90.6 89.5 18.5
bcc — 90.1 91.8 92.2 18.5
fcc — 90.2 91.7 92.1 18.5
axi/sc 95.4 — — — 18.6/18.6
axi/bcc 87.9 — — — / 18.4/18.8
axi/fcc 88.1 — — — 18.1/19.0
2D/PST 85.5 — — — 0.334
2D/PSE 98.7 — — —
3D/sr 92.4 — — — 0.326
3D/eac 92.2 — — — 0.327 18.5

direction and 3D/eac uses ensemble averaging over 3 arrangements of 20 particles each
followed by evaluation of the closest isotropic elasticity tensor as discussed in Section
5.4), however, fall within both the Hashin–Shtrikman and the three-point bounds. The
individual predictions for the moduli used for obtaining result 3D/eac differed by nearly
1.5%, which is somewhat in excess of what might be expected from the nonlocal estimates
of Drugan and Willis (1996). The above value also gives an idea of the elastic anisotropy
of such volume elements.

The nonlinear macroscopic mechanical responses of particle-reinforced composites tend

to be more sensitive to the phase arrangements than is their elastic behavior. This is
demonstrated in fig. 6.19 where unit cell predictions for the macroscopic uniaxial tensile
elastoplastic stress–strain responses of the same Al matrix as used in fig. 6.9 (von Mises
plasticity with linear hardening), reinforced by 40 vol.% of spherical, linear elastic SiC
particles (identical to the ones in table 6.3), are compared. For the sc, bcc and fcc cubic
arrays the responses to tensile loads acting in the <100>, <110> and <111> each are

134
 7x107

6x107

APPLIED STRESS [MPa]

5x107

4x107
sc, <100>
7
sc, <110>
3x10 sc, <111>
bcc, <100>
bcc, <110>
2x107 bcc, <111>
fcc, <100>
fcc, <110>
1x107 fcc, <111>
MP/eav
matrix
0
0 0.001 0.002 0.003 0.004 0.005 0.006
STRAIN []

Figure 6.19: Elastoplastic response of a particle-reinforced SiC/Al MMC (ξ=0.4, matrix as in

fig. 6.9, particles as in table 6.3) to uniaxial loading as predicted by PMA models using sc, bcc
and fcc arrays as well as multi-particle volume elements (MP/eav).

shown together with the ensemble averaged predictions for loads in the 3 edge directions of
two volume elements containing 20 randomly positioned particles each, 3D/ea. Somewhat
similarly to the case of fiber-reinforced composites, fig. 6.9, the macroscopic hardening
behavior of the multi-particle models is considerably more pronounced than that of the
cubic arrays in most situations, then main exception being the sc array subjected to tensile
uniaxial loading in <100>, which is excessively stiff.

Like fiber-reinforced MMCs, particle-reinforced composites typically display highly in-

homogeneous microscopic stress and strain distributions, especially in nonlinear regimes.
As an example, fig. 6.20, shows the predicted equivalent plastic strains of the elastoplastic
matrix inside the multi-particle unit cell model depicted in fig. 6.18. This behavior tends
to give rise to microscopic “structures” that can be considerably larger than individual
particles, leading to longer ranged interactions between inhomogeneities104 . The latter, in
turn, underlie the need for larger volume elements for studying composites with nonlinear
matrix behavior mentioned in Chapter 5.

Periodic volume elements are highly suitable for carrying out numerical experiments
that explore the influence of different aspects of the microgeometry of particle-reinforced
composites on their microscopic and macroscopic responses. For example, Rasool and
Böhm (2012) and Böhm and Rasool (2016) studied model composites that contain equal
volume fractions of randomly positioned and, where applicable, randomly oriented, identi-
104
Due to the filtering effect of periodic phase arrangements mentioned in Section 6.1, the length of such
features can exceed the side length of the unit cell only in certain directions, which may induce a spurious
direction dependence into the predicted large-strain behavior of sub-RVE unit cells. Cubic arrays give
rise to highly regular, long range patterns of the microfields that may strongly favor such structures; the
maxima of microfields in cubic arrays depend on the loading direction both in their location and magnitude.

135
 SECTION FRINGE PLOT
EPS.EFF.PLASTIC INC.12

5.0000E-02
4.0000E-02
3.0000E-02
2.0000E-02
1.0000E-02

SCALAR MIN: 6.6202E-04

SCALAR MAX: 8.1576E-02

Figure 6.20: Predicted distribution of equivalent plastic strain in the matrix of a particle-
reinforced MMC (ξ=0.2) subjected to uniaxial tensile loading (in left–right direction) obtained
for a unit cell with 20 spherical particles in a quasi-random arrangement (Böhm and Han, 2001).

cal particles having the shapes of spheres, cubes, regular octahedra and regular tetrahedra,
respectively. Figure 6.21 shows the resulting predictions for the responses to a single, non-
symmetric, uniaxial stress loading cycle, clear effects of the particle shape on the hardening
behavior and on the residual strains being evident.
NORMALIZED MACROSCOPIC STRESS []

SPH
1.5 OCT
CUB
1 TET

0.5

-0.5

-1

-1.5

-2

-2.5
-0.005 0 0.005 0.01 0.015
MACROSCOPIC STRAIN []

Figure 6.21: PMA predictions for the responses to a uniaxial load cycle of elastoplastic, particle-
reinforced MMCs (ξ=0.2) containing identical spherical (SPH), cube-shaped (CUB), octahedral
(OCT) and tetrahedral (TET) inhomogeneities, respectively.

136
 Periodic multi-inhomogeneous volume elements can be directly applied to studying
composites reinforced by coated particles, compare Böhm (2019). Resolving thin inter-
phases, however, may require rather fine discretizations and lead to rather large models.

A comparison of the modeling approaches for studying the mechanical behavior of

particle-reinforced composites discussed in this section clearly brings out three-dimensional
multi-particle models as being superior to the other options for evaluating both macroscopic
responses and microscopic stress and strain fields. Their main drawback, however, are their
computational costs, especially for nonlinear problems.

6.7 Periodic Models for Porous and Cellular Materi-

als
Due to their relevance to the ductile damage and failure of metallic materials elastoplas-
tic porous materials have been the subjects of a considerable number of PMA studies105 .
Generally, modeling concepts for porous materials are closely related to the ones employed
for particle-reinforced composites, the main difference being that the shapes of the voids
may evolve significantly through the loading history106 . Axisymmetric cells of the types
discussed in Sections 6.5 and 6.6, compare, e.g., Koplik and Needleman (1988) or Gărăjeu
et al. (2000), and three-dimensional unit cells based on cubic arrangements of voids, see,
e.g., McMeeking and Hom (1990) or Segurado et al. (2002b), have been used in the ma-
jority of pertinent PMA studies. In addition, studies of void growth in ductile materials
based on multi-void cells and using geometry data from serial sectioning were reported
(Shan and Gokhale, 2001).

In cellular materials, such as foams, wood and trabecular bone, the volume fraction of
the solid phase is very low (often amounting to no more than a few percent) and the void
phase may be topologically connected (open cell foams), unconnected (closed cell foams,
syntactic foams), or both of the above (e.g., hollow sphere foams). The linear elastic
regime of cellular materials in many cases is limited to a very small range of macroscopic
strains. Furthermore, marked differences tend to be present between tensile and compres-
sive inelastic macroscopic responses: As strains increase, gross shape changes of the cells
typically take place on the microscale, with large bending deformations, the formation of
105
Many models and constitutive descriptions of the ductile damage and failure of metals, among them
contributions by Rice and Tracey (1969), Gurson (1977), Tvergaard and Needleman (1984), Gologanu
et al. (1997) as well as Benzerga and Besson (2001), are based on micromechanics-based considerations of
the growth of pre-existing voids in a ductile matrix.
106
The evolution of the shapes of initially spheroidal voids under non-hydrostatic loads has been the
subject of studies by mean-field type methods, compare Kailasam et al. (2000). Such models use the
assumption that initially spherical pores will stay ellipsoids throughout the deformation history, which ax-
isymmetric cell analysis (Gărăjeu et al., 2000) has shown to be an excellent approximation for axisymmetric
tensile load cases. For compressive loading, however, initially spherical pores may evolve into markedly
different shapes and contact between the walls of pores tends to play an increasing role as the void volume
fraction is more and more reduced (Segurado et al., 2002b).
Void size effects (Tvergaard, 1996) and void coalescence (Faleskog and Shih, 1997) introduce additional
complexity when the ductile damage and failure of metals is studied.

137
plastic hinges, elastic as well as plastic buckling, and brittle failure of struts and cells walls
play major roles, especially under compressive loading. For compression-dominated load
cases this regime tends to give rise to a stress plateau on the macroscale, which underlies
the favorable energy absorption properties of many cellular materials. At some elevated
strain the effective stiffness under compression typically rises sharply, the cellular structure
having collapsed to such an extent that many cell walls or struts are in contact and the
void volume fraction has decreased dramatically. No comparable behavior is present under
tensile or shear loading.

Periodic microfield methods are generally well suited to studying the thermomechanical
behavior of cellular materials. The widely used analytical results of Gibson and Ashby
(1988) were derived by analytically studying arrangements of beams (for open cell foams)
or plates (for closed cell foams). They give the macroscopic moduli and other physical
properties of cellular materials as power laws of the type
E∗  ρ∗ n
∝ (m) (6.20)
E (m) ρ
in terms of the relative density107 , ρ∗ /ρ(m) .

In Finite Element based models of cellular materials the solid phase may be described
either via continuum or via structural elements (shells for the cell walls of closed cell foams
and beams for the struts in open cell foams)108 . The influence of the gas filling closed
cells can also be accounted for within such an FE setting (Mills et al., 2009). In order to
model buckling and compaction phenomena, unit cells for cellular materials require explicit
provision for handling large deformations of and contact (including self-contact) between
cell walls or struts. Boundary conditions must be applied to the unit cells in such a way
that they do not interfere with relevant buckling modes. Specifically, models must be suf-
ficiently large for non-trivial deformation and buckling patterns to develop, or Bloch wave
theory (Gong et al., 2005) must be used to account for long wavelength buckling modes109 .

The geometrically simplest cellular materials are regular honeycombs, the in-plane be-
havior of which can be modeled with planar hexagonal cell models. Somewhat less ordered
two-dimensional arrangements have been used for studying the crushing behavior of soft
woods (Holmberg et al., 1999), and highly irregular planar arrangements, of the type shown
107
The exponent n in eqn. (6.20) allows inferring the dominant local deformation mode of a cellular
microgeometry, with the extreme cases n=1 and n = 2 implying deformation by axial stretching and by
bending, respectively, of struts in open cell foams. The Gibson–Ashby cell for open cell foams gives n=2
and, as a consequence, tends to underestimate the stiffness of actual open-cell foams.
Stretching-dominated cellular materials tend to be more weight-efficient than bending-dominated ones
(Deshpande et al., 2001).
108
At high levels of porosity it is typically necessary to account for the overlap of shell elements at edges
and of beam elements at vertices when evaluating the phase volume fractions of the discretized unit cells.
When structural finite elements are used in unit cell studies periodicity conditions must obviously be
enforced in terms of both translational and rotational degrees of freedom.
109
For perfectly regular structures such as hexagonal honeycombs the minimum size of a volume element
for capturing bifurcation effects can be estimated from extended homogenization theory (Saiki et al., 2002).
When symmetry boundary conditions are employed care must be taken that cell walls that may buckle do
not coincide with symmetry planes at the boundaries, and periodic contact may have to be provided for
if periodicity boundary conditions are used.

138
 Y

Z X

Figure 6.22: Planar periodic volume element for studying irregular cellular materials (Daxner,
2003).

in fig. 6.22, may be used to generically investigate aspects of the geometry dependence of
the mechanical response of cellular materials.

The modeling of typical closed cell foams, however, requires three-dimensional micro-
geometries. In the simplest cases generic phase arrangements based on cube-shaped cells,
see, e.g., Hollister et al. (1991), truncated cubes plus small cubes (Santosa and Wierzbicki,
1998), rhombic dodecahedra, regular tetrakaidecahedra (Grenestedt, 1998; Simone and
Gibson, 1998), compare fig. 6.23, as well as Kelvin, Williams and Weaire–Phelan (Kraynik
and Reinelt, 1996; Daxner et al., 2007) geometries110 may be used. Analogous regular
microgeometries have formed the basis for analytical and numerical “lattice models” of
open cell foams, see, e.g., Zhu et al. (1997), Shulmeister et al. (1998) and Vajjhala et al.
(2000) as well as fig. 6.24, the struts connecting the nodes of the cellular structures being
modeled by beam or solid elements111 . In addition, tetrahedral arrangements (Sihn and
Roy, 2005) and cubic arrangements of struts with additional “reinforcements” as well as
perturbed strut configurations (Luxner et al., 2005) have been covered by such studies.
Voronoi tesselations have become a common tool in modeling irregular open and closed
cell foams, and random Laguerre tessellations have been proposed for generating periodic
multi-cell models (Redenbach, 2009).

110
In Kelvin, Williams and Weaire–Phelan (Weaire and Phelan, 1994) foams some of the cell walls are
spatially distorted in order to minimize the total surface, whereas in polyhedral foams all cell walls are
planar. At low solid volume fractions these small distortions lead to noticeable differences in the overall
elastic response.
111
Models of foams employing regular arrangements of polyhedral cells are not macroscopically isotropic,
compare, e.g., Luxner et al. (2005). A specific issue of beam lattice models for open cell foams are the
limitations of Timoshenko beams that typically fully account for nonlinear material behavior in the axial
and bending stiffnesses only, but not in shear stiffness, see, e.g., Pettermann and Hüsing (2012).

139
Figure 6.23: Idealized closed-cell foam microstructure modeled by regular tetrakaidecahedra
(truncated octahedra).

The effects of details of the microgeometries of cellular materials (e.g., thickness distri-
butions and geometrical imperfections or flaws of cell walls or struts or the Plateau borders
formed at the intersections of cell walls), which can considerably influence the overall be-
havior, have been a fruitful field of research employing periodic homogenization, see, e.g.,
Grenestedt (1998) and Daxner (2003). Typically, unit cell analysis of cellular materials
with realistic microgeometries is rather complex and numerically demanding due to these
materials’ tendency to deform by local mechanisms and instabilities. However, analytical
solutions have been reported for the linear elastic behavior of some simple periodic phase
arrangements, see, e.g., Warren and Kraynik (1991) or Sullivan and Ghosn (2009).

The high flexibility of FE-based periodic homogenization has allowed, on the hand,
modeling a wide range of material behaviors of cellular materials, among them plasticity,
viscoelasticity (Pettermann and Hüsing, 2012) and metal creep (Oppenheimer and Dunand,
2007), and, on the other hand, studying fairly complex microgeometries. An example of
the latter describes hollow strut foams produced by coating a precursor cellular material
and then removing this “template”. Figure 6.24 shows a model for such a foam that is
based on a Weaire–Phelan geometry and was meshed by a combination of solid and shell
elements (Daxner et al., 2007).

Due to the inherent X-ray absorption contrast between matrix and voids, cellular ma-
terials are well suited to high-resolution computed tomography, giving access to “real-
structure” microgeometries that can be directly converted into voxel models (compare
Section 5.3) of open-cell or closed-cell foams112 , see, e.g., Maire et al. (2000) or Roberts
and Garboczi (2001). Alternatively, image processing methods my be used to generate a
surface model of the solid phase in the volume element to allow meshing with “standard”
FE techniques (Youssef et al., 2005; Young et al., 2008). Because the microgeometries
extracted from the experiments are, in general, not periodic, windowing approaches using
112
Roberts and Garboczi (2001) estimated the “systematic discretization error” to be of the order of 10%
in terms of the overall moduli for voxel-based models of elastic open-cell and closed-cell foams.

140
Figure 6.24: Weaire–Phelan model of an open cell foam with hollow struts (left) and detail of the
FE-model (right), from Daxner et al. (2007).

homogeneous or mixed uniform boundary conditions, compare Chapter 7, tend to be the

method of choice for such models, however.

A specific group of cellular materials amenable to PMA modeling are syntactic foams
(i.e., hollow spheres embedded in a solid matrix) and hollow sphere foams (in which the
spaces between the spheres are “empty”, giving them both an open-cell and a closed cell
flavor). Syntactic foams have been studied via axisymmetric or three-dimensional cell mod-
els that are based on closest packings of spheres (Rammerstorfer and Böhm, 2000; Sanders
and Gibson, 2003) and by multi-void volume elements (Böhm, 2019).

Recent advances in additive manufacturing have led to major interest in architected,

lattice or shell micro- and nanostructures, which are periodic by design at the microscale.
They are, accordingly, amenable to study by periodic homogenization, compare, e.g.,
Abueidda et al. (2017). However, for samples of finite size boundary effects may play
a major role in their behavior, which requires modeling by full structural models (Luxner
et al., 2005) or by second order homogenization (Desmoulins and Kochmann, 2017).

A further type of cellular material, trabecular (cancellous, spongy) bone, has attracted
considerable modeling interest for more than thirty years113 . Cancellous bone shows a
wide range of microgeometries, which can be idealized as beam or beam–plate configu-
rations (Gibson, 1985). In studying the mechanical behavior of trabecular bone, large
three-dimensional unit cell models based on tomographic scans of actual samples and us-
ing voxel-based or smooth discretization schemes have become fairly widely used, see, e.g.,
Hollister et al. (1991), Hollister et al. (1994) or van Rietbergen et al. (1999).

In addition to periodic homogenization, windowing methods, compare Chapter 7, and

embedding models, see Chapter 8, have been successfully employed for studying cellular
materials.
113
Like many materials of biological origin, bone is an inhomogeneous material at a number of length
scales. The solid phase of trabecular bone is inhomogeneous and can itself be studied with micromechanical
methods.

141
6.8 Periodic Models for Some Other Inhomogeneous
Materials
At the continuum level, the thermomechanical behavior of essentially any inhomogeneous
material can be studied by periodic homogenization techniques. For example, the elastic,
elastoplastic, creep and damage behaviors of polycrystals, of high speed steels and dual
phase steels, of intermetallics, of superalloys, and of graded materials have been the targets
of unit cell models. In addition, “smart materials”, such as piezoelectric composites and
shape memory alloys, and solid state phase transitions in general have been studied. A
comprehensive discussion of the models involved would by far exceed the present scope, so
that the present section is limited to a small number of examples.

Generic PMA Models for Clustered, Graded, and Interpenetrat-

ing Microgeometries
Many commercially important steels may be viewed as matrix–inclusion-type compos-
ites. For example, high speed steels (HSS) contain carbidic inclusions in a steel matrix,
with arrangements ranging from statistically homogeneous to highly clustered or layered
(meso)geometries. Such materials can be modeled like particle-reinforced MMCs. A
straightforward and flexible strategy for setting up generic planar model geometries for
such studies consists of tiling the computational plane with regular hexagonal cells which
are assigned to one of the constituents by statistical or deterministic rules. If required, the
shapes of the cells can be modified or randomly distorted and their sizes can be changed to
adjust phase volume fractions. Such a Hexagonal Cell Tiling (HCT) concept can be used,
e.g., to generate unit cells for analyzing layer structured HSSs, compare (Plankensteiner
et al., 1997) and fig. 6.25, and clustered or random arrangements of inhomogeneities in a
matrix. HCT and related models of matrix–inclusion topologies are, however, restricted to
low reinforcement volume fractions.

Figure 6.25: HCT cell models for a layer structured high-speed steel (left) and a functionally
graded material (right).

142
 Related concepts can be used to obtain generic geometries for planar models for study-
ing functionally graded materials114 (FGMs), in which the phase volume fractions run
through the full physically possible range (Weissenbek et al., 1997; Reiter et al., 1997), so
that there are regions that show matrix–inclusion microtopologies and others that do not,
compare fig. 6.25 (right). A further application of such models have been investigations of
the microstructure–property relationships in duplex steels, for which matrix–inclusion and
interwoven phase arrangements were compared115 , see Siegmund et al. (1995) as well as
Silberschmidt and Werner (2001). An FFT-based modeling scheme employing HCT-type
geometries was developed by Michel et al. (1999) explicitly for studying interpenetrating
microgeometries. There is no direct three-dimensional equivalent to HCT models, but
unit cells partitioned into cube-shaped or tetrakaidecahedral subregions can be used in
an analogous manner to obtain generic microgeometries (which, however, may be numeri-
cally expensive if meshes are chosen such that details of the microfields can be resolved).
A distantly related, but geometrically much more flexible approach based on tetrahedral
subregions was reported by Galli et al. (2008).

For topological reasons most interpenetrating phase composites (IPC) must be studied
with three-dimensional models. Beyond this restriction, however, fairly standard periodic
homogenization approaches can be used for investigating their effective thermomechanical
behavior, see, e.g., Dalaq et al. (2016) or Soyarslan et al. (2018).

PMA Models for Polycrystals

A large number of materials that are of technological interest are polycrystals and, ac-
cordingly, are inhomogeneous at some sufficiently small length scale. In the special case of
metals and metallic alloys phenomena associated with plastic flow and ductile failure can
be investigated at a number of length scales (McDowell, 2000), some of which are amenable
to study by continuum micromechanical methods.

Micromechanical methods, especially analytical self-consistent approaches, have been

applied to studying the mechanical behavior of polycrystals since the 1950s. The more re-
cent concept of studying polycrystals by full field approaches, among them periodic homog-
enization, appears quite straightforward in principle — a planar or three-dimensional vol-
ume element is partitioned into suitable subregions that correspond to individual grains116 .
For these grains appropriate material models must be prescribed, and suitable material as
well as orientation parameters must be assigned to them. The generation of appropriate
grain geometries may be based, e.g., on Voronoi tessellations using Poisson or hardcore dis-
114
In contrast to most other materials discussed in the present report, FGMs are statistically inhomo-
geneous by design, so that phase arrangements must be non-periodic in at least one direction, for which
free-surface boundary conditions are used on the unit cells. Clearly, volume averaging is of limited value
in modeling such materials.
115
In HSSs and FGMs particulate inhomogeneities are present at least in part of the volume fraction range,
so that planar analysis must be viewed as a compromise between modeling accuracy and computational
requirements. Duplex steels, in contrast, may show elongated aligned grains, which can be described well
by generalized plane strain analysis in the transverse plane.
116
Usually symmetry boundaries passing through grains must be avoided in models of this type because
they give rise to “twin-like” pairs of grains that are unphysical in most situations.

143
tributed seed points, or on modified Voronoi tessellations117 . Such microgeometries may be
linked to uniform and isotropic grain growth that stops where neighboring grains contact
each other, may be efficiently generated as well as meshed by appropriate software (Fritzen
et al., 2009), and can provide periodic or non-periodic volume elements. Alternatively, the
grain geometries can be based on experimental data obtained, e.g., by microscopy and
serial sectioning.

Anisotropic elasticity and crystal plasticity models are best suited to describing the
material behavior of the individual grains, the orientation of which can be described by
stochastic models. The number of grains required for achieving a given accuracy in terms
of the macroscopic elastic tensor has been shown to depend on the level of anisotropy of
the individual grains (Nygårds, 2003). Analysis involving crystal plasticity tends to pose
considerable demands on computational resources for three-dimensional models, compare
Quilici and Cailletaud (1999). For general discussions of the issues involved in microme-
chanical models in crystal plasticity and related fields see, e.g., Dawson (2000) and Roters
et al. (2010).

PMA Models for Two-Phase Single Crystal Superalloys

Nickel-base single crystal superalloys, which consist of a γ matrix phase containing aligned
cuboidal γ ′ precipitates, are of considerable importance due to their creep resistance at high
temperatures, their main field of application being the hot sections of gas turbines. Because
these materials show relatively regular microgeometries that change under load, due to a
process known as rafting, there has been considerable interest in micromechanical mod-
eling for elucidating their thermomechanical behavior and for better understanding their
microstructural evolution. The elastic behavior of two-phase superalloys can be handled
relatively easily by hexahedral unit cells with appropriately oriented anisotropic phases. In
the inelastic range, however, the highly constrained plastic flow in the γ channels may be
difficult to describe even by crystal plasticity models, see, e.g., Nouailhas and Cailletaud
(1996).

6.9 Periodic Models Models for Diffusion-Type Prob-

lems
Periodic microfield methods analogous to those discussed in Sections 6.1 to 6.8 can be
used for studying linear diffusion-type problems of the types mentioned in Section 2.10.
The Laplace solvers required for numerically-based homogenization in diffusion problems
are widely available in FE packages, usually for modeling heat conduction. Specialized
solvers are, however, required in some cases, e.g., when studying the frequency dependent
dielectric properties of composites via complex potentials, see, e.g., Krakovsky and My-
roshnychenko (2002).

117
Although Voronoi tessellations are commonly used for generating microgeometries for modeling poly-
crystals, the results may differ noticeably from actual microgeometries, see, e.g., Lazar et al. (2012).

144
 For the most common application, thermal conduction, the equivalent to eqn. (6.2)
takes the form
T (z + ck ) = T (z) + hdi ck , (6.21)
where the nomenclature of table 2.1 is followed and hdi is the volume averaged temperature
gradient, compare also Section 2.10. Periodicity boundary conditions can then be expressed
in terms of the nodal temperatures as

∆Tk = Tk+ − Tk− = T (sk + ck ) − T (sk ) = hdick (6.22)

in direct analogy to eqn. (6.3). Asymptotic homogenization for thermal conduction was
discussed, e.g., by Auriault (1983), Matt and Cruz (2002), or Tang and Yu (2007), and
the method of macroscopic degrees of freedom, compare Section 6.3, can be formulated in
terms of temperatures, thermal gradients and fluxes rather than displacements, strains and
stresses (Nogales, 2008). Furthermore, symmetry planes of the phase arrangement that are
oriented parallel or normally to far-field gradients can be used to specify “symmetry-like”
boundary conditions by not constraining temperatures at all or by setting them to a fixed
value, respectively118 . Volume and phase averages of fluxes are best evaluated according
to eqn. (5.8). Finite interfacial conductances can be handled within an FE framework by
using appropriate interface elements, compare, e.g., Matt and Cruz (2008) or Nogales and
Böhm (2008).

There are, however, some intrinsic conceptual difficulties in applying periodic homog-
enization to transport problems where nonlinear conduction or diffusion behavior of the
constituents is involved. Whereas in solid mechanics material nonlinearities are typically
formulated in terms of the microscopic stresses and strains, which may be viewed as “gen-
eralized gradient” and “generalized flux” fields (compare table 2.1), respectively, nonlinear
conductivities and diffusivities in transport problems typically depend on the “direct vari-
able” (or “potential”), e.g., the temperature in heat conduction. As is evident from fig. 6.1,
in PMAs the generalized gradients and fluxes are periodic (and have constant averages),
whereas the direct variables consist of fluctuating and linearly varying contributions that
accumulate from cell to cell. As a consequence, whereas in solid mechanics periodic homog-
enization involving nonlinear material behavior can be relied upon to pertain to identical
material properties for all periodically arranged unit cells even in the nonlinear case, there
may be conceptual contradictions in the case of transport problems. Accordingly, even
though solutions in terms of both microscopic (gradient and flux) fields and homogenized
macroscopic conductivities can be obtained from unit cell analysis involving material non-
linearities of the above type, the underlying model material is inconsistent119 . Temizer and
Wriggers (2011) carried out a detailed analysis of two-scale homogenization for conduc-
tion problems and showed that there is a general thermodynamical inconsistency for finite
deviations of microscopic temperatures from the macroscopic ones. Accordingly, the use
of periodic homogenization for studying nonlinear transport problems in inhomogeneous
118
As discussed by Duschlbauer et al. (2003a), symmetry planes oriented parallel to the far-field gra-
dient act as insulating surfaces, whereas symmetry planes oriented normally to the applied gradient are
isothermal planes. Gradient fields are symmetric with respect to the latter planes, temperature fields are
antisymmetric and flux fields non-symmetric.
119
The fact that periodicity implies that the direct variable can take values from −∞ to +∞ leads to
another inconsistency in PMAs for thermal conduction, where the variable T has a thermodynamical lower
bound at absolute zero. However, this issue does not appear to have any practical repercussions.

145
media has to be viewed with some reservation and windowing or embedding methods, see
Chapters 7 and 8, may be preferable for such work.

146
Chapter 7

Windowing Approaches

The aim of windowing methods lies in obtaining estimates for or bounds on the macro-
scopic properties of inhomogeneous materials on the basis of — typically non-periodic —
volume elements that are referred to as mesoscopic test windows, observation windows or,
in short, “windows”. These volume elements typically are too small to be proper RVEs
(i.e., they are SVEs in the sense of section 5.1), so that the results of windowing analysis
tend to pertain to specific samples rather than to a material and are, accordingly, referred
to as apparent (rather than effective) material properties.

Windowing implicitly assumes that the material to be studied is statistically homoge-

neous, so that windows can be extracted from it at random positions. For convenience,
windows are often chosen to be rectangles or right hexahedra, but other shapes may be
used just as well. If the material is known (or at least assumed) to be macroscopically
isotropic, windows are best extracted at random orientations, as shown in fig. 7.1. For
anisotropic materials, however, the orientations of windows must either be taken into con-
sideration in taking the samples or it must be accounted for explicitly in processing the
results of the analysis.

Figure 7.1: Schematic depiction of a composite that is statistically transversally isotropic and of
four rectangular windows of equal size in the transverse plane.

147
 Windowing methods subject window-type, inhomogeneous volume elements to stat-
ically uniform boundary conditions (SUBC), eqn. (5.3), kinematically uniform boundary
conditions (KUBC), eqn. (5.4), or mixed uniform boundary conditions (MUBC), eqn. (5.5)
for obtaining estimates on their macroscopic behavior.

Macrohomogeneous Boundary Conditions

SUBC and KUBC are collectively referred to as macrohomogeneous boundary conditions.
Compliance tensors obtained with SUBC and elasticity tensors obtained with KUBC pro-
vide lower and upper estimates on the macroscopic elastic tensors of volume elements,
respectively (Nemat-Nasser and Hori, 1993). For volume elements of equal size, ensemble
averages of these lower and upper estimates give rise to lower and upper bounds on the
overall apparent tensors (Huet, 1990); for unequally sized windows weighted averages must
be used. These bounds are sometimes referred to as mesoscale bounds and provide infor-
mation on the macroscopic properties pertinent to a collection of windows of given size.
For further discussions of these issues see, e.g., Ostoja-Starzewski (2008).

By definition, if the windows are of sufficient size to be proper RVEs the lower and up-
per estimates and bounds on the overall elastic properties must coincide (Huet, 1990; Sab,
1992), defining the effective behavior. Accordingly, hierarchies of bounds that are gener-
ated from sequences of sets of sub-RVE testing volume elements of increasing volume can
be used for assessing the size of proper representative volume elements. Actually achieving
coinciding lower and upper mesoscale bounds has been found to be very difficult in prac-
tice, however. This is at least partly due to boundary layers, see, e.g., Pahr and Böhm
(2008), that form due to the interaction of the uniform B.C. with phase boundaries inter-
secting the volume elements’ surfaces, compare the remarks in Section 5.2. Consequently,
using volumes that are shaped such that their boundaries stay within the contiguous phase
of matrix–inclusion composites, which reduces boundary perturbations, has been found to
improve the convergence of hierarchies of mesoscale bounds to a considerable extent (Salmi
et al., 2012; Acton et al., 2019). This expedient, however, may bias phase volume fractions
in the windows, which, in turn, may give rise to difficulties in identifying proper RVEs. As
another consequence of boundary perturbations, it is typically not advisable to combine
macrohomogeneous or mixed uniform B.C. with single-particle volume elements, compare
also the discussion in Vilchevskaya et al. (2021).

The concept of generating lower and upper estimates by windowing using SUBC and
KUBC can be shown to be also valid in the context of nonlinear elasticity and deformation
plasticity (Jiang et al., 2001). Discussions of windowing bounds in finite strain elastic-
ity, in viscoelasticity and in incremental plasticity settings were given by Khisaeva and
Ostoja-Starzewski (2006), Zhang and Ostoja-Starzewski (2015) as well as Li and Ostoja-
Starzewski (2006), respectively. Windowing methods are directly applicable to conduction
problems, with macrohomogeneous flux and gradient boundary conditions giving rise to
lower and upper estimates, respectively, as well as mesoscale bounds for the conductivity
tensor.

Generally, if the individual windows are small, considerable variations in the phase vol-
ume fractions will typically be present in different realizations and the macroscopic material

148
symmetry must be expected to be subject to considerable perturbations. Furthermore, for
relatively small windows, within which the phase arrangement deviates significantly from
statistical homogeneity, different boundary conditions, especially SUBC and KUBC, tend
to give rise to marked differences in the distributions of the microfields, especially the plas-
tic strains in elastoplastic models. When larger windows are used, however, increasingly
similar statistical distributions of the microstresses and microstrains are obtained (Shen
and Brinson, 2006).

For most combinations of microgeometries and phase materials macrohomogeneous

boundary conditions are fairly straightforward to implement in a Finite Element setting.
However, in general homogeneous macrostrains cannot be prescribed along boundaries that
intersect rigid inhomogeneities and homogeneous tangential tractions cannot be imposed
along boundaries that intersect voids.

Mixed Uniform Boundary Conditions

Equations (5.5) are fulfilled by a range of different sets of MUBC, resulting in different es-
timates for the apparent macroscopic elastic tensors, all of which lie between the lower and
upper estimates provided by the macrohomogeneous boundary conditions (Hazanov and
Huet, 1994). A specific set of mixed uniform boundary conditions for volume elements that
avoids prescribing non-zero boundary tractions was proposed by Pahr and Zysset (2008)
for obtaining accurate apparent elastic tensors of volume elements of cellular materials.
Table 7.1 lists these conditions which corresponding to six linearly independent mechani-
cal load cases for use with three-dimensional volume elements that have the shape of right
hexahedra aligned with the coordinate system. The distances l1 , l2 and l3 correspond to

Table 7.1: The six linearly independent uniform strain load cases making up the periodicity
compatible mixed uniform boundary conditions (PMUBC) proposed by Pahr and Zysset (2008)
and the pertinent boundary conditions for loading by a uniform temperature; East, West, North,
South, Top, and Bottom denote the faces of a right hexahedral volume element, compare fig. 6.4,
and the the distances li correspond to the cell’s edge lengths in the 1-, 2- and 3-directions.

East West North South Top Bottom

Tensile 1 u1 = εa11 l1 /2 u1 = −εa11 l1 /2 u2 = 0 u2 = 0 u3 = 0 u3 = 0
τ2a = τ3a = 0 τ2a = τ3a = 0 τ1a = τ3a = 0 τ1a = τ3a = 0 τ1a = τ2a = 0 τ1a = τ2a = 0
Tensile 2 u1 = 0 u1 = 0 u2 = εa22 l2 /2 u2 = −εa22 l2 /2 u3 = 0 u3 = 0
τ2a = τ3a = 0 τ2a = τ3a = 0 τ1a = τ3a = 0 τ1a = τ3a = 0 τ1a = τ2a = 0 τ1a = τ2a = 0
Tensile 3 u1 = 0 u1 = 0 u2 = 0 u2 = 0 u3 = εa33 l3 /2 u3 = −εa33 l3 /2
τ2a = τ3a = 0 τ2a = τ3a = 0 τ1a = τ3a = 0 τ1a = τ3a = 0 τ1a = τ2a = 0 τ1a = τ2a = 0
Shear 12 u2 = εa21 l1 /2 u2 = −εa21 l1 /2 u1 = εa12 l2 /2 u1 = −εa12 l2 /2 u3 = 0 u3 = 0
u3 = 0, τ1a = 0 u3 = 0, τ1a = 0 u3 = 0, τ2a = 0 u3 = 0, τ2a = 0 τ1a = τ2a = 0 τ1a = τ2a = 0
Shear 13 u3 = εa31 l1 /2 u3 = −εa31 l1 /2 u2 = 0 u2 = 0 u1 = εa13 l3 /2 u1 = −εa13 l3 /2
u2 = 0, τ1a = 0 u2 = 0, τ1a = 0 τ1a = τ3a = 0 τ1a = τ3a = 0 u2 = 0, τ3a = 0 u2 = 0, τ3a = 0
Shear 23 u1 = 0 u1 = 0 u3 = εa32 l2 /2 u3 = −εa32 l2 /2 u2 = εa23 l3 /2 u2 = −εa23 l3 /2
τ2a = τ3a = 0 τ2a = τ3a = 0 u1 = 0, τ2a = 0 u1 = 0, τ2a = 0 u1 = 0, τ3a = 0 u1 = 0, τ3a = 0
Thermal u1 = 0 u1 = 0 u2 = 0 u2 = 0 u3 = 0 u3 = 0
Expansion τ2a = τ3a = 0 τ2a = τ3a = 0 τ1a = τ3a = 0 τ1a = τ3a = 0 τ1a = τ2a = 0 τ1a = τ2a = 0

149
cell’s edge lengths in the 1-, 2- and 3-directions, compare Fig. 6.4. The components of
the prescribed strain tensor are denoted as εaij and those of the prescribed traction vector
as τia . Compared to macrohomogeneous boundary conditions the PMUBC lead to much
faster convergence of the predicted homogenized properties in terms of the size of the vol-
ume elements, similarly to periodicity boundary conditions, compare Section 6.2, and to
certain embedding schemes, compare Chapter 8.

The use of these MUBC is not limited to cellular materials, and when applied to peri-
odic volume elements of orthotropic effective behavior, they were found to give very similar
predictions for the macroscopic elasticity tensor as does periodic homogenization (Pahr
and Böhm, 2008). This led to their being named periodicity compatible mixed uniform
boundary conditions (PMUBC). Their behavior with periodic microgeometries indicates
that PMUBC can also be used to advantage for obtaining estimates from non-periodic
volume elements, at least when the sub-orthotropic contributions to the overall symmetry
are relatively small. Accordingly, these boundary conditions offer an attractive option for
obtaining estimates of the macroscopic elastic behavior and of macroscopic yield surfaces
(Panyasantisuk et al., 2016) on the basis of SVEs.

The concept of periodicity compatible mixed uniform boundary conditions can be ex-
tended to thermoelasticity by adding a load case that constrains all displacements normal
to the faces of the volume element, sets all in-plane tractions to zero, and applies a uniform
temperature increment ∆T , see table 7.1. This allows evaluating the volume averaged spe-
cific thermal stress tensor hλi, from which the apparent thermal expansion tensor can be
obtained as α = −Chλi, the apparent compliance tensor C being evaluated from the first
six equations of table 7.1.

Mixed uniform boundary conditions for diffusion-like problems were discussed, e.g., by
Jiang et al. (2001). Boundary conditions that show an analogous behavior to the PMUBC
in elasticity were reported by Jiang et al. (2002a) for thermal conduction in two-dimensional
orthotropic periodic media, see table 7.2. The handling of finite interfacial conductances
in the context of windowing was studied by Nogales (2008).

Table 7.2: The three linearly independent uniform gradient load cases making up the periodicity
compatible mixed uniform boundary conditions (PMUBC) in thermal conduction; East, West,
North, South, Top, and Bottom denote the faces of a hexahedral volume element, compare fig. 6.4,
the dai are the components of the applied temperature gradient, and the the distances li can be
obtained by scaling the cell’s edge lengths, ci , by a suitable common factor.

East West North South Top Bottom

Thermal 1 T = da1 l1 /2 T = −da1 l1 /2 q2a = 0 q2a = 0 q3a = 0 q3a = 0
Thermal 2 q1a = 0 q1a = 0 T = da2 l2 /2 T = da2 l2 /2 q3a = 0 q3a = 0
Thermal 3 q1a = 0 q1a = 0 q2a = 0 q2a = 0 T = da3 l3 /2 T = da3 l3 /2

In contrast to periodicity and macrohomogeneous boundary conditions, the PMUBC

listed in table 7.1 are strictly limited to handling these seven specific load cases if nonlinear
behavior is present and the superposition principle ceases to hold. This implies that gen-
eral load paths in stress and strain space cannot be followed, so that in nonlinear regimes

150
this type of model is restricted to materials characterization, compare (Pahr and Böhm,
2008).

Certain MUBC, e.g., ones that handle macroscopic uniaxial loading by specifying nor-
mal displacements plus zero tangential tractions for one pair of faces and setting all traction
components to zero for all other faces, compare, e.g., Galli et al. (2008), can be interpreted
physically as describing the behavior of small, inhomogeneous samples120 as discussed in
Section 1.5. Due to the weaker boundary constraints such sample-type (or “structure-
type”) MUBC can be expected to predict more compliant macroscopic behavior than do
the periodicity boundary conditions discussed in Section 6.2 or the PMUBC listed in table
7.1. As mentioned in Section 1.5, structure-type boundary conditions do not involve scale
transitions. At this point, it is worth noting that the symmetry BC discussed in Section
6.2 may also be interpreted as a set of MUBC with good convergence properties.

Important strengths of windowing methods lie in providing an approach to studying

the behavior of non-periodic volume elements and in being considerably easier to handle
than PMAs for periodic volume elements. The main reasons for the latter point are that
the meshing of homologous faces is not an issue in windowing and multi-point constraint
equations are not required.

120
In contrast to periodicity BC, these specific MUBC give different results for volume elements made
up of a number “base units” compared to models consisting of a single unit. This is in keeping with their
interpretation as “structural” rather than “material” models.
If the tangential strains rather than the tangential stresses are set to zero for the pair of faces controlling
the macroscopic axial deformation, the physical interpretation is that of a sample “sticking” to the base
plates, implying the presence of strain gradients at the length scale of the sample. Such models are not
capable of describing a proper uniaxial stress load case.

151
Chapter 8

Embedding Approaches and Related

Models

Embedded Cell Approaches (ECAs) aim at predicting the microfields in specific, geomet-
rically highly resolved (sub)regions of models of inhomogeneous materials or structures.
Such models consist of two parts, as is evident in figs. 8.1 and 8.2. On the one hand, there
is a core (or “local heterogeneous region”) consisting of a discrete microstructure (“mo-
tif”), which can range from rather simple to highly complex phase arrangements. On the
other hand, there is a homogeneous outer region (“embedding region”, “frame”, “effective
region”) into which the core is embedded and which serves mainly for transmitting the
applied loads. Embedding strategies tend to give rise to relatively complex models, but
they avoid the main drawback of PMAs, viz., the requirement that the microgeometry and
all microfields must be periodic.

An intrinsic feature of embedding models are boundary layers that occur at the “inter-
faces” between the core and the embedding region and perturb the local stress and strain
fields121 , see, e.g., Harper et al. (2012a) and compare Section 5.2. Provision must be made
to keep regions of specific interest, such as crack tips or process zones, at a sufficient dis-
tance from the boundary layers122 , which, in turn, implies that cores must exceed some
minimum size in order to provide useful results.

Most of the embedding approaches reported in the literature for tasks related to con-
tinuum micromechanics fall into two groups.
121
The interfaces between core and embedding region are a consequence of the modeling approach and
do not have any physical background or significance. In elasticity such boundary layers typically have a
thickness of at least the distance between the centers of neighboring inhomogeneities, but they may be
longer ranged for nonlinear material behavior.
122
Boundary layers and embedding regions may interfere with extended regions of concentrated strains,
with shear bands, or with the growth of damaged regions. Such difficulties can only be avoided or mitigated
by choosing a sufficiently large core region. In dynamic models the boundaries between core and embedding
region may lead lead to the reflection and/or refraction of waves. Special transition layers can, however,
be introduced to mitigate boundary layer effects in the latter case.

152
Embedding Region with Self-Consistently Determined Response
One group of embedding schemes employ the homogenized thermomechanical response
of the core for determining the effective behavior of the surrounding medium via self-
consistent procedures. By applying suitable uniform far-field loads to the outer boundaries
of such an embedding layer, as sketched in Fig. 8.1, models of this type can be used for
carrying out scale transitions in terms of homogenization and, with some precautions,
localization.

core

S.C.

embedding

Figure 8.1: Schematic depiction of the arrangement of core and embedding region in a self-
consistent embedded cell model subjected to a uniform far-field load.

As mentioned in Section 5.2 self-consistent embedding schemes employing macroho-

mogeneous or periodic boundary conditions123 provide predictions for the effective elastic-
ity tensors that lie between the lower and upper estimates on the apparent tensors and,
by extension, within the mesoscale bounds discussed in Chapter 7. Such approaches, in
which the embedding layer mainly serves for smoothening out fluctuations at the core’s
boundaries was termed the “Window Method” by some authors (Krabbenhøft et al., 2008;
Temizer et al., 2013), and the smeared-out embedding region has been termed an “aug-
mentation strip” (Babuška et al., 1999) or an “envelope surrounding the RVE” (Vazeille
and Laberge Lebel, 2024). In a similar vein, the use of a self-consistent embedding “buffer
layer” has been proposed for models in which periodicity boundary conditions are applied
to non-periodic volume elements (Makowski et al., 2013). Methods of this type typically
aim at carrying out scale transitions and, accordingly, the core should at least approach
the size of a proper RVE.

123
In such schemes either the full core (including the boundary layers) or core’s interior (the regions
affected by the boundary layer being discarded) may be used for evaluating the effective responses. In
their analysis of such self-consistent schemes (Temizer et al., 2013) take the former approach. For elastic
spherical inhomogeneities in an elastic matrix the thickness of the boundary layer may be estimated as
four particle radii (Buryachenko, 2007).

153
 Because they involve self-consistent iterations, self-consistent embedding methods tend
to be more expensive numerically than are periodic homogenization and windowing anal-
ysis (in the sense of Chapter 7). In the elastic regime, however, the required number of
iterations tends to be fairly low and the approach has been shown to be unconditionally
stable by Salit and Gross (2014). A number of issues in choosing the width of the embed-
ding layer for thermoelastic analysis were discussed by Temizer et al. (2013).

The use of self-consistency procedures is, of course, predicated on the availability of

suitable parameterizable constitutive laws for the embedding material that can follow the
core’s (instantaneous) homogenized behavior with high accuracy for all load cases and
loading histories considered. This requirement can typically be fulfilled easily in the linear
range, see e.g., Yang et al. (1994) and Chen et al. (1994), but may lead to considerable
complexity when at least one of the constituents shows elastoplastic or viscoplastic ma-
terial behavior124 . Accordingly, approximations may have to be used (the consequences
of which may be difficult to assess in view of the nonlinearity and path dependence of
such material behaviors) and models of this type are best termed “quasi-self-consistent
schemes”. Approaches of this type were discussed, e.g., by Bornert et al. (1994) and by
Dong and Schmauder (1996).

Self-consistent and quasi-self-consistent embedding models are not suitable for handling
“strong features”, such as localized cracks of the type shown in fig. 8.2, within the core and
their physical interpretation may be difficult when the core includes subregions subject to
damage.

Embedding Region with Prescribed Material Response

In the other group of embedding methods the material behavior of the outer region is
described via appropriate, pre-defined, “smeared-out” constitutive models. These typi-
cally take the form of semi-empirical or micromechanically based constitutive laws that
are prescribed a priori for the embedding zone and are chosen to represent the appropriate
(usually damage-free) material behavior125 . This way, conceptually simple models are ob-
tained that are very well suited to studying local phenomena such as the stress and strain
124
Typically, the effective yielding behavior of elastoplastic composites shows some dependence on the
first stress invariant, and, for low plastic strains, the homogenized response of the core tends to be strongly
influenced by the fractions of the elastoplastic constituent(s) that have actually yielded. In addition, in
many cases anisotropy of the yielding and hardening behavior is induced by the phase topology (e.g.,
aligned fibers) and/or by the phase arrangement of the core. Identifying constitutive laws that, on the one
hand, can satisfactorily account for such phenomena and, on the other hand, have the capability of being
easily adapted to the instantaneous homogenized responses of the core by adjusting free parameters, poses
an important challenge in applying self-consistent embedding schemes to nonlinear material behavior.
125
The constitutive model and the pertinent material parameters required for the embedding layer may
be obtained, e.g., from micromechanical modeling or from experiments (Ayyar et al., 2008).
The description chosen for the embedding region must reflect the material symmetry, e.g., for metal
matrices reinforced by continuous fibers the yield surface, the flow rule and the hardening behavior must
account for these materials’ marked anisotropy in the elastoplastic regime.
When the core is used for studying local damage and failure, special care should be exercised with respect to
its boundary regions — the perturbed local stress and strain fields there may conceivably trigger unphysical
behavior. However, if failure or localized damage are restricted to the core this may impact the generality
of models, compare, e.g., Bao (1992).

154
distributions in the vicinity of crack tips (Aoki et al., 1996), around local defects (Xia
et al., 2001) or at macroscopic interfaces in composites (Chimani et al., 1997), the growth
of cracks in inhomogeneous materials (van der Giessen and Tvergaard, 1994; Wulf et al.,
1996; Motz et al., 2001; Mishnaevsky, 2004; González and LLorca, 2007), the behavior of
composites close to an indenter (Shedbale et al., 2017), or the damage due to the processing
of composites (Monaghan and Brazil, 1997).

In applications of this type loads may be applied or displacement boundary conditions

may be specified that impose a far-field behavior obtained from a suitable analytical or
numerical solution (e.g., the displacements describing the far-field of a crack tip in elas-
ticity or small-scale plasticity) pertinent to the problem under study. Alternatively, the
embedding region may be chosen to represent the structure or sample to be considered,
with the core zooming in on a detail of special interest. Such approaches, which may be
viewed as a type of concurrent multiscale modeling scheme, allow complete specimens or
components to be studied via “simulated experiments” as sketched in fig. 8.2.

Figure 8.2: Schematic depiction of the arrangement of core and embedding region in an embedded
cell model of a tensile test specimen.

Embedding models making use of prescribed properties for the embedding region are
not the best choice for homogenization analysis — they may rather be viewed as methods
specialized towards localization. On the one hand, this arguably makes them the most
versatile and powerful tool for looking in detail at “strong features” in some region of
interest. Embedding models of this kind can be used without intrinsic restrictions at or
in the vicinity of free surfaces and interfaces, they can handle gradients in composition
and loads, and they can be employed for studying the interaction of macrocracks with
microstructures126 following fig. 8.2. It is worth noting that for such models typically it is
not of major importance whether the core is a proper RVE or not — embedding models
126
The requirement of sufficiently separated length scales, compare Section 1.1, does not necessarily apply
to embedded models. Care is, however, required, with highly nonlinear analyses, especially with models
involving damage, where the inherent perturbations at boundaries between core and embedding region
may trigger spurious behavior.

155
of this type do not involve scale transitions. In fact, it may make sense to use phase
arrangements for the core that are known (or suspected) to be critical for the behavior to
be studied rather than being representative, in the spirit of a worst case analysis. The core
must, however, be of sufficient size for keeping perturbations due to boundary layer effects
at a suitably low level in its critical regions.

Other Embedding Strategies, Submodeling Schemes and Related

Approaches
Another approach related to embedding uses discrete microstructures in both the core
region and in the surrounding material, the latter, however, being discretized by a much
coarser FE mesh, see, e.g., Sautter et al. (1993). Such models, which are also referred to
as “fine mesh window” approaches (Podgórski, 2011) and are essentially descriptions of
full samples or structures that contain a refined mesh in some region(s) of interest, can
avoid boundary layers to a considerable extent by using appropriately graded meshes. They
tend, however, to be rather expensive computationally due to the need for relatively highly
resolved “sacrificial” zones in the embedding region that only serve to insulate the core
from the boundary layers (Harper et al., 2012a). Mesh superposition techniques, which
use a coarse mesh over the macroscopic model of some structure or sample together with
a geometrically independent, much finer mesh in regions of interest (Takano et al., 2001)
are conceptually similar to the above modeling strategy.

A closely related approach makes use of submodeling techniques in which an inho-

mogeneous micromodel is weakly or strongly coupled to a homogeneous macromodel via
appropriate coupling conditions (which may be implemented via the boundary conditions
of the micromodel), compare, e.g., Heness et al. (1999) or Váradi et al. (1999). Similarly,
far-field strains obtained from an unperturbed periodic model subjected to appropriate
loads or loading histories may be used as boundary conditions for a region containing a
local microstructural perturbation, compare Aboudi and Ryvkin (2012). Such modeling
strategies are conceptually related, on the one hand, to embedding models of the type
shown in Fig. 8.2 and, on the other hand, to the “inhomogeneous structure models” men-
tioned in Section 1.5.

Additional Remarks on Embedding Schemes

When a periodic multi-inhomogeneity volume element is used as the core in a self-consistent
or quasi-self-consistent embedding scheme, the concomitant relaxation of the periodicity
constraints tends to make the overall responses of the embedded configuration more com-
pliant compared to proper periodic homogenization of the periodic arrangement (Bruzzi
et al., 2001)127 .

Some analytical methods such as classical and generalized self-consistent schemes, see
Section 2.3, can obviously be viewed as embedding schemes that combine relatively simple
127
In the study mentioned above, the difference between PMA and ECA results is relatively small.
Identical responses can be expected from the two types of model if the core takes the form of a proper
RVE (so that by definition the boundary conditions do not play a role in the effective behavior) and the
embedding material can fully describe the homogenized behavior of the core.

156
cores with self-consistently defined material behavior of the embedding region. When geo-
metrically more complex cores are considered in ECA-like frameworks, however, numerical
engineering methods are best suited for resolving the discrete phase arrangements in the
core region, compare Bornert (1996).

The core and embedding regions may be planar, axisymmetric or fully three-dimensional,
and symmetries present in the geometries can, as usual, be made us of for reducing the size
of the model, compare Xia et al. (2001). Effective and phase averaged stresses and strains
from embedded cell models are best evaluated via eqn. (1.4) or its equivalents, and it is
typically preferable to use only the central regions of the core for this purpose in order to
avoid perturbations from the boundary layers.

All embedding techniques discussed above can also be used in analogy for studying the
thermal conduction behavior of inhomogeneous materials.

157
Chapter 9

Hierarchical and Multi-Scale Models

The micromechanical methods discussed in Chapters 2, 6, and 7 are designed for han-
dling a single scale transition between a lower and a higher length scale (“microscale” and
“macroscale”), overall responses being obtained by homogenization and local fields by lo-
calization. Many inhomogeneous materials, however, show more than two clearly distinct
characteristic length scales, typical examples being laminated and woven composites (com-
pare Section 6.4), materials in which there are well defined clusters of particles, as well as
most biomaterials. In such cases an obvious modeling strategy is a hierarchical approach
that uses a sequence of scale transitions, i.e., the material response at any given length
scale is described on the basis of the homogenized behavior of the next finer one128 . Figure
9.1 schematically shows such a hierarchical model for a particle-reinforced composite with
a clustered mesostructure.

scale scale
transition transition
#1 #2

MACROSCALE MESOSCALE MICROSCALE

(sample) (particle clusters) (particles in matrix)

Figure 9.1: Schematic representation of a hierarchical approach to studying a material con-

sisting of clustered inhomogeneities in a matrix. Two scale transitions, macro←→meso and
meso←→micro, are used.

A hierarchical model can be viewed as involving a sequence of scale transitions, and

suitable micromechanical models, i.e., mean-field, unit cell, windowing and, to some extent,
128
Describing the material behavior at lower length scales by a homogenized model implies that char-
acteristic lengths differ by more than, say, an order of magnitude, so that valid homogenization volumes
can be defined (there may, however, be exceptions in the case of linear material behavior). Employing
hierarchical approaches within “bands” of more or less continuous distributions of length scales, as can be
found, e.g., in some metallic foams with highly disperse cell sizes, requires specialized methods.

158
embedding approaches, may be used as “building blocks” at any level within hierarchical
schemes. Such hierarchical modeling strategies have the additional advantage of allowing
the behavior of the constituents at all lower length scales to be assessed via the corre-
sponding localization relations. Multiscaling methods can be classified into concurrent and
sequential approaches (Chakraborty and Rahman, 2008; Tadmor and Miller, 2011), with
the former requiring the physical problem to be studied simultaneously at all length scales
involved.

Among the continuum mechanical hierarchical descriptions of the thermomechanical

behavior of inhomogeneous materials reported in the literature, some have combined mean-
field methods at the higher length scale with mean-field (Hu et al., 1998; Tszeng, 1998) or
periodic microfield approaches (González and LLorca, 2000) at the lower length scale. A
common strategy for hierarchical modeling, however, uses Finite Element based unit cell
or embedding methods at the topmost length scale, which implies that the homogenized
material models describing the lower level(s) of the hierarchy must be proper constitutive
laws (i.e., material descriptions that are capable of handling any loading conditions and any
loading history) that are evaluated at each integration point of the discretized model. For
this purpose, either explicit, micromechanically-based constitutive models may be used or
micromechanical models may be run at each integration point, the results of which provide
“implicit” descriptions of the local material behavior. The models used for such purposes
may be analytical, semi-analytical or numerical.

For the elastic range, generating a micromechanically based constitutive description at

each integration point of the “macroscopic” model typically does not pose massive compu-
tational requirements, even if different microscopic models are used for different regions of
the macromodel; furthermore, in such a setting decomposition techniques may be used to
formulate the problems at the finer length scales in such a way that they are well suited
for parallel processing, allowing the development of computationally highly efficient multi-
scale procedures, see, e.g., Oden and Zohdi (1997). Once the appropriate homogenized
elasticity and thermal concentration tensors have been generated any load case can be
handled, often in a sequential mode. For simulating the thermomechanical response of
inelastic inhomogeneous materials, however, essentially a full micromechanical submodel
has to be maintained and solved for at each integration point, which amounts to fully
concurrent procedures129 . Although within such frameworks the use of sophisticated nu-
merical models at the lower length scales tends to be an expensive proposition in terms of
computational requirements, this concept has elicited major research interest, especially
within the framework of multi-scale models (Belsky et al., 1995; Ghosh et al., 1996; Lee
and Ghosh, 1996; Zohdi et al., 1996; Smit et al., 1998; Feyel and Chaboche, 2000; Fish
and Shek, 2000; Ibrahimbegović and Markovič, 2003; Moës et al., 2003; Tan et al., 2020).
Pertinent overviews were given by Ghosh et al. (2001), Kanouté et al. (2009), Geers et al.
(2010) as well as Nguyen et al. (2011).
129
Some approaches have been reported that aim at decoupling the scales by parameterizing results from
microscopic full field analysis and using this data together with an interpolation scheme as an “approximate
constitutive model”, see, e.g., Terada and Kikuchi (1996), Gänser (1998) or Schrefler et al. (1999). A
major challenge in this type of sequential modeling strategy lies in handling the load history and load path
dependences required for elastoplastic local behavior of and/or microscopic damage to the constituents.
Appropriate representation of general multiaxial stress and/or strain states also tends to be difficult.

159
 It is worth noting that in models of this type (which are often referred to as FE2 pro-
cedures) no explicit material law is used on the macroscale, the full constitutive behavior
being determined concurrently at the microscale. Arguably, such approaches at present
represent the most sophisticated application of the concepts of continuum micromechanics.

Lower (but by no means negligible) computational costs can be achieved by using mean-
field models in lieu of constitutive models at the integration point level. Applications to
elastoplastic composites have included incremental Mori–Tanaka methods, see, e.g., Petter-
mann (1997) and mean-field based versions of the Transformation Field Analysis, compare,
e.g., Fish and Shek (1998) or Plankensteiner (2000). In addition, the semi-analytical non-
uniform TFA has been employed for this purpose (Michel and Suquet, 2004). Figure 9.2
shows a result obtained by applying a multiscale approach that uses an incremental Mori–
Tanaka method, compare Section 2.9, at the integration point level of a two-dimensional
meso-scale unit cell model for describing the elastoplastic behavior of a cluster-structured
high speed steel. The particle-rich and particle-poor regions used for the description at the
mesoscale are treated as particle-reinforced MMCs with appropriate reinforcement volume
fractions. Such a model not only predicts macroscopic stress–strain curves, but also al-
lows the mesoscopic distributions of phase averaged microscopic variables to be evaluated,
compare (Plankensteiner, 2000). For an example of an Extended Finite Element method
at the upper length scale coupled to a semi-analytical model at the lower one see Novák
et al. (2012).

EPS.EFF.PLAST.MTX.

3.5000E-03
3.0000E-03
2.5000E-03
2.0000E-03
1.5000E-03

SCALAR MIN: -6.8514E-04

SCALAR MAX: 1.3667E-02

Figure 9.2: Phase averaged microscopic equivalent plastic strains in the matrix within the
inhomogeneity-poor regions of a cluster-structured high speed steel under mechanical loading
as predicted by a mesoscopic unit cell model combined with an incremental Mori–Tanaka model
at the microscale (Plankensteiner, 2000).

Alternatively, multiscale approaches may be rather “loosely coupled”, essentially link-

ing together quite different models to obtain an overall result, see, e.g., (Onck and van der
Giessen, 1997).

160
 A further development are concurrent multi-scale methods in which the computational
domain is adaptively split into regions resolved at the appropriate length scale. In an FE-
based hierarchical framework developed by Ghosh et al. (2001) “non-critical” regions are
described by continuum elements with pre-homogenized material properties, and for zones
of clearly nonlinear behavior PMA models are automatically activated at the integration
points to monitor the phase behavior at the microscale. If local behavior exceeding the
capabilities of periodic homogenization is detected (e.g., progressive damage, mesoscopic
interfaces, free edges), embedded models of the fully resolved microgeometry may be acti-
vated at the appropriate positions, compare Raghavan and Ghosh (2004). Such strategies
allow detailed studies of critical regions in inhomogeneous materials, e.g., near free edges.
In an alternative approach, homogenization models employing generalized continua have
been proposed for handling regions close to free surfaces and zones with high local field
gradients (Feyel, 2003).

Finally, it is worth noting that hierarchical and multi-scale approaches are not limited
to using the “standard” methods of continuum micromechanics as discussed above. Espe-
cially the capability of the Finite Element method of handling highly complex constitutive
descriptions has been used to build hierarchical approaches that employ, among others,
material models based on crystal plasticity (McHugh et al., 1993) and discrete dislocation
plasticity (Cleveringa et al., 1997).

The linking of continuum and atomistic descriptions has been an important goal of mod-
eling work for a considerable time, involving, e.g., concurrent homogenization schemes,
compare Curtin and Miller (2003) and, for a view on recent and future developments,
van der Giessen et al. (2020). The equivalent of homogenization in such settings is often
referred to as “upwards linking”, that to localization being known as “downward linking”.

161
Chapter 10

Closing Remarks

The research field of continuum micromechanics of materials has enjoyed considerable suc-
cess in the past four decades in furthering the understanding of the thermomechanical
behavior of inhomogeneous materials and in providing predictive tools for engineers and
materials scientists. However, its methods are subject to some practical limitations that
should be kept in mind when employing them.

All the methods discussed in Chapters 2 to 7 implicitly use the assumption that the
constituents of the inhomogeneous material to be studied can be treated as homogeneous
at the lower length scale, which, of course, does not necessarily hold. When the length scale
of the inhomogeneities in a constituent is much smaller than the length scale of the phase
arrangement to be studied, the above assumption is valid and hierarchical as well as multi-
scale models (“successive homogenization”) as discussed in Chapter 9 can be used (Carrère,
2001). However, no rigorous theory appears to be available at present for handling scale
transitions in materials that do not fulfill the above requirement (e.g., particle-reinforced
composites in which the grain size of the matrix is comparable to or even larger than the
size of the particles130 ). Typically the best that can be done in such cases is either to use
sufficiently detailed models with resolved microgeometry (potentially, a computationally
very expensive proposition) or to employ homogenized phase properties and be aware of
the approximation that is introduced.

A major practical difficulty in the use of continuum micromechanics approaches (which

in many cases is closely related to the questions mentioned above) has been identifying
appropriate constitutive models and obtaining the required material parameters for the
constituents. Typically, available data pertain to the behavior of the bulk materials (as
measured from homogeneous macroscopic samples), whereas the actual requirement is for
parameters and, in some cases, constitutive theories that describe the in-situ response of
the constituents at the microscale. For example, with respect to MMCs, on the one hand,
there is considerable evidence that classical continuum plasticity theories (in which there
is no absolute length scale) cannot adequately describe the behavior of metallic materi-
als in the presence of strong strain gradients, where geometrically necessary dislocations
can markedly influence the material behavior (Hutchinson, 2000). On the other hand,
130
For this special case unit cell or embedding models employing anisotropic or crystal plasticity models
for the phases may be used which, however, can become very large when the statistics of grain orientation
are accounted for.

162
it is well known that the presence of reinforcements can lead to accelerated aging and
refinement of the grain size in the matrix (“secondary strengthening”). Furthermore, in
many cases reasonably accurate material parameters are essentially not available (e.g.,
strength parameters for most interfaces in inhomogeneous materials). In fact, the dearth
of dependable material parameters is one of the reasons why predictions of the strength of
inhomogeneous materials by micromechanical methods tend to be a considerable challenge.

Another point that should be kept in mind is that continuum micromechanical descrip-
tions in most cases do not have absolute length scales unless a length scale is introduced
explicitly, typically via the constitutive model(s) of one (or more) of the constituents131 .
In this vein, absolute length scales, on the one hand, may be provided explicitly via dis-
crete dislocation models (Cleveringa et al., 1997), via gradient or nonlocal constitutive
laws, see, e.g., Tomita et al. (2000) or Niordson and Tvergaard (2001), or via (nonlocal)
damage models. On the other hand, they may be introduced in an ad-hoc fashion, e.g.,
by adjusting the phase material parameters to account for grain sizes via the Hall–Petch
effect. Analysts should also be aware that absolute length scales may be introduced in-
advertently into a model by mesh dependence effects of discretizing numerical methods, a
“classical” example being strain localization due to softening material behavior of a con-
stituent. When hierarchical or multi-scale models are used special care may be necessary
to avoid introducing inconsistent length scales at different modeling levels.

In addition it is worth noting that usually the macroscopic responses of inhomoge-

neous materials (and, as a consequence of eqns. (2.8) to (2.12), the phase averages of the
microfields) are much less sensitive to the phase arrangement (and to modeling approxima-
tions) than are the distributions of the microfields. Accordingly, whereas good agreement
in the overall behavior of a given model with “benchmark” theoretical results or experimen-
tal data typically indicates that the phase averages of stresses and strains are described
satisfactorily, this does not necessarily imply that the higher statistical moments of the
stress and strain distributions (and thus the heterogeneity of these fields) are captured
sufficiently accurately.

It is important to be aware that work in the field of micromechanics of materials invari-

ably involves finding viable compromises in terms of the complexity of the models, which,
on the one hand, have to be able to account (at least to a reasonably good approximation)
for the physical phenomena relevant to the given problem, and, on the other hand, must be
sufficiently simple to allow solutions to be obtained within the relevant constraints of time,
cost, and computational resources. Obviously, actual problems cannot be solved without
recourse to various approximations and tradeoffs — the important point is to be aware of
them and to account for them in interpreting and assessing results. It is worth keeping in
mind that there is no such thing as a “best micromechanical approach” to all problems.

131
One important exception are models for studying macroscopic cracks (e.g., via embedded cells), where
the crack length does introduce an absolute length scale.
It is worth noting that the behavior of nanocomposites is typically dominated by interfaces (or interphases).
Appropriately accounting for this behavior may require the introduction of an absolute length scale. Within
the framework of mean-field methods this may be achieved by introducing surface terms into the Eshelby
tensor, see, e.g., Duan et al. (2005).

163
Bibliography

N. Abolfathi, A. Naik, G. Karami, and C. Ulven. A micromechanical characterization

of angular bidirectional fibrous composites. Comput.Mater.Sci., 43:1193–1206, 2008.
doi:10.1016/j.commatsci.2008.03.017.

J. Aboudi. The generalized method of cells and high-fidelity generalized method of

cells micromechanical models — A review. Mech.Adv.Mater.Struct., 11:329–366, 2004.
doi:10.1080/15376490490451543.

J. Aboudi. Damage in composites — modeling of imperfect bonding. Compos.Sci.Technol.,

28:103–128, 1987. doi:10.1016/0266-3538(87)90093-5.

J. Aboudi. Micromechanical analysis of composites by the method of cells. Appl.Mech.Rev.,

42:193–221, 1989. doi:10.1115/1.3152428.

J. Aboudi. Mechanics of Composite Materials. Elsevier, Amsterdam, 1991. ISBN 0-444-

88452-1.

J. Aboudi. Micromechanical analysis of composites by the method of cells — Update.

Appl.Mech.Rev., 49:S83–S91, 1996. doi:10.1115/1.3101981.

J. Aboudi and M. Ryvkin. The effect of localized damage on the behavior

of composites with periodic microstructure. Int.J.Engng.Sci., 52:41–55, 2012.
doi:10.1016/j.ijengsci.2011.12.001.

D.W. Abueidda, M. Bakir, A.K. Abu Al-Rub, J.S. Bergström, N.A. Sobh, and
I. Jasiuk. Mechanical properties of 3D printed polymer cellular materials with
triply periodic minimal surface architectures. Mater.Design, 122:255–267, 2017.
doi:10.1016/j.matdes.2017.03.018.

M.L. Accorsi. A method for modeling microstructural material discontinu-

ities in a finite element analysis. Int.J.Num.Meth.Engng., 26:2187–2197, 1988.
doi:10.1002/nme.1620261004.

J.D. Achenbach and H. Zhu. Effect of interfacial zone on mechanical behavior

and failure of fiber-reinforced composites. J.Mech.Phys.Sol., 37:381–393, 1989.
doi:10.1016/0022-5096(89)90005-7.

K. Acton, C. Sherod, B. Bahmani, and R. Abedi. Effect of volume element geometry on

convergence to a representative volume. J.Risk Uncert.Engng.Syst.B, 5:030907, 2019.

164
D.F. Adams and D.A. Crane. Finite element micromechanical analysis of a unidirectional
composite including longitudinal shear loading. Comput.Struct., 18:1153–1165, 1984.
doi:10.1016/0045-7949(84)90160-3.

D.F. Adams and D.R. Doner. Transverse normal loading of a uni-directional composite.
J.Compos.Mater., 1:152–164, 1967. doi:10.1177/002199836700100205.

S.G. Advani and C.L. Tucker. The use of tensors to describe and predict fiber orientation
in short fiber composites. J.Rheol., 31:751–784, 1987. doi:10.1122/1.549945.

D.H. Allen and J.W. Lee. The effective thermoelastic properties of whisker-reinforced
composites as functions of material forming parameters. In G.J. Weng, M. Taya, and
H. Abé, editors, Micromechanics and Inhomogeneity, pages 17–40, New York, NY, 1990.
Springer-Verlag. doi:10.1007/978-1-4613-8919-4 2.

H. Altendorf and D. Jeulin. Random-walk-based stochastic modeling of three-dimensional

fiber systems. Phys.Rev. E, 83:041804, 2011. doi:10.1103/PhysRevE.83.041804.

A. Anthoine. Derivation of the in-plane elastic characteristics of masonry through homoge-

nization theory. Int.J.Sol.Struct., 32:137–163, 1995. doi:10.1016/0020-7683(94)00140-R.

T. Antretter. Micromechanical Modeling of High Speed Steel. Reihe 18, Nr. 232. VDI–
Verlag, Düsseldorf, 1998. ISBN 3-18-323218-9.

S. Aoki, Y. Moriya, K. Kishimoto, and S. Schmauder. Finite element

fracture analysis of WC–Co alloys. Engng.Fract.Mech., 55:275–287, 1996.
doi:10.1016/0013-7944(96)00021-5.

P. Arbenz, G.H. van Lenthe, U. Mennel, R. Müller, and M. Sala. Multi-level µ-finite ele-
ment analysis for human bone structures. In B. Kagström, E. Elmroth, J. Dongarra, and
J. Wasniewski, editors, Applied Parallel Computing. State of the Art in Scientific Com-
puting, pages 240–250, Berlin, 2008. Springer-Verlag. doi:10.1007/978-3-540-75755-9 30.

R.T. Arenburg and J.N. Reddy. Analysis of metal-matrix composite structures

— I. Micromechanics constitutive theory. Comput.Struct., 40:1357–1368, 1991.
doi:10.1016/0045-7949(91)90407-D.

J.L. Auriault. Upscaling heterogeneous media by asymptotic expansions. J.Engng.Mech.,

128:817–822, 2002. doi:10.1061/(ASCE)0733-9399(2002)128:8(817).

J.L. Auriault. Effective macroscopic description for heat conduction in periodic composites.
Int.J.Heat Mass Transf., 26:861–869, 1983. doi:10.1016/S0017-9310(83)80110-0.

M. Avellaneda. Iterated homogenization, differential effective medium theory, and appli-

cations. Comm.Pure Appl.Math., 40:803–847, 1987. doi:10.1002/cpa.3160400502.

M.S. Axelsen and R. Pyrz. Correlation between fracture toughness and the microstruc-
ture morphology in transversely loaded unidirectional composites. In R. Pyrz, editor,
Microstructure–Property Interactions in Composite Materials, pages 15–26, Dordrecht,
1995. Kluwer Academic Publishers. doi:10.1007/978-94-011-0059-5 2.

165
A. Ayyar, G.A. Crawford, J.J. Williams, and N. Chawla. Numerical simulation of the effect
of particle spatial distribution and strength on tensile behavior of particle reinforced
composites. Comput.Mater.Sci., 44:496–506, 2008. doi:10.1016/j.commatsci.2008.04.009.

I. Babuška, B. Andersson, P.J. Smith, and K. Levin. Damage analysis of fiber composites.
Part I: Statistical analysis on fiber scale. Comput.Meth.Appl.Mech.Engng., 172:27–77,
1999. doi:10.1016/S0045-7825(98)00225-4.

Y. Bai, C.Z. Dong, and Z.Y. Liu. Effective elastic properties and stress states of dou-
bly periodic array of inclusions with complex shapes by isogeometric boundary element
method. Compos.Struct., 128:54–69, 2015. doi:10.1016/j.compstruct.2015.03.061.

D. Balzani, D. Brands, J. Schröder, and C. Carstensen. Sensitivity analysis of statis-

tical measurements for the reconstruction of microstructures based on the minimiza-
tion of generalized least-square functionals. Techn.Mech., 30:297–315, 2010. URL
https://journals.ub.ovgu.de/index.php/techmech/article/view/800.

P.K. Banerjee and D.P. Henry. Elastic analysis of three-dimensional solids with fiber inclu-
sions by BEM. Int.J.Sol.Struct., 29:2423–2440, 1992. doi:10.1016/0020-7683(92)90001-A.

L. Banks-Sills and V. Leiderman. Macro-mechanical material model for

fiber-reinforced metal matrix composites. Composites B, 30:443–452, 1999.
doi:10.1016/S1359-8368(99)00018-9.

Y. Bansal and M.J. Pindera. Finite-volume direct averaging micromechanics of het-

erogeneous materials with elastic-plastic phases. Int.J.Plast., 22:775–825, 2006.
doi:10.1016/j.ijplas.2005.04.012.

G. Bao. Damage due to fracture of brittle reinforcements in a ductile matrix. Acta.metall.

mater., 40:2547–2555, 1992. doi:10.1016/0956-7151(92)90324-8.

G. Bao, J.W. Hutchinson, and R.M. McMeeking. Particle reinforcement of ductile

matrices against plastic flow and creep. Acta metall.mater., 39:1871–1882, 1991.
doi:10.1016/0956-7151(91)90156-U.

R.B. Barello and M. Lévesque. Comparison between the relaxation spectra obtained
from homogenization models and finite elements simulation for the same composite.
Int.J.Sol.Struct., 45:850–867, 2008. doi:10.1016/j.ijsolstr.2007.09.002.

S. Bargmann, B. Klusemann, J. Markmann, J.E. Schnabel, K. Schneider, C. Soyarslan,

and J. Wilmers. Generation of 3D representative volume elements for heterogeneous ma-
terials: A review. Prog.Mater.Sci., 96:322–384, 2018. doi:10.1016/j.pmatsci.2018.02.003.

B. Barroqueiro, J. Dias-de-Oliveira, J. Pinho-da-Cruz, and A. Andrade-Campos.

Practical implementation of asymptotic expansion homogenisation in thermoelastic-
ity using a commercial simulation software. Compos.Struct., 141:117–131, 2016.
doi:10.1016/j.compstruct.2016.01.036.

J.F. Barthélémy. Simplified approach to the derivation of the relationship between Hill
polarization tensors of transformed problems and applications. Int.J.Engng.Sci., 154:
103326, 2020. doi:10.1016/j.ijengsci.2020.103326.

166
N.D. Barulich, L.A. Godoy, and P.M. Dardati. A simple procedure to evaluate Cauchy
stress tensor at the macro level based on computational micromechanics under general
finite strain states. Mech.Mater., 117:73–80, 2018. doi:10.1016/j.mechmat.2017.10.008.

J.L. Bassani, A. Needleman, and E. van der Giessen. Plastic flow in a composite: Com-
parison of nonlocal continuum and discrete dislocation predictions. Int.J.Sol.Struct., 38:
833–853, 2001. doi:10.1016/S0020-7683(00)00059-7.

V. Belsky, M.W. Beall, J. Fish, M.S. Shephard, and S. Gomaa. Computer-aided multiscale
modeling tools for composite materials and structures. Comput.Syst.Engng., 6:213–223,
1995. doi:10.1016/0956-0521(95)00019-V.

Y. Benveniste. A new approach to the application of Mori–Tanaka’s theory in composite

materials. Mech.Mater., 6:147–157, 1987. doi:10.1016/0167-6636(87)90005-6.

Y. Benveniste. Some remarks on three micromechanical models in composite media.

J.Appl.Mech., 57:474–476, 1990. doi:10.1115/1.2892016.

Y. Benveniste and G.J. Dvorak. On a correspondence between mechanical and thermal

effects in two-phase composites. In G.J. Weng, M. Taya, and H. Abé, editors, Mi-
cromechanics and Inhomogeneity, pages 65–82, New York, NY, 1990. Springer-Verlag.
doi:10.1007/978-1-4613-8919-4 4.

Y. Benveniste, G.J. Dvorak, and T. Chen. On diagonal and elastic symmetry of the
approximate effective stiffness tensor of heterogeneous media. J.Mech.Phys.Sol., 39:
927–946, 1991. doi:10.1016/0022-5096(91)90012-D.

A.A. Benzerga and J. Besson. Plastic potentials for anisotropic porous solids.
Eur.J.Mech. A/Solids, 20:397–434, 2001. doi:10.1016/S0997-7538(01)01147-0.

M.J. Beran and J. Molyneux. Use of classical variational principles to determine bounds
for the effective bulk modulus in heterogeneous media. Quart.Appl.Math., 24:107–118,
1966. doi:10.1090/qam/99925.

S. Berbenni. A time-incremental homogenization method for elasto-viscoplastic particulate

composites based on a modified secant formulation:. Int.J.Sol.Struct., 229:111136, 2021.

S.A. Berggren, D. Lukkassen, A. Meidell, and L. Simula. Some methods for cal-
culating stiffness properties of periodic structures. Appl.Math, 2:97–110, 2003.
doi:10.1023/A:1026090026531.

J.G. Berryman. Long-wavelength propagation in composite elastic media, II. Ellipsoidal

inclusions. J.Acoust.Soc.Amer., 68:1820–1831, 1980. doi:10.1121/1.385172.

J.G. Berryman and P.A. Berge. Critique of two explicit schemes for estimat-
ing elastic properties of multiphase composites. Mech.Mater., 22:149–164, 1996.
doi:10.1016/0167-6636(95)00035-6.

J.G. Berryman and G.W. Milton. Microgeometry of random composites and porous media.
J. Phys.D, 21:87–94, 1988. doi:10.1088/0022-3727/21/1/013.

167
M. Berveiller and A. Zaoui. A simplified self-consistent scheme for the plasticity of two-
phase metals. Res.Mech.Lett., 1:119–124, 1981.

A. Bhattacharyya and G.J. Weng. Plasticity of isotropic composites with randomly oriented
packeted inclusions. Int.J.Plast., 10:553–578, 1994. doi:10.1016/0749-6419(94)90014-0.

P. Bisegna and R. Luciano. Variational bounds for the overall properties of piezoelectric
composites. J.Mech.Phys.Sol., 44:583–602, 1996. doi:10.1016/0022-5096(95)00084-4.

J.E. Bishop, J.M. Emery, R.V. Field, C.R. Weinberger, and D.J. Littlewood. Direct
numerical simulations in solid mechanics for understanding the macroscale effects of
microscale material variability. Comput.Mech.Appl.Mech.Engng., 287:262–289, 2015.
doi:10.1016/j.cma.2015.01.017.

J.F.W. Bishop and R. Hill. A theory of the plastic distortion of a polycrystalline aggregate
under combined stress. London Edinburgh Dublin Phil.Mag.J.Sci., 42:414–427, 1951.
doi:10.1080/14786445108561065.

B. Bochenek and R. Pyrz. Reconstruction of random microstructures —

a stochastic optimization problem. Comput.Mater.Sci., 31:93–112, 2004.
doi:10.1016/j.commatsci.2004.01.038.

H.J. Böhm. A short introduction to continuum micromechanics. In H.J. Böhm, editor,

Mechanics of Microstructured Materials, pages 1–40. Springer-Verlag, CISM Courses and
Lectures Vol. 464, 2004. doi:10.1007/978-3-7091-2776-6 1.

H.J. Böhm. Comparison of analytical and numerical models for the thermoelastic behav-
ior of composites reinforced by coated spheres. Int.J.Engng.Sci., 142:216–229, 2019.
doi:10.1016/j.ijengsci.2019.06.009.

H.J. Böhm. A comparative study of analytical and numerical models for the elastic behavior
of composites reinforced by coated unidirectional fibers. Int.J.Sol.Struct., 264:112093,
2023. doi:10.1016/j.ijsolstr.2022.112093.

H.J. Böhm and W. Han. Comparisons between three-dimensional and two-dimensional

multi-particle unit cell models for particle reinforced metal matrix composites. Mod-
ell.Simul.Mater.Sci.Engng., 9:47–65, 2001. doi:10.1088/0965-0393/9/2/301.

H.J. Böhm and F.G. Rammerstorfer. Fiber arrangement effects on the microscale stresses of
continuously reinforced MMCs. In R. Pyrz, editor, Microstructure–Property Interactions
in Composite Materials, pages 51–62, Dordrecht, 1995. Kluwer Academic Publishers.
doi:10.1007/978-94-011-0059-5 5.

H.J. Böhm and A. Rasool. Effects of particle shape on the thermoelastoplas-

tic behavior of particle reinforced composites. Int.J.Sol.Struct., 87:90–101, 2016.
doi:10.1016/j.ijsolstr.2016.02.028.

H.J. Böhm, F.G. Rammerstorfer, and E. Weissenbek. Some simple models for microme-
chanical investigations of fiber arrangement effects in MMCs. Comput.Mater.Sci., 1:
177–194, 1993. doi:10.1016/0927-0256(93)90010-K.

168
H.J. Böhm, A. Eckschlager, and W. Han. Modeling of phase arrangement effects in high
speed tool steels. In F. Jeglitsch, R. Ebner, and H. Leitner, editors, Tool Steels in the
Next Century, pages 147–156, Leoben, 1999. Montanuniversität Leoben.

H.J. Böhm, A. Eckschlager, and W. Han. Multi-inclusion unit cell models for metal matrix
composites with randomly oriented discontinuous reinforcements. Comput.Mater.Sci.,
25:42–53, 2002. doi:10.1016/S0927-0256(02)00248-3.

H.J. Böhm, D.H. Pahr, and T. Daxner. Analytical and numerical methods for modeling
the thermomechanical and thermophysical behavior of microstructured materials. In
V.V. Silberschmidt, editor, Computational and Experimental Mechanics of Advanced
Materials, pages 167–223. Springer-Verlag, CISM Courses and Lectures Vol. 514, 2009.
doi:10.1007/978-3-211-99685-0 5.

N. Bonfoh, V. Hounkpati, and H. Sabar. New micromechanical approach of the coated

inclusion problem: Exact solution and applications. Comput.Mater.Sci., 62:175–183,
2012. doi:10.1016/j.commatsci.2012.05.007.

M. Bornert. Homogénéisation des milieux aléatoires: Bornes et estimations. In M. Bornert,

T. Bretheau, and P. Gilormini, editors, Homogénéisation en mécanique des matériaux
1. Matériaux aléatoires élastiques et milieux périodiques, pages 133–221, Paris, 2001.
Editions Hermès.

M. Bornert. Morphologie microstructurale et comportement mécanique; charactérisations

expérimentales, approches par bornes et estimations autocohérentes généralisées. PhD
thesis, Ecole Nationale des Ponts et Chaussées, Paris, 1996.

M. Bornert and P. Suquet. Propriétés non linéaires des composites: Approches par les
potentiels. In M. Bornert, T. Bretheau, and P. Gilormini, editors, Homogénéisation en
mécanique des matériaux 2. Comportements non linéaires et problèmes ouverts, pages
45–90, Paris, 2001. Editions Hermès.

M. Bornert, E. Hervé, C. Stolz, and A. Zaoui. Self consistent approaches and strain
heterogeneities in two-phase elastoplastic materials. Appl.Mech.Rev., 47:66–S76, 1994.
doi:10.1115/1.3122824.

M. Bornert, C. Stolz, and A. Zaoui. Morphologically representative pattern-based bounding

in elasticity. J.Mech.Phys.Sol., 44:307–331, 1996. doi:10.1016/0022-5096(95)00083-6.

M. Bornert, T. Bretheau, and P. Gilormini, editors. Homogénéisation en mécanique des

matériaux. Editions Hermès, Paris, 2001. ISBN 2-746-20199-2.

N. Bouhfid, M. Raji, R. Boujmal, H. Essabir, M.O. Bensalah, R. Bouhfid, and A.K. Qaiss.
Numerical modeling of hybrid composite materials. In M. Jawaid, M. Thariq, and
N. Saba, editors, Modeling of Damage Processes in Biocomposites, Fibre-Reinforced
Composites and Hybrid Composites, pages 57–101, Cambridge, UK, 2019. Elsevier.
doi:10.1016/B978-0-08-102289-4.00005-9.

D. Boussaa. Effective thermoelastic properties of composites with temperature-dependent

constituents. Mech.Mater., 43:397–407, 2011. doi:10.1016/j.mechmat.2011.04.004.

169
S. Boyd and R. Müller. Smooth surface meshing for automated finite ele-
ment model generation from 3D image data. J.Biomech., 39:1287–1295, 2006.
doi:10.1016/j.jbiomech.2005.03.006.
L. Brassart, L. Stainier, I. Doghri, and L. Delannay. Homogenization of elasto-(visco)plastic
composites based on an incremental variational principle. Int.J.Plast., 36:86–112, 2012.
doi:10.1016/j.ijplas.2012.03.010.
R. Brenner and R. Masson. Improved affine estimates for nonlinear viscoelastic composites.
Eur.J.Mech. A/Solids, 24:1002–1015, 2005. doi:10.1016/j.euromechsol.2005.06.004.
R. Brenner, O. Castelnau, and P. Gilormini. A modified affine theory for the overall
properties of nonlinear composites. C.R.Acad.Sci.Paris, série IIb, 329:649–654, 2001.
doi:10.1016/S1620-7742(01)01382-4.
L.C. Brinson and W.S. Lin. Comparison of micromechanics methods for effective
properties of multiphase viscoelastic composites. Compos.Struct., 41:353–367, 1998.
doi:10.1016/S0263-8223(98)00019-1.
S. Brisard and L. Dormieux. FFT-based methods for the mechanics of compos-
ites: A general variational framework. Comput.Mater.Sci., 49:663–671, 2010.
doi:10.1016/j.commatsci.2010.06.009.
L.M. Brown and W.M. Stobbs. The work-hardening of copper–silica. I. A model
based on internal stresses, with no plastic relaxation. Phil.Mag., 23:1185–1199, 1971.
doi:10.1080/14786437108217405.
D.A.G. Bruggemann. Berechnung verschiedener physikalischer Konstanten von heteroge-
nen Substanzen. I. Dielektrizitätskonstanten und Leitfähigkeiten der Mischkörper aus
isotropen Substanzen. Ann.Phys., 24:636–679, 1935. doi:10.1002/andp.19354160705.
M.S. Bruzzi, P.E. McHugh, F. O’Rourke, and T. Linder. Micromechanical modelling of
the static and cyclic loading of an Al 2141-SiC MMC. Int.J.Plast., 17:565–599, 2001.
doi:10.1016/S0749-6419(00)00063-2.
I. Bucataru and M.A. Slawinski. Invariant properties for finding distance in space of
elasticity tensors. J.Elast., 94:97–114, 2009. doi:10.1007/s10659-008-9186-9.
J.Y. Buffière, P. Cloetens, W. Ludwig, E. Maire, and L. Salvo. In situ X-ray tomography
studies of microstructural evolution combined with 3D modeling. MRS Bull., 33:611–619,
2008. doi:10.1557/mrs2008.126.
V.N. Bulsara, R. Talreja, and J. Qu. Damage initiation under transverse loading of uni-
directional composites with arbitrarily distributed fibers. Compos.Sci.Technol., 59:673–
682, 1999. doi:10.1016/S0266-3538(98)00122-5.
V. Buryachenko. Micromechanics of Heterogeneous Materials. Springer-Verlag. Berlin,
2007. ISBN 978-0-387-36827-6.
V. Buryachenko. Critical analysis of generalized Maxwell homogenization
schemes and related prospective problems. Mech.Mater., 165:104181, 2022b.
doi:10.1016/j.mechmat.2021.104181.

170
V.A. Buryachenko. General integral equations of micromechanics of heterogeneous
materials. J.Multisc.Comput.Engng., 13:11–53, 2015. doi:10.1615/IntJMultCom-
pEng.2014011234.

V.A. Buryachenko. Local and Nonlocal Micromechanics of Heterogeneous Materials.

Springer-Verlag. Cham, 2022a. ISBN 978-3-030-81783-1.

V.A. Buryachenko. The overall elastoplastic behavior of multiphase materials with isotropic
components. Acta Mech., 119:93–117, 1996. doi:10.1007/BF01274241.

V.A. Buryachenko, N.J. Pagano, R.Y. Kim, and J.E. Spowart. Quantitative description
and numerical simulation of random microstructures of composites and their elastic
moduli. Int.J.Sol.Struct., 40:47–72, 2003. doi:10.1016/S0020-7683(02)00462-6.

J. Byström, N. Jēkabsons, and J. Varna. An evaluation of different models for pre-

diction of elastic properties of woven composites. Composites B, 31:7–20, 2000.
doi:10.1016/S1359-8368(99)00061-X.

C. Calvo-Jurado and W.J. Parnell. The influence of two-point statistics on the Hashin–
Shtrikman bounds for three-phase composites. J.Comput.Appl.Math., 318:354–365, 2017.
doi:10.1016/j.cam.2016.08.046.

C.W. Camacho, C.L. Tucker, S. Yalvaç, and R.L. McGee. Stiffness and thermal expan-
sion predictions for hybrid short fiber composites. Polym.Compos., 11:229–239, 1990.
doi:10.1002/pc.750110406.

N. Carrère. Sur l’analyse multiéchelle des matériaux composites à matrice métallique:

application au calcul de structure. PhD thesis, Ecole Polytechnique, Palaiseau, 2001.

M.A.A. Cavalcante, M.J. Pindera, and H. Khatam. Finite-volume micromechanics of

periodic materials: Past, present and future. Composites B, 43:2521–2543, 2012.
doi:10.1016/j.compositesb.2012.02.006.

A. Cazzani and M. Rovati. Extrema of Young’s modulus for cubic and transversely isotropic
solids. Int.J.Sol.Struct., 40:1713–1744, 2003. doi:10.1016/S0020-7683(02)00668-6.

J.L. Chaboche and P. Kanouté. Sur les approximations “isotrope” et “anisotrope” de

l’opérateur tangent pour les méthodes tangentes incrémentale et affine. C.R.Mécanique,
331:857–864, 2003. doi:10.1016/j.crme.2003.08.002.

J.L. Chaboche, S. Kruch, J.F. Maire, and T. Pottier. Towards a micromechanics

based inelastic and damage modelling of composites. Int.J.Plast., 17:411–439, 2001.
doi:10.1016/S0749-6419(00)00056-5.

A. Chakraborty and S. Rahman. Stochastic multiscale models for fracture anal-

ysis of functionally graded materials. Engng.Fract.Mech., 75:2062–2068, 2008.
doi:10.1016/j.engfracmech.2007.10.013.

N. Charalambakis. Homogenization techniques and micromechanics. A survey and per-

spectives. Appl.Mech.Rev., 63:030803, 2010. doi:10.1115/1.4001911.

171
G. Chatzigeorgiou and F. Meraghni. Elastic and inelastic local strain fields in composites
with coated fibers or particles: Theory and validation. Math.Mech.Sol., 24:2858–2894,
2019. doi:10.1177/1081286518822695.

N. Chawla and K.K. Chawla. Microstructure based modeling of the deformation be-
havior of particle reinforced metal matrix composites. J.Mater.Sci., 41:913–925, 2006.
doi:10.1007/s10853-006-6572-1.

H.R. Chen, Q.S. Yang, and F.W. Williams. A self-consistent finite ele-
ment approach to the inclusion problem. Comput.Mater.Sci., 2:301–307, 1994.
doi:10.1016/0927-0256(94)90112-0.

Q. Chen, X. Chen, Z. Zhai, and Z. Yang. A new and general formu-

lation of three-dimensional finite-volume micromechanics for particulate rein-
forced composites with viscoplastic phases. Composites B, 85:216–232, 2016.
doi:10.1016/j.compositesb.2015.09.014.

T. Chen. Exact moduli and bounds of two-phase composites with coupled multifield linear
responses. J.Mech.Phys.Sol., 45:385–398, 1997. doi:10.1016/S0022-5096(96)00092-0.

B. Chiaia, A. Vervuurt, and J.G.M. van Mier. Lattice model evaluation of progres-
sive failure in disordered particle composites. Engng.Fract.Mech., 57:301–318, 1997.
doi:10.1016/S0013-7944(97)00011-8.

C.M. Chimani, H.J. Böhm, and F.G. Rammerstorfer. On stress singularities at free edges
of bimaterial junctions — A micromechanical study. Scr.mater., 36:943–947, 1997.
doi:10.1016/S1359-6462(96)00461-7.

R.M. Christensen. Sufficient symmetry conditions for isotropy of the elastic moduli tensor.
J.Appl.Mech., 54:772–777, 1987. doi:10.1115/1.3173115.

R.M. Christensen and K.H. Lo. Solutions for effective shear properties in
three phase sphere and cylinder models. J.Mech.Phys.Sol., 27:315–330, 1979.
doi:10.1016/0022-5096(79)90032-2.

R.M. Christensen and K.H. Lo. Erratum to Christensen and Lo, 1979. J.Mech.Phys.Sol.,
34:639, 1986. doi:10.1016/0022-5096(86)90043-8.

R.M. Christensen, H. Schantz, and J. Shapiro. On the range of validity of the Mori–Tanaka
method. J.Mech.Phys.Sol., 40:69–73, 1992. doi:10.1016/0022-5096(92)90240-3.

B.G. Christoff, H. Brito-Santana, S. Talreja, and V. Tita. Development of an ABAQUSTM

plug-in to evaluate the fourth-order elasticity tensor of a periodic material via homoge-
nization by the asymptotic expansion method. Fin.Elem.Anal.Design, 181:103482, 2020.
doi:10.1016/j.finel.2020.103482.

P.W. Chung, K.K. Tamma, and R.R. Namburu. Asymptotic expansion homogenization
for heterogeneous media: Computational issues and applications. Composites A, 32:
1291–1301, 2001. doi:10.1016/S1359-835X(01)00100-2.

172
H.H.M. Cleveringa, E. van der Giessen, and A. Needleman. Comparison of discrete dislo-
cation and continuum plasticity predictions for a composite material. Acta mater., 45:
3163–3179, 1997. doi:10.1016/S1359-6454(97)00011-6.

T.W. Clyne and P.J. Withers. An Introduction to Metal Matrix Composites. Cambridge
University Press, Cambridge, 1993. ISBN 0-521-48357-3.

I. Cohen. Simple algebraic approximations for the effective elastic moduli of cubic arrays
of spheres. J.Mech.Phys.Sol., 52:2167–2183, 2004. doi:10.1016/j.jmps.2004.02.008.

D.A. Colera and H.G. Kim. Asymptotic expansion homogenization analysis using two-
phase representative volume element for non-periodic composite materials. Mul-
tisc.Sci.Engng., 1:130–140, 2019. doi:10.1007/s42493-018-00014-w.

R.D. Cook and W.C. Young. Advanced Mechanics of Materials. Mcmillan, London, 1985.
ISBN 0-133-96961-4.

W.M.G. Courage and P.J.G Schreurs. Effective material parameters for compos-
ites with randomly oriented short fibers. Comput.Struct., 44:1179–1185, 1992.
doi:10.1016/0045-7949(92)90361-3.

B.N. Cox and G. Flanagan. Handbook of analytical methods for textile composites. Tech-
nical Report NASA–CR–4570, NASA, Washington, DC, 1997.

H.L. Cox. The elasticity and strength of paper and other fibrous materials. Brit.J.Appl.
Phys., 3:72–79, 1952. doi:10.1088/0508-3443/3/3/302.

M.E. Cruz and A.T. Patera. A parallel Monte-Carlo finite element procedure for the
analysis of multicomponent random media. Int.J.Num.Meth.Engng., 38:1087–1121, 1995.
doi:10.1002/nme.1620380703.

J. Cugnoni and M. Galli. Representative volume element size of elastoplastic and

elastoviscoplastic particle-reinforced composites with random microstructure. Com-
put.Model.Engng.Sci., 66:165–185, 2010. doi:10.3970/cmes.2010.066.165.

W.A. Curtin and R.E. Miller. Atomistic/continuum coupling in computa-

tional materials science. Model.Simul.Mater.Sci.Engng., 11:R33–R68, 2003.
doi:10.1088/0965-0393/11/3/201.

A.S. Dalaq, D.W. Abueidda, R. Abu Al-Rub, and I.M. Jasiuk. Finite element predictions of
effective elastic properties of interpenetrating phase composites with architectured sheet
reinforcements. Int.J.Sol.Struct., 83:169–182, 2016. doi:10.1016/j.ijsolstr.2016.01.011.

T.D. Dang and B.V. Sankar. Meshless Local Petrov–Galerkin formulation for problems in
composite micromechanics. AIAA J., 45:912–921, 2007. doi:10.2514/1.23434.

L.C. Davis. Flow rule for the plastic deformation of particulate metal matrix composites.
Comput.Mater.Sci., 6:310–318, 1996. doi:10.1016/0927-0256(96)00044-4.

P.R. Dawson. Computational crystal plasticity. Int.J.Sol.Struct., 37:115–130, 2000.

doi:10.1016/S0020-7683(99)00083-9.

173
T. Daxner. Multi-Scale Modeling and Simulation of Metallic Foams. Reihe 18, Nr. 285.
VDI–Verlag, Düsseldorf, 2003. ISBN 3-183-28581-5.
T. Daxner, R.D. Bitsche, and H.J. Böhm. Micromechanical models of metallic sponges
with hollow struts. In T. Chandra, K. Tsuzaki, M. Militzer, and C. Ravindran, edi-
tors, Thermec 2006, pages 1857–1862, Trans Tech Publications, Zurich, 2007. Materials
Science Forum 539–543. doi:10.4028/www.scientific.net/MSF.539-543.1857.
F. de Francqueville, P. Gilormini, and J. Diani. Representative volume elements for the
simulation of isotropic composites highly filled with monosized spheres. Int.J.Sol.Struct.,
158:277–286, 2019. doi:10.1016/j.ijsolstr.2018.09.013.
P.H. Dederichs and T. Zeller. Variational treatment of the elastic constants of disordered
materials. Z.Phys., 259:103–116, 1973. doi:10.1007/BF01392841.
L. Delannay, I. Doghri, and O. Pierard. Prediction of tension–compression cycles in mul-
tiphase steel using a modified mean-field model. Int.J.Sol.Struct., 44:7291–7306, 2007.
doi:10.1016/j.ijsolstr.2007.04.013.
V.S. Deshpande, M.F. Ashby, and N.A. Fleck. Foam topology: Bending
versus stretching dominated architectures. Acta mater., 49:1035–1040, 2001.
doi:10.1016/S1359-6454(00)00379-7.
A. Desmoulins and D.M. Kochmann. Local and nonlocal continuum modeling
of inelastic periodic networks applied to stretching-dominated trusses. Com-
put.Meth.Appl.Mech.Engng., 313:85–105, 2017. doi:10.1016/j.cma.2016.09.027.
S. Diebels, T. Ebinger, and H. Steeb. An anisotropic damage model of foams
on the basis of a micromechanical description. J.Mater.Sci., 40:5919–5924, 2005.
doi:10.1007/s10853-005-5043-4.
F. Dinzart, H. Sabar, and S. Berbenni. Homogenization of multi-phase composites based
on a revisited formulation of the multi-coated inclusion problem. Int.J.Engng.Sci., 100:
136–151, 2016. doi:10.1016/j.ijengsci.2015.12.001.
J. Dirrenberger, S. Forest, and D. Jeulin. Towards gigantic RVE sizes for 3D stochastic
fibrous networks. Int.J.Sol.Struct., 51:359–376, 2014. doi:10.1016/j.ijsolstr.2013.10.011.
I. Doghri and C. Friebel. Effective elasto-plastic properties of inclusion-reinforced com-
posites. Study of shape, orientation and cyclic response. Mech.Mater., 37:45–68, 2005.
doi:10.1016/j.mechmat.2003.12.007.
I. Doghri and A. Ouaar. Homogenization of two-phase elasto-plastic com-
posite materials and structures. Int.J.Sol.Struct., 40:1681–1712, 2003.
doi:10.1016/S0020-7683(03)00013-1.
I. Doghri and L. Tinel. Micromechanical modeling and computation of elasto-plastic ma-
terials reinforced with distributed-orientation fibers. Int.J.Plast., 21:1919–1940, 2005.
doi:10.1016/j.ijplas.2004.09.003.
M. Dong and S. Schmauder. Modeling of metal matrix composites by a self-consistent em-
bedded cell model. Acta mater., 44:2465–2478, 1996. doi:10.1016/1359-6454(95)00345-2.

174
A. Drago and M.J. Pindera. A locally exact homogenization theory for periodic microstruc-
tures with isotropic phases. J.Appl.Mech., 75:051010, 2008. doi:10.1115/1.2913043.

W. Dreyer, W.H. Müller, and J. Olschewski. An approximate analytical 2D-solution for

the stresses and strains in eigenstrained cubic materials. Acta Mech., 136:171–192, 1999.
doi:10.1007/BF01179256.

W.J. Drugan and J.R. Willis. A micromechanics-based nonlocal constitutive equation and
estimates of representative volume element size for elastic composites. J.Mech.Phys.Sol.,
44:497–524, 1996. doi:10.1016/0022-5096(96)00007-5.

D.X. Du and Q.S. Zheng. A further exploration of the interaction direct derivative (IDD)
estimate for the effective properties of multiphase composites taking into account inclu-
sion distribution. Acta Mech., 157:61–80, 2002. doi:10.1007/BF01182155.

H.L. Duan, J. Wang, Z.P. Huang, and Z.Y. Luo. Stress concentration ten-
sors of inhomogeneities with interface effects. Mech.Mater., 37:723–736, 2005.
doi:10.1016/j.mechmat.2004.07.004.

H.L. Duan, B.L. Karihaloo, J. Wang, and X. Yi. Effective thermal conductivities of
heterogeneous media containing multiple inclusions with various spatial distributions.
Phys.Rev. B, 73:174203, 2006. doi:10.1103/PhysRevB.73.174203.

H.L. Duan, J. Wang, and Z. Huang. Micromechanics of composites with interface effects.
Acta Mech.Sinica, 38:222025, 2022. doi:10.1007/s10409-022-22025-x.

C.F. Dunant, B. Bary, A.B. Giorla, C. Péniguel, M. Sanahuja, C. Toulemonde, A.B. Tran,
F. Willot, and J. Yvonnet. A critical comparison of several numerical methods for
computing effective properties of highly heterogeneous materials. Adv.Engng.Softw., 58:
1–12, 2013. doi:10.1016/j.advengsoft.2012.12.002.

M.L. Dunn and H. Ledbetter. Elastic–plastic behavior of textured short-fiber composites.

Acta mater., 45:3327–3340, 1997. doi:10.1016/S1359-6454(96)00401-6.

M.L. Dunn and M. Taya. Micromechanics predictions of the effective electroe-

lastic moduli of piezoelectric composites. Int.J.Sol.Struct., 30:161–175, 1993.
doi:10.1016/0020-7683(93)90058-F.

D. Duschlbauer. Computational Simulation of the Thermal Conductivity of MMCs under

Consideration of the Inclusion–Matrix Interface. Reihe 5, Nr. 561. VDI–Verlag, Düssel-
dorf, 2004. ISBN 3-183-69105-1.

D. Duschlbauer, H.J. Böhm, and H.E. Pettermann. Numerical simulation of the thermal
conductivity of MMCs: The effect of thermal interface resistance. Mater. Sci. Technol.,
19:1107–1114, 2003a. doi:10.1179/026708303225004305.

D. Duschlbauer, H.E. Pettermann, and H.J. Böhm. Mori–Tanaka based evaluation of

inclusion stresses in composites with nonaligned reinforcements. Scr.mater., 48:223–228,
2003b. doi:10.1016/S1359-6462(02)003-1.

175
D. Duschlbauer, H.J. Böhm, and H.E. Pettermann. Computational simulation of
composites reinforced by planar random fibers: Homogenization and localization
by unit cell and mean field approaches. J.Compos.Mater., 40:2217–2234, 2006.
doi:10.1177/0021998306062317.

D. Duschlbauer, H.E. Pettermann, and H.J. Böhm. Modeling interfacial effects on the
thermal conduction behavior of short fiber reinforced composites. Int.J.Mater.Res., 102:
717–728, 2011. doi:10.3139/146.110515.

G.J. Dvorak. Micromechanics of Composite Materials. Springer-Verlag, Dordrecht, 2013.

ISBN 9-400-79781-8.

G.J. Dvorak. Transformation field analysis of inelastic composite materials. Proc.Roy.Soc.

London, A437:311–327, 1992. doi:10.1098/rspa.1992.0063.

G.J. Dvorak and Y.A. Bahei-el Din. A bimodal plasticity theory of fibrous composite
materials. Acta Mech., 69:219–241, 1987. doi:10.1007/BF01175723.

G.J. Dvorak, T. Chen, and J. Teply. Thermomechanical stress fields in high-temperature

fibrous composites. I: Unidirectional laminates. Compos.Sci.Technol., 43:347–358, 1992.
doi:10.1016/0266-3538(92)90058-B.

G.J. Dvorak, Y.A. Bahei-el Din, and A.M. Wafa. The modeling of inelastic composite
materials with the transformation field analysis. Modell.Simul.Mater.Sci.Engng., 2:571–
586, 1994. doi:10.1088/0965-0393/2/3A/011.

M.P. Echlin, W.C. Lenthe, and T.M. Pollock. Three-dimensional sampling of material
structure for property modeling and design. Integ.Mater.Manuf.Innov., 3:21, 2014.
doi:10.1186/s40192-014-0021-9.

R.F. Eduljee and R.L. McCullough. Elastic properties of composites. In R.W. Cahn,
P. Haasen, and E.J. Kramer, editors, Materials Science and Technology Vol.13:
Structure and Properties of Composites, pages 381–474, Weinheim, 1993. VCH.
doi:10.1002/9783527603978.mst0155.

F. El Houdaigui, S. Forest, A.F. Gourgues, and D. Jeulin. On the size of the representative
volume element for isotropic elastic copper. In Y.L. Bai and Q.S. Zheng, editors, IUTAM
Symposium on Mechanical Behavior and Micro-Mechanics of Nanostructured Materials,
pages 171–180, Dordrecht, 2007. Springer-Verlag. doi:10.1007/978-1-4020-5624-6 17.

M. El Mouden and A. Molinari. Thermoelastic properties of composites

containing ellipsoidal inhomogeneities. J.Therm.Stresses, 23:233–255, 2000.
doi:10.1080/014957300280425.

J.A. Elliott and A.H. Windle. A dissipative particle dynamics method for modeling the
geometrical packing of filler particles in polymer composites. J.Chem.Phys., 113:10367–
10376, 2000. doi:10.1063/1.1322636.

O. Eroshkin and I. Tsukrov. On micromechanical modeling of particulate com-

posites with inclusions of various shapes. Int.J.Sol.Struct., 42:409–427, 2005.
doi:10.1016/j.ijsolstr.2004.06.045.

176
J.D. Eshelby. The determination of the elastic field of an ellipsoidal inclusion and related
problems. Proc.Roy.Soc.London, A241:376–396, 1957. doi:10.1098/rspa.1957.0133.

J.D. Eshelby. The elastic field outside an ellipsoidal inclusion. Proc.Roy.Soc.London, A252:
561–569, 1959. doi:10.1098/rspa.1959.0173.

Y. Estrin, S. Arndt, M. Heilmaier, and Y. Bréchet. Deformation behaviour of particle-

strengthened alloys: A Voronoi mesh approach. Acta mater., 47:595–606, 1999.
doi:10.1016/S1359-6454(98)00362-0.

J. Faleskog and C.F. Shih. Micromechanics of coalescence — I.: Synergistic effects of

elasticity, plastic yielding and multi-size-scale voids. J.Mech.Phys.Sol., 45:21–50, 1997.
doi:10.1016/S0022-5096(96)00078-6.

G. Fang, B. El Said, D. Ivanov, and S.R. Hallett. Smoothing artificial stress concen-
trations in voxel-based models of textile composites. Composites A, 80:270–284, 2016.
doi:10.1016/j.compositesa.2015.10.025.

Y.F. Fangye, N. Miska, and D. Balzani. Automated simulation of voxel-based microstruc-

tures based on enhanced finite cell approach. Arch.Appl.Mech., 90:2255–2273, 2020.
doi:10.1007/s00419-020-01719-x.

M. Fergoug, A. Parret-Fréaud, N. Feld, B. Marchand, and S. Forest. A general boundary

layer corrector for the asymptotic homogenization of elastic linear composite structures.
Compos.Struct., 285:115091, 2022. doi:10.1016/j.compstruct.2021.115091.

M. Ferrari. Asymmetry and the high concentration limit of the Mori–Tanaka effective
medium theory. Mech.Mater., 11:251–256, 1991. doi:10.1016/0167-6636(91)90006-L.

F. Feyel. A multilevel finite element method (FE2 ) to describe the response of highly
non-linear structures using generalized continua. Comput.Meth.Appl.Mech.Engng., 192:
3233–3244, 2003. doi:10.1016/S0045-7825(03)00348-7.

F. Feyel and J.L. Chaboche. FE2 multiscale approach for modelling the elastoviscoplastic
behaviour of long fiber SiC/Ti composite materials. Comput.Meth.Appl.Mech.Engng.,
183:309–330, 2000. doi:10.1016/S0045-7825(99)00224-8.

S. Firooz, S. Saeb, G. Chatzigeorgiou, F. Meraghni, P. Steinmann, and A. Javili. System-

atic study of homogenization and the utility of circular simplified representative volume
element. Math.Mech.Sol., 24:2961–2985, 2019. doi:10.1177/1081286518823834.

H.F. Fischmeister and B. Karlsson. Plastizitätseigenschaften grob-zweiphasiger Werkstoffe.

Z.Metallkd., 68:311–327, 1977. doi:10.1515/ijmr-1977-680501.

J. Fish and K. Shek. Multiscale analysis of composite materials and structures. Com-
pos.Sci.Technol., 60:2547–2556, 2000. doi:10.1016/S0266-3538(00)00048-8.

J. Fish and K. Shek. Computational plasticity and viscoplasticity for composite materials
and structures. Composites B, 29:613–619, 1998. doi:10.1016/S1359-8368(98)00015-8.

177
S. Fliegener, M. Luke, and P. Gumbsch. 3D microstructure modeling of
long fiber reinforced thermoplastics. Compos.Sci.Technol., 104:136–145, 2014.
doi:10.1016/j.compscitech.2014.09.009.

C. Fond, A. Riccardi, R. Schirrer, and F. Montheillet. Mechanical interaction between

spherical inhomogeneities: An assessment of a method based on the equivalent inclusion.
Eur.J.Mech. A/Solids, 20:59–75, 2001. doi:10.1016/S0997-7538(00)01118-9.

S. Forest, G. Cailletaud, D. Jeulin, F. Feyel, I. Galliet, V. Mounoury, and S. Quilici.

Introduction au calcul de microstructures. Méc.Indust., 3:439–456, 2002. URL
https://www.sciencedirect.com/science/article/pii/S1296213902011879.

P. Franciosi. Uniformity of the Green operator and Eshelby tensor for hy-
perboloidal domains in infinite media. Math.Mech.Sol., 25:1610–1642, 2020.
doi:10.1177/108128652091119.

H. Fricke. A mathematical treatment of the electric conductivity and capacity of dis-

perse systems I. The electric conductivity of a suspension of homogeneous spheroids.
Phys.Rev., 24:575–587, 1924. doi:10.1103/PhysRev.24.575.

C. Friebel, I. Doghri, and V. Legat. General mean-field homogenization schemes for vis-
coelastic composites containing multiple phases of coated inclusions. Int.J.Sol.Struct.,
43:2513–2541, 2006. doi:10.1016/j.ijsolstr.2005.06.035.

F. Fritzen and T. Böhlke. Periodic three-dimensional mesh generation for particle rein-
forced composites with application to metal matrix composites. Int.J.Sol.Struct., 48:
706–718, 2011. doi:10.1016/j.ijsolstr.2010.11.010.

F. Fritzen and O. Kunc. Two-stage data driven homogenization for nonlinear

solids using a reduced order model. Eur.J.Mech. A/Solids, 69:201–220, 2018.
doi:10.1016/j.euromechsol.2017.11.007.

F. Fritzen, T. Böhlke, and E. Schnack. Periodic three-dimensional mesh generation for

crystalline aggregates based on Voronoi tessellations. Comput.Mech., 43:701–713, 2009.
doi:10.1007/s00466-008-0339-2.

S.Y. Fu and B. Lauke. The elastic modulus of misaligned short-fiber-reinforced polymers.

Compos.Sci.Technol., 58:389–400, 1998. doi:10.1016/S0266-3538(97)00129-2.

H. Fukuda and T.W. Chou. A probabilistic theory of the strength of short-fibre com-
posites with variable fibre length and orientation. J.Mater.Sci., 17:1003–1007, 1982.
doi:10.1007/BF00543519.

H. Fukuda and K. Kawata. Stress and strain fields in short fibre-reinforced composites.
Fibre Sci.Technol., 7:129–156, 1974. doi:10.1016/0015-0568(74)90025-6.

J. Gager and H.E. Pettermann. Numerical homogenization of textile composites

based on shell element discretization. Compos.Sci.Technol., 72:806–812, 2012.
doi:10.1016/j.compscitech.2012.02.009.

178
M. Galli, J. Botsis, and J. Janczak-Rusch. An elastoplastic three-dimensional homoge-
nization model for particle reinforced composites. Comput.Mater.Sci., 41:312–321, 2008.
doi:10.1016/j.commatsci.2007.04.010.
M. Galli, J. Cugnoni, and J. Botsis. Numerical and statistical estimates of the represen-
tative volume element of elastoplastic random composites. Eur.J.Mech. A/Solids, 33:
31–38, 2012. doi:10.1016/j.euromechsol.2011.07.010.
H.P. Gänser. Large Strain Behavior of Two-Phase Materials. Reihe 5, Nr. 528. VDI–Verlag,
Düsseldorf, 1998.
H.P. Gänser, F.D. Fischer, and E.A. Werner. Large strain behaviour of two-
phase materials with random inclusions. Comput.Mater.Sci., 11:221–226, 1998.
doi:10.1016/S0927-0256(98)00007-X.
M. Gărăjeu, J.C. Michel, and P. Suquet. A micromechanical approach of damage in
viscoplastic materials by evolution in size, shape and distribution of voids. Com-
put.Meth.Appl.Mech.Engng., 183:223–246, 2000. doi:10.1016/S0045-7825(99)00220-0.
M.R. Garnich and G. Karami. Finite element micromechanics for stiffness
and strength of wavy fiber composites,. J.Compos.Mater., 38:273–292, 2004.
doi:10.1177/00219983040392.
D. Garoz, F.A. Gilabert, R.D.B. Sevenois, S.W.F Spronk, and W. Van Paepegem.
Consistent application of periodic boundary conditions in implicit and explicit fi-
nite element simulations of damage in composites. Composites B, 168:254–266, 2019.
doi:10.1016/j.compositesb.2018.12.023.
A.C. Gavazzi and D.C. Lagoudas. On the numerical evaluation of Eshelby’s tensor
and its application to elastoplastic fibrous composites. Comput.Mech., 7:12–19, 1990.
doi:10.1007/BF00370053.
M.G.D. Geers, R.A.B. Engelen, and R.J.M. Ubachs. On the numerical modelling of ductile
damage with an implicit gradient-enhanced formulation. Rev.Eur.Elem.Fin., 10:173–191,
2001a. doi:10.1080/12506559.2001.11869246.
M.G.D. Geers, V.G. Kouznetsova, and W.A.M. Brekelmans. Gradient enhanced ho-
mogenization for the micro–macro scale transition. J.Physique IV, 11:145–152, 2001b.
doi:10.1051/jp4:2001518.
M.G.D. Geers, V.G. Kouznetsova, and W.A.M. Brekelmans. Multi-scale computational
homogenization: Trends and challenges. J.Comput.Appl.Math., 234:2175–2182, 2010.
doi:10.1016/j.cam.2009.08.077.
G.M. Genin and V. Birman. Micromechanics and structural response of functionally
graded, particulate–matrix, fiber-reinforced composites. Int.J.Sol.Struct., 46:2136–2150,
2009. doi:10.1016/j.ijsolstr.2008.08.010.
A. Ghazavizadeh, M. Haboussi, A. Abdul-Latif, A. Jafari, and H. Bousoura. A
general and explicit Eshelby-type estimator for evaluating the equivalent stiffness
of multiply coated ellipsoidal heterogeneities. Int.J.Sol.Struct., 171:103–116, 2019.
doi:10.1016/j.ijsolstr.2019.04.023.

179
S. Ghosh and D.V. Kubair. Exterior statistics based boundary conditions for repre-
sentative volume elements of elastic composites. J.Mech.Phys.Sol., 95:1–24, 2016.
doi:10.1016/j.jmps.2016.05.022.
S. Ghosh and S. Moorthy. Three dimensional Voronoi cell finite element model for
microstructures with ellipsoidal heterogeneities. Comput.Mech., 34:510–531, 2004.
doi:10.1007/s00466-004-0598-5.
S. Ghosh, K.H. Lee, and S. Moorthy. Two scale analysis of heterogeneous elastic-plastic
materials with asymptotic homogenization and Voronoi cell finite element model. Com-
put.Meth.Appl.Mech.Engng., 132:63–116, 1996. doi:10.1016/0045-7825(95)00974-4.
S. Ghosh, K.H. Lee, and P. Raghavan. A multi-level computational model for multi-
scale analysis in composite and porous materials. Int.J.Sol.Struct., 38:2335–2385, 2001.
doi:10.1016/S0020-7683(00)00167-0.
S. Ghosh, D. Dimiduk, and D. Furrer. Statistically equivalent representative vol-
ume elements (SERVE) for material behaviour analysis and multiscale modelling.
Int.Mater.Rev., 68:1158–1191, 2023. doi:10.1080/09506608.2023.2246766.
E. Ghossein and M. Lévesque. A fully automated numerical tool for a comprehensive
validation of homogenization models and its application to spherical particles reinforced
composites. Int.J.Sol.Struct., 49:1387–1398, 2012. doi:10.1016/j.ijsolstr.2012.02.021.
E. Ghossein and M. Lévesque. Random generation of periodic hard ellipsoids based on
molecular dynamics: A computationally-efficient algorithm. J.Comput.Phys., 253:471–
490, 2013. doi:10.1016/j.jcp.2013.07.004.
L.V. Gibiansky and S. Torquato. Connection between the conductivity and bulk mod-
ulus of isotropic composite materials. Proc.Roy.Soc.London, A452:253–283, 1996.
doi:10.1098/rspa.1996.0015.
L.J. Gibson. The mechanical behavior of cancellous bone. J.Biomech., 18:317–328, 1985.
doi:10.1016/0021-9290(85)90287-8.
L.J. Gibson and M.F. Ashby. Cellular Solids: Structure and Properties. Pergamon Press,
Oxford, 1988. ISBN 0-521-49911-9.
A. Gillman, G. Amadio, K. Matouš, and T.L. Jackson. Third-order thermo-mechanical
properties for packs of Platonic solids using statistical micromechanics. Proc.Roy.Soc.,
A471:20150060, 2015. doi:10.1098/rspa.2015.0060.
P. Gilormini. Insuffisance de l’extension classique du modèle auto-cohérent au comporte-
ment non-linéaire. C.R.Acad.Sci.Paris, série IIb, 320:115–122, 1995.
S. Giordano. Differential schemes for the elastic characterization of disper-
sions of randomly oriented ellipsoids. Eur.J.Mech. A/Solids, 22:885–902, 2003.
doi:10.1016/S0997-7538(03)00091-3.
S. Giordano. Order and disorder in heterogeneous material microstructure: Electric and
elastic characterisation of dispersions of pseudo-oriented spheroids. Int.J.Engng.Sci., 43:
1033–1058, 2005. doi:10.1016/j.ijengsci.2005.06.002.

180
A. Giraud, C. Gruescu, D.P. Do, F. Homand, and D. Kondo. Effective thermal conduc-
tivity of transversely isotropic media with arbitrary oriented ellipsoidal inhomogeneities.
Int.J.Sol.Struct., 44:2627–2647, 2007. doi:10.1016/j.ijsolstr.2006.08.011.

I.M. Gitman. Representative Volumes and Multi-Scale Modelling of Quasi-Brittle Materi-

als. PhD thesis, Technische Universiteit te Delft, Delft, 2006.

I.M. Gitman, H. Askes, and L.J. Sluys. Representative volume: Existence and size determi-
nation. Engng.Fract.Mech., 74:2518–2534, 2007. doi:10.1016/j.engfracmech.2006.12.021.

R. Glüge, M. Weber, and A. Bertram. Comparison of spherical and cubical statistical

volume elements with respect to convergence, anisotropy, and localization behavior.
Comput.Mater.Sci., 63:91–104, 2012. doi:10.1016/j.commatsci.2012.05.063.

M. Gologanu, J.B. Leblond, G. Perrin, and J. Devaux. Recent extensions of Gurson’s

model for porous ductile materials. In P. Suquet, editor, Continuum Micromechanics,
pages 61–130, Vienna, 1997. Springer-Verlag, CISM Courses and Lectures Vol. 377.
doi:10.1007/978-3-7091-2662-2 2.

B. Gommers, I. Verpoest, and P. van Houtte. The Mori–Tanaka method

applied to textile composite materials. Acta mater., 46:2223–2235, 1998.
doi:10.1016/S1359-6454(97)00296-6.

L. Gong, S. Kyriakides, and N. Triantafyllidis. On the stability of Kelvin cell foams under
compressive loads. J.Mech.Phys.Sol., 53:771–794, 2005. doi:10.1016/j.jmps.2004.10.007.

C. González and J. LLorca. A self-consistent approach to the elasto-plastic behav-

ior of two-phase materials including damage. J.Mech.Phys.Sol., 48:675–692, 2000.
doi:10.1016/S0022-5096(99)00057-5.

C. González and J. LLorca. Virtual fracture testing of composites: A com-

putational micromechanics approach. Engng.Fract.Mech., 74:1126–1138, 2007.
doi:10.1016/j.engfracmech.2006.12.013.

M. Goudarzi and A. Simone. Discrete inclusion model for reinforced composites: Compar-
ative performance analysis and modeling challenges. Comput.Meth.Appl.Mech.Engng.,
355:535–557, 2019. doi:10.1016/j.cma.2019.06.026.

J.L. Grenestedt. Influence of wavy imperfections in cell walls on elastic stiffness of cellular
solids. J.Mech.Phys.Sol., 46:29–50, 1998. doi:10.1016/S0022-5096(97)00035-5.

D. Gross and T. Seelig. Bruchmechanik mit einer Einführung in die Mikromechanik.

Springer-Verlag, Berlin, 2001. ISBN 978-3-662-46736.

C. Grufman and F. Ellyin. Numerical modeling of damage susceptibility of an in-

homogeneous representative material volume element of polymer composites. Com-
pos.Sci.Technol., 68:650–657, 2008. doi:10.1016/j.compscitech.2007.09.018.

H. Gu, J. Réthore, M.C. Baietto, P. Sainsot, P. Lecomte-Grosbras, C.H. Venner, and A.A.
Lubrecht. An efficient MultiGrid solver for the 3D simulation of composite materials.
Comput.Mater.Sci., 112:230–237, 2016. doi:10.1016/j.commatsci.2015.10.025.

181
F.J. Guild and A.J. Kinloch. Modelling the properties of rubber-modified epoxy polymers.
J.Mater.Sci, 30:1689–1697, 1995. doi:10.1007/BF00351597.

R.E. Guldberg, S.J. Hollister, and G.T. Charras. The accuracy of digital image-based finite
element models. J.Biomech.Engng., 120:289–295, 1998. doi:10.1115/1.2798314.

G. Guo, J. Fitoussi, D. Baptiste, N. Sicot, and C. Wolff. Modelling of damage behavior

of a short-fiber reinforced composite structure by the finite element analysis using a
micro–macro law. Int.J.Dam.Mech., 6:278–299, 1997. doi:10.1177/105678959700600304.

A.L. Gurson. Continuum theory of ductile rupture by void nucleation and growth: Part
I — Yield criteria and flow rules for porous ductile media. J.Engng.Mater.Technol., 99:
2–15, 1977. doi:10.1115/1.3443401.

A.A. Gusev. Controlled accuracy finite element estimates for the effective stiff-
ness of composites with spherical inclusions. Int.J.Sol.Struct., 80:227–236, 2016.
doi:10.1016/j.ijsolstr.2015.11.006.

A.A. Gusev. Representative volume element size for elastic composites: A numerical study.
J.Mech.Phys.Sol., 45:1449–1459, 1997. doi:10.1016/S0022-5096(97)00016-1.

A.A. Gusev, P.J. Hine, and I.M. Ward. Fiber packing and elastic properties of a trans-
versely random unidirectional glass/epoxy composite. Compos.Sci.Technol., 60:535–541,
2000. doi:10.1016/S0266-3538(99)00152-9.

R.M. Haj-Ali and J. Aboudi. Formulation of the high-fidelity generalized method of cells
with arbitrary cell geometry for refined micromechanics and damage in composites.
Int.J.Sol.Struct., 47:3447–3461, 2010. doi:10.1016/j.ijsolstr.2010.08.022.

J.C. Halpin and J.L. Kardos. The Halpin–Tsai equations: A review. Polym.Engng.Sci.,
16:344–351, 1976. doi:10.1002/pen.760160512.

L.T. Harper, C. Qian, T.A. Turner, S. Li, and N.A. Warrior. Representative volume
elements for discontinuous carbon fibre composites — Part 1: Boundary conditions.
Compos.Sci.Technol., 72:225–234, 2012a. doi:10.1016/j.compscitech.2011.11.006.

L.T. Harper, C. Qian, T.A. Turner, S. Li, and N.A. Warrior. Representative volume
elements for discontinuous carbon fibre composites — Part 2: Determining critical size.
Compos.Sci.Technol., 72:204–210, 2012b. doi:10.1016/j.compscitech.2011.11.003.

A.M. Harte and J.F. Mc Namara. Use of micromechanical modelling in the material char-
acterisation of overinjected thermoplastic composites. J.Mater.Process.Technol., 173:
376–383, 2006. doi:10.1016/j.jmatprotec.2005.12.010.

R. Hashemi, R. Avazmohammadi, H.M. Shodja, and Weng G.J. Compos-

ites with superspherical inhomogeneities. Phil.Mag.Lett., 89:439–451, 2009.
doi:10.1080/09500830903019020.

Z. Hashin. The elastic moduli of heterogeneous materials. J.Appl.Mech., 29:143–150, 1962.

doi:10.1115/1.3636446.

182
Z. Hashin. Viscoelastic behavior of heterogeneous media. J.Appl.Mech., 32:630–636, 1965.
doi:10.1115/1.3627270.

Z. Hashin. Complex moduli of viscoelastic composites: I. General theory

and application to particulate composites. Int.J.Sol.Struct., 6:539–552, 1970.
doi:10.1016/0020-7683(70)90029-6.

Z. Hashin. Theory of fiber reinforced materials. Technical Report NASA–CR–1974, NASA,

Washington, DC, 1972.

Z. Hashin. Analysis of composite materials — A survey. J.Appl.Mech., 50:481–505, 1983.

doi:10.1115/1.3167081.

Z. Hashin. The differential scheme and its application to cracked materials. J.Mech.Phys.
Sol., 36:719–733, 1988. doi:10.1016/0022-5096(88)90005-1.

Z. Hashin and B.W. Rosen. The elastic moduli of fiber-reinforced materials. J.Appl.Mech.,
31:223–232, 1964. doi:10.1115/1.3629590.

Z. Hashin and S. Shtrikman. Note on a variational approach to the theory of composite

elastic materials. J.Franklin Inst., 271:336–341, 1961. doi:10.1016/0016-0032(61)90032-1.

Z. Hashin and S. Shtrikman. A variational approach to the theory of the effective

magnetic permeability of multiphase materials. J.Appl.Phys., 33:3125–3131, 1962.
doi:10.1063/1.1728579.

Z. Hashin and S. Shtrikman. A variational approach to the theory of the

elastic behavior of multiphase materials. J.Mech.Phys.Sol., 11:127–140, 1963.
doi:10.1016/0022-5096(63)90060-7.

B. Hassani and E. Hinton. Homogenization and Structural Topology Optimization. Springer-

Verlag. London, 1999. ISBN 1-447-11229-6.

D.P.H. Hasselman and L.F. Johnson. Effective thermal conductivity of composites

with interfacial thermal barrier resistance. J.Compos.Mater., 21:508–515, 1987.
doi:10.1177/002199838702100602.

H. Hatta and M. Taya. Equivalent inclusion method for steady state heat conduction in
composites. Int.J.Engng.Sci., 24:1159–1172, 1986. doi:10.1016/0020-7225(86)90011-X.

S. Hazanov. Hill condition and overall properties of composites. Arch.Appl.Mech., 68:

385–394, 1998. doi:10.1007/s004190050173.

S. Hazanov and M. Amieur. On overall properties of elastic bodies smaller

than the representative volume. Int.J.Engng.Sci., 33:1289–1301, 1995.
doi:10.1016/0020-7225(94)00129-8.

S. Hazanov and C. Huet. Order relationships for boundary condition effects in heteroge-
neous bodies smaller than the representative volume. J.Mech.Phys.Sol., 41:1995–2011,
1994. doi:10.1016/0022-5096(94)90022-1.

183
Q.C. He and A. Curnier. A more fundamental approach to damaged elastic stress–strain
relations. Int.J.Sol.Struct., 32:1433–1457, 1997. doi:10.1016/0020-7683(94)00183-W.

G.L. Heness, B. Ben-Nissan, L.H. Gan, and Y.W. Mai. Development of a finite ele-
ment micromodel for metal matrix composites. Comput.Mater.Sci., 13:259–269, 1999.
doi:10.1016/S0927-0256(98)00127-X.

A. Henkes and H. Wessels. Three-dimensional microstructure generation using genera-

tive adversarial neural networks in the context of continuum micromechanics. Com-
put.Meth.Appl.Mech.Engng., 400:115497, 2022. doi:10.1016/j.cma.2022.115497.

J.J. Hermans. The elastic properties of fiber reinforced materials when the fibers are
aligned. Proc.K.Ned.Akad.Wet., B70:1–9, 1967.

E. Hervé and A. Zaoui. Modelling the effective behavior of nonlinear matrix–inclusion

composites. Eur.J.Mech. A/Solids, 9:505–515, 1990.

E. Hervé and A. Zaoui. n-layered inclusion-based micromechanical modelling. Int.J. En-

gng.Sci., 31:1–10, 1993. doi:10.1016/0020-7225(93)90059-4.

E. Hervé and A. Zaoui. Elastic behavior of multiply coated fibre-reinforced composites.

Int.J.Engng.Sci., 33:1419–1433, 1995. doi:10.1016/0020-7225(95)00008-L.

E. Hervé, C. Stolz, and A. Zaoui. A propos de’l assemblage des sphères composites de
Hashin. C.R.Acad.Sci.Paris, série II, 313:857–862, 1991.

E. Hervé-Luanco and S. Joannès. Multiscale modelling of transport phenomena for materi-

als with n-layered embedded fibres. Part I: Analytical and numerical-based approaches.
Int.J.Sol.Struct., 97–98:625–636, 2016. doi:10.1016/j.ijsolstr.2016.05.015.

P.A. Hessman, F. Welschinger, K. Hornberger, and T. Böhlke. On mean field homogeniza-

tion schemes for short fiber reinforced composites: Unified formulation, application and
benchmark. Int.J.Sol.Struct., 230–231:111141, 2021. doi:10.1016/j.ijsolstr.2021.111141.

R. Hill. A theory of the yielding and plastic flow of anisotropic metals.

Proc.Phys.Soc.London, 193:281–297, 1948. doi:10.1098/rspa.1948.0045.

R. Hill. The elastic behavior of a crystalline aggregate. Proc.Phys.Soc.London, A65:349–

354, 1952. doi:10.1088/0370-1298/65/5/307.

R. Hill. Elastic properties of reinforced solids: Some theoretical principles. J.Mech.Phys.

Sol., 11:357–372, 1963. doi:10.1016/0022-5096(63)90036-X.

R. Hill. Theory of mechanical properties of fibre-strengthened materials: I. elastic be-

haviour. J.Mech.Phys.Sol., 12:199–212, 1964. doi:10.1016/0022-5096(64)90019-5.

R. Hill. Continuum micro-mechanics of elastoplastic polycrystals. J.Mech.Phys.Sol., 13:

89–101, 1965a. doi:10.1016/0022-5096(65)90023-2.

R. Hill. A self-consistent mechanics of composite materials. J.Mech.Phys.Sol., 13:213–222,

1965b. doi:10.1016/0022-5096(65)90010-4.

184
R. Hill. The essential structure of constitutive laws for metal composites and polycrystals.
J. Mech. Phys. Sol., 15:79–95, 1967. doi:10.1016/0022-5096(67)90018-X.

R. Hill. On constitutive macro-variables for heterogeneous solids at finite strain. Proc.

Roy.Soc.London, A326:131–147, 1972. doi:10.1098/rspa.1972.0001.

S. Hoffmann. Computational Homogenization of Short Fiber Reinforced Thermoplastic

Materials. PhD thesis, Universität Kaiserslautern, Kaiserslautern, 2012.

S.J. Hollister, D.P. Fyhrie, K.J. Jepsen, and S.A. Goldstein. Application of homogeniza-
tion theory to the study of trabecular bone mechanics. J.Biomech., 24:825–839, 1991.
doi:10.1016/0021-9290(91)90308-A.

S.J. Hollister, J.M. Brennan, and N. Kikuchi. A homogenization sampling procedure for
calculating trabecular bone effective stiffness and tissue level stress. J.Biomech., 27:
433–444, 1994. doi:10.1016/0021-9290(94)90019-1.

S. Holmberg, K. Persson, and H. Petersson. Nonlinear mechanical behaviour

and analysis of wood and fibre materials. Comput.Struct., 72:459–480, 1999.
doi:10.1016/S0045-7949(98)00331-9.

C.L. Hom and R.M. McMeeking. Plastic flow in ductile materials containing a cubic array
of rigid spheres. Int.J.Plast., 7:255–274, 1991. doi:10.1016/0749-6419(91)90035-W.

M. Hori and S. Nemat-Nasser. Double-inclusion model and overall moduli of multi-phase

composites. Mech.Mater., 14:189–206, 1993. doi:10.1016/0167-6636(93)90066-Z.

H. Horii and S. Nemat-Nasser. Overall moduli of solids with microcracks: Load induced
anisotropy. Int.J.Sol.Struct., 21:731–745, 1985. doi:10.1016/0022-5096(83)90048-0.

G.K. Hu. Composite plasticity based on matrix average second order stress moment.
Int.J.Sol.Struct., 34:1007–1015, 1997. doi:10.1016/S0020-7683(96)00044-3.

G.K. Hu and G.J. Weng. The connections between the double-inclusion model and the
Ponte Castañeda–Willis, Mori–Tanaka, and Kuster–Toksoz models. Mech.Mater., 32:
495–503, 2000. doi:10.1016/S0167-6636(00)00015-6.

G.K. Hu, G. Guo, and D. Baptiste. A micromechanical model of influence of particle

fracture and particle cluster on mechanical properties of metal matrix composites. Com-
put.Mater.Sci., 9:420–430, 1998. doi:10.1016/S0927-0256(97)00166-3.

G.K. Hu, X.N. Liu, and T.J. Lu. A variational method for nonlinear micropolar composites.
Mech.Mater., 37:407–425, 2005. doi:10.1016/j.mechmat.2004.03.006.

J.H. Huang. Some closed-form solutions for effective moduli of composites con-
taining randomly oriented short fibers. Mater.Sci.Engng. A, 315:11–20, 2001.
doi:10.1016/S0921-5093(01)01212-6.

J.H. Huang and J.S. Yu. Electroelastic Eshelby tensors for an ellipsoidal piezoelectric
inclusion. Compos.Engng., 4:1169–1182, 1994. doi:10.1016/0961-9526(95)91290-W.

185
Y. Huang and K.X. Hu. A generalized self-consistent mechanics method for solids contain-
ing elliptical inclusions. J.Appl.Mech., 62:566–572, 1995. doi:10.1115/1.2895982.

Z.M. Huang. A unified micromechanical model for the mechanical properties of two con-
stituent composite materials Part I: Elastic behavior. J.Thermoplast.Compos.Mater.,
13:252–271, 2000. doi:10.1177/089270570001300401.

Z.M. Huang and S. Ramakrishna. Micromechanical modeling approaches for the stiffness
and strength of knitted fabric composites: A review and comparative study. Composites
A, 31:479–501, 2000. doi:10.1016/S1359-835X(99)00083-4.

C.O. Huber, M.H. Luxner, S. Kremmer, S. Nogales, H.J. Böhm, and H.E. Pettermann.
Forming simulations of MMC components by a micromechanics based hierarchical
FEM approach. In J.M.A.C. de Sá and A.D. Santos, editors, Proceedings of NUMI-
FORM 2007, pages 1351–1356, New York, NY, 2007. American Institute of Physics.
doi:10.1063/1.2740997.

C. Huet. Application of variational concepts to size effects in elastic heterogeneous bodies.

J.Mech.Phys.Sol., 38:813–841, 1990. doi:10.1016/0022-5096(90)90041-2.

C. Huet, P. Navi, and P.E. Roelfstra. A homogenization technique based on Hill’s mod-
ification theorem. In G.A. Maugin, editor, Continuum Models and Discrete Systems,
pages 135–143, Harlow, 1990. Longman.

M.I. Hussein, M.J. Leamy, and M. Ruzzene. Dynamics of phononic materials and struc-
tures: Historical origins, recent progress, and future outlook. Appl.Mech.Rev., 66:040802,
2014. doi:10.1115/1.4026911.

J.W. Hutchinson. Plasticity on the micron scale. Int.J.Sol.Struct., 37:225–238, 2000.

doi:10.1016/S0020-7683(99)00090-6.

J.W. Hutchinson. Elastic-plastic behavior of polycrystalline metals and composites. Proc.

Roy.Soc.London, A319:247–272, 1970. doi:10.1098/rspa.1970.0177.

A. Ibrahimbegović and D. Markovič. Strong coupling methods in multi-phase and

multi-scale modeling of inelastic behavior of heterogeneous structures. Com-
put.Meth.Appl.Mech.Engng., 192:3089–3107, 2003. doi:10.1016/S0045-7825(03)00342-6.

L. Iorga, Y. Pan, and A.A. Pelegri. Numerical characterization of material elastic

properties for random fiber composites. J.Mech.Mater.Struct., 3:1279–1298, 2008.
doi:10.2140/jomms.2008.3.1279.

N. Ishikawa, D. Parks, S. Socrate, and M. Kurihara. Micromechanical modeling of ferrite–

pearlite steels using finite element unit cell models. ISIJ Int., 40:1170–1179, 2000.
doi:10.2355/isijinternational.40.1170.

Y. Ismail, D.M. Yang, and J.Q. Ye. Discrete element method for generating random
fibre distributions in micromechanical models for fibre reinforced composite laminates.
Composites B, 90:485–492, 2016. doi:10.1016/j.compositesb.2016.01.037.

186
H. Ismar and F. Schröter. Three-dimensional finite element analysis of the mechan-
ical behavior of cross ply-reinforced aluminum. Mech.Mater., 32:329–338, 2000.
doi:10.1016/S0167-6636(00)00008-9.
T. Iung and M. Grange. Mechanical behavior of two-phase materials investigated by the
finite element method: Necessity of three-dimensional modeling. Mater.Sci.Engng. A,
201:L8–L11, 1995. doi:10.1016/0921-5093(95)09891-7.
D.S. Ivanov and S.V. Lomov. Modeling of 2D and 3D woven composites. In P. Irving
and C. Soutis, editors, Polymer Composites in the Aerospace Industry, pages 23–57.
Woodhead Publishing, Duxford, UK, 2020. doi:10.1016/B978-0-08-102679-3.00002-2.
S.K. Iyer, C.J. Lissenden, and S.M. Arnold. Local and overall flow in
composites predicted by micromechanics. Composites B, 31:327–343, 2000.
doi:10.1016/S1359-8368(00)00011-1.
N. Järvstråt. Homogenization and the mechanical behavior of metal composites. In H. Fu-
jiwara, T. Abe, and K. Tanaka, editors, Residual Stresses — III, pages 70–75, London,
1992. Elsevier.
N. Järvstråt. An ellipsoidal unit cell for the calculation of micro-stresses in short fibre
composites. Comput.Mater.Sci., 1:203–212, 1993. doi:10.1016/0927-0256(93)90012-C.
K. Jayaraman and M.T. Kortschot. Correction to the Fukuda–Kawata Young’s modu-
lus theory and the Fukuda–Chou strength theory for short fibre-reinforced composite
materials. J.Mater.Sci., 31:2059–2064, 1996. doi:10.1007/BF00356627.
D. Jeulin. Random structure models for homogenization and fracture statistics. In D. Jeulin
and M. Ostoja-Starzewski, editors, Mechanics of Random and Multiscale Microstruc-
tures, pages 33–91, Vienna, 2001. Springer-Verlag, CISM Courses and Lectures Vol. 430.
doi:10.1007/978-3-7091-2780-3 2.
M. Jiang, M. Ostoja-Starzewski, and I. Jasiuk. Scale-dependent bounds on effective
elastoplastic response of random composites. J.Mech.Phys.Sol., 49:655–673, 2001.
doi:10.1016/S0022-5096(00)00034-X.
M. Jiang, I. Jasiuk, and M. Ostoja-Starzewski. Apparent thermal conductivity
of periodic two-dimensional composites. Comput.Mater.Sci., 25:329–338, 2002a.
doi:10.1016/S0927-0256(02)00234-3.
W.G. Jiang, S.R. Hallett, and M.R. Wisnom. Development of domain superposition tech-
nique for the modelling of woven fabric composites. In P.P. Camanho, C.G. Dávila,
S.T. Pinho, and J.J.C. Remmers, editors, Mechanical Response of Composites, pages
281–291, Berlin, 2008. Springer-Verlag. doi:10.1007/978-1-4020-8584-0 14.
Y. Jiao, F.H. Stillinger, and S. Torquato. Modeling heterogeneous materials via two-point
correlation functions: Basic principles. Phys.Rev. E, 76:031110, 2007. doi:10.1103/Phys-
RevE.76.031110.
N. Jiménez Segura, B.L.A. Pichler, and C. Hellmich. Concentration tensors preserv-
ing elastic symmetry of multiphase composites. Mech.Mater., 178:104555, 2023.
doi:10.1016/j.mechmat.2023.104555.

187
X. Jin, D. Lyu, X. Zhang, Q. Zhou, Q. Wang, and L.M. Keer. Explicit analytical solutions
for a complete set of the Eshelby tensors of an ellipsoidal inclusion. J.Appl.Mech., 83:
121010, 2016. doi:10.1115/1.4034705.

M. Jirásek and S. Rolshoven. Comparison of integral-type nonlocal plastic-

ity models for strain softening materials. Int.J.Engng.Sci., 41:1553–1602, 2003.
doi:10.1016/S0020-7225(03)00027-2.

B. Johannesson and O.B. Pedersen. Analytical determination of the average Eshelby tensor
for transversely isotropic fiber orientation distributions. Acta mater., 46:3165–3173, 1998.
doi:10.1016/S1359-6454(98)00003-2.

R.M. Jones. Mechanics of Composite Materials. Taylor & Francis, Philadelphia, PA, 1999.
ISBN 1-560-32712-X.

J.W. Ju and L.Z. Sun. Effective elastoplastic behavior of metal matrix composites contain-
ing randomly located aligned spheroidal inhomogeneities. Part I: Micromechanics-based
formulation. Int.J.Sol.Struct., 38:183–201, 2001. doi:10.1016/S0020-7683(00)00023-8.

J.W. Ju and L.Z. Sun. A novel formulation for the exterior point Eshelby’s tensor of an
ellipsoidal inclusion. J.Appl.Mech., 66:570–574, 1999. doi:10.1115/1.2791090.

M. Kabel, A. Fink, and M. Schneider. The composite voxel technique for inelastic problems.
Comput.Meth.Appl.Mech.Engng., 322:396–418, 2017. doi:10.1016/j.cma.2017.04.025.

M. Kachanov and I. Sevostianov. On quantitative characterization of mi-

crostructures and effective properties. Int.J.Sol.Struct., 42:309–336, 2005.
doi:10.1016/j.ijsolstr.2004.06.016.

M. Kachanov and I. Sevostianov. Effective Properties of Heterogeneous Materials. Springer-

Verlag, New York, NY, 2013. ISBN 978-94-007-5714-1.

M. Kachanov and I. Sevostianov. Micromechanics of Materials, with Applications. Springer

International. Cham, 2018. ISBN 978-3-319-76203-6.

M. Kachanov, I. Tsukrov, and B. Shafiro. Effective moduli of solids with cavities of various
shapes. Appl.Mech.Rev., 47:151–S174, 1994. doi:10.1115/1.3122810.

M. Kailasam, N. Aravas, and P. Ponte Castañeda. Porous metals with developing

anisotropy: Constitutive models, computational issues and applications to deformation
processing. Comput.Model.Engng.Sci., 1:105–118, 2000. doi:10.3970/cmes.2000.001.265.

A.L. Kalamkarov, I.V. Andrianov, and V.V. Danishevs’kyy. Asymptotic homoge-

nization of composite materials and structures. Appl.Mech.Rev., 62:030802, 2009.
doi:10.1115/1.3090830.

S. Kammoun, I. Doghri, L. Adam, G. Robert, and L. Delannay. First pseudo-grain failure

model for inelastic composites with misaligned short fibers. Composites A, 42:1892–1902,
2011. doi:10.1016/j.compositesa.2011.08.013.

188
S.K. Kanaun. Efficient homogenization techniques for elastic composites:
Maxwell scheme vs. effective field method. Int.J.Engng.Sci., 103:19–34, 2016.
doi:10.1016/j.ijengsci.2016.03.004.

S.K. Kanaun and V.M. Levin. Effective field method in mechanics of matrix
composite materials. In K.Z. Markov, editor, Recent Advances in Mathematical
Modelling of Composite Materials, pages 1–58, Singapore, 1994. World Scientific.
doi:10.1142/9789814354219 0001.

H. Kang and G.W. Milton. Solutions to the Pólya–Szegö conjecture and the weak Eshelby
conjecture. Arch.Rat.Mech.Anal., 188:93–116, 2008. doi:10.1007/s00205-007-0087-z.

T. Kanit, S. Forest, I. Gallier, V. Mounoury, and D. Jeulin. Determination of the size

of the representative volume element for random composites: Statistical and numerical
approach. Int.J.Sol.Struct., 40:3647–3679, 2003. doi:10.1016/S0020-7683(03)00143-4.

T. Kanit, F. N’Guyen, S. Forest, D. Jeulin, M. Reed, and S. Singleton. Apparent and

effective physical properties of heterogeneous materials: Representativity of samples
of two materials from food industry. Comput.Meth.Appl.Mech.Engng., 195:3960–3982,
2006. doi:10.1016/j.cma.2005.07.022.

P. Kanouté, D.P. Boso, J.L. Chaboche, and B.A. Schrefler. Multiscale meth-
ods for composites: A review. Arch.Comput.Meth.Engng., 16:31–75, 2009.
doi:10.1007/s11831-008-9028-8.

P. Karimi, A. Malyarenko, M. Ostoja-Starzewski, and X. Zhang. RVE problem: Mathe-

matical aspects and related stochastic mechanics. Int.J.Engng.Sci., 146:103169, 2020.
doi:10.1016/j.ijengsci.2019.103169.

P. Kenesei, A. Borbély, and H. Biermann. Microstructure based three-dimensional finite el-

ement modeling of particulate reinforced metal matrix composites. Mater.Sci. Engng. A,
387:852–856, 2004. doi:10.1016/j.msea.2004.02.076.

E.H. Kerner. The elastic and thermo-elastic properties of composite media.

Proc.Phys.Soc.London, B69:808–813, 1956. doi:10.1088/0370-1301/69/8/305.

Y.M Khalevitsky and A.V. Konovalov. A gravitational approach to modeling the rep-
resentative volume geometry of particle reinforced metal matrix composites. En-
gng.w.Comput., 35:1037–1044, 2019. doi:10.1007/s00366-018-0649-8.

Z.F. Khisaeva and M. Ostoja-Starzewski. On the size of RVE in finite elasticity of random
composites. J.Elast., 85:153–173, 2006. doi:10.1007/s10659-006-9076-y.

D.W. Kim, J.H. Lim, and S. Lee. Prediction and validation of the transverse
mechanical behavior of unidirectional composites considering interfacial debond-
ing through convolutional neural networks. Composites B, 225:109314, 2021.
doi:10.1016/j.compositesb.2021.109314.

K. Kitazono, E. Sato, and K. Kuribayashi. Application of mean-field approximation to

elastic-plastic behavior for closed cell metal foams. Acta mater., 51:4823–4836, 2003.
doi:10.1016/S1359-6454(03)00322-7.

189
J. Koplik and A. Needleman. Void growth and coalescence in porous plastic solids.
Int.J.Sol.Struct., 24:835–853, 1988. doi:10.1016/0020-7683(88)90051-0.
V.G. Kouznetsova, M.G.D. Geers, and W.A.M. Brekelmans. Multi-scale constitutive mod-
eling of heterogeneous materials with a gradient-enhanced computational homogeniza-
tion scheme. Int.J.Num.Meth.Engng., 54:1235–1260, 2002. doi:10.1002/nme.541.
V.G. Kouznetsova, M.G.D. Geers, and W.A.M. Brekelmans. Multi-scale second-
order computational homogenization of multi-phase materials: A nested finite el-
ement solution strategy. Comput.Meth.Appl.Mech.Engng., 193:5525–5550, 2004.
doi:10.1016/j.cma.2003.12.073.
K. Krabbenhøft, M. Hain, and P. Wriggers. Computation of effective cement paste diffu-
sivities from microtomographic images. In V.V. Kompiš, editor, Composites with Micro-
and Nano-Structure: Computational Modeling and Experiments, pages 281–297, Berlin,
2008. Springer-Verlag. doi:10.1007/978-1-4020-6975-8 15.
I. Krakovsky and V. Myroshnychenko. Modelling dielectric properties of composites by
finite element method. J.Appl.Phys., 92:6743–6748, 2002. doi:10.1063/1.1516837.
A.M. Kraynik and D.A. Reinelt. Elastic-plastic behavior of a Kelvin foam. In D. Weaire,
editor, The Kelvin Problem. Foam Structures of Minimal Surface Area, pages 93–108,
London, 1996. Taylor & Francis.
E. Kröner. Self-consistent scheme and graded disorder in polycrystal elasticity. J.Phys.F,
8:2261–2267, 1978. doi:10.1088/0305-4608/8/11/011.
V.I. Kushch. Micromechanics of Composites: Multipole Expansion Approach. Butterworth–
Heinemann, Kidlington, UK, 2013. ISBN 978-0-12-407683-9.
V.I. Kushch, I. Sevostianov, and L. Mishnaevsky. Stress concentration and the ef-
fective stiffness of aligned fiber reinforced composite with anisotropic constituents.
Int.J.Sol.Struct., 45:5103–5117, 2008. doi:10.1016/j.ijsolstr.2008.05.009.
G.T. Kuster and M.N. Toksöz. Velocity and attenuation of seismic waves in two-phase
media: I. Theoretical formulation. Geophysics, 39:587–606, 1974. doi:10.1190/1.1440450.
N. Lahellec and P. Suquet. Effective behavior of linear viscoelastic com-
posites: A time-integration approach. Int.J.Sol.Struct., 44:507–529, 2007.
doi:10.1016/j.ijsolstr.2006.04.038.
N. Lahellec and P. Suquet. Effective response and field statistics in elastoplastic and elasto-
viscoplastic composites under radial and non-radial loadings. Int.J.Plast., 42:1–30, 2013.
doi:10.1016/j.ijplas.2012.09.005.
F. Larsson, K. Runesson, S. Saroukhani, and R. Vafadari. Computational homog-
enization based on a weak format of microperiodicity for RVE-problems. Com-
put.Meth.Appl.Sci.Engng., 200:11–26, 2011. doi:10.1016/j.cma.2010.06.023.
E.A. Lazar, J.K. Mason, R.D. McPherson, and D.J. Srolovitz. Complete topology of cells,
grains, and bubbles in three-dimensional microstructures. Phys.Rev.Lett., 109:095505,
2012. doi:10.1103/PhysRevLett.109.095505.

190
H.K. Lee and S. Simunovic. Modeling of progressive damage in aligned and randomly
oriented discontinuous fiber polymer matrix composites. Composites B, 31:77–86, 2000.
doi:10.1016/S1359-8368(99)00070-0.

H.W. Lee, A. Gillman, and K. Matouš. Computing overall elastic constants of polydisperse
particular composites from microtomographic data. J.Mech.Phys.Sol., 59:1838–1857,
2011a. doi:10.1016/j.jmps.2011.05.010.

K.H. Lee and S. Ghosh. Small deformation multi-scale analysis of heterogeneous ma-
terials with the Voronoi cell finite element model and homogenization theory. Com-
put.Mater.Sci., 7:131–146, 1996. doi:10.1016/S0927-0256(96)00072-9.

W.J. Lee, Y.J. Kim, N.H. Kang, I.M. Park, and Y.H. Park. Finite-element modeling of the
particle clustering effect in a powder-metallurgy-processed ceramic-particle-reinforced
metal matrix composite on its mechanical properties. Mech.Compos.Mater., 46:639–648,
2011b. doi:110.1007/s11029-011-9177-y.

G. Legrain, P. Cartraud, I. Perreard, and N. Moës. An X-FEM and level set com-
putational approach for image based modelling: Application to homogenization.
Int.J.Numer.Meth.Engng., 86:915–934, 2011. doi:10.1002/nme.3085.

B.A. Lerch, M.E. Melis, and M. Tong. Experimental and analytical analysis of the
stress–strain behavior in a [90/0]2S SiC/Ti-15-3 laminate. Technical Report NASA–
TM–104470, NASA, 1991.

M. Lévesque, M.D. Gilchrist, N. Bouleau, K. Derrien, and D. Baptiste. Numerical inversion

of the Laplace–Carson transform applied to homogenization of randomly reinforced linear
viscoelastic media. Comput.Mech., 10:771–789, 2007. doi:10.1007/s00466-006-0138-6.

V.M. Levin. On the coefficients of thermal expansion of heterogeneous materials. Mech.Sol.,

2:58–61, 1967.

A. Levy and J.M. Papazian. Elastoplastic finite element analysis of short-fiber-reinforced

SiC/Al composites: Effects of thermal treatment. Acta metall.mater., 39:2255–2266,
1991. doi:10.1016/0956-7151(91)90008-O.

H. Li, B. Zhang, and G. Bai. Effect of constructing different unit cells on

predicting composite viscoelastic properties. Compos.Struct., 125:459–466, 2015.
doi:10.1016/j.compstruct.2015.02.028.

M. Li, S. Ghosh, O. Richmond, H. Weiland, and T.N. Rouns. Three dimensional charac-
terization and modeling of particle reinforced metal matrix composites, Part I: Quan-
titative description of microstructural morphology. Mater.Sci.Engng. A, 265:153–173,
1999. doi:10.1016/S0921-5093(98)01132-0.

S. Li and Z. Zou. The use of central reflection in the formulation of unit cells for microme-
chanical FEA. Mech.Mater., 43:824–834, 2011. doi:10.1016/j.mechmat.2011.08.014.

S.G. Li. On the nature of periodic traction boundary conditions in micromechanical

FE analyses of unit cells. SIAM J.Appl.Math., 77:441–450, 2012. doi:10.1093/ima-
mat/hxr024.

191
S.G. Li and A. Wongsto. Unit cells for micromechanical analyses of particle-reinforced
composites. Mech.Mater., 36:543–572, 2004. doi:10.1016/S0167-6636(03)00062-0.

W. Li and M. Ostoja-Starzewski. Yield of random elastoplastic materials.

J.Mech.Mater.Struct., 1:1055–1073, 2006. doi:10.2140/jomms.2006.1.1055.

Y.Y. Li and J.Z. Cui. The multi-scale computational method for the mechanics parameters
of the materials with random distribution of multi-scale grains. Compos.Sci.Technol.,
65:1447–1458, 2005. doi:10.1016/j.compscitech.2004.12.016.

G. Lielens, P. Pirotte, A. Couniot, F. Dupret, and R. Keunings. Prediction of thermo-

mechanical properties for compression-moulded composites. Composites A, 29:63–70,
1998. doi:10.1016/S1359-835X(97)00039-0.

D.S. Liu, C.Y. Chen, and D.Y. Chiou. 3-D modeling of a composite material rein-
forced with multiple thickly coated particles using the infinite element method. Com-
put.Model.Engng.Sci., 9:179–192, 2005a. doi:10.3970/cmes.2005.009.179.

L. Liu, R.D. James, and P.H. Leo. Periodic inclusion–matrix microstruc-

tures with constant field inclusions. Metall.Mater.Trans., 38A:781–787, 2007.
doi:10.1007/s11661-006-9019-z.

Y.J. Liu, N. Nishimura, Y. Otani, T. Takahashi, X.L. Chen, and H. Munakata. A fast
boundary element method for the analysis of fiber-reinforced composites based on rigid-
inclusion model. J.Appl.Mech., 72:115–128, 2005b. doi:10.1115/1.1825436.

J. LLorca. A numerical analysis of the damage mechanisms in metal-matrix

composites under cyclic deformation. Comput.Mater.Sci., 7:118–122, 1996.
doi:10.1016/S0927-0256(96)00070-5.

M. Lo Cascio, A. Milazzo, and I. Benedetti. Virtual element method for computational

homogenization of composite and heterogeneous materials. Comput.Struct., 232:111523,
2020. doi:10.1016/j.compstruct.2019.111523.

S.V. Lomov, D.S. Ivanov, I. Verpoest, M. Zeko, T. Kurashiki, H. Nakai, and S. Hiro-
sawa. Meso-FE modelling of textile composites: Road map, data flow and algorithms.
Compos.Sci.Technol., 67:1870–1891, 2006. doi:10.1016/j.compscitech.2006.10.017.

B.D. Lubachevsky, F.H. Stillinger, and E.N. Pinson. Disks vs. spheres: Contrasting prop-
erties of random packing. J.Statist.Phys., 64:501–524, 1991. doi:10.1007/BF01048304.

V.A. Lubarda and X. Markenscoff. On the absence of Eshelby property for non-ellipsoidal
inclusions. Int.J.Sol.Struct., 35:3405–3411, 1998. doi:10.1016/S0020-7683(98)00025-0.

H.R. Lusti, P.J. Hine, and A.A. Gusev. Direct numerical predictions for the elastic and
thermoelastic properties of short fibre composites. Compos.Sci.Technol., 62:1927–1934,
2002. doi:10.1016/S0266-3538(02)00106-9.

M.H. Luxner, J. Stampfl, and H.E. Pettermann. Finite element modeling concepts and
linear analyses of 3D regular open cell structures. J.Mater.Sci., 40:5859–5866, 2005.
doi:10.1007/s10853-005-5020-y.

192
F. Maggi, S. Stafford, T.L. Jackson, and J. Buckmaster. Nature of packs used in propellant
modeling. Phys.Rev. E, 77:046107, 2008. doi:10.1103/PhysRevE.77.046107.
J.M. Mahishi. An integrated micromechanical and macromechanical approach to frac-
ture behavior of fiber-reinforced composites. Engng.Fract.Mech., 25:197–228, 1986.
doi:10.1016/0013-7944(86)90218-3.
E. Maire, F. Wattebled, J.Y. Buffière, and G. Peix. Deformation of a metallic foam
studied by X-ray computed tomography and finite element calculations. In T.W. Clyne
and F. Simancik, editors, Metal Matrix Composites and Metallic Foams, pages 68–73,
Weinheim, 2000. Wiley–VCH. doi:10.1002/3527606165.ch11.
M. Majewski, M. Kursa, P. Holobut, and K. Kowalczyk-Gajewska. Micromechanical and
numerical analysis of packing and size effects in elastic particulate composites. Compos-
ites, 124B:158–174, 2017. doi:10.1016/j.compositesb.2017.05.004.
M. Majewski, M. Wichrowski, P. Holobut, and K. Kowalczyk-Gajewska. Shape and packing
effects in particulate composites: Micromechanical modelling and numerical verification.
Arch.Civ.Mech.Engng., 22:86, 2022. doi:10.1007/s43452-022-00405-9.
P. Makowski, W. John, G. Kuś, and G. Kokot. Multiscale modeling of the simplified tra-
becular bone structure. In Mechanika: Proceedings of the 18th International Conference,
pages 156–161, Kaunas, 2013.
J. Mandel. Généralisation de la théorie de la plasticité de W.T. Koiter. Int.J.Sol.Struct.,
1:273–295, 1965. doi:10.1016/0020-7683(65)90034-X.
J. Mandel. Plasticité classique et viscoplasticité. CISM lecture notes, vol. 97, Springer–
Verlag, Wien, 1972.
C. Mareau and C. Robert. Different composite voxel methods for the numerical homoge-
nization of heterogeneous inelastic materials with FFT-based techniques. Mech.Mater.,
105:157–165, 2017. doi:10.1016/j.mechmat.2016.12.002.
S. Marfia and E. Sacco. Multiscale technique for nonlinear analysis of
elastoplastic and viscoplastic composites. Composites B, 136:241–253, 2018.
doi:10.1016/j.compositesb.2017.10.015.
F. Marketz and F.D. Fischer. Micromechanical modelling of stress-assisted
martensitic transformation. Modell.Simul.Mater.Sci.Engng., 2:1017–1046, 1994.
doi:10.1088/0965-0393/2/5/006.
K. Markov. Elementary micromechanics of heterogeneous media. In K. Markov and
L. Preziosi, editors, Heterogeneous Media: Micromechanics Modeling Methods and Sim-
ulations, pages 1–162, Boston, 2000. Birkhäuser. doi:10.1007/978-1-4612-1332-1 1.
R. Masson. New explicit expressions of the Hill polarization tensor for general anisotropic
elastic solids. Int.J.Sol.Struct., 45:757–769, 2008. doi:10.1016/j.ijsolstr.2007.08.035.
R. Masson, M. Bornert, P. Suquet, and A. Zaoui. An affine formulation for the prediction
of the effective properties of nonlinear composites and polycrystals. J.Mech.Phys.Sol.,
48:1203–1227, 2000. doi:10.1016/S0022-5096(99)00071-X.

193
C.F. Matt and M.A.E. Cruz. Application of a multiscale finite-element approach to cal-
culate the effective thermal conductivity of particulate media. Comput.Appl.Math., 21:
429–460, 2002.

C.F. Matt and M.A.E. Cruz. Effective thermal conductivity of composite materials with 3-
D microstructures and interfacial thermal resistance. Numer.Heat Transf., A53:577–604,
2008. doi:10.1080/10407780701678380.

A.Y. Matveeva, G. Kravchenko, H.J. Böhm, and F.W.J. van Hattum. Investigation of the
embedded element technique for modelling wavy CNT composites. Comput.Mater.Cont.,
42:1–23, 2014. doi:10.3970/cmc.2014.042.001.

J.C. Maxwell. A Treatise on Electricity and Magnetism. Clarendon Press, Oxford, 1873.
doi:10.1017/CBO9780511709333.

D.L. McDowell. Modeling and experiments in plasticity. Int.J.Sol.Struct., 37:293–310,

2000. doi:10.1016/S0020-7683(99)00094-3.

P.E. McHugh, R.J. Asaro, and C.F. Shih. Computational modeling of metal-matrix com-
posite materials — I. Isothermal deformation patterns in ideal microstructures. Acta
metall.mater., 41:1461–1476, 1993. doi:10.1016/0956-7151(93)90255-Q.

R. McLaughlin. A study of the differential scheme for composite materials. Int.J.Engng.

Sci., 15:237–244, 1977. doi:10.1016/0020-7225(77)90058-1.

R. McMeeking and C.L. Hom. Finite element analysis of void growth in elastic-plastic
materials. Int.J.Fract., 42:1–19, 1990. doi:10.1007/BF00018610.

C. Meng, W. Heltsley, and D.D. Pollard. Evaluation of the Eshelby solution

for the ellipsoidal inclusion and heterogeneity. Comput.Geosci, 40:40–48, 2012.
doi:10.1016/j.cageo.2011.07.008.

S.D. Mesarovic and J. Padbidri. Minimal kinetic boundary conditions for simulations of dis-
ordered microstructures. Phil.Mag., 85:65–78, 2005. doi:10.1080/14786430412331313321.

J.C. Michel and P. Suquet. Computational analysis of nonlinear composite structures

using the nonuniform transformation field analysis. Comput.Meth.Appl.Mech.Engng.,
193:5477–5502, 2004. doi:10.1016/j.cma.2003.12.071.

J.C. Michel, H. Moulinec, and P. Suquet. Effective properties of composite materials with
periodic microstructure: A computational approach. Comput.Meth.Appl.Mech.Engng.,
172:109–143, 1999. doi:10.1016/S0045-7825(98)00227-8.

J.C. Michel, H. Moulinec, and P. Suquet. Composites à microstructure périodique. In

M. Bornert, T. Bretheau, and P. Gilormini, editors, Homogénéisation en mécanique des
matériaux 1. Matériaux aléatoires élastiques et milieux périodiques, pages 57–94, Paris,
2001. Editions Hermès.

N.J. Mills, R. Stämpfli, F. Marone, and P.A. Brühwiler. Finite element micromechanics
model of impact compression of closed-cell polymeric foams. Int.J.Sol.Struct., 46:677–
697, 2009. doi:10.1016/j.ijsolstr.2008.09.012.

194
T. Miloh and Y. Benveniste. A generalized self-consistent method for the effective con-
ductivity of composites with ellipsoidal inclusions and cracked bodies. J.Appl.Phys., 63:
789–796, 1988. doi:10.1063/1.340071.

G.W. Milton. The Theory of Composites. Cambridge University Press, Cambridge, 2002.
ISBN 0-521-88125-6.

G.W. Milton. Bounds on the electromagnetic, elastic, and other properties of

two-component composites. Phys.Rev.Lett., 46:542–545, 1981. doi:10.1103/Phys-
RevLett.46.542.

L.L. Mishnaevsky. Three-dimensional numerical testing of microstructures of particle rein-

forced composites. Acta mater., 52:4177–4188, 2004. doi:10.1016/j.actamat.2004.05.032.

D. Missoum-Benziane, D. Ryckelynck, and F. Chinesta. A new fully coupled two-scales

modelling for mechanical problems involving microstructure: The 95/5 technique. Com-
put.Meth.Appl.Mech.Engng., 196:2325–2337, 2007. doi:10.1016/j.cma.2006.10.013.

B. Mlekusch. Thermoelastic properties of short-fibre-reinforced thermoplastics. Compos.

Sci.Technol., 59:911–923, 1999. doi:10.1016/S0266-3538(98)00133-X.

M. Moakher and A.N. Norris. The closest elastic tensor of arbitrary symmetry to an elastic
tensor of lower symmetry. J.Elast., 85:215–263, 2006. doi:10.1007/s10659-006-9082-0.

N. Moës, M. Cloirec, P. Cartraud, and J.F. Remacle. A computational approach to handle

complex microstructure geometries. Comput.Meth.Appl.Mech.Engng., 192:3163–3177,
2003. doi:10.1016/S0045-7825(03)00346-3.

A. Molinari, G.R. Canova, and S. Ahzi. A self-consistent approach for large deformation
viscoplasticity. Acta metall., 35:2983–2984, 1987. doi:10.1016/0001-6160(87)90297-5.

J. Monaghan and D. Brazil. Modeling the sub-surface damage associated with

the machining of a particle reinforced MMC. Comput.Mater.Sci., 9:99–107, 1997.
doi:10.1016/S0927-0256(97)00063-3.

I. Monetto and W.J. Drugan. A micromechanics-based nonlocal constitutive equa-

tion and minimum RVE size estimates for random elastic composites contain-
ing aligned spheroidal heterogeneities. J.Mech.Phys.Sol., 57:1578–1595, 2009.
doi:10.1016/S0022-5096(03)00103-0.

T. Mori and K. Tanaka. Average stress in matrix and average elastic en-
ergy of materials with misfitting inclusions. Acta metall., 21:571–574, 1973.
doi:10.1016/0001-6160(73)90064-3.

Z.A. Moschovidis and T. Mura. Two ellipsoidal inhomogeneities by the equivalent inclusion
method. J.Appl.Mech., 42:847–852, 1975. doi:10.1115/1.3423718.

O.F. Mossotti. Discussione analitica sull’influenza che l’azione di un mezzo dielettrico ha

sulla distribuzione dell’elettricitá alla superficie di piú corpi elettrici disseminati in eso.
Mem.Mat.Fis.Soc.Ital.Sci.Modena, 24:49–74, 1850.

195
C. Motz, R. Pippan, A. Ableidinger, H.J. Böhm, and F.G. Rammerstorfer. Deformation
and fracture behavior of ductile aluminium foams in the presence of notches under tensile
loading. In J. Banhart, M.F. Ashby, and N.A. Fleck, editors, Cellular Metals and Metal
Foaming Technology, pages 299–304, Bremen, 2001. Verlag MIT Publishing.

H. Moulinec and P. Suquet. A fast numerical method for computing the linear and nonlinear
mechanical properties of composites. C.R.Acad.Sci.Paris, série II, 318:1417–1423, 1994.

H. Moulinec and P. Suquet. A numerical method for computing the overall response of
nonlinear composites with complex microstructure. Comput.Meth.Appl.Mech.Engng.,
157:69–94, 1997. doi:10.1016/S0045-7825(97)00218-1.

H. Moussaddy, D. Therriault, and M. Lévesque. Assessment of existing and introduc-

tion of a new and robust efficient definition of the representative volume element.
Int.J.Sol.Struct., 50:3817–3828, 2013. doi:10.1016/j.ijsolstr.2013.07.016.

W.H. Müller. Mathematical vs. experimental stress analysis of inhomogeneities in solids.

J.Phys.IV, 6:C1–139–C1–148, 1996. doi:10.1051/jp4:1996114.

T. Mura. Micromechanics of Defects in Solids. Martinus Nijhoff, Dordrecht, 1987. ISBN

9-024-73256-5.

T. Nakamura and S. Suresh. Effects of thermal residual stresses and fiber packing
on deformation of metal-matrix composites. Acta metall.mater., 41:1665–1681, 1993.
doi:10.1016/0956-7151(93)90186-V.

A. Needleman. A continuum model for void nucleation by inclusion debonding.

J.Appl.Mech., 54:525–531, 1987. doi:10.1115/1.3173064.

S. Nemat-Nasser. Averaging theorems in finite deformation plasticity. Mech.Mater., 31:

493–523, 1999. doi:10.1016/S0167-6636(98)00073-8.

S. Nemat-Nasser and M. Hori. Micromechanics: Overall Properties of Heterogeneous Solids.

North–Holland, Amsterdam, 1993. ISBN 0-444-89881-6.

V.D. Nguyen, E. Béchet, C. Geuzaine, and L. Noels. Imposing periodic boundary condi-
tions on arbitrary meshes by polynomial interpolation. Comput.Mater.Sci., 55:390–406,
2012. doi:10.1016/j.commatsci.2011.10.017.

V.D. Nguyen, L. Wu, and L. Noels. Unified treatment of microscopic boundary conditions
and efficient algorithms for estimating tangent operators of the homogenized behav-
ior in the computational homogenization method. Comput.Mech., 59:483–505, 2017.
doi:10.1007/s00466-016-1358-z.

V.P. Nguyen, M. Stroeven, and L.J. Sluys. Multiscale continuous and discontinuous mod-
eling of heterogeneous materials: A review on recent developments. J.Multiscal.Model.,
3:229–270, 2011. doi:10.1142/S1756973711000509.

G.L. Niebur, J.C. Yuen, A.C. Hsia, and T.M. Keaveny. Convergence behavior of high-
resolution finite element models of trabecular bone. J.Biomech.Engng., 121:629–635,
1999. doi:10.1115/1.2800865.

196
C.F. Niordson and V. Tvergaard. Nonlocal plasticity effects on the tensile prop-
erties of a metal matrix composite. Eur.J.Mech. A/Solids, 20:601–613, 2001.
doi:10.1016/S0997-7538(01)01149-4.

S. Nogales. Numerical Simulation of the Thermal and Thermomechanical Behavior of

Metal Matrix Composites. Reihe 18, Nr. 317. VDI–Verlag, Düsseldorf, 2008. ISBN
3-18-140118-8.

S. Nogales and H.J. Böhm. Modeling of the thermal conductivity and thermomechan-
ical behavior of diamond reinforced composites. Int.J.Engng.Sci., 46:606–619, 2008.
doi:10.1016/j.ijengsci.2008.01.011.

A.N. Norris. The isotropic material closest to a given anisotropic material. J.Mech.Mater.
Struct., 1:223–238, 2006. doi:10.2140/jomms.2006.1.223.

A.N. Norris. A differential scheme for the effective moduli of composites. Mech.Mater., 4:
1–16, 1985. doi:10.1016/0167-6636(85)90002-X.

D. Nouailhas and G. Cailletaud. Finite element analysis of the mechanical be-

havior of two-phase single crystal superalloys. Scr.mater., 34:565–571, 1996.
doi:10.1016/1359-6462(95)00547-1.

J. Novák, L. Kaczmarczyk, P. Grassl, J. Zeman, and C.J. Pearce. A micromechanics-

enhanced finite element formulation for modelling heterogeneous materials. Com-
put.Meth.Appl.Mech.Engng., 201–204:53–64, 2012. doi:10.1016/j.cma.2011.09.003.

J.F. Nye. Physical Properties of Crystals, Their Representation by Tensors and Matrices.
Clarendon, Oxford, 1957. ISBN 0-198-51165-5.

M. Nygårds. Number of grains necessary to homogenize elastic materials with cubic sym-
metry. Mech.Mater., 35:1049–1057, 2003. doi:10.1016/S0167-6636(02)00325-3.

R.J. O’Connell and B. Budiansky. Seismic velocities in dry and saturated cracked solids.
J.Geophys.Res., 79:5412–5426, 1974. doi:10.1029/JB079i035p05412.

J.T. Oden and T.I. Zohdi. Analysis and adaptive modeling of highly hetero-
geneous elastic structures. Comput.Meth.Appl.Mech.Engng., 148:367–391, 1997.
doi:10.1016/S0045-7825(97)00032-7.

S. Onaka. Averaged Eshelby tensor and elastic strain energy of a super-

spherical inclusion with uniform eigenstrains. Phil.Mag.Lett., 81:265–272, 2001.
doi:10.1080/09500830010019031.

P. Onck and E. van der Giessen. Microstructurally-based modelling of inter-

granular creep fracture using grain elements. Mech.Mater., 26:109–126, 1997.
doi:10.1016/S0167-6636(97)00020-3.

S.M. Oppenheimer and D.C. Dunand. Finite element modeling of creep deformation in
cellular materials. Acta mater., 55:3825–3834, 2007. doi:10.1016/j.actamat.2007.02.033.

M. Ostoja-Starzewski. Lattice models in micromechancis. Appl.Mech.Rev., 55:35–60, 2002.

doi:10.1115/1.1432990.

197
M. Ostoja-Starzewski. Material spatial randomness: From statistical to
representative volume element. Probab.Engng.Mech., 21:112–131, 2006.
doi:10.1016/j.probengmech.2005.07.007.

M. Ostoja-Starzewski. Microstructural Randomness and Scaling in Mechanics of Materials.

Chapman & Hall, Boca Raton, FL, 2008. ISBN 1-584-88417-7.

M. Ostoja-Starzewski, X. Du, Z.F. Khisaeva, and W. Li. Comparisons of the size of the
representative volume element in elastic, plastic, thermoelastic, and permeable random
microstructures. Int.J.Multiscal.Comput.Engng., 5:73–82, 2007. doi:10.1615/IntJMult-
CompEng.v5.i2.10.

T. Ozturk, C. Stein, R. Pokharel, C. Hefferan, H. Tucker, S. Jha, R. John, R.A. Lebenssohn,

P. Kenesei, and R.M. Suter. Simulation domain size requirements for elastic re-
sponse of 3D polycrystalline materials. Modell.Simul.Mater.Sci.Engng., 24:015006, 2016.
doi:10.1088/0965-0393/24/1/015006.

D.H. Pahr. Experimental and Numerical Investigations of Perforated FRP-Laminates.

Reihe 18, Nr. 284. VDI–Verlag, Düsseldorf, 2003. ISBN 3-183-28418-9.

D.H. Pahr and S.M. Arnold. The applicability of the generalized method of cells for
analyzing discontinuously reinforced composites. Composites B, 33:153–170, 2002.
doi:10.1088/0965-0393/24/1/015006.

D.H. Pahr and H.J. Böhm. Assessment of mixed uniform boundary conditions for predicting
the mechanical behavior of elastic and inelastic discontinuously reinforced composites.
Comput.Model.Engng.Sci., 34:117–136, 2008. doi:10.3970/cmes.2008.034.117.

D.H. Pahr and F.G. Rammerstorfer. Buckling of honeycomb sandwiches: Peri-

odic finite element considerations. Comput.Model.Engng.Sci., 12:229–242, 2006.
doi:10.3970/cmes.2006.012.229.

D.H. Pahr and P.K. Zysset. Influence of boundary conditions on computed apparent
elastic properties of cancellous bone. Biomech.Model.Mechanobiol., 7:463–476, 2008.
doi:10.1007/s10237-007-0109-7.

J. Panyasantisuk, D.H. Pahr, and P.K. Zysset. Effect of boundary conditions on yield
properties of human femoral trabecular bone. Biomech.Model.Mechanobiol., 15:1043–
1053, 2016. doi:10.1007/s10237-015-0741-6.

S.D. Papka and S. Kyriakides. In-plane compressive response and crushing of honeycomb.
J.Mech.Phys.Sol., 42:1499–1532, 1994. doi:10.1016/0022-5096(94)90085-X.

A. Paquin, H. Sabar, and M. Berveiller. Integral formulation and self consistent modelling
of elastoviscoplastic behaviour of heterogeneous materials. Arch.Appl.Mech., 69:14–35,
1999. doi:10.1007/s004190050201.

W.J. Parnell. The Eshelby, Hill, moment and concentration tensors for ellipsoidal inho-
mogeneities in the Newtonian potential problem and linear elastostatics. J.Elast., 125:
231–294, 2016. doi:10.1007/s10659-016-9573-6.

198
W.J. Parnell and C. Calvo-Jurado. On the computation of the Hashin–Shtrikman
bounds for transversely isotropic two-phase linear elastic fibre-reinforced composites.
J.Engng.Math., 95:295–323, 2015. doi:10.1007/s10665-014-9777-3.
M.V. Pathan, V.L. Tagarielli, S. Patsias, and P.M. Baiz-Villafranca. A new algorithm
to generate representative volume elements of composites with cylindrical or spherical
fillers. Composites B, 110:267–278, 2017. doi:10.1016/j.compositesb.2016.10.078.
C.B. Pedersen and P.J. Withers. Iterative estimates of internal stresses in short-fibre metal
matrix composites. Phil.Mag., A65:1217–1233, 1992. doi:10.1080/01418619208201506.
O.B. Pedersen. Thermoelasticity and plasticity of composites — I. Mean field theory. Acta
metall., 31:1795–1808, 1983. doi:10.1016/0001-6160(83)90126-8.
C. Pelissou, J. Baccou, Y. Monerie, and F. Perales. Determination of the size of the
representative volume element for random quasi-brittle composites. Int.J.Sol.Struct.,
46:2842–2855, 2009. doi:10.1016/j.ijsolstr.2009.03.015.
P. Peng, M. Gao, E. Guo, H. Kang, H. Xie, Z. Chen, and T. Wang. Deformation behavior
and damage in B4 Cp /6061 Al composites: An actual 3D microstructure based modeling.
Mater.Sci.Engng.A, 781:139169, 2020. doi:10.1016/j.msea.2020.139169.
X. Peng, S. Tang, N. Hu, and J. Han. Determination of the Eshelby tensor in mean-field
schemes for evaluation of mechanical properties of elastoplastic composites. Int.J.Plast.,
76:147–165, 2016. doi:10.1016/j.ijplas.2015.07.009.
V. Pensée and Q.C. He. Generalized self-consistent estimation of the apparent isotropic
elastic moduli and minimum representative volume element size of heterogeneous media.
Int.J.Sol.Struct., 44:2225–2243, 2007. doi:10.1016/j.ijsolstr.2006.07.003.
E.S. Perdahcıoğlu and J.M. Geijselaers. Constitutive modeling of two phase materials
using the mean field method of homogenization. Int.J.Mater.Form., 4:93–102, 2011.
doi:10.1007/s12289-010-1007-6.
H.E. Pettermann. Derivation and Finite Element Implementation of Constitutive Material
Laws for Multiphase Composites Based on Mori–Tanaka Approaches. Reihe 18, Nr. 217.
VDI–Verlag, Düsseldorf, 1997. ISBN 3-183-21718-X.
H.E. Pettermann and J. Hüsing. Modeling and simulation of relaxation in vis-
coelastic open cell materials and structures. Int.J.Sol.Struct., 49:2848–2853, 2012.
doi:10.1016/j.ijsolstr.2012.04.027.
H.E. Pettermann and S. Suresh. A comprehensive unit cell model: A study of cou-
pled effects in piezoelectric 1–3 composites. Int.J.Sol.Struct., 37:5447–5464, 2000.
doi:10.1016/S0020-7683(99)00224-3.
H.E. Pettermann, H.J. Böhm, and F.G. Rammerstorfer. Some direction dependent prop-
erties of matrix–inclusion type composites with given reinforcement orientation distri-
butions. Composites B, 28:253–265, 1997. doi:10.1016/S1359-8368(96)00055-8.
N. Phan-Thien and G.W. Milton. New third-order bounds on the effective moduli of
n-phase composites. Quart.Appl.Math., 41:59–74, 1983. doi:10.1090/qam/700661.

199
N. Phan-Thien and D.C. Pham. Differential multiphase models for polydispersed spheroidal
inclusions: Thermal conductivity and effective viscosity. Int.J.Engng.Sci., 38:73–88,
2000. doi:10.1016/S0020-7225(99)00016-6.

C. Pichler and R. Lackner. Upscaling the viscoelastic properties of highly

filled composites: Investigation of matrix–inclusion-type morphologies with power-
law viscoelastic material response. Compos.Sci.Technol., 69:2410–2420, 2009.
doi:10.1016/j.compscitech.2009.06.008.

O. Pierard, C. Friebel, and I. Doghri. Mean-field homogenization of multi-phase thermo-

elastic composites: A general framework and its validation. Compos.Sci.Technol., 64:
1587–1603, 2004. doi:10.1016/j.compscitech.2003.11.009.

O. Pierard, C. González, J. Segurado, J. LLorca, and I. Doghri. Micromechanics of elasto-

plastic materials reinforced with ellipsoidal inclusions. Int.J.Sol.Struct., 44:6945–6962,
2007. doi:10.1016/j.ijsolstr.2007.03.019.

A.F. Plankensteiner. Multiscale Treatment of Heterogeneous Nonlinear Solids and Struc-

tures. Reihe 18, Nr. 248. VDI–Verlag, Düsseldorf, 2000. ISBN 3-183-24818-2.

A.F. Plankensteiner, H.J. Böhm, F.G. Rammerstorfer, V.A. Buryachenko, and G. Hackl.
Modeling of layer-structured high speed steel. Acta mater., 45:1875–1887, 1997.
doi:10.1016/S1359-6454(96)00327-8.

J. Podgórski. Criterion for angle prediction for the crack in materials with random struc-
ture. Mech.Contr., 30:229–233, 2011.

P. Ponte Castañeda. The effective mechanical properties of nonlinear isotropic composites.

J.Mech.Phys.Sol., 39:45–71, 1991. doi:10.1016/0022-5096(91)90030-R.

P. Ponte Castañeda. Bounds and estimates for the properties of nonlinear heterogeneous
systems. Phil.Trans.Roy.Soc., A340:531–567, 1992. doi:10.1098/rsta.1992.0079.

P. Ponte Castañeda and P. Suquet. Nonlinear composites. In E. van der Giessen and T.Y.
Wu, editors, Advances in Applied Mechanics 34, pages 171–302, New York, NY, 1998.
Academic Press. doi:10.1016/S0065-2156(08)70321-1.

P. Ponte Castañeda and J.R. Willis. The effect of spatial distribution on the effective
behavior of composite materials and cracked media. J.Mech.Phys.Sol., 43:1919–1951,
1995. doi:10.1016/0022-5096(95)00058-Q.

A. Pontefisso, M. Zappalorto, and M. Quaresimin. Effectiveness of the random sequen-

tial absorption algorithm in the analysis of volume elements with nanoplatelets. Com-
put.Mater.Sci., 117:511–517, 2016. doi:10.1016/j.commatsci.2016.02.024.

G.L. Povirk, S.R. Nutt, and A. Needleman. Analysis of creep in thermally cycled Al/SiC
composites. Scr.metall.mater., 26:461–66, 1992. doi:10.1016/0956-716X(92)90630-W.

C.P. Przybyla and D.L. McDowell. Microstructure-sensitive extreme value probabili-

ties for high-cycle fatigue of Ni-base superalloy IN100. Int.J.Plast., 26:372–384, 2010.
doi:10.1016/j.ijplas.2009.08.001.

200
M. Ptashnyk and B. Seguin. Periodic homogenization and material symmetry in linear
elasticity. J.Elast., 124:225–241, 2016. doi:10.1007/s10659-015-9566-x.

S.M. Qidwai, D.M. Turner, S.R. Niezgoda, A.C. Lewis, A.B. Geltmacher, D.J. Rowen-
horst, and S.R. Kalidindi. Estimating the response of polycrystalline materials us-
ing sets of weighted statistical volume elements. Acta mater., 60:5284–5299, 2012.
doi:10.1016/j.actamat.2012.06.026.

Y.P. Qiu and G.J. Weng. A theory of plasticity for porous materials and particle-reinforced
composites. J.Appl.Mech., 59:261–268, 1992. doi:10.1115/1.2899515.

J. Qu and M. Cherkaoui. Fundamentals of Micromechanics of Solids. John Wiley, New

York, NY, 2006. ISBN 0-471-46451-1.

S. Quilici and G. Cailletaud. FE simulation of macro-, meso- and mi-

croscales in polycrystalline plasticity. Comput.Mater.Sci., 16:383–390, 1999.
doi:10.1016/S0927-0256(99)00081-6.

J.A. Quintanilla. Microstructure and properties of random heterogeneous materials: A re-

view of theoretical results. Polym.Engng.Sci., 39:559–585, 1999. doi:10.1002/pen.11446.

D. Raabe. Computational Materials Science. VCH, Weinheim, 1998. ISBN 3-527-29541-0.

F.K.F. Radtke, A. Simone, and L.J. Sluys. A partition of unity finite element method
for simulating non-linear debonding and matrix failure in thin fibre composites.
Int.J.Numer.Meth.Engng., 86:453–476, 2011. doi:10.1002/nme.3056.

P. Raghavan and S. Ghosh. Adaptive multi-scale modeling of composite materials. Com-

put.Model.Engng.Sci., 5:151–170, 2004. doi:10.3970/cmes.2004.005.151.

F.G. Rammerstorfer and H.J. Böhm. Finite element methods in micromechanics of com-
posites and foam materials. In B.H.V. Topping, editor, Computational Mechanics for
the Twenty First Century, pages 145–164, Edinburgh, 2000. Saxe–Coburg Publications.

A. Rasool and H.J. Böhm. Effects of particle shape on the macroscopic and microscopic
linear behaviors of particle reinforced composites. Int.J.Engng.Sci., 58:21–34, 2012.
doi:10.1016/j.ijengsci.2012.03.022.

C. Redenbach. Microstructure models for cellular materials. Comput.Mater.Sci., 44:1397–

1407, 2009. doi:10.1016/j.commatsci.2008.09.018.

T.J. Reiter, G.J. Dvorak, and V. Tvergaard. Micromechanical models for graded composite
materials. J.Mech.Phys.Sol., 45:1281–1302, 1997. doi:10.1016/S0022-5096(97)00007-0.

Z.Y. Ren and Q.S. Zheng. Effects of grain sizes, shapes, and distribution on minimum
sizes of representative volume elements of cubic polycrystals. Mech.Mater., 36:1217–
1229, 2004. doi:10.1016/j.mechmat.2003.11.002.

A. Reuss. Berechnung der Fließgrenze von Mischkristallen auf Grund der Plastizitätsbe-
dingung für Einkristalle. ZAMM, 9:49–58, 1929. doi:10.1002/zamm.19290090104.

201
A. Riccardi and F. Montheillet. A generalized self-consistent method for solids con-
taining randomly oriented spheroidal inclusions. Acta Mech., 133:39–56, 1999.
doi:10.1007/BF01179009.

J.R. Rice and D.M. Tracey. On the ductile enlargement of voids in triaxial stress fields.
J.Mech.Phys.Sol., 17:201–217, 1969. doi:10.1016/0022-5096(69)90033-7.

M. Rintoul and S. Torquato. Reconstruction of the structure of dispersions. J.Colloid

Interf.Sci., 186:467–476, 1997. doi:10.1006/jcis.1996.4675.

A.P. Roberts and E.J. Garboczi. Elastic moduli of model random three-dimensional closed-
cell cellular solids. Acta mater., 49:189–197, 2001. doi:10.1016/S1359-6454(00)00314-1.

A.P. Roberts and E.J. Garboczi. Elastic properties of a tungsten–silver compos-

ite by reconstruction and computation. J.Mech.Phys.Sol., 47:2029–2055, 1999.
doi:10.1016/S0022-5096(99)00016-2.

F. Robitaille, A.C. Long, I.A. Jones, and C.D. Rudd. Automatically generated geomet-
ric descriptions of textile and composite unit cells. Composites A, 34:303–312, 2003.
doi:10.1016/S1359-835X(03)00063-0.

G.J. Rodin. The overall elastic response of materials containing spherical inhomogeneities.
Int.J.Sol.Struct., 30:1849–1863, 1993. doi:10.1016/0020-7683(93)90221-R.

G.J. Rodin and G.J. Weng. On reflected interactions in elastic solids containing inhomo-
geneities. J.Mech.Phys.Sol., 68:197–209, 2014. doi:10.1016/j.jmps.2014.04.001.

R. Roscoe. Isotropic composites with elastic or viscoelastic phases: General bounds

for the moduli and solutions for special geometries. Rheol.Acta, 12:404–411, 1973.
doi:10.1007/BF01502992.

B.W. Rosen and Z. Hashin. Effective thermal expansion coefficients and

specific heats of composite materials. Int.J.Engng.Sci., 8:157–173, 1970.
doi:10.1016/0020-7225(70)90066-2.

F. Roters, P. Eisenlohr, L. Hantcherli, D.D. Tjahjanto, T.R. Bieler, and D. Raabe.

Overview of constitutive laws, kinematics, homogenization and multiscale methods in
crystal plasticity finite element modeling: Theory, experiments, applications. Acta
mater., 58:1152–1211, 2010. doi:10.1016/j.actamat.2009.10.058.

F. Saadat, V. Birman, S. Thomopoulos, and G. Genin. Effective elastic properties of

a composite containing multiple types of anisotropic ellipsoidal inclusions, with ap-
plication to the attachment of tendon to bone. J.Mech.Phys.Sol., 82:367–377, 2015.
doi:10.1016/j.jmps.2015.05.017.

K. Sab. On the homogenization and the simulation of random materials.

Eur.J.Mech. A/Solids, 11:585–607, 1992.

G. Sachs. Zur Ableitung einer Fließbedingung. Z.VDI, 72:734–736, 1928.

202
I. Saiki, K. Terada, K. Ikeda, and M. Hori. Appropriate number of unit cells
in a representative volume element for micro-structural bifurcation encountered
in a multi-scale modeling. Comput.Meth.Appl.Mech.Engng., 191:2561–2585, 2002.
doi:10.1016/S0045-7825(01)00413-3.

V. Salit and D. Gross. On the convergence of the iterative self-consistent embedded cell
model. Comput.Mater.Sci., 81:199–204, 2014. doi:10.1016/j.commatsci.2013.08.014.

M. Salmi, F. Auslender, M. Bornert, and M. Fogli. Apparent and effective mechanical prop-
erties of linear matrix–inclusion random composites: Improved bounds for the effective
behavior. Int.J.Sol.Struct., 49:1195–1211, 2012. doi:10.1016/j.ijsolstr.2012.01.018.

W. Sanders and L.J. Gibson. Mechanics of hollow sphere foams. Mater.Sci.Engng. A, 347:
70–85, 2003. doi:10.1016/S0921-5093(02)00583-X.

A. Sangani and W. Lu. Elastic coefficients of composites containing spherical inclusions in

a periodic array. J.Mech.Phys.Sol., 35:1–21, 1987. doi:10.1016/0022-5096(87)90024-X.

A. Sangani and G. Mo. Elastic interactions in particulate composites with per-

fect as well as imperfect interfaces. J.Mech.Phys.Sol., 45:2001–2031, 1997.
doi:10.1016/S0022-5096(97)00025-2.

S.P. Santosa and T. Wierzbicki. On the modeling of crush behavior of

a closed-cell aluminum foam structure. J.Mech.Phys.Sol., 46:645–669, 1998.
doi:10.1016/S0022-5096(97)00082-3.

M. Sautter, C. Dietrich, M.H. Poech, S. Schmauder, and H.F. Fischmeister. Finite element
modelling of a transverse-loaded fibre composite: Effects of section size and net density.
Comput.Mater.Sci., 1:225–233, 1993. doi:10.1016/0927-0256(93)90014-E.

D. Savvas, G. Stefanou, and M. Papadrakakis. Determination of RVE size for random

composites with local volume fraction variation. Comput.Meth.Appl.Mech.Engng., 305:
340–358, 2016. doi:10.1016/j.cma.2016.03.002.

R.A. Schapery. Approximate methods of transform inversion in viscoelastic stress analysis.

In Proceedings of the Fourth US National Congress on Applied Mechanics, pages 1075–
1085, New York, NY, 1962. ASME.

R.A. Schapery. Viscoelastic behavior and analysis of composite materials. In G.P.

Sendeckyj, editor, Mechanics of Composite Materials 2, pages 84–168, New York, NY,
1974. Academic Press.

L. Scheunemann, D. Balzani, D. Brands, and J. Schröder. Design of 3D statistically similar

representative volume elements based on Minkowski functionals. Mech.Mater., 90:185–
201, 2015. doi:10.1016/j.mechmat.2015.03.005.

J. Schjødt-Thomsen and R. Pyrz. Influence of statistical cell description on the local

strain and overall properties of cellular materials. In S. Ghosh, J.M. Castro, and J.K.
Lee, editors, Proceedings of NUMIFORM 2004, pages 1630–1635, Melville, NY, 2004.
American Institute of Physics. doi:10.1063/1.1766763.

203
S. Schmauder, J. Wulf, T. Steinkopff, and H. Fischmeister. Micromechanics of plasticity
and damage in an Al/SiC metal matrix composite. In A. Pineau and A. Zaoui, edi-
tors, Micromechanics of Plasticity and Damage of Multiphase Materials, pages 255–262,
Dordrecht, 1996. Kluwer. doi:10.1007/978-94-009-1756-9 32.

K. Schneider, B. Klusemann, and S. Bargmann. Fully periodic RVEs for technologically

relevant composites: Not worth the effort! J.Mech.Mater.Struct., 12:471–484, 2017.
doi:10.2140/jomms.2017.12.471.

M. Schneider. The sequential addition and migration method to generate representative

volume elements for the homogenization of short fiber reinforced plastics. Comput.Mech.,
59:247–263, 2017. doi:10.1007/s00466-016-1350-7.

M. Schneider. A review of nonlinear FFT-based computational homogenization methods.

Acta Mech., 232:2051–2100, 2021. doi:10.1007/s00707-021-02962-1.

M. Schneider. Superaccurate effective elastic moduli via postprocessing in computational

homogenization. Int.J.Num.Meth.Engng., 123:4119–4135, 2022. doi:10.1002/nme.7002.

M. Schneider, M. Josien, and F. Otto. Representative volume elements for matrix–inclusion

composites — A computational study on the effects of an improper treatment of particles
intersecting the boundary and the benefits of periodizing the ensemble. J.Mech.Phys.Sol.,
158:104652, 2022. doi:10.1016/j.jmps.2021.104652.

P.J. Schneider and D.H. Eberly. Geometric Tools for Computer Graphics. Morgan–
Kaufmann, San Francisco, 2003. ISBN 1-558-60594-0.

B.A. Schrefler, M. Lefik, and U. Galvanetto. Correctors in a beam model for unidirectional
composites. Compos.Mater.Struct., 4:159–190, 1997. doi:10.1080/10759419708945879.

B.A. Schrefler, U. Galvanetto, C. Pellegrino, and F. Ohmenhäuser. Global non-linear

behavior of periodic composite materials. In H.A. Mang and F.G. Rammerstorfer, ed-
itors, Discretization Methods in Structural Mechanics, pages 265–272, Dordrecht, 1999.
Kluwer. doi:10.1007/978-94-011-4589-3 31.

J. Segurado, J. LLorca, and C. González. On the accuracy of mean-field approaches

to simulate the plastic deformation of composites. Scr.mater., 46:525–529, 2002a.
doi:10.1016/S1359-6462(02)00027-1.

J. Segurado, E. Parteder, A. Plankensteiner, and H.J. Böhm. Micromechanical studies

of the densification of porous molybdenum. Mater.Sci.Engng. A, 333:270–278, 2002b.
doi:10.1016/S0921-5093(01)01853-6.

J. Segurado, C. González, and J. LLorca. A numerical investigation of the effect of particle

clustering on the mechanical properties of composites. Acta mater., 51:2355–2369, 2003.
doi:10.1016/S1359-6454(03)00043-0.

A. Selmi, I. Doghri, and L. Adam. Micromechanical simulation of biaxial yield, hardening

and plastic flow in short glass fiber reinforced polyamide. Int.J.Mech.Sci., 53:696–706,
2011. doi:10.1016/j.ijmecsci.2011.06.002.

204
I. Sevostianov. On the shape of effective inclusion in the Maxwell homogeniza-
tion scheme for anisotropic elastic composites. Mech.Mater., 75:45–59, 2014.
doi:10.1016/j.mechmat.2014.03.003.
I. Sevostianov and A. Giraud. Generalization of Maxwell homogenization scheme for elastic
material containing inhomogeneities of diverse shape. Int.J.Engng.Sci., 64:23–36, 2013.
doi:10.1016/j.ijengsci.2012.12.004.
I. Sevostianov and M. Kachanov. Explicit cross-property correlations for
anisotropic two-phase composite materials. J.Mech.Phys.Sol., 50:253–282, 2002.
doi:10.1016/S0022-5096(01)00051-5.
I. Sevostianov and M. Kachanov. Effect of interphase layers on the overall elastic
and conductive properties of matrix composites. Application to nanosize inclusion.
Int.J.Sol.Struct., 44:1304–1315, 2007a. doi:10.1016/j.ijsolstr.2006.06.020.
I. Sevostianov and M. Kachanov. Elastic fields generated by inhomogeneities: Far-
field asymptotics, its shape dependence and relation to the effective elastic properties.
Int.J.Sol.Struct., 48:2340–2348, 2011. doi:10.1016/j.ijsolstr.2011.04.014.
I. Sevostianov and M. Kachanov. On some controversial issues in effective field ap-
proaches to the problem of overall elastic properties. Mech.Mater., 69:93–105, 2014.
doi:10.1016/j.mechmat.2013.09.010.
I. Sevostianov and M. Kachanov. Compliance tensors of ellipsoidal inclusions. Int.J.Fract.,
96:L3–L7, 1999. doi:10.1023/A:1018712913071.
I. Sevostianov, M. Kachanov, and T. Zohdi. On computation of the compliance and stiffness
contribution tensors to non ellipsoidal inhomogeneities. Int.J.Sol.Struct., 45:4375–4383,
2008. doi:10.1016/j.ijsolstr.2008.03.020.
I. Sevostianov, S.G. Mogilevskaya, and V.I. Kushch. Maxwell’s methodology of es-
timating effective properties: Alive and well. Int.J.Sol.Struct., 140:35–88, 2019.
doi:10.1016/j.ijengsci.2019.05.001.
Z. Shan and A.M. Gokhale. Micromechanics of complex three-dimensional microstructures.
Acta mater., 49:2001–2015, 2001. doi:10.1016/S1359-6454(01)00093-3.
A.S. Shedbale, I.V. Singh, and B.K. Mishra. Heterogeneous and homogenized models
for predicting the indentation response of particle reinforced metal matrix composites.
Int.J.Mech.Mater.Design, 13:531–552, 2017. doi:10.1007/s10999-016-9352-3.
H. Shen and L.C. Brinson. A numerical investigation of the effect of boundary conditions
and representative volume element size for porous titanium. J.Mech.Mater.Struct., 1:
1179–1204, 2006. doi:10.2140/jomms.2006.1.1179.
H. Shen and C.J. Lissenden. 3D finite element analysis of particle-reinforced aluminum.
Mater.Sci.Engng. A, 338:271–281, 2002. doi:10.1016/S0921-5093(02)00094-1.
L. Shen and S. Yi. An effective inclusion model for effective moduli of heteroge-
neous materials with ellipsoidal inhomogeneities. Int.J.Sol.Struct., 38:5789–5805, 2001.
doi:10.1016/S0020-7683(00)00370-X.

205
P. Sheng, J. Zhang, and Z. Ji. An advanced 3D modeling method for concrete-like particle-
reinforced composites with high volume fraction of randomly distributed particles. Com-
pos.Sci.Technol., 134:26–35, 2016. doi:10.1016/j.compscitech.2016.08.009.

M.S. Shephard, M.W. Beall, R. Garimella, and R. Wentorf. Automatic construc-

tion of 3/D models in multiple scale analysis. Comput.Mech., 17:196–207, 1995.
doi:10.1007/BF00364081.

M. Sherburn. Geometrical and Mechanical Modelling of Textiles. PhD thesis, University

of Nottingham, Nottingham, 2007.

D.L. Shi, X.Q. Feng, Y.Y. Huang, K.C. Hwang, and H. Gao. The effect of nanotube
waviness and agglomeration on the elastic property of carbon nanotube-reinforced com-
posites. J.Engng.Mater.Technol., 126:250–257, 2004. doi:10.1115/1.1751182.

V. Shulmeister, M.W.D. van der Burg, E. van der Giessen, and R. Marissen. A numerical
study of large deformations of low-density elastomeric open-cell foams. Mech.Mater., 30:
125–140, 1998. doi:10.1016/S0167-6636(98)00033-7.

T. Siegmund, F.D. Fischer, and E.A. Werner. Microstructure characterization and FE-
modeling of plastic flow in a duplex steel. In R. Pyrz, editor, Microstructure–Property
Interactions in Composite Materials, pages 349–360, Dordrecht, 1995. Kluwer Academic
Publishers. doi:10.1007/978-94-011-0059-5 29.

T. Siegmund, R. Cipra, J. Liakus, B. Wang, M. LaForest, and A. Fatz. Processing–

microstructure–property relationships in short fiber reinforced carbon–carbon com-
posite system. In H.J. Böhm, editor, Mechanics of Microstructured Materi-
als, pages 235–258. Springer-Verlag, CISM Courses and Lectures Vol. 464, 2004.
doi:10.1007/978-3-7091-2776-6 7.

S. Sihn and A.K. Roy. Modeling and prediction of bulk properties of open-cell carbon
foam. J.Mech.Phys.Sol., 52:167–191, 2005. doi:10.1016/S0022-5096(03)00072-3.

V.V. Silberschmidt and E.A. Werner. Analyses of thermal stresses’ evolution in ferritic–
austenitic duplex steels. In Y. Tanigawa, R.B. Hetnarski, and N. Noda, editors, Thermal
Stresses 2001, pages 327–330, Osaka, 2001. Osaka Prefecture University.

N. Silnutzer. Effective Constants of Statistically Homogeneous Materials. PhD thesis,

University of Pennsylvania, Philadelphia, PA, 1972.

A.E. Simone and L.J. Gibson. Effects of solid distribution on the stiffness and strength of
metallic foams. Acta mater., 46:2139–2150, 1998. doi:10.1016/S1359-6454(97)00421-7.

R.J.M. Smit, W.A.M. Brekelmans, and H.E.H. Meijer. Prediction of the mechanical be-
havior of non-linear heterogeneous systems by multi-level finite element modeling. Com-
put.Meth.Appl.Mech.Engng., 155:181–192, 1998. doi:10.1016/S0045-7825(97)00139-4.

S. Soghrati, A.M. Aragón, C.A. Duarte, and P.H. Geubelle. An interface-enriched gener-
alized FEM for problems with discontinuous gradient fields. Int.J.Numer.Meth.Engng.,
89:991–1008, 2012. doi:10.1002/nme.3273.

206
G. Soni, R. Singh, M. Mitra, and B. Falzon. Modelling matrix damage and
fibre–matrix interfacial decohesion in composite laminates via a multi-fibre multi-
layer representative volume element (M2 RVE). Int.J.Sol.Struct., 51:449–461, 2014.
doi:10.1016/j.ijsolstr.2013.10.018.
N.J. Sørensen. A planar type analysis for the elastic-plastic behaviour of continuous fibre-
reinforced metal-matrix composites under longitudinal shearing and combined loading.
Int.J.Sol.Struct., 29:867–877, 1992. doi:10.1016/0020-7683(92)90022-L.
N.J. Sørensen, S. Suresh, V. Tvergaard, and A. Needleman. Effects of reinforcement
orientation on the tensile response of metal matrix composites. Mater.Sci.Engng. A,
197:1–10, 1995. doi:10.1016/0921-5093(94)09739-9.
C. Soyarslan, S. Bargmann, M. Pradas, and J. Weissmüller. 3D stochastic bicontinuous
microstructures: Generation, topology and elasticity. Acta mater., 149:326–340, 2018.
doi:10.1016/j.actamat.2018.01.005.
M. Stroeven, H. Askes, and L.J. Sluys. Numerical determination of representative vol-
umes for granular materials. Comput.Meth.Appl.Mech.Engng., 193:3221–3238, 2004.
doi:10.1016/j.cma.2003.09.023.
N. Sukumar, D.L. Chopp, N. Moës, and T. Belytschko. Modeling holes and inclusions by
level sets in the extended finite element method. Comput.Meth.Appl.Mech.Engng., 190:
6183–6900, 2001. doi:10.1016/S0045-7825(01)00215-8.
R.M. Sullivan and L.J. Ghosn. Shear moduli for non-isotropic, open cell foam us-
ing a general elongated Kelvin foam model. Int.J.Engng.Sci., 47:990–1001, 2009.
doi:10.1016/j.ijengsci.2009.05.005.
P. Suquet. Elements of homogenization for inelastic solid mechanics. In E. Sanchez-Palencia
and A. Zaoui, editors, Homogenization Techniques in Composite Media, pages 194–278,
Berlin, 1987. Springer-Verlag.
P. Suquet. Overall properties of nonlinear composites: A modified secant moduli theory
and its link with Ponte Castañeda’s nonlinear variational procedure. C.R.Acad.Sci.Paris,
série IIb, 320:563–571, 1995.
P. Suquet. Effective properties of nonlinear composites. In P. Suquet, edi-
tor, Continuum Micromechanics, pages 197–264, Vienna, 1997. Springer-Verlag.
doi:10.1007/978-3-7091-2662-2 4.
T. Suzuki and P.K.L. Yu. Complex elastic wave band structures in three-
dimensional periodic elastic media. J.Mech.Phys.Sol., 46:115–138, 1998.
doi:10.1016/S0022-5096(97)00023-9.
S. Swaminathan and S. Ghosh. Statistically equivalent representative volume elements
for unidirectional composite microstructures: Part II — With interfacial debonding.
J.Compos.Mater., 40:605–621, 2006. doi:10.1177/0021998305055274.
S. Swaminathan, S. Ghosh, and N.J. Pagano. Statistically equivalent representative vol-
ume elements for unidirectional composite microstructures: Part I — Without damage.
J.Compos.Mater., 40:583–604, 2006. doi:10.1177/0021998305055273.

207
E.B. Tadmor and R.E. Miller. Modeling Materials: Continuum, Atomistic and Multiscale
Techniques. Cambridge University Press, Cambridge, 2011. ISBN 0-521-85698-1.

N. Takano, M. Zako, and T. Okazaki. Efficient modeling of microscopic heterogeneity and

local crack in composite materials by finite element mesh superposition method. JSME
Int.J.Srs.A, 44:602–609, 2001. doi:10.1299/jsmea.44.602.

D.R.S. Talbot and J.R. Willis. Variational principles for inhomogeneous non-linear media.
J.Appl.Math., 35:39–54, 1985. doi:10.1093/imamat/35.1.39.

D.R.S. Talbot and J.R. Willis. Upper and lower bounds for the overall response of an
elastoplastic composite. Mech.Mater., 28:1–8, 1998. doi:10.1016/S0167-6636(97)00012-4.

V.B.C. Tan, K. Raju, and H.P. Lee. Direct FE2 for concurrent multilevel mod-
elling of heterogeneous structures. Comput.Meth.Appl.Mech.Engng., 360:112694, 2020.
doi:10.1016/j.cma.2019.112694.

G.P. Tandon and G.J. Weng. The effect of aspect ratio of inclusions on the elastic
properties of unidirectionally aligned composites. Polym.Compos., 5:327–333, 1984.
doi:10.1002/pc.750050413.

G.P. Tandon and G.J. Weng. Average stress in the matrix and effective mod-
uli of randomly oriented composites. Compos.Sci.Technol., 27:111–132, 1986.
doi:10.1016/0266-3538(86)90067-9.

G.P. Tandon and G.J. Weng. A theory of particle-reinforced plasticity. J.Appl.Mech., 55:
126–135, 1988. doi:10.1115/1.3173618.

T. Tang and W. Yu. A variational asymptotic micromechanics model for predict-

ing conductivities of composite materials. J.Mech.Mater.Struct., 2:1813–1830, 2007.
doi:10.2140/jomms.2007.2.1813.

T. Tang and W. Yu. Asymptotic approach to initial yielding surface and elasto-
plasticity of heterogeneous materials. Mech.Adv.Mater.Struct., 18:244–254, 2011.
doi:10.1080/15376494.2010.483324.

X. Tang and J.D. Whitcomb. General techniques for exploiting periodicity and symmetries
in micromechanics analysis of textile composites. J.Compos.Mater., 37:1167–1189, 2003.
doi:10.1177/00219983030370130.

M. Tashkinov. Multipoint stochastic approach to localization of microscale elas-

tic behavior of random heterogeneous media. Comput.Struct., 249:106474, 2021.
doi:10.1016/j.compstruc.2020.106474.

M. Taya and T. Mori. Dislocations punched-out around a short fiber in MMC

subjected to uniform temperature change. Acta metall., 35:155–162, 1987.
doi:10.1016/0001-6160(87)90224-0.

M. Taya, W.D. Armstrong, M.L. Dunn, and T. Mori. Analytical study on dimensional
changes in thermally cycled metal matrix composites. Mater.Sci.Engng. A, 143:143–
154, 1991. doi:10.1016/0921-5093(91)90734-5.

208
G. Taylor. Plastic strain in metals. J.Inst.Metals, 62:307–324, 1938.

C. Tekoğlu, L.J. Gibson, T. Pardoen, and P.R. Onck. Size effects in foams: Experiments
and modeling. Prog.Mater.Sci., 56:109–138, 2011. doi:10.1016/j.pmatsci.2010.06.001.

I. Temizer and P. Wriggers. Homogenization in finite thermoelasticity. J.Mech.Phys.Sol.,

59:344–372, 2011. doi:10.1016/j.jmps.2010.10.004.

I. Temizer, T. Wu, and P. Wriggers. On the optimality of the window

method in computational homogenization. Int.J.Engng.Sci., 64:66–73, 2013.
doi:10.1016/j.ijengsci.2012.12.007.

J.L. Teply and G.J. Dvorak. Bounds on overall instantaneous properties of elastic-plastic
composites. J.Mech.Phys.Sol., 36:29–58, 1988. doi:10.1016/0022-5096(88)90019-1.

K. Terada and N. Kikuchi. Microstructural design of composites using the ho-

mogenization method and digital images. Mater.Sci.Res.Int., 2:65–72, 1996.
doi:10.2472/jsms.45.6Appendix 65.

K. Terada, T. Miura, and N. Kikuchi. Digital image-based modeling applied to the

homogenization analysis of composite materials. Comput.Mech., 20:331–346, 1997.
doi:10.1007/s004660050255.

K. Terada, M. Hori, T. Kyoya, and N. Kikuchi. Simulation of the multi-scale conver-

gence in computational homogenization approach. Int.J.Sol.Struct., 37:2285–2311, 2000.
doi:10.1016/S0020-7683(98)00341-2.

K. Terada, I. Saiki, K. Matsui, and Y. Yamakawa. Two-scale kinematics and linearization

for simultaneous two-scale analysis of periodic heterogeneous solids at finite strain. Com-
put.Meth.Appl.Mech.Engng., 192:3531–3563, 2003. doi:10.1016/S0045-7825(03)00365-7.

W. Tian, L. Qi, J. Zhou, J. Liang, and Y. Ma. Representative volume element for compos-
ites reinforced by spatially randomly oriented discontinuous fibers and its applications.
Compos.Struct., 131:366–373, 2015. doi:10.1016/j.compstruct.2015.05.014.

K. Tohgo and T.W. Chou. Incremental theory of particulate-reinforced composites in-

cluding debonding damage. JSME Int.J.Srs.A, 39:389–397, 1996. doi:10.1299/js-
mea1993.39.3 389.

Y. Tomita, Y. Higa, and T. Fujimoto. Modeling and estimation of deformation behav-

ior of particle reinforced metal-matrix composite. Int.J.Mech.Sci., 42:2249–2260, 2000.
doi:10.1016/S0020-7403(00)00006-0.

S. Torquato. Random Heterogeneous Media. Springer-Verlag, New York, NY, 2002. ISBN
1-475-76357-3.

S. Torquato. Random heterogeneous media: Microstructure and improved bounds on

effective properties. Appl.Mech.Rev., 44:37–75, 1991. doi:10.1115/1.3119494.

S. Torquato. Effective stiffness tensor of composite media: I. Exact series expansions.

J.Mech.Phys.Sol., 45:1421–1448, 1997. doi:10.1016/S0022-5096(97)00019-7.

209
S. Torquato. Effective stiffness tensor of composite media: II. Applications to isotropic dis-
persions. J.Mech.Phys.Sol., 46:1411–1440, 1998a. doi:10.1016/S0022-5096(97)00083-5.

S. Torquato. Morphology and effective properties of disordered heterogeneous media.

Int.J.Sol.Struct., 35:2385–2406, 1998b. doi:10.1016/S0020-7683(97)00142-X.

S. Torquato and S. Hyun. Effective-medium approximation for composite media: Realiz-

able single-scale dispersions. J.Appl.Phys., 89:1725–1729, 2001. doi:10.1063/1.1336523.

S. Torquato and F. Lado. Improved bounds on the effective moduli of random arrays of
cylinders. J.Appl.Mech., 59:1–6, 1992. doi:10.1115/1.2899429.

S. Torquato and D.M. Rintoul. Effect of the interface on the properties of composite media.
Phys.Rev.Lett., 75:4067–4070, 1995. doi:10.1103/PhysRevLett.75.4067.

S. Torquato, C.L.Y. Yeong, M.D. Rintoul, D. Milius, and L.A. Aksay. Char-
acterizing the structure and mechanical properties of interpenetrating boron car-
bide/aluminum multiphase composites. J.Amer.Ceram.Soc., 82:1263–1268, 1999.
doi:10.1111/j.1151-2916.1999.tb01905.x.

C. Toulemonde, R. Masson, and J. El Gharib. Modeling the effective elastic behavior of

composites: A mixed finite element and homogenization approach. C.R.Mécanique, 336:
275–282, 2008. doi:10.1016/j.crme.2007.11.024.

D. Trias. Analysis and Simulation of Transverse Random Fracture of Long Fibre Reinforced
Composites. PhD thesis, Universitat de Girona, Girona, 2005.

D. Trias, J. Costa, A. Turon, and J.F. Hurtado. Determination of the critical size of a
statistical representative volume element (SRVE) for carbon reinforced polymers. Acta
mater., 54:3471–3484, 2006. doi:10.1016/j.actamat.2006.03.042.

A. Trofimov, B. Drach, and I. Sevostianov. Effective elastic properties of com-

posites with particles of polyhedral shapes. Int.J.Sol.Struct., 120:157–170, 2017.
doi:10.1016/j.ijsolstr.2017.04.037.

T.C. Tszeng. The effects of particle clustering on the mechanical behavior of particle rein-
forced composites. Composites B, 29:299–308, 1998. doi:10.1016/S1359-8368(97)00031-0.

C.L. Tucker and E. Liang. Stiffness predictions for unidirectional short-fiber

composites: Review and evaluation. Compos.Sci.Technol., 59:655–671, 1999.
doi:10.1016/S0266-3538(98)00120-1.

V. Tvergaard. Debonding of short fibres among particulates in a metal matrix composite.

Int.J.Sol.Struct., 40:6957–6967, 2003. doi:10.1016/S0020-7683(03)00347-0.

V. Tvergaard. Analysis of tensile properties for a whisker-reinforced metal-matrix com-

posite. Acta metall.mater., 38:185–194, 1990. doi:10.1016/0956-7151(90)90048-L.

V. Tvergaard. Fibre debonding and breakage in a whisker-reinforced metal. Mater.Sci.

Engng. A, 190:215–222, 1994. doi:10.1016/0921-5093(95)80005-0.

210
V. Tvergaard. Effect of void size difference on growth and cavitation instabilities.
J.Mech.Phys.Sol., 44:1237–1253, 1996. doi:10.1016/0022-5096(96)00032-4.
V. Tvergaard and A. Needleman. Analysis of the cup-cone fracture in a round tensile bar.
Acta metall., 32:157–169, 1984. doi:10.1016/0001-6160(84)90213-X.
A.J. Urbański. Unified, finite element based approach to the problem of homogenisation
of structural members with periodic microstructure. In W. Wunderlich, editor, Solids,
Structures and Coupled Problems in Engineering, Munich, 1999. TU München.
S. Vajjhala, A.M. Kraynik, and L.J. Gibson. A cellular solid model for modulus reduc-
tion due to resorption of trabeculae in bone. J.Biomech.Engng., 122:511–515, 2000.
doi:10.1115/1.1289996.
E. van der Giessen and V. Tvergaard. Development of final creep failure in polycrystalline
aggregates. Acta metall.mater., 42:952–973, 1994. doi:10.1016/0956-7151(94)90290-9.
E. van der Giessen, P.A. Schultz, N. Bertin, V.V. Bulatov, W. Cai, G. Csányi, S.M. Foiles,
M.G.D. Geers, C. González, M. Hütter, W.K. Kim, D.M. Kochmann, J. LLorca, A.E.
Mattson, J. Rottler, S. Shluger, R.B. Sills, I. Steinbach, A. Strachan, and Tadmor E.B.
Roadmap on multiscale materials modeling. Model.Simul.Mater.Sci.Engng., 42:043001,
2020. doi:10.1088/1361-651X/ab7150.
B. van Rietbergen, R. Müller, D. Ulrich, P. Rüegsegger, and R. Huiskes. Tissue stresses
and strain in trabeculae of a canine proximal femur can be quantified from computer
reconstructions. J.Biomech., 32:165–173, 1999. doi:10.1016/S0021-9290(98)00150-X.
K. Váradi, Z. Néder, K. Friedrich, and J. Flöck. Finite-element analysis of a poly-
mer composite subjected to ball indentation. Compos.Sci.Technol., 59:271–281, 1999.
doi:10.1016/S0266-3538(98)00066-9.
T.J. Vaughan and C.T. McCarthy. A combined experimental–numerical ap-
proach for generating statistically equivalent fibre distributions for high
strength laminated composite materials. Compos.Sci.Technol., 70:291–297, 2010.
doi:10.1016/j.compscitech.2009.10.020.
F. Vazeille and L. Laberge Lebel. Envelope enrichment method for ho-
mogenization of non-periodic structures. Compos.Struct., 329:117819, 2024.
doi:10.1016/j.compstruct.2023.117819.
A.S. Viglin. A quantitative measure of the texture of a polycrystalline material–texture
function. Sov.Phys.Solid State, 2:2195–2207, 1961.
E.N. Vilchevskaya, V.I. Kushch, M. Kachanov, and I. Sevostianov. Effective proper-
ties of periodic composites: Irrelevance of one particle homogenization techniques.
Mech.Mater., 159:103918, 2021. doi:10.1016/j.mechmat.2021.103918.
W. Voigt. Über die Beziehung zwischen den beiden Elasticitäts-Constanten isotroper
Körper. Ann.Phys., 38:573–587, 1889. doi:10.1002/andp.18892741206.
W.K. Vonach. A General Solution to the Wrinkling Problem of Sandwiches. Reihe 18,
Nr. 268. VDI–Verlag, Düsseldorf, 2001. ISBN 3-183-26818-3.

211
K. Wakashima, H. Tsukamoto, and B.H. Choi. Elastic and thermoelastic properties of
metal matrix composites with discontinuous fibers or particles: Theoretical guidelines
toward materials tailoring. In The Korea–Japan Metals Symposium on Composite Ma-
terials, pages 102–115, Seoul, 1988. The Korean Institute of Metals.

L.J. Walpole. On bounds for the overall elastic moduli of inhomogeneous systems — I.
J.Mech.Phys.Sol., 14:151–162, 1966. doi:10.1016/0022-5096(66)90035-4.

J. Wang, J.H. Andreasen, and B.L. Karihaloo. The solution of an inhomogeneity in a finite
plane region and its application to composite materials. Compos.Sci.Technol., 60:75–82,
2000. doi:10.1016/S0266-3538(99)00103-7.

R. Wang, L. Zhang, D. Hu, C. Liu, X. Shen, C. Cho, and B. Li. A novel approach to
impose periodic boundary condition on braided composite RVE model based on RPIM.
Compos.Struct., 163:77–88, 2017. doi:0.1016/j.compstruct.2016.12.032.

X. Wang, Z. Guan, S. Du, G. Han, and M. Zhang. A long-range force based random method
for generating anisotropic 2D fiber arrangement statistically equivalent to real compos-
ites. Compos.Sci.Technol., 180:33–43, 2019. doi:10.1016/j.compscitech.2019.05.013.

Y.C. Wang and Z.M. Huang. Bridging tensor with an imperfect interphase. Eur.J.Mech.
A/Solids, 56:73–91, 2016. doi:10.1016/j.euromechsol.2015.10.006.

Y.C. Wang and Z.M. Huang. Analytical micromechanics models for elastoplastic behavior
of long fibrous composites: A critical review and comparative study. Materials, 11:
ma11101919, 2018. doi:10.3390/ma11101919.

Y.U. Wang, Y.M. Jin, and A.G. Khachaturyan. Phase field microelasticity theory and
modelling of elastically and structurally inhomogeneous solid. J.Appl.Phys., 92:1351–
1360, 2002. doi:10.1063/1.1492859.

Z. Wang, R.J. Oelkers, K.C. Lee, and F.T. Fisher. Annular coated inclusion model and
applications for polymer nancomposites — Part II: Cylindrical inclusions. Mech.Mater.,
101:59–60, 2016. doi:10.1016/j.mechmat.2016.07.005.

Z.Y. Wang, H.T. Zhang, and Y.T. Chou. Characteristics of the elastic field of a rigid line
inhomogeneity. J.Appl.Mech., 52:818–822, 1985. doi:10.1115/1.3169152.

W.E. Warren and A.M. Kraynik. The nonlinear elastic behavior of open-cell foams.
J.Appl.Mech., 58:376–381, 1991. doi:10.1115/1.2897196.

D.F. Watt, X.Q. Xu, and D.J. Lloyd. Effects of particle morphology and spacing on
the strain fields in a plastically deforming matrix. Acta mater., 44:789–799, 1996.
doi:10.1016/1359-6454(95)00209-X.

D. Weaire and R. Phelan. A counter-example to Kelvin’s conjecture on minimal surfaces.

Phil.Mag.Lett., 69:107–110, 1994. doi:10.1080/09500839408241577.

G. Wei and S.F. Edwards. Effective elastic properties of composites of ellipsoids (I). Nearly
spherical inclusions. Physica A, 264:388–403, 1999. doi:10.1016/S0378-4371(98)00465-8.

212
D. Weidt and L. Figiel. Finite strain compressive behavior of CNT/epoxy nanocom-
posites: 2D versus 3D RVE-based modelling. Comput.Mater.Sci., 82:298–309, 2014.
doi:10.1016/j.commatsci.2013.10.001.

E. Weissenbek. Finite Element Modelling of Discontinuously Reinforced Metal Matrix

Composites. Reihe 18, Nr. 164. VDI–Verlag, Düsseldorf, 1994. ISBN 3-18-316418-3.

E. Weissenbek, H.J. Böhm, and F.G. Rammerstorfer. Micromechanical investigations of

arrangement effects in particle reinforced metal matrix composites. Comput.Mater.Sci.,
3:263–278, 1994. doi:10.1016/0927-0256(94)90141-4.

E. Weissenbek, H.E. Pettermann, and S. Suresh. Elasto-plastic deformation of com-

positionally graded metal-ceramic composites. Acta mater., 45:3401–3417, 1997.
doi:10.1016/S1359-6454(96)00403-X.

E.W. Weisstein. Sphere packing, 2000. URL http://mathworld.wolfram.com/SpherePacking.html.

G.J. Weng. The theoretical connection between Mori–Tanaka theory and

the Hashin–Shtrikman–Walpole bounds. Int.J.Engng.Sci., 28:1111–1120, 1990.
doi:10.1016/0020-7225(90)90111-U.

O. Wiener. Die Theorie des Mischkörpers für das Feld der stationären Strömung.
Abh.Math-Phys.Kl.Königl.Sächs.Ges.Wiss., 32:509–604, 1912.

J.R. Willis. The overall response of nonlinear composite media. Eur.J.Mech. A/Solids, 19:
S165–S184, 2000.

J.R. Willis. Bounds and self-consistent estimates for the overall moduli of anisotropic
composites. J.Mech.Phys.Sol., 25:185–202, 1977. doi:10.1016/0022-5096(77)90022-9.

J. Wippler, S. Fünfschilling, F. Fritzen, T. Böhlke, and M.F. Hoffmann. Homogeniza-

tion of the thermoelastic properties of silicon nitride. Acta mater., 59:6029–6038, 2011.
doi:10.1016/j.actamat.2011.06.011.

P.J. Withers. The determination of the elastic field of an ellipsoidal inclusion in a trans-
versely isotropic medium, and its relevance to composite materials. Phil.Mag., A59:
759–781, 1989. doi:10.1080/01418618908209819.

P.J. Withers, W.M. Stobbs, and O.B. Pedersen. The application of the Eshelby method of
internal stress determination to short fibre metal matrix composites. Acta metall., 37:
3061–3084, 1989. doi:10.1016/0001-6160(89)90341-6.

L.A. Wittig and D.H. Allen. Modeling the effect of oxidation on damage in
SiC/Ti-15-3 metal matrix composites. J.Engng.Mater.Technol., 116:421–427, 1994.
doi:10.1115/1.2904308.

L. Wu, L. Noels, L. Adam, and I. Doghri. A combined incremental–secant mean-field

homogenization scheme with per-phase residual strains for elasto-plastic composites.
Int.J.Plast., 51:80–102, 2013. doi:10.1016/j.ijplas.2013.06.006.

T.T. Wu. The effect of inclusion shape on the elastic moduli of a two-phase material.
Int.J.Sol.Struct., 2:1–8, 1966. doi:10.1016/0020-7683(66)90002-3.

213
J. Wulf, T. Steinkopff, and H. Fischmeister. FE-simulation of crack paths in the
real microstructure of an Al(6061)/SiC composite. Acta mater., 44:1765–1779, 1996.
doi:10.1016/1359-6454(95)00328-2.

Z. Xia, Y. Zhang, and F. Ellyin. A unified periodical boundary conditions for representative
volume elements of composites and applications. Int.J.Sol.Struct., 40:1907–1921, 2003.
doi:10.1016/S1359-6454(00)00317-7.

Z.H. Xia, W.A. Curtin, and P.W.M. Peters. Multiscale modeling of failure in metal matrix
composites. Acta mater., 49:273–287, 2001. doi:10.1016/S1359-6454(00)00317-7.

F.X. Xu and X. Chen. Stochastic homogenization of random elastic multi-phase composites

and size quantification of representative volume element. Mech.Mater., 41:174–186, 2009.
doi:10.1016/j.mechmat.2008.09.002.

Q.S. Yang, L. Tang, and H.R. Chen. Self-consistent finite element method: A new method
of predicting effective properties of inclusion media. Fin.Elem.Anal.Design, 17:247–257,
1994. doi:10.1016/0168-874X(94)90001-9.

Q.S. Yang, X. Tao, and H. Yang. A stepping scheme for predicting effective
properties of the multi-inclusion composites. Int.J.Engng.Sci., 45:997–1006, 2007.
doi:10.1016/j.ijengsci.2007.07.005.

Y.M. Yi, S.H. Park, and S.K. Youn. Asymptotic homogenization of viscoelastic
composites with periodic microstructures. Int.J.Sol.Struct., 35:2039–2055, 1998.
doi:10.1016/S0020-7683(97)00166-2.

P.G. Young, T.B.H. Beresford-West, S.R.L. Coward, B. Notarberardino, B. Walker, and

A. Abdul-Aziz. An efficient approach to converting three-dimensional image data into
highly accurate computational models. Phil.Trans.Roy.Soc.London, A366:3155–3173,
2008. doi:10.1098/rsta.2008.0090.

S. Youssef, E. Maire, and R. Gaertner. Finite element modelling of the actual structure
of cellular materials determined by X-ray tomography. Acta mater., 53:719–730, 2005.
doi:10.1016/j.actamat.2004.10.024.

W. Yu and T. Tang. Variational asymptotic method for unit cell homogeniza-

tion of periodically heterogeneous materials. Int.J.Sol.Struct., 44:3738–3755, 2007.
doi:10.1016/j.ijsolstr.2006.10.020.

J. Yvonnet. Computational Homogenization of Heterogeneous Materials with Finite Ele-

ments. Springer-Verlag, Cham, 2019. ISBN 978-3-030-18382-0.

J. Zangenberg and P. Brøndsted. Quantitative study on the statistical properties of fibre

architecture of genuine and numerical composite microstructures. Composites A, 47:
124–134, 2013. doi:10.1016/j.compositesa.2012.11.015.

A. Zaoui. Plasticité: Approches en champ moyen. In M. Bornert, T. Bretheau, and

P. Gilormini, editors, Homogénéisation en mécanique des matériaux 2. Comportements
non linéaires et problèmes ouverts, pages 17–44, Paris, 2001. Editions Hermès.

214
A. Zaoui. Continuum micromechanics: Survey. J.Engng.Mech., 128:808–816, 2002.
doi:10.1061/(ASCE)0733-9399(2002)128:8(808).

J. Zeman. Analysis of Composite Materials with Random Microstructure. PhD thesis,

Czech Technical University, Prague, 2003.

J. Zeman and M. Šejnoha. Numerical evaluation of effective elastic properties of

graphite fiber tow impregnated by polymer matrix. J.Mech.Phys.Sol., 49:69–90, 2001.
doi:10.1016/S0022-5096(00)00027-2.

J. Zeman and M. Šejnoha. From random microstructures to representa-

tive volume elements. Modell.Simul.Mater.Sci.Engng., 15:S325–S335, 2007.
doi:10.1088/0965-0393/15/4/S01.

C.M. Zener. Elasticity and Anelasticity of Metals. University of Chicago Press, Chicago,
IL, 1948. ISBN 0-226-98054-5.

H. Zhang, P. Sheng, J. Zhang, and Z. Ji. Realistic 3D modeling of concrete compos-

ites with randomly distributed aggregates by using aggregate expansion method. Con-
str.Buildg.Mater., 225:927–240, 2019. doi:10.1016/j.conbuildmat.2019.07.190.

J. Zhang and M. Ostoja-Starzewski. Mesoscale bounds in viscoelasticity of random com-

posites. Mech.Res.Comm., 68:98–104, 2015. doi:10.1016/j.mechrescom.2015.05.005.

R. Zhang and R. Guo. Modeling of progressive debonding of interphase–matrix interface

cracks in particle reinforced composites using VCFEM composites. Engng.Fract.Mech.,
248:107734, 2021. doi:10.1016/j.engfracmech.2021.107734.

X.X. Zhang, B.L. Xiao, H. Andrä, and Z.Y. Ma. Homogenization of the aver-
age thermo-elastoplastic properties of particle reinforced metal matrix composites:
The minimum representative volume element. Compos.Struct., 113:459–4681, 2014.
doi:10.1016/j.compstruct.2014.03.048.

Q.S. Zheng, Z.H. Zhao, and D.X. Du. Irreducible structure, symmetry and average
of Eshelby’s tensor fields in isotropic elasticity. J.Mech.Phys.Sol., 54:368–383, 2006.
doi:10.1016/j.jmps.2005.08.012.

S.J. Zhou and W.A. Curtin. Failure of fiber composites: A lattice Green function model.
Acta metall.mater., 43:3093–3104, 1995. doi:10.1016/0956-7151(95)00003-E.

H.X. Zhu, J.F. Knott, and N.J. Mills. Analysis of the elastic properties of
open-cell foams with tetrakaidecahedral cells. J.Mech.Phys.Sol., 45:319–343, 1997.
doi:10.1016/S0022-5096(96)00090-7.

R.W. Zimmerman. Hashin–Shtrikman bounds on the Poisson ratio of a composite material.

Mech.Res.Comm., 19:563–569, 1992. doi:10.1016/0093-6413(92)90085-O.

T.I. Zohdi. Constrained inverse formulations in random material design. Comput.Meth.

Appl.Mech.Engng., 192:3179–3194, 2003. doi:10.1016/S0045-7825(03)00345-1.

215
T.I. Zohdi and P. Wriggers. A model for simulating the deterioration of structural-scale
material responses of microheterogeneous solids. Comput.Meth.Appl.Mech.Engng., 190:
2803–2823, 2001. doi:10.1016/S0045-7825(00)00367-4.

T.I. Zohdi, J.T. Oden, and G.J. Rodin. Hierarchical modeling of het-
erogeneous bodies. Comput.Meth.Appl.Mech.Engng., 138:273–289, 1996.
doi:10.1016/S0045-7825(96)01106-1.

216