CISM COURSES AND LECTURES

Series Editors:

The Rectors of CISM

Sandor Kaliszky - Budapest
Mahir Sayir - Zurich
Wilhelm Schneider - Wien

The Secretary General of CISM

Giovanni Bianchi - Milan

Executive Editor
Carlo Tasso- Udine

The series presents lecture notes, monographs, edited works and

proceedings in the field of Mechanics, Engineering, Computer Science
and Applied Mathematics.
Purpose of the series is to make known in the international scientific
and technical community results obtained in some of the activities
organized by CISM, the International Centre for Mechanical Sciences.
INTERNATIONAL CENTRE FOR MECHANICAL SCIENCES

COURSES AND LECTURES - No . 377

CONTINUUM MICROMECHANICS

EDITED BY

P. SUQUET
LABORATOIRE DE MECANIQUE ET D'ACOUSTIQUE

~ Springer-Verlag Wien GmbH

 Le spese di stam pa di questo volume sono in parte coperte da
contributi del Consiglio Nazionale delle Ricerche.

This volume contains 73 illustrations

This work is subject to copyright.

AII rights are reserved,
whether the whole or part of the material is concerned
specifically those of translation, reprinting, re-use of illustrations,
broadcasting, reproduction by photocopying machine
Of similar means, and storage in data banks.
© 1997 by Springer-Verlag Wien
Originally published by Springer-Verlag Wien New York in 1997

In order to make this volume available as economically and as

rapidly as possible the authors' typescripts have been
reproduced in their original forms. This method unfortunately
has its typographical Iimitations but it is hoped that they in no
way distract the reader.

ISBN 978-3-211-82902-8 ISBN 978-3-7091-2662-2 (eBook)

DOI 10.1007/978-3-7091-2662-2
 PREFACE

The rapid growth of new materials, as well as the need for a deeper
understanding of classical materials, has put in evidence the importance of
small-scale mechanisms governing the overall behavior of materials. At the
same time that Continuum Mechanics reaches its limits of flexibility at the
macro level where introduction of internal variables is required to account for
microstructural effects, it finds new applications at the micro level where these
internal variables have a clear physical meaning. This application of
Continuwn Mechanics at the microlevel is called Continuum Micromechanics
as illustrated in the celebrated article of Hill which was published in 1965
and entitled "Continuwn micro-mechanics of elastoplastic polycrystals ".

This topic and closely related ones have received considerable attention
in the last forty years from different groups of mechanicians and applied
mathematicians. Although their initial motivations were different and covered
various areas such as the deformation of polycrystalline aggregates, the
effective behavior of composite materials and mathematical asymptotic
methods to give rigorous foundations to Micromechanics, it soon became
clear that all these different approaches could benefit from closer
interactions.

The present course was organized with the aim of presenting to advanced
students, researchers and research engineers working at the forefront of the
field some of the progress made during the past ten years. These include the
transformation analysis (G. Dvorak and Y. Bahei-El-Din), the role of
microvoids in ductile rupture and gradient effects (M. Gologanu, J.B.
Leblond, G. Perrin, J. Devaux), the effective behaviour of nonlinear
composites ( P. Ponte Castaiieda, P. Suquet) with applications to
metal-matrix composites (G. Dvorak and Y. Bahei-El-Din, P. Suquet),
microstructure evolution (P. Ponte Castaneda), dynamics of composites (J.
Willis), morphology and constitutive behaviour (A. Zaoui).

It is a pleasure to thank the lecturers and the 53 attendees who came

from 15 different countries and actively participated in the course. The
efficient help of the administrative staff of CISM is also gratefully
acknowledged.

P. Suquet
 CONTENTS

Page

Preface

Inelastic Composite Materials: Transformation Analysis and Experiments

by G.J. Dvorak and Y.A. Bahei-El-Din ............................................................. 1

Recent Extensions of Gurson's Model for Porous Ductile Metals

by M. Gologanu, 1-B. Leblond, G. Perrin and J. Devaux ................................ 61

Nonlinear Composite Materials: Effective Constitutive Behavior

and Microstructure Evolution
by P. Ponte Castaneda ................................................................................... 131

Effective Properties of Nonlinear Composites

by P. Suquet .................................................................................................... 197

Dynamics of Composites
by J.R. Willis ................................................................................................... 265

Structural Morphology and Constitutive Behaviour

of Micro heterogeneous Materials
by A. Zaoui ...................................................................................................... 291
 INELASTIC COMPOSITE MATERIALS:
TRANSFORMATION ANALYSIS AND EXPERIMENTS

G.J. Dvorak
Rensselaer Polytechnic Institute, Troy, NY, USA
A. Bahei-El-Din
Cairo University, Cairo, Egypt

ABSTRACT
A brief survey of micromechanical averaging procedures and bounds on overall
elastic moduli introduces the analysis of the effects of combined overall mechanical
loads and internal transformation fields, such as eigenstrains and eigenstresses, on the
overall response and internal stress fields in heterogeneous solids. The transformation
field analysis method relates the internally generated, piecewise uniform distributions
of eigenstrains or eigenstresses to the local fields through certain influence functions
that are related to the Green's function. However, their evaluation can be simplified
by connections with the mechanical influence functions and concentration factors.
The method of uniform fields is used to establish such connections for two-phase and
muti-phase systems. The transformation method is then extended to analysis of
inelastic composites, with elastic-plastic, viscoplastic, or viscoelastic phases.
Governing equations for the local fields are formulated and solved in illustrative
examples. Finally, experimentally detected yield surfaces and plastic strains of a
unidirectional boron-aluminum composite loaded along a complex plane-stress path
are interpreted with several models. The yield surfaces are well predicted by the
bimodal plasticity theory, and the plastic strains by a refined unit cell model.
2 G.J. Dvorak and Y.A. Bahei-El-Din

1. INTRODUCTION
1.1 Definitions and notation
Composite materials and polycrystals are, in general, heterogeneous solids
consisting of two or more constituents or constituent phases (r = 1, 2, ... N), bonded
at their interfaces. The constituent phases are homogeneous elastic solids, possibly
anisotropic, with certain elastic properties represented by coefficients of (6x6)
compliance or stiffness matrices Lr and Mr such that Lr Mr = I, where I is an identity
matrix. The elastic constitutive relations of the phases are written as,

(1.1)

Here, and in what follows, stresses and strains are written in the contracted notation,
as components of (6x1) vectors,

(1.2)

In a given Cartesian system of coordinates, the coefficients L~j and M~j are constant.
The phases may occupy simply or multiply connected volumes of various shapes.
Fibers, matrices, polycrystal grains or subgrains, coatings, interphases, etc., are
examples of constituent phases.
Volume fractions of the constituents are defined as cr = Vr/V, where Vr is the
volume of the phase (r) and V = E Vr is the total volume; E cr = 1. In
micromechanical models which evaluate distributions of local stress and strain fields
in the phases of the aggregate, the phase volumes may be divided into several
subvolumes. The above volume fraction definitions then apply to the subvolumes.
The standard assumption is that the composite aggregate is statistically
homogeneous, such that the volumes of each phase are of similar size and their spatial
distribution provides for constant phase volume fraction cr in any volume V >> Vr.
On the macroscale, such aggregates respond as homogeneous solids.
Representative volume element (RYE) of the statistically homogeneous
aggregate is defined such that (a) it contains a sufficient number of phase volumes or
subvolumes Vr to be of typical composition, and (b) under overall macroscopically
uniform stress, strain, or temperature change, its overall properties are not dependent
Transformation Analysis and Experiments 3

on its size [1]. Many RVE geometries and loading .conditions are used. Composite
cylinder or sphere assemblages, averaging model geometries, and various unit cell
models of periodic or partly periodic microstructures provide specific examples.
The typical purpose of micromechanical modeling is to evaluate overall
properties and internal stress and strain fields in the heterogeneous aggregate, in
terms of constituent properties, microstructural geometry and phase volume
fractions. Accordingly, the boundary conditions applied to the RVE are
homogeneous, and of two distinct types. For a RVE of volume V with surfaceS,
(a) Surface tractions tO(x), which are in equilibrium with a certain uniform overall
stress ell are prescribed on S. If ll(x) is a unit outside normal vector at S(x), then

t~(x) = a~. v.(x) XES. (1.3)

1 1J J

(b) Surface displacements uO(x), derived from a certain uniform overall strain field Eo
are prescribed on S(x). The displacement field is spatially linear,

u~(x) = f~. x. XES. (1.4)

1 1J J

Under either the (a) or (b) boundary conditions, the actual local fields in the
aggregate are not uniform. Their exact evaluation is often not practical, however,
useful approximations can be obtained in terms of volume averages of local fields,

(1.5)

0' = ~ I q (x) dV = ~ ~r qr = ~ cr qr (1.6)

v
f = ~I f (x) dV = ~ ~r fr = ~ Cr fr (1.7)
v
where u (x) and E (x) are the actual, variable local fields, u and E are the overall
volume averages, ur and fr are the volume averages in the constituents or in their
subvolumes. As long as the interfaces are perfectly bonded, all displacement fields
are continuous and strains are compatible; exceptions exist in damaged aggregates,
 4 G.J. Dvorak and Y.A. Bahei-El-Din

where displacement jumps may be found at internal cracks or debonds. In either

case, boundary and interface tractions are continuous and stresses are in equilibrium.
Suppose that a representative volume V of a heterogeneous medium with
phases r = 1, 2, ... N, is subjected to a uniform overall strain fo or stress ~, given in
(1.4) and (1.3) above. Then, the above local fields in the phases can be expressed in
terms of certain mechanical influence functions [1],

f (x) = A (x) fo u(x) = B (x) ~ (1.8)

Using the averaging relations (1.5), the phase volume averages are found as,

(1.9)

_:_ JB (x) ~ dV = (1.10)

Vr

the Ar and Br are the mechanical strain and stress concentration factors, (6x6)
matrices, not diagonally symmetric,

(1.11)

Since &rur = u and !:crfr = £, the concentration factors must satisfy

(1.12)

In a two-phase aggregate, r = 1, 2, there is c 1 A 1 + c 2 A 2 = I and c 1 B 1 + c 2 B 2 = I,

hence only one concentration factor is needed to find averages of the local fields.
An extensive body of literature exists on evaluation of the concentration
factors. Apart from numerical evaluations in unit-cell models of the microstructure,
estimates of these factors are usually derived for single inclusions embedded in a
certain surrounding medium. For ellipsoidal shapes, the solution can be constructed
as discussed below.
 Transformation Analysis and Experiments 5

2. ESTIMATES OF OVERALL ELASTIC PROPERTIES AND LOCAL FIELDS

2.1 General results

Since the aggregate in the RYE is statistically homogeneous, the overall
constitutive relations can be written as those for a homogeneous body. If surface
displacements or overall strain tO are prescribed,

(2.1)

If surface tractions or overall stress oC are prescribed,

(2.2)

where the L and M = L-l are overaU stiffness and compliance. The coefficients are
expressed in terms of overall or effective elastic moduli, according to the type of
material symmetry or anisotropy of the aggregate. To evaluate the overall stiffness L
in terms of the concentration factors, recall (1.1) and (1.9), prescribe an overall strain
tO, and write,

(2.3)

(2.4)

(2.5)

Similarly, if surface tractions or overall stress oC are prescribed, one finds that,

(2.6)

In most actual systems, the geometry of the microstructure is known only in

terms of volume fractions and approximate shape of the phases. Therefore, the
concentration factors and the overall stiffness and compliance can not be found
exactly. First-order estimates are often sufficient, and are provided, for example, by
the Mori-Tanaka [2, 3] and self-consistent [4] methods. Rigorous Hashin-8htrikman
bounds on Land M [5, 6] are also available.
 6 G.J. Dvorak and Y.A. Bahei-El-Din

In all such procedures, the phases are regarded as inhomogeneities of

ellipsoidal shape, embedded in a large volume of a certain surrounding elastic medium
L8m which is remotely loaded by certain uniform fields of the type (1.3) or (1.4). For
problems of this kind, Eshelby [7) established the important result that the strain
field in the inhomogeneity is also uniform. The concentration factors for ellipsoidal
inhomogeneities can be evaluated with Eshelby's equivalent inclusion method, or the
following procedure suggested by Hill [4).
Different L8m is chosen for each of the above estimates, but in any case, the
results can be reduced to the form,

(2.7)

The Lr and Mr are the elastic properties of the inhomogeneity and the L5m and M5 m
are those of the surrounding medium, the L* and M* are the stiffness and compliance
of the cavity containing the inhomogeneity. If the cavity wall is deformed by surface
displacements derived from a uniform strain field E*, then the surface tractions are in
equilibrium with a uniform stress o* such that o* =- L*E*. Conversely, if tractions
derived from a uniform field are applied at the cavity wall, then the surface
displacements follow from the uniform strain field E* =- M*o*.
Noting that the second parts of the rhs of (2.7) depend only on Lsm or M5 m,
one can introduce the tensors,

p = (L
*+ Lsm) -1 = p
T *
Q = (M + Msm)
-1
= Q
T
(2.8)

that depend only on the properties of the surrounding medium, and rewrite (2. 7) as,

(2.9)

(2.10)

One can show that P L + M Q = I, and that the connections with the Eshelby
tensorS of a transformed homogeneous inclusion in L5m are,

L8m(I- S) = L* S = Q S M5 m = (I - S) M* = P (2.11)
The problem of finding the estimates (2. 7) is thus reduced to evaluation of P or Q;
this has been done, for example, by Walpole [8) for typical shapes.
 Transformation Analysis and Experiments 7

In particular, for a sphere in an isotropic medium, the non-vanishing

coefficients of P are, in terms of elastic moduli of L5 m,

p 44 = p 55 = p 66 = TJ (2.12)

'1/J = (1 + v)/[9K(1 - v)] TJ = 2(4- 5v)/[15G(1- v)]

For a circular cylinder (prolate spheroid with long axis parallel to x 1) in a

transversely isotropic surrounding medium described by Hill's [9] moduli k, m and p,
the non-vanishing coefficients are:

P22 = Paa = (k + 4m)/~ P 44 = 4(k + 2m)/~

(2.13)
~ = 8m(k + m)

For a penny-shaped circular disk, with a normal parallel to x 1, surrounded by

transversely isotropic matrix with Hill's moduli p and n,

P 11 = 1/n (2.14)

Note that P depends on the aspect ratios of the ellipsoid, and on the elastic moduli of
the surrounding medium, which were used in (2.12) to (2.14). Since P is often
singular, (2.7) are best evaluated from (2.10).
Using (2.7) in (2.5) and (2.6), one can establish that the estimates of the
overall stiffness L and compliance M can be evaluated using the expressions [6, 10],

-l
N

L = [L/r (L* + Lr) 1

- 1* (2.15)
r =1
N

M = .1-J cr (M * + Mr) -1] -1- M*

[~ (2.16)
r =1

where cr are the phase volume fractions, Lr and Mr are phase elastic stiffness and
compliance, and the L* and M* are Hill's constraint tensors for the surrounding
medium containing the inhomogeneity Lr.
 8 G.J. Dvorak and Y.A. Bahei-El-Din

2.2 Estimates and bounds

The different estimates of L and M are arrived at by choosing different
properties of the surrounding medium. In particular, Dvorak and Benveniste [10]
show that the Mon- Tanaka estimate of the the Land M is obtained if,

(2.17)

in (2.7)- (2.16), where L 1 is the stiffness of the matrix. In the original formulation,
intended for matrix-based composites, the Mori-Tanaka method approximates
interaction between the phases by assuming that each reinforcing phase is embedded,
in turn, in a large volume of matrix that is remotely loaded by the average matrix
stress. The matrix stress is determined by the method in terms of the overall applied
stress. The results can be reduced to those given by (2.7) and (2.15), (2.16), with the
choice (2.17). Since L1 is known and usually isotropic, the above relations provide
explicit expressions for the concentration factors and overall properties. Specific
evaluations are available for many systems, e. g., in [3, 11].
The self-consistent method approximates interaction between the phases by
assuming that each phase is embedded, in turn, in a large volume of an effective
medium with elastic properties L and M of the composite. Remote loading is equal to
the applied overall stress or strain. The choice of the surrounding medium is then,

(2.18)

Since the overall properties are not initially known, (2.15) and (2.16) provide a
system of implicit equations for L or M. Again, specific solutions have been derived
for many systems [4, 8].
Walpole [6] derived the expressions (2.15) and (2.16) for evaluation of the
Hashin-Shtrikman bounds on the actual stiffness Lac or compliance Mac of composite
aggregates. The properties of the surrounding medium are now chosen as,

(2.19)

where L0 is a certain homogeneous "comparison medium" selected from the

coefficients of Lr. Approximate choices are La ~ Lr for the upper bound on Lac• or
La S Lr for the lower bound on on Lac• where the known phase properties Lr are used.
 Transformation Analysis and Experiments 9

More specifically, Walpole proves that the overall actual stiffness Lac and
compliance Mac are bracketed as follows :
If (L 0 - Lr) is for all r positive (negative) semi definite, then so is (L- Lac)·
If (M 0 - Mr) is for all r positive (negative) semi definite, then so is (M- Mac)·
Implications of these results for bounding the ordered eigenvalues, and the diagonal
and off-diagonal terms of Lac or Mac were discussed by Teply and Dvorak [12].
Comparing these results with those found by the Marl-Tanaka method using (2.17),
one can see that in systems where the matrix r = 1 is the weaker phase such that
(L 1 - Lr) is for all r positive semi definite, the M-T estimate of L is equal to the
lower bound on Lac· Analogous relations exist for the compliances.
Hashin [13], Hashin and Rosen [14] and Walpole [6, 8] give specific bounds for
common compositions, these are the best bounds obtainable in terms of volume
fractions and elastic moduli of the phases. Higher-order bounds that incorporate
additional information about microstructural geometry are now available [15]. The
above estimates and bounds do satisfy the consistency condition L M = I, and, as can
be readily seen from (2.15) and (2.16), are diagonally symmetric. However, as noted
by Benveniste, Dvorak and Chen [16], the diagonal symmetry of the Marl-Tanaka
and self-consistent estimates obtains only if a single L* or M* is admitted.
Therefore, any inclusion shape and alignment are permitted in a two-phase system.
In systems with more than two phases, all inclusions must have the same shape and
alignment. However, diagonally symmetric estimates can also be derived for
aggregates with randomly oriented anisotropic inhomogeneities [11].

3. TRANSFORMATION FIELDS
3.1 Definitions and examples
Internal stress fields and overall strains of heterogeneous media can be caused
not only by mechanical loads, but also by internal deformations of the phases.
Thermal strains involving volume changes, or phase transformations that produce
both volume and shape changes in phases, as well as inelastic strains, are examples of
such deformations. If the transforming homogeneous phase were separated from the
aggregate, then it would deform freely, without any contribution from any
mechanical loads, and therefore, free of stress. However, due to the constraints
10 G.J. Dvorak and Y.A. Bahei-El-Din

imposed by the surrounding material, internal stresses and deformations are

generated in phases and the surrounding aggregate; these are usually referred to as
the residual fields. The stress-free strains that cause the residual fields are called
eigenstrains or transformation strains.
Consider first a volume V of some convenient shape, containing a
homogeneous elastic material of stiffness Land compliance M = L-t. Let the external
boundary S be free of constraints and loaded by an overall stress u on S. In addition,
a stress-free transformation strain 11 is introduced uniformly through V. If the total
strains are small and subject to an additive decomposition, then the constitutive
relation can be written as,

t=Mu+l1 (3.1)

·Thus, if u = 0, then t = 11 is the total strain.

Next, consider again the above transformed volume V, under displacement
boundary conditions consistent with some uniform overall strain t. The constitutive
relation now reads,

u= L t +~ (3.2)

Under rigid constraint, t = 0, the total stress u = ~, which is referred to as the

eigenstress, a deformation-free stress. From (3.1) and (3.2) one finds,

11=-M..\ (3.3)

The volume V can be made as small as desired, hence the elastic constitutive
relations that account for local transformation fields can be written as,

t(:x) = M(:x) o(:x) + J1{:x) o(:x) = L(:x) t(:x) + ~(:x) (3.4)

where the rhs o(:x) and t(:x) are superpositions of the mechanical fields and
transformation fields. In the absence of applied mechanical loads, the mechanical
fields are generated only by the transformations, and (3.4) are then the residual
fields.
Transformation Analysis and Experiments 11

The transformation fields can be classified as

(a): Physically activated fields of known magnitude, such as uncoupled thermal

strains and phase transformations. Volume changes due moisture absorption are
sometimes included, provided of course, that they are regarded as independent of the
local hydrostatic stress.
(b): Physically activated, but of initially unknown magnitude. Inelastic strains
resulting from the additive decomposition tin = E - M u are an example of
transformation strains that depend on past loading history. Coupled
electromechanical deformations, or strains caused by shape memory effects are
additional examples of transformation fields that depend, often in a complex manner,
on evolution of other physical fields present in the material.
(c): Equivalent eigenstrain fields are artificial fields introduced to simulate certain
desirable effects. Eshelby's [7) equivalent strain applied to a homogeneous solid to
represent an inhomogeneity in the equivalent inclusion method is and example, and
so are the uniform fields [17) discussed briefly below.
In small strain approximations, the transformation fields can be additively
superimposed,

p{x) = ~1 (x) + J.&ix) + ... ~(x) (3.5)

l(x) = l 1(x) + l 2(x) + ... llf(x) (3.6)

3.2 Transformation field analysis

In the presence of transformation fields, the local and overall constitutive
relations (1.1) and (2.1), (2.2) of the composite aggregate must be expanded to
conform with (3.1) and (3.2). The local fields in a phase volume or subvolume Vr of a
phase are then written as,

(3.7)

where both the mechanical and transformation fields are volume averages over Vr'
and therefore uniform. The definition of the representative volume is expanded to
specify that the local transformation fields in the RVE are distributed such as to
cause only uniform overall eigenstrain or eigenstress.
 12 G.J. Dvorak and Y.A. Bahei-El-Din

The overall constitutive relations then become,

(3.8)

where the overall I' and ~ are produced by the transformation and residual fields.
The connection between the local and overall quantities follows from Levin's [18]
formula, originally derived for the thermal eigenstrains and generalized in [10]. For
the actual fields in phase volumes VP the results are,

Jl =~ fB~(:1:) ~~p(x) dV ~ = ~ fA~(x) ~p(x) dV (3.9)

vP vP
For the case of piecewise uniform eigenstrains in phase volumes or subvolumes V r'

(3.10)

where ~ and Br are the mechanical concentration factor tensors, evaluated as

volume averages in Vr of the influence functions in (1.8).
To evaluate the residual fields caused by the transformations, consider first
the representation of the latter by continuous functions. The result is,

E{x) = f 0(x)-
v
f (3.11)
r(x, x' )[(L (x')- L 0) f(x') - L(x') 1-'(x' )] dx'

0
where f (x) denotes the strain field that would exist in a comparison homogeneous
medium LO under the same boundary conditions, a.nd

rijkl (x, x') = -!(Gikj, 1(x, x') + Gjk ,n(x, x')) (3.12)

0
where Gik is the Green's function of the homogeneous medium L that satisfies

qjkl Gkp ,lj(x, x') + 6ip 6(x- x') = 0, x, x' E V, (3.13)

where 6ip is the Kronecker symbol, and o(x- x') is the Dirac delta function.
Transformation Analysis and Experiments 13

In actual solutions, the field Eix) and the eigenstrain field J.&r(x) are replaced
by piecewise uniform approximations in the phases or in phase subvolumes. Then,
(3.11) is reduced to a system of linear algebraic equations for the local average
strains. Berveiller, et al., [19] and Walker, et al., [20], among others, had followed
such procedures.
The approach adopted here starts with the assumption that the
transformation fields are represented by distributions which are piecewise uniform,
either in the phases, or in local volumes Vr within those phases. The total local
strain caused by the uniform overall strain E or stress u, and a piecewise uniform
distribution of eigenstrains l's or eigenstress ~ 5 , both within Vr and in other volumes
V5 is sought in the following form [10, 17],
N

Er(x) = Ar(x) E + L Dr (x) l's

s =1
5 (3.14)
I; s = 1, 2, ... N ,
N

ur(x) = Br(x) u + L
s =I
Fr5(x) ~s {3.15)

The Drr(x) and Dr5 (x) are, respectively, the self-induced and transmitted eigenstrain
influence functions; the Frr(x) and Fr5 (x) are the corresponding eigenstress influence
functions. In analogy with (1.9) and (1.10), the local volume averages in Vr of these
functions may be referred to as the eigenstress or eigenstrain concentration factor
tensors. Collectively, these will be called the transformation influence functions and
concentration factors. The coefficients of these tensor functions are dimensionless.
The transformation fields that may be admitted must produce a uniform
overall body force ~ or eigenstrain I' in the representative volume. This is always the
case for each single component of l"s in a subelement of a properly defined unit cell,
and it is also assured in the evaluation of the self-<:onsistent and Mori-Tanaka
estimates of the transformation factor tensors. However, an eigenstrain distribution
corresponding, for example, to a temperature gradient in V would not be admissible.
In terms of the integral equation formulation of the problem, consider any
eigenstrain field J.'(x) in a representative volume of a heterogeneous medium under
zero overall strain, and define the transformation strain influence function D(x, x') as
14 G.J. Dvorak and Y.A. Bahei-El-Din

E(x) = D(x, x') * p.(x') (3.16)

Its evaluation then follows from (3.11) as

D(x, x') * p.(x') = - f r(x, x') [(L(x') - 1°) D(x~ x') - L(x' )]1-' (x') dx'
v
(3.17)
Note that for a transformed homogeneous inclusion of ellipsoidal shape, D55 = S, the
Eshel by tensor.
For any micromechanical model adopted, the transformation influence
functions must satisfy certain general connections, derived by Dvorak and Benveniste
[10],
I I

L D r(x) = I- A (x)
r • 1
5 5 L F r(x) = I- B (x)
r =1
5 8 (3.18)

I I

L
r • 1
D6r(x) Mr = 0 L F r(x) Lr = 0
r=1
6 (3.19)

(3.20)

L
r =1
cr Frs =0 (3.21)

with r, s = 1, 2, ... N everywhere. All these relations are exact, but (3.19) and (3.20)
provide (3.21), hence only (3.18) and one of the remaining three are independent.
Therefore, there are only (2 x N) independent relations for the (N x N) unknown
transformation influence functions. Thus one can solve the system and find exact
expressions for the transformation functions in terms of their mechanical counterparts
only in two-phase materials.
 Transformation Analysis and Experiments 15

3.3 Elements of uniform field theory

A useful technique for analysis of transformation fields is the theory of uniform
fields in heterogeneous media [17). Suppose that the initially stress free
representative volume V of a multiphase medium, has been loaded by a certain
uniform overall stress u = ~. The goal is to modify the nonuniform local fields by
superposition with certain auxiliary eigenstrains, introduced in the local volumes such
that a uniform stress field ur = ~, together with a uniform strain field are created
everywhere in V.

The problem can be solved in the following way. The uniform stress u0 is
prescribed in all local volumes to create a piecewise uniform but incompatible strain
field. Compatibility is restored by superposition of a piecewise uniform eigenstrain
field l'r which makes the strains uniform everywhere in V; stress equilibrium is
already guaranteed by the prescribed uniform stress. The local strains thus become

(3.22)
r = 1, 2, ... N.
Similarly, if a uniform overall strain field £0 is prescribed at S, an eigenstress
field is sought such that its superposition with the stress field caused by Eo will lead
to uniform stress and strain fields in V. The strain Eo is applied to all phases, and a
piecewise uniform eigenstress field .\r is added to make the stresses uniform in V.

(3.23)
r = 1, 2, ... N.
One solution of these equations that will be useful in §3.4 can be found as follows.
The aggregate is initially stress free, but uniform eigenstrains are introduced in two
local volumes r = p, q, as 11p = ~~po, and JSq = Pl Uniform eigenstresses .\p = lpo,
and lq = .\qO may be applied instead. The goal is to find the overall stress u or
~ ~

strain f that need to be imposed on S, and the eigenstrains l'r or eigenstresses .\r in
the remaining local volumes such that the local fields become uniform.
 16 G.J. Dvorak and Y.A. Bahei-EI-Din

The solution is
A A A

lit= P.qo- (Mr- Mq) u = Jl.po- (Mr- Mp) u

(3.24)

= Er = Mr + r-r = MP +
A

= Mq + r-q•0
A A A A

E (1 IL (1 IL-. 0
r-p (1 IL

and

= Aq 0 -
A

= Ap 0 -
A A

Ar (Lr - Lq) E (Lr - Lp) E

(3.25)
A A

= Ur = Lr E + Ar = Lp E + Ap 0 = Lq E + Aq 0•
A A A

Since by definition, Ap 0 = - Lp Jl.p 0 and Aq 0 = - Lq P.q 0, one can show that the two
fields are identical.

3.4 Estimates of transformation concentration factors

Here we illustrate the application of the uniform field theory in a
self-consistent evaluation of the concentration factors D55 and D5 r in (3.14). The
self--consistent method estimates of the transformation concentration factor tensors
from the solution of a dilute problem in which the inclusion 1 5 is bonded to a large
volume of the effective medium L. An overall uniform strain t: is prescribed at
infinity, together with the uniform eigenstrain lls in V5 • The expressions for the D55
and D5r are sought in terms of A5 , and are found by creating first in the entire
A

volume the uniform strain field t: suggested by (3.25 2),

(3.26)
A

Next, the overall strain is reduced from t: to the prescribed magnitude t:. Starting
with this uniform strain field, one applies an overall strain - t: and then the
prescribed strain E. This yields the estimate of the local strain,

(3.27)
Transformation Analysis and Experiments 17

Finally, using {3.3) and {3.10), one finds,

(3.28)

A comparison with {3.14) then provides the eigenstrain concentration factors,

{3.29)

A simple rearrangement, and an analogous derivation of Fsr yield,

{3.30)

where 65 r is the Kronecker symbol, but no summation is indicated by repeated

subscripts. Some algebra confirms that the results satisfy the connections {3.18) to
(3.21). Dvorak and Benveniste [10] show that the above results are valid both for the
self-eonsistent and Mori-Tanaka estimates, providing that the concentration factors
~ and Br, and the overall L and M are estimated by the same method.
For two-phase systems, r = a, {3, the uniform field theory provides these
results for the actual influence functions [10, 17],

{3.31)
Fr0 {x) = {I- Br{x)){M 0 - ~)-)fa
-1
Fr(i(x) = (I- Br(x)){M0 - ~) ~

No estimates were used in this derivation, hence these connections between

mechanical and transformation influence functions in two-phase systems are exact.
 18 G.J. Dvorak and Y.A. Bahei-El-Din

3.5 Influence functions for unit cell models

The above estimates of the transformation concentration factors of individual
phases were constructed for uniform eigenstrains and/or eigenstress in each phase or
inclusion. Such uniform fields may be caused by temperature changes or by phase
transformations. Of course, the eigenstrains are not uniform if they depend on the
past or current stress state, as they do, for example, during inelastic deformation.
The above estimates are then likely to prove inaccurate, unless the phases are
subdivided so that the local transformation and total fields can be replaced by
sufficiently refined piecewise constant approximations.
Unit cell model of the composite material are typically used is such
applications. If a periodic arrangement of the microstructure is selected, then the
unit cell is subjected to appropriate periodic boundary conditions, and the local fields
are evaluated by the finite element method. The domain under consideration is
subdivided by constant strain 3D elements 0'11, 1J = 1, 2, ... M, interconnected at
nodes i, j, ... = 1, 2, ... R. Many elements may contain the same phase material so
that their elastic properties will be identical, however, the local eigenstrains may be
different in each element.
Coefficients of the mechanical strain or stress concentration factor tensors ~
or B'll in (1.8) are found from solutions of six successive elasticity problems, where
overall strain E, or stress u, that have only one nonzero component of unit magnitude
are prescribed in turn; the eigenst.rains vanish in the entire domain. The ( 6x 1) strain
or stress vector found in the element n'll is the column of the (6x6) matrix~ or B'll,
corresponding to the selected nonzero component. The transformation concentration
factors are obtained in a similar way, by applying in turn a single component of the
eigenstrain J1p of unit magnitude in element nP, within a domain subjected to zero
overall strain and zero eigenstrains in other elements. The local strains found in nP
and 0'11 represent one column of the (6x6) matrices DPP and D'llP in those elements.
The unit local eigenstrains applied in such evaluations may be generated using any
suitable program, as unit thermal strains, by an appropriate selection of nonzero
thermal coefficients and temperature change.
A more efficient evaluation of these matrices is possible using the initial strain
formulation [21]. The displacement field within 0'11 is approximated by,

(3.32)
 Transformation Analysis and Experiments 19

where NTJ = [Ni N~ .. ] and ~ = [ai a~ .. ]T denote certain prescribed shape functions
and nodal displacements of the element OTJ. The strains at any point within OTJ can
be found as,

(3.33)
T] T]
where S is the kinematic linear differential operator. The Ni, Nj, .. , etc., shape
functions must be linear in the cartesian coordinates so that element strain field is
uniform; the coefficient matrix ~ is constant.
The stress uTJ in the element and the tractions on the element surface are
reproduced by subjecting the element to the nodal force statically equivalent to the
element stresses and boundary tractions, found from the principle of virtual work. In
the absence of body forces, the equivalent nodal force is given by the volume integral,

~ = f A~ uTJ dOTJ {3.34)

n
T]

where qTJ = [qi q~ .l.

For the constant stress and strain fields admitted in the
elements, ~is constant, and the integral can be evaluated in closed form. Using the
constitutive equation (3.7) and the strain-displacement relation given in (3.33), we
one write the equivalent nodal forces associated with node i of element nTJ,

(3.35)

{3.36)

~ =- f[AiJT ~ ~ dOTJ {3.37)

n
T]

where P is the total number of element nodes, Kij is the ij-partition of the element
stiffness matrix, and fi represents the forces at node i caused by the uniform
eigenstrains ~·
20 G.J. Dvorak and Y.A. Bahei-El-Din

The sum of the forces at a generic node i contributed by the elements

connected to that node, and by the loads Pi directly connected to that node must
vanish; at each such node,

Lq~ =
M

Pi- 0 i = 1, 2, .. R (3.38)
, =1

where M is the total number of elements, and R is the total number of nodes in the
domain. From (3.35) and (3.38),
M p

LL Kij a~ i =I, 2, .. R (3.39)

TJ=1j=1

These R matrix equations can be augmented and written as,

L
M

Ka=f Kij = Kij (3.40)

,=1 Tj=1

where K is the overall stiffness matrix; in the absence of large temperature changes,
its coefficients are constant. The overall load vector f is, in general, a function of
time and must be updated, as required by the ODE solver used in evaluation of the
nodal displacements llj_ ( i = 1, 2, .. R) and the stress and strain in the elements.
The transformation factors DPTJ (p,1J = 1,2, .. M) can now be evaluated from
the overall displacements found by solving (3.40) for 6M independent f vectors; each
corresponds to an eigenstrain l4rJ = ik, k = 1, 2, .. 6; 1J = 1, 2, .. M, where ik is the kth
column of the (6x6) identity matrix. Recall that according to the definition (3.14) of
DP"l , its evaluation from the local strain field requires that overall strains vanish.
Consequently, (3.40 1) must be subjected to displacement boundary conditions which
are derived from a null overall strain vector.
The overall stiffness matrix K can be evaluated for the unit cell domain n, and
then decomposed to lower and upper triangular factors, KL and Ku, using Cholesky's
method; since K is symmetric, Ku = Kl. The overall nodal displacements follow
from forw~d and backward solutions. The element nodal displacements ~ are
extracted from the solution and substituted in (3.33) to obtain the strains E1] which
represent, respectively, the columns of DPTJ' p,TJ = 1,2, .. M.
Transformation Analysis and Experiments 21

Let matrix P p. relate the nodal displacements of element p to the overall

displacement vector such that,

ap = p p alil p = 1, 2, .. M i = 1, 2, .. 6M (3.41)

From (3.33),

E
p
=Ap Pp alil (3.42)

For a specific vector f which corresponds to unit eigenstrain ~ = ik, k = 1, 2, .. 6, the

strain represents the kth column d~11 of the DP,r This can be written symbolically as,

(3.43)

The stress transformation factors F11 P follow from a similar procedure, but for f
vectors which correspond to 6M unit eigenstress vectors ~ = ik, k = 1, 2, .. 6. Since
~ = - ~~, (3.37) provides,

1J = 1, 2 ... M k = 1, 2 ... 6 (3.44)

where Aik is the kth column of matrix [Aif According to (3.15), the overall stress
inust vanish when evaluating the stress transformation factors from the local stress

field. Symbolically, the kth column ~11 , of the F PTJ factor corresponding to~ = ik can
be written as,

(3.45)

where t5PTJ is the Kronecker symbol, and f is assembled from the load vectors in (3.44)
using (3.40 3).
Additional details regarding the unit cell influence functions, and the efficiency
of their evaluation, can be found in [21].
 22 G.J. Dvorak and Y.A. Bahei-El-Din

3.6 Periodic hexagonal array model

In unit cell models which approximate the actual material geometry with a
periodic distribution of fibers or inclusions in a matrix, there are certain internal
symmetries that facilitate evaluation of the transformation factors. For example, the
periodic hexagonal array (PHA) model [12, 22], shown schematically in Figs. 1-3 has
a 3-fold axis of rotational symmetry, parallel to the fiber, that coincides with the
cartesian coordinate direction x~, and three planes of symmetry perpendicular to the
transverse x 2x 3-plane. This suggests a subdivision of the unit cell into six similar
triangular subdomains A, B, ... F, Fig. 3. The geometry of the finite element mesh in
each of these subdomains is selected to be invariant under 120° rotations about the
x 1 :: x1 :: x1'-axis, and under reflections about the x 2-axis, the x2-axis, and the
x2 '-axis. Such selection does not rule out elements that cross the symmetry axes,
c. f., Figs. 2, 3.
Such periodic models imply that all unit cells are loaded by identical
eigenstrains or eigenstresses under given overall loading conditions. We recall that
the eigenstrain factor Djl'T] is evaluated as the strain field fp in all subelements OP E 0,
p = 1, 2, ... M, caused by a unit eigenstrain ~ = ik, k = 1, 2, .. 6, prescribed in any
subelement 0 11 E n, 1J = 1, 2, ... M, while the composite material is subjected to zero
overall strain. The eigenstress concentration factors F P'l are found for similarly
specified subelement eigenstresses in a composite subjected to zero overall tractions.
These boundary conditions can be reproduced on the unit cell of the PHA model.
With the symmetries indicated above, both the mesh and the boundary
conditions remain invariant under the above described rotations and reflections, and
it is then possible to show that the Djl'T] and F jl'T] matrices found for element
eigenstrains or eigenstresses prescribed within only one subdomain of the unit cell can
be transformed into concentration factor matrices for eigenstrains or eigenstresses
specified in any other subdomain.
For the Djl'T] matrices, where p, 1J = 1, 2, ... M refers to all subelements Min n,
we denote subsets of the complete set of the DP11 matrices as Dpa> DPf3' DP"f' DP 5, DPE
and DPIP" These are found for unit eigenstrains introduced in the subdomains A, B, ...
F in Fig. 3. The one subset that needs to be constructed is selected here as the Dpo.>
for unit eigenstrains 1-'a. = ik, a= 1, 2, .. N, k = 1, 2, .. 6 in the subdomain A, defined
in the coordinate system xj, j = 1, 2, 3; N denotes the number of subelements in A.
Transformation Analysis and Experiments 23

The strain vector fP due to the single component l'o. = ik of a prescribed unit

eigenstrain component represents the kth column d~o. of the matrix Dpo.·
Consider now the subdomain D in the local coordinates xj, and locate the
subelement volume n 6 E D that has the same position relative to the system xj as the
volume no. E A has in the coordinates xj. Introduce in n 6 a uniform unit eigenstrain
vector 1'6 = ik, k = 1, 2, .. 6, defined in the xj coordinates, to produce local strains
DP6 1'6 in each of the subelements nP E n, p = 1, 2, .. M. The applied unit eigenstrain
1'6 transforms into the eigenstrain 1'6 in the xj-<:oordinates as,

(3.46)

where 'IIi is the transformation matrix, and th_ is its kth column. For the 120° cw
rotation implied in Fig. 3, and strain vectors defined as in (1.2), 'IIi is given by,

1 0 0 0 0 0
0 1/4 3/4-./3/4 0 0

•=
0 3/4 1/4 /J/4 0 0
0 /J/2 -./3/2 -1/2 0 0 (3.47)
0 0 0 0 -1/2 -./3/2
0 0 0 0 /J/2 -1/2

The three--fold rotational symmetry axis x 1 implies that local unit eigenstrains
l'o. and l's introduced in volumes no. and n 6 cause local strain fields in the respective
xj and x j coordinate systems that are invariant under the coordinate transformation

from the xj system to xj system. Therefore, the strain f~' in the xj-<:oordinates in
any selected volume nP, caused by the eigenstrain (3.46), can be found as,

(3.48)

where n<. is the subelement volume defined as that with the same position relative to
the xj coordinate system as the selected volume nP has in the x j coordinate system.
Then, the kth column d~ 6 of the influence coefficient DP6, which relates the strain in
volume OP, p = 1, 2, .. M, to the unit eigenstrain ~prescribed in 0 0, is given by the
24 G.J. Dvorak and Y.A. Bahei-El-Din

""....
Q)

....u::l
::l 'N
........ ......
0""
....
u ""....

- :g
~
"§ t;:l
Q)
~
'"C u
s
0

< =
=
:::l
>.
ll..
u
~
=
Q)

....
...c::
..... 0
bD
0 Cl!
uQ)
""
.9=
....
'"C
0
'"C
u '"C
=
Q)

"" Cl!
""""0.... ~
u
Q)
=
0
bD
til Cl!
!il Q)><
>
"" ...c::
= ....
Cl! ...c::
.... -~
E-1

,.....;
Transformation Analysis and Experiments 25

2. Geometry and supports of the unit cell of the PHA model [12].
26 G.J. Dvorak and Y.A. Bahei-El-Din

3. Subdivision of the unit cell into six subdomains [21].

Transformation Analysis and Experiments 27

transformation of the strain vector (3.48) from the Xj to the xr-coordinates. This
can be written as

k = 1, 2, .. 6 (3.49)

and the transformation concentration factor matrix DP11 then follows from

a, 6 = 1, 2, .. N; p, ( = 1, 2, .. M (3.50)

Recall that regardless of the method used to determine the transformation

concentration factors, the result must satisfy the general connections (3.18-3.21).
These may provide a useful check on the accuracy of the numerical results.

4. INELASTIC COMPOSITE MATERIALS

4.1 Governing equations

Here we show how the transformation concentration factors for a selected
micromechanical model can be used to formulate a system of governing equations for
evaluation of the inelastic local fields. Suppose that a certain constitutive relation
has been selected for evaluation of the inelastic strains in the subvolumes 0"1 of the
inelastic phase or phases, which can be regarded, for example, as elastic-plastic,
viscoplastic or viscoelastic For piecewise uniform fields in the local volumes, one can
formally write such constitutive relations in the general form,

(4.1)

where the elastic strain and stress are connected as in (1.1), and,

~n(t) = f(u"l(t-r),O(t-r)) (4.2)

~e(t) = g(~(t-r), O(t-r)) (4.3)

where ~n( t) is the inelastic strain, written here as a function of the past stress

history u"l(t- r) and temperature history 0 (t- r); the ~e(t) is the corresponding
28 G.J. Dvorak and Y.A. Bahei-El-Din

relaxation stress that accumulates under prescribed strain history E.q(t- r) as well as
the temperature history and 0 (t- r).
Observing the formal similarity of (4.1) and (3.7), including the connections

(3.3), ,r,;e(t) =- ~ ~n(t), ~n(t) =- ~ cr,;e(t), one can substitute into (3.14) and
(3.15) to find the governing equations for the total local fields [23],
y

Ep(t) + I.. ~
Dp1J ~ g (E.q(t- r), O(t- r)) = 0
APE (t) + 0
aP 0 (t) ( 4.4)
fJ=l

L
y

up(t) + FPTJ ~ f(uTJ(t- r), O(t- r)) = BP u 0(t) + bP 8°(t) (4.5)

fJ=l

where the thermal concentration factors are evaluated as,

y y

aP = L Dp1J ~ = - L DPTJ ~ lq (4.6)

fJ=l fJ=l
y y

bp = I, FPTJ ~ = - I, Fp1J ~ ~ (4.7)

fJ=l fJ=l

Recall that the mechanical and transformation concentration factors depend

only on elastic phase moduli and on local geometry. If those remain constant, the
governing equations can be differentiated and used for evaluation of stress and strain
increments, or their time rates. In this manner, the inelastic deformation problem for
any heterogeneous medium is reduced to evaluation of the various concentration
factor tensors or matrices, and to solution of one of the equations (4.4) or (4.5), and
their rate equivalents.

4.2 Elastic-plastic phases

The elastic-plastic constitutive relation can be represented as,

dur = ~r dEr + ~r dO dEr = ~ r + «r dO (4.8)

where ~rand ~rare the instantaneous stiffness and compliance tensors, and /r,
Transformation Analysis and Experiments 29

•r are the thermal stress and strain tensors that typically describe the consequence
of a variation of yield stress with temperature. All these tensors depend in some
given form on the past deformation history, hence in an actual material the
magnitudes of their coefficients will vary within each local volume V r or 0 11 • To
prevent large errors in evaluation of the local response, it is advisable to choose
material models that permit refined subdivisions of the representative volume.
Decomposing (4.8) according to (4.1),

(4.9)

where

(4.10)

and IDr is a (6x1) thermal strain vector of linear coefficients of thermal expansion
with lr =- Lr IDr· Substituting (4.9) into (4.4) and (4.5), one finds the following two
systems of equations for local fields in elastic-plastic heterogeneous media,
M M

dfP + I,nP1'J M, ~,P d~ = AP df + (ap- I,nP11 M, ~P) dO

1J=1 1J=1
(4.11)
M M

dup + LF P1'J L, .6~ du

1J=1
11 = BP du + ( bp- ll
1J=1
P11 1, -~ ) dO

where the df or du and dO are parts of the prescribed overall loading path. In actual
numerical solutions, one may reduce these to the matrix forms,

{dfp} = (diag (I)+ [DP11 M, ~11 PJf{[Ap] df- {[DP11 M,]{~ + / 11 P} dO}
(4.12)

{dup} = (diag (I)+ [FP11 1, A~]r{[Bp] du-{[FP11 L,]{m, +-~}dO}

 30 G.J. Dvorak and Y.A. Bahei-El-Din

If one of these systems is solved for the local fields, then the overall response of the
representative volume V follows from (1.5) to (1.7) and (3.9) with (3.8).
Closed form solutions can be found for coarse subdivisions of the
microstructure; for example, in a two-phase system r = o:, {3, one can find the
solution of the above equations in the form [23],
0 0
dEr= ...lrdE + ardO (4.13)

where, for r = o:,

(4.14)

(4.15)

...4'13 , a 13 , etc., follow from an exchange of subscripts.

Once the local fields are known, the overall instantaneous response is described
in analogy with (3.1), (3.2) and (2.5), (2.6) as,

du = .z'dE +£dO, dE= ~du+ «dO (4.16)

L
M M
.z'= I, cr .z'r ...~r, £= cr( .z'r ar + ~) (4.17)
r =1 r =1

L
M M

.-6=
r =1
Cr .,6 r .!1 r ' •= I, cr( .,6r .1r + •r) (4.18)
r =1

These are potentially convenient expressions, however, it should be kept in mind that
the subdivision into only two subvolumes r = o:, {3 may not provide for accurate
evaluation of the local inelastic response and overall behavior.
Transformation Analysis and Experiments 31

The instantaneous local fields and overall properties of multiphase aggregates,

or unit cells subdivided into many elements, can be evaluated in a similar way. First,
(4.13) are augmented for all volumes OP, p = 1, 2, .. M, and written in matrix form,

Then, one finds the augmented instantaneous strain concentration factors [vi p] and
the thermal strain concentration factors { ap},

(4.20)

Substituting [vi p] into (4.17) then provides the overall instantaneous stiffness matrix
for the aggregate,

(4.21)

The instantaneous thermal stress matrix is found by substituting in a similar manner,

{~} = (cp ..t'p)[diag (I)+(DP'l ~ ,.t''IJPJf{{ap}-[DP'IJ ~){~P}} +

M

+ {l>'l] ~}
'1]=1
(4.22)

The overall instantaneous compliance matrix and the thermal strain matrix
are found in an analogous way as,

(4.23)

{ •} = [cp ~ p) (diag (I)+[FP'IJ ~ ~~J) -!{ {bp}-[FP'IJ ~]{ -~}} +

M

+ { Lc'IJ •'IJ} (4.24)

'1]=1
 32 G.J. Dvorak and Y.A. Bahei-EI-Din

These expressions provide the overall instantaneous moduli and compliances

that may be used in analysis of elastic-plastic composite structures, such as
laminated plates and shells. Integration of (4.19) along the thermomechanical
loading path specified by the history of overall strain EO( t) or stress o{l( t) and
temperature OO(t) provides the piecewise uniform local strains or stresses. Efficient
solutions have been constructed with the Runge--Kutta formula of order two [21].
As an example, consider a layer of a boron/aluminum fibrous system, with an
elastic fiber (E = 379.2 GPa, v = 0.21) and an elastic-plastic matrix (E = 68.9 GPa,
v = 0.33) of the Mises type, with a tensile yield stress Y = 24 MPa and constant
plastic tangent modulus H = 14 GPa, which follows the Prager-Ziegler hardening
rule. The PHA model with 16 elements was used. The prescribed loading path in
transverse tension a 22 and shear a 23 , and the overall response appear in Fig. 4.
Detailed comparisons of the above results with ABAQUS finite element
solutions for the same system show close agreement. The relative efficiency of the
two procedures is illustrated in Fig. 5, which indicates for various meshes the number
of multiplications required for the two methods in evaluation of the concentration
factors and during 100 loading steps. The implication is that the transformation
method is advantageous when small meshes are used. However, the cost of both
methods depends on many other factors, such as the specified tolerance, and the
number of iterations to achieve that tolerance. These can be evaluated only in actual
applications. The transformation method can readily accommodate any physically
admissible constitutive relation, whereas a separate UMAT routine would be needed
in an ABAQUS implementation.

4.3 Viscoplastic systems

In composite materials with viscoplastic phases, the local constitutive
equations may be specified using a certain unified theory that connects the local

inelastic strain rate £~n( t) to the local stress history, or the local relaxation stress

rate ir,;e( t) to the local strain history. For example, the local inelastic strain rate
may be specified by a power law of an internal stress variable R, [24],

(4.25)
 400 r n._
3 4
300 r Boron/ Aluminum
c,- 0 45
350 ~
250 V>

300
0'
200
3
!:?.
Iii 250 Iii a·
CL CL ::s
~200 ~ 150 >
M ::s
N $»
1:) 150 1:)
"'"'
100 -<
~-
V>
100 $»
::s
Q.
tT1
><
'"0
-o- ~
100 150 200 250 300 1 2 3 4 5 6 7 :::1.
a 22 (MPa) E22 (x10-3) 3
~
a
V>

400 .. 5
Boron/Aluminum

350'
c,- 0.45
300

.3

2200
i"'l
j150 J
1001-
X1

~
~
0005 001 0.015 0.02 0025 003

2. '* E23

4. The transverse tension and shear loading path and the resulting overall
w
response for a boron/aluminum fiber composite [21). w
 34 G.J. Dvorak and Y.A. Bahei-El-Din

No. of Multiplications x 10 6
No. of
Elements Col')centration FEM TFA
Factors
For 1oo Load Steps
I
16 0.57 8.49 3.n I
I
I

-.....
I
45 ·6.10 43.49 28.85 I
I
I
<0 93 32.06 138.67 128.56 I

-
I
0 300 I
)(
159 112.32 344.87 394.80 I
I
I
I
I
I
- - - - TFA I
I
-FEM I
I
I.
- · - Concentration Factors

-0

-· .-
z
0

____ .,. -· -·
.
-· -· ,;'
-·"'

............ ...........
OL----~--.u~~~--~·~---------~------~------._------~---~
o ro ~ ~ ~ 100 ,~ ,~ ,~

No. of Elements

5. Estimated number of multiplications required in the TF A and FE methods for

evaluation of the transformation concentration factors, and in 100 load steps
applied to the PHA models with different number of elements.
 Transformation Analysis and Experiments 35

where ~( 8), p'I'J( 8) are material parameters for the element volume n'I'J, and ~
specifies the direction of the inelastic strain rate in the local stress space. When

substituted into the rate form of (4.5) with ~e(t) =- ~ £~n(t), (4.25) provides the
following differential equation for the local stresses,

up(t) = BP U(t)- f
'1')=1
FP'I'J ~ {~ 0°(t) + ~(8) R~'I'J( 8) ~} (4.26)

where o(t) is the prescribed overall stress history and O(t) is the temperature history.
The power law (4.25) leads to a stiff differential equation which requires integration
by an implicit multistep method, such as the Adams method, with Newton iteration
to solve the resulting nonlinear system of equations, and backward differentiation to
evaluate the functional derivatives found in the Newton's method.
Using material constants specified in [21], the overall response of a fibrous ply
of a SCS6/Ti-15-3 composite was evaluated for an overall combined biaxial stress-
time path, Fig. 6. The results show the time rates of the applied stress components,
and the two total overall strain component history during the applied loading
program.

4.4 Viscoelastic systems

In these systems, the present approach provides for easy implementation of
both linear and nonlinear constitutive models, and avoids the difficulties involved in
using the correspondence principle in linear viscoelasticity. In the following
illustrations, we use the established linear model [25].
For a history of applied stress that starts from zero stress at t = 0, the phase
constitutive equations is taken in the form,

f"r(t) = J
0
Jr(t- r)
do"r( r)
dr
dr (4.27)

where Jr(t- r) is the creep compliance, with symmetries Jj"jkl=J]ikl=Jijik=J~Iij·

We assume the existence of a nonvanishing compliance Jr(O) that defines the
instantaneous elastic response, and thus provides property magnitudes needed in
evaluation of the transformation influence functions.
 512000MCI
•IIOOOHCI 3 0 11 ·00MPAI'MC 5 VJ
500 600 • I 0\
., 0

•so
...
500 2 :-------~~~~~~--05
I
• I
~ I
I
1:!350 I
10
0
-
300
'I
~· •
~
-
t)
I
I
I
1-

3
i
~
2!0

200
1750MCI i
-
300
~3
__ .,,
! 200 :~
'
- - - - "D
1130000C) 2f700oocl p-------~2 '
, 1 00
;:.
i ,
,.,,
,.,
,,
100 200 300 500 1100 200 100 1000 1200 1 - 1100 11100 2000
Axilll Sirna, o 1 1 (MP•l - - - Time, I(MC.j

5
scsem-15-3
c1 - 0.3
• 412"C

~
0
.!!.
,...J __
j ,,,.A2
a
'-
2
I ,' 0
p" <
/
, 0
, ;J
;><;"
0~ I I I I I I I I t
..,::::
0 200 - 1100 - 1000 1200 1 - 1100 · - 2000 0..
>-<

6. Biaxial tension path applied to the SCS6/Ti-15-3 composite with a

>
..,co
:::r
viscoplastic matrix, with the time history of the applied stresses, and the ro
'
predicted overall strains. The integration was performed with the GEAR tp
solver with tolerance of 0.001 [21). 9.
::::
Transformation Analysis and Experiments 37

Similarly, for a history of local strain starting at zero strain at t = 0, the total
stresses are given by

(4.28)

where Gr(t- r) is the local relaxation function with the same symmetries that were
indicated for Jr(t- r), and Gr(O) 'f 0 describes the elastic response.
The corresponding rates are obtained from the above relations as,

£r(t) = Mr o"r(t) + jr(O) o"r(t) + f

0
Jr(t- r) ur(r) dr ( 4.29)

o'r(t) = Lr £r(t) + Gr(O) Er(t) + f

0
Gr(t- r) Er(r) dr (4.30)

The elastic and inelastic parts of the total strain and stress, and of the
respective rates, are easily separated, and the latter are substituted into appropriate
total or rate forms of ( 4.4) or (4.5). For example, if the composite is loaded by a
history of overall stress u( t) and temperature history 0( t ), the local stresses and their
rates follow from [21],

irp(t) = BP ir(t)- ± ~ {~
1]=1
FP"l O(t) + jTJ(O) uTJ(t) + f
0
JTJ(t- r) uTJ(r) dr}

(4.31)

The terms on the right are functions of the current state and applied load.
Integration along a specified loading path yields the local stresses in the local volumes
OP, p = 1, 2, .. M. The corresponding strains are found by integration of (4.29).
 38 G.J. Dvorak and Y.A. Bahei-El-Din

In addition to the relations for the stress rates, one can also find governing
equations for the total local stresses. The result is,

L
M

O'p(t) = Bp u(t)- F P'1 ~ {~ O(t)- ~ 0'1](t) +

1]=1
t

f
+ J11(t -r) (du11 (r)/dr) dr}
0
(4.32)

Again simple closed-form relations for the local fields can be written in
two-phase systems, subdivided into only two local volumes r = a, {3. In this case,
differentiation of (4.4) provides,

(4.33)

This can be easily solved using relations that follow from (3.31),

caDaa + c13DI3a = 0 ~ Daa =- (c13 /ca) Dl3a

Dl3aDaa =- (cl3/ca) Dl3a Dl3a = Daa Dl3a (4.34)

E = Ca Ea + C13 E13 ~ Ea = (E- C13E13)/ca

Then, after some algebra one finds the following equation for the local strain rate in
one phase, say r = {3,

+ (D13aDa13- DaaD1313) J13(0) f

0
Gr3(t- r) Er3( r) dr- (4.35)

.
- [Dr3a (Aa- (1/ca)I)- Daa ~] E0(t)- (Dr3aaa- Daa~) 0°(t)
Transformation Analysis and Experiments 39

A similar procedure involving total fields rather than rates gives the total local
strain in the phase {3 as,

= (D(3o.D~- Do.o.D(3(3) J(3(0) f

0
G(3(t- r)(dE(3(r)fdr) dr- (4.36)

The total stress fields, or strain fields, can be found at a given time t by
solving one of the above systems of nonlinear equations. Either the Newton's method
or the fixed point solution may be used. Both methods require repeated evaluation of
the nonlinear functions J'IJ or G(3 , and of the time derivative of the stress or strain
fields. Moreover, Newton's method requires an evaluation of a Jacobian matrix from
the nonlinear functions in (4.32) and (4.36), and of its inverse. This complicates the
numerical procedure for evaluation of the local fields, and thus argues against writing
the governing equations in the total form.
A more convenient and efficient procedure for evaluation of local fields is given
by integration of the governing rate equations (4.31) or (4.35). This calls for
u ..

evaluation of the integrands [J'IJ(t- r) u'IJ( r) dr] or [G'IJ(t - r) ~( r) dr], 17 = 1, 2, ..

M, from r = 0 to r = t. The local stress and strain histories are known up to the
..
current time t, while derivatives of the creep compliance J and of the relaxation

function G must be provided by the viscoelastic phase model. The local fields at time
(t + ~t) are then found by integration of the differential equations (4.32) or (4.35),
using the Runge-Kutta formula of order two, for specified histories of the overall

stress rate u0( t) or the overall strain rate £0( t ), and the temperature rate 0°( t) [21].
40 G.J. Dvorak and Y.A. Bahei-El-Din

5. EXPERIMENTS IN PLASTICITY OF FIBROUS COMPOSITES

5.1 Bimodal plasticity theory

The unit cell models discussed above have been designed to evaluate in
sufficient detail the inelastic local fields and overall response of composite materials,
for many constitutive theories prescribed for the phases. However, in plasticity of
fibrous composites one can obtain much useful information about the shape and
position of the initial and subsequent yield surfaces with a simpler model that relies
on certain assumptions about internal deformation modes [26]. This bimodal
plasticity theory admits two distinct inelastic deformation modes in the fiber system,
the fiber-dominated and matrix-dominated modes. Each mode is activated by a
different overall stress state and is associated, through a certain yield condition, with
a branch of the overall yield surface; the internal envelope of the two branches is the
overall yield surface.
In the fiber-dominated mode (FDM), the composite is regarded as a
heterogeneous medium where both the fiber and the matrix participate in carrying
the applied load. The local stresses in the phases may be estimated, for example, by
the self--<:onsistent or Mori-Tanaka methods, discussed in Section 2.2. The overall
yield surface indicates the magnitudes of the overall stress that cause local stress
averages to satisfy the matrix yield condition. In particular, suppose that the fiber
remains elastic while the matrix follows the Mises or another yield condition given in
the form f( um) = 0, then the yield condition of the composite in the overall stress
space u becomes F( u) = f(Bmu), where Bm is the elastic stress concentration factor of
the matrix phase (1.8, 1.10). The Bm may be found using one of the averaging
methods discussed in 2.2. For example, iff( um) is represented by the Mises form and
displays kinematic hardening, one finds the FDM branch of the overall yield surface,

1 T T 2
F(u- a)=- (u- a) (Bm Q Bm)(u- a)- T 0 = 0 (5.1)
2

where a denotes the current position of the center of the yield surface, To is the
matrix yield stress in shear, and Q is a 6x6 matrix with Q11 = Q22 = Q33 = 2/3, Q12
= Q13 = Q23 =- 1/3, Q44 = Q55 = Q66 = 1, while the remaining coefficients vanish.
Transformation Analysis and Experiments 41

In the matrix-dominated mode (MDM) of plastic deformation, all applied load

is assumed to be carried by the matrix, while the fiber constrains the matrix plastic
deformation to simple shear straining on planes that are parallel to the fiber axis x 1.
The matrix~ominated mode is thus represented by a variant of the continuum
slip-model. For plane state of stress applied in the experiments discussed later, the
slip systems that may operate in the matrix mode are indicated in Fig. 7.
The initial yield condition on any potential slip plane (k) is taken as,

(5.2)

where r ns denotes the resolved shear stress. The active slip system is defined by the
normal ni to the slip plane, and by the slip direction sj, so that the resolved shear
stress is,

r<kl
ns
= n(kl
1
a..
~
s(kl
J
= (r2-r2)1/2
1 2 Ckl
(5.3)

7'1 = a21 cos /31 7'2 = ~ a22 sin 2/31 (5.4)

The max r ns(/3) is evaluated from,

(5.5)

For the slip system on the plane k = 1,

/3 1 = ~cos- 1 q 2 for lql ~ 1; /31 =
0 for lql ~ 1 (5.6)
where q = a 2tf a 22 if a 22 # 0, and the angle 01 between the slip direction and x 1 is
given by

tan 01 = ~sin /3 1 (5.7)

The conjugate system on the plane k = 2 in Fig. 7 is specified by the angles {32
and 02 that are related to /3 1 and 01 by
42 G.J. Dvorak and Y.A. Bahei-El-Din

Substituting the slip system parameters that assure the maxima of the
resolved shear stress under the applied overall plane stress state, and assuming
kinematic hardening for the matrix, one finds the overall yield condition for the
MDM mode as
fa(u> = [(12\: a21r + [(12\: a22 :1: 1r -1 =o for lql ~ 1
(5.9)

These relations suggest that the MDM yield surface is an open-ended cylinder
of oval crossection in the overall plane stress space, with generators parallel to the u 11
direction. Its crossection in the u21 u22-plane for the initial state a 21 = a 22 = 0, is
shown in Fig. 8. Note that this surface is independent of phase moduli and volume
fractions.
The internal envelope of the two branches (5.1) and (5.9) of the overall surface
appears Fig. 9 for the boron aluminum system examined in the experiments. The
FDM segment (5.1) of the overall yield surface resembles an ellipsoid in the overall
plane stress space, providing the end caps of the cylindrical MDM surface.
The experimental results that follow suggest kinematic hardening in both
modes, with evolution of the back stress components of the matrix-dominated mode,

forlql~1
(5.10)
for lql ~ 1

For the fiber mode, the experiments suggest compliance with the Phillips
hardening rule,

da= du (5.11)

Since both modes operate on the same local volume of the matrix, hardening
in either mode affects the entire overall surface, and there is no relative translation of
the two branches during deformation. The bimodal theory predicts quite closely the
shape and position of the observed yield surfaces. However, normals to these surfaces
do not provide reliable estimates of actual directions of the plastic strain increments.
 Transformation Analysis and Experiments 43

7. Geometry of the conjugate slip systems of the matrix-dominated deformation

mode [26].

'· '·

-3 2 3
uulro

-1

-2
'·
8. Transverse cross-section of the initial yield surface of the matrix-dominated
mode [26].
 44 G.J. Dvorak and Y.A. Bahei-El-Din

9. The plane stress initial yield surface predicted by the bimodal plasticity theory
[26).
Transformation Analysis and Experiments 45

5.2 Interpretation of experimental results

Here we briefly summarize the main conclusions obtained in our experimental
studies of inelastic behavior of boron-aluminum fiber systems [27-30], and in the
interpretation of the results . The experiments were conducted on thin-walled tube
specimens shown in Fig. 10. The tubes were fabricated by diffusion bonding of
monolayer 6061-Al/B sheets by Amercom in Chatsworths, CA. The finished tube
had seven layers of fiber in the wall, all aligned in the axial direction, with fairly
regular fiber distribution, Fig. 11. The fiber diameter was 142tt, volume fraction cr =
0.45. All specimens were annealed at 4000C for two hours, and C-scanned for
damage before testing. Steel collar grips were adhesively bonded at both faces to the
ends of the tube. The specimens were instrumented with strain gages and :1:: 450
rosettes for axial, hoop and shear strain measurement. Loading was applied in
tension/compression and by internal pressure and torque in a computer-controlled
MTS machine. The loading rate was 4.2 MPafmin. in any direction. Accuracy of
measurements was 0.1 MPa for stress, and 1x10-6 for strain.
The specimens were subjected to various loading programs. Here we consider
the results found for loading by internal pressure and torque; this produced a
transverse normal stress u22 and a longitudinal shear stress u21 in the tube wall.
Since no loading compensation was made for the axial tension caused by the internal
pressure, an axial normal stress u11 = u22 /2 was induced as well. The measurements
were thus made in a stress plane inclined to the u21 u22-plane ; the appropriate scale
on the projection of the the u22 axis into that plane is (.J5/2)u22 • This scale is used in
Fig. 12 and later. The application of the pressure-induced axial stress was accounted
for in modeling.
The objective of the experiments was to detect initial and several subsequent
yield surfaces along the loading path, and the plastic strain magnitudes after each
load increment. At each point on any yield surface, the yield stress was defined by
back extrapolation from a small excursion into the plastic region, at zero offset strain.
The experimental yield surfaces were constructed by fitting the appropriate sections
of the bimodal yield surface of Figs. 8 and 9 to the detected yield points [28, 29].
Some of the results are shown in Fig. 12. The surfaces are of somewhat different size,
the same effect was observed in neat matrix experiments [27].
In addition to the bimodal yield surfaces, initial and subsequent clusters of
46 G.J. Dvorak and Y.A. Bahei-El-Din

1
12.5mm

T
50mm
end attachment
(diameter= 76 mm)

tubular specimen
(diameter =41.5mm}
(wall th ick.= 1.3mm)

100mm •

circumferential
strain gage
adhesive bonding

materials:
50mm tu be: 8-AI, c 1 =0.45

1
grip : steel

12.5mm
T
10. Dimensions and instrumentation of the composite tube specimen [23].

U. Transverse section of the composite tube specimen [23}.

 Transformation Analysis and Experiments 47

6.8

x,
J-~
8-AI, Ct= 0.45
Ill

80 110 120 130

1cr 22 (MPa)

• xeo Experimental
Bimodal Theory (MOM)
<f> Center of Yield Surfaces
Loading Sequence:
Surface 1( x); 0 -3; Surface II (e); 4 -s; Surface Ill ( •); 6-11; Surface IV (O)

12. Loading path in the transverse tension-longitudinal shear stress space, with
experimental yield points and matrix-dominated yield surfaces [28].
48 G.J. Dvorak and Y.A. Bahei-El-Din

yield surfaces were constructed for the subelements of the unit cell of the PHA model,
using a refined mesh with 222 elements. Each surface in the cluster was generated by
substituting the stress concentration factor Bm for a given finite element into {5.1),
together with the measured matrix yield stress, and plotting the result in the
u21,{.j5/2)u22-fJtress plane. All solutions were obtained using the ABAQUS program.
The twO-fJurface plasticity theory used in this simulation, and the material
parameters used are summarized in (30].
Figure 13 shows the initial yield surfaces found in the experiments and from
(5.1). As expected, the latter are seen to cross inside the experimental surface,
indicating presence of micro yielding during loading to the experimentally detected
yield points. The resulting rearrangements of the cluster are not shown. Figure 14
shows the cluster after extensive loading in transverse tension, together with the
computed and measured plastic strain increment vectors. Figure 15 shows a similar
situation after complete unloading from point 2 and reloading along the path leading
to point 6. The cluster rearrangements are caused by local hardening in response to
the applied and plastic strain-induced residual stress fields in the domain. The load
vector intersects with all local yield surfaces at the current loading point, indicating
that plastic yielding is taking place in all elements. The plastic strain increments are
contained in the cone of normals to the element yield surfaces at the loading point
[31, 32], in agreement with the normality rule resulting from the Drucker's postulate.
However, since the strain field in the matrix is not uniform, normality to the
experimentally detected surface is not required. In fact, the experimental surface
merely joins the vertices formed by the cluster of the local surfaces at intersection
with the load vector, when onset of yielding is detected in the experiment. Moreover,
the element yield surfaces become intercepted during loading excursions from the
current loading point on the main path toward the individual points on the current
yield surface. This causes micro yielding and addition&l rearrangement of the cluster.
After such rearrangement, the experimentally detected yield surface may not
necessarily pass through the loading point, such as #6 on the main path in Fig. 15.
This phenomenon, piercing of the current yield surface, is often observed in
experiments of this kind.
Figures 16 to 18 show comparisons between the predictions of plastic strains
along the path of Fig. 12, obtained from the PHA model with the two different
meshes, the matrix mode of the bimodal theory, the Mori-Tanaka version of the Hill
 ~
l'l
::l
"'8'
<J 21 (MPaJ
XJ 3~
B-Al. c 1 =0.45 5"
::l

>-
::l
l'l
~
x2 ::;;·
"'
l'l
::l
a.
tr1
><
X1 '"0
~
::l.
~
::?.
"'
60 60 70 80 90 100 110 120 1)0

/j a 22 (MPa l
2

o Expenmental
B1modal Theory (MOM)
Local Y1eld Surface Protecuon
from PHA Model
~ Center ot B1modal Y1eld Surface

13. The cluster of initial PHA yield surfaces, experimental yield points , and the
bimodal yield surface [30] . ~
\0
 v.
0
x3

60
8-AI. c, = 0.45
so
x2
4 1)

30 X1

20

10 .........
<11
a.. Cone of Normals
~ 0
0

tl"' •o
-10
Jj cr22 (MPa)
- 20 2

- 30 Plastic Stram Increment Otrec11on

a
--- --• Experimental
'-
- 40
···· ·······• Numencal 0
<
0
- SO ~
I»
::l
- 60 Loadmg Sequence Local Yteld Surface ProteCtiOn c.
0-1-2 from PHA Model --<
Loading Path ~
to
I»
::r
~-
14. The cluster of the local PHA yield surfaces at loading point 2, and the ~
observed and predicted plastic strain increment directions [30]. 6:;·
 x3
~
s:»
:I
CJ>
0'
...,
3
::;
xz o·
:I

>
:I
s:»
X1 '-CJ><"'
c;;·
s:»
:I
0-
tTl
;.<
"0
(1)
...,
~-
a
CJ>

1()0 1 10 , 10 • )0

~ a 22 iMPa)
2
Plasllc Srraon Increment D•recroon
--· · · • Expenmenral
· ·· .. Numerteal

o Expenmenral
Bm1odal Theory IMDMl
Local Y~eld Surface Pro1ec110n
Load•ng Sequence from PHA Model
-60
0-1-2-3-4-5-6 * Center of B•modal Y•eld Surface
Loading Path

15. The cluster of the local PHA yield surfaces at loading point 6, with
experimental yield points and a matrix-dominated mode yield surface, and
observed and predicted plastic strain increment directions [30]. VI
 52 G.J. Dvorak and Y.A. Bahei-El-Din

x3
8-AI, c 1 = 0.45
120
x2
-· -· -·
Path 0- 11
-
~---·--·2
100
ca x1
a..
~
-
~C\1
C\1 80 .
6.7 9
--·6
(/J

--a
(/J
Q)
60
~

en 5
4
40 ....... PHA Model (Refined Mesh)
E
~ - PHA Model (Coarse Mesh) 10 10
0
z 20
-·-Bimodal Theory (MOM)
- .. - Mori· Tanaka
····Experimental 11

0
0.00 0.02 . 0.~ 0.06 0.08 0.10 0.18

Plastic Normal Strain, £~2 (1 0 '2 )

16. Transverse normal stress-plastic strain response of the composite tube during
loading along the path 0-11 [30].
 Transformation Analysis and Experiments 53

x2
B-AI, c t =0.45 x1

Path 0- 11

- -------1I
6

--
ca
a.. 7,9 t 7.9
:2 10 r 10
I
t:)C\1 I
8 ra
(/)
I

-...
(/)
I
~ I
en I
I
ca
Q) 0 I
-··- Mori·Tanlka. I
~
en ---· Experi"*"-1 I
-10 I

-~~~-----~------4-----~-----~~~~~_.
,,I
.....--.....
I

._~

0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.~ 0.22

Plastic Shear Strain, 2£~1 ( 10 ·2 )

17. Longitudinal shear stress-plastic strain response of the composite tube during
loading along the path 0-11, as observed and predicted by various methods
[30].
 54 G.J. Dvorak and Y.A. Bahei-El-Din

8-AI, c, = 0.45
0.20 6.7.9.10

o- 11 ,{e
--?fl. Path
•••··•• PHA Model (Refined Mesh) I
I
I
11 x1

Q.~ 0.15 -PHA Model (Coarse Mesh) I

-·-Bimodal Theory (MOM) 1 6.7.9. 10
~ I e
-··- Mon-Tanaka

-
c: ---· Expenmental I
'(ij I 11
·~
I
en 0.10
I
~ I
ctS I
Q)
..c I
en 6

·-
(J

(/). 0.50
ctS
a:
o.oo Lt-...~~-~-~--~-~-::i·~L_L,___ ____j

0.00 0.50 0.10 0.15 0.20

Plastic Normal Strain, e ~2 ( 1o·2 )

18. Plastic strains in the composite tube during loading along the path 0-11, as
observed and predicted by various methods [30].
Transformation Analysis and Experiments 55

model [33], and from experimental measurements. All the results were obtained with
the same constitutive equations. In the bimodal theory, these equations were applied
to the active slip systems, with the plastic strain increment direction given by the
normal to the current yield surface.
The comparisons indicate that, as expected, both the Mori-Tanaka and
bimodal theories, relying as they do on the normality to a single yield surface, fail to
approximate the actual plastic strains. In contrast, the PHA model provides a
reasonable approximation, although there is a significant deviation along the path
segment 5-6 in Fig. 12. However, the overall agreement is probably better than that
one might find in comparisons of the two-surface plasticity theory with similar
experiments on unreinforced tubular specimens loaded along an open path.
Comparisons of the transformation method predictions with results of cyclic
loading experiments on viscoplastic titanium matrix laminates reinforced by silicone
carbide fibers are reported in [34. 35].

6. CLOSURE
The transformation field analysis is presented here as a general method for
solving both mechanical and transformation loading problems for heterogeneous
media with many interacting inhomogeneities. The method accommodates any
uniform overall loading path and inelastic constitutive equations that conform with
the additive decomposition of local strains, as well as any micromechanical model
that offers connections between local and overall fields in representative volumes of
elastically deforming heterogeneous aggregate. Indeed, many existing approaches to
analysis of inelastic heterogeneous media, such as the unit cell models, and the
self-consistent, Mori-Tanaka and other methods based on the Eshelby solution are
included as special cases of the TF A procedure. The elastic-plastic, viscoelastic and
viscoplastic deformation problems are also treated in a unified manner, by solving a
system of governing equations for either total local fields or for their rates. While the
nature of the governing equations and the solution methods may vary, their
derivation is accomplished in a standard form that incorporates local geometry effects
through the transformation influence functions, and local material response through
the selected constitutive relations. Moreover, the transformation method eliminates
the need to solve inelastic inclusion problems, indeed, such solutions are associated
with models that violate the generalized Levin formula [23].
 56 G.J. Dvorak and Y.A. Bahei-El-Din

In comparisons with experimentally measured plastic strains, only the unit cell
models appear to provide reasonable approximations to the observed strain
magnitudes. The main reason is the ability of these models to monitor the correct
direction of the plastic strain increments. The models that rely on normality to a
single overall yield surface are bound to provide misleading predictions, because no
such single overall surface actually exists. The experimentally found surfaces are seen
as loci of vertices of yield cones formed as inner envelopes of the clusters formed by
multiple local branches of the overall yield surface. Normality is thus satisfied only
within the cone of normals to the local surfaces, in agreement with Drucker's stability
postulate [36]. Only those micromechanical models that can reproduce the plastic
strain increments within these cones may, in general, provide reliable predictions of
the observed behavior. These objectives can be accomplished by several techniques,
for example with the fast Fourier transform, as shown by Moulinec and Suquet [37,
38]. We also note the work of DeBotton and Castaneda [39] which provides a more
general model for prediction of the bimodal yield surfaces and plastic strains.

ACKNOWLEDGEMENT
The work surveyed in this paper was supported by grants from the Air Force
Office of Scientific Research, and by the Mechanics Division of the Office of Naval
Research. Drs. George Haritos, Walter Jones and Yapa Rajapakse served as program
monitors. Drs. Himanshu Nigam, Rahul Shah and Amr Wafa contributed to the
various research results described here.

REFERENCES

1. Hill, R., 1963, "Elastic properties of reinforced solids: Some theoretical

principles," Journal of the Mechanics and Physics of Solids, Vol. 11, pp. 357-372.
2. Mori, T. and Tanaka, K., 1973, "Average stress in matrix and average energy of
materials with misfitting inclusions," Acta Metallurgica, Vol21, pp. 571-574.
3. Benveniste, Y.,1987, "A new approach to the application of Mori-Tanaka's theory
in composite materials," Mechanics of Materials, Vol.6, pp. 147-157.
4. Hill, R., 1965, "A self-consistent mechanics of composite materials," Journal of
the Mechanics and Physics of Solids, Vol. 13, pp. 213-222.
 Transformation Analysis and Experiments 57

5. Hashin, Z. and S. Shtrikman, 1963, "A variational approach to the theory of the
elastic behavior of multi phase materials," Journal of the Mechanics and Physics of
Solids, Vol. 11, pp.127-140.
6. Walpole, L. J., 1966, "On bounds for the overall elastic moduli of inhomogeneous
systems -!,"Journal of the Mechanics and Physics of Solids, Vol. 14, pp. 151-162;
Part II, ibid. Vol.14, pp. 289-301.
7. Eshelby, J. D., 1957, "The determination of the elastic field in an ellipsoidal
inclusion and related problems," Proc. Roy. Soc. London, Vol A241, pp. 376-396.
8. Walpole, L. J., 1969, "On the overall elastic moduli of composite
materials," Journal of the Mechanics and Physics of Solids, Vol. 17, pp. 235-251.
9. Hill, R., 1964, "Theory of mechanical properties of fibre-strengthened materials:
I. Elastic behaviour," Journal of the Mechanics and Physics of Solids, Vol. 12,
pp. 199-212.
10. Dvorak, G. J. andY. Benveniste, 1992, "On transformation strains and uniform
fields in multiphase elastic media," Proceedings A of the Royal Society, London, Vol.
A437, pp. 291-310.
11. Chen, T., G. J. Dvorak andY. Benveniste, 1992, "Mori-Tanaka estimates of the
overall moduli of certain co~posite materials," Journal of Applied Mechanics, Vol.
59, pp. 539-546.
12. Teply, J.L. & Dvorak, G.J., 1988, "Bounds on overall instantaneous properties of
elastic-plastic composites," J. Mech. Phys. Solids, Vol. 36, pp. 29-58.
13. Hashin, Z., 1965, "On elastic behaviour of fibre-reinforced materials with
arbitrary transverse phase geometry," J. Mech. Phys. Solids, Vol. 13, pp. 119-134.
14. Hashin, Z. and B. W. Rosen, 1964, "The elastic moduli of fiber reinforced
materials," Journal of Applied Mechanics, Vol. 31, pp. 223-232.
15. Torquato, S., 1991, "Random heterogeneous media: Microstructure and improved
bounds on effective properties," Applied Mechanics Reviews, Vol. 40, pp. 37-76.
16. Benveniste, Y., G. J. Dvorak and T. Chen, 1991, "On diagonal and elastic
symmetry of the approximate stiffness tensor of heterogeneous media," J. Mech.
Phys. Solids, Vol. 39, pp. 927-946.
17. Dvorak, G.J., 1990, "On uniform fields in heterogeneous media," Proc. R. Soc.
Lond. A 431, pp. 89-110.
 58 G.J. Dvorak and Y.A. Bahei-EI-Din

18. Levin, V. M., 1967, 11 Thermal expansion coefficients of heterogeneous materials, 11

Mekhanik:a Tverdogo Tela, Vol. 2, pp.88-94, English translation, Mech. of Solids,
Vol. 11, pp.58--el.
19. Berveiller, M. and Zaoui, A., 1979, 11 An extension of the self-consistent scheme
to plastically flowing polycrystals. 11 J. Mech. Phys. Solids, Vol. 26, pp. 325-344.
20. Walker, K. P., E. H. Jordan and A. D. Freed, 1991, 11 Equivalence of Green's
function and the Fourier series representation of composites with periodic structure, 11
in Micromechanics and Inhomogeneity - The T. Mura 65th anniversary volume (G.J.
Weng, M. Taya and H Abe, eds.), Springer Verlag, New York, 535-558.
21. Dvorak, G. J., Y. A. Bahei-El-Din and A. Wafa, 1~94, 11 Implementation of the
transformation field analysis for inelastic composite materials," Computational
Mechanics Vol14, pp. 201-228.
22. Dvorak, G.J. & J.L. Teply, 1985, 11 Periodic hexagonal array models for plasticity
analysis of composite materials, 11 In Plasticity Today: Modeling, Methods and
Applications, W. Olszak Memorial Volume (ed. A. Sawczuk & V. Bianchi),
pp. 623--e42. Elsevier Science Publishers, Amsterdam.
23. Dvorak, G. J., 1992 "Transformation field analysis of inelastic composite
materials," Proceedings of the Royal Society, London, Vol. A437, pp. 311-327.
24. Bahei-El-Din, Y.A., R.S. Shah and G.J. Dvorak, 1991, 11 Numerical analysis of
the rate-dependent behavior of high temperature fibrous composites," In Mechanics
of Composites at Elevated and Cryogenic Temperatures (ed. S.N. Singhal, W.F.
Jones, T. Cruse & C.T. Herakovich), ASME AMD-vol. 118, pp. 67-78. American
Society of Mechanical Engineers, New York, NY.
25. Christensen, R.M. 1971, Theory of viscoelasticity, An introduction. Academic
Press, New York.
26. Dvorak, G. J. and Y. A. Bahei-El-Din, 1987, 11 A bimodal plasticity theory of
fibrous composite materials, 11 Acta Mechanica, Vol. 69, pp. 219-241.
27. Dvorak, G. J., Y. A. Bahei-El-Din, Y. Macheret and C. H. Liu, 1988, "An
experimental study of elastic-plastic behavior of a fibrous boron-aluminum
composite, 11 J. Mech. Phys. Solids, Vol. 36, pp. 65H87.
28. Nigam, H., G. J. Dvorak and Y. A. Bahei-El-Din, 1994, 11 An experimental
investigation of elastic-plastic behavior of a fibrous boron-aluminum composite:
I. Matrix-dominated mode, 11 lntl. J. Plasticity, Vol. 10, pp. 23-48.
 Transformation Analysis and Experiments 59

29. Nigam, H., G. J. Dvorak and Y. A. Bahei-El-Din, 1994, "An experimental

investigation of elastic-plastic behavior of a fibrous boron-aluminum composite:
II. Fiber-dominated mode," Intl. J. Plasticity, Vol. 10, pp. 49-{)2.
30. Dvorak, G. J., Y. A. Bahei-El-Din, R. S. Shah and H. Nigam, 1991,
"Experiments and modeling in plasticity of fibrous composites," in Inelastic
deformation of composite materials, (ed. G. J. Dvorak), Springer-Verlag, New York,
pp. 283-306.
31. Dvorak, G. J., 1991, "Plasticity theories for fibrous composite materials," in
Metal matrix compositesj Mechanisms and properties, (ed. R. K. Everet and R. J.
Arsenault), Academic Press, Boston, pp. 1-77.
32. Hill, R., 1967, "The essential structure of constitutive laws for metal composites
and polycrystals," J. Mech. Phys. Solids, Vol. 15, pp. 79-95.
33. Lagoudas, D. C and A. C. Gavazzi, 1990, "Incremental elastoplastic behavior of
metal matrix composites based on averaging schemes," in Inelastic deformation of
composite materials, (ed. G. J. Dvorak), Springer-Verlag, New York, pp. 465-485.
34. Dvorak, G. J., H. Nigam andY. A. Bahei-El-Din, 1996, "Isothermal fatigue of
Sigma/Timetal 21S laminates, I. Experimental results," Mechanics of Composite
Materials and Structures, to appear.
35. Bahei-El-Din, Y. A. and G. J. Dvorak, 1996, "Isothermal fatigue of
Sigma/Timetal 21S laminates, II. Modeling and numerical analysis," Mechanics of
Composite Materials and Structures, to appear.
36. Drucker, D. C., 1951, "A more funbdamental approach to plastic stress strain
relations," in Proc. 1st. U.S. Natl. Cong. Appl. Mech., ASME, New York,
pp. 487-491.
37. Moulinec, H. and P. Suquet, 1994, "A fast numerical method for computing the
linear and nonlinear mechanical properties of composites," C. R. Acad. Sci. Paris,
t.318, Serie II, p. 1417-1423.
38. Moulinec, H. and P. Suquet, 1995, "A FFT-based numerical method for
computing the mechanical properties of composites from images of their
microstructures,"IUTAM Symposium on Microstructure-Property Interactions in
Composite Materials, Aalborg University, Denmark, Kluwer Acad. Publishers,
Dordrecht.
39. deBotton, G. and P. Ponte Castaneda, 1993, "Elastoplastic constitutive relations
for fiber-reinforced solids, Intl. Jnl. Solids Structures, 30, 1865-1890.
 RECENT EXTENSIONS OF GURSON'S MODEL
FOR POROUS DUCTILE METALS

M. Gologanu and J-B. Leblond

University of Paris VI, Paris, France
G. Perrin
Bureau de ContrOle des Chaudieres Nucleaires, Dijon, France
J. Devaux
FRAMASOFT +CSI, Lyon, France

ABSTRACT
This paper is devoted to two distinct extensions of Gurson's (1977) famous model for
plastic voided metals. Gurson 's work was based on an approximate limit-analysis of
a typical elementary volume in a porous material, namely a hollow sphere subjected
to conditions of arbitrary homogeneous boundary strain rate. The first extension
envisaged consists in considering a more general geometry, namely a spheroidal vol-
ume containing some spheroidal confocal cavity. The aim here is to incorporate void
shape effects into Gurson 's model. The second extension again considers a hollow
sphere, but now subjected to conditions ofinhomogeneous boundary strain rate. The
goal is to account for possible strong variations of the macroscopic mechanical fields
at the scale of the representative cell (i.e. of the void spacing), as encountered near
crack tips.

INTRODUCTION
The most famous "homogenized" model for porous ductile metals is that of
Gurson (1977). In the authors' view, its major advantage over its competitors
is that unlike them, it was not derived from purely heuristic arguments but from
some micromechanical analysis. Gurson's approach indeed consisted of an approxi-
mate limit-analysis of some hollow sphere (typical representative volume element in
a porous medium) made of rigid ideal-plastic von Mises material and subjected to an
arbitrary (not necessarily axisymmetric as is often mistakenly thought) loading via
62 M. Gologanu et al.

conditions of homogeneous boundary strain rate. This limit-analysis was based on

two trial velocity fields only, one representing an isotropic incompressible expansion
of the sphere and the other one a uniform deviatoric strain rate. Although the "up-
per bound" character of Gurson's approach seemed to be lost in some "uncontrolled"
approximation, it was proved by Leblond and Perrin (1995) that that approxima-
tion in fact preserves it. Therefore Gurson's criterion represents a rigorous upper
bound for the overall criterion of a hollow sphere subjected to conditions of arbitrary
homogeneous boundary strain rate. That upper bound was recently improved by
Garajeu (1995) by replacing the trial velocity field corresponding to a uniform devi-
atoric strain rate by the exact elastic solution for a purely deviatoric loading. The
resulting modification of the yield criterion is important in the case of large porosi-
ties, in that region of the stress space where the deviatoric part dominates over the
mean one.
The present work is devoted to two distinct extensions of Gurson's approach
and model. The first one is concerned with the incorporation of void shape effects
in Gurson's model. This topic was considered in several recent papers ( Gologanu et
al., 1993, 1994; Fonte-Castaneda and Zaidman, 1994; Garajeu, 1995). The second
work heavily relies on the concept of a linear comparison material and the famous
Hashin-Shtrikman theory. It can incorporate the effect of two shape parameters ( ar-
bitr~ry ellipsoidal cavities). However its applicability is limited to the case of small
or moderate values of the overall triaxiality (ratio of the mean macroscopic stress
over the von Mises equivalent macroscopic stress), because for high ones, there are
considerable differences between the behaviours of porous elastic and plastic solids.
For instance, the following (admittedly somewhat paradoxical, but nonetheless real)
phenomenon was evidenced by Budiansky et al. (1982) on the basis of numerical
simulations: an initially spherical cavity located in an infinite plastic medium sub-
jected to some axisymmetric loading with predominant axial stress tends to become
oblate, and not prolate as in elasticity, if the triaxiality is sufficiently high. The
other works are more directly related to that of Gurson in that they are based on
plasticity theory (more specifically, limit-analysis) from the beginning to the end. In
fact, they stand as extensions of it to more complex geometries: they consider prolate
or oblate axisymmetric spheroidal cavities (so that there is a single shape parame-
ter) embedded in a finite representative volume which has the shape of a confocal
spheroid. However, recent numerical studies of Mazataud (1995) and Sovik (1995)
have evidenced some defects of the models proposed. We shall endeavour here to
solve these difficulties by proposing a more refined model, which will be obtained by
broadening the choice of possible trial velocity fields used in the limit-analysis.
The second extension to be discussed is quite different from the first one (and
no attempt will be made to combine the two). It consists in considering again a
hollow sphere as in Gurson's original work, but now subjected to conditions of in-
homogeneous boundary strain rate. The motivation here is as follows. It is well
known that the use of constitutive models involving softening, like that of Gurson,
Recent Extensions of Gurson's Model 63

raises difficulties connected with the tendency of damage to concentrate within lay-
ers of measure zero, resulting in a zero dissipated energy at failure. In practical
finite element simulations of structures undergoing ductile rupture, this problem is
generally obviated by putting a lower limit on the size of the elements. This means
that the element size is no longer a mere mathematical artifact but acquires some
physical meaning; namely, it represents the typical dimension of the "elementary"
volumes to which the "homogenized" model used is deemed to be applicable, which
is of the order of the void spacing. Another, more elaborate solution was proposed by
Leblond et aJ. (1994), following an earlier suggestion of Pijaudier-Cabot and Bazant
( 1987) made in a more general context; that solution was also adopted by Tvergaard
and Needleman ( 1995). It consisted in assuming the porosity (parameter governing
damage and therefore softening) to be of non-local nature through the appearance of
some spatial convolution integral in its evolution equation. However, both of these
solutions were of purely heuristic nature. In contrast, the solution proposed here will
be based on some micromechanical approach, just as Gurson's original work. More
specifically, we shall attempt to account for the possibility of non-negligible variations
of the macroscopic mechanical fields at the scale of the representative volume (that
is, over distances of the order of the void spacing), due to the tendency of damage to
localize, by incorporating more complex boundary conditions into Gurson's original
limit-analysis. As a result, the (mean) void spacing will appear as an extra material
parameter in the macroscopic constitutive equations.

PART I: GURSON-LIKE MODELS INCORPORATING THE INFLU-

ENCE OF VOID SHAPE

1.1. Overview
Just as in the works of Gologanu et al. (1993, 1994) and Garajeu (1995), we shall
endeavour to incorporate void shape effects into "homogenized" models for porous
plastic metals by extending Gurson's (1977) classical limit-analysis of a hollow sphere
to some axisymmetric prolate or oblate spheroid containing a spheroidal confocal
cavity. We shall essentially focus our attention on axisymmetric loadings, although
some generalization of the model to arbitrary ones will be presented at the end.
After some geometric preliminaries, we present the family of axisymmetric ve-
locity fields proposed by Lee and Mear (1992), duly supplemented with some extra
fields in order to account for the finiteness of our representative volume (the medium
considered by Lee and Mear was infinite).
We next explain how the exact macroscopic yield locus or some approximate
one can be derived from there. Following the Hill (1967) - Mandel (1964) classi-
cal approach, we enforce conditions of homogeneous boundary strain rate on the
64 M. Gologanu et al.

elementary "cell" considered. To find the exact overall yield locus, one must then
numerically solve a minimization problem involving a large number of parameters
tied through some linear constraints. The procedure to find an analytic, approxi-
mate yield function is simpler provided that two trial velocity fields only (as in the
works of Gurson, 1977, Gologanu et al., 1993, 1994 and Garajeu, 1995) are used.
Indeed the overall plastic dissipation can then be calculated as a function of the two
independent components (in the axisymmetric case envisaged) of the macroscopic
strain rate without any minimization, since the values of these components fully de-
termine those of the multiplicative factors of the trial fields, and the expression of
the approximate criterion readily follows from there.
The next section explains the choice of the two trial velocity fields for the theo-
retical analysis. One of these fields is the usual one corresponding to some uniform
deviatoric strain rate. The other one, which describes' the expansion of the void, is
an a priori free 1 combination of four Lee-Mear fields, whereas it was a fixed combi-
nation of only two in the works of Gologanu et al. (1993, 1994) and Garajeu (1995);
it is this enrichment that allows for the derivation of improved criteria. Even with so
few fields, some approximations (which are different for prolate and oblate cavities)
are necessary to obtain some fully explicit results. The coefficients of the superpo-
sition of Lee-Mear velocity fields defining the "expansion" field are "optimized" in
some sense at the end, and this defines the values of some of the parameters ap-
pearing in the analytic criterion. Those of the remaining ones are determined in a
slightly different way, by forcing the analytic yield locus to pass through some special
points which represent an exact solution in several limit-cases and especially that of
a "penny-shaped" crack. The resulting approximate criterion reproduces a number
of well-known exact, or approximate but widely accepted, results.
Comparisons with the results of some numerical minimizations using a much
larger number of trial fields given in the next section evidence the improvements
brought with respect to previous models. Two kinds of comparisons are made. First,
we compare the yield loci themselves. The new analytic criterion for oblate cavities
notably yields very good results even for a penny-shaped crack, in contrast to that
proposed by Gologanu et al. (1994). Second, we compare porosity rates, which
are directly tied to the direction of the normal to the yield locus. The new theory
predicts, in the prolate case, a much stronger influence of the shape of the void upon
its growth rate than before, in agreement with numerical results.
The problem of predicting the evolution of void shape is envisaged next. This

1 In a sense, this means that we are in fact using more than two trial fields. Our procedure does

preserve, however, the essential feature of a two-field approach, namely the fact that the components
of the macroscopic strain rate unambiguously fix the multiplicative factors of the trial fields, because
the momentarily free coefficients of the combination of Lee-Mear fields defining the "expansion" will
be fixed at the end independently of these components.
 Recent Extensions of Gurson's Model 65

is done first by using a two-field theoretical approach, like for the derivation of some
approximate criterion. However, the use of a velocity field corresponding to some
uniform deviatoric strain rate here proves quite insufficient to correctly reproduce the
numerically computed evolution of the void shape in the case of a purely deviatoric
loading. In the spirit of the works of Ponte-Castaneda and Zaidman (1994) and
Garajeu (1995), we then replace that field by the exact elastic solution of Eshelby
(for an infinite medium). This does improve the theoretical predictions, but these
predictions still fail to capture the strong influence of the macroscopic triaxiality,
which is a typically plastic effect. This influence is finally incorporated through the
introduction of some heuristic factor (the value of which is deduced from numerical
simulations) into the evolution equation of the shape parameter.
Comparisons are then presented with some finite element simulations of Maza-
taud (1995) and Sovik (1995) of the behaviour of cylinders containing an initially
spheroidal void and subjected to some axisymmetric loadings. The main difference
with respect to the numerical minimizations described earlier is that these simulations
follow the evolution of the representative cell in time. Also, the comparison with the
model developed does not involve a single feature of this model (the criterion, the
evolution equation for the shape parameter, ... ) but the whole of it. It is found that
its use instead of the old ones considerably improves the quality of the agreement
with numerical results.

We finally present a generalization of the analytic criterion to arbitrary stress

states, that can be used for practical finite element simulations of rupture of structures
made of ductile materials containing non-spherical voids.

1.2. Geometric and kinematic preliminaries

I.2.a. Spheroidal coordinates

We consider a spheroidal (axisymmetric) prolate or oblate cavity with semi-axes

a 1 (along Ox3) and b1 (along Ox 1 and Ox 2 ), embedded in a representative "volume
element" or "cell" which has the shape of a confocal spheroid with semi-axes a 2
(along Ox3) and b2 (along Ox 1 and Ox2) (Figure I.l). Let c = Ia~- b~l =
112

I a~ - b~ 112
denote the focal distance and e1 and e2 the eccentricities of the inner
1

and outer spheroids. Such a geometry can be characterized, up to some homothetical

transformation, by two dimensionless parameters, for instance the porosity f and the
shape parameter S defined by

(1.1)

The inner and outer eccentricities can be calculated in terms of these parameters
66 M. Gologanu et al.

through the formulae

_1_ = e2ISI. (1- eDn- ~ (1- ei)n - { 1 (p) (1.2)

1 - e~ ' e~ - f e~ ' n - 1/2 ( o)

where the symbols (p) and ( o) indicate the cases of a prolate or an oblate void
respectively.

Figure I.l: The geometry considered. Case a: prolate void; case b: oblate void

We shall incidentally use both cylindrical coordinates p, c.p, z (associated or-

thonormal basis ep,e<p,ez) and spherical ones r,B,c.p (associated orthonormal ba-
sis en e 8 , e'P). However we shall mainly employ spheroidal (orthogonal) coordinates
>.., fJ, c.p (associated orthonormal basis e>., e,a, e"') defined by

p = c sinh >.. sin fJ p = c cosh >.. sin fJ

{ { (o) (1.3)
z = c cosh ).. cos fJ (p) ; z = c sinh >.. cos fJ

with).. E [0, +oo[, fJ E [0, n] and cp E [0, 2n]. The iso->.. surfaces are confocal spheroids
with semi-axes and eccentricity denoted a, band e respectively and given by

a= c cosh>.. a= c sinh>..
{ b = c sinh>.. (p) { b = c cosh>.. ( 0 ). (1.4)
e = cfa = 1/cosh >.. e = cfb = 1/cosh >..

In particular, the surface of the void and the external boundary are iso->.. surfaces
corresponding to some values >.. 1 and >.. 2 respectively. The iso-/3 surfaces are hyper-
boloids orthogonal to the iso->.. surfaces. The nonzero metric coefficients for this
Recent Extensions of Gurson's Model 67

system of coordinates are given by

( 0 ).

(I.5)
I.2. b. The Lee-Mear velocity fields
Lee and Mear (1992) have proposed a family of axisymmetric incompressible
velocity fields "adapted" to spheroidal coordinates. The nonzero components of these
fields in the basis e>,, ef3, e'P are given by

v>,(A, ;3) = ( c2 / J9,U){Boo/sinh A

+ L:t~, 4 ... L:~~o k(k + 1) [BkmQ~(w) + CkmP~(w)] Pk(u)}
(p );
vf3(A,j3)= (c 2 /J9,U)x
L:t~, 4 ... l::~~l m(m + 1) [BkmQm(w) + CkmPm(w)J Pl(u)
(I.6)
U>,(A, (3) = (c2 / J9,\:\) {Boo/cosh A
+ L:t~, 4 ... L:~~o k(k + 1)im [iBkmQ;,(w) + CkmP~(w)] Pk(u)}
(a)
vf3(A, ;3) = (c ~ / J9,\:\) x
L:t~, 4 ... L~~1 m(rn + 1)im [iBkmQm(w) + CkmPm(w)] Pf(u)
where

_ { cosh A (p) . . _
w = i sinh A ( o) ' u = cos j3 · (I. 7)

In these expressions, the Pn, P~, Qn, Q~, are the associated Legendre functions of
the first and second kinds and of order 0 and 1 (see Gradshteyn and Ryzhik, 1980),
and the Bkm and Ckm arbitrary coefficients 1 .
A number of interesting particular fields can be obtained by specializing the
values of the coefficients Bkm and Ckm· For instance, the well-known Eshelby solution
for a spheroidal cavity in an infinite, incompressible elastic medium is a combination
of the four fields corresponding to the coefficients Boo, B 20 , B 22 and C 22 , all remaining
coefficients being zero. Other examples will be given below.

1 ln fact, the coefficients Ckm were omitted in the work of Lee and Mear (1992); this was
because the corresponding terms in the expression of the velocity failed to tend to zero at infinity,
as required for the infinite medium considered by these authors. These terms must be included here
since the representative volume studied is finite.
 68 M. Gologanu et al.

The nonzero components, in the basis e.A, e;3, e'f', of the strain rate d associated
to the Lee-Mear fields are given for instance in Gologanu (1996).

!.3. Principle of the determination of the macroscopic yield locus for an

axisymmetric loading
I.3.a. Constraints arising from boundary conditions
Following the Hill (1967) - Mandel (1964) approach to homogenization, we now
enforce conditions of (axisymmetric) homogeneous boundary strain rate on the rep-
resentative cell considered:

v(x) = D.x for,\= Az; Dn = D22 -:f. 0, D33 -:f. 0, other Dij = 0. (1.8)

The tensor D here is classically termed the macroscopic strain rate. Using spheroidal
coordinates and associated Legendre functions, one puts this equation in the form

v.A(,\ = Az, (3) = azbz [Dm + D~ 3 Pz(u)] / y'9)S..

{ (1.9)
v;3(,\ = Az,f3) = (a~D33- b~Du)P{(u)/(3y'9>:>..)
where the subscript m and the prime indicate the mean and deviatoric parts of a
tensor as usual (Dm =
(2Du + D33)/3, D~ 3 =
2(D33 - Du )/3). Comparison with
eqs. (1.6) then shows, since the Pk on the one hand, and the P1 on the other hand,
form independent families, that

c 3 Boo= azb~Dm; 6c 2 Fz(,\z) = azbzD~ 3 ; 3c 2 Gz(Az) = a~D33- b~Du; (I.lO)

Fk(>.z) = Gk(,\z) = 0 , k = 4, 6, 8, ... (1.11)

where

{
=I:;~';0 [BkmQ~(w) + CkmP;,(w)]
Fk(,\)
(p );
Gk(,\) =2:~:1 m(m + 1) [BkmQm(w) + CkmPm(w)]
(I.12)
Fk(,\) =I:;~:o im [iBkmQ~(w) + CkmP;,(w)]
{
Gk(,\) =L~:l m(m + 1)im [iBkmQm(w) + CkmPm(w)]
( 0 ).

Eliminating Du and D33 in eqs. (1.10) and expressing a 2 /c and b2 /c in terms of e2 ,

one concludes that the coefficients Bkm and Ckm necessarily verify, in addition to
(1.11), the following relation where the components of D no longer appear:

e~Boo/(3(1- eD) + (3- eDFz(,\z)/J1- e~ - Gz(Az) = 0 (p)

{ (1.13)
-e~Boo/(3J1- e~) + (3- 2eDFz(Az)/ J1- e~ - Gz(,\z) = 0 (o) .
Recent Extensions of Gurson's Model 69

One can check that if Bzz = Boo/6 in the prolate case or Bzz = -Boo/6 in
the oblate case, all other coefficients being zero, eq. (I.13I) or (I.13z) is verified
whatever the value of >. 2 • We thus get, with such a choice, some velocity field satisfying
conditions of homogeneous boundary strain rate on every spheroid confocal with
the void. The existence and uniqueness of such a field, up to some velocity field
corresponding to a uniform deviatoric strain rate, was established independently,
without any reference to Legendre associated functions, by Gologanu et al. (1993,
1994), who proposed it to describe the expansion of the cavity.
Also, if the only nonzero coefficients are Boo and Bzo and those coefficients
satisfy eq. (I.13!) or (!.13 2 ), we get some velocity field satisfying conditions of ho-
mogeneous boundary strain rate and having a zero f)-component, which means that
it is orthogonal to all spheroids confocal with the cavity. Again, the existence and
uniqueness (up to some uniform deviatoric strain rate) of such a field was proved
independently by Garajeu (1995) (for a prolate cavity), who used it as an "expansion
field".
I.3.b. Numerical determination of the exact yield criterion
We now suppose that the matrix is made of some rigid ideal-plastic von Mises
material whose flow rule obeys the normality property. The microscopic plastic dis-
sipation 1r( d) is classically defined, for every traceless d, by

( d) = S * . d = { O'odeq in the matrix (!.14)

7r - up 0' · 0 in the voids

=
where the supremum is taken over all plastically admissible u*, and O'o and deq (2d :
d/3) 1 12 represent the yield stress in uniaxial tension and the equivalent microscopic
strain rate, respectively. The following inequality then holds for all "macroscopic
stresses" :E and macroscopic strain rates D (Suquet, 1982):

:E: D:::; II( D):::: Inf (1r( d(x)))c (!.15)

where ( )c denotes an average value taken over the elementary cell C considered and
the infimum is taken over all incompressible velocity fields satisfying conditions of
homogeneous boundary strain rate corresponding to the macroscopic strain rate D.
The function II(D) is called the "macroscopic plastic dissipation" (Suquet, 1982).
For any set of parameters Bkm and Ckm satisfying eqs. (!.11) and (!.13), i.e.
any velocity field satisfying conditions of homogeneous boundary strain rate for some
(unspecified) D, one can compute the overall dissipation (1r( d(x) ))c corresponding to
the velocity field considered through numerical integration over C. Let us now assume
that we wish to determine the intersection of the macroscopic yield locus, which is
(in the axisymmetric case considered here) a curve in the plane (~u, ~ 33 ), with some
straight line of equation ~ 33 = X~u where X is a given parameter. Inequality (1.15)
 70 M. Gologanu et al.

then yields
" (1r( d(x)))c
::::}un<----------'--
- 2Dll + XD33
(I.16)
The problem is thus to determine the minimum of the right-hand side in the last
inequality over all sets of coefficients Bkm, Ckm satisfying the linear constraints (I.ll)
and (1.13). The values of D 11 and D 33 here can be deduced from those of these
coefficients by eqs. (I.lO). Since the quantity to be minimized is obviously positively
homogeneous of degree 0 with respect to the Bkm and Ckm, one can arbitrarily
decide that its denominator must be equal to unity; this introduces an additional
linear constraint on these coefficients. The quantity to be minimized is then simply
(1r(d(x)))c.
In actual computations, the values of k that we consider are 2, 4, 6, 8, 10 and
those of m vary between 0 and 4, which makes a total of 50 trial velocity fields.
I.3.c. Analytical determination of some approximate yield criterion
A classical consequence of eq. (1.15) is that the macroscopic yield locus can be
defined by the equation

(I.17)

where the components of D act as parameters. (In fact, the function II( D) being pos-
itively homogeneous of degree 1) the function (an I aD) (D) is positively homogeneous
of degree 0; thus it depends only on the ratios of 5 of the 6 components of D over the
last one, so that eq. (!.17) defines a 5-dimensional surface in a 6-dimensional space,
as required). Now let us consider the axisymmetric case and assume that we use only
2 trial fields for simplicity. Since the loading is also characterized by 2 parameters,
D 11 and D 33 , these parameters fully determine the multiplicative factors of the trial
fields. Hence the calculation of the upper estimate of II(D) resulting from the varia-
tional approach, which we shall still denote II(D) for simplicity, is greatly simplified
because it no longer requires any minimization. (We shall see, however, that even
so, it remains a difficult task that will require some approximations). Formula (1.17)
then yields some parametric equations of the estimated yield locus. Elimination of
the components of D in this formula, provided that it reveals feasible, then yields
the analytic, explicit (non-parametric) equation of that yield locus.

1.4. An analytical two-field estimate of the macroscopic yield criterion for

an axisymmetric loading
I.4.a. Choice of trial velocity fields
Just as in the work of Gurson (1977), we now decide to use only 2 trial fields
denoted vA and v 8 . The second one will simply be that corresponding to some
Recent Extensions of Gurson's Model 71

uniform deviatoric (axisymmetric) strain rate:

(1.18)

(In Lee and Mear's decomposition, this is the field corresponding to the coefficient
C 22 ). Thus we do not follow here Garajeu's (1995) suggestion to replace Gurson's
simple field vB by the exact elastic solution for a purely deviatoric loading. The
reason is that as mentioned in the Introduction, the improvement brought to the
approximate yield criterion is important only for high porosities, and we are mainly
interested in small ones here. (In contrast, such a replacement will be seen to be
essential in the search for the evolution equation of the shape parameter).
We now have to choose some field vA representing the expansion of the void.
We shall be led here by the following arguments. Firstly, eq. ( I.lOI) shows that Dm
is tied solely to the coefficient B 00 ; in other words, all fields corresponding to the
other coefficients involve some change of shape of the void but no expansion. Hence
it is necessary to take B 00 /:- 0 to describe that expansion. Secondly, some of the
coefficients B 2 m or C 2 m must then also be nonzero, otherwise condition (1.13t) or
(1.13 2 ) would not be satisfied, resulting in the violation of conditions of homogeneous
boundary strain rate, whatever the value of D. However, conditions (1.13) do not
forbid to take Bkm = Ckm = 0 for k = 4, 6, 8, ... and the same is true of eqs.
(1.11 ). Thirdly, in the important particular case of a vanishingly small porosity, the
expansion field looked for must tend to zero at infinity. This imposes to take nil
coefficients Ckm in that case, since they are multiplied, in the expression (1.6 2 ) or
(1.6 4 ) of Vf3, by the polynomials Pm(w), which of course do not vanish at infinity
(contrary to the non-polynomial functions Qm(w)). Fourthly, acceptable (in a first
approximation) expansion fields were proposed by Gologanu et al. (1993, 1994) and
Garajeu (1995) (for prolate cavities) by taking B 00 and B 22 in the first case, B 00 and
B 20 in the second one, to be the only nonzero coefficients. Fifthly, as already men-
tioned, in the famous Eshelby elastic solution, the only nonzero coefficients are again
B 00 , B 20 and B 22 (plus a term proportional to C 22 , but which represents a uniform
deviatoric strain rate, not an expansion of the cavity). All of these elements suggest
to take B 00 , B 20 , B 22 and by "interpolation" B 21 , as the only nonzero coefficients.
The coefficient B 00 is taken as unity in order to "normalize" the velocity field. The
values of the coefficients B2o, B21, B22 will not be specified for the moment. (As
already mentioned in the Overview, they will be fixed at the end independently of
those of D 11 and D 33 , so that the values of these components will completely fix
those of the multiplicative factors - noted A and B in the sequel - of the velocity
fields vA and vB, in agreement with what was said in Section 1.3.c). This (momen-
tary) liberty implies greater possibilities of choice than in the works of Gologanu et
al. (1993, 1994) and Garajeu (1995), which will reveal essential in the search for an
improved model solving the difficulties evidenced by the finite element simulations of
Mazataud (1995) and Sovik (1995) (see Section 1.7 below).
 72 M. Gologanu et al.

1.4. b. Two-field variational approach

We now consider trial velocity fields of the form

(1.19)

where A and B are arbitrary parameters. With obvious notations, the square of the
equivalent microscopic strain rate is given by

(1.20)

The value of d1q 2 can be deduced from those of the components of dA in spheroidal
coordinates (see Gologanu, 1996). Also,

4 A 4 1 A 1 A
B2
deq =1 ; 3d : d
B
= 3( -2d 11 - 2d22 + d33A ) = 2d33A (1.21)

df
where the property tr( dA) = 0 has been used. The value of 3 here can be obtained
from those of df>.., a:f], dff] through the usual transformation formulae from one
basis to another. Finally, the expression of the elementary volume in spheroidal
coordinates is dv = b gu sin j3 d>. dj3 d<p. The expression of the upper estimate of
the macroscopic dissipation is then

II(D) =
27ra [>..2
/C/ 0 }>., d>.
ltr
0
deq b 9>..>.. sin j3 dj3 (1.22)

where /C/ =47ra bV3 denotes the volume of the elementary cell C considered.
2

The components of the macroscopic strain rate associated to the field vA are
easily expressed in terms of F 2 (>. 2 ) using eqs. (I.lOI) and (1.10 2 ). Adding those due
to the field v 8 , one gets

Du = D22 = ADf1 + BD~ =A [c 3 /(azbD- 3c 2 Fz(>.z)/(azbz)]- B/2

{ (1.23)
D33 = ADf3 + BDfa =A [c 3 /(azbD + 6c 2 Fz(>.z)/(azbz)] +B.

The macroscopic stresses are given by eq. (1.17). It follows from that equation and
the preceding expressions of D11 and D33 that

Let us introduce the notation

(1.24)
Recent Extensions of Gurson's Model 73

where eqs. (I.23) have been used. The preceding expression of 8TI/ f) A then becomes

all 3c3 47rc3

f)A = a 2 b~ :Eh = lCf:Eh , :Eh = 2o:2:Eu + (1- 2o:2):E33 . (I.25)

A similar calculation yields

f)TI B B
f)B = 2Du :Eu + D33 :E33 = :E33 - :Eu . (1.26)

Equations (I.25) and (I.26) are the parametric equations of the approximate macro-
scopic yield locus (both 8TI/ f) A and 8TI/ f)B being functions of the ratio A/ B, which
acts as a parameter). But of course, for their being of any usefulness, one must carry
out the explicit calculation of the macroscopic dissipation given by (1.22).
I.4.c. Approximate yield criterion for prolate cavities
In fact, this calculation is untractable unless some approximations are made.
Unfortunately, those which will now be introduced are "uncontrolled" except in some
special cases like that of spherical cavities, which means that it is not clear in general
whether they do or do not preserve the "upper bound" character of the approach.
The first one is as follows:
A 1 : After having performed the change of variable u =
cos f3 in eq. (!.22), one
replaces u 2 by its average value over the interval of integration [-1,+1], i.e. 1/3.
This was exactly the approximation made by Gurson (1977) in the case of a spherical
cavity, which appeared as "uncontrolled" in his work but was in fact proved by
Leblond and Perrin (1995) to lead to some rigorous upper bound for the macroscopic
yield locus of a hollow sphere. It was also used in the previous works of Gologanu et
al. (1993, 1994) on void shape effects, and found to be quite acceptable for all values
of the shape parameter.
We now introduce the following change of variable in the integral of the right-
hand side of eq. (1.22):

(1.27)

where

deq(x) = y'A2 P(x) + B 2 + 2ABQ(x),

(1.28)
P(x) = [d1/(x)]u2=1 / 3 , Q(x) = [d:3(x)L2=l/ 3 .
All equations written up to now are valid for both prolate and oblate voids. From
now on, we consider (in this section) only prolate ones. The next step is to study the
74 M. Gologanu et al.

behaviour of the function P( x) in the limit-cases of spherical and cylindrical cavities.

For a spherical one, c--+ 0 and a~ b ~ r so that, by eq. (1.271), x--+ 0 (for fixed r).
In contrast, for a cylindrical one, c ~a--+ +oo and b ~ p so that x --+ +oo (for fixed
p). Thus this study will provide informations about the behaviour of the function
P(~) at the ends of its domain of definition. After some lengthy but elementary
calculations, we get

for x--+ 0 (spherical cavity)

for x --+ +oo (cylindrical cavity) .

On the other hand, if we refer to the work of Gurson (1977), we find that with obvious
notations,

pGurson( ) = { 4x 2 for x -t 0 (spherical cavity)

x - 3x 2 for x --+ +oo (cylindrical cavity) .

We want to recover Gurson's criterion in the special case of a spherical void, because
of the well-established qualities of that criterion. Therefore comparison of the preced-
ing expressions shows that one ought to take B 20 = 0 for such a void (characterized
by S = 0) 1 . However, P(x) behaves like x 2 for both x --+ 0 and x --+ +oo inde-
pendently of the value of B 20 , and this strongly suggests to introduce the following
approximation, whatever the shape of the void:
A 2 : P(x) being written in the form P(x) =F (x)x2, the function F(x) is replaced
2

by some constant F (to be defined precisely later) in the expression of deq( x ).

A similar study of the function Q( x) yields the following results:

{
=
Q( x) -4B2o ; QGurson( x) =0 for x --+ 0 (spherical cavity)
Q(x) := 0; QGurson(x) := 0 for x - t +oo (cylindrical cavity).

Again, we find that one should take B 20 = 0 for a spherical cavity in order to recover
Gurson's criterion in that case. These results make it tempting to completely neglect
the "crossed" term 2ABQ( x) in the expression (1.28 1 ) of deq( x ). This was exactly
what was done by Gologanu et al. (1993) and Garajeu (1995) in their previous studies
of prolate cavities. However, comparison between the approximate yield criteria
derived in that way and the (supposedly exact) criterion obtained numerically (see
Section 1.5 below) will show that the effect of that crossed term is not completely
negligible and that a better analytic criterion may be obtained by accounting for it.

1 Although the simple notation used does not underline it, the quantity f( X) defined by (1.282)

depends in fact also on S and f. Indeed the values of the coefficients B2o, B21, B22, which are
left free for the moment, will be fixed at the end and are bound to depend upon the geometric
parameters of the problem.
 Recent Extensions of Gurson's Model 75

One way to do it was proposed by Gologanu et al. (1994), in the context of oblate
cavities, but it is easy to adapt it to the case of prolate ones. It consists in writing
the term 2ABQ(x) and the expression (!.281) of deq(x) in the form

2ABQ(x) = 2ABF(x)G(x)x 2 , G(x) = Q(x)/(F(x)x 2);

(I.29)
deq(x) = J[AF(x) + BG(x)Fx 2 + B 2H 2(x), H(x) = y'1- G2(x)x2,
and in introducing the following final approximation:

A 3 : The functions G( x) and H ( x) are replaced by some constants G and H in tbe

expression of deq(x ).
This seemingly gross approximation may be justified in the following way. From
what precedes, one expects the effect of the crossed term in the expression of deq ( x)
to be weak, which means that the functions Q( x) and G( x) must be "small". Thus
the effect of the terms BG(x) and -G2(x)x 2 in the expressions AF(x) + BG(x) and
y'1 - G2( x )x 2 must be weak, and it is therefore reasonable to account for it in the
simplest possible way by replacing these terms by mere constants in the interval of
integration [x2,x 1 ] of the integral in the right-hand side of eq. (!.272)·
From that point, the calculations become almost identical to those of Gurson
(1977) for the spherical case. With approximations A2 and A3, the overall dissipation
takes the simple form

(I.30)

Calculation of that integral, then of the derivatives 8ITI 8A' and BIT I 8B' and elimi-
nation of the ratio A' I B' between the expressions found yields

-1 ( -arr ) 2 +2fcosh ( -1- -arr

-) -1-f 2 =0.
a5 8B' aox2 8A'

Finally, expressing 8IT/8A' and 8IT/8B' in terms of 8IT/8A and 8IT/8B using eqs.
(!.302, 3 ), and then in terms of the macroscopic stress components using eqs. (!.251)
and (!.26), one gets

(I.31)

where
C =11 H 2 ; ry = -3x2 G IF "' =31 F (I.32)
76 M. Gologanu et al.

Equation (I.31) represents the approximate, analytic criterion proposed; it only re-
mains to ascribe precise values to the parameters C, ry, ~ and 0:2 (which appears in
the expression (!.25 2) of 'Eh)i this will be done in Sections I.4.d and I.4.f below.
I.4.d. Approximate criterion for oblate cavities
As already mentioned, provided that one introduces the same approximation A 1
for oblate voids as for prolate ones, eqs. (I.27) and (I.28) remain valid for oblate
voids. Again, the next step is to examine the behaviour of the function P( x) for
both x ---+ 0 and x ---+ +oo. The first limit again corresponds to a spherical cavity; it
was studied above, and the conclusion was that P( x) behaves like x 2 then and that
one should take B 20 = 0 for such a cavity. The seco;nd limit no longer corresponds
to a cylindrical void but to a "sandwich" consisting of two plane, parallel layers of
material separated by an empty space; indeed in that case a :::::: z and c :::::: b ---+ +oo, so
that x---+ +oo by eq. (I.27t). The investigation of the behaviour of P(x) in that limit
is quite involved and has been carried out with the aid of the symbolic computation
program Maple. The result is that this function admits a finite limit given by

This implies that the approximation A 2 made for prolate cavities, which consisted
in replacing the function P( x) by some constant times x 2 , is not applicable to oblate
cavities because it is inappropriate for large values of x.
A solution to that problem was proposed by Gologanu et al. (1994); it consisted
in adopting a suitable homographic change of variable. Here we define the variable

y=--
X
( dy = dx) (I.33)
::::} 2 2 .
x+1 y X

=P(y) behaves like y

For x ---+ 0, y "" x and for x +oo, y ---+ 1. Thus, if considered as a function of y,
---+
P( x) for y ---+ 0 and tends to some finite limit for y ---+ 1. We
2

then introduce the variant, noted A~, of approximation A2, which is identical to that
approximation except for the replacement of the variable x by the variable y. Clearly,
it is the fact that the upper limit of the latter variable is 1 (instead of +oo for the
former) which makes the replacement of P(y) by some constant times y 2 reasonable
near that upper limit. But there is another advantage to the change of variable
(I.33t ), namely the fact that it preserves the form of the element of integration in the
integral of the right-hand side of eq. (!.27 2 ) (see eq. (1.33 2 )). This feature implies
that the simplified integral we shall obtain will again be of the same type as in the
work of Gurson, and thus that it will lead to some fully explicit, analytic approximate
criterion.
Recent Extensions of Gurson's Model 77

Similarly, with regard to the function Q( x ), we already know that it is zero in

the limit x --+ 0, and it can be checked that just like P( x ), it has a finite limit for
x --+ +oo given by

Q(x =+oo) = 12Bzl + 9Tr B22 .

This limit has no reason to be zero (in contrast to what happens for prolate cavities)
so that it becomes essential here to account for the crossed term 2ABQ( x) in the
expression of deq ( x). This is done in the same way as in the prolate case, except for
the replacement of the variable x by the variable y : we again define some functions
G(y) and H(y) by eqs. (!.29) with y instead of x and introduce the variant A~ of
approximation A 3 which is identical to that approximation but for the same replace-
ment. It can verified numerically that although the importance of the terms BG(y)
and -G 2 (y)y 2 in the expressions AF(y)+BG(y) and \/1- G2 (y)y 2 is much greater
than in the case of prolate cavities, replacing the functions G(y) and H(y) by some
constants remains reasonable. The fact that the upper limit of the variable y is 1
instead of +oo again plays a role here in that it lessens the importance of the term
y 2 near that upper limit.
From there, one follows the same steps as in the case of prolate voids. The
expression of II(D) is given by (I.30) with Yl, yz, y instead of x1, x 2 , x in the
integral (but not in the multiplicative factor 0" 0 x 2 ). Calculating that integral, eval-
uating aiii aA' and aiii aB'' eliminating A' I B' between the expressions found, and
expressing arr I aA' and arr I aB' in terms of arr I aA and arr I aB' then in terms of the
macroscopic stress components, one finally gets the following approximate criterion:

where
c3 4Trc 3 13 e~
g =Xz =azb~ = ICI =y'l- e~ (!.35)

and C, ry, K are given by the same formulae (I.32) as in the prolate case. The new,
seemingly superfluous notation g for the quantity x 2 is introduced here to underline
the fact that this parameter plays the role of some kind of "second porosity" in the
yield criterion (!.34). In fact, eq. (!.353) shows that it can be interpreted as the
fictitious porosity obtained by replacing the real spheroidal void by a spherical one
with radius equal to the focal distance c.
Just as in the case of prolate cavities, it remains to give precise values for the
parameters C, ry, K and a 2 .
I.4.e. Determination of the coefficients K and a 2
We now wish to derive some appropriate expressions for the coefficients K and a 2
which appear in the "cosh" term of the criterion, and first for K:, which is tied to the
 78 M. Gologanu et al.

constant F through eq. (!.32 3 ). The best possible value for F is that which, when
substituted for F(x) or F(y) in the integral defining II(D), yields the exact value for
that integral. Since this condition obviously cannot be satisfied for all values of the
ratio A/ B simultaneously, we just impose it forB= 0. This leads to

{
fxx21 F dxx =J:21 F(x)d: * F= -(Cnf)-1 fee21(3- e2)F(e)e(1~e2) (p)
(!.36)
fYt
Jy2 p t!JJ.y =Jy2 F(y)t!JJ.y *
fYl p = (Cn(yl/yz))-1 Jy2fYt F(y)dyy (o)
where use has been made of the relation x = c3 /(ab 2 ) = e3 /(1- e2) in the prolate
case.
The function F(x) or F(y) is tied to P(x) =
[d~/(~)L 2 = 1 / 3 and therefore de-
pends on B 20 , Bz 1 and Bzz; the time has now come to choose the values of these
parameters. One way to make such a choice is to minimize (1r(dA(x))) 0 , or equiva-
J:
lently the integral J:21 F(x)dx/x or 21 F(y)dyjy, with respect to these values, under
the constraint (!.13) which imposes that conditions of homogeneous boundary strain
rate be satisfied for some (unspecified) D. The physical meaning of this minimization
is not· as clear as it may seem at first sight, precisely because D is not fixed in the
procedure, whereas it is fixed in the definition (!.15 2) of II(D). In fact, it means
looking for an approximation of the purely hydrostatic stress (~ L;m 1) which lies =
on the yield locus. Indeed, for such a stress, the plastic dissipation inequality (I.15)
reads
(!.37)
Hence finding the maximum possible value of L;m (that which promotes plastic flow)
is equivalent to minimizing (7r(d(x)))c/(3Dm) or, as already discussed, {1r(d(x)))c
itself, with the additional constraint that Dm be fixed. But, by eq. (I.lOI), fixing Dm
was precisely what we did when we "normalized" the field vA by setting B 00 = 1.
Therefore minimizing (1r(dA(x))) 0 with respect to B 20 , B 21 and B 22 is equivalent to
looking for an upper bound for the purely hydrostatic stress lying on the yield locus
through some variational approach based on the restricted class of velocity fields
defined by eqs. (I.6) with B 00 = 1, B 20 , B 21 and B 22 as the only nonzero coefficients.
Unfortunately, performing such a minimization analytically reveals an impossible
task, and we are therefore forced to do it in an approximate way; the approximations
introduced are different in the cases of prolate and oblate cavities.
In the prolate case, our approach is based on the following observation, which has
been established with the aid of the program Maple: for any fixed value of e E [e2, e1] ,
the minimum value ofF( e) for all parameters B2o, Bz1, B22 satisfying the constraint
(I.13) is independent of e2 (in spite of the fact that this quantity appears m eq.
(I.13)) and given by

[F(e)]min = ~
2V3 . (1.38)
v3 + e4
 80 M. Gologanu et al.

and (1.36):

- g
gl = --.
g+1
(1.41)
With regard to the parameter o: 2 , which is given by (1.24) and thus depends on
the parameters B 20 , B 21 and B 22 , the simplest approach consists in calculating it by
adopting for the values of these parameters those (provided by the program Maple)
which achieve the minimum of F(e) under the constraint (I.13) fore= e2 • This leads
to
(1 + e~)/(3 + e~) (p)
{ (1.42)
2
o: = (1- eD(l- 2eD/(3- 6e~ 4e~)
+ (o) ·

The only way to assess the accuracy of these formulae is to check them against
supposedly exact results derived through some numerical minimization of the overall
dissip?-tion based on a large number of trial velocity fields, as sketched in Section
I.3.b. But, since the definition (1.24) of o: 2 makes reference to the "expansion field"
vA, this raises the problem of giving a precise definition of that field: we cannot
just accept here the "definition" used up to now that it is a field of type (1.6) with
B 00 = 1, B 20 , B 21 and B 22 as the only nonzero coefficients, first because it is too
vague and second because the other coefficients should precisely not be set to zero a
priori in a numerical minimization procedure. Two possibilities seem envisageable at
first sight: defining vA as the velocity field corresponding to some purely hydrostatic

=
macroscopic strain rate (D =:Dml) or to some purely hydrostatic macroscopic stress
(:E Em 1 ). To choose between these two alternatives, it suffices to consider the case
of a cylindrical void. It is then clear that vA should be identical to Gurson's field
defined by Vp = Cst.j p, v'P = Vz = 0. The corresponding D is not hydrostatic since
Daa is zero contrary to Du and D22· On the other hand, :E is purely hydrostatic:
this can be deduced from Gurson's criterion (which is exact for such a geometry
and an axisymmetric loading) plus the fact that the macroscopic flow rule obeys the
normality property (see Gurson, 1977). We are thus left with the only possibility of
regarding vA as the velocity field corresponding to a purely hydrostatic macroscopic
stress. By eq. (!.37), the problem is then simply to determine the minimum of
("7r(d(x)))c among all fields of type (!.6) subject to the constraints (I.11) and (I.13)
plus the additional condition that Dm, or equivalently B 00 , be fixed. (Note that even
the coefficient c22' which represents a uniform deviatoric strain rate, is not a priori
set to zero in the procedure).
Figure !.3 shows various functions o: 2 (e 2 ) for both prolate and oblate cavities.
These functions include those determined numerically for two porosities in the form
of discrete points, plus several theoretical ones represented by lines: those defined
 Recent Extensions of Gurson's Model 79

(Of course, the values of B2o, B 21 , B 22 which achieve the minimum depend on e).
It so happens that whatever the values of x 1 and x2, or equivalently e1 and e2,
=
the function [F(e)]min gives a fair approximation of the true function F(e) F(x)
(determined numerically) which minimizes the integral 2 F( x )dx I x with respect to
1

B2o, B21 and B22; this is illustrated in Figure I.2.a for typical values of e1 and e2.
J:
This figure also shows the approximation of F(e) adopted by Gologanu et al. (1993),
which is seen to be poorer than that given by [F(e)]min· Provided that the latter
approximation is adopted, eqs. (1.32 3) and (1.36) yield the following value for the
parameter K :

_1 _ 1 1 [( ro3 2) 0 e1 1 0 3+e~+2J3+e~ 0 v'3+J3+efl

"' - - + - vJ - .c.n- + -.c.n + .c.n--::,.---=--:===
v'3 Rnf e2 v'3 3 + ei + 2J3 + ef v'3 + J3 + e~
(1.39)

7 /
F(e) F(y) /
2.00 -- --::. . =--:- -
.....
a 6
--crack, g=0.14 (exact)
/
/
b
..... - - - [F(y)]
' '' '
app
1.95 /

' ' .\ 5 - ·- ·- Gologanu et al. /

/

1.90 exact ' \

'· \
/

1.85 [F(e)]min '· \

4
Gologanu et al. '·''.\
1.80 \ 3
\
1.75 e 2
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

Figure 1.2: Various functions F(e) or F(y). Case a: prolate void; case b: oblate void

The same approach unfortunately reveals impossible for oblate cavities, because
the function [F( e)]min is found to give a poor approximation of the true function F(y)
rl
minimizing Y2 F( y )dy I y' as determined numerically. However' it so happens that
this true F is relatively independent of the parameters Y1 and Y2 and, at least when
considered as a function of the variable y, can be approached by a simple function,
namely

[F(y)]app := 2(1 + Y + 2y 5 / 2 - 3y 5 ) ; (1.40)

this is illustrated in Figure I.2.b, which also shows the poorer approximation proposed
by Gologanu et al. (1994). With this replacement, one gets for K, using eqs. (1.323)
Recent Extensions of Gurson's Model 81

by (!.42), plus those proposed by Gologanu et al. (1993, 1994) and Garajeu (1995) 1 ,
the last of which are given by

CiGar- { 1/(3- eD (p)

(!.43)
2
- (1- e~)/(3- 2e~) (o) .

The numerical results shows that the approximation implicitely made in eqs. (!.42)
and (!.43) that a 2 is independent of f is reasonable. Also, these numerical results
being taken as a reference, Garajeu's (1995) proposal (eqs. (!.43)) appears as better
than our previous one but poorer than our present one (eqs. (!.42)). In particular,
in the oblate case, the last proposal is the only one that reproduces (although not
very faithfully) the "valley" that can be observed in the numerical results for large
eccentricities.

0.50 0.4
a2 a b
this paper
~ I I
Gologanu et al. I I
' ' '
0.45
''
Garajeu
I I 0.2 \
I I <> f=0.01 \ \
<> f=0.01 I I
I I !::. f=0.001
0.40 !::. f=0.001 'D. / I
<t.
/ /
0. this paper <>
/ /
/
Gologanu et al. ~ 0
0.35
Garajeu
e2 -0.2
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8

Figure 1.3: Various functions a2(e 2). Case a: prolate void; case b: oblate void

I.4.f. Some special points belonging to the exact two-field yield locus
The yield locus defined parametrically by eqs. (I.25) and (I.26) with II(D) given
by (!.22), which is obtained by considering only the two velocity fields vA and v 8 ,
but without any further approximation like A 1 , A 2 , etc., will be called the "exact
two-field yield locus" in the sequel 2 • It so happens that one can calculate analytically
some interesting points on that yield locus, namely those corresponding to A = 0.

1 In fact, as mentioned earlier, Garajeu considered only the case of prolate cavities; but it is
easy to define the equivalent of his field vA (which determines a2) for oblate cavities by requiring
it to be orthogonal to all spheroids confocal with the cavity, as he did in the prolate case.
2 It is approximate only in that it considers no more than two trial fields.
82 M. Gologanu et a!.

Indeed one gets in that case using eqs. (1.20), (1.21 2 ), (1.22), (1.25), (1.26) and (1.271)
and Gauss' theorem:

3x, E h = g~ :rr 1~ : egn( B )df, dv ~ 'gn( B) 1c1 (Is, vf!n, da - fs, vf!n, da)
{
:E33- :Eu - aB - ICI JM sgn(B)dv- sgn(B)a 0(1- f)

where sgn denotes the sign function, M the domain occupied by the matrix (de-
fined by )q :::; ). :::; ).2)' sl and 52 the surfaces ). = AI and ). = ).2 respectively,
and n the unit normal vector to these surfaces, oriented outwards. The expression
b
1 1
fs 2 vtn3.da is nothing else than the component D13 of the macroscopic strain rate
corresponding to the field VA. Similarly, if one defines the strain rate nv of the void
in a natural way by the formula

(1.44)

one sees that b fs

1
vtn3da is nothing else than f D3'{ By eqs. (1.101) (with Boo = 1
1 1

for the field vA) and the definition (1.24 1 ) of a 2 , D~ can be written in the form
3x 2 (1- 2a 2 ). Let us similarly define the quantity

DvA
a = (1.45)
2DvA + DvA
11
1-
1133

One then has f D3t = f(2D~f+D3't )(1-2ai) = (2Dii. +D~)(1-2ai) = 3x2(1-2ai)

where the incompressibility of the matrix has been used. Gathering these elements,
one concludes that the points defined by

(!.46)

belong to the exact two-field yield locus.

In general, these points have no reason to lie on the true yield locus. However,
there are some cases where they do, provided that a 1 takes suitable values. First,
for a cylindrical void (prolate case, f arbitrary, e 1 = e2 = 1), a 2 = 1/2 by eq.
(1.421); therefore, by the definition (1.25 2 ) of :Eh, provided that a 1 is also equal to
1/2, the points defined by (1.46) can be written as (:E 11 , :E 33 ) = [0, ±a0 ( 1 - !)].
These points are exact and correspond to some uniform axisymmetric deviatoric
strain rate, i.e. to the velocity field vB. Second, for a "sandwich" (oblate case, f
arbitrary, e 1 = e2 = 1), a 2 = 0 so that provided that a 1 is also zero, eqs. (!.46)
reduce to (:E 11 ,:E 33 ) = [=fa 0(1- !),0]; again, these are exact points obtained for a
velocity field identical to vB. Third, for a "penny-shaped crack" (oblate case, f = 0,
Recent Extensions of Gurson's Model 83

e1 = 1, e2 arbitrary)\ provided that a1 = 0, eqs. (1.46) yield (~n, ~33) = (=fao, 0),
which are again exact points obtained for the field vB. Also, recall that the points
defined by (1.46) were obtained within the two-field approach by setting A = 0, i.e.
by precisely taking vB as the trial field; in other words, the velocity field considered
by the two-field approach is also exact in the special cases just detailed. Thus, in
these special cases, and provided that a 1 takes suitable values, the exact two-field
locus not only meets the true yield locus at the points defined by (1.46), but is even
tangent to it there.
I.4.g. Determination of the coefficients C, ry and a1
We now come to the problem of giving formulae for the coefficients C and ry
which appear in the "square" term of the criterion (1.31) or (1.34). The most natural
approach here would consist in looking for some approximations for the constants
G and H appearing in their definitions (1.32 1 , 2 ), as we did for the parameter "'·
However, numerical experience reveals that a closer agreement with the true yield
locus can be obtained by forcing the approximate one defined by (1.31) or (1.34) to
meet the exact two-field yield locus at those points defined by (1.46) and be tangent
to it there, whatever the values of the geometric parameters f and S. (This is not
surprising, since by what precedes, this obliges the approximate yield locus to pass
through some exact points with the correct tangent in the special cases detailed
above). These conditions yield a system of two equations on the unknowns C and 1],
the solution of which is

"'Q(g + 1)(g + f)sh

ry=-----~------~~--~=-~~--~------~
(g + 1)2 + (g + !)2 + (g + 1)(g + f)["'Hsh- 2ch] '

( 1.47)
+ 1)(g + f)sh
C =-
K(g
(Q+ ryH)ry ; sh =.
smh(~~:H), ch = cosh(~~:H),

where g is conventionally taken as zero in the case of a prolate void and defined by
(1.35) in that of an oblate one, and Hand Q denote the special values of ~hfa 0 and
(~33- ~n)/ao given by (1.46).
Since both C and ry depend on a 1 , one must now give some formula for that
parameter. We make here the assumption that just like a 2 , a 1 is nearly independent
of f and thus depends only on e1. The function a 1 ( e 1 ) can then be determined by
considering the representative cell as infinite and therefore spherical(!- 0, e2 - 1).
In that case one can check that the factors of the coefficients B 20 and B 21 of the

1 There is an equivalent configuration in the prolate case, namely the infinitely thin, "needle-
shaped" cavity; but it is of no interest since the void has no effect then on the overall mechanical
behaviour.
 84 M. Gologanu et al.

field vA in eqs. (I.13) tend to infinity, which suggests to take them as zero. One
can then evaluate a 1, using its definition (!.45), as a function of B 00 = 1 and B 22
only; the calculation of the surface integrals defining the components of DvA (see eq.
(!.44)) is feasible analytically using a formula for associated Legendre functions due
to Neumann (see Gradshteyn and Ryzhik, 1980). It is found that a 1 is independent
of the value of the coefficient B 22 and given by

[e1- (1- ei)tanh- 1e1] /(2eD (p)

{ (I.48)
a1= [-e 1(1-ei)+J1-eisin- 1el]/(2ei) (o) ·

These formulae are identical to those proposed by Gologanu et al. (1993, 1994), but
in these works we also used them for the function a 2 (e 2 ), which we no longer do here.
It is easy to check that they verify the properties a1 = 1/2 for a cylindrical void and
a 1 = 0 for a "sandwich" and a penny-shaped crack, which were shown above to be
necessary for the points defined by (I.46) to belong to the exact yield locus for those
configurations.
The accuracy of this proposal can be assessed through comparison with some
numerical results, just as for the parameter a 1. Figure I.4 shows this comparison.
The numerical results (discrete points) show that a 1 depends only quite weakly on
f, as we assumed, and are well reproduced by formulae (I.48).

0.50 a1 b
-this paper 0.30 a1
0 f=0.01 0.25
0.45
L. f=0.001
0.20 0 f=0.01

0.40 0.15 L. f=0.001

0.10 - this paper

0.35 0.05
e1 e1
0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

Figure I.4: Various functions a 1 (e!). Case a: prolate void; case b: oblate void

I.4.h. Special cases

For a spherical void, it is easy to check that a1 = a2 = 1/3 =? ~h = ~m =
tr(:E)/3, g = 0, K = 3/2, ry = 0, C = 1, so that both criteria (I.31) (for prolate voids)
and (!.34) (for oblate ones) reduce to the usual "spherical" Gurson criterion. For a
 Recent Extensions of Gurson's Model 85

cylindrical one, a 1 = a2 = 1/2 ::::} Eh

=En, g = 0 (by definition), K = J3, ry = 0,
C = 1 so that the criterion (!.31) again reduces to Gurson's "cylindrical" criterion,
which is recalled to be exact for such a geometry and axisymmetric loadings. For
a "sandwich", g becomes infinite. Expanding the criterion (!.34) in powers of that
parameter, one finds that it reduces to lEn I :::; O"o(1- f), E33 = 0, which is trivially
exact (for axisymmetric loadings). Finally, for a penny-shaped crack, f = 0 so that
(!.34) reduces to

C
z(E33- En+ ryEh) 2 + 2g(g + 1)cosh ( K - Eh) - (g + 1) 2 - g2 = 0.
~ ~

Recall that because of the very method we used to evaluate the parameters C and 17,
this yield locus passes through the exact 1 points defined by (!.46), with the correct
tangent. It resembles that of Gurson, especially for small values of the parameter g,
which plays the role of an "equivalent" porosity (the real one being zero). We have
seen that it is in fact identical to the porosity defined by a spherical void with radius
equal to the focal distance, i.e. to the radius of the crack in the present case. This
provides some qualitative support to Pineau and Joly 's ( 1991) assumption that one
could use Gurson's usual "spherical" criterion in the case of penny-shaped cracks by
just replacing the cracks by spherical voids with the same radius. Pineau and Joly's
hypothesis is not completely correct, however, notably because it implies assuming
that the stress which essentially governs void growth is Em, and this is clearly wrong
for cracks which are more "sensitive" to E33 than to ~11 = E22·

1.5. Comparisons with some numerical 1ninimizations using a large num-

ber of velocity fields
I.5.a. Comparison of yield criteria
The principle of the numerical simulations has been expounded in Section I.3.b.
Figure I.5.a shows various criteria for prolate cavities with aspect ratio atfb 1 = 5
and several porosities: numerical (supposedly exact) ones in the form of discrete
points and theoretical ones in the form of lines. The dotted line represents our
former proposal (Gologanu et al., 1993), the dashed one that of Garajeu (1995)
and the full one our present proposal. All theoretical criteria are of comparable
quality, although that of Garajeu slightly errs on the horizontal Em/ O"o axis for
small porosities. However, Figure I.5.b shows a zoom of the region located near the
(E33- En)/O"o axis, for a porosity of 0.01. It clearly appears that the approximate
yield locus proposed here is the only one that remains approximately parallel to the
true one everywhere. This may seem of no importance, considering the smallness of
the differences on the vertical axis; it is important, however, because even such small

1 Since a1 = 0.
 86 M. Gologanu et a!.

a 0.990 b
0.5 .0.
.0.
··.' .0.
0.985 (:
.0.
("
0.0 .0.

0.980
-0.5
f = 0.01
-1.0
0 2 3 Lmla 0 -0.5 0.0 0.5 1.0

Figure I.5: Yield loci for prolate cavities

differences generate non-negligible variations of the normal to the yield locus, and
therefore of the porosity rate (for small triaxialities).
Figure I.6.a shows criteria for oblate cavities with aspect ratio bi/ a 1 = 5 and
several porosities, and Figure I.6.b for penny-shaped cracks for several values of the
"equivalent porosity" g. Again, discrete points indicate numerical results, the dotted
line our former proposal (Gologanu et al., 1994) and the full one our present one.
The superiority of the latter is particularly conspicuous for penny-shaped cracks.

(Lss- L11)/ cro (L33- L11)/ cro

1.0 b
1.0 a
0.5 0.5

0.0 0.0 :

-0.5 -0.5

-1.0 ~~~~~==-~--'"-_j -1.0 ~~~~=:=L-...o....~~

0 2 0 2

Figure I.6: Yield loci for oblate cavities (a) and penny-shaped cracks (b)

I.5.b. Comparison of porosity rates

Figures I. 7.a and b show the ratio Dm/ D:J:h as a function of aspect ratio for
Recent Extensions of Gurson's Model 87

prolate and oblate voids respectively, for various triaxialities Em/Eeq (1/3 for trian-
gles, 1 for circles, 2 for squares); D:!:h here denotes the value of Dm for a spherical
cavity and serves as a reference. In all cases, Em > 0, E33 > Eu and f = 0.01. The
symbols used are the same as in Figures 1.5 and 1.6. In the prolate case, both our
previous proposal (Gologanu et al., 1993) and that of Garajeu (1995) (not shown
here) predict an insufficient influence of the aspect ratio upon Dm, and frequently
err in the position of Dmf D:J:h with respect to unity, in contrast with our present
proposal. In the oblate case, both our old and new proposals are acceptable, but the
former predicts an insufficient influence of the triaxiality.

7
D I osph
m m

Figure I. 7: Dm/ D:!:h versus aspect ratio for prolate (a) and oblate (b) cavities

Figures I.8.a and b show the ratio nr:num I D:/:' where nr:num and D:/: represent
the numerical and theoretical values of Dm respectively, as a function of aspect ratio
for different triaxialities, in the prolate and oblate cases. The meaning of the symbols
is the same as in Figures 1.5, 1.6 and I. 7. Again, Em > 0, E33 > E 11 and f = 0.01
in all cases. No model yields a unity value, as one would naturally hope; therefore
the "best" model here is that that yields a value as constant as possible, so that
its defects can be corrected through the introduction of some constant empirical
factor analogous to Tvergaard's famous q1 parameter. The superiority of our present
proposal is clear in that respect.
A variant of the model proposed here is presented in the Appendix. The foun-
dations of that variant are somewhat hybrid because it consists in replacing, without
any satisfactory justification, some elements of the original model by some others
taken from the work of Ponte-Castaneda and Zaidman (1994). Its use improves the
agreement between numerical and theoretical values of the porosity rate for low tri-
axialities and in the case where :E 11 > :E 33 (and :Em > 0), which is admittedly of less
practical interest than the case where E33 > Eu.
 88 M. Gologanu et a!.

2.5 0 num1 0 th b
m m
................ z:l· ............... .

2.0

' ....... E)························

1.5~.-~

0.5 ~ ~ u

.,
···~.;-- :4
0.0 ...._~~---''---.·~·-=-_.·-~·. _ ..-_.·-·~~·
. ..,..__.. =---=··...:..o:-:_:
• .,...:..:..;J
••

4 4

Figure I.8: D~um / D~ versus aspect ratio for prolate (a) and oblate (b) cavities

1.6. Evolution of the shape parameter

We now come to the question of finding an evolution equation for the shape
parameter S (which plays the role of an internal parameter, just like f). It must at
once be noted that this is bound to be a more difficult problem than that of defining
the overall yield criterion. Indeed the criterion admits a variational characterization
in terms of the overall plastic dissipation (see (!.15)), whereas the rate of change
of the shape parameter does not. Hence, if we adopt, for each loading, some trial
velocity field which is not too far from the true solution, the estimated minimum of
the overall dissipation should be close to the true minimum, so that the estimated
yield locus should also be close to the exact one; on the other hand, nothing warrants
that the estimate of S will be of good quality.

The first problem that arises is that of the definition of S. By eq. (1.1 2 ), we get
=
S ada1 - bdb1. But this definition makes the implicit assumption that the void
remains spheroidal in the deformation, i.e. that it is strained homogeneously, and S
is then also given by

(1.49)

where the strain· rate nv of the void is defined by eq. (1.44). In reality, the void is
not strained homogeneously and these two definitions are different. We adopt the
second one (!.49) because it accounts for the movement of the whole boundary of the
void, not only for that of those special points located at f3 = 0, 1r /2 and 1r.
Recent Extensions of Gurson's Model 89

I.6.a. First two-field estimate of S

If we first adopt the same simple two-field approach as above, we get

where eq. (1.45) and the incompressibility of the matrix have been used. Now
inversion of eqs. (1.23) yields

Dm
A= DA ; B = -2(1- 2az)D 11 + 2a 2 D 33
m

where eq. (1.24 1 ) has been used. Inserting these values into the preceding equation,
we get
S· = D33 - Du + 3 (1-3alf + 3az - 1) Dm. (1.50)

I. 6. b. Particular cases
For a cylindrical cavity (a 1 = a 2 = 1/2), eq. (1.50) reduces to

(1.51)

Now it is easy to calculate the exact value of S for such a geometry (and an axisym-
metric loading). Indeed the velocity field is necessarily of the form Vp = 1/ p- 1' pl2,
v'P = 0, Vz = 1' z where 1 and 1' are constants. This field satisfies conditions
of homogeneous strain rate on both the external and internal boundaries. There-
fore the components of the macroscopic strain rate are given by Du = Dzz =
v p(pz) I pz = 1 I p~ - 1' 12, D33 = 1', and those of the strain rate of the void, by
Df1 = D22 = vp(PdiPl =II Pi- 1'12, D?,3 = 1'; P1 and pz in these expressions
denote the radii of the internal and external cylinders respectively. Eliminating 1
and 1' between these formulae and using the fact that f = pif p~ and the definition
(1.49) of S, one obtains eq. (1.51). Therefore eq. (1.50) is exact for a cylindrical
cavity.
For a "sandwich" ( a 1 = az = 0), eq. (1.50) yields

(1.52)

Again, S can be calculated exactly in that case. The velocity field is of the form
vP = -1' p/2, v'P = 0, Vz = 1sgn(z )+1' z where 1 and 1' are constants; the components
of D and Dv are given by Du = D22 = -1' /2, D33 = 1/hz +1', Df1 = D2z = -1' /2,
90 M. Gologanu et al.

D33 = "(I h1 +"( 1 where h1 and h2 denote the half-thicknesses of the empty space and
the "sandwich"; elimination of 'Y and "( 1 and use of the relation f = h1 I h2 lead to eq.
(1.52). Thus eq. (1.50) is also exact for a "sandwich".
The case of a penny-shaped crack is more intricate. If Dm = 0, eq. (1.50) reduces
to S = D 33 - D 11 , which is exact since the true velocity field is then v 8 so that the
strain rates of the void and the cell are identical. If Dm =/= 0, the left-hand side of
eq. (1.50) becomes infinite because of the presence of a1 = 0 in the denominator of
the expression (1.44) of nv, and the same is true of the right-hand side because f
is zero. Regarding the crack as the limit of an oblate cavity with vanishingly small
minor axis a 1, and considering dominant terms (proportional to 1I a 1) in both. sides
of eq. (1.50), we get

where use has been made of eq. (1.44) and of the fact that a1 = 0. The last equality
is correct since the left-hand side represents the rate of increase of the volume of the
void, and the right-hand side that of the volume of the cell, which are equal since the
matrix is incompressible. Hence eq. (1.50) is again exact for a penny-shaped crack.
The only remaining interesting special case is that of a spherical void. The
term proportional to Dm in eq. (1.50) is zero (since a1 = a2 = 113); this is correct
since the shape of an initially spherical void does not change for a purely hydrostatic
loading (D 11 = D33). On the other hand, there is no reason why the term D33- D 11
in eq. (!.50) should be correct; in fact, it is known to be incorrect, because numerical
studies (see for instance Gologanu et al., 1993, 1994) have shown that for a purely
deviatoric loading (Dm = 0), the strain rate of the void is close to twice that of the
cell, at least for very small porosities.
I.6.c. Second two-field estimate of S
There is no reason to modify the term proportional to Dm in eq. (1.50) since it
was observed to be exact in all special cases considered above. On the other hand,
the term proportional to D 33 - D 11 was observed to be correct for a cylindrical void,
a "sandwich" and a penny-shaped crack, but not for a spherical void. This suggests
to modify eq. (1.50) in the following way:

·
S=h(D33-Dn)+3 (1-f3al +3a2-1 ) Dm (I. 50')

where his an empirical factor (which of course should be unity for a cylindrical void,
a "sandwich" and a penny-shaped crack, but not for a spherical void).
It is easy to see that the fact that h was found to be unity in eq. (1.50) was tied
to the uniformity of the strain rate corresponding to the field v 8 . One simple idea,
 Recent Extensions of Gurson's Model 91

quite in the spirit of the works of Ponte-Castaneda and Zaidman (1994) and Garajeu
(1995), to remedy that deficiency is to momentarily forget that the material is plastic,
not elastic, and to adopt for h the value resulting from Eshelby's famous solution.
This is equivalent to replacing the simple field v 8 by the exact elastic solution for a
purely deviatoric loading (Em = 0). (Strictly speaking, this operation is licit only in
the limit f --+ 0, since the medium considered by Eshelby is infinite and his solution
does not obey conditions of homogeneous strain rate on confocal spheroids, except
on the boundary of the void). Using the expression of Eshelby's tensor provided by
Mura (1982), one then gets h = 1 + 3(al- afar)/(1- 3a 1 ), where afar is defined
by the same formula (!.43) as afar, but with e1 instead of e2 . This formula is not
completely satisfactory, however, because such an h can be verified to be equal to 5/3
for a spherical cavity, whereas a value of 2 is more correct, as was mentioned above.
This suggests to artificially modify the preceding formula into h = 1 +he where

(I. 53)

With such a definition, he is zero for a cylindrical void (a 1 =afar= 1/2), a "sand-
wich" and a penny-shaped crack (a 1 = afar = 0), and unity for a spherical void, so
that h takes the required values in all cases.
I.6.d. Comparison with some numerical results
Using the same numerical approach as in Section 1.5, one can evaluate the sup-
posedly exact values of S and h (using the values of a 1 and a2 given by eqs. (1.48)
and (1.42)). It is found that the improved formula (1.53) is still insufficient to re-
produce the results obtained; in fact h is found to depend on both f (even for
small values of this parameter) and the loading via the triaxiality T = Em/Eeq,
Eeq = (3:E' : :E' /2) 1 12 (where the subscript m and the prime indicate the mean
and deviatoric parts as above). In order to account for these extra dependencies, we
modify the previous expression of h into

(I. 54)

where he is still given by (I. 53) and the new empirical factors h f and hr are to
account for the influences of the porosity and the loading. Since, as was seen above,
h = 2 is a good value for a spherical void in an infinite medium subjected to some
purely deviatoric loading (T = 0), and he = 1 for such a void, ht and hr must be
unity for f = 0 and T = 0 respectively. Also, h f must tend to zero for f --+ 1 (which
admittedly is a completely academic case), because the void completely fills the cell
then so that their strain rates must be identical.
Figure I.9.a shows the influence of the porosity upon the numerical values of
h (discrete points) in the case of a spherical cavity and a zero triaxiality (so that
he = hr = 1). The influence of f is seen to be considerable: note that when it
92 M. Gologanu et al.

increases from 0 to only 0.01, h decreases from 2 to 1.85. The numerical results
can be reproduced by the simple empirical formula (which respects the condition
ht(l) = 0)
(1.55)
(continuous line). This formula being adopted, Figure I.9.b compares the numerical
(discrete points) and theoretical (continuous line) influences of the void shape upon
h, for f = 0.01 ( =? h f = 0.81) and again T = 0 ( =? hr = 1). The semi-empirical
formula (1.53) for he can be seen to give an acceptable approximation of the numerical
results, although its accuracy deteriorates for oblate voids and e 1 close to 1 (note
however that it reproduces the exact result for e 1 equal to 1).

2.0 h a 2.4 h
2.2
1.8
2.0
1.6 1.8

1.4 1.6
prolate
1.4
1.2 0
1.2
1.0
f 1.0
0.00 0.05 0. 10 0.15 0.20 0.25 0.30 2 0 2 1.0

Figure I.9: The parameter has a function off (a) and of e 1 (b)

Figure I.lO shows the influence of the triaxiality upon h, for a spherical cavity
(=? he = 1) and a porosity of 0.01 (=? ht = 0.81). The parameter E appearing in
this figure is defined as

(!.56)
In fact, I:m is taken as positive in all simulations, and 2::~ 3 = 2(2:: 33 - 2:: 11 )/3 in the
=
axisymmetric case, so that E sgn(2:: 33 - 2:: 11 ) here. Two cases can be distinguished
according to the value of e; E = +1 means that the loading consists of one major
(positive) principal stress and two identical minor (possibly negative) ones, whereas
E = -1 means two identical major (again po~itive) principal stresses and one minor
(again possibly negative) one. The numerical results (discrete points) evidence a
considerable influence ofT. They also show some effect of the value of E.
In order to better understand the origin of the latter effect, let us analyse in
more detail in which way the parameter h may be influenced by the loading. Since
 Recent Extensions of Gurson's Model 93

2 0 2

Figure !.10: The parameter h as a function of T fort= ±1

the geometry considered in Figure I.lO is spherical and therefore isotropic, h can
only depend, in this case, on the three invariants of :E, namely l:m, l:eq and det(:E),
or equivalently on T =
l:m/Eeq, Eeq and det(:E). But the (true) criterion, which is
also isotropic, determines Eeq as a function of the other two parameters 1 . Hence
h can only depend on T and det(:E). In the case of an axisymmetric loading, since
there are only two independent stress components tied by the criterion, it may seem
that the stress state is completely determined by the single parameter T and thus
that h should also be a function of T alone. This is not so however, because there
are in fact two axisymmetric stress states with the same triaxiality which obey the
criterion; these stress states can be distinguished through the value of det(:E) or t.
The fact that h is found to depend on t therefore means that it depends on the
third invariant det(:E). It is worth noting here that a similar influence of the third
invariant was observed by Suquet and Ponte-Castaneda (1993) in a related, though
slightly different, context (the influence of det(:E) was on the criterion, not on h, and
the contrast beween the phases was assumed to be small, whereas it is maximum in
the case of some second phase consisting of voids, as considered here).
To find suitable formulae for the parameter hr, it is useful to make the following
remark. Since we are dealing with metal plasticity, starting from physically admis-
sible (microscopic) velocity and stress :B.elds at a given instant, one can obtain some
other admissible fields, at the same instant, by just changing the signs of all velocities

1 Note that the reasoning does not rely on the assumption that Gurson's criterion, which is
independent of the third invariant, is applicable; the criterion considered here is the exact one,
which may depend on it.
94 M. Gologanu et al.

=
and stresses. In this operation, S D33 - D1 1 , D33 - Du and Dm all change sign;
therefore h, as defined by eq. (1.50'), must remain invariant. But the signs of ~m and
~~ 3 also change, so that € is unchanged whereas T changes sign. This means that
€ being fixed, h, and consequently hr, must be even functions ofT. It is therefore

reasonable to look for a formula for hr in the form of some polynomial expression of
T 2 • It is found that the formulae

for € = +1
(!.57)
for € = -1

reproduce the numerical results with a reasonable accuracy, although the agreement
tends to deteriorate for large triaxialities (see the continuous line in Figure 1.10).

1.7. Comparisons with some finite element simulations of the evolution of

representative volume elements
I. 7.a. The simulations of Sovik
Sovik (1995) has performed finite element calculations of the evolution of cylin-
drical cells containing initially spheroidal voids of different shapes and subjected to
axisymmetric loadings under constant triaxiality. The material considered was ideal-
plastic. The mean stress ~m was positive and ~ 3 3 was greater than ~ll in all cases.
The initial porosity was 0.0002 and the triaxiality 1. Figure I.ll.a shows his results
for the porosity as a function of Eeq = J:(2D 1(T) : D'(T)/3) 1 12 dT for the different
aspect ratios, and Figure I.ll. b shows the predictions of our old and new proposals.
Although none of these yields quite satisfactory quantitative results, the qualitative
trend of the curves is much better reproduced by our new proposal. A satisfactory
quantitative agreement could be obtained by introducing some Tvergaard-like heuris-
tic factor q1 > 1 into the model. The numerical results indicate that this parameter
would have to be a function of void shape in the prolate case, which is not surprising
since q1 = 1.5 and q1 = 1 are known to be good values for spherical and cylindri-
cal voids respectively (recall that for the latter voids, our criterion is identical to
Gurson's which is exact for q1 = 1).
I. 7. b. A simulation of Mazataud
Mazataud (1995) has performed the following finite element simulation. He
considers like Sovik a cylindrical cell containing now an initially spherical void, and
subjected to some axisymmetric loading. The originality of his calculation is that
he imposes a stress ~33 smaller than ~ll for Eeq :::; 0.4 but greater than it for
Eeq 2:: 0.4, the triaxiality remaining constant, equal to 2/3. The initial porosity is
0.0001. Figure I.12.a shows his result for f as a function of Eeq, and Figure I.12.b
the predictions of Gurson's usual model (dotted line), of our previous model (dashed
line) and our present one (full line). There is a sharp change of slope in the numerical
Recent Extensions of Gurson's Model 95

0.008 a,. 0.003

a1/b1=1/3 I
b
f I f a1/b1=1/2 /
-···a1/b1=113 .·new
0.006 0.002 a1/b1=1 / /
--- a1/b1=1/2 I a1/b1 =2 /
I / /
-adb1=1 I a1/b1=3 .. · /
I
0.004 -·-·-· a1/b1 =2 I 0.001 / /
_,. _,·'
I / _ ... ..... ·"
- - a1/b1=3 /
/ /

0.002 /·. /
/
0
,.,..~_'.,...,
.,..
_p_-::::::.:;:::_.~-:..:.·:..:- Eeq 0 Eeq
0.000
0.00 0.15 0.30 0.45 0.60 0.1 0.2 0.3 0.4 0.5 0.6

Figure I.ll: Numerical results of Sovik (a) and theoretical predictions (b)

results which is easily explained as follows: since 'Eu > 'E33 initially, the spherical
void grows into an oblate spheroid, the growth of which is obviously more "sensitive"
to 'E33 than to 'Eu. Therefore when one "switches" from 'Eu > 'E33 to 'E33 > 'Eu,
there must be some increase of the porosity rate. Gurson's model completely fails
to capture the effect because it only relates void growth to the triaxiality and does
not make any difference between the cases 'Eu > 'E33 and 'E33 > 'Eu. Our previous
proposal does capture the effect but to an insufficient extent. The best result again
corresponds to our present proposal. Again, the agreement with numerical results
could be improved by introducing some q1 factor into the model, but the value to be
used would then have to be smaller than 1; this is related to the fact that Mazataud
considers a hardenable material with a rather high hardening exponent (N = 0.3),
and it is known that q1 is a decreasing function of N (see Koplik and Needleman,
1988).

1.8. Conclusion: constitutive equations for ductile porous metals incorpo-

rating the influence of void shape
The final task is to extend and complete the model in order for it to be applicable
to finite element simulations of actual structures undergoing ductile rupture. We
shall adopt here a purely heuristic approach as in our previous works ( Gologanu
et al., 1993, 1994) for simplicity, although some more rigorous one is possible (see
Gologanu, 1996).

The first thing to do is to extend the criterion (1.31) or (1.34) to arbitrary

(non-axisymmetric) loadings. The generalized criterion looked for must verify three
(minimal) requirements: it must be invariant upon change of sign of the stresses,
reduce to that of von Mises for f --+ 0 (or g --+ 0 for penny-shaped cracks), and
96 M. Gologanu et al.

0.0010 0.0010
f
a f
b
0.0008 0.0008

0.0006 0.0006 old/

/
new / .·
.·
.· .··
/
0.0004 0.0004 /
/

,_;..··· Gurson
0.0002
Eeq
0.0002 ..... -- _.. ?··

Eeq
0 0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

Figure !.12: Numerical results of Mazataud (a) and theoretical predictions (b)

respect tranverse isotropy perpendicularly to the axis of the voids. Fulfilling the first
requirement is made more difficult by the appearance of L:h in the "square" term,
which forbids to just replace L:33 - L:u by L:eq as in Gurson's criterion. It is possible
to respect it, however, as well as the third one, by adopting the following criterion:

In this equation, 11.11 denotes the von Mises norm (IITII = (3T : T/2) 1 12 ) and the
tensor X is defined by

(I. 59)

The parameters"'' C, ry, a 1 , a 2 are given by eqs. (!.39) (prolate case) or (!.41) (oblate
case), (!.47), (1.48) and (!.42). Also, g is zero for prolate voids and given by (!.35)
for oblate ones, and the expression of Eh is

(!.60)

this is the natural extension of eq. (!.25 2 ) respecting transverse isotropy to arbitrary
stress states. Recall finally that the eccentricities e1 and e2 can be calculated in
terms of the porosity and the shape parameter by eqs. (I.2).
It can be verified, using all of these equations, that the criterion (!.58) does
reduce to that of von Mises in the limit f --t 0 or g --t 0, as desired.
The reader is referred to Gologanu (1996) for a slightly more elaborate proposal.
Gologanu's approach is not heuristic but relies on the same kind of micromechanical
Recent Extensions of Gurson's Model 97

analysis as above, the velocity field v 8 being replaced by some uniform deviatoric,
but otherwise arbitrary (not necessarily axisymmetric) strain rate. The approximate
criterion obtained is identical to (1.58) but for an additional term proportional to
1 - C, which is generally negligible because C is close to unity.
Elasticity can be introduced in the usual way, by adding to the macroscopic plas-
tic strain rate DP, given by the normality rule (see Gurson, 1977), some hypoelastic
strain rate De tied to some objective derivative (that of J aumann being the simplest)
of the stress.
Finally, the evolution equation for the porosity is the usual one: j = 3( 1- f)Dfn,
which derives from matrix plastic incompressibility, and that of the shape parameter
(which again extends (1.50') while respecting transverse isotropy) is

S. -_ 2
3 P1
hD 33 + 3
( 1 - 3a 1
J + 3a2 - 1
) DP
m (1.61)

The parameter h here is defined by (1.54), (1.53) (with a?ar

given by (1.43) with e1
instead of e 2), (1.55), (1.56) and (1.57). Also, one must keep track of the direction e 3
collinear to the axis of the voids. The simplest hypothesis here consists in assuming
that it rotates with the same velocity as the matter:

( 1.62)

where n denotes the same rotation rate (antisymmetric tensor) as in the formula
defining the objective derivative of the stress.

PART II: A GURSON-LIKE MODEL INCLUDING THE EFFECT OF

STRONG GRADIENTS OF MACROSCOPIC FIELDS

11.1. Overview
As already mentioned in the Introduction, we now wish to derive an approximate
model for ductile porous solids including the effect of possible notable gradients of
macroscopic mechanical fields at the scale of the void spacing (arising from the ten-
dency of damage to localize). This model will be derived from some micromechanical
approach extending Gurson's (1977) famous limit-analysis of a hollow sphere sub-
jected to conditions of homogeneous boundary strain rate, to more general boundary
conditions.
We begin by extending the classical Hill (1967) - Mandel (1964) approach to
homogenization to the case where the velocity imposed on the boundary of the rep-
resentative volume element considered is no longer linear but quadratic with respect
98 M. Gologanu et al.

to the coordinates. The macroscopic mechanical description resulting from homoge-

nization is formally very close to "micromorphic" theories proposed by former authors
like Mindlin (1964), Mindlin and Eshel (1968) and Germain (1973a,b) on purely phe-
nomenological grounds. There is also some connection with a very recent work of
Drugan and Willis (1996) on elastic composites, based on homogenization like the
present work, but not on the Hill- Mandel approach. The description of internal forces
(at the macroscopic scale) involves not only ordinary macroscopic stresses (average
values of the microscopic stresses) but also quantities which homogenization shows
to be interpretable as "moments" of the microscopic stresses (average values of these
stresses times some position coordinate). Also, for a rigid ideal-plastic porous ma-
terial, the macroscopic yield criterion is given a variational characterization, and a
normality property is established for the macroscopic flow rule.

The core of the work is to be found in the next section, which describes a limit-
analysis of some hollow sphere made of rigid ideal-plastic von Mises material and
subjected to conditions of inhomogeneous, axisymmetric boundary strain rate. This
necessitates the introduction of two extra velocity fields in Gurson's analysis. These
fields are assumed to derive from some "stream function" of the form f (r) g( 8) in
spherical coordinates r, 8, IP· The function of 8 is identified through consideration of
the boundary conditions, and that of r is determined by requiring the overall plastic
dissipation to be minimum for the velocity field considered taken alone. This gives
rise to some minimization problem analogous to, though slightly more complex than,
those encountered in rational mechanics; its solution involves some Lagrange-like
differential equation on the fonction f(r), which is solved numerically. In spite of
the complications which arise from the presence of 4 velocity fields in the analysis,
instead of 2 as in Gurson's original work, it remains possible to derive a Gurson-
like approximation for the macroscopic yield locus. An interesting feature of the
analysis is that it is found that for a totally sound material (zero porosity), plastic
flow is possible for a purely "hydrostatic" moment, although it is impossible for
a purely hydrostatic (macroscopic) stress. The mathematical consequence of that
phenomenon is that the "traces" of the moment appear in the square instead of the
"cosh" as one would intuitively expect.

We next describe some numerical minimizations performed using a large number

of trial velocity fields, aimed at determining the exact overall yield criterion for
the geometry and boundary conditions considered. Comparison of the results of
these simulations and the approximate criterion proposed evidences the quality of
that criterion, although it fails to reproduce "corners" which can be observed on
the numerical yield locus. This comparison suggests some slight adjustments of
parameters in the analytic yield function which lead to enhanced agreement (although
the analytical yield locus unfortunately remains free of corners).

We also describe some finite element simulations of the evolution of some ini-
tially spherical representative volume element strained axisymmetrically and inhomo-
Recent Extensions of Gurson's Model 99

geneously, and compare their results to theoretical predictions. The main differences
between the model and the simulations is that the latter account for both the pro-
gressive change of shape of the representative volume and the coalescence of cavities,
in contrast to the former. It is found that the agreement between theoretical and
numerical results, which is initially quite acceptable, accordingly deteriorates to some
extent in time.
To conclude, we define a heuristic extension and modification of the model to be
used in practical finite element calculations of actual structures undergoing ductile
rupture. The necessary generalizations include extension of the yield criterion to arbi-
trary, non-axisymmetric stress and moment states, incorporation of strain hardening
and void nucleation effects, and of hypoelasticity. Also, the appearance of second
spatial derivatives of the velocity in the model, and the resulting seemingly unavoid-
able need for finite elements of class C1 , are removed using a trick quite analogous to
the transition from the Kirchhoff-Love plate theory to that of Mindlin and Reissner.

11.2. Extension of the Hill-Mandel approach to homogenization to condi-

tions of inhomogeneous boundary strain rate
II.2.a. Basic principles - Definitions of the macroscopic strain rate and strain rate
gradient
The classical approach of Hill ( 1967) and Mandel ( 1964) to homogenization is
based upon the assumption of a linear variation of the velocity v(x) over the boundary
of the representative "cell" C(X) considered; X here denotes the centroid of the cell.
In contrast, it will be assumed here that the variation of the boundary velocity is
quadratic:

(ILl)

where A and B are second- and third-rank tensors, respectively. Figure II.l il-
lustrates the difference between the classical and new approaches. It is harmless
to assume A to be symmetric because its antisymmetric part only represents some
rigid-body motion; also, one can of course assume B to be symmetric in its second
and third indices.
The macroscopic strain rate D(X) is defined as the average value of the micro-
scopic strain rate d(x) over C(X), the velocity field being extended in an arbitrary
way over the possible voids (which are supposed not to intersect 8C(X)). It follows
that
 100 M. Gologanu et al.

Classical approach

New approach

Figure II.l: Deformations of a lattice of elementary cells

= 2 1C~X)Ilac(X) [(A;kXk + ~BikhXkXh)nj(x) + (AjkXk + ~Bjkhxkxh)n;(x)] da

= IC(~)IL(X) [ A;j + ~(Bijk + Bjik)Xk] dv

where IC(X)I denotes the volume of C(X) and n the outward normal unit vector to
its boundary oC(X). This means that

1 1
D;j(X) = A;1 + 2 (Bijk + Bjik) Xk => D;j;k(X) = 2 (Bijk + Bj;k) (II.2)

where the position-vector X is treated as a continuous variable (at the macroscopic

scale) and a semicolon denotes a differentiation with respect to its components. Equa-
tions (11.2) show in particular that D(X) and VD(X) are independent of the manner
in which the velocity field is extended over the voids.
Equation (11.2 2 ) can be inverted by writing it for the triplets (i, j, k ), (j, k, i) and

!
(k,i,j):

D;j;k(X) = (Bijk + Bjik) /2

Djk;;(X) = (Bjki + Bkji) /2 => Bijk = (Dij;k- Djk;i + Dki;j)(X), (11.3)

Dki;J(X) = (Bkij + Bikj) /2

where the symmetry property of the coefficients B;jk has been used. For given tensors
D(X), VD(X), eqs. (11.21) and (11.3 2 ) fully define the tensors A and B, and therefore
the boundary conditions to be enforced on 8C(X).
Recent Extensions of Gurson's Model 101

Remark. The classical Hill-Mandel approach to homogenization does not say any-
thing about the existence of such a thing as the "macroscopic velocity" V(X), from
which the "macroscopic strain rate" D(X) should derive. It is only when the ho-
mogenization procedure is completed (i.e., the macroscopic constitutive equations
fully defined) and microscopic quantities henceforth forgotten that one introduces
the heuristic assumption that the "macroscopic strain rate", as defined through ho-
mogenization, can be identified with the symmetric part of the (macroscopic) gradient
of some "macroscopic velocity". However, this weakness of the Hill- Mandel theory
can be remedied to some extent, even in its extended version presented here, in the
following way. Let us assume the shape of C(X) to be simple (for instance ellipsoidal
or parallelepipedic) so that the centroids of C( X) and 8C( X) coincide, and let us de-
fine V(X), in an admittedly somewhat artificial manner, as the average value of v(x)
over 8C(X) (instead of C(X)). Then, using eq. (ILl) and writing x as X+ (x- X),
one gets

where ( )'D,S represents an average value taken over the domain 1) or surfaceS. The
last term in the right-hand side here is independent of X. It follows that
1 1
Vi;j(X) = Aii + BiikXk => "2 (V;;i + Vj;i) (X)= Aii + "2(Bijk + Biik)Xk = D;j(X)

(by eq. (11.2!) ), as desired.

II.2.b. Tbe Hill-Mandel lemma and tbe definition of macroscopic stresses and mo-
ments
Let the microscopic stress field u(x), which initially exists only in the matrix,
be defined as zero in the voids. This operation preserves its statically admissible
character, since it respects the condition of continuity of the traction across the
boundary of the voids. It follows that

(u(x) : d(x))c(X) = IC(~)/l(X) aij(x)vi,j(x)dv

= /C(~)/lc(X) aij(x)vi(x)nj(x)da
(by the principle of virtual work)

= /C(~)/lc(X) aij(x) ( AikXk + ~BikhXkXh) nj(x)da

= /C(~)/ L(X) aij(x) ( Aij + ~(Bijk + Bikj)Xk) dv
 102 M. Gologanu et al.

Writing Xk as xk + (xk- Xk), defining the quantities

and using eqs. (II.2), one readily puts this result in the form

(u(x) : d(x))c(x) = E;i(X)D;j(X) + Mklii(X)Dij;k(X)

= :E(X) : D(X) + M(X):\7D(X) .
(!1.5)

This formula stands as an extension of the classical Hill-Mandellemma for conditions

of homogeneous boundary strain rate, to conditions of inhomogeneous boundary
strain rate. We shall call the (symmetric) tensor :E the macroscopic stress, and the
(symmetric in its second and third indices) tensor M the macroscopic moment, for
obvious reasons.
II.2.c. The virtual powers of internal and external forces and the resulting macro-
scopic equilibrium equations
Let n denote the borly considered (made of "elementary cells" C(X)). The vir-
tual power of internal forces is

p(i) =- fo u(x): d(x) dv =- l:x(u(x): d(x))c(x)IC(X)I

(II.6)
=- j 0 [:E(X): D(X) + M(X):\7D(X)] dV,
X being again treated as a continuous variable.
The expression of the virtual power of external forces p(e) is not generally de-
duced from homogenization, which does not easily deal with boundary conditions; it
most often simply results from an ad hoc postulate. Here we assume that

p(e) =lao{ T(X).V(X) dA (II.7)

where T(X) represents some macroscopic surface traction. One could imagine adding
some term proportional to \7V to the integrand; the cofactor of that quantity would
represent a "surface moment". This would make sense, however, only if there were
significant variations of the microscopic surface traction over elementary cells, i.e.
over distances of the order of the void spacing. We suppose that such is not the case
here, because strong gradients of macroscopic fields arise from the softening character
of the material (due to the porosity), not from some pathological character of the
boundary conditions.
Application of the principle of virtual work yields, upon double integration by
parts of the expression of p(i), the following equilibrium equations:

(!1.8)
Recent Extensions of Gurson's Model 103

which are identical to eqs. (2.19) of Mindlin and Eshel (1968), except for the body
force and acceleration terms which are disregarded here. The boundary conditions
can also be obtained from the same principle; they are given for instance in (Germain,
1973a). They will not be detailed here because they are complex and will not be
needed in the sequel.
II.2.d. Elimination of rigid-body motions and restrictions on macroscopic moments

Let V'D and M denote the "deviators" of V'D and Mover their non-symmetric
indices:
-
D ij;k =Dij;k- 4 (o;kDjh;h + DjkDih;h) ; M klii =Mklii- 4 (o;kMhlih + DjkMhlih) .
1 - 1

(II.9)
With these definitions, V'D and M possess the same symmetries as V'D and M,
namely D ij;k = D ji;k and M klij = M klji, and they are traceless (hence the term
"deviator"):
D ij;j = 0 ; M ilii = 0 . (11.10)
The condition for these deviators to be zero is that Dij;k or Mklii be of the form

(II.ll)

for some vector U.

In the sequel, we shall mainly consider spherical elementary cells (in spite of
the fact that strictly speaking, such cells cannot be stacked up to build a lattice, at
least if it is to be periodic). Of course, we shall take the centre of the cell studied
as the origin of space. It so happens that for such a cell, tensors V'D having a zero
deviator only correspond to rigid-body motions; indeed, if V'D is of the form (II.ll),
Bijk= 2DjkUi by eq. (II.32) so that the velocity arising from V'D on the boundary of
the cell is of the form v;(x)= DjkUiXjXk= b2 U;= Cst. where b denotes the radius of
the cell. It follows that such strain rate gradients have no effect on the constitutive
equations and can therefore be disregarded in the search for those equations.
By eqs. (II.6), (II.9) and (11.10),

......._
__,
. L[
p(a)--
- 0
E •1.. D •1.. + - -
M ki" .. k + -M·1·
•1 D •J;
1 1 •1·D·k
2
J
• ; k dV •
104 M. Gologanu et al.

Since p(i) must be zero for a rigid-body motion (D = 0, VlD = 0), the cofactors of
the "traces" Dik;k must be zero:

(11.12)

This result can also be obtained through direct consideration of the equilibrium
of the (spherical) cell. Indeed, if S(X = 0, r) =
S(r) denotes the sphere of radius r
centered at the origin, global equilibrium of that sphere implies that

and one gets (11.12) from there through integration over r, using the definition (11.4 2 )
of the moments.
II.2.e. Limit-anB.lysis for a rigid ideal-plastic material
We now consider a porous rigid ideal-plastic von Mises material (with yield stress
a 0 in simple tension) whose flow rule obeys the normality property. The (microscopic)
plastic dissipation 1r( d) is defined by the same formula (1.14) as in Part I. Then, by
eq. (11.5),

:E: D + M:VlD = (u(x): d(x))c(x):::; (7r(d(x)))c(X)

for any plastically admissible pair (:E, M) so that

:E: D + M:VlD :S II(D, VlD) =: lnf (7r(d(x)))c(X) (11.13)

where the infimum is taken over all incompressible kinematically admissible (i.e. sat-
isfying boundary conditions corresponding to the prescribed values of D and VlD)
velocity fields. II(D, VlD) is called the macroscopic plastic dissipation; this defini-
tion is a natural extension of the classical one in the usual Hill-Mandel theory: see
Suquet (1982). It follows from (11.13) that the domain C of plastically admissible
macroscopic pairs (:E, M) lies in the intersection of the semi-spaces (depending upon
the parameters D and VlD) :E: D + M:VlD :::; II(D, VlD); in fact, a classical result
in limit-analysis asserts that it is identical to this intersection. The boundary 8C of
this domain, i.e. the macroscopic yield locus, is defined by the equations

arr arr (11.14)

E; 1 = -0 (D, VlD) ; Mklij = -(D, VlD)
-!::\
D;j uD;j;k

where D and VlD act as parameters.

Recent Extensions of Gurson's Model 105

A comparison with the work of Drugan of Willis (1996) on elastic composites

is now in order. These authors define some stresses s and t by the formulae 1 Sij =
8W/8(e)iii tiik = 8W/8(e)ii;k where the function W((e), V'(e)) plays the role of an
elastic energy; e here denotes the (small) strain. These equations are quite analogous,
in the context of elasticity, to our eqs. (II.14), W playing the role of II, (e)ij that of
D;j, (e)ij;k that of Dij;k, Sij that of ~ij, and tijk that of Mklij (note the interchange
of indices). Also,the equilibrium equations read Sij;j- tijk;jk = 0 and are therefore
completely analogous to our eqs. (II.8).
However, there is a puzzling difference between the two works. In the present one,
:E is the average value of u. In that of Drugan and Willis, sis not; in fact (a)ij = Sij-
tijk;k· The equilibrium equations satisfied by (u) are accordingly different: (a;j);j-
Mklij;jk = 0 here, (a;j);j = 0 in Drugan and Willis' work. The key to this apparent
enigma is that the meaning of the symbol ( ) is different in the two works. Here it
represents a volume average; it was shown above rigorously (at least for boundary
conditions of type (ILl)) that (u) does not then obey the equation V'.(u) = 0. On
the other hand, average values in Drugan and Willis' work are ensemble averages,
that is, average values taken over a large number of possible "realizations" of the
composite; since the equation V'.u = 0 is then satisfied for each composite, it is also
obviously satisfied on average.
Since at the microscopic scale, the normality property is supposed to be obeyed,
Hill's principle of maximum plastic work is satisfied; that is, (u- u*) : d ~ 0 where
u and u* denote arbitrary microscopic stress tensors respecting the criterion and d
the microscopic strain rate corresponding to u through the flow rule. It then follows
from eq. (II.5) that
(:E- :E*): D + (M- M*):V'D ~ 0 (II.15)
where (:E, M) and (:E*, M*) denote arbitrary macroscopic plastically admissible pairs
and (D, VD) the "generalized macroscopic strain rate" corresponding to (:E, M). A
classical reasoning then leads to the following conclusions: first, the set C of macro-
scopic plastically admissible pairs (:E, M) is convex; second, (D, VD) is orthogonal
to the macroscopic yield locus 8C at the point (:E, M).
The analytic expression of the latter property is more complex than may be
thought at first sight, for the following reason. The pairs (:E, M) and (:E*, M*) in
inequality (II.15) do not span the whole space £ of pairs obeying only the symmetry
properties ~ij = ~ji, Mklii = Mklji, of dimension 24, but its subspace :F defined
by the restrictive conditions (II.12), of dimension 21. This implies that both C and
8C lie in :F, so that the space V of pairs (D, V'D) orthogonal (in £) at some point
(:E, M) to ac, which is a surface of dimension 20, is not of dimension 1 as usual,

1 We have replaced Drugan and Willis' comma in the second formula by a semicolon in order to
conform to the present notation.
106 M. Gologanu et al.

but· of dimension 4. Hence it is not simply generated as usual by the "gradient" of

components f)iPj8"E,ij, 8if!I8Mklij where if!(:E, M, f)= 0 (f: porosity) represents the
equation of the yield locus.
In order to analytically express the condition (D, VD) E V, let us note that
in practice, the expression of the yield function if! will be given for arbitrary pairs
(:E, M) E E, in spite of the fact that pairs not lying in :F are physically meaning-
less (for the spherical representative volume that will be considered). This means
arbitrarily extending the surface ac into some surface ac of dimension 23 out of the
subspace :F. Vectors collinear to the "gradient" (8if! I ar,ij, 8if! I 8Mklij) are orthogonal
to 8C and hence to 8C at the point (:E, M). By adding to them vectors orthogonal
to :F·in E, i.e. of the form (D = 0, 'VD) with VD given by (11.11), one obtains all
vectors orthogonal (in E) to 8C at the point (:E, M). Hence the analytic expression
of the condition (D, 'VD) E V is

(II.16)

for some scalar H which must be non-negative because of inequality (11.15), and some
arbitrary vector U.

11.3. Approximate overall yield criterion for a hollow sphere subjected to

conditions of axisymmetric, inhomogeneous boundary strain rate
We now consider a hollow sphere of centre 0, external radius b (which represents
a kind of average half-spacing between voids) and internal radius a= bjll 3 , made of
some rigid, ideal-plastic von Mises material obeying the normality rule.
II.3.a. Boundary conditions
This sphere will be subjected to boundary conditions of type (11.1) satisfying
the extra requirements of rotational invariance about the Ox 3 axis and symmetry
with respect to all planes containing that axis. The first task is to find which are
the nonzero components of the most general tensors D and 'VD compatible with
these requirements. The result for D is classical and quite obvious: its sole possibly
nonzero components are Du = D22 and D 33 . For VD, one can reason as follows.
Because of the second requirement, the indices 1 and 2 must appear an even number
of times in the nonzero components Dij;k· If neither of them appears, we are left with
the component D 3 3; 3 , which respects axisymmetry. If one of them, say 1, appears
twice, the third index cannot be 2, so that we get two possibilities: Du;3 and D 13;l
(= Dal;I). Axisymmetry then implies that D22;3 = Du;3 and D23;2 = D13;1· However,
+
we have seen that the "trace" D3i;i = 2D 13 ; 1 D 33 ; 3 only represents some rigid-body
motion. Hence it can be fixed arbitrarily, for instance by taking D 13 ;1 = D23;2 = 0.
The only possibly nonzero components Dij;k are then Du;3 = D22;3 and D33;3·
Recent Extensions of Gurson's Model 107

If we take V'D = 0, B = 0 by eq. (II.3 2 ) and A= D by eq. (II.21), so that the

boundary conditions (II.l) read

v1(x) = Dux1 = (Dm- D: 3/2) x1

{ (II.17)
vz(x) = Duxz = (Dm - D 33 /2) xz

v3(x) = D33x3 = (Dm + D; 3) x3

(for xi + x~ + x~ = b2 ), where the subscript m and the prime denote the mean and
deviatoric part of a second-rank tensor, as usual.
If now D = 0, A = 0 by eq. (II.21) (with X = 0). Calculating the tensor B
using eq. (IL3z), we get the following nonzero components: B113 = B1 31 = B223 =
B232 = Du;3i B311 = B322 = -Du;3i B33;3 = D33;3· The boundary conditions (11.1)
then read

(II.l8)
v3(x) = -Du;3 (xi+ x§) /2 + D33;3xV2
= Dm;3 (x~- xi- xD /2 + n;3;3 (xi+ X~+ 2xD /4
(for xi+ x§ + x~ = b2 ). The subscript m and the prime here denote the "mean" and
"deviatoric" part of a third-rank tensor, taken over its symmetric indices (unlike the
"deviators" defined by (II.9)). These boundary conditions can be written in spherical
coordinates r, 8, t_p in the following way:

vr(r = b,8,t_p) = ~Dm;3 b2 cos () + ~D; 3 ; 3 b2 (3 cos 28 -1) cos()

{
vo(r = b,8,t_p) = ~Dm;3 b2 sin ()- ~D; 3 ; 3 b2 (3 cos 28 + 1) sin() (Il.l9)

vcp(r = b,8,t_p) = 0.

Figure II.2 shows the deformations of the sphere arising from the terms proportional
to Dm; 3 and D; 3;3.
Of course, the most general boundary conditions are obtained by linear super-
position of those given by eqs. (II.17) and ((11.18) or (II.l9)).
II.3. b. Choice of trial velocity fields
Since the boundary conditions (11.17) and (II.l9) depend upon 4 parameters, we
need at least 4 trial velocity fields; in fact we shall just use 4 in our search for an
approximate, analytic criterion (a much larger number will be used to numerically
 108 M. Gologanu et al.

Figure II.2: Deformations of tbe sphere generated by Dm;a and D~ 3 ; 3

determine the supposedly exact criterion). The first two, which will be denoted vA
and vB and satisfy eqs. (II.17) for (Dm,D~ 3 )= (1,0) and (0, 1) respectively, are
exactly the same as in the work of Gurson:

A
v (x) = r~2 er; B
v (x) = -2e1-
XI X2
2e2 + xaea. (II.20)

Thus, like in Part I, we do not follow Garajeu's (1995) suggestion to replace the field
vB by the exact elastic solution for a purely. deviatoric loading in the search for an
analytic criterion, because, again, the improvement brought is important only for
high porosities, which are not of interest here.
It remains to define velocity fields v 0 and vD satisfying (11.19) for (Dm;a, D~ 3 ; 3 )
= (1/b, 0) and (0, 1/b) respectively. (The 1/b factor is introduced here in order that
the components of these fields be homogeneous to a length, just as those of vA and
vB). Incompressible fields satisfying conditions of axisymmetry and symmetry with
respect to all vertical planes containing the axis Ox 3 can be represented in the form

v(x) = \7 x [w(r, B)ecp] (II.21)

where w(r, B) is an arbitrary "stream" function. That function will be assumed here
to be of the form

w(r, B)= F(r~G(B) (11.22)

r sm B
(in fact equivalent to w(r, B)= f(r)g(B); the (r sinB)- 1 factor is introduced only in
order to simplify the equations below). The nonzero components of v simply become
then
vr(r,B,cp) = F(r)G'(B)/(r 2 sin B)
{ (11.23)
v9(r,B,<p) == -F'(r)G(B)/(r sin B).
Recent Extensions of Gurson's Model 109

The fields v 0 and vD being taken in that form, comparison with eqs. (11.19) shows
that, with obvious notations,

(11.24)

(11.25)

where some "normalization" conditions Fc(b) = Fv(b) = 1 have been introduced.

Equations (11.24) and (11.25) do not fully specify the functions Fc(r) and Fv(r).
Since the overall yield criterion is to be evaluated through minimization of the aver-
age plastic dissipation (7r(d(x)))c(X), it is sensible to determine these functions by
requiring the quantities (7r(d 0 (x)))C(X) and (7r(dD(x)))c(X), considered separately,
to be minimum. These average values are double integrals of the form

uo
1b 2 111' . 3uo 1b 2
41l'b3 / 3 a 21l'r dr 0 deq(r,O) sm (} d(} =----,;-- a (deq)rr dr,

where (f), is a short notation for (f(x))as(r)' i.e. the average value of the function
fover the spherical shell of radius r.

In order to simplify these expressions, we replace (deq) r by ( ~q) ~/ 2 , which

preserves the "upper bound" character of the approach since the latter quantity
cannot be smaller than the former. The calculation of the integrals over (} then
becomes feasible analytically and yields

(11.26)

( dD 2) = }!__ ( 104 F_b + 101 Ff/ + 13 FL) 2 _ 88 FvFb + ll FvFL) _ 13 FbFL))

eq r 630 r6 4 r4 8 r2 r5 r4 r3 2
(11.27)
where the argument of the functions has been discarded for simplicity. The expres-
sions to be minimized are then no longer double, but simple integrals of the form

I= 1b .C[F(r),F'(r),F"(r); r]dr , .C[F(r),F'(r),F"(r);r] = r 2 (d~q)~ 12 . (11.28)

Of course, when minimizing, one must account for the fact that the values of both F
and F' at the point r =bare fixed by eqs. (II.24 1 , 2 ) or (11.25 1 ,2 ).
Prior to calculating the functions Fe and Fv which achieve the minimum of the
corresponding integrals I, it is useful to make the following remark. It is obviously
 110 M. Gologanu et al.

possible to find functions Fe, Fv satisfying conditions (11.241,2) and (11.251,2) and
tending to zero with r quickly enough for the quantities (d~ 2)r, (d~ 2)r to be finite
(or even zero) in that limit. The corresponding integrals I are then finite. Thus
the average plastic dissipations corresponding to the fields ve and vD can be made
finite even for a -+ 0, i.e. for f -+ 0. This implies that the overall yield limit under
an arbitrary loading of "moment" type (arising from boundary conditions involving
some nonzero 'VD) will never be infinite, even for a zero porosity. This marks a
striking difference between loadings of "stress" and "moment" type, since the overall
yield limit under some purely hydrostatic stress becomes infinite in the limit f -+ 0.
Using the calculus of variations, it is easy to show that the "Lagrange equa-
tions" and "initial conditions" for the function F achieving the minimum of I while
respecting the condition that both F( b) and F' (b) be fixed are

d~22 ( ::,, ) -! (:~,) + ~~ = Oi [ ::,, L=a = Oi [! (::,, ) - :~, L=a = O.

(11.29)
Equation (11.29 1) is a 4th order differential equation on the function Fe(r) or Fv(r)
which, duly completed by the 4 conditions ((11.241,2) or (11.251,2)) and (11.292,3),
allows to calculate it; of course, this can only be done numerically. The procedure
is as follows. Since C is positively homogeneous of degree 1 with respect to the
arguments F, F', F", the differential equation (11.29 1) is positively homogeneous of
degree 0 with respect to the function F and its successive derivatives, and so are
also conditions (11.29 2,3), which means that if F(r) is a solution, >.F(r), where>. is
a positive constant, is also a solution. One can thus momentarily fix the value of
F(a) in an arbitrary way, and adjust (using Newton's method) that of F'(a) so as
to get, upon integration of (I1.291), the required value of the ratio F'(b)/F(b) (see
eqs. (I1.241,2) or (II.251,2)). Integration is possible because eqs. (I1.29 2,3 ) provide
the values of F" (a) and F"' (a) once those of F( a) and F' (a) are known. It only
finally remains to multiply the function F by the factor 1/ F( b) in order to respect
the "normalization" condition (11.241) or (11.251).
In practice, since we are interested only in small porosities, we consider small
values of the ratio ajb; the results depend only weakly on this parameter anyway
(provided that it is small). The values ajb = 0.05 and 0.001 are used for the functions
Fe and Fv respectively. (Use of smaller values for the function Fe proved impossible
because of numerical problems ni.;,r the point r = 0). Figure II.3 shows the resulting
quantities ( dfq 2) ~/ 2 , ( d~ 2) ~/ 2 , plotted as functions of ( r / b)3.
II.3.c. Derivation of some approximate overall criterion
We now consider a "general" velocity field of the form

(11.30)
Recent Extensions of Gurson's Model 111

( D 2) 1/2
deq r

0
(2/3) (de: d 0 ) r
-1
0.0 0.2 0.4 0.6 0.8 1.0

(It is recalled that the fields ve and v 0 correspond to (Dm;3, D~ 3 ; 3 ) = (1/b, 0)

and (0, 1/b) respectively; thus the b factor here ensures that the Dm;3 and D~ 3 ; 3
corresponding to v are precisely those appearing in the right-hand side). Then, if we
still denote II(D, V'D) what is in fact only an upper bound for that quantity,

II(D, V'D) _= (7r(d(x)))e(X) = 47l"b

CTo / 2 1 b
I 2 )1/2 ( 3)
CTo aa \deq
3 3 a (deq)r .47l"r dr :S b3 r dr 1~

(II.31)
where, by the definition of deq,

( d~q) T = v;, (d~q 2) r + v;~ (d~ 2) + b2 v-;,;3 (d~ 2) + b2 D~~;3 ( d~ 2)

T T T

(II.32)
+2b 2 Dm;3D~ 3 ; 3 .~ (de: d 0 )r.

"Crossed terms" other than 2 (de : dD) r /3 have been omitted here because they are
all zero. The values of ( d~q 2 ) r and ( d~ 2 ) r are easily found to be

fdA
\ eq
2) r -- 4 b6 . fdB
r6 ' \ eq
2) r -- 1
. (II.33)

Those of (d~ 2 )r and (d~ 2 )r are given by eqs. (II.26) and (II.27). Finally, calcula-
tion of 2 (de: d 0 )r /3 yields

-2(de.. dD) - b6
_ - ( FeFv FbFb 1 F(;F}!J
14--+5--+---- FcFb
8- -
3 r 90 6 4 r 2 5 r 2 r r

(II.34)
 112 M. Gologanu et al.

F' F
- 8 ____!2__P_ F. F"
+ ____£____!2_ F" F -
+ ____!2__P_ F' F"
__s______Q_ -
F" F' )
__s______Q_
r5 r4 r4 r3 r3
0

In order to get an analytic, approximate expression of the last integral in eq.

(11.31 ), we now introduce the approximation which consists in replacing the functions
(d~ 2)r, (d~ 2)r and 2 (d 0 : dD)r /3 in eq. (11.32) by some constants denoted /:).cc,
/:).DD and b..cv respectively; the precise value to be ascribed to these constants will
be discussed later. Figure II.3 shows that this is indeed reasonable for values of
(r jb )3 greater than, say 0.5, since these functions then vary between values which
are neither zero nor infinite. On the other hand, this approximation becomes grossly
erroneous for small values of ( r j b)3, since it fails to capture the behaviour of the
functions for r - t 0, namely the fact that they tend to zero in that limit. This is of
no importance, however, since for such values of (r/b) 3,.all functions are completely
overwhelmed by the term D~ ( d~ 2) r =4D~ b6 j r 6 anyway. With this approximation,
the expression (11.32) of ( a;q) r becomes

(·~q) r = 4D;. rb66 + P 2 , P2 =D~~ + b b..ccD;..'3 + b b..vvD;~.'3 + 2b2/:).cvDm·3D;3.

2 2
' '
3,
(11.35)
and the last integral in eq. (11.31) is then of the same type as that encountered
by Gurson, which can be calculated analytically. If we still denote II(D, V'D) that
integral, which is now in fact only an approximation of an upper bound, we thus get,
upon evaluation of the derivatives 811j8Dm and 811j8P and elimination of the ratio
Dm/ P between the expressions found:

2
21 ( -811 ) +2fcosh ( -
1 -811
- ) -1-f 2 =0. (11.36)
a0 8P 2ao 8Dm

Equation (11.36) represents the analytic expression of the approximate criterion; it

only remains to relate the derivative 811j8P to 811/8D~ 3 , 811/8Dm;3 and 811/8D~ 3 ; 3 ,
and then 8ITJ 8Dm and these three derivatives to the nonzero components of the
macroscopic stress :E and moment M.
To relate 811j8P to 811/8D~ 3 , 811j8Dm;3 and 811/8D~ 3 ; 3 , we note that eq.
(11.352) can be written in the form

- ( U. c .U )1;2 =_ (C01 pU 01 Up )112

P= (11.37)

where the 3-dimensional vector u =(u 01 ) and matrix C =(Cap) are defined by
/:).CD]
/:),.DD .
(1!.38)
Recent Extensions of Gurson's Model 113

It follows that
an an aP

c-1 an an = c-1 Ca,u, CpoU6 (an) 2= C,ou,uo (an) 2= (an) 2

a/3 aua aui3 a./3 P P ap P2 aP aP
where c;J denotes the (a, (3) component of the inverse matrix c- 1. Since, by eq.

(11.38 2 ), ;;_, ~ ~ b~' 1!.,~'] , thi' mean> that

[

( an)
2 ( an ) 2 ~ - 1 ( an ) 2 ~ -1 an ) 2 2 ~ - 1 an an
(
aP = aD~ 3 + b~c aDm;3 + ~D aD~ 3 ; 3 + b~D aDm;3 aD~ 3 ; 3
( 1!.39)
where use has also been made of eq. (11.38 1).
We shall now relate the derivatives anjaDm, anjaD~ 3 , anjaDm;3, anj8D~ 3 ; 3
to the quantities L;m,
L;33 - L;ll =
3L;~ 3 /2, M3lm• M3l33 - M3111 3M~I 33 /2 (the =
meaning of the subscript m and the prime here is the same as for the quantities D
and 'VD : they denote the mean and deviatoric parts, taken over symmetric indices if
the tensor considered is of rank 3). It must be noted here that we have not evaluated
n(D, 'VD) for arbitrary D and 'VD, but only for axisymmetric ones. That is, what
we have calculated is in fact

n(Dm,D~ 3 ,Dm;3,D~ 3 ; 3 ) =II(Dn,D22 = Dll,D33,Dll;3,D22;3 = D11;3,D33;3)

where zero components of D and 'VD have been discarded in the right-hand side.
Since Du = Dm- D~3/2, D33 = Dm + D~ 3 , Dn;3 = Dm;3- D~ 3 ; 3 /2 and D33;3 =
Dm;3 + D~ 3 ; 3 , one thus gets, using eqs. (1!.14):

and similarly

Inserting; these results into (11.39), then (11.39) into (11.36), one gets the final expres-
sion of the criterion:

_ 21 (( L;33 - L;ll) 2 + -Q22 )

<P(:E, M, f) = 3 L;m)
+ 2fcosh ( - - - 1- f2 = 0 , (11.40)
~0 b 2 ~0
 114 M. Gologanu et al.

where

It should be noted that although the component M 1 131 = M 2 132 does not appear
here, it is nonzero and equal to - M3 133 /2 by eq. (II.12). Also, in the limit f--+ 0,
the term "cosh" disappears so that its argument can take arbitrary values. One
would intuitively expect M 3 1m to appear in that term, in analogy with l:m, and thus
to become unlimited for f --+ 0. The fact that it appears in the square, just like
M31 3 3 - M3111, means that (as already mentioned) the yield limits under all loadings
of "moment" type remain finite for f--+ 0.
II.3.d. Values of the coefficients of the quadratic form Q2
It remains to give precise definitions of .6.cc, .6.DD and .6.cD· The best possible
values of .6.cc and .6.DD are those which, when substituted for (dfq 2 )r and (d~ 2 )r
in the expression (11.32) of (d;q)r, yield the exact result for the last integral in eq.
(II.31) for Dm = 0, D; 3 = 0 and (Dm;3,D; 3;3) = (1/b,O) and (0,1/b) respectively.
These conditions are satisfied for

- 1 b3
D 2 1/2 3
]2
.6.DD = [
b3- a313 (deq )r d(r )
(II.42)
Also, a good value for .6.c D is the average value of 2dc (x) : d D ( x) /3 over the matrix,

1
that is
b3

.6.c D = b3
1
3
-a a3 3
- I\ d C : d D) r d( r 3) .
2 (II.43)

Numerical computation of these integrals (using the same, small values of a/bas for
the calculation of the functions Fe and FD) and inversion of the resulting matrix ~
yields
.6.(7~ = 1.205 ; .6.[ib = 11.02 ; .6.C:b = 2.616 . (11.44)

11.4. Numerical determination of the exact criterion for a hollow sphere

strained axisyn,tmetrically and inhomogeneously
II.4.a. Principle of the numerical calculation
We now consider axisymmetric velocity fields defined by eq. (II.21 ), with "stream

l
functions" w( r, B) of the form

~ [
w(r, B)=- r sin B Boo cos B + t; m~oo
+oo +oo
Bkm G) m sin B Pi(cos B) (II.45)
Recent Extensions of Gurson's Model 115

(which can be verified to span the whole space of acceptable stream functions); the
nonzero components of the velocity field are then

vr(r,8) = b [Boo(b/r) 2 + 2:t~ 2::~:-oo k(k + 1)Bkm(rjb)m- 2Pk(u)] _

{ 'u =cos 8.
ve(r, 8) = b 2:t~ 2::~:-oo mBkm(r /b)m- 2 Pf( u)
(11.46)
Pk and Pf here are the associated Legendre functions of the first kind and order 0
and 1, like in Part I (see Gradshteyn and Ryzhik, 1980).
The coefficients Bkm are tied through linear constraints arising from the fact that
v must respect boundary conditions corresponding to the linear superposition of eqs.
(II.17) and (11.18). Using spherical coordinates and associated Legendre functions,
one readily puts these boundary conditions in the form

vr(b, 8) = Dmb + D~3bP2( u) + (Dm;3/2 + D~3;3/5) b2Pt ( u) + 3D~3;3b 2 P3( u )/10

{
ve(b,8) = D~ 3 bPi(u)j2 + (-Dm;3/2 + 2D; 3;3/5) b2Pl(u) + D~ 3 ; 3 b 2 P}(u)/10.
Comparison with eqs. (II.46) and identification of the terms corresponding to the
various values of k shows that for

2 2::~:-oo Btm = b (Dm;3/2 + D~3;3/5)

k = 0: Boo = Dm; k = 1: { ;
L~:-oo mBlm = b ( -Dm;3/2 + 2D~3;3/5)
6 2::~':-oo B2m = D;3 { 12 2::~':-oo B3m = 3bD;3;3/10
k = 2: { . 1.: = 3.
2::~:-oo mB2m = D; 3 /2 ' . . 2::~:-oo mB3m = bD~ 3 ; 3 /10
k(k + 1) 2::~:-oo Bkm = 0
k > 4: { (II.47)
- 2::~':-oo mBkm = 0 .

Elimination of the coefficients D; 3, Dm; 3 and D~ 3 ; 3 between these equations yields

L~:_ 00 (m- 3)B2m = 0

{ "L:~:_ 00 (m- 4)B3m = 2::~:-oo Bkm = 0
0 { fork 2 4.
' 2::~:-oo mBkm = 0
2::~:-oo [(m + 2)Btm- 24B3m] = 0
(II.48)
One can numerically compute the overall plastic dissipation (1T'(d(x)))c(X) for
any set of coefficients obeying eqs. (II.48). The procedure for numerical deter-
mination of discrete points belonging to the exact macroscopic yield locus is then
 116 M. Gologanu et al.

as follows. Like in Part I, we look for the intersection of this yield locus with a
given straight line; that is, if we assume E 33 - En, for instance, to be nonzero,
we fix the ratios T = Em/(E33 - Eu), X1 =
(M31 33 - M3111 )/(b(E33 - Eu)),
=
X2 M31m/(b(E33- Eu)). Then, by inequality (II.13),

(II.49)

The problem is thus, like in Part I, to minimize the right-hand side of this inequality
with respect to all sets of coefficients Bkm subject to the linear constraints (II.48) (the
values of D~ 3 , Dm, bD~ 3 ; 3 , bDm;3 here can be deduced from those of the coefficients
by eqs. (II.4 7) ). Again, because of the positive homogeneity of degree 0 of this ratio
with respect to the Bkm, one can arbitrarily decide that the denominator must be
unity, which imposes an extra linear constraint, and simply minimize (1r( d(x) ))c(X)
then.
In practice, the integers k and m are taken in the intervals [1, 10] and [-1, +8]
respectively (numerical experience has revealed that considering more negative values
of m is useless), which makes a total of 100 trial velocity fields. All calculations are
performed for a single porosity, f = 0.01.
II.4. b. Numerical results and adjustment of the coefficients of the quadratic form Q2
Even for axisymmetric loadings, which are the only ones considered here, the
macroscopic yield function depends on 4 mechanical variables, Em, E 33 - E 11 , M3 1m
and M3133 - M 3111 . Thus the yield locus is a 3 dimensional surface in a 4 dimensional
space. It is obviously impossible to numerically explore the entirety of this surface.
We shall therefore examine only the intersections of the yield locus with the "coor-
dinate planes", i.e. the 2 dimensional spaces obtained by setting 2 of the 4 variables
to zero. Each of these intersections is simply a curve in a plane. There are 6 such
curves, but we shall not consider that obtained in the (Em, E 33 - En) plane, for
M3lm = M3j33 - M3111 = 0, because the yield criterion of a hollow sphere in the ab-
sence of macroscopic "moments" has already been the subject of many investigations
(see for instance Gurson, 1977 and Gologanu et al., 1993).
Figure !!.4 shows the numerical yield locus in the plane (M3Im• M3l33 -M3Iu), for
Em= E33- Eu = 0, in the form of discrete points. (In view of the obvious symmetry
with respect to the origin, only the upper half is actually represented). The analytical
approximation defined by eqs. (II.40), (II.41) and (II.44) is also shown in the form of
a continuous line. That approximation is seen to be acceptable but numerical results
Recent Extensions of Gurson's Model 117

suggest the following slight modifications of the values of fl.c;.~, S 0b and fl.cb which
lead to the best possible agreement (see the second continuous line):

fl."C~ = o.992; f11Jb = 1o.1s; fl."Cb = 1.917. (II.44')

All theoretical curves will be represented hereafter for these new, improved values of
A -1 A -1 d A -1
ucc• uvv an ucv·

o.s (M3133- M3111)/(bcro)

0.4

0.2

0.1

0.0~~--~~--~~--~--~~--~~--~~
-0.6 -0.4 -0.2 -0.0 0.2 0.4 0.6

Figure II.4: Yield loci for f = 0.01, Em = E33 - Eu = 0

Remark. Recall that the values of ~C~• f11Jb and fl.cb given by (1!.44) correspond
to formulae (1!.42) for fl.cc and fl.vv and (II.43) for fl.cv. These formulae are such
that when fl.cc and ~DD are substituted for ( d~ 2 ) r and ( d~ 2 ) r in the expression of
(a:_q) r, they yield the exact result for the last integral in eq. (1!.31) for Dm = D~ 3 = 0
and (Dm;3, D; 3;3) = (1/b, 0) and (0, 1/b) respectively. That integral being simply
denoted II(D, VD) as above, we then get, by eq. (11.13), the following rigorous
inequalities:

3M3im/b ~ II(D = 0, Dm;3 = 1/b, D; 3;3 = 0)

{
(M3I33- M3lu)/b ~ II(D = O,Dm;3 = O,D~ 3 ; 3 = 1/b).
These inequalities provide rigorous upper bounds for M 3 1m independently of the
values of Em, E33 - Eu and M 3133 - M 3111 , and for M 3133 - M 3111 independently of
the values of Em, E33- Eu and M 31m· Geometrically, this means that in Figure II.4,
the approximate yield criterion corresponding to eqs. (1!.44) should overestimate
the maximum possible values of M 31m and M 3133 - M 3111 . This is readily verified to
be indeed true. On the other hand, this property can be observed not to hold for
the approximate yield criterion corresponding to eqs. (1!.44'), which is not surprising
 118 M. Gologanu et a!.

since the corresponding values of 6.cc, D.vv and 6.cv only provide an "uncontrolled"
approximation of the last integral in eq. (11.31).

Prior to giving other numerical results, it is useful to make the following remark.
Starting from some axisymmetric velocity and stress fields respecting the microscopic
constitutive equations, and applying a symmetry with respect to the plane x 3 = 0 to
these fields, one obtains some new fields still respecting the constitutive equations. In
this operation, Em = (2E 11 + E 33 )/3 and E 33- E 11 remain unchanged but M 31m =
(2M3I11 +M3I33)/3 and M3l33 -M3I11 change sign. Therefore the (exact) yield locus
is invariant in the transformation (~, M) ----+ (~, -M). This means that the curves
representing the intersections of this yield locus with those "coordinate planes" which
involve both a "stress" component and a "moment" component are symmetric with
respect to the "stress" axis. Since they are obviously symmetric with respect to
the origin, they also admit the "moment" axis as an axis of symmetry. Thus it is
sufficient to represent them in a quarter of a plane only.

Figures II.5.a &J.d b, II.6.a and b represent the numerical and theoretical yield
loci in the planes (Em, M3lm) (for E33 -En = 0, M3l33 - M3111 = 0), (Em, M3l33 -
M3ln) (for E33-Eu = 0, M3lm = 0), (E33-Eu,M3Im) (for Em= 0, M3I33-M3I11 =
0) and (E33 -En, M31 33 -M31 11 ) (for Em = 0, M3lm = 0) respectively. The agreement
between numerical results (discrete points) and theoretical ones (continuous lines) is
acceptable in all cases. The discrepancy is maximum for the last curve; it arises
from the fact that the tangent to the theoretical curve is vertical on the horizontal
E33 - Eu axis, whereas the tangent to the numerical yield locus is not (this has
been carefully checked through more detailed examination of the region near the
horizontal axis). In view of the symmetry with respect to the horizontal axis, the
latter property means that the exact (numerical) yield locus has an angular point on
the (horizontal) E33 - En axis in the (E33 - Eu, M 3133 - M 3111 ) plane. A similar
phenomenon can be observed in the (Em, M 31m) plane, although the corner is less
sharp. Unfortunately, the approximate yield function proposed cannot capture such
features of the exact one, whatever the values of the parameters .6.(;~, .6. vb and
t.-.c;b, since it is quadratic with respect to the components of the moment.
11.5. Comparisons with some finite element simulations of the evolution
of a hollow sphere strained axisymmetrically and inhomogeneously

II.5.a. Principle of the simulations

We shall now describe the results of some finite element calculations of the
behaviour of a hollow elastic-plastic sphere strained axisymmetrically and inhomoge-
neously. There are two main differences with respect to the numerical minimizations
presented in the preceding section, which arise from the fact that these minimizations
were based on limit-analysis, whereas the present simulations use the full constitutive
law of the matrix material. The first novelty is that the present simulations follow the
 Recent Extensions of Gurson's Model 119

a
0.3

0.2

0.1 f=0.01

0.0 '---~~'-''----'---'~---L~----'-~----
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0

Figure II.5: Yield loci for L:33- L:n = 0 and M3l33- M3111 = 0 (a) or M3lm = 0 (b)

Figure II.6: Yield loci for L:m = 0 and M3l33- M3111 = 0 (a) or M3lm = 0 (b)

evolution of the cell and its void in time, and in particular account for their change of
shape and the coalescence of cavities. The second one is that because of the amount
of calculation required, we shall not attempt here to explore even a small portion
of the macroscopic yield criterion, but concentrate instead on just two mechanical
histories.

Between any two successive instants of the calculation, one imposes boundary
120 M. Gologanu et a!.

!
increments of displacement ~u of the form, in cylindrical coordinates p, i.p, z :

~up(p, z) =~En p+ ~E11;3 p(z- Z)

~u"'(p,z) = 0 (II.50)

~uz(p, z) = ~E33(z- Z)- ~El1;3 p 2 /2 + ~E33;3(z- Z) 2 /2

where Z denotes the vertical coordinate of the centroid of the cell (which is zero
only initially, when the cell is spherical). Equations (II.50) are identical to the linear
superposition of eqs. (II.17) and (ILlS) except for the use of cylindrical coordinates,
=
the replacement ofv and D by ~u and ~E, and that of x 3 z by z-Z. The values of
b~E11;3/ ~E~ 3 and b~E33;3/ ~E~ 3 (where ~E~ 3 =
2(~E33- ~E11 )/3) are fixed dur-
ing the entire calculation1 . Two cases are envisaged: (b~E11;3/ ~E~ 3 , b~E33;3/ ~E~ 3 )
= (2.5,0) (case a) and (0, 10) (case b). (These values are chosen in such a way
that the components of :E and M/b be of the same order of magnitude). The
value of ~E 11 / ~E33 is adjusted at every step so as to enforce a constant triaxi-
ality Em/(E 33 - E 11 ) of 3, with E 33 > E 11 . (Only high triaxialities are of interest
here, since high gradients of macroscopic mechanical fields are encountered only near
crack tips).
The initial porosity is f = 0.01 as before. The matrix is ideal-plastic, with pa-
rameters E (Young's modulus) = 200.000 MPa, v (Poisson's ratio) = 0.3 and O"o (yield
stress in uniaxial tension) = 400 MPa. The calculations are stopped once the overall
cumulated deformation Eeg =
J0t(2D': D' /3) 112(T)dT reaches a value of 0.1. In the
absence of "moments", coalescence of cavities is known to start, for such a material
and loading, around Eeq = 0.03 (Koplik and Needleman, 1988); thus one expects to
observe it here. The simulations are performed using the Large Deformation Plastic-
ity option of the SYSTUS finite element code developed by FRAMASOFT+CSI.
II.5.b. Numerical results and comparison with the theoretical model
In all subsequent figures, discrete points represent numerical values and contin-
uous lines theoretical ones. Figure II. 7 shows the macroscopic von Mises equivalent
stress Eeg =(3:E' : :E' /2) 1 12 as a function of Eeg for the two cases envisaged. The
results corresponding to the "reference case" where no moments are present are also
indicated with a symbol "R" 2 • The presence of moments can be seen to have a strong

1 Rigorously speaking, the definition of the parameter b here is unambiguous only at the start of
the simulation, when the cell is spherical. However, since the overall deformation to be undergone
by the cell is moderate, it is reasonable to consider that parameter as a constant.
2 One may be surprised not to detect any sign of coalescence on the curves in that case; but it

was no more visible on those obtained by Koplik and Needleman (1988). For such high triaxialities
(and in the absence of moments), the only reliable way to detect coalescence is to observe when the
cell ceases to deform laterally and its mode of deformation becomes a uniaxial extension.
Recent Extensions of Gurson's Model 121

influence on L:eq, and the theoretical predictions are acceptable in both cases at the
beginning of the simulations. In case a, both numerical and theoretical values are
pretty much stationary, which seems to imply that no coalescence takes place. In
case b, numerical values begin to notably decrease once Eeq reaches a value of about
0.04, which is an indication that coalescence occurs. Theoretical ones remain almost
constant, however, since the model developed does not account for coalescence.

a b
0.6 ~
0.6 ~
0.4

0.4
0.2
ooooooooooooooooooo
0.2 L...o...~.....L.....o~"'--'-~--'--'L....o-..~-::::..o-~....... 0.0 L....~...L.......~-'-~"'--'-~__.__.~...........,
0.00 0.02 0.04 0.06 Eeq 0.1 0 0.00 0.02 0.04 0.06 Eeq 0. 10

Figure II. 7: L:eq / ao as a function of Eeq

Figure 1!.8 shows the porosity as a function of Eeq· In case a, the presence
of moments has a paramount influence on void growth, which can be observed to
be completely hindered in the numerical simulation. The model does capture the
effect, although to a somewhat insufficient extent. The influence of moments is also
notable in case b. Here the agreement between the theory and the simulation again
deteriorates for Eeq 2:: 0.04, due to coalescence.
Figure II.9 shows the nonzero components of the moment. The agreement be-
tween theoretical and numerical results is acceptable at the beginning of the simu-
lation but tends to deteriorate in time. This is certainly due to the fact that the
evolution of the cell shape is not accounted for by the model. In fact, by the very
definition (II.4 2 ) of the moments which involves the distance to the centroid of the
cell, it is clear that they are more sensitive than the (macroscopic) stresses to that
shape. Coalescence also probably has some influence in case b.
Figure II.lO shows the deformed mesh at the end of both simulations. The "pear"
shape of the void in case b clearly indicates a concentration of the deformation in
the ligament located slightly above the middle horizontal plane, which is a sign that
coalescence takes place. No such feature can be observed in case a.
In conclusion, the most notable feature evidenced by these simulations is the
strong influence of moments (or of gradients of the macroscopic strain rate) upon
122 M. Gologanu et a!.

0.10 0.08
f a f b~
~
~
0.08 ~

~~
0.06 ~
~

~ ~
~
~
0.06 ~
~~ R~ ~ ~
~
R ~~ 0.04 ~~
~
~
~~ ~
0.04 ~~ ~~
~~
~

~~ ~~
~~ ~~ ~~
~~ 0.02 ~~
~~
0.02 ~
~~ ~

~ ~~~~~~~~~~~~~~~~
0.00 L........~..L...~._L_~......_._~~=---....._.__. 0.00 L........~..L...~..._....._~......_._~~=--~
0.00 0.02 0.04 0.06 Eeq 0.1 0 0.00 0.02 0.04 0.06 Eeq 0.1 0

Figure II.8: f as a function of Eeq

0.6 ~ ~~~~~~~~~~~~
~~~~~~

19
0.4 M31d(bcro)

~~~~~~~~~~~~~~~~~~~
0.2
M3133l(bcro)
0.0 -0.1
M1131/(b cro)
QQQQQYYYV VVYYYYYYV V

-0.2 -0.3L........~~~~~~~~~~~

0.00 0.02 0.04 0.06 Eeq 0. 10 0.00 0.02 0.04 0.06 Eeq 0.10

Figure II.9: M3in/(bao), M3l33/(bao) and M 1 13If(bao) as functions of Eeq

growth and coalescence of voids.

11.6. Conclusion: constitutive equations for ductile porous metals incor-

porating the effect of strong gradients of macroscopic fields
II.6.a. Generalized criterion for arbitrary loadings
When looking for some heuristic extension of the criterion defined by (II.40)
and (II.41) to arbitrary, non-axisymme tric loadings, one must respect isotropy (since
the representative volume element considered is isotropic), which implies that only
invariants of ::E and M can appear in the expression of the yield function <I>(::E, M, f).
 Recent Extensions of Gurson's Model 123

a b

Figure II.lO: Deformed meshes at the end of the simulations

Also, <I> must be convex with respect to :E and M.

With regard to :E, these requirements are easily met by replacing (L: 33 - L: 11 ?
in (II.40) by L:~q· With regard to M, it proves exceedingly difficult to respect the
condition of convexity of lll if non-quadratic invariants are used. This suggests to just
replace, in eq. (II.40), the quadratic form Q2 given by (II.41 ), which only applies
to the axisymmetric case, by some more general quadratic form. It is easy, though
tedious, to check that there are only 3 independent quadratic invariants for general
tensors M obeying the symmetry property Mklii= Mklji and the condition (II.12):

1M M - 23 M' M' 3 M'

- 2
M I ;:::: g ilii ilkk = M ilm M ilmM II =
; iljk iljk .' M III = iljkMjlik
I
·
(II. 51)
Thus the general expression of Q2 should be of the form

(II.41')

where AI, All and All I are coefficients. Of course, these coefficients must be de-
termined through identification of formulae (II.41) and (II.41') in the axisymmetric
case.
However, calculation of the 3 above invariants in the axisymmetric case shows
that they happen not to be independent then; that is, M I I I can be written in the form
o:MI +,BMII in that case. Thus identification of (II.41) and (II.41') can only provide
124 M. Gologanu et a!.

the values of AI+aAui and Au+,BAuJ but not those of A1, Au, Au! individually.
Hence we arbitrarily decide to take Au 1 = 0. The following problem then arises.
Expression (II.41), which depends on 3 parameters, 6 0 ~, 6[;b and 60b, obviously
cannot be matched with only 2 constants, AI and Au. To investigate that question,
it is convenient to consider (in the axisymmetric case) the quantities X andY defined
by X = 2M3 Ill+ M3l33 ( = 3M3 1m) and Y =
-M3Ill + 2M3I33· Simple calculations
show that expressions (II.41) and (II.41') (considered in the axisymmetric case) can
be written in the forms

respectively, where

Identification of these expressions is possible if the "crossed" term proportional to XY

in the first one is negligible, i.e. if a).;y ~ lax xayy I; happily enough, this happens
to be true for the values which result from eqs. (1!.44') ( a).;y /(ax xayy )= 0.002).
The values of A1 and Au (and Au I) are then, by what precedes,

A1 = 9 ( 6 0 ~- 510 6[;b- ~6cb) = 0.194; Au= ~6[;b = 6.108; Au!= 0.

(II. 52)
In conclusion, the generalized criterion proposed reads

<I>(~, M, f) _ 1(
= 175 2
l::eq
Q2 ) + 2fcosh
+ {;2 (32-;;:;;-
l::m) - 1- f2 =0, (II.40')

Q2 being given by (II.41'), (II.51) and (II.52). It can be checked that the quadratic
form Q 2 is positive-definite over the space of tensors M obeying the symmetry con-
dition Mklii = Mklii and eq. (I1.12). Thus the function <I> is convex over the space
of physically admissible pairs (~, M), as required.
II.6.b. Introduction of strain hardening and void nucleation
We now drop the hypothesis of ideal-plasticity and assume that the matrix is
subject to some strain hardening of isotropic type, characterized by some function
17( t) that gives the true (Cauchy) stress as a function of logarithmic strain in a uni-
axial tensile test. We follow Gurson's heuristic suggestion to replace the real porous
material, strained inhomogeneously, by some fictitious one in the matrix of which the
=
equivalent cumulated strain f. is uniform. The yield stress a 17(f.) is then also uni-
form in the matrix so that the criterion derived earlier remains applicable but for the
replacement of O"o by a. The evolution equation for the parameter f. is also derived in
Recent Extensions of Gurson's Model 125

the same way as in Gurson's approach, that is by identifying the macroscopic plastic
dissipations of the real and fictitious materials. The only difference with respect to
the work of Gurson is that here, because of eq. (II.5), the expression of the former
dissipation is no longer simply :E : D but :E : D+M:vn. We thus get

(II. 53)

Also, void nucleation can be introduced in the usual way, by decomposing the
porosity rate into two contributions due to growth and nucleation of voids respec-
tively:
(II.54)
where A(€) is a given (for instance Gaussian) function. The expression of j 9 here is
the same as usual (in particular it does not involve 'VD), since it simply results from
incompressibility of the matrix.
II.6.c. Introduction of bypoelasticity
Although elasticity effects are generally quite small in ductile rupture, (hypo)-
elasticity must be incorporated into the constitutive equations for practical, numerical
simulations to be feasible, since finite element programs cannot accomodate rigid-
plastic materials.
It is thus postulated that both D and 'VD can be decomposed in the form

(II. 55)

Note that the elastic and plastic parts ('VDY, ('VD)P of the gradient of D have no
reason to coincide with the gradients 'V(De), 'V(DP) of the elastic and plastic parts
ofD. .

The quantities DP and ('VD)P are given by the flow rule (II.16), which is assumed
to apply to them instead of D and 'VD :

acp . (v D )Pi).k =HaMacp + VikuJ

D pij -- H ._. , • + 81kui· (II.16')
8 6ij klij

Also, oe and ('VD)e are supposed to be tied to the Jaumann derivatives (for
instance) D:Ej Dt, DMJ Dt of the stress and moment tensors through some elastic
moduli. The expression of D:EJ Dt is well known. To find that of DM/ Dt, let us
denote 0 = ('VV- ('VVf) /2 (V : macroscopic velocity) the Jaumann rotation
rate. Also, let us decompose M in the basis ( e1, e2, e3) "tied to the matter", which
rotates at the rate 0 (that is, ei = O.ei = OJieJ): M = MkliJek ®ei ®eJ· Then the
time-derivative of M in the reference frame is
 126 M. Gologanu et al.

whereas that in the matter-tied frame is

(the basis vectors are to be considered as constant here by the very definition of that
time-derivative). It follows that

DM/Dt = M- Mklij (ek l8l e; l8l ej + ek l8l e; l8l ej + ek l8l e; l8l ej)
= M- Mklij (flhkeh l8l e; l8l ej + ek l8l nhieh l8l ej + ek l8l e; l8l nhjeh) (11.56)

where the antisymmetry of n has been used.

DY.:, / Dt being assumed to be tied to De by the familiar elasticity law, it remains
to define the elastic moduli connecting DM/ Dt and (\7D)e. Some crude approxima-
tion is quite sufficient here. Such an approximation can be obtained by considering
the case of a purely elastic medium in small strain, and trial displacement fields
corresponding to a rigid-body motion for macroscopic strain gradients of the form
(II.ll ), and of type (ILl) (everywhere in the spherical cell considered, not only on
its boundary) for "deviatoric" ones in the sense defined by eq. (II.9I). The resulting
relation between DM/ Dt and (\7Dl is easily found to be

( DD~) klij = ~ ( .Ac5;j( \7 D )/,hk- ~6ik( \7 D )hhj - ~6jk( \7 D )/,hi+ 211( \7 D )ijk)
(11.57)
where ,\ and J1 are the Lame coefficients and a multiplicative 1 - f factor has been
neglected.
It must finally be mentioned that in the presence of (hypo )elasticity, D and \7D
are to be replaced by DP and (\7D)P in eqs. (11.53) and (II.54 2 ).
II.6.d. Elimination of second derivatives of the velocity and of the resulting need for
C1 elements
One very serious drawback of the model just proposed lies in the appearance of
second derivatives of the macroscopic velocity in the constitutive equations. Indeed
this makes it necessary to use finite elements of class C1 , which are complex and
generally not available in standard codes, for numerical simulations of ductile rupture
of structures.
However, this problem is quite similar to that encountered in the numerical
implementation of the Kirchhoff-Love plate or shell theory, and can be remedied in
the same way, through some slight alteration of the theoretical formulation. Indeed,
 Recent Extensions of Gurson's Model 127

let us introduce, in addition to the macroscopic velocity V(X), an extra kinematic

degree of freedom in the form of a symmetric second-rank tensor W(X); this tensor
differs from D(X) a priori, but will be forced in fine to be almost equal to it. This
procedure is analogous to dropping the Kirchhoff-Love hypothesis and considering
the rotation of the segments joining the lower and upper surfaces bounding the plate
or shell as a new degree of freedom independent of the displacement of the middle
surface. Next we introduce some mechanical symmetric second-rank tensor P(X)
"dual" to the "strain rate" (W- D)(X) in the sense of the principle of virtual work;
that is, we modify the expression (II.6) of p(i) into

p(i) =:-in [:E(X): D(X) + M(Xfv'W(X) + P(X): (W- D)(X)]dV . (II.6')

This is equivalent, in the context of plates and shells, to introducing the notion of
transverse shear. We next replace \7D, (\7D)e and (\7D)P everywhere in the preced-
ing constitutive equations by \7W, (\7W)e and (\7W)P. The new theory obtained
in that way plays the same role with respect to the old one as the Reissner-Mindlin
plate or shell theory with respect to that of Kirchhoff and Love. Finally we oblige W
to be (almost) equal to D by enforcing the following relation between W - D and
the Jaumann derivative of P :

( DP) = !!:_(W· - D· ·)
Dt . . E •J •J
(II.58)
•J

where E is a small parameter. This is equivalent to letting the thickness of the plate
or shell go to zero so as to make the transverse shear stiffness become very large as
compared to the bending stiffness.
Numerically, W will be treated as a nodal degree of freedom, and the evaluation
of \7W at the Gauss points of the elements will only require the use of first derivatives
of the shape functions, whereas that of \7D =\7 (\7V + (\7Vf) /2 would have
required that of second derivatives. Thus the need for finite elements of class C1 is
removed.
It is instructive to examine the equilibrium equations and boundary conditions
resulting from the new expression (II.6') of p(i). Application of the principle of virtual
work yields, upon integration by parts of this expression and use of eq. (II. 7):

0 = p(i) + p(e) =in [(~ii;i- P;j;j)Vi + (Mklii;k- P;j)W;j] (X)dV

+ { [(( -~;i + P;j)nj + T;) Vi- Mkliink W;j] (X)dA
lan
* { ~ij;j(X)- P;j;j(X) =_0 inn; (II. 59)
Mklij;k(X) - P;j(X) - 0
 128 M. Gologanu et a!.

(~ii(X)- Pii(X))ni(X) = Ti(X)

{ on an. (11.60)
Mklii(X)nk(X) =0
It is easy to eliminate Pij;j(X) between eqs. (11.591) and (11.59 2 ); the result is eq.
(11.8). Thus the equilibrium equations of the new theory are the same as those of
the old one.
Equations (11.59) and (11.60) are interesting because they allow to understand in
a qualitative manner why infinite concentration of damage will be prevented within
the new theoretical framework developed (which is indeed the ultimate goal pursued
in developing it). Indeed certain components of M are zero on the lips of a crack by
eq. (11.60 2 ). If the surrounding damaged zone were infinitely thin, those components
would take finite, nonzero values as soon as one would shift the observation point away
from these lips. Hence certain components of VM would be infinite. But, because
of the equilibrium equations (11.59), \7~ behaves like VVM, so that ~behaves like
VM. Therefore certain components of ~ would be infinite, which is prohibited by
the criterion. Thus the thickness of the damaged zone cannot be zero. Of course,
in practice it will be of the order of the characteristic length (typical half-spacing
between voids) b introduced into the model.

REFERENCES
Budiansky, B., Hutchinson, J.W., and Slutsky, S., 1982, Void growth and collapse
in viscous solids, in Mechanics of Solids, Hopkins and Sewell, eds., Pergamon Press,
Oxford, pp. 13-45.
Drugan, W.J., and Willis, J.R., 1996, A micromechanics-based nonlocal con-
stitutive equation and estimates of representative volume element size for elastic
composites, J. Mech. Phys. Solids, 44, 497-524.
Garajeu, M., 1995, Contribution a l'etude du comportement non lineaire
de milieux poreux avec ou sans renfort, These de Doctorat, Universite de la
Mediterranee, Marseille.
Germain, P., 1973a, La methode des puissances virtuelles en mecanique des
milieux continus. Premiere partie: Theorie du second gradient, J. Mecanique, 12,
235-274.
Germain, P., 1973b, The method of virtual power in continuum mechanics. Part
2: Microstructure, SIAM J. Appl. Math., 25, 556-575.
Gologanu, M., 1996, These de Doctorat, in preparation, Universite Rierre et
Marie Curie, Paris.
Gologanu, M., Leblond, J.B., and Devaux, J., 1993, Approximate models for
ductile metals containing non-spherical voids - Case of axisymmetric prolate ellip-
soidal cavities, J. Mech. Phys. Solids, 41, 1723-1754.
Gologanu, M., Leblond, J.B., and Devaux, J., 1994, Approximate models for
Recent Extensions of Gurson's Model 129

ductile metals containing non-spherical voids - Case of axisymmetric oblate ellipsoidal

cavities, ASME J. Eng. Mat. Tech., 116, 290-297.
Gradshteyn, I.S., and Ryzhik, I.M., 1980, Table of Integrals, Series, and
Products, Academic Press, New-York.
Gurson, A.L., 1977, Continuum theory of ductile rupture by void nucleation and
growth: Part I - Yield criteria and flow rules for porous ductile media, ASME J.
Eng. Mat. Tech., 99, 2-15.
Hill, R., 1967, The essential structure of constitutive laws for metal composites
and polycrystals, J. Mech. Phys. Solids, 15, 79-95.
Koplik, J., and Needleman, A., 1988, Void growth and coalescence in porous
plastic solids, Int. J. Solids Structures, 24, 835-853.
Leblond, J.B., and Perrin, G., 1995, Introduction a Ia mecanique de Ia
rupture ductile des metaux, Cours de l'Ecole Polytechnique, Paris.
Leblond, J.B., Perrin, G., and Devaux, J., 1994, Bifurcation effects in ductile
metals with nonlocal damage, ASME J. Appl. Mech., 61, 236-242.
Lee, B.J., and Mear, M.E., 1992, Axisymmetric deformation of power-law solids
containing a dilute concentration of aligned spheroidal voids, J. Mech. Phys.
Solids, 40, 1805-1836.
Mandel, J ., 1964, Contribution tbeorique a l'etude de l'ecrouissage et des lois
d'ecoulement plastique, in Proceedings of the 11th International Congress on
Applied Mechanics, Springer, Berlin, pp. 502-509.
Mazataud, P., 1995, private communication.
Mindlin, R.D., 1964, Microstructure in linear elasticity, Archives Rat. Mech.
Anal., 16, 51-78.
Mindlin, R.D., and Eshel, N.N., 1968, On :first strain-gradient theories in linear
elasticity, Int. J. Solids Structures, 4, 109-124.
Mura, T., 1982, Micromechanics of Defects in Solids, Martinus Nijhoff
Publishers, The Hague.
Pijaudier-Cabot, G., and Bazant, Z.P., 1987, Nonlocal damage theory, ASCE
J. Eng. Mech., 113, 1512-1533.
Pineau, A., and Joly, P., 1991, Local versus global approaches of elastic-plastic
fracture mechanics. Application to ferritic steels and a cast duplex stainless steel,
in Defects Assessment in Components - Fundamentals and Applications,
Blauel and Schwalbe, eds., ESIS, EGF publication 9.
Ponte-Castaneda, P., and Zaidman, M., 1994, Constitutive models for porous
materials with evolving microstructure, J. Mech. Phys. Solids, 42, 1459-1495.
Sovik, O.P., 1995, private communication.
Suquet, P., 1982, Plasticite et homogeneisation, These de Doctorat d'Etat,
Universite Pierre et Marie Curie, Paris.
Suquet, P., and Ponte-Castaneda, P., 1993, Small-contrast perturbation expan-
sions for the effective properties on nonlinear composites, C. R. Acad. Sci. Paris,
Serie II, 317, 1515-1522.
 130 M. Gologanu et al.

Tvergaard, V., and Needleman, A., 1995, Effects of nonlocal damage in porous
plastic solids, Int. J. Solids Structures, 32, 1063-1077.

APPENDIX
A variant of the model proposed in Part I can be obtained by simply changing the
values of the parameters Q and H in eqs. (I.4 7) without modifying these equations
themselves. Recall that the values H::::::: "£h/O'o, Q::::::: ("£33- "£ 11 )/0'o provided by
eqs. (!.46), which are supposed to be employed in eqs. (!.47), correspond to points
belonging to our exact two-field yield locus. These points were obtained for A = 0,
i.e. for Dm = 0 since. D~ = 0. If we consider the points ("£ 11 , "£ 33 ) belonging to Ponte-
Castaneda and Zaidman's (1994) yield locus and also corresponding to Dm = 0, and
calculate the ensuing values of Q and H, we get

{
Q = (1- nj(l- 3ai)/ [1- 3al + 3f(ai- afar)]
(I.46')
H = 2(afar- a2)Q

where the notations have been defined in the text. Using these values in eqs. ( I.4 7),
we get a yield locus that crosses that of Ponte-Castaneda and Zaidman and is even
tangent to it at those points of the latter yield locus corresponding to Dm = 0. The
practical merits of this yield locus are explained in Section I.5.b of the text.
 NONLINEAR COMPOSITE MATERIALS:
EFFECTIVE CONSTITUTIVE BEHAVIOR
AND MICROSTRUCTURE EVOLUTION

P. Ponte Castaneda
University of Pennsylvania, Philadelphia, PA, USA

ABSTRACT

This article is a review of some recent developments in the field of nonlinear composite
materials. More precisely, it is an attempt to summarize and relate certain methods that have
been developed over the past 10 years for estimating the effective, homogenized or overall
constitutive behavior of composite materials with nonlinear constituents. In addition to their
ability to handle constitutive nonlinearity, these homogenization methods have also been
extended in a consistent fashion to account for geometric nonlinearity. In other words, they
are able to incorporate the effect of changes in the microstructure, due to the presence of
finite deformations, on the effective constitutive behavior of the composites. These
homogenization procedures- like some of the earlier, now classical procedures- make
use of a "linear comparison composite," for which the effective behavior can be assumed to
be readily available from the extensive literature in the field of linear composite materials. In
particular, the use of a linear comparison composite provides a way of bringing in higher-
order statistics (beyond volume fractions) into the description of the effective behavior of
composites with random microstructures, whether they be "particulate" or "granular" in
character. However, the choice of the linear comparison composite in these modern
homogenization procedures is dictated by appropriate approximations in the context of
rigorous variational principles for the effective energy of the composite. This simple fact
goes a long way toward explaining the significant improvements achieved by the new
methods over the more classical procedures, where the choice of the comparison composite
was not derived from a variational principle.
132 P. Ponte Castaneda

1. INTRODUCTION

By now, much is known about the characterization of the effective properties of

composite materials with linear constitutive behavior. When their microstructure is periodic,
the effective properties may be determined exactly in terms of unit cell problems with
appropriate boundary conditions [1-2]. For random microstructures, the problem is not
deterministic and the effective properties cannot be determined exactly. Instead, the goal
becomes to determine the range of possible effective behavior in terms of bounds depending
on the known statistics- usually only up to two-point- of the microstructure. Various
homogenization methods have been developed for this purpose including the averaging
methods of Voigt and Reuss, which were shown to lead to rigorous bounds on the effective
moduli of N-phase composites with prescribed volume fractions, and of polycrystals with
given orientation distriLution functions, by Hill [3] and Paul [4]. More refined bounds
incorporating two- and three-point statistics were made possible by the works of Hashin &
Shtrikman [5-7] and Beran [8], respectively (see the reviews of Willis [9-10]). Alternatively,
ad hoc procedures have been proposed to estimate the effective behavior of composites with
special classes of microstructures. Perhaps the best known among these is the self-consistent
method [11-15].
By comparison, much less is known about the effective behavior of composites with
nonlinear constitutive behavior. For periodic microstructures, once again the problem is
deterministic and the effective properties may be determined in terms of unit cell problems
with appropriate boundary conditions [ 16-17], at least provided that the underlying potentials
are convex [18]. However, in spite of their simplicity from a theoretical point of view, the
periodic hypothesis has several limitations including the exclusion of many naturally
appearing and man-made microstructures, as well as difficulty in achieving certain standard
material symmetries, such as isotropy. For composites with random microstructures, the
only rigorous information available until very recently was given by the bounds of Bishop &
Hill [19-20] for rigid perfectly plastic polycrystals, one of which had been proposed as an
approximation much earlier by Taylor [21]. The Taylor-Bishop-Hill estimates constituted a
generalization of the Voigt-Reuss-Hill "microstructure-independent" bounds for nonlinear
composites and polycrystals. In addition, various generalizations of the self-consistent
procedure were proposed over the years, including the incremental procedure of Hill [22]
and the (simplified) secant methods of Hutchinson [23] and Berveiller & Zaoui [24 ], as well
Nonlinear Composites and Microstructure Evolution 133

as more specialized results including the dilute estimates of Gurson [25] for porous
materials. It was not until the work of Willis & Talbot [10, 26, 27], which provided
extensions of the Hashin-Shtrikman variational principles for nonlinear composites, that the
computation of rigorous nonlinear bounds incorporating two-point information first became
possible. This pioneering contribution was followed by the introduction of several new
variational principles making use of appropriately chosen "linear comparison composites,"
which allowed the determination of Hashin-Shtrikman, as well as more general bounds and
estimates - directly from corresponding estimates for linear composites. These included the
variational principles of Ponte Castaneda [28, 29] and Talbot & Willis [30] for general
classes of nonlinear composites, as well as those of Suquet [31] and Olson [32] for power-
Jaw and perfectly plastic composites, respectively. These new variational principles and their
application to the estimation of bounds and estimates for nonlinear composites is the subject
of the first part of this review.
The second part of this review is concerned with the effect of changes in the
microstructure, due to the presence of finite deformations, on the effective constitutive
behavior of nonlinear composite materials. Although this feature, introducing geometric
nonlinearities into the problem of estimating effective behavior, has been recognized to be
important in the context of texture evolution of polycrystals, going back to the early work of
Taylor [21 ], it has been almost consistently ignored in the context of more general types of
composite materials. Perhaps the only exception to this observation is the Gurson [25] model
for porous plastic materials, which includes an evolution Jaw for the kinematically driven
changes in the porosity. However, even this model, which is known to be very accurate for
nearly hydrostatic loading conditions, is inappropriate for Iow-triaxiality conditions because
it neglects the development of anisotropy due to changes in the average shape of the voids.
Similarly, the Taylor model, incorporating only first-order statistical information (i.e., the
crystallographic orientation distribution function), has been found to be rather inaccurate for
relatively large strains because it neglects the effect of additional microstructural features,
such as changes in the average shape of the grains, which is known to be present in many
standard forming operations. In this article, ways of accounting for microstructural changes
involving the shape and orientation of inclusions in particulate composites, or of grains in
granu:ar composites are reviewed following the work of Ponte Castaneda & Zaidman [33],
where the proposal was made to make use of Hashin-Shtrikman-type estimates for the
average strain rate in the inclusions to develop kinematically based evolution Jaws for the
134 P. Ponte Castaneda

average shape of the inclusions in the context of composites with particulate microstructures.
This idea provided a way of generalizing Eshelby's [34] procedure for linearly viscous
composites with dilute concentrations of inclusions to account for interaction effects, at least
approximately. In addition, this idea also allowed the extension to nonlinearly viscous and
perfectly plastic composites by means of a consistent use of the variational procedure of [28,
29] to handle the (nonlinear) homogenization by means of the linear comparison composite.
Application of these models and their generalizations [35, 36] to account for distribution and
orientation effects have been found to be in agreement with certain experimental and
numerical observations, such as the prediction of void collapse strains during uniaxial
compression and overall material stability during uniaxial tension of porous perfectly plastic
materials, thus improving on the corresponding predictions of Gurson's model accounting
only for the evolution of the porosity in these materials. Finally, it is noted that an alternative
approach- one that has the advantage of reducing to the Gurson model for high triaxiality,
but which is more restrictive in use, dealing exclusively with porous materials and allowing
only the development of transversely isotropic symmetries -has been proposed recently by
Gologanu, Leblond and Devaux [37, 38].

2. NONLINEAR COMPOSITES AND EFFECTIVE BEHAVIOR

2.1 Nonlinear constitutive behavior

We are interested in materials with local constitutive behavior characterized by a
potential, or energy-density function w(x,E), such that the stress 0' and the infinitesimal
strain £, at point x in the specimen Q, are related by

(2.1)

or, alternatively, by

£=
au
Jcr(x,cr), (2.2)

where u is the complementary energy function of w. In this work, it will be assumed that the
potential w is a.sufficiently smooth, convex function of £, such that its Legendre-Fenchel
transform is given by (see, for example, [39, 40])

w*(x,cr) = max{£·0'- w(x,e)}. (2.3)

£
 Nonlinear Composites and Microstructure Evolution 135

Note that the unique maximizing condition in this expression is precisely the stress-strain
relation (2.1), so that u is in fact the Legendre transform of w, or u = w·. The
complementary potential u is also convex (in a), and therefore

w(x,E) =max{E· a- u(x,a)}, (2.4)

(J

which leads, in tum, to the maximizing condition (2.2).

This constitutive model is commonly used to represent a number of nonlinear
mechanical phenomena, including the time-independent plastic deformation, as well as the
time-dependent viscous deformation of metals. For isotropic materials, with elastic
compressibility, sufficiently general forms of wand u are given by
w(x,E) =~1\(x)e! + lf>(x,e.). (2.5)
2
and

u(x,a) = _.!__() cr;, + 'l'(x,cr.). (2.6)

21\ X
where 1\ denotes the bulk modulus of the material, em= !trE and crm =!tr<J are the
hydrostatic strain and stress, e, = ~1e · e and cr, =~-!s · s are the equivalent strain and
stress (here e = E- eml and s = a- crml are the strain and stress deviators and I the
identity), and where if> and ljl are convex functions of e, and cr,, respectively, such that
ljl =l/>*. In addition, it is assumed that lf>(x,e.) ~ 0, ljl(x,cr.) ~ 0 "\/ x,e,,cr, and that
lf>(x,O) = ljl(x,O) = 0 "\/ x. (Note that dependence on the third invariants has been neglected,
for simplicity.) This framework is consistent with the so-called Jz-deformation theory of
plasticity [41 ]. In particular, the choice

'I'{ Xcr ) =_1_cr

' e6J.L(x)
2+ cr,.(x) [~]n+I
n + 1 cr.r(x)
e '
(2.7)

where J.l is the shear modulus, cr,. is a reference stress and n (0 < n < oo) is the power
exponent, corresponds to the Ramberg-Osgood model.
Another example is provided by the nonlinearly viscous material, which is a good
model for the high-temperature creeping behavior of metals. In this case, the deformations
are finite and a and E should be identified with the Cauchy stress and Eulerian strain rate (or
rate of deformation tensor), respectively. Then, neglecting elastic effects and assuming
incompressibility, relation (2.1) applies with w given by (2.5) taking 1\ ~ oo, so that if>
 136 P. Ponte Castaneda

characterizes the local dissipation function of the material. One choice for c/J that is commonly
used to characterize creep is the power-law form
3n I+_!_
c/J(x,e,) = -ry(x)e, n' (2.8)
n+l
where 1J is the nonlinear viscosity of the material and n is the creep exponent. (The
corresponding expression for u is given by (2.6) and (2.7) with J.l ~ oo and aY = 31].)
Finally, a special case of interest here is the limit of the power-law viscous material as
n ~ oo, which defines the Mises rigid/ideally plastic material. For this material, it is useful to
introduce a yield function
<I>(x,a,) =a,- ay(x) , (2.9)
such that u(x,o) =0 if <I>(x,a,) ~ 0 and u(x,o) =oo if <I>(x,a,) > 0. Then, interpreting the
derivatives in (2.2) in a generalized sense (as the subdifferentials of convex analysis; see [39,
40]) leads to the conclusion that
.a<J>
E= A a(J (x,CJ), (2.10)

where i =0 if <I>(x,a.) < 0 and i ~ 0 if <I>(x,a,) =0. (This is the normality condition.)
2.2 Effective constitutive behavior
As discussed in the Introduction, this work is concerned with the estimation of the
effective, or homogenized behavior of composites with nonlinear constitutive properties. By
a "composite" is meant a heterogeneous material with variations in its local properties
occurring at a length scale that is small compared to size of the specimen and to the scale of
variation of the applied loads. The distribution of inhomogeneities in the specimen -which
may be deterministic, as in a periodic composite, or random - is usually referred to as the
"microstructure." Assuming that the microstructure is uniform through the specimen, in the
deterministic context, or statistically uniform in the random context, it makes sense to define
the average stress and strain (or strain rate) fields, and to attempt to characterize the relation
between these average stress and average strain (rate) quantities. This homogenized stress-
strain relation defines the effective constitutive behavior of the composite. It corresponds
precisely to what the experimentalist observes in the laboratory when measuring the stress-
strain relation for a bar of steel, which is in fact a composite made of single crystal grains of
the metal (i.e., a polycrystal).
It is possible that the microstructure is not uniform through the specimen, but that it
varies smoothly at an intermediate length scale that is still large relative to the size of the
 Nonlinear Composites and Microstructure Evolution 137

inhomogeneities. This may be the case, for example, in an extrusion process where the
porosity ahead and behind the die may be different. In this case, it still makes sense to define
the effective constitutive relation for the composite (a porous material), provided that the
averages of the stress and strain fields are evaluated over the intermediate length scale. A
mathematical definition of the effective behavior of nonlinear composites with the class of
local constitutive behavior introduced in the previous section is given below. Consistent with
the above comments, the loads are assumed to be applied slowly, so that inertial effects may
be neglected. However, it is noted that when finite deformations are involved, as may be the
case for the nonlinearly viscous and rigid perfectly plastic materials described above, the
evolution of the microstructure in time will need to be taken into account. In this case, it
suffices to carry out the homogenization process with fixed microstructure and to integrate
the resulting constitutive relation in time, supplemented by appropriate evolution equations
for the relevant microstructural variables. For more details on the definition of effective
properties for composites with linear constitutive behavior, the reader is referred to the
monographs by Sanchez-Palencia [1] and Bessoussan et al. [2] in the context of periodic
composites and to the review articles by Willis [9, 10] in the random context.
Consider an N-phase composite occupying the domain Q, with each phase occupying
sub-domains Q(r) (r = 1, ... , N), and let the potentials w(x,E) and u(x,a), from (2.1) and
(2.2), be expressed in terms of the homogeneous phase potentials w<'J(E) and u<'J(<J),· via
n n

w(x,E) = Li'l(x)w<'J(E) and u(x,O') = Li'l(x)u<'l(a), (2.11)

r=l r=l
where
X(r) (X) =1 if X E n,<rJ, or 0 otherwise, (2.12)
is the characteristic function of phase r. These functions define the microstructure.
Then, the strain (or strain-rate) tensor £, which is required to satisfy the relations
I
f.If =-(u .. +u.)
2 1,] ],1
(2.13)

in terms of the displacement (or velocity) field u, is related to the stress a, satisfying the
equilibrium equations
(] ... =0,
1),)
(2.14)
via the constitutive relation (2.1) or (2.2). The commas in equations (2.13) and (2.14) denote
differentiation, and the summation convention has also been used in equation (2.14). The
phases are assumed to be perfectly bonded, so that the displacement (or velocity) is
 138 P. Ponte Castaneda

continuous across the interface boundaries. However, the strains and, therefore, the stresses
may be discontinuous across such boundaries, and hence equation (2.14) must be interpreted
in a weak sense, requiring continuity of the tractions across interface boundaries.
The statement of the problem is completed by the selection of the boundary condition:
U; =l;jxj, X E an, (2.15)
where an denotes the boundary of the composite and i is a given constant tensor. This
affine condition has some useful properties. Thus, letting
E=(E) and a=(o") (2.16)
denote the respective averages of the actual strain and stress fields in the composite, where
the triangular brackets denote volume average over n, e.g.,

(a)= l~l{aij(x)dV, (2.17)

it follows that
(2.18)
and that

G;j = J~a;knkx)dS. (2.19)

This means that'the average strain E depends only on the applied boundary condition (2.15),
and that the average stress a can be "measured" in terms of the boundary tractions only. The
effective behavior of the composite then follows from the relation between E and a.
An alternative characterization of the effective behavior of the composite is obtained
from the principle of minimum potential energy, which is a variational characterization of the
above boundary value problem, described by equations (2.1 ), or (2.2), and (2.13) to (2.15),
under the assumption of convexity of the nonlinear potential w. Thus, an effective potential
W is defined for the composite from the average minimum potential energy
W(E) = min (w(x,E)), (2.20)
£(x)eK
where
(2.21)
is the set of kinematically admissible strains. Note that W depends only on E; in fact, W
can be shown to be a convex function of E (see Appendix A of [42]). Also note that, in
general, W will be an anisotropic function of E. Then, it can be shown (Hill [43]) that
a= ~!(E), (2.22)
Nonlinear Composites and Microstructure Evolution 139

if W happens to be differentiable; otherwise, the derivative should be interpreted as the

subdifferential of convex analysis. Relation (2.22) yields the effective constitutive relation
for the composite in terms of the average variables a and £. Given this connection, the
problem of determining the effective stress-strain relation of the composite reduces to that of
determining the scalar potential W.
A dual formulation of this problem may be obtained by means of the principle of
minimum complementary energy in terms of the stress potential u(x,a), which, as
mentioned in the previous section, is related to the strain potential via the Legendre
transformation. In the context of this article, it is instructive to derive the expression for the
stress potential U directly from the Legendre relation (see [44])
U(a) = m~x{a· £- W(£)}. (2.23)
£

Note that W may be written in the form

W(E)= min (max{<J·£-u(x,o)})
£(X)EK <J

=max{ min (a· e)- (u(x,a))}.

<J £(x)EK

= max {£·(a)-(u(x,a))}
o;V·o=O

where the interchange of the maximum ~d the minimum is allowed (see [40]) by the Saddle
Point theorem on account of the fact that the function is convex in e and concave in CJ. The
last step follows from the fact that the inner minimum (over e) is -oo unless V · <J = 0.
Then, putting this expression for W into (2.23), it is deduced that
U(a) = min (u(x,a)), (2.24)
<J(X)ES
where
S={aiV·a=O inn, and(a)=a} (2.25)
is the set of statically admissible stresses satisfying an average stress "boundary" condition.
(This additional constraint on a was a consequence, once again, of the interchange of the
maximum over£ and the minimum over <J.) It then follows that

__ au<_)
e- aa a· (2.26)

where now a is equal to the prescribed average stress and the average strain £ can be
"measured" in terms of the displacement on the boundary. Note that [; is also convex.
 140 P. Ponte Castaneda

It is common to replace the average stress condition in the definition (2.25) by the
uniform stress condition: an= an on an , where n is the normal to an. This substitution
is appropriate provided that the stress potential u is strictly convex. However, it fails for
rigid/perfectly plastic materials, when the potential uloses strict convexity (see [45]).
The specialization of the above potentials for two limiting cases is worthy of separate
mention. For linear elastic (or linearly viscous) composites, the effective strain and stress
potentials may be written in terms of the effective modulus (viscosity) tensor L and
compliance tensor M =L-1 , respectively, so that
W(£)=~£·(L£) and U(a)=~a·(Ma). (2.27)

On the other hand, for rigid/perfectly plastic behavior, the effective dissipation function W
may be obtained directly from expressions (2.20) and (2.21), with the local dissipation
function w given by (2.5) with (2.8) and K: ~ oo, n ~ oo. In the description of the dual
potential U in (2.24), it is convenient to introduce [ 17, 19] an effective yield function ci>, or
extremal yield surface
ci>(a) =o, (2.28)
such that U(a) = 0 if ci>(a) ~ 0, or oo otherwise. In terms of ci> one has [ 17, 19] that

£=A~: (a), (2.29)

where A is a non-negative parameter that is to be determined from the consistency condition

ci> =0. Note that the convexity of u, and therefore of U, implies convexity of ci> and that
relation (2.29) is a statement of the fact that normality at the local level (2.1 0) implies
normality at the global level. Also, the effective dissipation function W is homogeneous of
degree 1 and is related to ci> by
W(£) = Q).ax {a·£} . (2.30)
a, <l>(a)so
At this point, a few remarks are in order. If the microstructure of the composite is
deterministic as in a periodic composite, W and U can be determined exactly in terms of the
solution of a nonlinear boundary problem on a unit cell with periodic boundary conditions
[16-18]. On the other hand, if the microstructure of the composite is random, usually only
partial statistical information is available in the form of the volume fractions of the phases,
or, less frequently, some higher-order information such as overall isotropy for the
composite. It is then not possible to determine the effective properties of any given composite
exactly, and it is useful to reinterpret W and U as the effective potentials of appropriate
Nonlinear Composites and Microstructure Evolution 141

classes of composites with prescribed statistics of the microstructure. In this event, it makes
sense to attempt to delimit the effective behavior of composites by specifying bounds for W
and (; in terms of the prescribed microstructural information. In some cases, however, as
when the bounds are too far apart to be useful, it may be possible to try to identify
approximations that in some sense best characterize the properties of certain families of
composites. In the context of composites with linear constitutive behavior much progress has
been made along these lines over the past 50 years, including rigorous bounds, such as the
Voigt-Reuss-Hill, Hashin-Shtrikman and Beran bounds, and approximations such as the
self-consistent estimates. In the following section, a brief account of these various estimates
is presented. (For a more comprehensive account, see [9].) In later sections, use will be
made of these estimates for linear composites to generate, by means of appropriate variational
principles, corresponding estimates for nonlinear composites.

3. SOME SELECTED RESULTS FOR LINEAR COMPOSITES

3.1 The Voigt and Reuss bounds

In the context of linear elastic composites, Hill [3] and Paul [4] made the observation
that the choice of a uniform strain trial field in the variational statement (2.20) for W lead to
the rigorous upper bound
-(-) :5-E·LE,
WE L -- (3.la)
2
which implies that the Voigt estimate

(3.lb)
r=l

where the c<rl = (i'l) are the volume fractions of the phases, is a rigorous upper bound for
L (in the sense of (3.la)).
Similarly, the choice of a uniform stress field in the variational statement (2.23) for [;
leads to the upper bound
u-(-) l_ - -
0' ~-O'·MO', (3.2a)

r
2
Making use of the Legendre relation (2.26), this relation implies that the Reuss estimate

[tc'''(L'''f' (3.2b)

is a rigorous lower bound for L.

 142 P. Ponte Castaneda

It is noted that the Voigt-Reuss bounds are exact to first order in the contrast for weakly
inhomogeneous composites. For this reason, sometimes they are referred to as first-order
bounds. In addition, it is noted that the Voigt-Reuss bounds do not incorporate any additional
information beyond the volume fractions of the phases. Therefore, they apply to the class of
all microstructures with prescribed volume fractions, including, for example, laminated
microstructures. These bounds are known to be sharp in the sense that they can be attained
by specific microstructures loaded in certain special directions, but are not optimal because
the bounds are cannot be attained in general (see [46]). Next, a different procedure for
determining bounds is discussed, one which will bring in statistical information of higher
order than the volume fractions, namely of order two.

3.2 The Hashin-Shtrikman variational principles [27, 36]

Hashin and Shtrikman [5-7] introduced variational procedures to estimate the effective
modulus tensor L of linear-elastic composites with statistically isotropic microstructures.
Through the well-known analogy between the governing equations for a linear-elastic solid
and a linearly viscous fluid, it is clear that the Hashin-Shtrikman (HS) procedure may also be
used to estimate the instantaneous effective viscosity tensor of a two-phase Newtonian fluid,
provided that surface tension, buoyancy and dynamical effects are neglected. In the context
of linear elastic media, the HS variational representation makes use of polarization fields
relative to a homogeneous reference material with modulus tensor L<OJ to obtain upper and
lower bounds for L, when L<OJ is chosen to be equal to the "maximum" and "minimum"
phase modulus tensors L(r), respectively. A sufficiently general version of the HS variational
procedure is given below, following Willis [9], Talbot & Willis [27] and Kailasam & Ponte
Castaneda [36]. This version of the HS variational procedure also works for nonlinear
composites. However, as discussed in Section 2, in this work we will make use of an
alternative procedure requiring the HS estimates for L to obtain corresponding estimates for
the effective potentials W of nonlinear composites.
Define w< to be the potential function of a reference linear material with uniform
0>

modulus (viscosity) tensor L<0>, and assume that L<OJ ~max{L<rl}, in the sense that
r

£ · (L<r>- L(O>)e ~ 0 for all £::;:. 0 (r = 1, ... , N), so that the "difference" potential w- w< 0 J is
a concave function. The Legendre-Fenchel transform of this difference potential is defined as
( w- w< 0 >).(x, 't) = mjn{'t· E- [w(x,E)- w< 0 l(E) ]}. (3.3)
Nonlinear Composites and Microstructure Evolution 143

It then follows from the fact that the difference potential is concave that
w(x,E)- w< 0 l(E) = m}n{'t· E- ( w- w< 0 l).(x, 1:)}. (3.4)

which is nothing more than the statement that w- w<OJ =(w- w<OJ )••. Therefore, substituting
(3.4) for w in expression (2.20) for W, and interchanging the order of the minima over E
and 't, one arrives at
W(£) = m}n{~ip(w< 0 l(E) + 't· E)- ((w- w< 0 l).(x, 1:))}. (3.5)

Taking the "polarization field" 1: in the "inner" minimum problem as given, it follows that
the displacement field u, related to the minimizing E by (2.13), solves the boundary value
problem:
~ -
E,ijkluk.lj - -'rij.j
.
lll Q, and
_-
U; -l;jXj On an. (3.6)
Then, making use of the Green's function G<OJ associated with the system (3.4) in the
domain n, it is possible to obtain the following expressions for the strain (strain rate) and
rotation (spin) tensors:
(3.7)
where E and iil are the average strain and rotation tensors associated with I, and where :r<0J
and N°l are linear integral operators
r<oJ't = fnr< 0l(x,x')[ 1:(x')- 't]dV' and /'( 0l't = 1/'( 0 l(x,x')[ 't(x')- 't]dV', (3.8)

whose kernels are related to G<oJ via

a2G(OJ
r<o) _ __ d A(o)
azc<Ol
ik_
ijkl - ax .ax'
1 I (ij).(kl)
an ijkl = ax .ax'
1
ik

I [ij].(kl)
' (3.9)

respectively. In these relations, 't denotes the mean value of 1: over n, and the parentheses
and square brackets (enclosing the subscripts) denote symmetric and anti-symmetric parts,
respectively.
Of course, the polarization field in (3.5) is difficult to determine exactly. Because of this,
it is helpful to make the approximation of piecewise constant polarization, namely,
N

't(x) =l:z<'l(x)'t(rl. (3.10)

r=l
Then, the finite-body kernels in (3.8) can be replaced by the corresponding kernel
constructed from the infinite-body Green's function under the no long-range order
hypothesis for the distribution of the phases (see [9]). One important consequence of this
hypothesis is that the kernels in (3.8) become functions of (x- x') only.
 144 P. Ponte Castaneda

Recalling that the c<rJ = (x<'J) denote the volume fractions of the phases, and making
use of the fact that the average of a tensor field T over phase r is given by
(T)(rJ = ((i'J jc<'J)T), it follows from (3.7) and (3.8) that the average over phase r of the
strain and rotation tensors, E<rJ = (E)(rJ and ro<rJ = (ro)<rJ, are given by
N N
£<rJ = £- _1_ ~ E<rsJ't<sJ and ro<rJ = ro- _1_ ~ F<rsJ't<sJ (3 .11)
d~~ d~~ ,
s=l s=l

respectively, where
E(rs) =(tdV' x<r) (X)[ i"J (x')- c<s)] r<OJ (X- x')),

F(rsJ = (tdV' i'l(x)[i"l(x')- c<"J] A<.OJ(x- x')) (3.12)

(r, s = 1,... , N) are tensors which depend only on the microstructure of the composite and on
L<OJ. Furthermore, they are known to be symmetric in the superscripts rands, and are not all
independent, satisfying the relations

L =L
N
E(rs)
N
E(rs) =0 (3.13)
r=l s=l r=l s=l

t. t.t.
It also follows that

IV(£) ~ ,~·d IV'"(£) +'t £- ~ c1' 1<''1 [( H

M.''')' .,< ' 1 J}, (3 .14)
.,<' 1 [ E1" 1<''1

L c<rJ't<rJ . Then, optimizing over the polarizations

2
where L\L(rJ = L(rJ - L<OJ and 1: = 't(rJ
r=l

(r = 1,... , N), one obtains the relations

N

( L(r)- L(0))-1 't(r) + _(1) ~ E(rs)'t(s) = E, (r = l, ... ,N), (3.15)

c' ~
s=l
so that
(3 .16)
Note that the aforementioned symmetry of the geometric tensors E<rsl ensures the
corresponding symmetry of L in (2.27a).
The upper bound implied by (3.16) for L assumes that L<0J ;::>: max{L<'>}. If, instead, it
r

is assumed that L<OJ:::; min{L<'l}, changing the minima to maxima in (3.3) and (3.4), it is
r

found that (3.15) and (3.16) still apply, except that the direction of the inequality must be
changed to give a lower bound for W. However, if neither hypothesis on L<Ol is satisfied,
Nonlinear Composites and Microstructure Evolution 145

the right-hand side of (3.16) still provides a (stationary) variational estimate for W, which is
neither an upper nor a lower bound on W. Regardless of the choice of L(o), the Hashin-
Shtrikman estimates for L are known (see [9]) to be exact to second-order in the contrast for
weakly inhomogeneous composites. For this reason, they are often referred to as second-
order estimates.
The estimate for W provided by (3.16), together with (3.15), is explicit, except for the
geometric parameters E<rs), which need to be determined for specific classes of
microstructures. Similarly, expressions (3.11), with the optimal choices for the 't(r) from
(3.15), provide estimates for the average strain and rotation tensors in the. phases. They
depend on the geometric tensors E(rs) and F<rs). Explicit estimates are given in the following
sections for classes of particulate and granular microstructures with prescribed two-point
correlation functions for the statistics of the distribution of the phases.
Finally, it is noted that, using similar ideas, a corresponding variational principle for
U may be obtained (see [27]). However, the resulting estimates for L
the effective potential
can be shown to be equivalent to those described above from the variational principle for W.
For this reason, further discussion of these dual variational principles is omitted.

3.3 Hashin-Shtrikman bounds for random composites with ellipsoidal symmetry [47]
For composites with random microstructures, it follows by interchanging the order of
integration in expressions (3.12) that E<'·'·l and F<rsl depend on the probability function of
finding simultaneously phase rat x and phase s at x'
p<'·'·l(x,x') = (x<'l(x)xc'l(x')). (3.17)
In fact, since statistical homogeneity has been assumed, so that p<rs)(x,x') = p<'·'l(x- x'),
expressions (3.12) reduce to
E(rs) J
= dV"[p<rs)(x")-c<'V')]r<Ol(x"),

F(rs) = JdV" [P<'·'l(x")- c(rV')] N°l(x"), (3.18)

where the integrals are now taken over all space, because of no long range order. If it is
further assumed that the two-point probability function has "ellipsoidal symmetry" so that
p<rs)(x) = p<rs)(lzxl),
for some constant tensor Z. Note that the special case when Z =I, so that the p<rs) depends
only on jxj corresponds to statistical isotropy. Thus, ellipsoidal symmetry may be thought of
 146 P. Ponte Castaneda

as the class of geometries that would be obtained by uniformly stretching statistically

isotropic microstructures.
Under this statistical hypothesis, Willis [47] has shown that the tensors E<rsl reduce to
E(rs) = c(r)( 8,, - c(s)) p<Ol, (3.19)

where
p<Ol =_I_ Ju<Ol(l;)iz-1~-3 dS. (3.20)
4JriZI 1!;1= 1
Here u<o> is a tensor with components H;:~/(l;)= B~ 0 >(l;)~ ~~~
,
, where B<OJ the inverse of
I ] (ij),(kl)

the acoustic tensor A <o>. which has components A;~o>(l;) = L~~1~~j~L· Note that p<Ol is related to
the well-known tensor s(O) of Eshelby [34] by p<Ol = s<Ol(L(O)t.
Similarly, it was shown in [36] that the tensors F(rs) reduce to
F(rs) = c(r) ( ors - c(s)) R(O), (3.21)
where the tensor R<o; is like the tensor p<Ol, except that the symmetric part with respect to the
indices i, }, in the expression for the tensor H, must be replaced by the corresponding anti-
symmetric part. Note that R<OJ is related to the well-known tensor n<OJ of Eshelby [34] by
R <o> = n(O> (L(Ol f' .
The above expressions for the E(rs) and F(rs) allow further simplification of the
expressions (3.11) for e<r> and ro<'> to obtain
(3.22)

r{t c'''[I+ P"'(L'''- L'"')r} '

where

A''' = [I+ P'"'(L'''- L"') (3.23)

and
(3.24)
In addition, one also ~as that

(3.25)
r=l

As mentioned previously, the choices L<OJ ~ max {L<'l} and L<OJ :-: ; min{ L<'>} lead to rigorous
r r

upper and lower bounds for L, respectively. It is emphasized that such bounds, although
appearing to depend only on volume fractions, incorporate two-point statistics, and therefore
have the potential of being sharper than the Voigt-Reuss bounds. For two-phase composites
Nonlinear Composites and Microstructure Evolution 147

(N = 2) with L0 > ~ L< 2>, it is known that the best lower and upper bounds are obtained by
choosing IJ0> = L< 1> and L< 2>, respectively. When the phases are isotropic and the
microstructure is statistically isotropic with Z =I, the celebrated bounds of Hashin &
Shtrikman [7] are recovered. In this case, making use of sequentially laminated
microstructures, Francfort & Murat [48] have shown that the HS bounds are optimal,
provided that the phases are well-ordered ((J1< 2 >- J1°>)(7C< 2 >-7C 0 >) ~ o). (A similar result was
shown earlier in [7, 56] for the bulk modulus making use of a coated sphere microstructure.)
For more than two phases, the HS bounds are known to be not optimal, in general, although
they can be optimal in some cases (see [49] and references therein).
Finally, it is important to mention (Willis [47]) that if L<o> is chosen equal to L, as given
by (3.25), an implicit equation is obtained for L<o>, which when solved leads to an estimate
for L. Note that, from (3.15), it follows that the average polarization associated with this
prescription is identically zero. Also, it is of interest that the above prescription reproduces
the self-consistent estimates of Budiansky [15] and Hill [14] for two-phase composites,
when the shape of the two-point probability function is chosen to be the same as the shape of
the inclusions in [14-15]. Similarly, the above prescription also reproduces the self-
consistent estimates of Hershey [ 12] and Kroner [ 13] for polycrystalline aggregates, without
the need for making the explicit assumption that the grains are ellipsoidal in shape. Thus, the
variational estimates (3.25) for L also provide a generalization of the classical self-consistent
estimates for random composites with prescribed two-point statistics.

3.4 Hashin-Shtrikman estimates for composites with particulate microstructures [50-51]

Consider an N-phase composite medium with particulate microstructure consisting of a
random distribution of aligned ellipsoidal inclusions of phases r (r =2, ... , N) in a matrix of
phase I, as shown in Fig. I. The inclusions of each phase, which are identical in shape and
orientation, are characterized by reference ellipsoids n<r> ={x : iz<r>xi2 ~ 1}, for some tensors
z<r> with principal values I I a<r>, 1I b<r>, I I c<r> defining two aspect ratios w}'> = a<r> I c<r>
and wi'> = b<r> I c<r>, and principal directions defining a coordinate frame x' whose
orientation relative to the fixed laboratory frame x is determined by the Euler angles ¢<rl,
lfl<r>, e<r>. Consistent with prior hypotheses, it will be assumed that the distribution of the
centers of the inclusions is statistically homogeneous, in such a way that the volume fractions
of the phases take on fixed values c<r>. In addition, the distribution of the inclusions will be
 148 P. Ponte Castaneda

assumed to exhibit "ellipsoidal symmetry", that is, it will be assumed that the joint probability
functions p<rs) for finding inclusions centered at x and y depend only on z = y- x through
the combinations iz<rs)zl, for some tensors z<rs), symmetric in the superscripts, defining

distributional ellipsoids n<rs) = {X: lz(rs)xl 2~I}. The shape of these ellipsoids is

characterized by aspect ratios w?") and wirs), and their orientation is characterized by frame
x"' oriented relative to the fixed laboratory frame X by Euler angles 1/>(rs)' lfl(rs)' e<rs).

(a) (b)

Fig. I. Particulate microstructures. (a) Reference inclusion of type r occupying reference domain n'''. (b) The
inclusions of type rare distributed randomly in n with centers at points x~' (a = I, ... , n''' ).

The application of the Hashin-Shtrikman variational procedures for the above class of
particulate microstructures was considered by Willis [50] and Ponte Castaneda & Willis [51].
In this case, it is reasonable to choose L(O) equal to the modulus tensor of the matrix phase
L0 ), so that the polarization field in the matrix -r0 ) vanishes exactly. With this choice for
L(o), the only approximation in expression (3.25) for L is that associated with the choice of a
uniform polarization in the inclusions. However, it is known from the work of Eshelby [34]
that the stress and strain fields in the inclusions of a linear elastic composite with a dilute
concentration of aligned ellipsoidal inclusions is uniform. This means that, at least for the
limiting case of dilute concentrations of inclusions, the estimate (3.25) is exact. For larger
concentrations of inclusions, the fields within the inclusions will no longer be uniform, but
Nonlinear Composites and Microstructure Evolution 149

the estimate (3.25) is probably still good, provided that the inclusion concentration is not so
large that the inclusions start coming into contact with each other. Thus, the HS estimate
(3.25) for L is exact in the dilute limit , but may also be expected to be accurate in the low to
moderate range of volume fraction for the inclusions.
Under the statistical assumptions of homogeneity and ellipsoidal symmetry for the
distribution of the aligned ellipsoidal inclusions, supplemented by the hypotheses of no
overlap of the inclusions and no long-range order in their distribution, which allows the
replacement of G<o> by its infinite-body counterpart, Ponte Castaneda and Willis [51] (see
also Bornert et al. [52] for closely related results) have shown that the microstructural tensor
E(rs> and F(rs> in (3.12) simplify to:
E<r.•> = c<r>(8rs p(r>
1
- c(.•>p<rs>)
d and F<r.•> = C <r>(8rs R<r>
1
- c<"'>R<rs>)
d (3.26)
where the tensors Pir>, py-·> and R~r>, R~-'> are given by expressions similar to (3.20) with
L<O> = L 0 >, with appropriate symmetries and choices for the tensors Z. They depend on the
shape and orientation of the inclusions and distribution functions, through the variables wlr>,
w(r) n,(r) lf/(r) e<r> and w<rs) w<r.<) n,(r.•) lfl(rs) e<rs) respectively
2 ' 'f' ' ' ' I ' 2 ' 'f' ' ' ' ·
For a two-phase composite with aligned ellipsoidal inclusions of phase 2 in a matrix
of phase 1, the microstructural variables reduce to two aspect ratios wl;>, w~n and three
angles l/Ju>, lflu>, e<n for the inclusions, and two aspect ratios wldl, w~d> and three angles
l/J(d)' lfl(d)' e<d> for the distribution function. Then, L can be written in the form
L = L< 0 + c< 2>(L< 2>- L0 >)A <2>, (3.27)
in terms of the strain-rate concentration tensor A <2>, such that

(3.28)
Similarly, the spin-concentration tensor em is given by
c<z> =(R~I}- c(2}R~>)[(L( 2 )- IY>( + (P?)- c(2}Pj 1>)r. (3.29)

Note that the no-overlap hypothesis places restrictions on the maximum volume fractions of
the inclusions, which depend on the aspect ratios and orientations of the ellipsoidal inclusions
and distribution ellipsoids. To visualize (see Fig. 2) this class of microstructures it is helpful
to think of a generalization of the Hashin composite spheres, with ellipsoidal inclusions of
some given shape and orientation being surrounded by "shells" of the matrix material with
some other given shape and orientation. (Note, however, that the above estimates do not
 150 P. Ponte Castaneda

provide exact answer for such microstructures, as in the Hashin construction for the bulk
modulus of an isotropic two-phase composite.)

wd< w wd= 1

(a) (b) (c)

Fig. 2. Three different types of distributions containing spheroidal particles with aspect ratio w;" = w;" = w.
(a) Aligned particles d.istributed with spheroidal symmetry with aspect ratio wd < w. (b) Aligned particles
distributed with spheroidal symmetry with aspect ratio wd > w. (c) Randomly oriented particles with aspect
ratio w distributed with spherical symmetry ( wd = I).

The variational estimates (3.27) for two-phase particulate composites will be upper and
lower bounds for L, depending on whether L0 > > L(2), or L< 2> > L0 >, respectively.
However, the corresponding expressions for N-phase composites, obtained from (3 .25) and
(3.26) will only be a bound if the matrix phase (I) happens to be the one with the largest, or
smallest modulus tensor. Otherwise, such an expression for L will only be a variational
estimate, but one which is expected to be fairly accurate for low to intermediate volume
fractions of inclusions, as discussed earlier.
The expressions (3.27), (3.28) and (3.29) for L, A <Z> and em can be seen to reduce,
in the limit as c< -t 0, to the corresponding expressions of Eshelby for composites with
2>

dilute concentrations of inclusions. (Note that the distributional effects, as characterized by

P5'> and R~>, disappear in this limit, as expected.) In addition, when the inclusions are
aligned with the distributional ellipsoid and have the same aspect ratios, the above expression
for L, A(2) and C(2) can be seen to reduce to those of Section 3.2.1, as first shown by
Willis [50]. This last observation is in agreement with the earlier comment that the upper and
lower HS bounds for general ellipsoidal microstructures are "attained" by composites with
particulate microstructures.
Nonlinear Composites and Microstructure Evolution 151

3.5 Self-consistent estimates for random composites with granular microstructures [47]
Provided that the shape and orientation of the inclusions and distribution ellipsoids are
kept equal in the estimates of Section 3.4, it is possible to consider the limit as the volume
fraction of the matrix phase vanishes. The resulting microstructure is granular in character
being composed of aligned ellipsoidal grains of identical shape and orientation, but of
varying size so as to fill space. In this case, there is no advantage in choosing the modulus
tensor L<OJ of the homogenous reference material to be equal to that of any of the phases,
since all the phases are now inclusions, and there is no reason to believe that the strain in any
of the inclusions will be uniform, in general. Then, perhaps the only sensible prescription
[47] is to choose L< 0 > equal to L in expression (3.25) with (3.23) and (3.26), so that the
average polarization in the composite vanishes identically. The resulting estimate for L is of
course not a bound, but has the advantage that it treats all the phases symmetrically, which
may be appropriate for certain classes of composites, such as polycrystals. Also, perhaps not
surprisingly, the resulting estimate is found to agree (once the phase indices are relabeled)
with the self-consistent estimates for the general microstructures with prescribed two-point
statistics. This suggest that the self-consistent estimates may be the most appropriate for
granular types of microstructures, in much the same way as the Hashin-Shtrikman estimates
of Section 3.3 are most appropriate for composites with particulate microstructures.

3.5 Third-order bounds for random composites [8, 55]

Beran [8] introduced a procedure to obtain variational estimates depending on up to
three-point statistics, starting from the minimum energy representations for the effective
potentials. These estimates were simplified and improved in various ways by Kroner [53],
Milton [54] and Willis [9]. However, for N-phase composites, these estimates are fairly
complicated involving microstructural tensors analogous to E<rs> and F<'·'>, but which depend
on the three-point probability functions p<rsr> (see, for example, [9]). Note that such bounds
are know to be exact to third-order in the contrast for weakly inhomogenous composites.
Here, for simplicity, only the simplification of Milton & Phan-Thien [55] for two-
phase composites with statistically isotropic microstructures will be given. Denoting the bulk
and shear moduli of the phases by 7C<rl and J.l<rJ, respectively, the effective moduli of the

[L (r)( [L (r)(
bulk and shear moduli of the composite i( and [l may be written in the forms

-1-
;;.
_ 2
c - 1+ -
,..(r)
1
,;..
)-Il-l --
,;.. '
1-
-
i/
_ 2
c -1+ -
1/(r)
1
II
)-Il-l 1
A' (3.30)
~ ~~ ~ ~ ~ r ~ r r /1
 152 P. Ponte Castaneda

where K: and [L are functions of the volume fractions c(rl, the phase moduli IC(r) and Jl(r)
and of some third-order microstructural parameters 7] 0> and s(l) lying in the interval [0, 1].
For example, the third-order upper and lower bounds on K: are obtained by taking K: =
-1
4 2 4 2 (r)
3Ls(r)Jl(r) and 3 L'(r)
[ )
, respectively. (Here, 7]( 2>=1-7] 0 > and s( 2l=1-s0 >.)
r=l r=l J1
Similarly, upper and lower bounds on ji are obtained by taking [L = 8/6 and 3/6, where 8
and 3 are given by rather complicated expressions (see eqs. (48) and (49) of [55]).
Assuming for definiteness that IC 0 >< IC< 2>and J1°> < J1< 2>, it is interesting to remark that
the above third-order upper and lower bounds for K: and ji reduce exactly to the
corresponding Hashin-Shtrikman upper and lower bounds in the limits as 7] 0 > and s0>

approach 0 and l, respectively. In particular, this means that the HS estimates are exact for
statistically isotropic microstructures with 7] 0> and s 0> either equal to 0 or 1. Examples of
such microstructures are the Hashin composite-spheres [56] and sequentially laminated
microstructures of [48]. Note that such microstructures are particulate in character.

4. VARIATIONAL METHODS FOR NONLINEAR COMPOSITES

Improved bounds for nonlinear composites were first obtained by Willis [26] for
nonlinear dielectrics and Ponte Castaneda & Willis [42] for nonlinearly viscous composites
using an extension of the Hashin-Shtrikman variational principles for nonlinear materials due
to Willis [10] and Talbot & Willis [27]. The basic result is embodied in (3.5), where the
homogeneous reference material is still chosen to be linear, so that advantage can still be
taken of the representations (3.7) in terms of Green's functions. As mentioned earlier, an
alternative approach will be reviewed here - one which makes use of new variational
principles allowing the representation of the effective potential of a nonlinear composite in
terms of the effective potential of an appropriately chosen linear comparison composite.
This approach has the advantage that it not only permits the determination of nonlinear
Hashin-Shtrikman bounds, but also allows the use of any other bound and estimate for linear
composites to generate corresponding bounds and estimates for nonlinear composites.

4.1 Variational principles for nonlinear composites with isotropic phases [28, 29]
4.1.1 Minimum potential energy formulation. Ponte Castaneda [28, 29] proposed an
alternative variational representation for the effective potential of a nonlinear composite in
 Nonlinear Composites and Microstructure Evolution 153

terms of the potential of an appropriately chosen "linear comparison composite". The new
variational principle centers around the change of variables p = e;,
defining a function f,
such that the function l/J in the expression (2.5) for the potential w may be written as

l/J(x,e,) = f(x,p). (4.1)

By the properties of l/J, f is a nonnegative function with same dependence on x as l/J, and
such that f(x,O) = 0. Furthermore, f is continuous, but not necessarily convex (in p). In
fact, because of our interest in applications in plasticity, it will be assumed in this section that
w has weaker-than-quadratic growth in E, or, equivalently, that/ has weaker-than-linear
growth in p, which is consistent with concavity of f. Thus, for example, for an
incompressible power-law material, w- e!+IIn (I ~ n < oo ), and f- p<n+I)J 2n is a concave
function, even though w is itself convex.
Next, defining the concave polar function of f by (see Appendix I of [29])

f.(x,q)=min{pq- f(x,p)}, (4.2)

p'i?.O
so that
J..(x,p) = min{pq- f.(x,q)}, (4.3)
q~O

it is known (see [39, 40]) that, in general,

f ~f. •.
This last relation has the geometric interpretation that f.. is the upper concave envelope off,
that is the lowest function lying above f which is concave. It follows that f =f.., iff is
concave, or using (4.3) that
f(x,p) = min{pq- f.(x,q)}, iff is concave. (4.4)
q~O

Then, introducing a linear comparison composite, with the same bulk modulus 1C as the
nonlinear potential (2.5), and with shear modulus Jlo, such that its quadratic potential is
given by
(4.5)

and identifying q with (3/2)J1 0 , it follows, from (4.1) and (4.4), that the potential of the
nonlinear composite w, as defined by (2.5), may be given the representation:
w(X,E) = 2.7C(x)e! + min{l/Jo(x,e,) + v(X,J10 )} =min{ W 0 (X,E) + v(X,J10 ) } , (4.6)
2 ~~ ~~

where v(X,J10 )=-J.(x,-fJ1.), or, from (4.2),

v(X,J10 ) =max{ l/J(x,e,) -l/Jo(x,e,)} =max{ w(x,E)- W 0 (X,E)}. (4.7)
~ E
154 P. Ponte Castaneda

It also follows that the local stress-strain relation (2.1) for the nonlinear composite may now
be written
<J(x) =3K(x)cml + 2jlo(x)e,
where P-o is the optimal value of the shear modulus flo of the linear comparison composite,
as determined by the minimizing condition in (4.6), namely by,
3 2 _ dv(x,jlo)
-£,-- . (4.8a)
2 dflo
Note that P-o is a function of£ - in fact, P-o is the local "secant modulus" of the nonlinear
composite [41]. This can be easily checked by inverting the previous relation, or, using
Legendre duality, by considering the optimality condition in (4.7), which leads to
) ae
flo~ (X) =-3
1 dljJ (
-:;--- X,£, =-3 ' (4.8b)
£, a£, £,
where the factor of 3 arises because of the different definitions of a, and £,.

Next, by substituting the expression (4.6) for w into the expression (2.20) for W, and
interchanging the order of the minima over£ and flo, it foUows that
W(£)= min {W,(£)+(v(X,fl 0 ))},
(4.9)
.U,(x);,Q

where
W,(£) = min
E(x)EK
(wo(x,e)) (4.1 0)

is the effective potential associated with the linear comparison composite with local potential
W 0 , as given by (4.5). It is emphasized that, under the hypothesis of concave f, which is

consistent with the constitutive models of Section 2.1, the new variational representation
(4.9) and the classical representation (2.20), which is based on the principle of minimum
potential energy, are completely equivalent. Thus, under these circumstances, the
variational statement (4.9) can be given the following interpretation: the effective potential W
of the nonlinear heterogeneous solid, as defined by (2.20), can be alternatively determined
from the effective energy of a linear heterogeneous comparison solid, with bulk modulus
K(x) and with shear modulus fl 0 (x), whose precise variation is determined by the variational
principle (4.9). Thus, letting Lo(flo) denote the effective modulus tensor associated with the
effective potential of the linear comparison composite W,, the optimal choice of the
comparison moduli flo, denoted P-o, is given by
}_E· [dLo(P-o) El =-(Jv(x,jlo))·
2 dflo dflo
 Nonlinear Composites and Microstructure Evolution 155

It then follows, from (2.22), using the previous result, that

cr =L"(.Uo)£, (4.11)
where Lo(.Uo) can be interpreted as the effective secant modulus of the nonlinear composite.
It is well worth emphasizing that whereas the bulk modulus of the linear comparison
composite is constant in each phase (the same as for the nonlinear material), the
corresponding shear modulus will be nonuniform within each phase. Because of this, the
variational principle (4.9) is, in general, at least as difficult to implement as the original
classical variational principle (2.20). However, the variational principle (4.9) has the
advantage that it allows useful approximations that are not accessible directly from the
classical variational principle. One possible approximation in (4.9) is to replace the minimum
over the set of arbitrarily variable, nonnegative shear moduli by the smaller set of piecewise
constant, nonnegative moduli. The application of this approximation in the determination of
bounds and other estimates for the effective behavior of nonlinear composites will be
considered in a later section. This type of approximation was introduced directly by Ponte
Castaneda [28].
Finally, it is also worth noting that the local strain field E in the nonlinear composite
may be obtained by evaluating the strain field in the linear comparison composite at the
optimal distribution of shear moduli J.l.o(x), i.e., those solving the variational statement (4.9).

4.1.2 Minimum complementary energy formulation. As shown in [29], it is possible to start

from the complementary energy representation (2.24) for (; to obtain a corresponding dual
version of the linear comparison composite representation. This requires the change of
variables p =a;, defining a function g, such that the function 1f1 in the expression (2.6) for
the potential u may be written as
lfl( x,a.) = g(x, p).
Then, assuming that the function g is convex (which is consistent with the constitutive
hypotheses of Section 2.1 ), nonnegative and such that g(x, 0) = 0, it follows similarly that

u(x,O') = ~()a!+ max{ lflo(x,a.)- v(X,J.l.0 )} = max{u0 (X,O')- v(x,J..L,)}, (4.12)

21( X ll.~o Jlo~O

where
(4.13)
156 P. Ponte Castaneda

and
v(x,JLJ = max{V'o(x,o-.) -ljl(x,o-.)} =max{ u0 (X,<J)- u(x,a)}. (4.14)
~ a
Note that ito, the maximizer in (4.12), is once again the secant modulus of the nonlinear
composite, as can be shown by a procedure completely analogous to that of Section 4.1.1.
Then, making use of expression (4.12) for u in expression (2.24) for U, and
interchanging the minimum over <J and the maximum over JL0 , which is allowed by the
appropriate Saddle Point theorem, it is concluded that
U(a)= max{Uo(o)-(v(X,JL0 ) ) } .
jl,(x)~O
(4.15)

where
(4.16)

is the effective potential of the linear comparison composite with local potential (4.13).
It is worthwhile emphasizing that the function v appearing in both expression (4.9) for
W and expression (4.15) for U are precisely one and the same function, provided that w
and u are related by the Legendre transformation (2.4). Indeed, starting from (4.14) and
using the fact that the order of maxima can be interchanged, it follows that
v(X,JL0 ) = m:x{ m:x{E · <J- W (X,E)}- u(x,a)}
0

= m:x{m:X{E· <J- u(x,a)}- w)x,E)} (4.17)

= max{w(x,E)- W 0 (X,E)},
£

which is in agreement with (4.7).

Next, a more direct derivation of statement (4.15) for U is given. Thus, making use of
the expression (2.23) for the Legendre transform of W and of expression (4.9) for W, one
has that
U(a) = m:x{£· a- W(£)}

=m_!lX{E·O- min {W,(£)+(v(X,Jl0 ) ) } }

£ l'o(X)~O
(4.18)
= max {m_!lX{E·O- W,(£)}-(v(X,JL0 ))}
l'o(X)2:0 £

= max {uo(a)- (v(x,JLo))}.

jl0 (x)~O
 Nonlinear Composites and Microstructure Evolution 157

where, once again, use has been made of the fact that the order of the maxima can be
interchanged. This means that the two formulations (4.9) and (4.15) are completely
equivalent (i.e., there is no duality gap).
Finally, it is remarked that it was proved in [29] that the convexity of g hypothesis is
implied by the concavity off hypothesis. For this reason, it will be assumed here, unless
otherwise specified, that the potentialw is convex and such that/ is concave. Under this
hypothesis, both the exact variational representations (4.9) for W and (4.15) for U are
exactly equivalent to each other and to the classical representations of Section 2.2.

4.2 A generalization for composites which violate the concavity of f hypothesis [57]
It is clear that the equality in the variational statement (4.9) will fail to be met if the
concavity off hypothesis is relaxed, so that f ~ J.. and the equality (=) in (4.6) must be
replaced by an inequality (~). Similarly, the equality in (4.15) will also fail to be satisfied if
the convexity of g is relaxed, so that g 2:: g**. Although the standard constitutive models in
plasticity are consistent with the concavity of/hypothesis, it is important from a theoretical
point of view to provide a generalization of the above variational principles for the case where
f is not concave, although w will still be required to be convex. This is particularly important
in order to make contact with the work of other authors on the subject.
Recalling the definition (4.5) for the local potential W 0 of a linear comparison
composite, consider the identity
(w (X,E)) =(w(x,E)) + (w (X,E)- w(x,E)),
0 0 (4.19)
from which it follows that
min (w0 (X,E)) 2:: min (w(x,E)) + min (w0 (X,E)- w(x,E))
e(x)eK e(x)eK e(x)eK
(4.20)
2:: min(w(x,E))+ min((wo-w)**(x,E)).
e(x)eK £(x)eK

Then, recalling the definitions (2.20) and (4.1 0) for the effective potentials of the nonlinear
and linear comparison composites, respectively, one arrives at
WC£) ~ min
(x)<:O
)l0
{W,(e)- e(x)eK
min ((wo-w)** (x,E))}. (4.21)

This variational statement for the nonlinear potential Winvolves the potential W, of a
linear comparison composite, just like the variational statement (4.9). However, it also
involves the computation of the effective potential associated with the nonlinear convex
potential (w(J - w r' which is at least as difficult to determine as the original nonlinear
 158 P. Ponte Castaneda

effective potential W. As it will be seen in the next section (see also [57, 58]), the variational
statement (4.21) has the advantage over the original formulation (2.20) for W in that it
allows the use of a crude bound for the nonlinear potential (W 0 - w )**, combined with a more
refined bound for the linear comparison potential ~ to yield improved bound for W. This
approach is similar to the so-called "translation method" [59, 60], where a homogeneous,
quadratic, quasi-convex functional is used to obtain improved bounds for linear composites.
However, the idea to make use of a linear heterogeneous comparison material, as in the
variational procedure of Ponte Castaneda [28], is relatively new [57, 58].
The main advantage of the statement (4.21) over (4.9) that it has the capability of
recovering the original nonlinear potential w, when applied to a homogeneous material,
even if the concavity of !hypothesis is violated. To see this note that, for a homogeneous
r
material, the optimality condition in (4.21) reduces to (W 0 - w = 0, or flo = 2cp(x,Ee)j E;.
Note also that this optimality condition is different from the corresponding optimality
condition in (4.6), which lead to expression (4.8b) for the secant modulus.
On the other hand, it is obvious from the fact that

. min (( W 0
E(x)eK
- w)** (x,E)) ~ min((wo- w)**cx,E)) =min(w"(x,E)- w(x,E)).
E(x) E(x)
(4.22)

that the variational statement (4.21) implies that

W(E) min {W, (E)+ (max (w- W

~ .U,(x);,Q 0 ))} , (4.23)
E

the right-hand-side of which is nothing more than the variational statement (4.9) (recall
(4.5)). Therefore, the variational statement (4.21) has the potential of improving the
estimation of W for materials that violate the concavity of!hypothesis, at the expense of a
more difficult calculation, although it obvious from the equality in (4.9) that (4.21) cannot
improve on the estimate (4.9) for materials satisfying the concavity offhypothesis.
Finally, it is noted that there is a corresponding generalization of the variational
statement (4.15) for materials violating the convexity of g hypothesis, namely,

U(a) ~ max
.U,(x);,Q
{uo(a) + min ((u- uo}** (x,a))},
<J(x)eS
(4.24)

where Uo is the effective potential (4.16) of a linear comparison composite with local
potential U0 , given by (4.13).
 Nonlinear Composites and Microstructure Evolution 159

4.3 Special variational principles for power-law composites [31]

If the potential function w of the nonlinear composite is chosen to be of the power-law

type (2.5) with (2.8), such that the associated function/is concave (recall n ~ 1). Therefore,

the variational statement (4.9) applies exactly, with v, from (4.7), given by
n+l

( ) _ 3(n -1) (.Uo)l-n

V X,,Uo - 2n • (4.25)
2(n+l) (7J)I-n

Next, letting .Uo(x) =ll.uoii.Uo(x), for some appropriate norm ll.uoll. and noting that W0 and v

are homogeneous of degrees 1 and (n + 1)/(1- n) in ,U (X), respectively, it follows from the
0

variational statement (4.9) that

W(£) .min min{ll.uoll W,(£) + ll.uoll~:~ (v(x,,U))} •

=.U,(x)~O (4.26)
l.u,ll
where W,, is the same as W,, in (4.9) with ,U0 (X) replaced by ,il"(x). Then, the maximum
over ll.uolf can be easily evaluated to obtain the result that

W(£) = _l!!:_ .min { [W,(£)]n2:

n + 1.UO(xl~O
1
[nn -+II (v(X,,t1 ))J!ff}·
0 (4.27)

This is precisely the variational principle of Suquet [31], which was originally derived by
means of Holder's inequality. The present derivation, directly from (4.15), shows that the
variational principle of Suquet is strictly a special case of the variational principle of [29].
Corresponding versions of this variational statement for [; and <i> may be obtained similarly,
but they are omitted in the interest of brevity.

4.4 Generalization of the variational principles of Section 4.1 for polycrystals [61]

The prior variational statements were derived under the assumption of isotropy for the
constituent phases. Mathematically speaking, there are various possible anisotropic
extensions of the nonlinear constitutive models introduced in Section 2. Of these various
possibilities, in this section, only one extension is considered for nonlinear composites with
anisotropic phases, which is suitable to characterize creep of polycrystalline materials.
 160 P. Ponte Castaneda

Consider a reference single crystal which is capable of undergoing creep on a set of K

preferred crystallographic slip systems. Each of the systems is characterized by a second-
order tensor Jl(k) (k = l, ... ,K), such that

Jl(k) =±(m(k) ® n(k) + n(k) ® m(k))' (4.28)

where n(k) and m(k) denote respectively the unit vectors normal to the slip plane and along
the slip direction of the kth system. Then, the resolved shear stress on the kth slip system of
the reference crystal is given by
'l"(k) =(J· Jl(k)' (4.29)
where <J is the stress field acting on the crystal.
Following standard practice (see Hutchinson [23]), it is. assumed that the shear strain-
rate in the kth slip system y(k) depends on the applied stress, only through the resolved shear
stress -r(k) on that system, in such a way that
- dcp(k) ( ) (4.30)
rck> -~ -r(k),
(k)
where qJ(k) is a convex function (Rice [62], Mandel [63]), satisfying the condition
qJ(k)(O) =0. In the above relation, it has been assumed implicitly that the slip potential
functions qJ(k) are differentiable; however, even when they are not differentiable, as would be
the case for rigid/perfectly plastic crystals (e.g., cp(kl = 0, if -r(k) ~ -r"; and qJ(k) = oo , if
'l"(k) > 'l"" ), it is still possible to relate y(k) and -r(kJ via the normality condition.
It follows from the definition of Yckl that the overall strain rate £ in the single crystal is
given by the relation

L rck>llck> .
K

e= (4.31)
k=I

Alternatively, this relation may be written

au< c)
£= ()(J (<J), (4.32)
where
K

u(c)(<J) =LqJ(k)(-r(k)) (4.33)

[l
k=I

is the stress potential of the crystal. The pure-power law choice

n+l
_ Yo(-ro)(k) 'l"
(4.34)
'P<k>( -r)- (n + 1) (-r")(kJ '
Nonlinear Composites and Microstructure Evolution 161

where n ~ 1 is the creep exponent, (ro)<kl is a reference stress and Yo is a reference strain is
commonly made [23] for the slip potential functions cp(k)• but in this work, general functions
cp(k) (satisfying the above-mentioned convexity hypotheses) will be considered.
For the purposes of this section, a polycrystal is an aggregate of a large number of
identical single crystal grains, with generally distinct orientations. The size of the typical
grain is assumed to be much smaller than the size of the specimen of interest, and the grains
are assumed to be perfectly bonded at their boundaries. It is further assumed that the
geometric arrangement of the grains within the composite is statistically uniform and
isotropic. Then, the grains within the polycrystal are characterized by their orientation relative
to the reference crystal, as given by rotation tensors Q{rl (r = 1,... , N), in such a way that
phase r will be defined to be the region of the polycrystal occupied by all crystals with
orientations Q{rl. Therefore the local potential u of the polycrystal is defined by relation
(2.11 ), where x(rl is the characteristic function of phase r, as defined by (2.12), and where
U(r)( <J) =U(c)( Q(r)T <JQ(r)) (4.35)

denotes the stress potential of the rth phase. It is remarked that u(r) may be rewritten in the
form

L cp(k)( -z-i;i)
K

u<rl( cr) = (4.36)

k=l
where
(4.37)
The variational statement developed in this section is a direct extension of the
variational principle (4.15), for nonlinear heterogeneous materials with locally isotropic
behavior, to classes of nonlinear heterogeneous materials with a special type of locally
anisotropic behavior, broad enough to include the nonlinear polycrystalline aggregates
discussed at the beginning of this section. Thus, an exact variational representation is given
below for the effective stress potential of N-phase heterogeneous materials with phase
potential functions (4.36), except that the slip potential functions cp{kl will not be assumed to
be the same for each phase, and therefore will be replaced by phase-dependent functions
cpi;~. The specialization of the relevant results for the polycrystals of the previous section
(with identical grains) is straightforward.
Following the development of deBotton & Ponte Castaneda [61], the variables
p=lrl 2 , (4.38)
 162 P. Ponte Castaneda

are introduced and the generalized slip potential functions cp(;; are assumed to be such that
there exist corresponding functions gi;~ satisfying the relation

P) = cp(kJ (r)(
(r)( )
g(kJ r · (4.39)
The functions g(;; clearly satisfy the condition g(<;;(o) = 0, and are further assumed to be
convex. For example, the power-law functions (4.34), with n ~ 1, satisfy this last
hypothesis.
Next, defining the convex Legendre transform of g~;~ via

(4.40)

it is observed that

g(;;(p) = ~:0x{pa- (gf;;)* (a)}, (4.41)

which, on account of (4.39), may be rewritten as

(4.42)

where
v(;j(a) =m;lX{alrl 2
- q,(;j(r)}. (4.43)

{'f 'f
It then follows from (4.36) that

(r)( (r)l (r)l 2 (4.44)

u cr) -_ ~E~~ ~ a(k)
(r)( (r))}
r(k) - ~ v(k) a(k) ,

and therefore, from (2.11 ), also that

N K N K }
u(x,cr)- ~N~~N{ ~~X (x)a(k)lr(k)l -~~X
_ (r) (r) (r) 2 (r) (r) (r)
(x)v(k)(a(k)) . (4.45)

l=I, ... ,K

It is emphasized that the equalities in (4.42), (4.44), and (4.45) would need to be replaced by
inequalities (~) if the functions cp(;; were not such that the g(;; functions are convex.
In view of relations (4.44) and (4.45), it is expedient to introduce a "linear comparison
po1ycrysta1" with potential function
N

u"(x,cr) = Li'l(x)u~'l(cr), (4.46)

r=l
Nonlinear Composites and Microstructure Evolution 163

such that the quadratic potential of phase r, in the linear comparison polycrystal, is given by
K

u~'l(<J) =~<J·(M~><J) = :Lai;~lr~;~l 2 (4.47)

k=!
where
K

M<'l- ,£..J a<'lu<'>

" - 2~ (k)r(k)
® r(k)
u<'> • (4.48)
k=!

Comparing (4.36) and (4.47), it is observed that the linear comparison polycrystal has the
same distribution of phases as the original nonlinear polycrystal. In addition, the grains
belonging to a given phase r of the linear comparison polycrystal have exactly the same slip
systems as the corresponding grains of the nonlinear polycrystal. However, it is emphasized
that, even when all the grains of the nonlinear polycrystal are identical (except, of course, for
their orientations), the corresponding grains of the linear comparison polycrystal need not be
identical. This is because the "slip system compliances" ai;J
of the linear comparison
polycrystal depend not only on the corresponding slip potentials cpf;~. but also on the actual
state of stress within the grains, which in general will vary from grain to grain.
Finally, defining the function v via
N K
v(x,aii'/) = LLX<'l(x)v~;~(ai;D. (4.49)
r=! k=!

so that an expression like (4.12) may be written for u, and following a procedure exactly
analogous to that used in the development of the variational statement (4.15), the following
expression is derived for U, namely,
U(<J) = max {u,(a)- (v(x,a<<i'll))}. (4.50)
a/;'/(x)~O
.<=! •...• N
I=! ....• K

where U, is the effective potential of the linear comparison polycrystal, which is defined by
(4.16) with u, given by (4.46) and (4.47). It is emphasized that the optimization variables
aii'/ (k =1, . .. ,K) are actually sets of nonnegative functions ofx, each set of functions being
defined over its corresponding phase (s = I, ... , N). Because of this variability of the slip
compliances ab'J-not only from grain to grain-but also within each grain, the effective
potential of the associated class of linear comparison polycrystals would be extremely
difficult to compute in general. For these reasons, it is extremely difficult to make direct use
of the above variational statement in the computation of (; for a specific nonlinear
polycrystal. However, as it was found by Ponte Castaneda [29] for the corresponding locally
164 P. Ponte Castaneda

isotropic composites, the above variational statement lends itself to a useful approximation,
which will be discussed in the next section.
Finally, it is remarked that a corresponding generalization of (4.24) may be obtained
for the nonlinear polycrystals. It takes the form

U(a) ~ max
al!/(x)~O
{uo (a)+ min ((u- uJ* (x, a>)}
a(x)eS
(4.51)
s=l .... ,N
l=l, ... ,K

where u0 will still be characterized by (4.46) and (4.47). With this choice of U0 , it follows
that the variational statement (4.51) provides an appropriate generalization of the variational
statement (4.50) for polycrystals with potential functions that are not such that the associated
g functions are convex.

5. BOUNDS AND ESTIMATES FOR NONLINEAR COMPOSITES

5.1 Classical bounds [19-20]

As was the case for linear composites, the minimum energy principles may be used also
for nonlinear composites to obtain rigorous bounds for the effective potentials. This was
recognized by Bishop and Hill [19-20] in the context of rigid/plastic polycrystals and
Drucker [64] for porous and particle-reinforced composites with perfectly plastic matrix
phases. Hutchinson [23] carried out analogous calculations for power-law creeping
polycrystals. Thus, use of a uniform trial strain field in the expression (2.20) for the potential
W leads to the rigorous upper bound
N
W(£) ~ :Lc<r>w<r>(£), (5.1a)
r=l

t c'''(u''')'J
which can be rewritten in the form

U(a) [ ~ (G) (5.1 b)

[t t
Yet another form of this bound may be obtained by noting that

c''' (u''1)' J =ml"'{a·

(G) E- c''' "!,~ [E.G''' - u'''(iT'')]}

=~.? { t c''1u''1(iT'') + ml"'[E( (j- t ll

c'''a''1)
Nonlinear Composites and Microstructure Evolution 165

where use has been made of the Saddle Point theorem to interchange the order of the
maximum over £ and the minimum over (j(rl. Next, noting that the inner maximum is

t.
positive infinity unless the term in parentheses is zero, it is concluded that

ii(li) ~ • n;.~ { c'"u'''(a'")}. (5.I c)

I,c<'>cf' 1=ii
r=l

t. t
For composites with isotropic phases with potentials (2.6), this last result specializes to

ii(li) ~ • ":}~ { c'"'l'''' (jl - w''' r,)} + ~ [ c'''K''J' if~, (5.I d)

l;cVlwVl=O
r=l

which was given in [29]. The more general result (5.1c) was given in [61].
Similarly, the choice of a uniform stress field in the variational statement (2.23) for U
leads to the upper bound

u(<J) ~ 2:N
c(')u(r) (<J), (5.2a)
r=l

t r
which can be rewritten in the dual form

W(£) ~ [ c'''( w''' )' (£) , (5.2b)

or

W(£)~ •~\~ {t.c'''w'''(£''')}. (5.2c)

I.c<'l£<rl=£
'"''

l e;.
Finally, for composites with isotropic phases (2.5), it follows that
N -1
l N (r)

W(£) ~ N ~~~
{
~/r)q,(r)(ll- (()(r)le,) + 2 ~ ;(r)} [
(5.2d)
lc1'lw!'l=O
r=l

In the context ofpolycrystalline materials, the bound (5.1) is commonly referred to as

the Taylor bound, after the work of Taylor [21] who first proposed its use of (5.1) to
estimate the overall response of ductile polycrystals. The bound (5.2) is often associated with
the name of Sachs [65].
Like the Voigt/Reuss bounds for linear composites, the Taylor/Sachs bounds for
nonlinear composites make use of limited statistical information - the volume fractions of
 166 P. Ponte Castaneda

the phases. For this reason, it is important to develop alternative methods to bound and
estimate the effective behavior of nonlinear composites which are capable of bringing in
additional statistical information. In this context, second-order bounds for nonlinear
composites were first obtained by Willis [26] for nonlinear dielectrics and Ponte Castaneda &
Willis [42] for nonlinear viscous composites using an extension of the Hashin-Shtrikman
variational principles for nonlinear materials due to Willis [10] and Talbot & Willis [27]. The
basic result is embodied in (3.5), where the homogeneous reference material is still chosen to
be linear, so that advantage can still be taken of the representations (3.7) in terms of the
Green's function for the homogeneous reference material. As mentioned earlier, an
alternative approach will be used here- one that makes use of the variational principles of
Section 4 allowing the representation of the effective potential of a nonlinear composite in
terms of the effective potential of an appropriately chosen linear comparison composite.
This approach has the advantage that it not only permits the determination of nonlinear
Hashin-Shtrikman bounds, but also allows the use of any other bound and estimate for linear
composites to generate corresponding bounds and estimates for nonlinear composites.

5.2 Bounds for composites with isotropic phases via the variational principles of§ 4.1

As mentioned in Section 4.1, the variational statement (4.9) is an exact representation

for the effective potential W of a nonlinear composite material. However, it requires the
computation of the effective potential ~~ of a linear comparison composite with an arbitrary
distribution of shear moduli f.lo(x), as well as the solution of an infinite-dimensional
optimization problem for JIJx), both of which problems are quite difficult. One possible
approximation that simplifies both of these problems, at the expense of introducing some
error in the estimation of the exact W, is to consider the minimization over the set of
piecewise constant moduli
N

f.lo(x) = Li'l(x)f.l~r)' (5.3)

r=l
where the f.l~rJ are constants. With this approximation, the local potential w" for the linear
comparison composite (4.5) may be written in the same form (2.lla) as that for the nonlinear
potential w, in terms of some phase potentials

(5.4)
Nonlinear Composites and Microstructure Evolution 167

In view of the reduction in the set of trial functions, it then follows [28-29] from (4.9) that

W(£) min {w (£) + ~

~ J.l.~,>~o; o~
c<"'V' (J..l-'1
o
1)}. (5.5)
r=l, ... ,N s=l

where W, is now the effective potential (4.10) of a linear comparison composite with the
same microstructure as the nonlinear composite. Even though the phases of the linear
comparison composite are isotropic, the overall behavior of the composite may be anisotropic
if the distribution of the phases in the composite is not statistically isotropic. Then, the
effective potential of the linear comparison composite may be written in the form (see
(2.27a))
(5.6)

where L" is the effective modulus tensor of the composite, which depends on the phase
moduli J.1~' 1 and l(<rl (r = 1, ... , N), as well as on the microstructure. Also, the functions
v<rl, such that the function v in (4,7) is given in terms of the v<rl by an expression analogous
to (2.11) for w, are given by

v<' 1 (J.1~' 1 ) = ~~x{ cp<'1(e,)-% J.1~' 1 e;} =m:x{w<'1(E)- w~' 1 (E)}. (5.7)

Thus, given any nonlinear composite with known microstructure, the expression (5.5)
provides an upper bound for its effective potential W in terms of the effective modulus L"
of a linear composite with precisely the same distribution of phases as the given nonlinear
composite. As discussed earlier, the microstructure of a random composite is not typically
known with precision - only some of its statistics are, usually only up to second or third
order. In this context, it is useful to consider classes of composites with prescribed
microstructural statistics and to seek bounds for its effective potentials. If an upper bound is
available for the effective modulus tensor L" of a given class of linear composites, such as,
for example, the Hashin-Shtrikman bound (3.25) for composites with ellipsoidal symmetry,
then (5.5) may be used to generate an upper bound for the effective potential W of the
corresponding class of nonlinear composites. If on the other hand, a lower bound is available
for L", then, obviously, (5.5) may not be used to generate a lower bound for W. Finally, if
L" of a composite
an ad hoc estimate, such as the self-consistent estimate, is available for the
(or class of composites), (5.5) may be used to generate a corresponding estimate for W,
knowing that such an estimate is likely to be an overestimate.
 168 P. Ponte Castaneda

If the minimizing moduli /l~'J (r = 1,... , N) in relation (5.5) are determined (as they

[- l ()
usually are) by the set stationarity conditions
l_ aLO (A(l) A(N))- (r)av' (A(r))-o (
2 E. aJl~rJ Jlo ' ... , Jlo E +c aJl~rJ Jlo - r = 1,... , N) ' (5.8)

then noting that the effective stress-strain relation (2.22) may be written in the form
- -L- (A A(N))- __
1 .[ _
aL-o { A(J) A(N)) ] _ a (.•)
(s)_v_( A(s)) }~
a
a a
(I) N {
~ A(.<)
CJ- o Jlo , ... ,Jlo £+ £..J 2E (s) Jlo , ... ,Jlo E+c (s) Jlo a£,
s=i 'Jlo 'Jlo
where the inequality in (5.5) has been replaced by an equality in the sense of an
approximation, it follows that (e.g., [66])
-CJ-LoJlo''"'Jlo
_- (A (I)A(N))-E. (5.9)
It is important to emphasize that, in spite of its appearance, this relation corresponds to
nonlinear overall behavior for the nonlinear composite. This is because the set of minimizing
comparison moduli fl~'J depend on the applied average strain £ through the set of relations
(5.8). The effective modulus tensor L" of the linear comparison composite in (5.9)
corresponds to an approximation to the "effective secant modulus" of the nonlinear
composite, the exact one already having been defined by (4.11 ).
As mentioned earlier in Section 4.1, the exact strain field E (x) in the nonlinear
composite may be computed from the exact strain field in the linear comparison composite, as
determined by the variational statement (4.1 0), evaluated at the exact minimizing comparison
moduli Jlo(x), which is determined in turn by the exact version of the variational statement
(4.9). However, consistent with the piecewise constant approximation (5.3) for the linear
comparison moduli Jlo(x), an approximation for the strain field in the nonlinear composite
may be obtained from the strain field in the linear comparison composite evaluated at the
optimal comparison moduli fl~'J from relations (5.8). In particular, the average strains in the
phases may be written in the form (see [33])
E(rl =A~;l(fl~n, ... ,fl~Nl)e, (5.10a)
where the tensors A~J are the strain concentration tensors (3.22a) of the linear comparison
composite, evaluated at the optimal values fl~'J of the comparison moduli. Similarly, the
average spin tensors over the phases may be approximated by the expressions
ro(') =m-c~;)(/l~ 1 l, ... ,fl~N))e, (5.10b)
where the C~J denote the spin concentration tensors (3.22b) of the linear comparison
composite, evaluated at the fl~'J. Note that- just as was the case for the effective secant
Nonlinear Composites and Microstructure Evolution 169

modulus Lo -the strain and spin concentration tensors in general depend on the prescribed
average strain (through the jt~'l).
It is useful to note that theN stationary conditions in (5.8) are not all independent.
Indeed, noting from (4.1 0) and (5.6) that Lo is a homogenous tensor function of degree 1 of

l
the comparison phase moduli fl~'> and K<rJ, it follows from Euler's theorem that

~(u<"> at" + 7(<"> ato

,£..J r-o = L()' (5.11)

lL
a,<s) al((s)
s=l t""o
so that
1
_-, [ L- {~(I) ~(N))-L
N aL- o {~(I) ~(N))-
N a (.v)
(s)~(s)_v_(~(s))=O
2
E o flo , ... ,flo 1( (s) _
a1("() flo , ... ,flo E+
_
C flo :1 ()
d,/1 ..
flo · (5.12)
s=l s=l ro

The above estimates for the effective behavior of nonlinear composites were first given
by Ponte Castaneda [28] in the form (5.5). Of course, the idea of using a linear comparison
composite to estimate the effective behavior of nonlinear composites is not new and has been
used by Hill [22] in the context of his incremental self-consistent procedure, which interprets
the nonlinear composite as a linear one at each increment of the applied average strain.
Similarly, Berveiller and Zaoui [24] (see also [67]) proposed a "secant modulus theory"
providing an estimate for the nonlinear composite akin to (5.9), where the secant moduli it~'>
of the phases where chosen through some ad hoc relations, different from (5.8). The
interesting feature of the choice (5.8) in [28] for the comparison moduli it~'> is that they were
derived from a rigorous variational principle, thus making the choice (5.8) the best possible
within the class of piecewise constant secant moduli. In particular, this suggests that the
predictions of the "secant modulus theory" [24, 67] should be less accurate than those
arising the variational theory of [28-29]. This has been demonstrated explicitly through
examples [68, 69].
Next, an alternative form of the bound (5.5) for W is developed. In view of relations

t
(5.7), the bound (5.5) may be rewritten

IV<£) ~ .~l~~{±£ [L, (1':", ..,I':N) )j£ + c'' [ ~,,, <<!'' l- ~ !';''(E!'' )' l},
where the variables £~'> are obtained in terms of the fl~'l through the optimality conditions in
(5.7), namely,

(5.13)
170 P. Ponte Castaneda

(Note that these conditions uniquely determine the e~'l in terms of the f.l~'l, because of the
concavity of/hypothesis.) In tum, the optimality conditions for the fl~'l may be rewritten

.!_ _. [ JL (A (1)
£ (r)
0 A(N))]-£ _~
f.lo ,... ,f.lo C
(r)[ A(r)]2
ee +c
(r)[JrjJ(r) ( A(r)) _ 3 A(r) A(r)l Je;r)
'l ee f.lo ee 'l (r)
=O'
2 df.lo 2 at:. 0 1-lo

from which it follows, via (5.13), that

(5.14)

Finally, defining

(5.15)

and making use of relation (5.11 ), which implies that

±E. [Lo(/1~1) ,... ,fl~N)) ]£=%I c(s)l((s)[ e~')r +%I c(s) /1~")[ e_;s)r'
N

s=l
N

s=l
(5.16)

leads to the result

W(£) ~% Ll((s)(e~) )2 + L
N N
c(s)rf>(s)( e_;s))' (5.17)
s=l s=l

where the variables fl~r)' e~r) and f_~) are determined from the relations (5.13), (5.14) and
(5.15). Note that e~'l and e~> happen to be "second moments" of the strain (see [70-73]).
The alternative form (5.17) of the bound (5.5) for the effective potential of the
nonlinear composite is essentially due to Suquet [70-71 ], although this author provided a
different derivation of the result. However, bounds of the form (5.17) for various special
cases have been given earlier in other works for porous materials [28, 30, 31, 72], rigidly
reinforced composites [28, 30, 31], laminates [66], and more general two-phase composites
[29, 75, 76] (see also [73, 74] for related results for porous and rigidly reinforced materials
with incompressible phases). Note, however, that the simplified form of the Hashin-
Shtrikman estimates of [29] for N-phase (N ~ 3) systems (see eqs. 4.30 in that reference)
may give incorrect results depending on values of material parameters and volume fractions,
although, of course, the original form (5.5) of the results is always correct. On the other
hand, the simplified results (4.34) of [29] for N = 2 are always correct and equivalent to
(5.17). In this sense, the alternative form (5.17) is a "better" simplified form of the results-
one which works for any type of estimates, Hashin-Shtrikman, or otherwise.
Nonlinear Composites and Microstructure Evolution 171

Next, for completeness, the corresponding forms of the bounds for the dual potential

t.
U are given as follows

ii (a) ~ :r:~. {~a [M. (tt~", ... ,~~~NJ )a]- c'''v'''(tt!'')}, (5.18)

where Mo is the effective compliance tensor of the linear comparison composite, which
depends on the phase moduli Jl~'> and IC<r> (r = 1, ... , N), as well as on the microstructure,
and where the functions v<r> are given by

v<'>(Jl~'>) = max{ 6 \,>a} -lfl<'>(a".)} = max{u~'>(a)- u<'>(a)}.

(Je Jlo 0'
(5.19)

_!_-.
2 (J [aM()-
Then, given the conditions
(A (I)
aJl~r) Jlo ,... ,Jlo
l av()
A(N)) - - (r)
0"-C
r ( A(r))
aJl~r) Jlo (r = 1, ... , N), (5.20)

determining the optimal [l~'>, the effective stress-strain relation (2.26) may be written as
- - (A (I) A(N))- (5.21)
E=MO Jlo , ... ,Jlo 0".

Finally, the bound (5.18) may alternatively be written

N N
UCCi) ~ _!_2 £.
~-~-(a-<"'>)2
l((s) m
~ c<"'>lfl<"'>(a<"'e >) '
+ £. (5.22)
s=l s=l
where

a-<rl
e
3 - [aM:"
= F>(J. (Ao> A(Nl)]-(J'
aJl~r) Jlo , ... ,Jlo (5.23)

(5.24)

and where
(5.25)

5.3 Bounds for composites with isotropic phases via the variational principles of§ 4.2
The variational principle (4.21) has the capability to give improved bounds by making
use of the classical lower bound of Section 5.1 for W w, namely of

['f
0 -

min ((w"- w)** (x,e))

£(x)EK
~ £. c<"'>( w~··>- w<"'>)*]* (E). (5.26)
s=l
 172 P. Ponte Castaneda

It then follows, from (4.21 ), that

W(E)~ min {w(£)-[~ c(.<)(w<s) -w(.<)r]*(E)}.

J.l~'):a:O; o ~ o
(5.27)
r=l •... ,N .•=1

Recalling (5.6), the minimizing conditions for the ji~'> (r = 1,... , N) are given formally by

_!_-
2E
0 [

aatO(r)
'flo
a'flo [~
(f..loA(l) ,... ,f..LoA(N))]-E _c _a(r)
~c (r)
s=1
(S)( (S)-
wo W
(.<))*]*: 0
' (5.28)

so that here also the relation (5.9) holds, with the ji~'> now determined by (5.28).
The bound (5.27) for W is similar in form to the bound (5.5), and although it is more
difficult to compute (the double Legendre transform in (5.27).being more difficult than the
minima (5.7) in the v<r> functions), the bound (5.27) has the advantage that it reproduces the
nonlinear potential function w, for a homogeneous material, even when the concavity of f
hypothesis is violated.

[t. t t
It is easy to see from the inequality

c''' (w:'' - w''' )' J (E) 2 - c'''(w:'' - w''' )' (0) = - c'''v''' (1':'') (5.29)

that the bound (5.27) is, in general, at least as good as the bound (5.5). However, the bound
(5.27) does not improve on the bound (5.5) for materials satisfying the concavity off
hypothesis. To see this note that relation (5.16), together with (5.14) and (5.15) (without the

t t r
hats), allow the bound (5.27) to be rewritten in the form

IV(£) ,; ~~~ ~ { t c'''K'''[e~' ]' + ~ c'''1':''[e!'' ]' - [ c'''( w:'' - w''' )' (E)}.( 5. 30)

Then, making use ofthe form (5.2c) for the classical lower bound of W0 - w, it follows that
N

W(E) ~ min max ~ c<•>{.2.K"<•>[e<">] 2 + 'i J.L<•>[£<">] 2 - (w<·•>- w<">)**(£<'>)}. (5.31)

~n<!O; £<'> ~ 2 m o e 2 o
r=I, ... ,N N s=1
:I;c<''£<'>2:£

- I ()
r:(

Finally, interchanging the minimum and the maximum in (5.31), and noting that aL" aJ.L"'
is homogenous of degree 0 in the comparison moduli, so that
'f
f:t c<•>{.2.K"<•>
2
_j_[e<•>]2 + 'i ,<s> _a_[e<">J2} = 0
af..l~r) 2 m af..l~r) ro e '
(5.32)

leads once again to the result (5.17). (Note also that the second moments of the strain e~'>
and£~> do indeed satisfy the constraint on the £<'>).
 Nonlinear Composites and Microstructure Evolution 173

5.4 Bounds for nonlinear polycrystals

The use of the approximation of piecewise constant slip system compliances in the
variational statement (4.50) for U leads to the result [61] that

U(O') ~ max {_!_[M 0 (a<'l)a]

aw~o 2 (k)
0'- .£... .£... c<"'V(/)"'>(a<"'
0
'f 'f
(/) l)} ' (5o33)
r=1oo .. ,N .<=1 1=1
k=1, ... ,K

where Mo is the effective compliance tensor of the linear comparison polycrystal with grain
compliances M~J, as given by (4.48) in terms of the slip compliances at;l, and the functions

_!_
2
_o
aa(r) a(l) (J
~)
l
v{;~ are given by expressions (4.43)0 The optimality conditions for the at;l are given by

[aM- {A(-'))- o-- c(r)~{A(rJ)-O (J

av(rJ
aa<r) a(k)
~)
- '
(5034)

from which it follows that

- o (auJ
D =M A(s))-
O"o (5035)
Since Mo is homogeneous of degree 1 on the comparison compliances at;j, it follows
that
N K -
~~ (s) (r))
.£....£...a(I) aa<s) a(k)
aMo (
=M (a(k)(r)) '
0
(5o36)
.<=1 1=1 (/)

from which it is deduced that

N K
![M:
2 () (a<'>)a]
(k}
0 (i =
.£... ~ c<s>a<"'
~ .£... (/) >[f(s>]2
(/) ' (5037)
.<=1 1=1

where

'l'A(rJ
(k} -
_ - 1-[aMo
--a
2c(r} aa<rl
( (/)(s})-]-
<Jo<J
0
(5038)
(k}

Similarly, since a~~ is homogeneous of degree 0 in at;j, it follows that

aa(k)

N K a[f(s)]2
~~
.£... .£...
c<s>a<sl(/) (ll - 0
aa<r) - 0
(5o39)
s=1 1=1 (k}

Then, making use of (5037) in (5o33), it is concluded that

N K
U(O') >- max ~ ~ c<s>{a<s>[f(s)]2- (s>(a<s>)}
!'>>o £..i £..i (I) (I) v(IJ (IJ '
(5.40)
a!>)-
r=1o ... ,N s=1 1=1
k=1, ... ,K
174 P. Ponte Castaneda

which, in view of (4.42) and (5.39), leads to

N K
u(a)~ LL/')q>li'/(f(i'l), (5.41)
s=l 1=1

where
d (r)
~ (r) _ 1 q>(k) ( ~(r))
(5.42)
a(k) - 2f(r) Jr(r) r(k)
(k) (k)

is to be used in conjunction with (5.38) to determine the f(~l·

It is important to note that the variables f((;l in (5.38) are all independent (except that
(5.37) must be satisfied). Thus, it is not true, in general, th~t the variables f(;l may be
written in the form f{;l = o(rJ · Jl~~~ for some appropriate set of variables o(rJ. In other words,
it is not possible to reduce the number of optimization variables a(~l by expressing them in
terms of some Taylor-like stress in each grain of the polycrystal. This could, of course, be
done as an approximation, but one that is unlikely to give as good estimates for U.
Finally, it is remarked that a possible generalization of these bounds for materials not
satisfying the hypothesis of convexity of the g(;; functions is obtained from (4.51) by
making use of piecewise constant a{;l. The result may be written

U(iJ) ~ .~~ { MM.(ai;:Ja] iJ+[t c (u'''- u;'')'Jta+

1' 1 (5.43)

k=I ..... K

where the u(r) and u~rJ are given by expressions (4.36) and (4.47), respectively.
This result has the advantage that it can reproduce the Taylor bound with the simplest
choice a(~l = 0 for all rand k. This is not the case for the simpler bound (5.33), which
requires maximization over a full set of slip system compliances a(;l to recover the Taylor
bound (see [61]). On the other hand, it follows by arguments completely analogous to those
used in Section 5.3 for composites with isotropic phases that the bound (5.43) is identical to
the bound (5.33) for polycrystals with phases with convex g(;; functions. In this context, it
is important to remark that Bonnet et al. [77] proposed the use of a bound similar to (5.43),
but with a(~l =a for all rand k. When use is made of this so-called "constrained stress"
bound in conjunction with a Hashin-Shtrikman bound for the linear comparison polycrystals,
the resulting bound for the nonlinear polycrystal is not as good as the corresponding bound
of deBotton & Ponte Castaneda [61], obtained from (5.33) by optimizing over all the aW.
Nonlinear Composites and Microstructure Evolution 175

However, the "constrained stress" bound (with at;~= a) is significantly sharper than the
bound that would be obtained from (5.33) with all the a~;~ =a.
A disadvantage of (5.43) is that the difficulty associated with the computation of the
double Legendre transform in (5.43). For this reason, the following "relaxation" of the
bound (5.43) has been proposed by Bonnet et al. [77] (recall that these authors made use of
the additional assumption that at;~ =a for all rand k), following Dendievel et al. [78],
namely,

ii(a)
k~I •... ,K
(j t.
~ ,~t, {MM. (a[;i)al + [ ~ c'''(~!;,' - ~:ch)}a+ (5.44)

where cpri·(!lr) = a[!/(-r( However, again this bound (with a full set of optimization
variables a~;l) can be shown to be equivalent to the bound (5.33), which is still simpler to
evaluate.
Finally, it is noted that Bhattacharya & Kohn [58] have recently made use of bounds of
the type (5.43) in the context of shape-memory polycrystals. However, it appears that these
authors also did not optimize over the full set of at;l,
raising the possibility that their bounds
may not be as good as they could be, if the optimizations had been carried out over the full
t f (r)
se o a<*l'

5.5 Some illustrative examples

Several examples of applications of the above variational estimates for the effective
behavior of various nonlinear composite systems have been given over the past few years.
These include Hashin-Shtrikman upper (lower) bounds for W (U) [28-31, 33, 72, 79] for
porous materials (see also [42, 80] making use of the Talbot-Willis extension of the Hashin-
Shtrikman variational principles), Hashin-Shtrikman, self-consistent and three-phase
estimates for rigidly reinforced systems [28-32] (see also [42]), Hashin-Shtrikman bounds
and estimates for two-phase perfectly plastic systems [31, 32, 81 ], fiber- and particle-
reinforced composites [69, 75, 76, 81] (see also [83] making use of Talbot-Willis variational
principles], Hashin-Shtrikman estimates for polycrystals [61] (see also [78] making use of
the Talbot-Willis variational principles). Below, two examples which are representative of
the above-mentioned results are included for illustrative purposes.
 176 P. Ponte Castaneda

5.5.1 A Beran bound for statistically isotropic porous nonlinear composites

Denoting the incompressible matrix phase of a porous material with stress potential
(2.6) (with K'Ol --7 oo) by the superscript I and the void phase by superscript 2, it follows
from expressions (3.30) for the effective moduli of a composite with statistically isotropic
microstructure and prescribed third-order parameters 77°) and ~Ol, together with expression
(5.18) or, alternatively, from (5.22-5.25), that the effective potential of the nonlinear porous
composite is given by

U-(-) > Ol
(J _c lfl
(!)[(i)I 2 (2))-
( 1+-(-l)c 2 9 c( 2) -21
(je +-(l)(jm . (5.45)
c 37] 4~
The weakest possible bound is obtained in the limit when 7]( 1) and ~(!) --7 I, corresponding
to the statistically isotropic particulate microstructures of Section 3.4 (with
w}il = w~il = wid) = w~d) = I) with the void phase occupying the inclusions. The resulting
Hashin-Shtrikman bound

U(a) ~c 0 llf/ol( )n (I+ ~c(2) )a;+ ~c(2 )0'~) (5.46)

was first given in [28] (see also [29, 30, 42, 72, 73, 74, 79, 80] for alternative derivations of
this, or closely related results). The more general bound (5.45) was given in [29], but in less
explicit form.

5.5.2 Hashin-Shtrikman estimates for particulate power-law and ideally plastic composites.
In this section, use is made of the Hashin-Shtrikman estimates (3.27) for linearly
viscous composites, specialized to two-phase isotropic, incompressible constituents, to
generate via (5.18) estimates for the effective potential of nonlinearly viscous composites
with pure power-law constituents of the type described by (2.7) with Jl--7 oo. In order to
simplify the expression (3.27) for this case, the following two observations are made. First,
by virtue of the incompressibility and isotropy of the phases of the linear comparison
composite, and taking the superscript 2 to denote the matrix phase, expression (3.27) for M
may be written in the form
M=-~--(
3J1(2) mY
(I)) (5.47)

where m depends on the linear comparison viscosities J1°l and J1< 2l, only through the ratio
y(l) = (y( 2 )r'
= J1( 2) I Jl(l). Second, because of the pure power-law hypothesis, one has, from
(2.7), that
Nonlinear Composites and Microstructure Evolution 177

2n
( 1) (aO>)n-1 2n n+I
V(l)(.u(I)) = n- Y n+l = (z(2) )n-1 (ll )n-1 V(2)(.u(2)) (5.48)
2(n + 1) (3,u(ll)n-I

where z< 2> =(z<l) t =a~.l) j a~. >

2 is the ratio ofthe reference stresses of the two phases.

It then follows that expression (3.27) may be rewritten in the form

U(a) ~ max{[c( l + c( 1 l(z( 2 l)•2~1 (y( 1 l):~: ]max{-

)'(il;,o
2 1-(8.(1)) 2 - V( l(,u( l)}}·
6.U(2)
~1ll;,o e
2 2 (5.49)

where

(5.50)

from which it follows, via (5.19), that

U(a) ~ ~~~{[c(2) + c(ll( z(2) )•~I (ll ):~:] 11'(2)[a~l)(ll)]}. (5.51)

These results for power-law composites may be specialized for rigid/perfectly plastic
composites, by noting that 11'(2)(a,)= 0 if a, :5: a~2 >, or oo otherwise, which leads to the

r
following estimate for the corresponding effective yield function (2.28):
<i>(cr) :5: 91~~{[ c(2) + c(ll( l2))2 y<l) (j. [ m(y(l>)cr)- (a~2)r}. (5.52)

Estimates (5.51) and (5.52) may also be obtained directly from Suquet's [31]
variational principle (4.27) for power-law composites by making use of piecewise constant
viscosities. The derivation given here follows that of [84]. For perfectly plastic composites
with a statistically isotropic distribution of spherical inclusions ( w~il = w;il
= w~"> = w;"> = 1)
result (5.52) yields the following estimate for the effective yield strength &,. in tension of the
composite, namely,

2 (I)) 2 (I)
( 1+ _c_ (a<z>)z - _c_(a<I>)z if z<2> < ~
~
2
3 r 3 .\' 2 2 + 3c(l) '

"f (2)>5 ~
I Z -2~~·
 178 P. Ponte Castaneda

This result, first given in [81] and independently in [31] and [32], can also be interpreted as a
rigorous upper bound for the yield strength of composites with general statistically isotropic
microstructures.
Results have also been given in [81] and [86] for fiber-reinforced composites
( wii) = wiil =widl =widl --7 oo ). In this case, there are essentially three modes corresponding
to axisymmetric loading, transverse shear and longitudinal shear. Denoting the yield strength
in shear for these three modes by id, iP and in, respectively, it follows that

As can be seen from the Fig 3, for combined loading, the yield surfaces are "bi-modal" [85]
in qualitative agreement with the numerical simulations of Moulinec & Suquet [86], as well
as with the experimental results of Dvorak & Bahei-El-Din [87].

c( 1l = c( 2 l =0.5
0.5

4 5

Fig. 3. Comparison of the predictions (continuous lines) of the nonlinear homogenization procedure of [28]
with the periodic homogenization results (dots) of Moulinec and Suquet [86] for fiber-reinforced composites
with rigid/perfectly plastic phases. The stress invariant fJ corresponds to the axisymmetric mode (e.g.,
tension along .the fibers) and f, corresponds to shear transverse to (or along) the fibers ( <" and <" refer to
the yield stress in shear of the fiber and matrix, respectively).
Nonlinear Composites and Microstructure Evolution 179

6. MICROSTRUCTURE EVOLUTION

When a composite is subjected to finite deformation on its boundary, it is clear that

its microstructure will not remain fixed but instead will change at every stage of the
deformation process. In general, the problem of characterizing the evolution of the
microstructure is an extremely difficult one. However, for simplified descriptions of the
microstructure, it is possible to develop consistent models for its evolution. For example, for
linearly viscous composites with dilute concentrations of inclusions, subjected to uniform
loading conditions, Eshelby [34] has shown (as already mentioned) that the strain rate within
the inclusions is uniform. This implies that initially ellipsoidal inclusions will deform into
ellipsoidal inclusions with possibly different size, shape and orientation. This fact has been
used by numerous authors (e.g., [88, 89]) to characterize the evolution of the size, shape and
orientation of a void or inclusion in a composite with dilute concentrations of such voids or
inclusions. In the context of polycrystalline materials, it is commonplace to evaluate the
texture development from the Taylor estimates for the spin of the grains in the polycrystals
[21 ]. In this section, more general theories are developed for composite with both particulate
and granular microstructures, which attempt to incorporate the evolution of microstructure
while accounting for interaction effects in the context of the second-order Hashin-Shtrikman
and self-consistent estimates of Section 3.

6.1 Particulate microstructures

For linearly viscous composites with the general types of particulate microstructures
defined in Section 3.4, it will be assumed in this work that the initially aligned ellipsoidal
inclusions also deform - in an appropriate average sense - into aligned ellipsoidal
inclusions of different size, shape and orientation. However, instead of using the dilute
estimates of Eshelby for the strain rate and spin in the inclusions to determine the evolution of
the inclusion size, shape and orientation, it is proposed to use the corresponding "non-dilute"
HS expressions (3.22) for the average strain rate and spin in the inclusions. At finite volume
fractions, the strain rate in the inclusions will no longer be uniform in the inclusions and,
therefore, the inclusions will not keep their ellipsoidal shape. However, to the level of
accuracy of the Hashin-Shtrikman description for the effective properties of the composite,
the strain-rate field in the inclusions is uniform, and therefore it is consistent with this
180 P. Ponte Castaneda

approximate description to assume that the ellipsoidal inclusions will deform into ellipsoidal
inclusions. Furthermore, the internal variables used in the Hashin-Shtrikman description of
the effective behavior of the composite are averaged statistical variables, so there is no need
to determine precisely how each particle evolves - only the average size, shape and
orientation of the particles are important. In addition, it will be assumed that the shape and
orientation of the distributional ellipsoid will be governed by the average strain rate and spin
in the composite. This is required to keep the integrity of the composite.
The suggestion to make use of the HS estimates for the average strain rate in the
inclusions to estimate the change in shape of the inclusions in particulate composites was
made by Ponte Castaneda and Zaidman [33], in the context of triaxial loading conditions,
when the orientation of the inclusions remains fixed. In addition, these authors made the
simplifying assumption that the shape of the distributional ellipsoid evolves in precisely the
same fashion as the ellipsoidal inclusions. This was done in order to be able to use the HS
estimates of Willis [47, 50] for composites with identical shape for the inclusions and
distributional ellipsoids. More realistic results, allowing for independent changes in the shape
of the inclusions and pair-distribution function were developed recently by Kailasam et al.
[35], using the more general HS estimates of Ponte Castaneda and Willis [51], but still
keeping the hypothesis of triaxial loading conditions to ensure that the inclusions remain
fixed in orientation. Finally, the extension to more general boundary conditions, including
boundary conditions such as simple shear loading which lead to the rotation of the
inclusions, was proposed recently by Kailasam and Ponte Castaneda [36].
Restricting our attention to two-phase systems with only one set of aligned, self-
similar, isotropic inclusions, distributed randomly in an isotropic matrix, the basic "internal
variables" characterizing the state of the microstructure are the volume fraction of the
inclusions c(i) = c( 2 ), the two aspect ratios of the ellipsoidal inclusions w~i), wi;), the two
aspect ratios of the pair-distribution function of the inclusions w~d), wid), the three Euler
angles of the inclusions (j>U), 'I'U), eu) and the three Euler angles of the pair-distribution
function (j>(d), 'l'(d), e<d). Of course, more complex systems, including, for example,
composites with distributions of differently oriented particles, could also be considered, but
the two-phase aligned-inclusion case is a simple prototype of the more general case. Having
identified the relevant microstructural variables for the two-phase system, the appropriate
"evolution laws" for these variables are derived next.
Nonlinear Composites and Microstructure Evolution 181

6.1.1 Linearly viscous composites

An evolution law for the inclusion volume fraction is obtained from the exact
kinematical relation
(6.1)
where £( 1) and £(2l denote the average strain rates in the matrix and inclusion phases,
respectively. However, when the matrix is incompressible, tr£< 1l =0 and therefore (recall
that c< 1l£(Jl + c< 2l£< 2l = £)
c<il = (1- c<il)tr£. (6.2)

This last expression for a composite with an incompressible matrix provides the evolution
law for the volume fraction of inclusions c<;J in terms of the average strain rate in the
composite. For the more general case when the matrix is compressible, the evolution law for
c<il is obtained by making use of the estimate (3.28) for the average strain rate in the
inclusions £< 2l =A <2l£.
Consistent with the assumption that the ellipsoidal inclusions deform into ellipsoidal
inclusions with the average strain rate £( 2) in the inclusions, the change in the aspect ratios of
the inclusions is governed by the kinematical relations
w(i)-
I -
w<il(e<
I D
2/- e<II2/) and w(i)- 2 D2/- c.22
2 - w<i)(e< c;( 2)
1
)
• (6.3)

where the primes denote components relative to the x' axes which instantaneously coincide
with the principal directions of the ellipsoidal inclusions. Here again the average strain rate
£< 2l in the inclusions is obtained from £< 2l =A<2l£ in terms of the Hashin-Shtrikman
estimates (3.28) for A<2l.
The evolution of the inclusion orientation angles is determined from the spin of the
inclusions w<il through the relations:

(jjl~) =-lif(i)- tiJ<il cose(i)'

=-iJ<il cos "'(i) - tiJ<il sin "'(i) sin e(i)'
(jj~~) (6.4)
m~;> =-iJU> sin l/f(i) + tiJU> COS l/f(i) sin 8(i).

In turn, the spin of the inclusions m(il is related to the spin of the material in the inclusions
m< 2l, as given by m< 2l =m- c< 2l£ in terms of expression (3.29) for C(2), by the well-
known relation
2
=m<2J - (w<iljwu.o>) + (w<iljw<i.o>) e(2l
1 1
2 1
w<il
kl kl (
k

Wk
k I I
(i)/ Wk(i,0))2 - (WI(i)/WI(i,0))2 kl '
k
* z' (6.5)
 182 P. Ponte Castaneda

where wy> and wY' 0l (k = 1, 2) are the current and initial values of the inclusion aspect ratios
and wjil = wY 0 l = 1.
It follows from the hypothesis that the distribution of the inclusions is controlled by the
bulk flow that the evolution of the aspect ratios of the pair distribution function of the
inclusions is given by
. (d) -
wi - WI
(d)(- " - ")
£33 -£II an d . (d)-
W2
(d)(- "
- W2 £~3
-
- f22
")
' (6.6)

where the double primes denote components relative to the distributional coordinate system
x". Finally, the evolution of the distributional angles is determined by equations similar to
(6.4)- with the superscript (i) replaced by (d)- together with

-(d( - I )2 + (wjdl I w,<d.o) )2 ;:; "

" - ( Wkdl Wkd.OJ
'k=F-[, (6.7)
( Wkdl I Wkd.OJ) - ( wjdl I w,<d.o))
OJkl =OJkl 2 2ck/

where Wkdl and wid.o) have a meaning completely analogous to wiil and wii.OJ. Note that
these last two evolution laws can not be expected to hold for strains that are large enough to
bring the particles into contact with each other, nor for situations where the relative motions
of the particles are so large that three- and higher-point interaction effects become significant.
In summary, denoting the relevant internal variables by s<al (a= 1, ... , 11), the above
evolution equations may be written as a set of 11 coupled ordinary differential equations of
the form
s<al = u<al(,uol, ,u<2l;s<al) ·E + K<al(,uol ,,u<2ljal). ffi, (6.8)
where the tensors u<al and K<aJ are obtained from the expressions (3.28) and (3.29) for
A (2 ) and C(2).
Then, the effective response of the linearly viscous composite is given by
cr=L(,u<l),,u< 2J;s<a))£, (6.9)
where the internal variables s<al are determined as functions of time t by integration of (6.8).
Note that because of the implicit dependence of the s<al on E, the effective constitutive
relation (6.9) is in fact nonlinear (i.e., the relation between cr and E is nonlinear), even
though the constitutive behavior of the phases themselves has been assumed to be linear.

6.1.2 Nonlinearly viscous composites

Recalling expressions (5.10) for the average strain rate and spin in the phases of the
nonlinear composite in terms of the strain-rate and spin concentration tensors A~;J and c~;J in
the linear comparison composite, and observing that the basic kinematical expressions (6.1)
 Nonlinear Composites and Microstructure Evolution 183

to (6.7) remain valid for the nonlinearly viscous composite, it follows that the evolution laws
for the microstructural variables of the nonlinearly viscous composite subjected to finite
deformations are given by expressions similar to (6.8), except that the viscosities J1°> and
J1< 2> of the linearly viscous composite must be replaced by the optimal viscosities [l~n and

{l~2 > of the linear comparison composite, as determined by expressions (5.8), so that
s(a) = H(a)(JlA(l)
o o'o'
J1A(2), s<a>) ·E + K(a)(JlA(l) J1A(2), s<a>). (ij
o o'o' · (6.10)
With these expressions serving to characterize the evolution of the microstructural variables,
the effective response of the nonlinear composite is determined by expression (5.9) or, more
precisely, by
O=L 0
(J1A(l)
0 ,
JlA(Z),S(a))£
(} , 0
(6.11)

It is important to emphasize that although the form of the effective stress-strain relation and
evolution laws are formally the same for the linear and nonlinearly viscous composites, they
are different in fact- because of the dependence of the comparison viscosities [l,~1 > and {l~ 2 >
on the applied strain £. In particular, note that the effective behavior of the nonlinearly
viscous composite is nonlinear for two reasons: 1) the constitutive nonlinearity of the
phases, as described by the dependence of [l~n and {l~2 > on £ in relations (5.8); and 2) the
geometric nonlinearity associated with the evolution of the microstructure, as characterized
by the relations (6.10). Application of this model under simple loading conditions for power-
law porous materials and two-phase perfectly plastic systems have been considered by
Kailasam & Ponte Castaneda [90] and Zaidman & Ponte Castaneda [91]. More realistic
examples will be reported elsewhere.
Finally, note that uniform boundary conditions have been assumed for the purpose of
carrying out the homogenization procedure. However, once this is accomplished, the
resulting constitutive model can be used for general non-uniform boundary conditions,
provided that the scale of variation of the applied loading conditions is still large compared to
the size of the typical heterogeneity. For non-uniform boundary conditions, the constitutive
relation (6.11) and the evolution equations (6.10) for the now position-dependent
microstructural variables must be supplemented by the equilibrium and compatibility
conditions for the average stress a and the average strain rate £,respectively. In addition,
the convected terms due to the nonuniformity of the internal variables must be included in
evaluating their material time derivatives.
184 P. Ponte Castaneda

6.2 Granular microstructures

For viscous composites with the granular microstructures of Section 3.5, it will be
assumed that the distribution function of the phases is controlled by the mean flow. In other
words, the "shape" and "orientation" of the grains are assumed to be controlled by the
average strain rate £ and spin ro in the composite. This is required to preserve the integrity
of the composite - for example, if every grain was to deform with the average strain and
spin tensor in the grains, the composite would likely disintegrate, as each grain would in
general be expected to have a different strain and spin field. In addition, if the material inside
the grains is anisotropic, as would be the case for the polycrystals of Section 4.4, it is also
necessary to keep track of the orientation of the material within each grain, which will be
assumed here to be controlled by the average spin of the material within the grains, as
determined by the self-consistent estimates of Section 3.5 for linearly viscous composites, or
by their generalization of Section 5.4 for nonlinearly viscous polycrystals. Although a
continuum distribution of orientations could be considered, for simplicity, only a discrete set
of N orientations, corresponding to N different "phases", will be considered here. In any
event, the continuum case would have to be discretized for the purpose of performing the
required numerical computations.

6.2.1 Linearly viscous composites

For the linearly viscous composites with granular microstructures of Section 3.5, the
basic "microstructural variables" are the orientation of the symmetry axes of the (anisotropic)
grain viscosity tensors L(rl (r = 1,... , N) , as determined by the sets of Euler angles cf>(rl,
lfl(r), e(r) (r = 1, ... , N). Note that, of course, for standard fluids, frame indifference requires

the L(r) to be isotropic, in which case the variables l/>(r), lfl(r), e(r) become irrelevant.
However, there may be situations, as for liquid crystals (which themselves possess a
microstructure, at a smaller length scale), where it becomes necessary to keep track of the
orientation of the material in the grains. In addition, there is also the shape and orientation of
the self-similar grains, as determined by the aspect ratios of the grains wig), wig) (or more
accurately, of the two-point grain distribution function) and the three Euler angles
characterizing the grain orientation lf>(g), ljl(g), (}(g). Note that the orientation of the grains and
of the axes of anisotropy of the material inside the grains are different in general - the
second one is usually referred to as the crystallographic texture, the first could be referred to
 Nonlinear Composites and Microstructure Evolution 185

as "geometric" texture [83] . Having identified the relevant microstructural variables,

appropriate "evolution laws" for these variables are derived next.
Consistent with the hypotheses that the ellipsoidal grains deform into ellipsoidal grains
with the average strain rate £, the change in the aspect ratios of the grains is governed by the
kinematical relations
w(g) - w(g)(e
I - I 33
I - E I)
II
and w(g) - w(g)(e
2 - 2 33
I - E22 I) • (6.12)

where the primes denote components relative to the x' axes which instantaneously coincide
with the principal directions of the ellipsoidal grains.
Similarly, the evolution of the grain orientation angles is determined from the average
spin ro in the composite through the relations:
rol~) = -lit(g)- ~(g) cose(g)'
roi:) = -iJ(g) cos ll'(g) -~(g) sin ll'(g) sin (}(g)' (6.13a)
rojfl = -iJ(g) sin ll'(g) +~(g) cos ll'(g) sin e<g)'
where
I I (w<gl jw<g.o>)2 + (w<gl jw<g.Ol)2
(.ij(g) - (.ij - k k I I E k :;:. l (6 13b)
kl - kl ( (g)/ (g,0))2 - ( (g)/ (g,0))2 kl ' • •
Wk Wk WI WI

On the other hand, the crystallographic orientations of the material in the various
grains is determined from the "microstructural spin" tensor dm(r) of the material in each grain
r, so that
drol<;> = -lit(r)- ~(r) cose(r)'
~wi;> = -iJ<r> coslJI<rl- ~(r) sin lJI(rl sine<r>, (6.14)
dwj~> = -iJ<r> sin lJI<r> + ~<r> cos lJI<r> sin e<r>,

where microstructural spin dm(r) depends on the average "continuum" spin m(r) in a fashion
that depends on the physics of the problem. However, the average spin ro(r) in the grains is
related to the average spin m and strain rate £ through relation (3.22b) in terms of the spin
concentration tensor c< 2>, which is in tum given by expression (3.24) with L< 0>= L.
Then, the effective response of the linearly viscous composite is given by
cr=L(L<'>;s<a>)£, (6.15)
where L is given by (3.25) in terms of the concentration tensors A (r), as given by (3.23)
with L<o> = L. Here, again, the internal variables s<a> are determined as functions of time t
by integration of the appropriate evolution laws above. Note, once again, that due to the
implicit dependence of the s<a> on £ and ro, the effective relation (6.15) is in fact nonlinear.
186 P. Ponte Castaneda

6.2.2 Nonlinearly viscous polycrystals

For the nonlinear polycrystals of Sections 4.4 and 5.4, the evolution of the shape and
orientation of the grains, as determined by the aspect ratios w1<gJ, wi
8 l and Euler angles l/J< 8 J,

ljl< 8 l, o<gl, is given by the same equations (6.12) and (6.13), as for the linearly viscous
composite.
On the other hand, the crystallographic orientations of the material in the various grains
is determined by the average ("microstructural" or "elastic") spin tensor sm<r> of the lattice in
each grain r, so that
Smi;) =-tiJ(r) - ~(r) cos(}(r)'
sm<rl = -iJ<rl cos m<rl - ,.i.,<rl
23 'I'
'f'
sin 11/rl sin o<rl
'f' . '
(6.16)
Smi~) = -iJ<rl sin ljf(r) + ~(r) cos ljf(rl sin o<rl,

where it is recalled (see [21, 63, 92, 93]) that the spin of the lattice in each grain is given as
the difference between the "continuum" spin m(r) and the "plastic" spin ro~)' or
sro<rJ = ro<rJ- m~l. (6.17)
In this last relation, the average continuum spin ro<rJ in the grains is related to the average
spin ro and strain rate £ through relation (3.22b) in terms of the spin concentration tensors
c~) (with L~) = Lo ), given by expression (3.24b ), and evaluated at the optimal from ag;
(5.34). On the other hand, the plastic spin ro~l is given by expressions
K
m(r) - ~ r(r)(m(r) ® n(r) - n(r) ® m(r)) (6.18)
pi - .LJ (k) (k) (k) (k) (k) '
k=l

in terms of slip rates ri~~ on the slip systems. These slip rates are most easily determined via
expressions (4.30) and (4.37) in terms of the grain average stresses cr<rl = B~:>cr, where the
stress concentration tensors B~> are evaluated at the optimal aiti from (5.34). Alternatively,
they could also be estimated from (5.38). The concentration tensors B~:> may be obtained
from a Hashin-Shtrikman analysis completely analogous to that of Section 3.2 for the

r,
average strain rate e<rJ =A (rl£. For the granular microstructures of this section (see [9]), it

t
turns out that

B~' ~[I+ Q~"(M~'- M~")r' { c'''[I+ Q;"(M;''- M~"')r' (6.19)

where Q<OJ
o
=L<OJ -
o
0 lp< 0 lL<OJ
L<o o o·

Finally, the effective response of the nonlinearly viscous polycrystal is given by

£ = M0 (a<sJ.
(I) '
s<a>)cr ' (6 • 20)
 Nonlinear Composites and Microstructure Evolution 187

where

(6.21)
r=I

with M~0 > =M", denotes the self-consistent estimate for the effective viscous compliance of
the linear comparison polycrystal, and where the ag; satisfy the relations (5.34), and the s(a)

denote the above-defined microstructural variables.

6.3 Illustrative results

Figure 4 shows some results (see [33] for full details) for the application of the
simplified (as in Section 6.1 with w~il = w~dJ =w;) model of Ponte Castaneda and Zaidman
(PCZ) to uniaxial tension and compression of porous materials with initially isotropic
distributions of spherical pores ( 15 % volume fraction) in a rigid/perfectly plastic matrix.
Results are shown (continuous lines) for evolution of the porosity ( c(i) =f), the aspect ratios
( w1 = w2 = w on account of the axisymmetry of the problem), the uniaxial stress-strain
relation, and the effective hardening rate (H). (Note that the orientation of the pores is fixed
for these boundary conditions.) Results are also shown for the corresponding predictions of
the Gurson model (long dash lines) and for the PCZ model neglecting shape changes (short
dash lines). Careful study of these plots leads to the following conclusions: 1) the evolution
of pore shape has a direct effect on the effective behavior of the porous material (this is
mostly through its implications for the overall anisotropy of the porous material); 2) the
evolution of pore shape also has an indirect effect on effective behavior through its effect on
the evolution of porosity (note that neglecting the pore shape changes leads to large
overestimates for the increases in porosity under tensile conditions and to large overestimates
in the void collapse strain for compressive conditions); 3) the combined implication of these
two effects is that the effective hardening predictions may be significantly different if pore
shape changes are accounted for (for example, in tension, the PCZ model leads to overall
hardening, whereas the Gurson model predicts strong overall softening); 4) the predictions of
the model without pore shape evolution are very similar to those of the Gurson model
suggesting that the differences between the new model (with pore shape evolution) and the
Gurson model arise as a consequence of the change in pore shape; 5) the predictions of the
PCZ model for total void collapse are in much better agreement with experimental and
analytical results (see [94, 95]) than the predictions of the Gurson model (more than 250% ).
 188 P. Ponte Castaneda

f
0.3 ......-~....-.-~....--~~-.-~-.....,
w
6.-----~----~----~--~

,.,G)··<
.......
Tension .... ....:::.-····Hs
0.2 ... :::;...... w=I.O
'IC-~-;::....-· 4
HS

·----~-~--
0.1 ""''::.~----- w =HS1.0
----~y:.:.
Compression

Compression
0 o\-~-,..~~"'"""..,_=~o.7=-:5....---llle3)1

(a) (b)

10'331
Gy
H
1,......--~....--~---T--~--~--~

Compression

-- -....---
0.9
GS ---

----··········
0.8 .. : .. :")·c:.···
............................... _
Hs···-.. . . . . . . . . . . . . . .§
0.7 w =1.0 ····-~---. HS !J~
-0.5 ----.w.:::.lQ________ ~\\
Tension
- ns---- -·=·=-=1- ·-----··
---
(c) (d)

Fig. 4. Application of the model of Ponte Castaneda and Zaidman [33] to uniaxial tension and compression
of porous materials with a rigid/perfectly plastic matrix, and initially isotropic distributions of spherical pores
in 15 percent volume fraction. (a) evolution of the porosity (j) (b) evolution of the aspect ratios ( w, = W 2 = w)
(c) uniaxial stress-strain relation (d) evolution of the effective hardening rate (H). The results for the model
with evolution of the pore shape are designated by continuous lines; the corresponding results for the model
with fixed aspect ratios are designated by short-dash lines, and those for the Gurson model with fixed aspect
ratios are designated by long-dash lines, respectively.
Nonlinear Composites and Microstructure Evolution 189

7. CONCLUDING REMARKS

The nonlinear homogenization procedures reviewed in Sections 4 and 5 exhibit a

favorable balance of accuracy versus computational effort. However, of course, they are not
free of limitations, and there may be situations where the development of more accurate
procedures may be warranted. Thus, in spite of their advantages, the homogenization
procedures of Sections 4 and 5 suffer from the following two limitations. The first has to do
with the fact that these variational procedures only give bounds in one direction, whereas
bounds on the other direction may also be of interest. In this connection, recent progress
along these lines has been made in the recent works of Talbot & Willis [96-97]. The second
limitation of the variational procedures is their inability to recover the small-contrast
asymptotic estimates of [98-99], in particular, for statistically isotropic microstructures.
While the Hashin-Shtrikman estimates for linear composites are known to be exact to second
order in the contrast (a parameter measuring the degree of inhomogeneity in the composite),
the corresponding predictions of the variational procedures for nonlinear composites are only
exact to first order in the contrast. Related to this fact is the inability of the variational
estimates of Sections 5 and 6 to discern the effect of the third invariant (determinant) in
statistically isotropic composites with isotropic phases. A new variational procedure making
use of anisotropic linear comparison composites, with the tangent moduli of the phases
evaluated at appropriate estimates for the average strain in the phases, which has the
capability to reproduce the small-contrast results of [98-99] to second-order in the contrast
has been proposed recently by Ponte Castaneda [100]. The accuracy of this new procedure
appears to be much better than that of the variational procedures of Section 5 and 6, although
the resulting estimates for this new procedures are not known to be rigorous bounds.
However, this is a direction which may eventually lead to improved procedures resulting in
second-order bounds in a true Hashin-Shtrikman sense.
The procedure suggested in Section 6 to account for the evolution of microstructure in
composite materials and polycrystals is consistent with the homogenization procedures of
Sections 3 to 5 in the following two senses. First, the procedure accounts for the evolution
of all relevant microstructural variables in the context of Hashin-Shtrikman and self-
consistent descriptions (including up to two-point statistical information) of the effective
behavior of linearly viscous composites with random microstructures of the particulate and
granular type, respectively, although the specific evolution laws for the pertinent
 190 P. Ponte Castaneda

microstructural variables are different for both types of microstructures. In particular, the
procedure generalizes the procedure of Eshelby [34] for particulate composites with non-
dilute concentrations of particles. Second, the approximations that are made in the
characterization of the evolution laws - basically the hypothesis that they are controlled by
the averages of the fields in the various phases of the composite - is consistent with the
Hashin-Shtrikman approximation of piecewise constant polarization fields in the linearly
viscous composite, as well as the approximation of piecewise constant comparison moduli in
the nonlinear homogenization procedures of Sections 5. As a result, the model is expected to
be accurate for the same situations for which the Hashin-Shtrikman and self-consistent
approximations are thought to be accurate, i.e., for particulate composites with low to
moderate volume fractions of inclusions and for composites with granular microstructures,
respectively. Finally, it is emphasized that the resulting constitutive models for
heterogeneous media, fitting within a standard internal-variable framework, can easily be
implemented in the context of a finite-deformation finite element program, which would
allow the numerical analysis of forming processes involving large deformations and
accounting for variations in the microstructural variables (at a length scale that is still large
relative to that of the heterogeneities). These local variations in the local effective properties
of the medium resulting in the possible development of anisotropy and softening may very
well trigger the instabilities that are commonly observed in standard forming processes in the
vicinity of their forming limits. The homogenized constitutive models would of course
become invalid in the presence of such localized instabilities, but they may still serve a useful
purpose by identifying conditions for which the development of these instabilities becomes
unlikely. In conclusion, these constitutive models accounting for the evolution of the
microstructure may prove to be useful tools in the analysis of deformation processes
involving heterogeneous media, given that direct numerical simulations of realistic problems
may either be intractable or impractical.

ACKNOWLEDGMENTS

This article was completed with the support of the National Science Foundation, under grant
No. CMS-9622714.
Nonlinear Composites and Microstructure Evolution 191

REFERENCES

1. Sanchez-Palencia, E. (1980). Non-homogeneous Media and Vibration Theory,

Lecture Notes in Physics, No. 127, Springer-Verlag, Heidelberg.
2. Bensoussan, A., Lions, J.L. and Papanicolaou, G. (1978). Asymptotic Analysis for
Periodic Structure, North-Holland, Amsterdam.
3. Hill, R. (1952). The elastic behavior of a crystalline aggregate. Proc. Phys. Soc. A 65,
349-354.
4. Paul, B. (1960). Prediction of elastic constants of multiphase materials. Trans. of the
ASME 218, 36-41.
5. Hashin, Z. and Shtrikman, S. (1962). On some variational principles in anisotropic and
nonhomogeneous elasticity. J. Mech. Phys. Solids 10, 335-342.
6. Hashin, Z. and Shtrikman, S. ( 1962). A variational approach to the theory of the elastic
behaviour of polycrystals. J. Mech. Phys. Solids 10, 343-352.
7. Hashin, Z. and Shtrikman, S. ( 1963). A variational approach to the theory of the elastic
behavior of multiphase materials. J. Mech. Phys. Solids 11, 127-140.
8. Beran, M. ( 1965). Use ofthe variational approach to determine bounds for the effective
permittivity of a random media. Nuovo Cimento 38, 771-782.
9. Willis, J. R. (1981). Variational and related methods for the overall properties of
composites. In Advances in Applied Mechanics (C. S. Yih, ed.), Academic Press,
New York, 1-78.
10. Willis, J. R. (1983). The overall response of composite materials. ASME J. Appl.
Mech. 50, 1202-1209.
11. Bruggeman, D. A. G. (1935). Berechnung verschiedener physikalischer Konstanten
von heterogenen Substanzen. I. DielektriziUitskonstanten und Leitfahigkeiten der
Mischkorper aus isotropen Substanzen,Ann. Phys. 24, 636-679.
12. Hershey, A. V. (1954). The elasticity of an isotropic aggregate of anisotropic cubic
crystals. ASME J. Appl. Mech. 21, 236-240.
13. Kroner, E. (1958). Berechnung der elastischen konstanten des vielkristalls aus den
konstanten des einkristalls. Z. Phys. 151, 504-518.
14. Hill, R. ( 1965). A self-consistent mechanics of composite materials. J. Mech. Phys.
Solids 13, 213-222.
15. Budiansky, B. (1965). On the elastic moduli of some heterogeneous materials. J. Mech.
Phys. Solids 13, 223-227.
16. Suquet, P. (1983). Analyse limite et homogeneisation. C. R. Acad. Sci. Paris II 295,
1355-1358.
17. Suquet, P. (1985). Elements of homogenization for inelastic solid mechanics. In
Homogenization techniques for composite media, Lecture notes in Physics 272 (E.
Sanchez-Palencia and A. Zaoui, ed.), Springer-Verlag, New-York, 193-278.
18. Marcellini, P. ( 1978 ). Periodic solutions and homogeneization of nonlinear variational
problems, Ann. Mat. Pura Appl. 4, 139-152.
19. Bishop, J. F. W. and Hill, R. (1951). A theory of the plastic distortion of a
polycrystalline aggregate under combined stresses. Phil. Mag. 42,414-427.
20. Bishop, J. F. W. and Hill, R. (1951). A theoretical derivation of the plastic properties of
a polycrystalline face-center metal. Phil. Mag. 42, 1298-1307.
21. Taylor, G. I. (1938). Plastic strains in metals. J. Inst. Metals 62, 307-324.
 192 P. Ponte Castaneda

22. Hill, R. (1965). Continuum micromechanics of elastoplastic polycrystals. J. Mech.

Phys. Solids 13, 89-10 I.
23. Hutchinson, J. W. (1976). Bounds and self-consistent estimates for creep of
polycrystalline materials. Proc. R. Soc. Land. A 348, I 01-127.
24. Berveiller, B. and Zaoui, A. (1979). An extension of the self-consistent scheme to
plastically flowing polycrystals. J. Mech. Phys. Solids 26, 325-344.
25. Gurson, A. L. ( 1977) Continuum theory of ductile rupture by void nucleation and
growth. J. Engng. Mater. Tech. 99, 2-15.
26. Willis, J.R. (1986). Variational estimates for the overall response of an inhomogeneous
nonlinear dielectric. In Homogeneization and Effective Moduli of Materials and
Media (J. L. Ericksen et al., eds.) Springer-Verlag, New York, 247-263.
27. Talbot, D.R.S. and Willis, J.R. (1985) Variational principles for nonlinear
inhomogeneous media. IMA J. Appl. Math. 35, 39-54 ..
28. Ponte Castaneda, P. (1991). The effective mechanical properties of nonlinear isotropic
composites. J. Mech. Phys. Solids 39, 45-71.
29. Ponte Castaneda, P. (1992). New variational principles in plasticity and their application
to composite materials. J. Mech. Phys. Solids 40, 1757-1788.
30. Talbot, D. R. S. and Willis, J. R. (1992). Some explicit bounds for the overall behavior
of nonlinear composites. Int. J. Solids Struct. 29, 1981-1987.
31. Suquet, P. (1993). Overall potentials and extremal surfaces of power law or ideally
plastic composites. J. Mech. Phys. Solids 41, 981-1002.
32. Olson, T. (1994). Improvements on Taylor's upper bound for rigid-plastic composites.
Mater. Sci. Eng. A 175, 15-19.
33. Ponte Castaneda, P. and Zaidman, M. (1994). Constitutive models for porous materials
with evolving microstructure. J. Mech. Phys. Solids 42, 1459-1497.
34. Eshelby, J.D. (1957). The determination of the elastic field of an ellipsoidal inclusion,
and related problems. Proc. R. Soc. Lond. A 241, 376-396.
35. Kailasam, M., Ponte Castaneda, P. and Willis, J. R. (1996). The effect of particle size,
shape, distribution and their evolution on the constitutive response of nonlinearly
visocus composites. Phil. Trans. R. Soc. Land A, to appear.
36. Kailasam, M. and Ponte Castaneda, P. (1996). General three-dimensional constitutive
models for linear and nonlinear media with evolving microstructure. J. Appl. Mech,
submitted for publication.
37. Golaganu, M., Leblond, J.-B. and Devaux, J. (1993) Approximate models for ductile
metals containing non-spherical voids - Case of axisymmetric prolate ellipsoidal cavities.
J. Mech. Phys. Solids 41, 1723-1754.
38. Golaganu, M., Leblond, J.-B. and Devaux, J. (1994) Approximate models for ductile
metals containing non-spherical voids - Case of axisymmetric oblate ellipsoidal cavities.
J. Engng. Mater. Tech. 116, 290-297.
39. Van Tiel, J. (1984). Convex Analysis. Wiley, New York.
40. Ekeland, I. and Temam, R. (1974). Analyse Convexe et Problemes variationnels,
Gauthier-Villars, Paris.
41. Budiansky, B. (1959). A reassessment of deformation theories of plasticity. J. Appl.
Mech. 26, 259-264.
42. Ponte Castaneda, P. and Willis, J. R. (1988). On the overall properties of nonlinearly
viscous composites. Proc. R. Soc. Land. A 416, 217-244.
 Nonlinear Composites and Microstructure Evolution 193

43. Hill, R. (1963). Elastic properties of reinforced solids: some theoretical principles. J.
Mech. Phys. Solids 11, 357-372.
44. Willis, J. R. (1989). The structure of overall constitutive relations for a class of
nonlinear composites. IMA J. Appl. Math. 43, 231-242.
45. Bouchitte, G. and Suquet, P. ( 1987). Charges limites, plasticite et homogeneisation. C.
R. Acad. Sci. Paris I 305, 441-444.
46. Kohn, R. V. and Lipton, R. (1988). Optimal bounds for the effective energy of a
mixture of isotropic, incompressible, elastic materials. Arch. Rational Mech. Anal. 102,
331-350.
4 7. Willis, J. R. (1977). Bounds and self-consistent estimates for the overall properties of
anisotropic composites. J. Mech. Phys. Solids 25, 185-202.
48. Francfort, G. A. and Murat, F. (1986). Homogenization and optimal bounds in linear
elasticity. Arch. Rat. Mech. Anal. 94, 307-334.
49. Milton, G. W. and Kohn, R. V. (1988). Variational bounds on the effective moduli of
anisotropic composites. J. Mech. Phys. Solids 36, 597-629.
50. Willis, J.R. (1978). Variational principles and bounds for the overall properties of
composites. In Continuum Models for Discrete Systems, (J.W. Provan, ed.),
University of Waterloo Press, 185-215.
51. Ponte Castaneda, P. and Willis, J. R. (1995). The effect of spatial distribution on the
effective behavior of composite materials and cracked media. J. Mech. Phys. Solids 43,
1919-1951.
52. Bornert, M., Stolz, C. and Zaoui, A. ( 1996). Morphologically representative pattern-
based bounding in elasticity. J. Mech. Phys. Solids 44,307-3331.
53. Kroner, E. ( 1977). Bounds for the effective elastic moduli of disordered materials. J.
Mech. Phys. Solids 21, 9-17.
54. Milton, G. W. ( 1982). Bounds on the elastic and transport properties of two-component
composites. 1. Mech. Phys. Solids 30, 177-191.
55. Milton, G. W. and Phan-Thien, N. (1982). New bounds on the effective elastic moduli
of two-component materials. Proc. R. Soc. Land. A 380, 305-331.
56. Hashin, Z. (1962). The elastic moduli of heterogeneous materials. J. Appl. Mech. 29,
143-150.
57. Ponte Castaneda, P. ( 1996). Variational methods for estimating the effective behavior of
nonlinear composite materials. In Continuum Models and Discrete Systems (K. z.
Markov, ed.), World Scientific, 268-279.
58. Bhattacharya, K. and Kohn, R. V. (1996). Elastic energy minimization and recoverable
strains of polycrystalline shape-memory materials. Arch. Rational Mech. Anal.,
submitted.
59. Lurie, K. and Cherkaev, A. ( 1986). Exact estimates of conductivity of a binary mixture
of isotropic components. Proc. Roy. Soc. Edinburgh A 104, 21-38.
60. Milton, G.W. (1990). On characterizing the set of possible effective tensors: the
variational method and the translation method. Commun. Pure Appl. Math. 43, 63-125.
61. deBotton, G. and Ponte Castaneda ( 1995). Variational estimates for the creep behavior
of polycrystals. Proc. R. Soc. Land. A 448, 421-442.
62. Rice, J.R. (1970). On the structure of of stress-strain relations for time-dependent
plastic deformation in metals. J. Appl. Mech. 37, 728-737.
63. Mandel, J. (1972). Plasticite et Viscoplasticite. CISM Udine Courses and Lectures,
No. 97, Springer-Verlag, New York.
194 P. Ponte Castaneda

64. Drucker, D.C. (1966). The continuum theory of plasticity on the macroscale and the
microscale. J. Mater. 1, 873-910.
65. Sachs, G. (1928). Zietschrift Verein. Deut. Ing. 72, 734-736.
66. deBotton, G. and Ponte Castaneda (1992). On the ductility of laminated materials. Int.
J. Solids Structures 29, 2329-2353.
67. Tandon, G. P. and Weng, G. J. (1988). A theory of particle-reinforced plasticity. J.
Appl. Mech. 55, 126-135.
68. Gilormini, P. (1995). A shortcoming of the classical nonlinear extension of the self-
consistent model. C. R. Acad. Sci. Paris, Serie lib 320, 115-122.
69. Suquet, P. (1996). Overall properties of nonlinear composites. In Micromechanics of
Plasticity and Damage of Multiphase Materials (IUTAM Symposium) (A. Pineau
and A. Zaoui, eds.), Kluwer, New York, to appear.
70. Suquet, P. (1995). Overall properties of nonlinear composites: a modified secant
modulus theory and its link with Ponte Castaneda's nonlinear variational procedure. C.
R. Acad. Sci. Paris, Serie lib 320, 563-571.
71. Suquet, P. ( 1996). Overall properties of nonlinear composites: Secant moduli theories
and variational bounds. In Continuum Models and Discrete Systems (K. Z. Markov,
ed.), World Sr.ientific, 290-299.
72. Michel, J.C. and Suquet, P. (1992) The constitutive law of nonlinear viscous and
porous materials. J. Mech. Phys. Solids 40, 783-812.
73. Buryachenko, V.A. (1990). Effective characteristics of elastic physically nonlinear
composites. Prikladnaya. Mekh. 26, 9-13.
74. Qiu, Y.P. and Weng, G.J. (1992). A theory of plasticity for porous materials and
particle-reinforced composites. J. Appl. Mech. 59, 261-268.
75. deBotton, G. and Ponte Castaneda, P. (1993). Elastoplastic constitutive relations for
fiber-reinforced solids. Int. J. Solids Structures 30, 1865-1890.
76. Li , G. and Ponte Castaneda, P. (1993). The effect of particle shape and stiffness on the
constitutive behavior of metal-matrix composites. Int. J. Solids Struct. 30, 3189-3209.
77. Bonnet, G., Dendievel, R. and Willis, J. R. (1991 ). New bounds for polycrystalline
materials. In Constitutive relations for finite deformation of polycrystalline
materials. (R. Wang and D. C. Drucker, eds.), Springer, New York, 77-86.
78. Dendievel, R., Bonnet, G. and Willis, J. R. (1991). Bounds for the creep behaviour of
polycrystalline materials. In Inelastic Deformation of Composite Materials (G. J.
Dvorak, ed), Springer-Verlag, New York, 175-192.
79. Ponte Castaneda, P. (1991). Effective properties in power-law creep. In Mechanics of
Creep Brittle Materials 2, edited by A.C.F. Cocks and A.R.S. Ponter, Springer, New
York, 218-229.
80. Willis, J. R. (1991 ). On methods for bounding the overall properties of nonlinear
composites. J. Mech. Phys. Solids 39, 73-86. ·
81. Ponte Castaneda, P. and deBotton, G. (1992). On the homogenized yield strength of
two-phase composites. Proc. R. Soc. Lond. A 438, 419-431.
82. Ponte Castaneda, P. (1991). The effective properties of brittle/ductile incompressible
composites. In Inelastic Deformation of Composite Materials (G. J. Dvorak, ed),
Springer-Verlag, New York, 215-231.
83. Talbot, D. R. S. and Willis, J. R. (1991). The overall behavior of a nonlinear fiber
reinforced composite. In Inelastic Deformation of Composite Materials (G. J.
Dvorak, ed), Springer-Verlag, New York, 527-545.
 Nonlinear Composites and Microstructure Evolution 195

84. Ponte Castaneda, P. and Zaidman, M. (1996). The finite deformation of nonlinear
composite materials-!. Instantaneous constitutive relations. Int. J. Solids Struct. 33,
1271-1286.
85. Dvorak, G. J. and Bahei-El-Din, Y. A. (1987). A bimodal plasticity theory of fibrous
composite material, Acta Mechanica 69, 219-241.
86. Moulinec, H. and Suquet, P. (1994). A FfT-based numerical method for computing the
mechanical properties of composites from images of their microstructures. In JUTAM
Symposium on Microstructure-Property Interactions in Composite Materials (R.
Pyrz, ed.), Kluwer, 235-246.
87. Dvorak, G. J., Bahei-El-Din, Y. A., Macheret, Y. and Liu, C. H. (1988). An
experimental study of elastic-plastic behavior of a fibrous Boron-Aluminum composite,
J. Mech. Phys. Solids 36, 655-687.
88. Bilby, B.A. and Kolbuszewski, M.L. (1977). The finite deformation of an
inhomogeneity in two-dimensional slow viscous incompressible flow. Proc. R. Soc.
Lond. A 355, 335-353.
89. Budiansky, B., Hutchinson, J.W. and Slutsky, S. (1982). Void growth and collapse in
viscous solids. In Mechanics of Solids, The Rodney Hill Anniversary Volume, (H.G.
Hopkins and M.J. Sewell, eds.), Pergamon Press, Oxford, 13-45.
90. Kailasam, M. and Ponte Castaneda, P. (1996). Constitutive relations for porous
materials: the effect of void shape and orientation. In Micromechanics of Plasticity and
Damage of Multiphase Materials (IUTAM Symposium) (A. Pineau and A. Zaoui,
eds. ), Kluwer, New York, to appear.
91. Zaidman, M. and Ponte Castaneda, P. (1996). The finite deformation of nonlinear
composite materials-H. Evolution of the microstructure. Int. J. Solids Struct. 33,
1287-1303.
92. Asaro, R.J. (1983). Micromechanics of crystals and polycrystals. In Advances in
Applied Mechanics (1. W. Hutchinson and T. Y. Wu, eds.) Academic Press, New
York, 1-114.
93. Bassani, J. L. (1994). Plastic flow in crystals. In Advances in Applied Mechanics (J.
W. Hutchinson and T. Y. Wu, eds.), Academic Press, New York, 191-258.
94. Wang, P. T. (1988). Powder Technology 54, 107-118.
95. Mear, M. and Lee, B. J. ( 1994) Studies of the growth and collapse of voids in viscous
solids. J. Engng. Mater. Tech. 116, 348-358 ..
96. Talbot, D. R. S. and Willis, J. R. (1994). Upper and lower bounds for the overall
properties of a nonlinear composite dielectric. I. Random microgeometry. Proc. R. Soc.
Lond. A 447, 365-384.
97. Willis, J. R. ( 1994). Upper and lower bounds for nonlinear composite behavior. Mater.
Sci. Engng. A 175, 7-14.
98. Suquet, P. and Ponte Castaneda, P. (1993). Small-contrast perturbation expansions for
the effective properties of nonlinear composites. C. R. Acad. Sci. Paris, Shie II 317,
1515-1522.
99. Ponte Castaneda, P. and Suquet, P. (1995). On the effective mechanical behavior of
weakly inhomogenous nonlinearly materials. Eur. J. Mech. A/Solids 14, 205-236.
100. Ponte Castaneda, P. ( 1996). Exact second-order estimates for the effective mechanical
properties of nonlinear isotropic composite materials. J. Mech. Phys. Solids 44, 827-
862.
 EFFECTIVE PROPERTIES OF NONLINEAR COMPOSITES

P. Suquet
L.M.A./ C.N.R.S., Marseille, France

Abstract: These lectures describe several procedures commonly used or recently devel-
oped to predict the overall behavior of nonlinear composites from the behavior of their
individual constituents and from statistical information about their microstructure. Se-
cant methods are discussed in section 2. A modified method based on the second-order
moment of the strain field is proposed and compared with the classical secant method
in specific situations, composites with large or small contrast and power-law materials.
The incremental method is presented in section 3. It appears much stiffer than the
two secant methods. Its predictions for isotropic two-phase power-law composites even
violate a rigorous upper bound when the nonlinearity is strong. A variational procedure
leading to rigorous upper bounds for the effective potential of the composite is presented
in section 4. Specific forms for voided or rigidly reinforced power-law composites are
given first. Then a general upper bound applying to a general class of nonlinear compos-
ites is derived. The variational procedure coincides with the secant approach based on
second-order moments and with the variational procedure of Ponte Castaneda. These
different schemes are applied in section 5 to predict the overall behavior of metal-matrix
composites. A simplified model based on the variational procedure is proposed. Its pre-
dictions compared well with simulations performed by the Finite Element Method.
 198 P. Suquet

1 :Definitions and general principles governing the

effective properties of nonlinear composites
1.1 Outline
These lectures are devoted to the prediction of the effective properties of nonlinear
composites. The term composite refers to any nonhomogeneous material whatever the
nature of its individual constituents. For instance a metal containing voids is considered
a composite since it is composed of two different phases, the metal matrix and the voids.
Similarly a polycrystal is considered a composite since all families of grains with given
shapes and given orientations can be treated as different phases.
Composites are composed of N individual constituents with a known behavior and
perfectly bonded at interfaces. It is further assumed that statistical information about
their microstructure is available. The aim of the theories that will be developed in this
study is to combine the two types of information, behavior of the individual constituents
and microstructure, to predict the effective behavior of the composite. Beyond the
effective properties, it is often useful to have some insight into the local mechanical
state of the individual phases. This is an even more difficult problem, somewhat related
to that of the effective properties, that we will try to address.

There is an enormous body of literature devoted to the case in which the individual
phases forming the composite have a linear behavior. Some of these results will be
used here. They are mainly related to the Hashin-Shtrikman bounds ( [21]), the self-
consistent scheme (HILL [24]), and their various extensions (see for instance WALPOLE
[60], WILLIS [61], MILTON and KOHN [34] and BORNERT et al [4]). It is assumed that
the reader is familiar with these linear theories.
The topic of nonlinear composites has also received great attention. To cite just
a few contributions : the bounds of BISHOP and HILL [3] for polycrystals, the varia-
tional approach to the strength of composite materials developed by DRUCKER [11],
the self-consistent analysis of HILL [24] [25] and HUTCHINSON [26] for composites
and polycrystals, the secant method used by CHU and HASHIN [9], BERVEILLER and
ZAOUI [2] and TANDON and WENG [59]. But it is only recently that variational pro-
cedures have been systematically developed after the pioneering studies by TALBOT
and WILLIS [56] and WILLIS [62]. These studies extended the Hashin-Shtrikman varia-
tional procedure to nonlinear composites by using a homogeneous comparison material.
Then PONTE CASTANEDA, in a series of papers starting from [39], introduced a non-
homogeneous reference material which made it possible to use linear bounds different
from the Hashin-Shtrikman bounds. SUQUET [51], also using a nonhomogeneous lin-
ear comparison material, proposed a derivation of the bounds for power-law materials,
leading to simpler results than the previous ones. OLSON [38] used a similar method
for polycrystals.
 Effective Behavior of Nonlinear Composites 199

These lectures are organized as follows. Section 1 recalls the main formulations
of the constitutive law of nonlinear constituents. Section 2 investigates the secant
methods. The classical one is based on the first-order moment of the strain in each
individual constituent. A modified method, based on the second-order moment is pro-
posed following SUQUET [52] {theories based on a similar principle have been proposed
by QIU and WENG [47] and BURYACHENKO [7]). Section 3 examins the incremental
method. Section 4 presents a variational procedure. The coincidence between this
variational procedure, the modified secant method and PONTE CASTANEDA's varia-
tional procedure is proven. In section 5 the different methods are applied to the case of
Metal Matrix Composites (MMC) and their accuracy is assessed by comparison with
cell calculations performed through the Finite Element Method (FEM).

Notations : Barred letters IL, M, A refer to fourth order tensors and bold-face letters
rr, e, u refer to first or second order tensors. Their components are denoted with
ordinary letters Lijkh, a;j, u;.

1.2 Description of phases.

Consider a representative volume element (r.v.e.) V of the composite material where
the inhomogeneities are small compared to V. The composite is made up of N homo-
geneous phases V,., r = 1, .. , N. x(rl denotes the characteristic function of phase r, c(r)
its volume fraction, < . >r the spatial averaging over V,., < . > the spatial averaging
over V
1 { ( )d
W (r) -<
- -
W >r- IV..I Jvr W X X,

1 [ N
W =< w >= -lVI Jv w(x) dx = I:c(r) < w >r.
V r=l

The individual constituents are nonlinear. As schematically depicted in Figure 1, their

individual behavior under uniaxial tension can be described in several ways within the
context of small strains :

Secant formulation
a= E.(c) c,
Incremental formulation

Energetic formulation

'lj.J(a) = 1" eda.

These different formulations can be extended to triaxial states of stress and strain.
Most of the present study (except sections 3 and 5) deals with nonlinear hyperelastic
 200 P. Suquet

300 ,_--------~----------~----.

(]"

100 <p(c)

0.02 0.04

Figure 1: Different formulations of the nonlinear behavior of individual phases

materials within the context of small strains. These materials are characterized by two
dual potentials, a strain energy density r.p, or equivalently a stress energy density ~'
such that
a~
e = ou(u), r.p(e) + ~(u) = u: e. (1)

The potentials r.p and ~ are assumed to be convex and with a sufficient growth at
infinity so that classical results in homogenization apply (MARCELLINI [30]). Because
of the correspondence between the equations governing steady-state creep and those
governing nonlinear elasticity at small strains, the above relations apply equally well to
the case where all individual constituents are purely viscous, provided that all strains
involved in the problem are converted into strain-rates (the assumption of infinitesimal
strains consequently drops).

1.3 A class of nonlinear composites

In most of this study the constitutive law of the individual constituents can be put as

(2)
Effective Behavior of Nonlinear Composites 201

where Um and em are the first invariants of the stress and strain, s and e are the stress
and strain deviators,
1 1
Um = 3 tr(u), em= 3 tr(e),

Sij = Uij - UmOij, e;j = Cij - emOij,

Ueq and eeq are the Von Mises equivalent stress and equivalent strain

J.ls is the secant shear modulus

Ueq
J.ls=--.
3eeq
The response of the phases is therefore assumed to be linear for purely hydrostatic
loadings and nonlinear in shear. This assumption is made to simplify the developments
but is by no means a restriction of the methods described below. The constitutive law
(2) derives from the potential

rxl/2
J(x) = Jo 3J.t 8 (e)e de. (3)

Note that
J.ls(eeq) = ~f'(e~q), (4)
where f' denotes the derivative off with respect to its scalar argument x.

1.3.1 Secant formulation

The constitutive law (2) can be alternatively written in a compact form

u = L.(e): e, (5)

with
(6)
L. is the secant stiffness tensor. The relation (6) shows its decomposition on the two
projectors J and lK:

Jiikh = ~o;iokh, K;jkh = ~(o;koih + o;hoik)- Jijkh, }

(7)
J :J = J, K : K = IK, K : J = 0, J +K = n,
where nis the identity in the space of symmetric fourth order tensors.
202 P. Suquet

Remark 1: lL8 is not uniquely defined (GILORMINI [18]). One can add to this isotropic
tensor an anisotropic contribution which does not modify the constitutive relation. Set

e
e=-.
A

Ceq

Then lL~ : e = lL8 : e for any a and lL~ can also serve as a secant tensor satisfying (5) .
However lLs is the only isotropic tensor satisfying the constitutive relation.

1.3.2 Incremental formulation

By differentiation of (5) with respect to time, the constitutive law (2) takes the incre-
mental form
(8)
with
e
e-- (9)
A

-
ceq
'
where p' and f" denote the first derivative of J.L and the second derivative of f with
respect to their scalar arguments ceq and x. With notations introduced by PONTE
CASTANEDA [42], 1L1 can be re-written as

lLt = 3k .If + 2p1 IF + 2A E, (10)

where
2
E = 3e0 IF=K-E, (11)
A A

e,
IE and IF are orthogonal projectors. They are also orthogonal to J.

E : E = E, IF : IF = IF, E : IF = 0, E : J = 0, IF : J = 0.

The tangent stiffness lL1 of the phases is anisotropic in general (except when the strain
is purely spherical). It differs from the secant tensor when the material behavior is
nonlinear:

1.4 Power-law materials

We shall often make reference to the specific class of incompressible power-law materials
for the purpose of comparing predictive schemes. The two potentials <p and 'ljJ read in
this case
m+l
(e)= O'oco ( ceq ) when tre = 0, r.p(e) = +oo otherwise, (12)
'P m +1 co
 Effective Behavior of Nonlinear Composites 203

and
n+l
t/J (tT) = O"oco ( 0"eq )
(13)
n +1 cro
The constitutive law reads
s 2 e = (Ceq)m
tT = O"ml + s, - =--,
O"eq 3 Ceq
O"eq
cro co
(14)

1
The exponents m and n are related by m = -.
n
- When m = n = 1 the above potentials describe the behavior of a linear incom-
pressible material with shear modulus p, = !!.E._.
3co
- When m = 0, n = +oo, the potentials describe the behavior of a rigid-plastic
material with flow stress cr0 •

In most applications considered here the exponent m varies between 0 and 1. The
secant tensor IL. given by (6) and the incremental tensor ILt given by (10) read

(15)

and
ILt = +oo J + 2p,t(ceq) (lF + mlE), P,t(ceq) = p,.(ceq), A= ffi/1-t· (16)
Remark 2 : A straightforward calculation shows that the tensor .;;- Lt can also be used
as a secant tensor to define the constitutive law (14).

1.5 The local problem. Boundary conditions

The potentials r.p and t/J depend on the position x inside the r.v.e. and take constant
values r.p(r) and .,p(r) in each individual phase (r):
N
I: r.p(r)(e) x(r)(x),
N
r.p(x, e)= t/;(x,u) =I: .,p(r)(u) x(r)(x). (17)
r=l r=l

The local stress and strain fields within the r.v.e. solve the so-called local problem
consisting of the constitutive equations (1 ), compatibility conditions met by e, and
equilibrium equations satisfied by u :
1
e = 2 (Y'u + V'u 1) , div(u) = 0. (18)

The problem is ill-posed in absence of boundary conditions. Classically three classes

of boundary conditions can be considered on av:
204 P. Suquet

uniform tractions
u(x).n(x) = I:.n(x) on av, (19)
affine displacements
u(x) = E.x on 8V, (20)
or periodicity conditions

u(x) = E.x + u*(x), u* #, u.n - #, (21)

where the notations # and -# denote periodic and anti-periodic fields respectively.
Only one set of boundary conditions can be applied on 8V. ·The difference between
the three types of boundary conditions will not be discussed here (see SUQUET [49] for
a more detailed discussion). The assumption that the potentials cp and '1/J are convex
allows us to consider that the three types of boundary conditions are equivalent when
the r.v.e. is large with respect to the inclusions (the limiting case of rigid-plastic
materials requiring a specific treatment as in BOUCHITTE and SUQUET [5]). The
overall stress and strain are the averages of the local stress and strain fields (with usual
extensions in the case of voided or rigidly reinforced materials) :

I: =< u >, E =< e > .

1.6 Effective potentials

The solutions u and u of the local problem (1) (17) (18) (20) have the following._.
variational properties
< cp(e(u)) >= inf < cp(e(v)) >, (22)
ve.qE)

where
X::(E) = {v = E.x on 8V}.
Alternatively (assuming that {19) and {20) are equivalent in the limit of a large r.v.e.)

< '1/J(u) >= inf < '1/J(T) >, (23)

TES(E)

where
S(I:) = {T, div(T) = 0 in v, T.n = I:.n on av}.
The overall behavior of the composite is also governed by two potentials, related to the
infima in (22) and (23). Define
\li(I:) = TES(E)
inf < '1/J(T) >, <I>(E) = vEX:(E)
inf < cp(e(v))) >, (24)

Then {WILLIS (62] for instance),

aw 8<1>
E = 8I:(I:), I:= 8E(E).
Effective Behavior of Nonlinear Composites 205

Choosing v = E.x in (22) and T = ~ in (23) yields the classical bounds of Voigt and
Reuss
<I>(E) ~< cp >(E) (Voigt), Ill(~)~< 1/J > (~) (Reuss). (25)
In the case of power-law materials the Reuss and Voigt bounds read

0'0R co ( ceq )m+l ~ <I>(E) ~ O'oV co ( Ceq )m+l ,

m+l ~ m+l ~

When m = 0 (rigid-plastic materials) the Reuss bound simplifies to

0'~ = inf O'~r).

r=l, .. ,N

2 Secant methods
All materials considered in this section are governed by the constitutive law (2).

2.1 General procedure

Recall that the local stress and strain fields solve the following problem

e: =! (Vu + \7u 1), div(u) =0, +boundary conditions, }

u(x) = IT...~r)(x): e:(x) in phase r,

with
IT....(x) = 3k(x) .If+ 2fl,(x) IK,
k(x) = k(r), fls(x) = flirl(ce 9 (x)) in phase r.
In a nonlinear composite, the secant modulus fls varies from one point to another
even iri the same phase. The nonlinear composite behaves like a linear composite with
infinitely· many phases with elastic moduli ( k(r), fls( X)). The spatial fluctuations of
these moduli depend strongly on the type of nonlinear behavior of the phases and on
the quantity of overall strain which is applied. This statement is illustrated in Figure 2
where the response of a nonlinear matrix reinforced by aligned elastic fibers is simulated
numerically (MOULINEC and SUQUET [35] [37]). The two-dimensional cross-section of
the composite is shown in Figure 2 a. The unit cell is periodic both in the horizontal
and in the vertical directions. The maps of equivalent strain (Figure 2 b,c,d) show
that the strain fields depend significantly on the type of behavior of the matrix. When
the matrix is purely elastic, the strain can be reasonably considered homogeneous
in the matrix (Figure 2 b). When the matrix is elastic ideally plastic, the plastic
strain concentrates in very thin zones (white zones in Figure 2 c). The secant moduli
206 P. Suquet

are therefore very low in these highly deformed regions, whereas they are comparable
to the elastic modulus in the rest of the unit cell. The case of an elastic plastic
matrix with linear isotropic hardening shown in Figure 2 d is intermediate between the
two preceeding extreme cases. The secant modulus does fluctuate (according to the
fluctuations in the strain field), but does not exhibit the abrupt variations of the ideally
plastic case. As illustrated in this example and as recognized by HILL [25] forty years
ago, it is impossible to analytically determine the local response of these nonlinear
composites. An approximation has to be introduced to make analytic calculations
more feasible.
The approximation for secant methods consists in replacing the secant tensors IL.(x)
by tensors which are constant over each individual phase :
IL.(x) = JL(r) in phase r.
In addition, it is reasonable to assume that the tensors JL(r) are given by the relations
(6) evaluated at some "effective strain" E~JJ
][)r) = JL(r)(E(r) )
s ef f •

E~'if is in charge of picking up the main features of the strain field in phase r. Assuming
for a while that this effective strain is known, the approximate version of the local
problem reads

e =~ (\?u + \7u 1) , div(u) = 0, u(x) = JL(r) : e(x) in phase (r), boundary conditions.

Given the tensors JL(r), the above problem is now a classical one for a linear elastic
material. Note however that these tensors themselves depend on the strain field e
(through the effective strain E~j1 ) which itself depends on the tensors JL(•),s = l...N.
The problem is therefore nonlinear, but instead of having infinitely many nonlinear
problems to solve (at each point x), there are "only" N nonlinear problems to solve,
corresponding to the N individual phases :
JL(r) = IL~r) (E~'if ), E~'if = function of JL(•), s = 1, ... , N.
Once these N nonlinear problems are solved, the stiffness tensors JL(r) and the effective
strains E~JJ of each phase are known. Both depend nonlinearly on the overall strain E.
The corresponding linear composite is sometimes referred to as the linear comparison
composite. The effective stiffness JLhom of this linear composite can then be estimated
by a linear predictive scheme appropriate to the microstructure of the composite.
JLhom _
= JLhom(c(•) , JL(•)I s-J..,N ).
Finally, the overall constitutive law then reads
:E = JLhom(E): E,
where the dependence of the overall secant tensor JLhom on the overall strain E (through
E~'if and JL(r)) has been emphasized. Only the "effective strain" E~'if remains to be
defined.
Effective Behavior of Nonlinear Composites 207

(a) (b)

(c) (d)

Figure 2: Composite reinforced by continuous aligned fibers. Numerical simulations.

(a) : cross section of the composite. Maps of the equivalent strain ceq for different
behaviors of the phases. (b) : both phases are elastic (fibers E(!) = 400000 MPa,
v(ll = 0.23, matrix E( 2 ) = 68900 MPa, v( 2 ) = 0.35). (c) : the matrix is elastic- ideally
plastic (Von Mises flow theory a0 = 68.9 MPa) and the fibers are elastic. (d): the
matrix is elastic plastic with linear isotropic hardening (flow theory) and the fibers are
elastic. Th e applied loading is transverse uniaxial tension (generalized plane strains).
Overall strain Eu = 1%. 0% strains are in black, 10% strains (and more) are in white.
 208 P. Suquet

2.2 The classical secant method: a theory based on average

strains
In the most commonly used secant method (called classical here), the effective strain
of phase ( r) is the average strain of this phase
E(r) = < e >r.
The average strain in the linear comparison material can be obtained by means of the
strain concentration tensor
e(x) = A(x) : E, E(r) = A(r) : E, A(r) =< A >r .
Although it was not made explicit that the problem is nonlinear, it is clear that the
strain concentration tensor A(r) depends on the stiffness of all phases, which themselves
depend on the local strain and thus on the strain concentration tensors. Therefore the
classical secant method involves the following steps:
1. A linear theory providing an expression for [}om and the N strain concentration
tensors in each individual phase as functions of the secant tensors ][_.{•) ls= 1, ... N and
of the microstructure.
2. The resolution of N nonlinear problems for theN unknown secant tensors JL(r)(E(r)):
JL(r) = IL~rl(E(rl), E(r) = A(r) : E, A(r) = A(rl(JL(•l, s = l...N) (26)

3. Once the N nonlinear problems (26) are solved, the overall stress-strain relation
is given by :
(27)
When the constitutive relations of the individual constituents are given by (2), the
secant modulus depends only on the equivalent strain J.Ls(C:eq) and the quantity needed
to measure the effective strain of the individual phases is indeed a scalar
1/2 t E (r)
E(r) = ( 2t:'(r) £(:)/3 ) f(r) = E(r) - _r_ _ l
eq •1 •J ' 3 .
The secant modulus in phase r is then given by J.L(r) = J.Lir)(E~;>).

2.3 Two-phase isotropic composites

The strain concentration tensors in two-phase composites are determined by the effec-
tive stiffness JLhom (this is a straightforward consequence of a result derived by LEVIN
[29])
_A(l) = c~1) (JL(l) - ][)2l) -1 (ILhom -JL(2))' }
(28)
A( 2 l = ) 2 > (IIPl - n:.P>r 1 (II}om - II.Pl) .
 Effective Behavior of Nonlinear Composites 209

The nonlinear system of equations to solve in the classical secant procedure reads

E(l) = )~l (ll.Pl- n.Plrl (Lhom- L(2l) : E,

E(2) = c~2) (ILPl- L(ll) -I (Lhom - L(Il) : E, (29)

[.,(I) = L~I)(E(ll), [.,(2) = L~2\E(2l), [.,hom= Lhom(L(I) [.,(2) c(ll)·

' , '
When the individual phases are isotropic and when the composite has overall isotropy,
all stiffness tensors take the form (6). The deviators of the average strain in each phase
are colinear to the deviator £ of the overall strain

£' = £'(1) = £'(2) £'(r) _ £(r) ' £

' - (r)' £=-E. (30)
Eeq eq

The systems of nonlinear equations (29) reduce to

_1_ khom - k(2)

(I) - 1 khom k(l)
(2)__ -
am - c(l) k(I) - k(2) ' am - c(2) k(2) - k(I) '

(2)-
1 ,..,11hom -,..,11(1) (31)
aeq - c(2) p(2) - pP) ,

If in addition the phases are incompressible, these nonlinear equations can be simplified
to give
(I)£
Eeq(I) -_ aeq £(2) _ (2)£
eq, eq - aeq eq,

(32)

1-lhom = 1-lhom (p (I), 1-1(2), c(l l).

Any linear theory can be used to compute the overall elastic stiffness [.,hom of the
comparison material. Some of them are recalled below but the method is not limited
to these specific schemes.
 210 P. Suquet

2.4 Linear predictive schemes

2.4.1 Hashin-Shtrikman and Willis bounds
The Hashin-Shtrikman bounds provide rigorous upper and lower bounds for the linear
effective properties of composites. Initially developed for isotropic phases with overall
isotropy ([21]), they were extended to anisotropic media arranged according to an
"ellipsoidal symmetry" (WILLIS [61]). For a two-phase composite with "well-ordered"
phases, the Hashin-Shtrikman (or Willis) bound reads

where
(34)

with
K;~l = L~]kh~i~h, N( 2J = II~pl- 1 , Hg1h(e) = N;~l~i~hl(ii)(kh)·
e.z.e = 1 is the ellipsoid defining the "ellipsoidal symmetry". The expression (33)
is a lower bound when phase 1 is stiffer than phase 2 (IL.Pl 2: [)2l, inequality in the
sense of energies). It is an upper bound when phase 2 is stiffer than phase 1. The
tensor JP( 2) has the diagonal symmetry of an elasticity tensor. It is therefore ea~icr
to use than the Eshelby tensor §( 2 ) of the ellipsoidal inclusions of phase 1 in phase
2 to which it is related by JP'( 2l = §( 2) : M( 2l. M( 2l = (ILPlr 1 is the compliance of
phase 2. The Hashin-Shtrikman bounds also provide reasonably good estimates of
the overall linear properties of two-phase isotropic composites made of inclusions of
a first phase dispersed in the second phase which plays the role of a matrix. The
inclusions are assumed to have an ellipsoidal shape and to be arranged according to
an ellipsoidal symmetry. When both the shape and the arrangement are defined by
the same ellipsoid the adequate estimate is given by (33) (WILLIS [61]). When they
are defined by different elliptic forms there exists a more general estimate given by
PONTE CASTANEDA and WILLIS [46]). The present study is limited to arrangement
of inclusions that has the same symmetry as the inclusions themselves.
When the individual constituents are isotropic and when the composite has overall
isotropy, the tensor JP>( 2) reduces to

JP>(2J = _1_ .IT + _1_ K,

3kp 2{lp
(35)
5{l( 2 ) 3k( 2 ) + 4{l< 2 l
2{lp = -3- k(2) + 2{l(2)

and the Hashin-Shtrikman lower bound (assuming for definiteness that phase ( 1) 1s
Effective Behavior of Nonlinear Composites 211

stiffer than phase (2)) can be expressed as

ll...HS- = 3kHS_ .JJ + 2f1HS- K, )

k(!) - k(2)
kHS_ = k(2) + C(l) f1HS- = /1(2) + C(l) /1(1) - /1(2)
I 2 .
(2) k(l) - k( 2)' (2) /1( ) - fl( )
1
+c kp 1 +c f1p
(36)
Remark 3: Unidirectional composites, made up of phases which are invariant by trans-
lation along a given direction (taken to be the third direction here), are also of practical
interest. Limiting the attention to transversally isotropic unidirectional composites un-
dergoing plane strains, note that two in-plane effective constants of the composite can
be bounded by expressions in the form (36). Define in dimension d ( d = 2 or 3)

- E;;
Em-d' £ = E -m,
E I

where the repeated indices i and j vary between 1 and d. Then the elastic constitutive
law of an isotropic composite (d = 3) or a transversally isotropic composite under plane
strains (d = 2) can be written as

2 £
:Eeq J Eeq'
khom and f1hom denote the bulk and shear moduli (d = 3) or in-plane bulk and shear
moduli (d = 2) of the composite. The Hashin-Shtrikman lower bounds for unidirec-
tional composites (phase ( 1) is assumed to be stiffer than phase (2)) can be put in the
form (36) with

(37)

The corresponding lower bounds are appropriate to predict the effective behavior of a
matrix (phase (2)) containing dispersed fibers of phase ( 1). Both cases (2-dimensional
(35) and 3-dimensional (37)) can be put in the more compact form
k(2) k(2) 2 k( 2 ) + 2J1( 2 )
kp=-, = -----,--,----,---
{3 = 2 + d k(2) + 2 ( dd!) /1(2).
0'
0' k(2) + 2 (ddt) /1(2)'
(38)

2.4.2 Bounds for composite sphere assemblages

When the composite microstructure is an assemblage of composite spheres, bounds
which are tighter than the Hashin-Shtrikman bounds have been proposed by HASHIN
212 P. Suquet

[19] and improved by HERVE, STOLZ and ZAOUI [22] (see also BORNERT et a! [4] and
ZAOUI [64]). The refined bounds of [22] for isotropic two-phase composite spheres
assemblages read (for definiteness it is assumed that J.l(I) ~ J.l( 2))

where J.ll_! 5 is given by (36)(35) and

(39)
F ( c, I' /O) = ( 2/3) 2
2 1 c 1-c
5(1- c)+ :.y=-1- 10 19(1- 1) 713 10 25
- c +- + -.,.-------,-
21 16 + 191 21 24 (/ - 1)
0

2.4.3 Estimates by self-consistent schemes

The linear effective propPrties of "disordered" composites can be estimated by self-
consistent schemes. The classical self-consistent scheme is appropriate for phases that
are "perfectly disordered" (in the sense of KRONER [28]). The effective moduli are
solutions of the systems of two coupled equations

SC _ (2) (I) J.l(l)- J.l( 2)

J.l - J.l +C (I) (2) . (40)
l+c(2)fl -p.
J.l.~c

k$0 and J.t$0 are given by (38) where k( 2) and J.l( 2) are replaced by the unknown effective
properties k 50 and p. 50 . When the composite is a disordered assemblage of composite
spheres (spherical particles of phase 1 in a spherical shell of phase 2), the generalized
self-consistent scheme of CHRISTENSEN and Lo [8] is appropriate to estimate the ef-
fective properties. If we limit our attention to incompressible individual phases and to
overall isotropy, the shear modulus J.t 3"' predicted by this thre~-phase model is obtained
by solving the following implicit equation (HERVE, STOLZ and ZAOUI [22]):

3<p _
!!:...._ ( 3<p)
!!:...._
1 +c
(I) (I)
J.l( 2) - F c , ,, J.l( 2) . (41)

2.5 Classical secant method. Power-law materials

The composites considered in this section are two-phase power-law materials with over-
all isotropy. The individual constituents are governed by the relations (2). They have
the same exponents n and m and the same strain c: 0 but different flow stresses a 0 . The
Effective Behavior of Nonlinear Composites 213

Hashin-Shtrikman bound (36) is used. Detailed calculations are presented here for 3-
dimensional isotropic composites. Most of the results are also valid for unidirectional
composites (or can be easily extended).

The nonlinear problem (32) can be further simplified by using the homogeneity of
degree m of the constitutive law and the explicit expression of the secant modulus Jl$
given in (15) . The nonlinear equations reduce to the scalar equation

E( 1)
eq
1
= -----:--;:~-:- (E (1))
Eeq
(2)
m-1
(42)
E(2) ( (1) ) •
1 + {3 ~(2) - 1
eq
eq

where (3 = _2_d. This equation can be rearranged into

2+
E(1) (1) ( E(l)) m
( 1 -fJ) ~~) +f3(112) ~~) = 1. (43)
Eeq CT0 Eeq

Remark 4 : Interestingly, the volume fractions of the phases do not enter the above
nonlinear equation. In other words, the classical secant method predicts that the ratio
between the average strains in the two phases is independent of the volume fraction of
the phases.

Equation (43) cannot be solved in a closed form, except in specific cases (see below).
It has to be solved numerically. Although more refined numerical methods can be used,
E~!l J.l(l) .
a straightforward substitution gives satisfactory results. E( 2 ) and ('2) bemg known
eq Jl
at iterate i, their values at iterate i + 1 are deduced from the relations (42) and this
process is repeated until convergence is reached.
For the purpose of comparison between the classical secant theory and other theo-
ries, it will be useful to consider four limiting cases corresponding to an infinite contrast
between phases (rigid inclusions or voids in a deformable matrix), to rigid-plastic phases
(m = 0) with arbitrary contrast and to phases with a small contrast.

2.5.1 Rigid inclusions

When phase 1 is rigid, the nonlinear equation (42) and the average condition

c( 1 ) E(ll
eq
+ c( 2) E(eq2) = Eeq
can be combined to show that

1
lim E(l)
eq
= 0' .
IIm E(2)
eq = (2)•
Eeq
1
.
1Im (1) (E(l))m
CTo eq
= _!._aCTo(2) (E(2))m
eq •
a~ )-Hoo a~1)-t+oo C a~ )-Hoo fJ
214 P. Suquet

The deviator of the overall stress is

where, using (30)

S(1) _
-
1.
1m
~ (1)
a0
( Eeq(1))m c "'
u~') -Hoo 3 co

Finally
2 hom (E )m-1
s = -~ __!!!.. £,
3 co co
with
( 1 + _2d c(ll)
17 hom _ 17 (2) -'----:-:-:':::-~ (44)
o - o - c(2)m

2.5.2 Porous materials.

When the inclusions are voids, the effective properties of the linear comparison com-
posite are charactenzed by a bulk and a shear modulus

khom- 2(d- 1) c(2) (2) 1-"hom = 1-"(2) (--c-;;(2,....)- ) (45)

- d c( 1) I" ' 1 + Lc(l)
1-,6
'

and (31) (36) imply that the effective strain in the matrix is determined without solving
any nonlinear problem

The overall constitutive law is therefore (these relations are valid for an arbitrary
behavior of the matrix, power-law or not)

_ hom _ 2(d- 1) c( 2l (2 ) ( Eeq )

~m - k trE - d c( 1) fl. 1 + ~c(l) trE,
(46)
_ hom _ 3c( 2) (2) ( Eeq )
~eq - 3J.L Eeq - 1 + ~c(l) 1-"s 1 + ~c(l) Eeq·

Two serious limitations of the classical secant method can be seen from this result.
1. ( GARAJEU [17]) The constitutive law predicted by the secant method does not
derive from a potential (although it is known from section 1.6 that the actual
behavior of the composite derives from a potential). Indeed the existence of a
potential <I> defining the constitutive law (46) would imply that
8<1> 811>
~m = 8trE' ~eq = 8Eeq'
 Effective Behavior of Nonlinear Composites 215

and therefore that

ar.m 82 .P ar..q
8Eeq - 8trE 8Eeq = 8trE'
This relation is not satisfied by {46).

2. {QIU and WENG [47]). When the porous material is subjected to a macroscopic
hydrostatic strain, the average strain in the matrix is hydrostatic and the effective
strain E!~) is 0 whatever the intensity of the overall hydrostatic strain. Therefore
the secant modulus fL~2 ) keeps its initial value fL~2 ){0). The prediction by the
secant method (46) is a linear response of the hollow sphere under hydrostatic
deformation. When the matrix is power-law, the initial secant modulus is +oo
and the secant method predicts that the voided material behaves rigidly under
hydrostatic loading. This prediction is clearly incorrect. The reason for this flaw
is that the average strain of the matrix is not able to pick up local fluctuations
in shear, which average to 0 but which are responsible for the nonlinear overall
response of the composite.

2.5.3 Rigid-plastic phases

When the phases are rigid-plastic (m = 0), the nonlinear equation (43) can be solved
analytically and gives
(1) 1
CTo
when - <-
(2)- a'
O'o fJ

The composite is a rigid-plastic body with flow stress agom

0'( 1 ) d +2
_o_ < --
(2)- 2 '
O'o
{47)
when

2.5.4 Small contrast expansions

When the contrast between the individual phases of the composite is small, one can
derive asymptotic expansions of the effective properties in powers of the contrast. Both
phases are assumed to be power-law materials with the same exponent. The flow stress
of phase 1 differs from the flow stress of phase 2 by a small quantity

a0
(2)
= ao, (1)
a0 = CTo + aa,
l:
8a « a0 •
216 P. Suquet

The contrast between the phases is measured by the small parameter du « 1. Define
uo
w such that
E(tl E(2l
__::!_ = 1 + c( 2lw, __::!_ =1- c(tlw. ( 48)
Eeq Eeq
(Note that the average condition c(ll E~!l + c( 2) E~~) = Eeq is satisfied). Then the
nonlinear equation (43) implies that the ratio between the average strains in each
individual phases is close to 1
E(ll da 1
~~) = 1 + w, w « 1, w '::::!. J3 --.----,...
ao l+f3(m-l)
(49)
Eeq
The overall constitutive law reads
s 2 E
=--,
Eeq 3 Eeq

with

(50)

Combining (49), (48), and (50) finally yields the second-order expansion of the overall
flow stress under the form
2 2
~ _ hom (Eeq)m hom
ao >-
C' < Uo > - < Uo >
' (51)
""'eq- Uo ' uo '::::!.<
co < uo >
with

and
C' = J3 m (1 + ~(m- 1))
(52)
(1 + j3(m- 1)) 2
This final result calls for the following comments:

1. When m -t 0 the constant C' goes to 0 and the prediction of the secant method
coincides with the Voigt bound to second order in the contrast, in agreement
with (47).

2. Since the. composite has overall isotropy, it is expected that the overall constitu-
tive law depends on the three invariants of the overall strain E. The dependence
on the first invariant is ruled out by the incompressibility condition. The secant
method does not predict an influence of the third invariant.
Effective Behavior of Nonlinear Composites 217

2.6 A modified secant method based on the second order mo-

ment of the strain field
In the modified theory proposed by SUQUET [52] the effective strain which serves to
estimate the strain field in phase r is related to the "second-order moment" of the
strain field in this phase
(53)
The secant tensors can be computed in each individual phase with this new effective
strain
f-L(r) = f-L~r)(€~~)), ][)r) = n)rl(k(r)' f-L(rl).

Note that the effective strain (53) is non zero as soon as the equivalent strain ceq is
non zero at some point in the r.v.e. The second moment of the strain can be computed
merely from the overall stiffness of the composite. Its expression follows from the
following general result (previously used by several authors, see e.g. KREHER [27] or
BURYACHENKO [6]):
Consider an N phase linear composite. Let ( k, JL) denote for simplicity the set of
2N elastic coefficients (k(r), p(rl),r = 1, ... N. Let [..hom(k,p) be the overall stiffness of
the composite as a function of the elastic coefficients of the individual phases. Then

=(r)2 2 1 an...hom
ceq =< ceq >r= 3c(r) E : 8p(r) ( k, f-L) : E. (54)

A proof of this result is given in Appendix 1; the expression of the full second moment
< c;jC:kh >r of the strain field in each individual phase (r) is also given. (54) is used to
compute the second-order moment of the strain in the linear comparison composite.
Finally, the modified secant method involves the following steps:

1. A linear theory providing an expression for [..hom([.(•) ls=l, ... N) and its derivatives
with respect to the individual shear moduli p(r).

2. The resolution of N nonlinear problems for theN unknown secant tensors [..(r)(€~:)) :

=(r) 1 an... hom ) 1/2

n...(r) = n.)rl(€(r))
s eq ' ceq = ( 3c(r)E: 8p(r) (k,p): E (55)

3. Once the N nonlinear problems (55) are solved, the overall stress-strain relation
is finally given by :
(56)
It should be pointed out that the modified theory is at least as easy to implement as
the classical one since it does not require knowing the concentration tensors A_(r).
218 P. Suquet

2.6.1 Isotropic two-phase composites. Hashin Shtrikman lower bound

When the composite is made up of two isotropic phases and has overall isotropy, the
effective stiffness of the comparison linear composite depends only on a bulk and a
shear modulus. The resulting expressions for the "effective strains" €~~) are

(57)

These relations can be used with any linear prediction for khom and phom. We give ex-
plicit expressions for the case where the Hashin-Shtrikman bounds (36) are appropriate
to estimate the effective linear properties of the comparison composite (e.g. phase 1
dispersed in phase 2). The effective strains read
=(1) (1) (1)
Ceq = Eeq = aeq Eeq,

(58)

where the strain concentration factors a~) ... are given above.

2. 7 Modified method. Two-phase power-law materials

2.7.1 Nonlinear equation
In this section we apply the modified method to incompressible power-law materials
with the same exponent. The nonlinear problem (55) reduces to a scalar equation
which takes the following form (except for porous materials where the problem is even
simpler, see section 2. 7.3 below)

ll
(1)
Uo
(1) (S(l))
eq
m-1
(59)
p(2) = (7(2) =(2)
0 ceq

When the effective shear modulus phom of the comparison solid is estimated by the
Hashin-Shtrikman lower bound, this equation can be rearranged to give

2
€(2)) ( ( (1) ))2 ( (1) )2
( €~~) = 1 + f3 ~( 2 ) - 1 + /3(1 - f3)c( 1) ~( 2 ) - 1 (60)
Effective Behavior of Nonlinear Composites 219

Remark 5: The volume fractions of the phases enter the above nonlinear equation : the
ratio between the average strains in the two phases predicted by the modified secant
method depends on the volume fraction of the phases.

Equation (60) cannot be solved in a closed form, except when the phases are rigid-
plastic (m = 0) (see below) and is solved numerically.

2. 7. 2 Rigid inclusions

When phase 1 is rigid l~~) = E~!) = 0. The energy equation

J.lhom = c(t)J.l(t) (l~~)) 2 + c(2)J.l(2) (l~~)) 2 ,

Eeq Eeq

and the nonlinear equation (59) show that

1 + !.=Qc(l) )
1/2
(
I. =(2) !3 E
1m Ceq = (2) eq,
"~1)-t+oo C

Finally the effective constitutive law reads

2 hom
s- -~ (E )m-1 £
__.!:!.
- 3 co co '
with
( 1 + _2d c(t))
!!!±l

a 0hom = a 0(2) -'---"'=-:::;-'---

c(2)m
2
(61)

2.7.3 Porous materials.

When the inclusions are voids the effective strain in phase 2 is determined by the
relations (58) without solving any nonlinear problem.

=(2) (2(d-1) 2 1 2) 1/ 2
ceq = 3dc(t) trE + 1 + ~c(t) Eeq (62)

The overall constitutive law is therefore

"' = khom
"-'m
trE = 2( d d- 1) c(2)
c(1) J.ls
(2) (:;;;(2))
C.eq trE,
(63)
3 (2)
Eeq- 1 + ~c(l)J.l 8
I; _ hom _ C (2) (=(2))
eq- 3j.t Ceq Eeq·

The modified method is free of the two limitations of the classical secant method
pointed out in section 2.5.2
220 P. Suquet

1. The constitutive law predicted by the secant method does derive from a potential.
This point will be proven in general in section (4) but it can be directly proven
here by setting
ci>(E) = c(2) J(~~~l\
where f and~~~) are given by (3)and (62). It can be easily checked that with this
definition of ci> :
8cl> 8cl>
~m = 8trE' ~eq = 8Eeq •

2. The overall response of porous materials under hydrostatic strains, as predicted

by the modified secant method, is nonlinear.

2. 7.4 Rigid-plastic phases

When the phases are rigid-plastic (m = 0), the nonlinear equation (59) is a quadratic
equation which can be solved analytically. Looking for positive solutions one obtains

(2) {3
O'o < {3 1 + ..!....=_c(1)
(1) - {3 '
O'o

~ (1 - (0'~1))
1 + 1 - {3 (2)
2)
O'o

1 - {3 + {3c(ll
(2)
1-{3
1 + -{3-c( 1 l. The composite is a rigid-plastic body with flow stress
O'o
when (ij 2 {3
O'o
O'0hom ..

1 + 1 - {3 c(ll (64)
{3 '
and

~ (1 - (0'~1))
1 + 1 - {3 (2)
2) ' (65)
O'o

(2)
1-{3
1 + _{3_c(1l.
O'o
when (ij 2 {3
O'o
Effective Behavior of Nonlinear Composites 221

2. 7. 5 Small contrast expansions

When the contrast between the individual phases of the composite is small, the overall
effective flow stress ugom can be asymptotically expanded in powers of the contrast.
The calculation is tedious but straightforward. First the expansion to second order of
the ratio between the effective strains in the phases can be found from the nonlinear
equation (59)

=(1)
C:eq •
00'
= 1+a-+b
( 00'
'
-
)2 , a= f3 _ ma 2 a3 ( 1 - f3 (1) _ )
-( 2 )
l eq O"o O"o 1 +f3(m- 1),b- 2 +/3 f3 c m. (66)

The energy equation

and the relation l:eq = 3phom Eeq imply that l:eq = ugom ( !f;) m with

Then the expansion to second-order in the contrast of the overall flow stress follows
from (66) and from the expansion to second order of e~~) = E~~l. Finally one obtains
2 2
hom
O'o :::< O"o > - C" < O'o > - < O'o > , (67)
< O'o >
with
C"= f3(m+1) {68)
2 (1 + f3(m- 1))
2.8 Comparison between the classical and the modified secant
method
The explicit results derived for power-law materials can be used to compare the two
secant methods based on the average strain and on the second moment of the strain
respectively.
1. The dilute expressions (small c< 1l) of the overall flow stresses (44) and (61) ob-
tained for rigidly reinforced composites are

classical method
O'hom
- 0-
(2)
"'
-
1+
(d-2 + m) c< 1)
'
O'o
 222 P. Suquet

(m-2+-1 -2 +
modified method
d )
1+
O'hom
- 0- "' m c(t)
(1~2 ) - .
For isotropic rigid-plastic materials {m = 0, d = 3) the predicted slopes are
~ (classical method) and ~ (modified method). BAO, MAC MEEKING and
HUTCHINSON [1], and MICHEL and SUQUET [33] performed cell calculations by
the Finite Element Method (a spherical inclusion placed in a cylindrical volume
of matrix) and found that the slope is approximately 0.375. The results of [33]
are shown in Figure 3. Both methods overestimate the reinforcement effect of

1.6 ....
~--.--.--....-,-.--r-.--.-.-...--.--.--.--,.....--.-......-,···"71
········ Classical secant method ....······
- - Modified secant method

F.E.M. ///,,//
a 1.4

-~

,•'•'
.. ....···
,•'
•'

..·······
..······•···

... .
...·············
1.0 """'-_._....................._._.....__.__.__.__........._._....__..__.__.__...........__,
0.0 0.1 0.2 0.3 0.4
(1)
c

Figure 3: Spherical rigid inclusions in a rigid-plastic matrix {m = 0). Overall flow

stress as a function of the volume fraction of the rigid phase. Dotted line: classical
method based on the first moment of the strain, dashed line : modified method based
on the second moment of the strain, circles: cell calculations by the Finite Element
Method {from {33}).

rigid inclusions dispersed in a matrix, but the modified method is closer to the
numerical results than the classical one. For two-dimensional composites both
methods predict a positive slope (the slope is 1 for the classical method and 0.5
for the modified method). Numerical simulations (not reported here) show that
the slope is likely to be 0, which means that there is no reinforcement effect
in two-dimensional composites at low volume fraction of the reinforcing phase.
None of the present methods is capable of picking up this effect. However, PONTE
 Effective Behavior of Nonlinear Composites 223

CASTANEDA [41] [42] has recently proposed another predictive scheme which is
able to predict the correct slope.

2. Figure 4 shows the overall flow stress of isotropic rigid-plastic composites that is
predicted by the two secant methods and found numerically by cell calculations.
The overall flow stress is plotted as a function of the contrast between the overall
flow stresses of the individual phases. Both methods predict the threshold effect
observed in the numerical calculations: above a certain contrast, the particles do
not deform and behave as rigid inclusions. This threshold effect can be attributed
to the fact that the shear stresses in the vicinity of the particles are limited by
the Von Mises criterion in the matrix. Therefore the shear stresses transmitted
from the matrix to the particles are limited. When the flow stress of the particles
is high enough these particles cannot yield and behave as rigid inclusions.

(I)
c0 >= 0.15 1.8 c =0.3

- -
B~L2 B~ 1.6

j j 1.4
~1.1 ~
1.2

1.0 "-'-'......_......_._...._._........._.................._._...._._............, 1.0 .................................................................................................~

1 2 3 1 2 3 4
uoo> I uo<2> uo<•> I uo<2>
- Upper bound: Voigt - - Modified secant approach
········ Classical secant approach F.E.M.: Axisymmetric cell

Figure 4: Rigid plastic constituents (m = 0). Overall flow stress as a function of the
contrast between the flow stresses of the individual constituents. Dotted line: classical
method based on the first moment of the strain , dashed line : modified method based
on the second moment of the strain , circles: cell calculations by the Finite Element
Method (from {51}). a) c(l) = 0.15. b) c(l) = 0.3.

3. The exact expansion to second order of the overall flow stress of power-law com-
posites has been given by PONTE CASTANEDA and SUQUET [53] [45] :
2 2
>- C < O"o > - < O"o > ,
hom
ao ~< O"o (69)
< O"o >
224 P. Suquet

with

The constant C depends on the third invariant of the overall strain E. More
specifically the eigenvalues of E = E/ Eeg can be written in terms of an angle (}
(0<0<7r/3)
7r
"3 ), E3 =
A

E1 = cos(O- cos(O).

(} is related to the third invariant of E by det(E) = cos(30)/4. C is finite for

all finite values of n and is also finite when n = +oo, except when one of the
eigenvalues of E vanishes (the other two eigenvalues are opposite by incompress-
ibility, corresponding to pure shear). The variations of C with n and the third
invariant of E are shown in Figure 5 The predictions of both secant methods

2.0

1.8

1.6
1.4

1.2

Ul.O
0.8
0.6

0.4
0.2

0.0
0.0 0.1 0.2 0.3 0.4 0.5
e
Figure 5: Constant C giving the expansion to second-order for power-law composites
with a small contrast. Dependence on n and on 0, det(E) = cos(30)/4. From {53}.

(classical and modified) disagree with this exact result. They do not depend on
the third invariant of E. It is also found that the constant C' and C" are decreas-
ing functions of n, whereas the exact constant C is an increasing function of n.
For rigid-plastic phases the prediction of the classical method even coincides (at
second order) with the Voigt upper bound. This prediction is excessively high.
 Effective Behavior of Nonlinear Composites 225

4. The methods are not restricted to the Hashin-Shtrikman bounds and to power-law
materials. Figure 6 shows the overall response of a nonlinear composite predicted
by the two secant methods and by cell calculations. Both phases were assumed
to be incompressible. The matrix was governed by the following constitutive
stress/strain relation

Ceq = 3po
aeq
+
((aeq-h a 0 )+)n ' (71)

where (. )+ denotes the positive part of the quantity inside the parentheses. The

~0~--~----~----.----~----~

................-··········
....··
300 ......

IN200 1// ----------------------

,, ...
,, , c< 1>=0.3
Generalized self-consistent scheme
I
100 ........ Classical secant method
- - Modified secant method
-F.E.M.
--- Matrix
OL----'~---'----'"----'-----'
0.0 0.02 0.04 0.06 0.08 0.1
E

Figure 6: Spherical inclusions dispersed in a matrix. The inclusions are linear elastic.
The matrix is incompressible and nonlinear (governed by (71)). Volume fraction of in-
clusions c(l) = 0.3. The effective stiffness is estimated by the three-phase self-consistent
scheme.

material data used in the calculations were Jlo = 25 GPa, n = 2.567, h =

416.5 MPa, a 0 = 75 MPa. The particles were assumed to be elastic (and
incompressible) with an elastic shear modulus p(I) = 133 GPa and a volume
fraction c(l) = 0.3. The predictions of the classical secant method are stiffer than
the predictions of the modified secant method. The modified secant method is
in good agreement with the cell calculations.

5. There is no evidence that the prediction of the classical approach derives from
an elastic potential (as it should). Conversely, it will be seen in section 4 that
the prediction of the modified method always derives from a potential.
226 P. Suanet

2.9 Dual formulations

The compliance tensors M = (Lr 1 are sometimes more appropriate than the stiffness
tensors. This is the case for instance in the simulation of a uniaxial tensile test. The
overall stress :E is given but the overall strain E is unknown. The secant procedures
can be developed with the stress as the main variable instead of the strain. These
approaches (classical and modified) are completely parallel to the ones which have
been described, and we shall not go into the details of the derivation. In addition the
results of the approaches based on stresses (or compliances) are rigorously equivalent
to the predictions of the approaches based on strains (or stiffnesses ). The main results
are as follows. The constitutive law of each individual constituent reads

e = M.(u): u, M.(u) = ;.Jf + e.~u)][{,

1
where m = -k, e. = The compliance tensors in the linear comparison composite
f.ls
are assumed to be constant within each individual phase.

2.9.1 Classical approach

The "effective stress" of each individual phase is the average of the stress over this
phase given by the stress concentration tensors JR(r)

:E(r) =< 0" >n :E(r) = JR(r) : :E.

Again the secant method involves two major ingredients:

1. A linear theory providing an expression for the overall compliance Mhom and the
N stress concentration tensors JR(r) as functions of M(•) ls=l, ... N,

2. The resolution of N nonlinear problems for the N unknown secant tensors M(r)

(72)

3. Once the N nonlinear problems (72) are solved, the overall stress-strain relation
is finally given by :
E = Mhom (:E) : :E
Effective Behavior of Nonlinear Composites 227

When the individt,lal constituents and the composite are isotropic and when the Hashin-
Shtrikman bounds are relevant to the microstructure under consideration, one has
mhom ()hom
Mhom = - - J + --![{
3 2 '

(73)
()(1) - ()(2)
1 + ----,---
b(2)- ()Q
eq - ()(1) _ ()(2) '
1 + c< 2 l ----::---
()Q

m(2) ()(2)
where mQ = - - , OQ =---a·
1-o: 1-fJ

2.9.2 Modified approach

The "effective stress " in phase (r) is estimated by the second-order moment of the
stress field in the linear comparison composite

=(r)
(jeq =
(
< (jeq
2
>r )1/2 •
This second order moment can be related to the overall compliance by

=(r) 3 !:l~AThom ) 1/2

(j eq =
( c(r) :E : U!'(ll
/)()(r) ( m, 0) : :E (74)

The dual formulation of the modified secant method involves the following steps:

1. A linear theory providing an expression for Mhom (M(•) ls= 1,... N) and its derivatives
with respect to the individual shear compliances ()(r).

2. The resolution of N nonlinear problems for theN unknown secant tensors M(r) (a~~)) :

= M(r) (a(r)) =(r) ( 3 i)M[hom ) 1/2

M(r)
s eq ' (j eq = c<rl :E : ()()(r) ( m, B) : :E (75)
 228 P. Suquet

3. Once the N nonlinear problems (75) are solved, the overall stress-strain relation
is finally given by :

When the Hashin-Shtrikman lower bound is used to estimate Mhom the "effective
stresses" are given by
=(1) - ._,(1) - b(1)._,
aeq - L.Jeq - eq L.Jeq,
=(2) -
aeq -
( >'2
UL.Jm + b"2
L.Jeq
) 1/2
'

(76)

Remark 6 : QIU and WENG [47] have also proposed a theory based on the second-
order moment of the stress in each individual phase to predict the effective properties
of nonlinear composites. Their expression for the "effective stresses" is similar to (76),
except that coefficient b is estimated by

b = _1_ (ohom - c(l)()(l)b(1)2) (77)

c(2)()(2) eq •

Apparently these authors were not aware of the exact result (74). When the phases
are incompressible, Qiu and Weng's theory coincides with the present theory.

3 The incremental method

3.1 General procedure
When the constitutive law of the phases is written in an incremental form the local
problem reads

e = ~ (\lu + \7u 1) ' div(u) = 0, +boundary conditions, }

u= n:..~r)(e(x)): e(x).

The stiffness tensors L~r) can be different from one point to another since they depend
on the strain field e(x) ). Again, an approximation has to be introduced to simplify
the problem and the tangent stiffness tensors L 1 are replaced by tensors [.(r) which
are constant in each phase. These tensors are computed from the exact expressions of
the tangent tensor in each phase (r) evaluated at some effective strain for this phase.
Again the most immediate choice for the effective strain of phase r is the average strain
over this phase
(78)
 Effective Behavior of Nonlinear Composites 229

The local problem now reduces to

e= ~ (\7u + \7u 1)' div(u) = 0, +boundary conditions, }

(79)
u = L(rJ : e(x), L(rJ = Llrl(E(rJ).
It is a linear problem for the increments e and u which can be determined with any
linear theory appropriate to the morphology of the composite and capable of predicting
the effective properties of anisotropic linear composites. There is no nonlinear equation
to solve. At each timet, the average strains E(r) and the tangent tensors L(r) are known.
The overall elastic tangent tensor Lhom is computed and the overall constitutive law is
found in incremental form
(80)
The increments of average strain in each phase are computed by

where the tensors A(r) are the elastic strain concentration tensors associated with the
stiffness tensors L(r) and Lhom. The average strains E(r) are then updated and the
procedure continues step-by-step.

Remark 7: Under certain conditions, the above relations (80) are integrable in time
and the overall constitutive law can be expressed as a relation (nonlinear) between
stresses and strains. But nothing in the procedure guarantees that this integration
is possible and, when it is so, it is not clear that the resulting constitutive equations
derive from a potential.

3.2 Isotropic two-phase composites.

As mentioned above, the tangent tensors L 1 (8) are anisotropic. The linear theory used
to derive the effective tangent stiffness has to be able to predict the overall behavior
of anisotropic phases, even when the geometry of the constituents does not favor any
particular direction. The extension of Hashin-Shtrikman bounds by WILLIS [61] meets
these requirements and the expression (33) can be used to estimate Lhom. The main
difficulty then is to compute the tensor r( 2l from the general relations (34). One of the
problems with this determination is that the phases generally have different types of
anisotropy. Indeed the tangent tensors L(r) read

where JE(r) = 3£(r) ® £<r). In general the tensor JP'( 2 ) has to be computed numerically.
Interestingly this computation is simplified in specific situations, and results can be
derived in closed form. More specifically we assume that
 230 P. Suquet

Hl. the composites under consideration are two-phase composites and the arrange-
ment of the phases is isotropic (in the view of WILLIS this amounts to the state-
ment that the two-point correlation function of the characteristic functions of the
phases depends only on the distance between the points).

H2. the average strains in each phase have proportional deviators

This last statement, formulated here as an assumption and used previously by

HUTCHINSON [26], can be justified for spherical inclusions in a matrix, provided
the loading applied to the composite is purely radial.

When only the stress/strain response of the composite under a radial path of defor-
mation (or stress) is searched for, there is no need to compute the whole tensor P( 2).
Knowing £: P( 2 ) : £ is sufficient. This scalar quantity was computed by SUQUET and
PONTE CASTANEDA [51] for incompressible materials and by PONTE CASTANEDA
and SUQUET [45] for compressible materials. It was used by SUQUET [54] in a study
on the incremental method. However the most general result was given by PONTE
CASTANEDA [42] (with a correction in [44]) who derived the general form of the r
tensor for this class of composites {generalizing {35))

1 1 1
r (2) =-.If+
3kp
-IF+ -IE,
2J.Lp 2Ap

where

and
_1_ = _1 f i£.ei2- o:Pl(e.t.e)2 ds(e) {81)
2Ap 61r Je=1 J.l(2) + ~(A(2) _ J.l(2)) (Jt.e 12 _ (e.t.ep)
1 1 '
(2) - 3k(2) + J.l(2)
0: - 3k( 2) +4J.L( 2).
Effective Behavior of Nonlinear Composites 231

The expression of 1-"P given in [44] is not used here. The resulting strain concentration
tensors A(r) take the form

= 1 1 1
A(l) IF+ IE,
( )k(1)- k(2) J+ ( )1-"(1) -p(2) ( ) ,A(1) - ,A(2)
1+c 2 1+c 2 1+ c2
kp 1-"P Ap
k(1)- k(2) 1-"(1) -p(2) ,A(1)- ,A(2) {82)
1+ 1+ 1+
= kp 1-"P Ap
A(2) J+ IF+ IE.
( ) k(l) - k(2) ( )1-"(1) -p(2) ( ) ,A(l) - ,A(2)
1+ c2 1+ c2 1+ c2
kp 1-"P Ap

3.3 Incremental method. Incompressible power-law materials

The composites considered in this section are two-phase isotropic composites made
up of power-law materials with the same exponents m and n but with different flow
stresses u 0 • Their tangent stiffness tensors are then given by {16). Since phase 2 is an
incompressible power-law material, the above expressions reduce to

1 CCE,m)
{83)
2Ap m +1 p( 2 )

where C is the constant introduced in [53] and given by {70). The following result is
central in the discussion of the predictions of the incremental procedure for power-law
materials.
Under the above assumptions {Hl} and (H2}, the equations resulting from the incre-
mental procedure can be integrated in time and give

{84)

In other terms the incremental procedure coincides with a classical secant procedure
where the secant stiffness tensors are given by ~L~r) (this observation was made by
HUTCHINSON [26]). These secant stiffness tensors are anisotropic and differ from the
"isotropic" secant stiffness tensors used in section 2. It is therefore expected that
the effective properties predicted with these secant tensors differ from the effective
properties predicted with the isotropic secant tensors. Although this result is interest-
ing for theoretical purposes, it does not simplify the numerical implementation of the
procedure : the incremental procedure requires the step-by-step resolution of linear
problems, whereas the secant procedure requires the resolution of one nonlinear prob-
lem (which is achieved by solving a succession of linear problems). It should also be
kept in mind that this result is valid only under proportional loadings.
Proof: This is a detailed proof of the equivalence between {79) and {84) for power-
law materials. Consider the solution {u, u) of the secant problem associated with the
232 P. Suquet

secant tensors ~n:..~r)

e = ~ (V'u + V'u 1), div(u) = 0, u = E(t).x on 8V, }

(85)
0' = ~n:..~(E(r)): e(u).
These fields depend on t. The strain concentration tensors corresponding to the tensors
~[..~r.) are identical to those corresponding to the tensors n:..lr) and are therefore given by
(82) where the term multiplying .If drops (incompressibility) and where the coefficients
_x(r) and J.l(r) are related by _x(r) = mJ.l(r). The average strains in the two phases read

E(t) = E( 1 l(t)f. E( 2 ) = E( 2 l(t)f.

eq ' eq '
where the E~~) are solutions of the following nonlinear equation

E(t)
eq
1
= --.,--,----:-:- ( Eeq(1)) m-1 1
.X (2)
2 c
= m + 1 J.l(2). (86)
E(2) _x(t) _ _x(2)' Eeq
(2)
1+
p
eq A]'

This equation can be rearranged into

(1 2mC) E~!l 2mC a~1 l (E~!l) m

(87)
- m + 1 E~:) + m + 1 a~2 ) E~:) = 1'

Although the average strains E(t) and E( 2) depend on t, it follows from (87) that the
ratio of their norms does not depend on t. Then it follows from (82) and from the last
equations in (86) that

1 (1 -
E(eq2 l(t) = -c(2) c(t) E( 1 l(t))
eq ·

Therefore E(r)(t) = a(r) Eeq(t) where the second order tensor a!r) does not depend on
t. By homogeneity of degree m- 1, the stiffness tensors read

It follows that the field e(x, t) can be written as

e(x, t) = a(x)Eeq(t), (88)

where the field a does not depend on t. Taking the time derivative of (85) yields

· = _!__n:..(r)(E(r)) .. + _!__ an:..lr) . E. (r) . (89)

0' m t •e m 8E(r) . . e.
 Effective Behavior of Nonlinear Composites 233

But E(r) = !::~:~E(rl, and itfollows from (88) that !::~!~e(x, t) = e(x, t). Finally one
can again use the homogeneity of degree m- 1 of the stiffness tensor (Euler theorem)
8L(r)
to prove that _ t _ · E(r) = (m- 1)L(r)(E(r)) This last result combined with (89)
8E(r). t •
completes the proof of (84).
The equivalence between the incremental procedure and a classical secant procedure
can be used to obtain explicit results in the case of materials with infinite or small
contrast and in the case of rigid-plastic phases. Indeed the nonlinear equation (87) to
be solved is identical to (42) provided {3 is replaced by 2mC . We can check that all
m+ 1
the results established for the classical secant method in the case of rigid inclusions
and rigid-plastic bodies and the small contrast expansions remain valid for the present
anisotropic secant model. More specifically one obtains:
Rigid inclusions.

The overall flow stress is:

( c(2) + m2m±Cl c(l))
,..hom_ ,..(2) ....:.--7:7i~-.:....
uo - uo c(2)m (90)

Small-contrast expansions

< rr~ >- < rro >2

< Uo >=< Uo > -Cine ,
< rro >
with
2Cm m (1 + £j:T(m- 1))
(91)
Cine= m + 1 (1 + ~±";(m- 1))2
Rigid-plastic bodies.
When m goes to 0, an inspection of the integral defining C shows that

lim C(m,E) m
m-+0
= 0.

Therefore (88) gives in this limit E~!) = E~:) and by the average condition c(t) E~!) +
c(2lE(eq2) = E eq, one obtams
. E(t)
eq =
E(eq2) = E eq.

For rigid-plastic materials the incremental model reduces to the Voigt bound.
A similar observation was made by GILORMINI [18] with the self-consistent scheme
(instead of the Willis lower bound). The above result shows that the convergence of
the incremental model to the Voigt bound (or Taylor model) when m goes to 0 is not
due to the condition of self-consistency hut to the incremental procedure itself.
 234 P. Suquet

3.4 Comparison with the secant methods

The overall flow stresses predicted by the two secant methods and by the incremental
procedure are shown in Figure 7. The applied overall stress was uniaxial tension. The
ab ab
contrast between the phases was 1 ) j 2 ) = 8, and two different values n = 3 and
n = 10 were considered. The rigorous upper bound of PoNTE CASTANEDA [39] (in
the form given by SUQUET [51]) is also plotted for reference. The incremental model
is always stiffer than the two other procedures. In addition the incremental model
violates the rigorous upper bound for large values of n (m close to 0).

8 8
n= 10
7 7
......
.·
6 6
N' N'
~~5 ~~ 5 ...................
-.. .......
...··
.·
j e
4 ..8 4 ....··
~ ~
3 3 .....············/ .·
2 2 ..... _.... /

-~~:;_·;;.~-----=--·--·
1
0
c(l)

Upper bound: Voigt - - Incremental procedure --- Modified secant approach

......... Variational upper bound · · · · · · Oassical secant approach --·-· Reuss bound

Figure 7: Isotropic two-phase composites. The constituents are power-law with the same
exponent n. Overall flow stress as a function of the volume fraction of the strongest
phase. When the nonlinearity is strong enough the incremental procedure violates the
rigorous variational upper bound established in section 4 .

4 Variational procedures
The most recent procedures for predicting the overall nonlinear properties of composites
are the variational procedures. The first contributions by WILLIS [62] and TALBOT
and WILLIS [56] were followed by the studies of PONTE CASTANEDA [39] [40] [41]
and SUQUET [50] [51] [55]. Here [51] is considered. To simplify, we shall consider
first composites with infinite contrast (a nonlinear matrix containing pores or rigid
inclusions) and then power-law materials, before giving results for general nonlinear
constituents.
Effective Behavior of Nonlinear Composites 235

4.1 Nonlinear materials containing voids or rigid inclusions

The composites considered in this section consist of a nonlinear matrix with voids and
(or) rigid inclusions.

4.1.1 Variational procedure

The potential r.p of the matrix is assumed to be a function of an "elliptic equivalent
strain" C:e
r.p(e) = r.p(ce), C:e = (e: II..O: ef 12 ,
where rr..O is a fourth order tensor with diagonal symmetry. The case of an isotropic
incompressible matrix governed by the Von-Mises equivalent strain corresponds to

C:e
2
= ceq = ( 3C:ijC:ij
)1/2 when tr( e) = 0, C:e = +oo otherwise,

IL = 3k0 .1f + 2lOC, J1° = 1/3, k0 = +oo.

The main result of this section is the following.
Assume that r.p can be written as

where f is a concave and increasing function. Then the effective potential !I> can be
bounded

!I>(E) ::S C(m)r.p (Ee), Ee-

_(E : JLhom :
c(m)
E) 1/2 (92)

JLhom is the effective stiffness of a linear composite with the same geometry as the initial
nonlinear one (pores, rigid inclusions, matrix), but made of an incompressible linear
elastic matrix with stiffness IL 0 .

c(m) denotes the volume fraction of the matrix. (92) can be given in an equivalent dual
form. First note that the potential 7/J (dual of r.p) can be written

Then (92) is equivalent to

_ (~: MJ:hom: ~)1/2

L:e- c(m) ' (93)

Proof of {92) : The concavity off implies that for any function F defined on V

f( < F >m) 2:< f(F) >m .

236 P. Suquet

Then, for every v

It follows from the definition of <I> that

<I>(E) = veK:(E)
inf < <p(e(v))) >,

then
<l>(E) ~ vEK:(E)
inf c(m) f( < e~(v) >m) = c(m) f ( inf < e~(v) >m) .
veK:,(E)
(94)

(the last equality is a consequence of the fact that f is an increasing function. The
effective energy of the linear composite composed of a linear matrix with stiffness [.0
containing pores and rigid inclusions as in the original nonlinear composite is

E : ll.}om : E = inf c(m) < e(v) : ll...0 : e(v) >m= inf c(m) < e~(v) >m .
vEK:(E) veK:(E)

Then
E ·[..hom, E
inf < e~(v) >m= · ( ) · , (95)
vEK:(E) C m

and (92) follows from (94) and (95).

The inequality (92) can be used in two ways:
1. When rigorous upper bounds for [..hom are available, their incorporation into (92)
yields rigorous upper bounds for the effective potential <I>. The Hashin-Shtrikman
upper bound for isotropic composites and the Herve-Stolz-Zaoui upper bound for
composite sphere assemblages are two examples of such upper bounds. In specific
cases (e.g. power-law materials) a bound on the potential can be converted into
a bound on the effective properties (or the overall stress-strain curve).

2. Estimates for the effective potential <I> can be derived from (92) by incorporating
estimates for the effective stiffness tensor [..hom.

In both cases note that there is no nonlinear problem to solve and that the bound or
estimate on the potential is deduced from a linear theory.

4.1.2 Isotropic rigidly-reinforced materials

The composites considered in this section are made up of a nonlinear incompressible
matrix with rigid inclusions. Phase 1 denotes the inclusions and phase 2 the matrix.
The "elliptic strain" ee of the matrix is the Von Mises equivalent strain. The linear
comparison composite is incompressible and isotropic, and its effective elastic properties
 Effective Behavior of Nonlinear Composites 237

depend on a single elastic constant, the shear modulus p,hom. The estimate obtained
for the effective potential is
hom ) 1/2 )
<I>(E) ~ c(2)r.p ( ( :Oc{2) Eeq . (96)

This result can be interpreted as follows. The strain fluctuates in the matrix but it can
be roughly estimated by an "effective strain"
hom)1/2
Ee = (
:Oc( 2) Eeq• (97)

The rigid inclusions do not contribute to the overall potential which reduces to the
contribution of the matrix c< 2lr.p ( Ee)·
When the Hashin-Shtrikman lower bound is used in (97) to estimate p,hom, the
effective strain is
(1 + ~c(1lf/2
Ee = c{ 2 ) Eeq·
If in addition the matrix is a power-law material, the estimate of the effective potential

(E )m+1
IS

hom
<I>(E) = ao co ___!!!_ '
m +1 co
This last result coincides with (61). The Hashin-Shtrikman upper bound is +oo and
does not give any information on <I>.

4.1.3 Isotropic voided materials

The composites considered in this section are made up of a nonlinear incompressible
matrix containing voids. Phase 1 denotes the voids and phase. 2 the matrix. The "ellip-
tic strain" ce of the matrix is the Von Mises equivalent strain. The linear comparison
composite is compressible and isotropic, and its effective elastic properties depend on
two constants, the bulk and shear moduli khom and p,hom. The upper bound obtained
for the effective potential is

Ee -
_(-1( \1/2
2l J
(khom
0 trE
2
+
flhom
0
2 ) 1/2
Eeq (98)
C 1 3p, fL
When the Hashin-Shtrikman upper bounds are used in (97) to estimate khom and p,hom,
one obtains

Ee = ( AtrE
2
+ B Eeq2 ) 112 , A = 2( d- 1) _1
< l, B= 1 (99)
3d c 1 1 + ~c(ll ·
This result coincides with (62). It has been derived by PoNTE CASTANEDA [39], TAL-
BOT and WILLIS [57], MICHEL and SUQUET [32] and SUQUET [50] by three different
methods.
 238 P. Suquet

4.2 Power-law materials

4.2.1 Upper bound for the effective potential
The composites considered in this section are made up of power-law materials with the
same exponent n and the same reference strain eo, but different flow stresses o-0 • Recall
the variational characterization of the effective potential of the composite

II>(E) = veK(E)
inf ('P(x,e(v))), = inf { - 1 - - 1 (a0 (x)c:~+ 1 (v(x))) }·
vEK(E) m + 1 c: 0
A fictitious linear nonhomogeneous composite (linear comparison composite), which
is incompressible and isotropic with an arbitrary nonnegative shear modulus JL(X) at
point x and occupies the same volume element V as the given real nonlinear composite,
is introduced. Note that

The~ Holder's inequality,

1 1
(191 +-s = 1,
I I
(J.g)::; (lfnr 8
)•, -
r
applied with

J(x) 3
= ( "2JL(x)c:eq(v(x))
2 )~ , g(x) = ao(x) (32JL(x) )-~ , r = m 2+ , s = _2 m,
1 1

gives

{100)

Equality holds in {100) when JL(x) = 3a0 (x)c:~- 1 {v(x)). Therefore

(101)

The infimum of (aoc:~+l(v)) over all admissible fields v can be computed from (101)
by interchanging the order of the infima:

II> E =-
1 1 3
- - inf inf - c: 2 v .m±..!. (
2 3
-
.m±..!.
m-1 _2_ ) I 2m)
m + 1 c:0 l'(x)>O ( veK(E) ( 2JL eq( ))
17 1 -m
( ) (2JL) 0 ·
 Effective Behavior of Nonlinear Composites 239

The infimum over v amounts to finding the overall elastic behavior of a fictitious
composite consisting of elastic, isotropic and incompressible phases with shear modulus
11(x) at point x. Let Whom (11, E) denote the overall elastic energy of the comparison
linear composite,
(102)

then:
1-m}
I\ (~11)
!!!.±.!. -2-

<I>(E) = _1__ 1 { inf whom(l1, E)¥ m-1 O"~:m) (103)

m + 1 t: 0
0

1-'(xJ::::o 2

Comments:
1. The above relation expresses the nonlinear properties of the composite (namely
the potential <I>) in terms of the linear properties of a class of fictitious composites
(namely the energy whom).

2. <I> is exactly determined only when Whom(11, E) is known for every field 11(x). This
exact information is not available except in very specific situations (laminates),
but (103) provides an upper bound for <I> with any choice of the field 11· Rigorous
upper bounds for <I> can be obtained by taking rigorous upper bounds for whom 0

Estimates of Whom yield only estimates of <I>.

3. In most cases (103) is applied by assuming that the shear modulus is constant
in each phase: JL(x) = IL(r) in phase (r). The optimal value of the moduli IL(r) is
determined by solving an optimization problem derived from ( 103). The analogy
with the secant methods is now apparent : in both methods a linear comparison
composite is introduced and the elastic properties of the individual phases are
determined by the procedure according to a given criterion. However the criterion
apparently differs in the secant methods, where the elastic moduli are the secant
moduli for some "effective strain" of the phases, and in the variational methods,
where they are determined by an optimization procedure. This difference will be
examined in more details in section 4.4.3.

4. <I> is positively homogeneous of degree m + 1 with respect to E

<I>(,\E) = ,\m+ 1 <I>(E).

(102) and (103) compare two functions, <I> and w¥, which have the same degree
of homogeneity.

5. The upper bound (103) can be given in a dual and equivalent form
..1.1:L -2-

0";;-1 1-n}
w(:E) ;:::: -
n
CQ

+
•
mf { w*
ll'(x)::::o
hom
(JL, L)
!!..±1.
2 I\ (6JL)n-1
!!..±1.) '
240 P. Suquet

where w;om(Jl, :E) is the conjugate potential of Whom(Jl, E).

4.2.2 Isotropic power-law materials

The composites considered in this section are assumed to be isotropic. Both the ficti-
tious linear composite and the nonlinear composite are incompressible (all the phases
are incompressible and perfectly bonded), except for voided materials which have al-
ready been considered. The infimum in (103) is restricted to fields J1 which are constant
in each individual phase : J1 = Jl(r) in phase r . The fictitious composite is isotropic in
the overall sense and its effective energy whom can be expressed with only one material
parameter, the effective shear modulus Jlhom. ( 103) gives

<I>(E) S
hom
co
O"o
(E )m+l
___!!!!.. '
m+ 1 co

When the composites under consideration are composed of two phases, (104) takes the
form

(105)

{1)
(105) is a unidimensional optimization problem involving the variable ~(z). It cannot
be solved in closed form except in special cases.
Remark 8 : It follows from (103) that the overall potential of the composite <I> is
bounded from above by a power-law potential with a flow stress u3om. This does not
mean that the potential of the composite depends only on the Von Mises equivalent
strain. (104) does not rule out the possibility that <I> may depend on the third invariant
of E.

Except when the phases are rigid plastic, the optimization problem (105) has to be
solved numerically. Different predictions for ugom, corresponding to different choices
for Jlhom, are shown in Figure 8. Note that the prediction derived from the Hashin-
Shtrikman upper bound is a rigorous upper bound for all isotropic composites. Similarly
the prediction deduced from the Herve-Stolz-Zaoui upper bound IS a ngorous upper
bound for all isotropic composite sphere assemblages.

4.3 Rigid plastic phases.

4.3.1 Extremal surface
When the composite consists of rigid ideally plastic phases, <I> is a positively homoge-
neous function of degree one with respect to E. It is the plastic dissipation function of
Effective Behavior of Nonlinear Composites 241

Upper bound: HS+ Upper bound: H-S-Z

4 4

c:!,
~
..._3
§
.c: 0
b

2 2

~~~~~~~~~~~~ ~~~~~~~-L~~~~

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
c0 l - n=l cOl
······ n=3
--- n=5
· · · · n= 10
--n=oo

Estimate: Three-phase Estimate: HS.

4 self-consistent 4 f I
:I
:I
: I

:
.: II
.. .- I
.•
.. .- I
I .

. ....-,I :

~--
2 2

~~~~~~~~~~~~ ~~~~~~~~~~~~

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
c(Il (I)
c
Figure 8: Isotropic two-phase composites. Power-law phases with the same exponent.
Contrast between the phases 17~1) /a~2 ) = 5. Predictions of the overall flow stress by
( 105) when different upper bounds and estimates are used for f..lhom: rigorous upper
bounds deduced from the Hashin Shtrikman upper bound ( HS+) and the Herve-Stolz-
Zaoui upper bound (HSZ),estimates deduced from the three-phase self consistent scheme
and the Hashin-Shtrikman lower bound (HS_) (from {51}).
 242 P. Suquet

the overall strength domain of the composite defined as (DE BUHAN and TALIERCIO
[10], SUQUET [48] [49]) :

such that there exists D'(x) with (D') = ~. }

(106)
div(D'(x)) = 0, O'eq(x) ~ a0 (x), for every x in V}.
Then,
ci>(E) = Sup < ~ : E) > .
~ E phom
The extremal surface of the composite (HILL [25]) is the boundary of phom. It can
be defined regardless of the behavior of its individual constituents prior to the flow
stress. The elastic moduli of the phases and their hardening influence the macroscopic
hardening of the material and hence the transition to the ultimate flow surface of the
composite, but not the extremal surface itself. The phases can therefore be assumed
to be rigid-plastic without any loss of generality relative to the extremal surface of the
composite. This extremal surface depends on the flow stresses, on the volume fractions
and on the arrangement of the individual phases.
The general properties of phom will not be discussed here (the reader can refer
to SUQUET [49]), but it may be worth noting that (103) gives information about the
overall strer.gth domain. When m = 0, (103) simplifies to give

ci>(E) = Inf (Whom (p, E) ( ~~(~))) 112 (107)

p(x) ~ 0

The right-hand side of (107) is a positively homogeneous function at degree one with
respect to E. It is therefore the dissipation function of a convex set which can be
determined explicitly (SUQUET [51]). Set

P(JJ) = { ~ E R~xJ such that: W_"om(JJ, ~) ~ ( :;~:~)}.

Taking the infimum over JJ in (107) is equivalent to taking the intersection between the
sets P(JJ ):
phom = n P(JJ).
p(x) ~ 0
This last expression can be useful when one is looking for phom, rather than cl>, and
any choice of JJ gives an estimate of phom from the outside. A few applications of (107)
are examined now.
Effective Behavior of Nonlinear Composites 243

4.3.2 Two-phase isotropic composites

When m = 0, (105) becomes
I

p(2) + c(2)) 2 (108)

p(l)

A rigorous upper bound for the effective shear modulus of two-phase isotropic compos-
ites is given by the Hashin-Shtrikman upper bounds for general microstructures and
by the Hervb-Stolz-Zaoui upper bound for composite sphere assemblages. The infi-
mum in (108) can be computed analytically with the Hashin-Shtrikman upper bound.
Assuming that 0"~1 ) 2: 0"~2 ) for definiteness, it reads

O"{f S+
0"(2) = 1-
c(2)
{3 + {3c(2)
( 1 - {3)c(2)
+ 1 - {3 + f3c{2) ((]"((1))2
O"o2
)
+
_f3_c(2)
1 - {3
(( (1))2 1)
O"o2
(]"( )
- (109)
0 0 0

This result was first given by PONTE CASTANEDA and DE BOTTON [43]. In the case
of the Herve-Stolz-Zaoui upper bound, the infimum has to be evaluated numerically.
Estimates can also be obtained by using lower bounds or estimates for {thorn. When the
Hashin-Shtrikman lower bound is used, the optimization procedure can be performed
analytically. The estimate of the overall flow stress resulting from this calculation
is exactly (64) (65) : the prediction of the variational procedure coincides with the
prediction of the modified secant method. This is a general result proven in section
4.4.3.
The dependence of the bounds and estimates on the volume fraction of the inclusions
is examined in Figure 9 at a given value of the ratio O"~t) /0"~2 ) = 5. Surprisingly,
when c(l) approaches 1, three of the predictions do not coincide, as might be expected,
with the flow stress of phase 1. Two of the predictions are estimates (3-phase self-
consistent scheme and lower Hashin-Shtrikman bound) and cannot be used to draw
definite conclusions on the actual overall flow stress of the composite. However the
prediction based on the Hervb-Stolz-Zaoui upper bound is a rigorously accurate upper
bound, which proves that the flow stress in any assemblage of composite spheres does
not tend to the flow stress in phase 1 when the proportion of phase 1 tends to 1.
This effect is typical of perfect plasticity. It is reminiscent of an earlier result by
DRUCKER [11]. In a rigid-plastic composite, if a shear plane can be passed through
the weakest phase then the shear strength of the composite is the shear strength of the
weakest phase regardless of the volume fraction of the phases. This result indicates
that surfaces play a significant role in plasticity. In the present situation, although the
volume fraction of the matrix tends to 0, its "surface" (which reduces to the surface of
the spheres of phase 1) does not tend to zero.
 244 P. Suquet

-·?

c(ll
- - Upper bound: Voigt - Upper bound: H-S-Z --- Estimate: Three-phase
.......... Upper bound: HS+ · · · ·.. Estimate: HS_ self-consistent

Figure 9: Isotrop~c two-phase composites. Rigid plastic phases (m = 0). Contrast

17~ 1 ) I 17~2 ) = 5. Predictions of the variational procedure with different upper bounds and
estimates for 1-lhom.

4.3.3 Unidirectional composites

The composites considered in this section are unidirectional composites (aligned in the
third direction) composed of two rigid-plastic phases. Phase 1 consists of cylindrical
fibers with a circular cross section dispersed into phase 2 which is the matrix (see
Figure 2 for an example of the configurations considered here).
The extremal surface (boundary of phom) is determined numerically according to
MARIGO et al [31] and MICHEL and SUQUET [33]. Each phase is assumed to be given
an elastic ideally plastic behavior and, for a prescribed radial direction in the space
of stress, the response of the composite is computed along this direction of loading.
The overall stress reaches an asymptotic value which is on the extremal surface. The
overall stresses under consideration consist of the superposition of a uniaxial tension
and a transverse shear (this example was first considered by PoNTE CASTANEDA and
DE BOTTON [43])

Three contrast ratios between the strengths of the two phases 17~ 1 ) I 17~2 ) have been
investigated: 17a1 ) 117~2 ) = 2, 5, 10. For 172) 117a2 ) = 2, the computations were performed
with 11 configurations ([36]). For the other two ratios, the computations were per-
formed on a single configuration representative of the average of the predictions over
the whole set of configurations when 17~ 1 ) I 17~2 ) = 2. More specifically this configura-
Effective Behavior of Nonlinear Composites 245

6.0
O"oC1> I uoC2>_w
-

4.5 •
~ •
O"ou> I uoc > 5
=
N
~0 2
~ 3.0
....
~

1.5

* - HS-
• Computations
0.0 '----~-----i.....,_~--~-------'
0.0 0.375 0.75 1.125 1.5
~1l O'o(2l
Figure 10: Unidirectional composites (the direction of reinforcement is the third di-
rection). Overall stress I: = l:t (e 1 ® e1 - e2 ® e2) + l:3e3 ® e3. Extremal surface
in the plane (I:., 1:3 ) for different contrast between the phases. Comparisons between
numerical simulations (squares, circles and stars) on "random configurations" (simi-
lar to the configuration shown in Figure 2} and the estimate deduced from the lower
Hashin-Shtrikman bound and the variational procedure (from {36}).

tion approaches transverse isotropy and its overall strain/stress response is close to the
mean response of all configurations, both under multiaxialloading and under uniaxial
tension. The results are shown in Figure 10.
In all three cases, the shape of the extremal surface resembles a bimodal surface.
Bimodal surfaces have been used by HASHIN [20], DVORAK et al [14], DE BUHAN and
TALIERCIO [10], PONTE CASTANEDA and DE BOTTON [43] (among other authors) to
describe the first yield or the flow surface of unidirectional composites. The present
calculations confirm the validity of this assumption or observation. Inspecting the
failure modes at the local level reveals that the unit cell can fail under two possible
modes and confirms even further the bimodal shape of the overall flow surface. The
first mode, observable for low values of the axial stress 1:3, corresponds to shear bands
in the matrix with an inclination of approximately ±45° on the horizontal direction.
When 1:3 reaches a threshold corresponding to the vertex of the flow surface, the plastic
zone spreads throughout the unit cell. As 1:3 is increased, the plastic strain tends to be
more and more homogeneous and becomes fully homogeneous when 1:3 reaches< u 0 >.
These numerical calculations can be used to propose a closed form expression for the
 246 P. Suquet

bimodal surface :

(110)

where k(l) = 0'~1 ) / ../3 is the in-plane shear strength of phase 1, k. is the in-plane shear
strength of the composite. k. can be either fitted numerically ([36]) or taken as the
prediction of the variational procedure used with the Hashin-Shtrikman lower bound
([43]) : k. = 7JCT{/ 5- where CT{/ 5- is given by (64) (65) in dimension d = 2.

4.4 General procedure

The composites considered in this section are made up of individual constituents gov-
erned by the constitutive law (2). Their strain energy potential reads

It will be further assumed that J(r) is an increasing and concave function.

4.4.1 A simple variational upper bound.

A simple variational upper bound for ~ can be obtained by the concavity of J(r). Recall
the definition of the second moment of the equivalent strain 6~:)

(111)

The concavity of f(r) yields, for every function F,

and more specifically

Therefore

~(E) inf
vEK(E) ~
~ c(r) (!k(r)
2
< tre(v) 2 >r + < /(r)(c 2 (v)) > )
eq r
r=l

inf L c(r)
N
< ~(E)= vEK(E) .
(-21 k(r) < tre(v) 2 >r +/(r) (6~:)(v) 2 )). (112)
r=l

~ is an upper bound for ~.

Effective Behavior of Nonlinear Composites 247

4.4.2 Link with the modified secant approach.

The definition of the overall potential ~(E) involves a variational problem

<I>(E) ~ .~J!(./(v), J(v) ~ ~ c<'l GM'i <tre(v)' >, +j<>l (l~; 1 (v)'))
(113)
Under appropriate hypotheses on the growth of the function J(r), the variational prob-
lem (113) admits a unique solution u which satisfies

u E X:(E), (~~ (u), v*) = 0, Vv* E X:(O), (114)

where the derivative ~~ is the Frechet derivative of the functional J and where (,)
is to be understood in the sense of a duality product. A straightforward calculation
shows that
~(r)
2
( av
uceq *
(v), v ) = =(r) < e(v) : e(v*) >n (115)
3c:eq
and it follows that the variation of J with respect to v is
N
(~~(u), v*) = I>(r) (k(r) tre(u)tre(v*) + 2ji~rl(e~:)) < e(u): e(v*) >r), (116)
r=l

where ttir) is the secant shear modulus in phase (r) (see (3)): Jl~rl(e~:l) = ~f'(r)(e~:>\
By virtue of (116), the solution u of (114) solves the following system of equations :

u E K(E), div(u) = 0, }
(117)
u(x) k(x) tre(u) + 2J1 8 (x) e(u)(x),

where
N N
k(x) = :L k(r) x<r>(x), p.(x) = :L flir>(e~:>) x<r>(x).
r=l r=l
(117) is exactly the nonlinear problem encountered in the secant method based on
second-order moments of the strain. It can easily be checked that

~!(E)= n..,hom: E =< 0' >,

where n..,hom is the effective stiffness tensor computed in the modified secant approach.
In other terms the constitutive law predicted by the modified secant approach derives
from the potential ~.
248 P. Suquet

4.4.3 Link with Ponte Castaneda's variational procedure.

To understand the link of the above variational results with the variational procedure
of PONTE CASTANEDA ([39]), it is useful to introduce the dual function of f(r). First
J( r) (e), well defined when e ~ 0, is extended for negative values of e by - oo
f(r)(e) = f(r)(e) when e ~ 0, f(rl(e) = -oo when e ~ 0.

Jir) denotes the dual function of J(r) (in the sense of concave functions [16])

Jir) is itself a concave function such that Jir) (Jl) = -oo when J1 < 0. In addition

Then, following PONTE CASTANEDA ([40])

f(rl(£~~)(v)2) = ~"m~o (~Jl(r)£~~)(v)2 _fir) (~Jl(r))),

and

z=
N
inf c(r) (-21 k(r) < tre(v) 2 >r +-23 Jl(r) < c:;q(v) >r)
vEK(E) r=l

= ~E: Il}om: E.
2

whom is the elastic energy of a fictitious linear composite made up of phases with
shear moduli Jl(r) occupying the domains V,.. The equality (118) is nothing else than
the upper bound resulting from Ponte Castaneda's variational procedure ([39])

( 119)
Effective Behavior of Nonlinear Composites 249

with

The relation (119) coincides with (118) and the potential~ coincides with the upper
bound predicted by the variational procedure of [39]. The set of shear moduli J.l =
(1Pl, · · ·, J.l(N)) which give (118) an infimum value, satisfies

(120)

or, equivalently, noting that (J'~r)) -I = J'(r),

(121)

The set of nonlinear equations (121) to be solved in the variational procedure of PONTE
CASTANEDA is identical to the set of nonlinear equations (55) to be solved in the mod-
ified secant method. In conclusion the modified secant moduli approach coincides with
Ponte Castaneda's variational procedure (which yields results coinciding with other
variational theories in specific situations [63], [51]). In other terms, Ponte Castaneda's
variational procedure coincides with the modified secant approach.

5 Application to Metal Matrix Composites

5.1 Individual constituents
This section is devoted to Metal Matrix Composites (MMC) composed of an elastic-
plastic matrix with inclusions. The inclusions (phase 1) are assumed to be linearly
elastic. The matrix (phase 2) is assumed to be elastic plastic with isotropic hardening
and governed by a Jrflow theory

e(u) = ee + eP, ee = ~2 ) : U, 3. s
e·p = -p-, }
2 (J eq
(122)
. 3 1
p = 2 h(p) (s: s)+ when Cleq = CJy(p), p = 0 when Cleq < ay(p).

~2 ) is the initial elastic compliance of the matrix. For simplicity the elastic properties
of the matrix are assumed to be isotropic and characterized by a bulk modulus k~2 )
250 P. Suquet

and a shear modulus p~2 >. o-y(p) is the flow stress of the matrix which depends on a
scalar hardening parameter p. p is the cumulated plastic strain

p(t)
(2
= lot 3eP : eP )1/2 ds.

The hardening modulus his the slope of the curve (p, o-y) : h = aY(p).

5.2 Secant and incremental methods

5.2.1 Models based on a deformation theory
The secant methods can be applied to the deformation theory associated with the
incremental equations (122). Assuming that the stress direction ~(x, t) remains
O"eq
constant along the loading path (i.e. independent oft), (122) can be integrated in
time to give
3 s
eP = -p-.
2 O"eq
In this approximation, the constitutive law of the matrix can be written as

e = ~2 l(u): u, M2) = mo(2) .If

(2)(
Bs
)
O"eq K (123)
6 3 + 2 '
1 _1_
) =
+ 3p(aeq) .
(2) (} (2)(
mo = k(2)' • O"eq (2)
0 Jlo O"eq
p(aeq) is obtained by inverting the relation O"eq = ay(p). Both secant methods can
be applied to predict the overall response of the composite. The dual formulations in
section 2.9 are more appropriate to the form (123) of the constitutive law.

5.2.2 Model based on incremental relations

The constitutive equations of the matrix can be written in incremental form. Assuming
that there is no unloading in the matrix, the stress-rates and strain-rates are related
by
iT= L~ 2 ) : e, Lt = 3k~2 ) .If+ 2p~2 l IF+ 2.A( 2l JE, }

1 1 3 lE 3, (124)
= (2) + h(p ) ' = 2 s ® s.
A

_A(2)
Jlo
The incremental method described in section 3 can be applied, provided that the
directions of the average stress deviators are parallel in each phase

s(t) = s( 2> = s.
Effective Behavior of Nonlinear Composites 251

5.3 A model based on the 'Iransformation Field Analysis

For elastic-plastic constituents an alternate method based on the Transformation Field
Analysis (TFA) has been proposed by DVORAK eta/ [13] [15]. The method is recalled in
the general context of aN-phase composite. The plastic strain field can be considered
a field of eigenstrains according to the first equation in (122). The stress field lT can be
determined (at least formally) from the field of eigenstrains e'P by solving an elasticity
problem with eigenstrains. Then the plastic strain field is updated by the flow rule
(122). The overall plastic strain is then given by ([25], [49])

EP =< JRl : e'P >,

where !a denotes the elastic stress concentration tensor. When the field e'P varies from
one point to another, the determination of the stress field through the resolution of an
elastic problem with eigenstrains leads to an integral equation which cannot be solved
analytically. An approximation is introduced by assuming that the plastic strain is
constant within each constituent
N
e'P(x) =L EP(r) x(rl(x). (125)
r=l
The overall plastic strain is then

L c(r)Ja(r)T: E*l,
N
EP = Ja(r) =< Ja(x) >r.
r=l
The evolution of the plastic strain EP(r) is assumed to be governed by the average stress
over phase r
3 ·(r) _S(r)
Ep(r) _ _
- 2 p ._,(r)'
p(r) = ~
2
1
h(r)(p(r))
(s(r) .· s(r)) + '
L...eq
(126)
Under the assumption (125), the determination of the stress field u and of the average
stresses ~(r) can be viewed as a thermoelasticity problem for an N-phase composite
(with 6 independent thermal expansion coefficients). For a two-phase composite the
problem simplifies further and the average stresses in each phase can be expressed
in terms of the plastic strains EP(t) and EP( 2l ([12]) by means of the elastic stress
concentration tensors Ja(r)

~(1) = Ja{l) ~ + (Ja(l)- rrd)

= = (M(2)- M{l)rl = (Ep(2)- EP{ll), }
~(2) = Ja(2) ~ + (Ja(2) - rrd)
= = (M(2) - M{l)rl = (EP(2)- EP{ll). (127)

When the inclusions are purely elastic (i.e. EP(t) = 0) the average stress in the matrix
is given by
 252 P. Suquet

and the equations governing the overall and local plastic strains are
3 S( 2 )
EP(2) - - ·(2) -
- 2p ~(2)'
"-'eq
When the composite is isotropic the concentration tensor simplifies and the above
relation reads
b(2)
S( 2) = b( 2)S +2 eq - 1 EP( 2)
eq 0(2) - 0(1) '

where Ja( 2 ) = b~) .If + b~~)K The scalars b~) and b~~) (given for instance by (73) when
the hashin-Shtrikman lower bound is used) can be expressed in terms of the bulk and
shear compliances of the individual phases and of the composite. The last relation can
be rearranged to show the nature of the overall hardening of the composite (at least as
predicted by the model). Define

X= HE", 2 ( 1 )
H = c(2) b~~)
2
b~2q) -
0(2) - 0(1)'
1 O'hom(P) _ _
y
1 0'(2)
- b(2)
eq
y
(-p-)
C
(2)b(2)
eq
.

Then the overall constitutive law reads

E = Mhom : :E + E", · 3 · S-X }

E"=2P(S-X)e9 '
{128)
X= HE", (S- X)eq ~ O'~om(P).
The overall hardening of the composite as predicted by the TFA model when the matrix
is elastic plastic with isotropic hardening is the result of two contributions:
- a linear kinematic hardening due to the back stress X,
- an isotropic hardening due to the dependence of the yield function afom on the
cumulated plastic strain P.

5.4 An incremental model based on the variational procedure

A simple incremental model which applies to metal matrix composites reinforced by
particles can be proposed by neglecting the deformation of the particles. First recall
the construction of constitutive laws such as {122) by two thermodynamic potentials.
The model is defined by three state variables, e, e" and p in this specific situation
and two thermodynamic potentials, the Helmholtz free energy 1l and the dissipation
potential V of the material. The free energy is split into two terms, the elastic energy,
recoverable by unloading the material, and the stored energy:

1l(e,e",p) = 21 (e- e"): L! 2l: (e- e") + ~

Stored energy
' r.p(p) = 1" O'y(p) dp.
Elastic energy
Effective Behavior of Nonlinear Composites 253

The thermodynamic forces are defined by the state laws as

_ a11.. _ a'H..
R = --
ap = -ay(p),
(2). e
lT - - aeP - Lo . e ' (129)

The power dissipated in the irreversible processes taking place in the material is

v = u : eP + RfJ.
The evolution of the internal variables describing irreversible phenomena are given by
the complementary laws :

Yield function
F(u, R) = aeq + R,
flow rule
,a;:( R)
·p_
E: - A au (T l l

The approximate model for the overall behavior of MMC's is also based on the descrip-
tion of dissipative constitutive laws by two overall potentials, the overall free energy
of the composite and its dissipation potential. The approximations introduced to es-
timate these potentials are based on the observation that one can distinguish three
regimes in the deformation of the composites :

1. Elastic regime: the plastic strains in the matrix are smaller than the elastic
strains.

2. Elastic-plastic regime: the elastic and plastic strains in the matrix are of the same
order.

3. Plastic regime: the elastic strains in the matrix and in the inclusions are smaller
than the plastic strains in the matrix.

The elastic regime can be handled by any linear theory appropriate for the microstruc-
ture of the composite providing the overall stiffness JL,hom of the composite and, conse-
quently, its overall elastic energy. The elastic-plastic regime is neglected in the present
model. The plastic regime is handled by considering the particles as rigid inclusions
in a plastic matrix with hardening. The overall stored energy of the composite is es-
timated by considering that the matrix is a nonlinear elastic material with a strain
energy density ~.p( e:) (deformation theory of plasticity with elastic deformations ne-
glected). When the composite has overall isotropy, the overall stored energy potential
can be estimated by the right hand-side of (96) which simplifies due to the isotropy of
the linear comparison composite :

hom) 1/2 )
~( P) = c(2)i.p ( ( ~2) tto p ' (130)
 254 P. Suquet

where Jlhom is the effective shear modulus of a linear composite made up of an incom-
pressible matrix with shear modulus 11° containing rigid inclusions. The expression
(130) makes use of the fact that the overall strain reduces to the overall plastic strain
in the plastic regime E ~ EP and E~q = P. The two terms entering the Helmholtz free
energy of the composite, the elastic energy, and the stored energy are estimated

The state laws read

81-lhom
~ = ___ = Lhom , (E _ EP)
8EP . '
hom ) 1/2
C= ( ~ (131)
c( ) Jlo
2

The overall yield function :Fhom is the usual Von Mises criterion :

and the normality law at the overall scale reads

Finally, the proposed effective constitutive law of the composite is

E = Ee + EP, Ee = Mhom : :E 1 }

. 3 . s h • ( 132)
EP = -P~ when I:eq = 17/m(P), EP = 0 when I:eq < 0'"~0 m(P).
2 L..Jeq
This approximate model requires only one constant (C in (131)) which is deduced from
a linear theory.

5.5 Discussion
The above models were implemented with the following material constants

= 400 GPa,
E(l) v( 1 ) = 0.2,
£(2) = 75 GPa, v( 2 ) = 0.3,

O'"y(p) = O"o + Ap", O"o = 75 MPa, A= 416 MPa, o: = 0.3895.

The FEM cell calculations were performed with spherical particles. The overall elastic
stiffnesses used in the theoretical predictions (and therefore the corresponding stress
concentration factors) were computed by the Hashin-Shtrikman lower bound (secant
methods) or the Willis lower bound (incremental procedure).
 Effective Behavior of Nonlinear Composites 255

~0~----~~r-------~----------------,
c(IJ=0.3
Hashin-Shtrikman & Willis lower bound
--- Matrix

300

~200

-------------------------------
-------::.__ Transformation Field Analysis
/,,........ ............ Flow theory + Incremental formulation
100
I ·-·-·- F.E.M.
f ...... · Classical secant formulation
- - Modified secant formulation
Model (132)
0 L-----------------~----------------~
0.0 0.02 0.04
E
Figure 11: Metal matrix composite. Spherical inclusions dispersed in a matrix. Volume
fraction of inclusions c(l) = 0.3. The inclusions are linear elastic. The matrix is
governed by a J 2 -fiow theory with isotropic hardening. Uniaxial tension. Predictions
of the overall response of the composite by several schemes. The overall tensors are
calculated with the Hashin-Shtrikman or the Willis lower bound.

5.5.1 Monotone loadings

The results of the predictions are shown in Figure 11. The model with elastic accomo-
dation and the prediction of the incremental procedure are very stiff. This feature of
the incremental procedure has already been noted and will not be discussed further.
To understand how the model based on the TFA behaves by comparison with other
models, consider the case of rigid particles in an incompressible matrix. The kinematic
hardening modulus H and the overall yield stress are in this case
3 (1) + 2c( 2 ) (5c(ll + 2c(2l )
H = _.:_) 2 ) (5c(ll
2 c(2)f"'0
+ 2c( 2l) ' hom
G'y
_
(P)-
5c(ll
2
( 2)
G'y 2d2l p (133)

The kinematic hardening modulus H is about five times higher than the elastic shear
modulus of the matrix when c(ll = c( 2 ). This value is excessively high. It has long been
recognized (HILL [23], BERVEILLER and ZAOUI [2]) that the models based on elastic
accomodation, i.e. using the concentration tensors derived from the elastic properties
of the constituents, are too stiff. The model can be accurate provided each phase is
subdivided into subdomains where the plastic strain is assumed to be constant. This
was done by DVORAK et al [15] and gave accurate results together with a low computing
time. It is exact in the limit where each point xis considered a subdomain. However,
 256 P. Suquet

the subdivision into subdomains requires precise knowledge of the microstructure, such
as a periodic structure with a known geometry which is the situation considered in [15].
The model can therefore be seen as an alternative to FEM calculations.
The three other schemes (classical and modified secant methods and the approxi-
mate scheme (132) are in good agreement with the FEM computations. The prediction
of the modified secant method seems to slightly underestimate the stiffness of the com-
posite. This is because the Hashin-Shtrikman lower bound was chosen to estimate
the elastic properties of the linear comparison composite. As shown in Figure 7, the
three-phase self-consistent scheme would be more appropriate and would lead to a
more accurate description by the modified secant method. With this scheme however,
no explicit result would be available for the incremental method (where the tangent
moduli are anisotropic).

5.5.2 Cyclic loadings

············· ...............................
400

200

~~ 0 t-----t-----f-----f------1

-200

-400

-0.08 -0.04 0.04 0.08

- F.E.M. present model

Figure 12: Spherical inclusions dispersed in a nonlinear matrix. Cyclic uniaxial ten-
sion. Volume fraction of inclusions c< 1l = 0.3. The inclusions are linear elastic. The
matrix is governed by a J 2 -flow theory with isotropic hardening. Predictions of the
overall response of the composite by the approximate model {132).
Effective Behavior of Nonlinear Composites· 257

To further check the accuracy of the approximate model the case of a cyclic loading
has been considered. The results after one cycle are shown in Figure 12. Again the
predictions of the approximate model are satisfactory. The overall hardening of the
composite is well rendered except for the part of the stress/strain curve where kine-
matic hardening is observed on the F.E.M. calculations (yielding in compression). The
relative accuracy of the simple model (132) is rather surprising since one would expect
a significant kinematic hardening (due to internal stresses). When the particles become
fibers, explicit calculations can be performed (under uniaxial tension in the direction of
the fibers). Such calculations show that the elastic stresses in the fibers create residual
stresses in the matrix which are responsible for an overall kinematic hardening. This
effect is less pronounced for spherical particles (considered here).

6 Final remarks
A few conclusions can be drawn from the comparison between these schemes:

1. The predictions of the incremental method seem to be very stiff by compari-

son with the predictions of the secant methods and by comparison with FEM
simulations.

2. The predictions of the model based on the TFA are also very stiff when used with
no subdivision of the phases into subdomains. This subdivision is possible when
the information on the microstructure is precise enough. The accuracy of the
TFA increases with the number of subdomains considered in the analysis (the
model is exact in the limit of infinitely many domains).

3. The definite advantage of the TFA on the other methods, and especially on the
secant methods, is that it provides a full constitutive law applying to any loading
path, including unloading.

4. The main advantage of the secant methods is the good qualitative agreement
between their predictions and the results obtained by F.E.M. calculations. The
predictions of the classical and modified secant methods, based on the first-order
moments or second-order moments of the strain in each individual phase, are
rather close to one another for metal matrix composites. They differ significantly
for porous materials and for materials with strong nonlinearities. The constitutive
law predicted by the modified method has been found to derive from an overall
potential. The modified method is therefore recommended. The main limitation
of the secant methods is that they are based on a deformation theory of Plasticity
which is questionable for complex microstructures (local unloading can take place
under overall monotone loading) and inapplicable for complex loading paths. The
secant methods provide a rather good description of the composite response under
monotone loading but do not provide a full constitutive law for the composite.
 250 P. Suquet

5. The variational procedure of PONTE CASTANEDA has been developed indepen-

dently of the secant procedures. It coincides with a secant method based on the
second-order moment of the strain. Most materials are neither purely elastic nor
purely viscoplastic and the domain of application of the variational methods can
appear as rather limited. If one keeps in mind the construction of constitutive
laws from two thermodynamic potentials it is expected that exact results obtained
for simple systems (nonlinearly elastic or purely viscoplastic) could be extrapo-
lated to derive useful information on the overall thermodynamic potentials of
real composites. An example has been given in this text where the variational
procedure has been used to propose a simple, but accurate, model for the overall
behavior of particulate MMC's.

Many questions have not been adressed here, either because they are beyond the scope
of these notes or because they are still open. Briefly, these include :

1. The variational procedure gives upper bounds for the overall potential. The
question of finding improved lower bounds remains open (see however the recent
contribution of TALBOT and WILLIS [58]).

2. The incremental procedure has some nice qualitative features. It predicts an

influence of the loading path and an influence of the third invariant of the stress.
However its predictions with the average strain taken as the effective strain are
not in good quantitative agreement with the numerical results. Is it possible to
define an adequate "effective strain" leading to more accurate predictions? This
question is closely related to the proper definition of a nonlinear self-consistent
scheme respecting the variational bounds when they can be applied.

3. The approximate model proposed in section 5 presumes that the overall hardening
of the composite is only isotropic. This is a crude approximation. Is it possible
to estimate the part of isotropic and kinematic hardening in an MMC and to
model the kinematic hardening using the variational procedure?

Acknowledgements. I thank P. PONTE CASTANEDA for helpful discussions about

the relative merits of the secant and the variational methods, G. DVORAK for his
comments on section 5 of the manuscript, and J.C. MICHEL and H. MOULINEC for
their help in the numerical simulations shown here. Part of the computations were
carried out at the Institut Mediterraneen de Technologie in Marseille; the funds were
provided by the PACA region.
 Effective Behavior of Nonlinear Composites 259

References
[1] BAO G., HUTCHINSON J., and McMEEKING R.: The flow stress of dual-phase,
non-hardening solids. Mech. Materials, 12, 85-94, 1991.

[2] BERVEILLER M. and ZAOUI A.: An extension of the self-consistent scheme to

plastically-flowing polycrystals. J. Mech. Phys. Solids, 26, 325-344, 1979.

[3] BISHOP J. and HILL R.: A theory of the plastic distortion of a polycrystalline
aggregate under combined stresses. Phil. Mag., 42, 414-427, 1951.

[4] BORNERT M., STOLZ C., and ZAOUI A.: Morphologically representative pattern-
based bounding in elasticity. J. Mech. Phys. Solids, 44:307-331, 1996.

[5] BoUCHITTE G. and SUQUET P.: Homogenization, plasticity and yield design. In
Dal Maso G. and Dell'Antonio G. (eds), Composite Media and Homogenization
Theory, pp 107-133. Birkhauser, 1991.

[6] BURYACHENKO V.: Effective strength properties of elastic physically nonlinear

composites. In Marigo J.J. and Rousselier G. (eds), Micromechanics of Materials,
pp 567-578. Eyrolles, 1993.

[7] BURYACHENKO V.: The overall elastoplastic behavior of multiphase materials

with isotropic components. submitted, 1995.

[8] CHRISTENSEN R. and Lo K.: Solutions for effective shear properties in three
phase sphere and cylinder models. J. Mech. Phys. Solids, 27, 315-330, 1979.

[9] CHU T. and HASHIN Z.: Plastic behavior of composites and porous media under
isotropic stress. Int. J. Engng Sci., 9, 971-994, 1971.

[10] DE BUHAN P. and TALIERCIO A.: A homogenization approach to the yield

strength of composites. European J. Mechanics: A/ Solids, 10, 129-154, 1991.

[11] DRUCKER D.: On minimum weight design and strength of non-homogeneous

plastic bodies. In Olszag W. (ed.), Non-homogeneity in Elasticity and Plasticity.
Pergamon Press, New-York, 1959.

[12] DVORAK G.: On uniform fields in heterogeneous media. Proc. R. Soc. Lond. A,
431, 89-110, 1990.

[13] DVORAK G.: Transformation field analysis of inelastic composite materials. Proc.
R. Soc. Lond. A, 437, 311-327, 1992.

[14] DVORAK G., BAHEI-EL-DIN Y., MACHERET Y., and LIU C.: An experimental
study of elastic-plastic behavior of a fibrous boron-aluminum composite. J. Mech.
Phys. Solids, 36, 655-687, 1988.
 260 P. Suquet

[15] DVORAK G., BAHEI-EL-DIN Y., and WAFA A.: The modeling of inelastic com-
posite materials with the transformation field analysis. Modelling Simul. Mater.
Sci. Eng, 2, 571-586, 1994.

[16] EKELAND I. and TEMAM R.: Analyse convexe et problemes variationnels. Dunod,
Gauthier-Villars, Paris, 1974.

[17] GARAJEU M.: Private communication.

[18] GILORMINI P.: lnsuffisance de !'extension classique du modele auto-coherent au

comportement non lineaire. C. R. Acad. Sc. Paris lib, 320, 115-122, 1995.

[19] HASHIN Z.: The elastic moduli of heterogeneous materials. J. Appl. Mech., 29,
143-150, 1962.

[20] HASHIN Z.: Failure criteria for unidirectional fiber composites. J. Appl. Mech.,
4 7' 329-334, 1980.

[21] HASH IN Z. and SHTRIKMAN S.: A variational approach to the theory of the elastic
behavior of multiphase materials. J. Mech. Phys. Solids, 11, 127-140, 1963.

[22] HERVE E., STOLZ C., and ZAOUI A.: A propos de ]'assemblage des spheres
composites de Hashin. C. R. Acad. Sc. Paris, II, 313 , 857-862, 1991.

[23] HILL R.: Continuum micro-mechanics of elastoplastic polycrystals. J. Mech. Phys.

Solids, 13, 89-101, 1965.

[24] HILL R.: A self-consistent mechanics of composite materials. J. Mech. Phys.

Solids, 13, 213-222, 1965.

[25] HILL R.: The essential structure of constitutive laws for metal composites and
polycrystals. J. Mech. Phys. Solids, 15, 79-95, 1967.

[26] HUTCHINSON J .: Bounds and self-consistent estimates for creep of poly crystalline
materials. Proc. Roy. Soc. Lond., A348, 101-127, 1976.

[27] KREHER W.: Residual stresses and stored elastic energy of composites and poly-
crystals. J. Mech. Phys. Solids, 38, 115-128, 1990.

[28] KRONER E.: Bounds for effective elastic moduli of disordered materials. J. Mech.
Phys. Solids, 25, 137-155, 1977.

[29] LEVIN V.: Thermal expansion coefficients of heterogeneous materials. Mekh.

Tverd. Tela, 2, 83-94, 1967.

[30] MARCELLINI P.: Periodic solutions and homogenization of nonlinear variational

problems. Ann. Mat. Pura Appl., 117, 139-152, 1978.
Effective Behavior of Nonlinear Composites 261

[31] MARIGO J., MIALON P., MICHEL J., and SUQUET P.: Plasticite et homogeneisa-
tion : un exemple de prevision des charges limites d'une structure periodiquement
heterogene. J. Mica. Th. Appl., 6, 47-75, 1987.

[32] MICHEL J. and SUQUET P.: The constitutive law of nonlinear viscous and porous
materials. J. Mech. Phys. Solids, 40, 783 - 812, 1992.

[33] MICHEL J. and SUQUET P.: On the strength of composite materials: variational
bounds and numerical aspects. In Bendsoe M.P. and Mota-Soares C. (eds), Topol-
ogy Design of Structures, pp 355-374. Kluwer Pub., 1993.

[34] MILTON G. and KOHN R.: Variational bounds of the effective moduli of
anisotropic composites. J. Mech. Phys. Solids, 36, 597-629, 1988.

[35] MoULINEC H. and SUQUET P.: A fast numerical method for computing the linear
and nonlinear properties of composites. C. R. Acad. Sc. Paris II, 318, 1417-1423,
1994.

[36] MoULINEC H. and SUQUET P.: A FFT-based numerical method for computing
the mechanical properties of composites from images of their microstructure. In
Pyrz R. (ed. ), Microstructure-Property Interactions in Composite Materials, pp
235-246. Kluwer Acad. Pub., 1995.

[37] MoULINEC H. and SUQUET P.: A numerical method for computing the overall
response of nonlinear composites with complex microstructure. submitted, 1996.

[38] OLSON T.: Improvements on Taylor's upper bound for rigid-plastic bodies. Mater.
Sc. Engng, A 175, 15-20, 1994.

[39] PONTE CASTANEDA P.: The effective mechanical properties of nonlinear isotropic
composites. J. Mech. Phys. Solids, 39, 45-71, 1991.

[40] PoNTE CASTANEDA P.: New variational principles in plasticity and their appli-
cation to composite materials. J. Mech. Phys. Solids, 40, 1757-1788, 1992.

[41] PONTE CASTANEDA P.: Exact second-order estimates for the effective mechanical
properties of nonlinear composite materials. J. Mech. Phys. Solids, 44, 827-862,
1996.

[42] PoNTE CASTANEDA P.: A second-order theory for nonlinear composite materials.
C.R. Acad. Sc. Paris, 322, Serie II b, 3-10, 1996.

[43] PoNTE CASTANEDA P. and DEBOTTON G.: On the homogenized yield strength
of two-phase composites. Proc. R. Soc. London A, 438, 419-431, 1992.

[44] PONTE CASTANEDA P. and NEBOZHYN M. : Exact second-order estimates of the

self-consistent type for nonlinear composite materials. submitted, 1996.
 262 P. Suquet

[45] PONTE CASTANEDA P. and SUQUET P.: On the effective mechanical behavior of
weakly inhomogeneous nonlinear materials. Eur. J. Mech. A/Solids, 14, 205-236,
1995.

[46] PONTE CASTANEDA P. and WILLIS J.: The effect of spatial distribution on the
effective behaviour of composite materials and cracked media. J. Mech. Phys.
Solids, 43, 1919-1951, 1995.

[47] QIU Y. and WENG G.: A theory of plasticity for porous materials and particle-
reinforced composites. J. Appl. Mech., 59, 261-268, 1992.

[48] SUQUET P.: Analyse limite et homogeneisation. C.R. Acad. Sc. Paris, II, 296,
1355-1358, 1983.

[49] SUQUET P.: Elements of homogenization for inelastic solid mechanics. In Sanchez-
Palencia E. and Zaoui A. (eds), Homogenization Techniques for Composite Media,
Lecture Nutes in Physics 272, pp 193-278. Springer Verlag, 1987.

[50] SUQUET P.: On bounds for the overall potential of power law materials containing
voids with arbitrary shape. Mech. Res. Comm., 19, 51-58, 1992.

[51] SUQUET P.: Overall potentials and extremal surfaces of power law or ideally
materials. J. Mech. Phys. Solids, 41, 981-1002, 1993.

[52] SUQUET P.: Overall properties of nonlinear composites: a modified secant moduli
theory and its link with Ponte Castaneda's nonlinear variational procedure. C.R.
Acad. Sc. Paris, 320, Serie lib, 563-571, 1995.

[53] SUQUET P. and PONTE CASTANEDA P.: Small-contrast perturbation expansions

for the effective properties of nonlinear composites. C.R. Acad. Sc. Paris, 317,
Serie II, 1515-1522, 1993.

[54] SUQUET P.: Overall properties of nonlinear composites : Remarks on secant and
incremental formulations. In Pineau A. and Zaoui A. (eds), Plasticity and Damage
of Multiphase materials, pp 149-156. Kluwer Acad. Pub., 1996.

[55] SUQUET P : Overall properties of nonlinear composites : secant moduli theo-

ries and variational bounds. In Markov K. (ed.), Continuum Models of Discrete
Systems 8, pp 290-299. World Scientific, 1996.

[56] TALBOT D. and WILLIS J.: Variational principles for inhomogeneous nonlinear
media. IMA J. Appl. Math., 35, 39-54, 1985.

[57] TALBOT D. and WILLIS J.: Some explicit bounds for the overall behavior of
nonlinear composites. Int. J. Solids Struct., 29, 1981-1987, 1992.
 Effective Behavior of Nonlinear Composites 263

[58] TALBOT D. and WILLIS J.: Upper and lower bounds for the overall properties of
a nonHnear composite dielectric. I. random microgeometry. Proc. R. Soc. Lond.
A, 447, 365-384, 1994.

[59] TANDON G. and WENG G.: A theory of particle-reinforced plasticity. J. Appl.

Mech., 55, 126-135, 1988.

[60] WALPOLE L.: On bounds for the overall elastic moduli of inhomogeneous systems
-I and II. J. Mech. Phys. Solids, 14,151-162 and 289-301, 1966.

[61] WILLIS J .: Bounds and self-consistent estimates for the overall moduli of
anisotropic composites. J. Mech. Phys. Solids, 25, 185-202, 1977.

[62] WILLIS J .: The structure of overall constitutive relations for a class of nonlinear
composites. IMA J. Appl. Math., 43, 231-242, 1989.

[63] WILLIS J.: On methods for bounding the overall properties of nonlinear compos-
ites. J. Mech. Phys. Solids, 39, 73-86, 1991.

[64] ZAOUI A. This volume.

Appendix 1. Proof of (54)

A direct proof of (54) can be found in KREHER [27]. We give here a different and less
direct proof, valid for more general materials (including nonlinear materials), based on
a general result about the derivation of the overall energy of a composite with respect
to a parameter. This result was used by SUQUET and PONTE CASTANEDA [53] (see
also [45]) to obtain small contrast expansions of the overall properties of composites.
Consider a composite comprised of N linear or nonlinear phases individually governed
by potentials cp(r) depending on a scalar parameter t. Then the local field e 1 and the
overall potential depend on t

cfl(t, E)=< cp(t, et) >= inf < cp(t, e(v)) > . (134)
vEK(E)

Then
(135)

e e
To prove this result, let 1 denote the derivative of e 1 with respect to t. 1 = e(u1)
is a compatible strain field and its average is zero (since E does not depend on t)
Differentiating (134) with respect tot yields:

(136)
264 P. Suquet

The stress field tr 1 = ~(t, et) satisfies the equilibrium equations div(u 1 ) = 0 in V.
According to Hill's lem~a:

This completes the proof of (135). This result can be used to derive (54) by considering
N
t= L~jLh, cp(t, e) = L ~CijL~jLhCkh x(s)(x),
s=l

~~ (t, Et) = ~CijCkh x(r)(x), <I>(t, E)= ~E: n..,hom :E.

Then, according to ( 135)

1 an...,hom 1 c(r)
-E
2 ·• -
8L(r) - 2 < t:··t:kh
- ·• E-- IJ
x(rl(x) >--
- 2 < t:··t:kh
IJ
>r •
ijkh

The second moment of the strain field in phase (r) is given by

1 an...,hom
< CijCkh >r= c(r) E : -(-r)- : E. (137)
8Liikh

If the phases are isotropic

and

1 an...,hom 1 an...,hom aLnh

c(r) E : 81-L(r) : E = c(r) E : 8L~jlh : E ai-L(r) =< CijCkh >r 2Kijkh = 3 < €~q >r •

This completes the proof of (54).

 DYNAMICS OF COMPOSITES

J.R. Willis
University of Cambridge, Cambridge, UK

ABSTRACT
These notes provide a brief account of how to analyze the propagation of ultrasonic
waves in a composite; the methods and theory apply equally well to problems on the
terrestrial scale, such as the propagation of seismic waves through regions containing
arrays of cracks. The methods are less advanced, and the results less precise, than those
available for the statics of composites. Furthermore, a wider class of phenomena needs
to be addressed. Here, attention is restricted to methods that have their counterparts
in statics, to which they reduce in the limit of very slow deformations.
The basic equations of elastodynamics are first introduced, together with some fun-
damental concepts: plane waves, Green's functions, and variational principles. Then,
waves in random media are introduced, using perturbation theory, valid for weak vari-
ations in material properties. This is not the case of most practical interest, but it
is one for which "honest" information can be generated, so allowing some of the phe-
nomena to be expected in all randomly inhomogeneous media to be recognised. Later
sections treat dilute suspensions, and then the general case, for which the best that can
be hoped for is a strategy for developing systematic approximations. Approximations
based on variational principles are advocated, both for their intrinsic merit and because
they reproduce, in the static limit, results of proven utility, developed elsewhere in this
course.
Final sections discuss waves in composites which display nonlinear behaviour. Re-
search in this area has hardly been started, so only some preliminary indications can be
given. The main difficulty is that the nonlinearity of the composite, which is not known
a priori, exerts a crucial influence on the wave that is excited by the applied loading:
this cannot be assumed to be time-harmonic, for example, so the whole problem has to
be approached by a "bootstrap" procedure, in which mean constitutive relations and
the mean wave that is propagated through the medium have somehow to be determined
concurrently. One example, worked out very recently, is discussed for illustration. This
concerns a linearly elastic medium, containing cracks. The response of the cracks is
nonlinear, because the cracks are open during tension but closed during compression.
The entire medium is thereby rendered nonlinear, and all of the challenges alluded to
above have to be confronted.
266 J.R. Willis

1. BASIC CONCEPTS FROM ELASTODYNAMICS

1.1. Governing equations
The discussion is confined to the case of small deformations. Subject to this re-
striction, the equations of motion of any continuum can be written

-a
aaij

Xj
.
+ f; = p;, or divu + f = p, (1.1)

where u (with components CT;j) denotes Cauchy stress, f is body force and p is mo-
mentum density. The superposed dot represents differentiation with respect to time,
t.
It is required now to add a specification of the 111edium, in the form of constitu-
tive equations. One is so common that it is not usually thought of in these terms:
momentum density is mass times velocity, or

p=pu, (1.2)

where u denotes particle displacement. For elastic media (the only media considered
here), stress is related to strain by

(1.3)

where the strain tensor e has components

e;j = ~ (Ui,j + Uj,i), (1.4)

the subscript ,i representing differentiation with respect to x;. The tensor of elastic
constants c has the usual symmetries cijkl = cjikl = cklij·
Substitution of (1.2) and (1.3) together with (1.4) into the equations of motion (1.1)
gives
(1.5)
These equations govern the displacement field u(x, t) in a body, and have to be solved
over the domain n occupied by the body, for all t > 0, subject to initial conditions
prescribed at t = 0 and boundary conditions prescribed over the boundary an, for all
t > 0. In general, the body is heterogeneous, so that C and p are functions of position
x in the body.
1.2. Plane waves
Useful insight can be gained about the system of equations (1.5) by considering the
particular case of a uniform body, occupying all space (so that C and pare constants).
The equations then admit plane wave solutions, for which u has the form

u = Vf(t- n.xjc), (1.6)

Dynamics of Composites 267

where f is an arbitrary function, U and n are unit vectors, and c is a constant to be

determined (the speed of the wave, which propagates in the direction n). The form
(1.6) satisfies (1.5) if

(1. 7)

where the matrix L(n) has ik element C;1k 1n1n 1• It is symmetric and (by hypothesis)
positive definite, and so has three orthogonal eigenvectors UN (which therefore can
be taken orthonormal) and corresponding eigenvalues pc~ (N = 1, 2, 3). Such values
define plane waves which can propagate in the medium.
In the special case of an isotropic medium,

(1.8)

and it is easy to prove that

ci = (>. + 2fl)fp, U 1 = n, (1.9)

while
(1.10)
The waves with N = 1 are longitudinal, or P-waves. Those with N = 2 or 3 are shear
waves, or S-waves. Their associated "wave polarizations" U span the two-dimensional
subspace CJrthogonal to n, corresponding to the double eigenvalue.
1.3. Greerr's function
The response of a medium to a point impulse plays a crucial role in later devel-
opments. This is the Green's function (or tensor), G(x,x',t), whose component G;p
represents the i-component of displacement at x generated by a body force with i-
component /; = O;pO(x- x')o(t). Thus, it satisfies the equations

(1.11)

together with the initial condition G =0 for all t < 0, and suitable homogeneous
boundary conditions over 80. (the simplest being zero displacements, so that G(x, x', t) =
0 for x E 80.).
If G could be determined, then all problems would be solved! Of course this cannot
be done. A valuable building block, however, is provided by the Green's function
G 00 (x, t) for an infinite uniform medium. Translation-invariance allows the point force
to be placed at x' = 0, without loss of generality: the response to a force at general x'
is then G 00 (x- x', t).
Equations (1.11) will now be solved, in the special case of an infinite, uniform body,
with x' = 0. The Fourier transform of G 00 follows immediately. With the definition

G 00 (e,w) = j dx j dtG 00 (x,t)exp[i(e.x+wt)], (1.12)

268 J.R. Willis

equation (1.11) implies (in matrix notation)

(L(e) - pu?I)G 00 =I, (1.13)
so that
(1.14)
Causality is assured by regarding w as having small positive imaginary part. Then,
inversion with respect to w gives the value zero for G 00 for all t < 0. The inverse of
the matrix in (1.14) can be expressed in terms of the eigenvalues and eigenvectors of
L(e), introduced in the context of plane waves. Thus,
-00 - 3 UN(UNf
:L
G (e,w)- N=l p (iel2 eN
2 - W2)' (1.15)

Here, UN and eN are evaluated for the unit vector n = e/lel.

It is impossible to invert the transform completely in the general case but some
simplification can be effected. The inverse transform with respect toe can be given in
the form

Goo(x, w) = (2nt3 t f
N=l l1e1=1
dSVN (VNf foo dRR22 expl-i(e~x)~J.
lo peN(R - w feN)
(1.16)

Evaluation of the integral with respect toR is fairly straightforward. First, because
e
points and -e in the integrand contribute equally, it is possible to extend the range
of integration with respect toR from -oo to oo, and to compensate by dividing by 2.
Now,

1/oo dRR2 exp[-i(e.x)R] = 1 joo dR [-'(c )R] w2 joo dRexp[-i(e.x)R]

2
-oo R2 - w / eN
2 2
-oo exp 1 ., .x + 2eN2 -oo R2 - w2j eN
2

(1.17)

the second integral having been evaluated using Cauchy's theory of residues. The
consequent representation for the time-reduced Green's function

G""(x,w)
1
= -8 2 L (:
3
1. dS
UN (UNf {
2 o(e.x) + -2iw-exp(ile.xlw/eN) } (1.18)
1l' N=l l<,;l=l peN eN
was first given in [1]. A corresponding representation for the transient Green's function
follows by Fourier inversion with respect to w. The result is not needed here, but it
was derived, somewhat differently, in [2].
In the case of an isotropic medium, Green's function can be obtained by several
elementary methods. It was derived from the representation (1.18) in [3], for instance.
Perhaps the most convenient representation is

(1.19)
 Dynamics of Composites 269

where <I>N (N = 1, 2) satisfy scalar wave equations

(1.20)

[Verification of this result is left to the reader.] The representation (1.19) is applicable
in any number of dimensions, and applies also to the time-reduced case if tH(t) is
replaced by -1/w 2 , and time derivatives are replaced by -iw. Equation (1.20) has
solution, in three space dimensions,

(1.21)

where r = lxl. The corresponding time-reduced form is

(f) _ - exp(iwr /eN)

N- 41fw 2r · (1.22)

1.4 Variational principles

Complications will be avoided by discussing only problems where displacements are
prescribed over the boundary an, for all t > 0. The simplest variational principle that
delivers the equations of motion is Hamilton's principle:

b {T dt { dx { ~ [piui - O"ijeij] + fiu;} = 0, (1.23)

.fo .In
where p and ( j are related to u and e as in (1.2) and (1.3). The displacement u is the
field that can vary, and allowed variations have to conform to the given displacement
boundary conditions. In addition, the displacement has to be regarded as prescribed
at times t = 0 and t = T, which is artificial.
Less conventional, but useful, principles can be obtained by considering the gov-
erning equations, Fourier- or Laplace-transformed with respect to time. Considering,
for definiteness, the Laplace transform, the governing equations (1.5) become

(1.24)

and these can be associated immediately with a variational principle of "minimum

energy" type:
(1.25)

when the transform variable s is real; more generally, the functional on the left side
is stationary for the actual (transformed) displacement field, amongst trial fields that
satisfy the specified boundary conditions and the correct initial conditions. The latter
merit further discussion: the simplest way to accommodate them is to require that
u = 0 for all t < 0, and to allow the body force fi (x, t) to contain suitable delta-
function terms with respect to timet, at t = 0. The Fourier transform version of (1.25)
has s 2 replaced by -w 2 , and so it is only a stationary principle. A stationary principle
270 J.R. Willis

in the time domain can be obtained from (1.25) by transforming back: products become
time convolutions and the principle is

(1.26)

The expression on the left side of (1.26) is unconventional, in the sense that it is a
generalized function of time t. Principles of this type were first established by Gurtin
[4]. Part of their interest is that they incorporate the correct initial conditions and do
not involve unknown conditions at time t = T.
Dual variational principles, which characterize the stress and momentum, can also
be derived. The principle dual to (1.23) is

o {faT dt (~in dx [P:i - Sijklaijakl] + kn dA uiaijnj) - [in dxpiui[=J = 0,

(1.27)
where SiJkl are elastic compliances, so that SiJpqCpqkl = !(oikojl + oilojk)· The displace-
ment u is prescribed on an, n
and it is regarded as known over all of when t = 0 and
t = T. The variation is taken over fields of stress and momentum density that satisfy
the equations of motion (1.1). The principle dual to (1.25) is

~in dx [P:i + sijkfOijakl] - kn dA Uiaijnj = Min, (1.28)

amongst fields rr, p for which

divrr +f = sp ( 1. 29)
(with s real). Inverting the Laplace transform gives

{ r [Pi * Pi- + SijkLOij

o ~ Jn dx - p * akl ] r
- JandA ui * ai1 nj }= 0, (1.30)

amongst fields of stress and momentum density that satisfy (1.1).

A more general discussion of these and other variational principles is given in [5].

2. RANDOM MEDIA
Although some of the techniques to be discussed later apply also to media whose
microgeometry is known (media with periodic microstructure, for example), most com-
posites vary from sample to sample and are best considered as random media. These
are families of media, whose physical properties vary not only with position x but de-
pend also upon a parameter (or set of parameters), denoted o:, where a is a member of
a sample space A, over which a probability measure pis defined. Thus, the probability
that a medium has parameter a lying in some subspace A 1 c A is p(A 1 ), and the
expectation value of some property ¢, conditional upon a E A 1 , is

(2.1)
Dynamics of Composites 271

In particular, the ensemble average of¢ is its unconditional expectation value,

(¢) = L ¢(a)p(da). (2.2)

Most of the media considered in the sequel are made of n constituents, or phases, so
that material of type r has elastic constant tensor Cr and density Pr· Such media can
be described by a set of characteristic functions Xr, where Xr(x, a) = 1 if xis in phase
r in realization a, and is zero otherwise. Since one phase is present at each point,
n
L Xr(x, a)= 1. (2.3)
r=l

The probability Pr(x) that point x lies in material of type r is

Pr(x) = (Xr)(x) = L Xr(x, a)p(da). (2.4)

The probability Prs(x, x') that points x and x' lie in material of types r and s respec-

L
tively is
Prs(x, x') = Xr(x, a)xs(x', a)p(da), (2.5)
and so on.
If a medium is statistically uniform, its multipoint probabilities Pn Prs, etc. are
insensitive to translations. Thus, Pr(x) becomes a constant, Pn equal to the volume
fraction of material of type r, and Prs becomes a function of (x- x') only.
Ensemble averages of physical properties follow readily in terms of the definitions
just given. For example, the elastic constant tensor C(x, a) at position xis given as
n
C(x, a)= L Crxr(x, a), (2.6)
r=l

and its ensemble average is

(C)(x) = (~ CrXr) (x) = ~Pr(x)Cr. (2.7)

The ensemble average (u)r(x, t) of a displacement field u(x, t, a) at (x, t), conditional
upon x being in material of type r, is defined so that

Pr(x)(u)r(x, t) = L u(x, t, a)xr(x, a)p(da). (2.8)

Other averages are defined similarly.

 272 J.R. Willis

3. PERTURBATION THEORY
This section considers media that are weakly heterogeneous, so that
C(x) = c(O) + bC(x), p(x) = p(O) + pb(x), (3.1)
where C(o) and p(o) are constants and the perturbations bC(x), bp(x) are small. The
initial development applies equally to deterministic or random media. Implicitly, there-
fore, bC(x) and bp(x) depend also on a and the deterministic case is included by al-
lowing the possibility that the probability measure p could be concentrated at a single
point o:0 in A.
The objective is to solve, in the sense of developing a perturbation expansion, the
equations of motion (1.1) for x E 0, together with the constitutive relations (1.2)
and (1.1), when C and pare given by (3.1). Boundary and initial conditions are also
specified. Purely for simplicity, suppose u is prescribed over 80 for all t > 0 and that
u = u = 0 when t = 0. The displacement u(x, t) thus satisfies

(Cijkluk,l),j + fi = pui, (3.2)

plus boundary and initial conditions.

It is convenient to define a "stress polarization" T so that

Tij = bCijklekl =bCijkluk,l = bCijkl (u~~] + buk,l), (3.3)

and a "momentum polarization" 7r so that

1ri = upui
f .
= up
f
ui + uui
( . (0) f . )
. (3.4)
Then,
(3.5)
and
Pi = P(o)ui + 1ri. (3.6)
The equations of motion (1.1) then imply

dO)
ijkl Uk ,lJ + (f· + T IJ,J· · - p' ·)- p(O)U
I I - I>
(3.7)
and it follows that

up(x', t) = u~0 l(x', t) + kdxG)~l(x, x', t) * [rij,j- Pi](x, t), (3.8)

where u(O) solves the corresponding problem for the unperturbed medium, G(o) is
the Green's tensor for that medium and * denotes the operation of time-convolution.
Integration by parts now gives, formally 1

u = u(o)- Sr- M7r (3.9)

'
1 The representation (3.9) applies also to boundary conditions more general than those of prescribed

displacements. See [1].

Dynamics of Composites 273

where S and M are operators:

(3.10)

The symmetry c;~l(x,x',t) = G~~l(x',x,t) of the Green's function allows the kernel of
the operator S to be given in the alternative form

··(, )-~[ac~~l(x',x,t) ac~~\x',x,t)]

sP'J x , x, t -
2
aXj + aX; . (3.11)

Equations for T and 1!" can now be developed by differentiating (3.9) to give rep-
resentations for the strain e and velocity u and then subsituting into the definitions
(3.3), (3.4). Thus, in symbolic notation,
T + oC(SxT + Mx1l") = oCe(Ol,
1l"+Op(Str+Mt1l") = ou(o)_ (3.12)

The operators introduced here have kernels

(3.13)

The bracketing of suffixes implies symmetrization. Thus, for instance, (Sx)pqij =

(S,)qpii.
It is noted for future reference that the symmetry of G(o) implies

(Sr)pqij(x', x, t) = (Sx)ijpq(x, x', t),

(M,)M;(x', x, t) = (S1 );pq(x, x', t),
(Mt)p;(x', x, t) = (llf1);p(x, x', t). (3.14)

Equations (3.12) in fact are exact and apply even if oC and op are large. When oC
and Jp arc small, however, the equations can be solved in the form of a perturbation
('xpansion:
r = JCe(o)- Jc[s",(JCe(o)) + Mx(Jpu( 0 l)) + ... ,
1l" = Jpu(o)- Jp[St(JCe(o)) + MtOpu(o)] + .... (3.15)
274 J.R. Willis

There is, however, a difficulty. The expansions (3.15) may contain "secular terms",
analogous to the expansion of eiwt in the form

eiwt = 1 + iwt - ~w 2 t2 + ....

The truncated series provides a good approximation only if lwtl « 1.
At least for random media of a fairly general class, this difficulty is avoided by a
modest reformulation of the problem.
4.1. The method of smoothing
'J;'he device now to be deployed was introduced by J.B. Keller [6]. First take the
ensemble average of (3.9) to give

(u) = u(o) - S(r) - M\rr), (3.16)

with corresponding expressions for (e) and (u). Now employ these, to eliminate e( 0 )
and U.(O) from (3.12):

T + bC[Sx(T- \r)) + Mx(rr- \rr))] = bC(e),

rr + bp[St(r- \r)) + M 1 (rr- (rr))] = bp(u). (3.17)

Even though (e) and (u) are unknown, equations (3.17) can still be solved, iteratively,
to give

T = bC(e)- bC[Sx(bC- (bC))(e) + Mx(bp- (bp))\u)) + ... ,

rr = bp(u)- bp[St(bC- (bC))\e) + Mt(bp- (bp))\u)) + (3.18)

Note also that (3.5) and (3.6) give, upon ensemble averaging,

(3.19)

Therefore, also ensemble averaging (3.18),

(a-) = (C(o) + (JC) )\e)

-( (JC- (bC)) Sr(JC- (JC)) )\e)
-( (JC- (bC)) Mr(bp- (bp)) )\u) + ... ,
(p) = (p(o) + (bp) )\u)- ( (bp- (Jp) )St(bC- (bC)) )\e)
-( (bp- (Jp) )Mt(bp- (Jp)) )\u) + ... . (3.20)

Equations (3.20) are effective constitutive relations. In conjunction with the ensem-
ble averaged equation of motion (1.1),

(3.21)
Dynamics of Composites 275

they yield directly a system of equations that governs the mean wave (u)(x, t).
It should be noted that the perturbation expansion demonstrates that, inevitably,
the mean stress and mean momentum density are both coupled linearly to mean strain
and mean velocity. It is perhaps helpful to display one of the terms in (3.20) in more
explicit notation:

((JC- (JC))Sx(JC- (JC)))(e)--+

j dx'(Sx)pqij(X, x', t) * ([!SCrspq- (JCrspq)](x)[JCjk!- (JCjkt)](x'))(ekt)(x', t).
(3.22)

For media with no long-range order, the two-point mean ([JC- (JC)](x)[JC-
(JC)](x')) decays, typically exponentially, with a small "correlation length" l. The
integral with respect to x' is thus effectively localized to a neighbourhood within a
distance of order l of the point x, and convergence of the perturbation series (3.20)
can be expected, irrespective of the spatial (or temporal) variation of the mean wave
(u)(x, t).
For ann-phase medium of the type introduced in Section 2,
II.

JC(x) = ~]C,.- Co)Xr(x, a)

1"=1

and n
!Sp(x) = L)Pr- Pohr (x, a).
r=l

The two·-point means in (3.20) are thus expressible in terms of the geometrical
two -point probabilities

Wr_,(x, x') = Pr..(x, x') - p,.(x)p.,(x'). (3.23)

For instance,

([r5C- (r5C)](x)[c5C- (c5C)](x')) = L L CrCsWrs(x, x'). (3.24)

4.2. Some implications

Insight into the behaviour of the system of equations (3.21), with (3.20), which
govern the mean wave (u) can be gained by specializing to the case of an infinite
statistically uniform medium and seeking plane wave solutions. In the case of an n-
phasc medium, the mw point probabilities Pr become constants and the two-point
probabilities Ill,._, become functions of (x- x'). Green's function and the associated
operators S,, de. likewise have kernels that depend only on (x - x') and t. The
equations then support plane waves of the form

(u) (x, t) = Uexp[i(kn· x- wt)], (3.25)

276 J.R. Willis

for which the constitutive relations (3.20) imply

(CJ;i) = ikC;'ikpknl- iwM;ikuk,

(p;) = -ikMktiUknt - iwp:kuk, (3.26)

both multiplied by the exponential factor shown in (3.25).

Here,
n n n
C;'jkl = L Pr (Cr )ijkl - L L (Cr )ijpq
r=l r=ls=l

(! (Sx)pqab(x, w)Wrs exp( -ikn.x)(x)dx) (Cs)abkl +

n n
Mijk = - L L(Cr)ijpq
r=l s=l

n n n
P7k = LPrPr5ik- L L Pr
r=l r=ls=l

(3.27)

The symmetries (3.14) were invoked here. The tensors C;*jkl' M;jk, Pik are the Fourier
transforms, evaluated at ( -kn, w), of the non-local operators in the effective constitu-
tive relations (3.20), specialized to a statistically uniform medium.
Substitution of (3.26) into the equation of motion (3.21), with f = 0, now implies

(3.28)

which is an eigenvalue problem, yielding a dispersion relation between k and w, and

associated eigenvectors U. Note that the operators Sx etc. that appear in (3.27) are
here related to the infinite-body Green's tensor, which was developed in Section 1.3.
Consider now the special case of low frequency (or long wavelength). The repre-
sentation (1.18) for the Green's tensor shows that, as w ---+ 0, s.T takes on its static
form (the term independent of w), while Mx and M 1 tend to zero. Correspondingly,
M;jk ---+ 0 and Pik ---+ I:~= I PrPr5ik· The constitutive equations (3.26) then reduce to
(3.29)

where ceff denotes the static tensor of effective elastic moduli and (p) is the mean
density. Correspondingly, equations (3.28) reduce to

(3.30)

Although the matrix that multiplies U in (3.28) is symmetric, its entries neverthe-
less are complex (except in the limit w---+ 0), and the associated dispersion relation is
Dynamics of Composites 277

complex: if k is chosen real, then w will be complex, while if w is real, then k will be
complex. The branches that are acceptable physically are those for which the distur-
bance decays, either as time increases or in the direction of propagation. Thus, if k is
real, w has negative imaginary part, while if w is real and positive, the real and imagi-
nary parts of k have the same sign. This attenuation of the mean wave does not imply
loss of energy, however. It is symptomatic of the progressive scatter of energy into
incoherent waves, whose mean is zero, and demonstrates the value of computing other
statistics in addition to the mean. The simplest is the variance, and this is directly
related to the mean energy, which is discussed next.
4.3. Energy relations
Energy is conserved in any elastic medium. Consider any subdomain V C 0. The
total potential and kinetic energy in the material contained within V is

(3.31)

Its rate of change is

d£(V)
- d- =
t
J(C.
\1
.
ijkte;jekt
. .. )dx =-
+ P'Ui'l/.i j (.eu +up. ')dx,
I'
(3.32)

having used the symmetry Ciikt = Cktii of the elastic constant tensor. The symmetry
of the stress tensor allows the replacement of e;jCJij by

·ti·t,]·a··=
l]
(1i ' a·) ·- ·ti·a·
l] ,] '
· · = (·ti.·a·)
l),J '
·- 1i·(p·-
l] ,] l l
f·)' '
from the equations of motion (1.1). Application of the divergence theorem to this term
in (3.32) thus gives
(3.33)

The first term on the right side of (3.33) represents the rate of working of the body
force f over V. The second represents the flux of energy into V, across av. The energy
flux vector F is defined to be the vector with )-component

(3.34)
If total energy density c is defined to be

c = t(eu +up) (3.35)

and the density of the rate of working of the body force is denoted

(3.36)

equation (3.33) implies that

-8c
8t + d'lVF = r, (3.37)
278 J.R. Willis

since it has to hold for arbitrary subdomain V.

Equation (3.37) implies the corresponding relation for ensemble averages:
8(c) .
at+ d1v(F) = (r). (3.38)

It is of interest to study the individual terms in (3.38). The easy one is

(r) = (u;) j;, (3.39)

because methodology for obtaining (u) has been discussed already. The energy flux is
now considered.
It follows from equations (3.9) and (3.18) that

u= u(OJ - s{ JC(e)- oC[Sx(oC- (JC) )(e)+ Mx(op- (op) )(u)] + ... }

- M{ op(u)- op[St(oc- (oC) )(e)+ Mt(op- (op) )(u)] + ... }. (3.40)

It is useful to eliminate u(o) in favour of (u) and, in the interest of avoiding even longer
expressions, to choose C(o) = (C), p(O) = (p), and to introduce notation so that, for
any quantity q,
q' = q- (q). (3.41)
The required expression is then

u = (u) + { S [(C'SxC')'- C'] + M(p'S 1C')'}(e)

+ { S(C'Mxp')' + M [(p'Mt/)'- p'J}(u). (3.42)

Similarly,

a-= (a-) + { (C( 0 lST,- I) [(C'SxC')'- C'] + C( 0 )Mx(p'S 1C')'}(e)

+ { (C( 0 lSx- I)(C'Mxp')' + C( 0 )Mx(p'Mtp')'}(u). (3.43)

An expression for energy flux - which can then be ensemble averaged - follows from
the product of (3.42) (differentiated with respect to time) with (3.43). This will not
be pursued in detail. Two points, however, may be noted. First, the mean energy flux
is not simply -(u)(a-). Second, equations (3.42) and (3.43) demonstrate that, at least
within the context of perturbation theory, the local flux of energy is related to values
of the mean field at neighbouring points, so long as the integrals converge. Evidently,
similar analysis could be performed for the energy density E.
It is remarked finally that, if the energy balance relation (3.33) is applied when
V = 0, and if f and the prescribed boundary values of u are sure (i.e. prescribed
independent of the realization a, so that u = (u) on 80), then ensemble averaging
yields
(3.44)
Dynamics of Composites 279

and hence, by use of the divergence theorem and the mean equation of motion (3.21),

(3.45)

This does not imply, however, that the integrand equals d(c) /dt pointwise.

4. DILUTE SUSPENSIONS
4.1. General formulation
A different kind of perturbation theory is discussed now: the composite consists of a
uniform matrix, with tensor of elastic moduli C 1 and density p1 , in which is embedded
a dilute random array of inclusions. For simplicity, the discussion will be restricted to
a single set of inclusions, all with tensor of moduli C 2 and density p2 . They are also
taken to have identical size, shape and orientation, though positioned randomly. Thus,
if 0 0 represents a domain containing the origin, "inclusion A" occupies the domain

and will be described as being "centred at xA". The centres XA are randomly dis-
tributed, with number density v. Thus, the volume fraction of inclusions is c2 = vlrlol,
where IDol denotes the volume of no. Equivalently, the probability density for finding
an inclusion centred at XA is PA = IJ. The two-point probability, P.4.B, for finding
distinct inclusions centred at XA and x 8 has the property

P.4.B = ()
if XA and Xs are such that nAn nB is non-empty. Also, assuming no long-range order,

The basic problem ean be formulated in a similar way to that introduced in Sec-
tion 3. Choose the matrix as comparison mediuin and define stress and momentum
polarizations

rA(x, t) = (C 2 - CI)e(x, t), 1rA(x, t) = (p2- pi)u(x, t) (4.1)

in inclusion A, and define TA, 1rA to be zero elsewhere. Polarization fields can then be
defined throughout the domain n occupied by the composite:

r(x, t) = L rA(x, t), 1r(x, t) = L 1rA(x, t), (4.2)

A A

so that
(4.3)
 280 J.R. Willis

The equations of motion then imply equation (3. 7), with C(o) = C 1 , p(o) = p 1 , and their
solution is (3.9). Equations for TA, 1l'A follow from calculating e and u in inclusion A:

(C2- Clt 1TA + L(SxT 8 + Mx11' 8 ) = e(O),

B
(P2 - P1t 111'A + :l:(Str 8 + Mt11' 8 ) = u(oJ, x E oA. (4.4)
B

These can be reduced by the method of smoothing, to give

(C2- C1t 1TA + SxTA + Mx1l'A

+Sx(L T 8 - (r)) +Mx(L 11' 8 - (11')) =(e),
B#A B#A
(P2- pl)- 11l'A + StTA + Mt1l'A
+St(L r 8 - (r)) +Mt(L 11' 8 - (11')) = (u), x E OA, (4.5)
R#A B#A
where
(4.6)

with a similar definition for (11'). Here, (rA)A represents the expectation value of
r(x, t), conditional upon the presence of an inclusion covering x (x- XA E 0 0 for some
A).
Mean constitutive relations are sought, and this requires the deduction of (rA)A,
(11'A)A from equations (4.5). These equations are therefore ensemble averaged, condi-
tional on x E OA. This gives

(C2- ct)- 1 (rA)A + Sx(rA)A + Mx(11'A)A

+ J dxsSx ((r 8 )sAPBIA- (r 8 )sPs)

+ j dxsMx ((11' 8 )sAPBIA- (11' 8 )sPs) =(e), x E OA, (4.7)

with a similar equation for (11'A)A· Here, (r 8 )sA represents the expectation value of
r(x, t), conditional upon the presence of inclusions centred at xs and XA. The associ-
ated geometric probability density, for finding an inclusion centred at x 8 conditional
upon the presence of an inclusion centred at XA, satisfies

(4.8)

Equation (4. 7) is the first of an infinite hierarchy. However, it is reasonable to

assume that, as lxs - xAI --too,

(4.9)
Dynamics of Composites 281

if the composite has no long-range order. Then, also,

(4.10)

The integral with respect to XB thus converges and will be of order c2 viOol, as
c2 --+ 0. In this case, in the limit of a dilute suspension, the integral with respect to
XB can be discarded, leaving the equations

(C2- C1t 1rA + SxTA + Mx1rA = (e),

(P2- P1t 11rA + StrA + Mt?rA = (u), x E nA. (4.11)

These equations describe the perturbation of a field whose values in the vicinity of OA
are (e), (u), by a single inclusion centred at xA. Once this "single scattering" problem
is solved, mean constitutive relations follow, as in Section 3.3. The dilute suspension
approximation disregards effects of multiple scattering (contained in the integrals with
respect to xB), except through requiring that any one inclusion scatters the mean wave
rather than the wave u(o) that would exist in the absence of all scatterers.
4.2. The single scatterer problem
It has just been observed that equations (4.1) define a problem of scattering by a
single inclusion. Of course such a problem can also be formulated directly, to provide
alternative strategies for its solution. For example, scattering of a plane wave by
an isotropic spherical inclusion in an isotropic matrix can be solved by employing
expansions in spherical harmonics [7]. Here, contact with statics is again made, by
specializing to the case of time-harmonic waves, at low frequency. The operators SXl
St, Mx, Mt are needed. The last, Mt, has kernel given as -w 2 times the Green's
function and, for the case of an infinite (in general anisotropic) matrix, this is given by
(1.18). The kernel of Mx is, in this case,

and the kernel of Sx is

- -1 3
(Sx)pqij = -2 L1
8Jr N=l lei=!
dS
~(pu:rut~j) { II
2
PlCN
W2 iW3 . }
0 (e.y)- 2o(e.y)- - 3 exp(IIe.ylw/cN) '
eN 2cN
(4.13)
where y = x- x'. These relations were obtained using (3.13). The second of relations
(3.14) gives St from Mx.
Now as w--+ 0, the "incident wave" specified by (e), (u) has long wavelength and
asymptotically is uniform over OA (which can be placed at the origin, so that XA = 0,
without loss of generality). To lowest order, therefore, equations (4.11) give

(C 2 _ CI)-ITA + (e),
S~tatic 7 A =

(P2- pi)-l7rA = (u), (4.14)

282 J.R. Willis

where gstatic
X
has kernel

( sstatic)
X
..
PQ'J
-1 3
8 2 """"'
= _
L
1.
(;
1f N=l l<,l=l
~ UNUN(
dS (p q) (' ;) 6"(C )
2
P1CN
., .y . (4.15)

Equation (4.14) defines the perturbation of a static uniform remote field (e) by an
inclusion occupying OA. It is well-known that it is solved by a uniform polarization
TA, in the case that OA is an ellipsoid; see, for example, [8]. In the special case of a
spherical inclusion, an easy calculation confirms that

(4.16)

where
P
PQ'J - 4 L
3
- _!__ """"' 1.
(;
N N
dS~(puq) u(i ~j)
2 . (4.17)
1f N=l 1., 1=1 P1 eN

The corresponding tensor of effective moduli is

(4.18)

The ·solution of the second of equations (4.14) is immediate: 1rA = (p 2 - p1 )(u), and
leads directly to the effective density

(4.19)

where c1 = 1- c2.
The low-frequency approximation can be refined by iteration. A complex number
~ and hence attenuation ~ first appears non-trivially through the term multiplying
-iw 3 in (4.13). To this order, the exponential can be replaced by 1, so that Sx contains
the constant term
C UNUNC
· 3 3
1W 2 I: r ds"(P q) 5(i '>j).
161r N=l J1e1=1 P1CN

The perturbation theory is easy to do. It is not developed here but it can be found in
[9].

5. VARIATIONAL PRINCIPLE
A method is now presented for obtaining systematic approximations, when the
elastic constant and density contrasts are large, and volume fractions are arbitrary. For
the sake of conciseness, the general n-phase medium will be considered, but variants
of the. method to be given can be tailored to deal explicitly with matrix-inclusion
composites. The basic problem is that presented by equation (4. 7): what to do to
close the hierarchy, when the volume fraction is large? The prescription offered here is
to develop a stochastic variational principle, and then to substitute into it trial fields
which are to be chosen optimally.
Dynamics of Composites 283

Variational principles of "classical" type were presented in Section 1.4. One ap-
proach would be to substitute into these strain and velocity fields obtained from the
representation (3.9), and then to select the polarizations T, 1r as well as possible,
within some suitable subspace of functions. It is, however, possible also to develop a
dynamical variant of the Hashin-Shtrikman variational structure; this is outlined now.
For a general (n-phase) composite with elastic constant tensor
n
C(x, o:) = L CrXr(x, o:)
r=l

and density
n
p(x, o:) = L PrXr(x, o:),
r=l

choose a comparison medium with elastic constant tensor C(o) and density p(o), and
define polarizations
(5.1)
It follows that u can be represented in the form (3.9), and that T, 1r satisfy the
equations
(C- c(0)( 1T + SxT + Mx1l" = e(o),

(p- p( 0 ) ( 1rr + StT + Mt1l" = u(O)' XED. (5.2)

The symmetries (3.14) of the operators that appear in (5.2) ensure that these equations
are equivalent to the variational statement 2
61-l( T, rr) = 0, (5.3)
where
1-l(T, rr) = Ldx{~[T * (C- c(o))- 1T + T * STT + T * Mx1l"
+rr * (p- p( 0) ( 1rr + 1l" * StT + 1l" * Mtrr]- T * e(O)- 1l" * u(O) },(5.4)
the symbol * denoting, as before, convolution with respect to time, t.
Furthermore, the entire family of equations (5.2), for a random medium, is gener-
ated by the stationary principle

t{tl(T, rr)) = 0, (5.5)

where now T 1r are regarded as arbitrary functions of x, t and o:, over w x [0, oo) x A.
This can be seen by performing the first variation explicitly:

L p(do:) L dx{ OT * [(c- c(0)( 1T + SxT + Mx1l"- e(O)l

+ 01!" * [(p- p(O)tl p + StT + Mt1l"- u.(O)]} = 0. (5.6)

2 The stationary principle (5.3) can also be derived from the "classical" principles discussed in

Section 1.4: this was done in [5].

 284 J.R. Willis

Equations (5.2) follow, by taking br, 61r to be arbitrary functions of x and t, concen-
trated in a neighbourhood of a chosen value o: in A.
The strategy now is to seek a stationary point of the functional in (5.5), over a
subspace of functions T, 1r, defined over 0 x [0, oo) x A. The chosen subspace is that
for which
n

r(x, t, o:) = l:r,.(x, t)xr(x, o:),

r=l
n
1r(x, t, o:) = L 7rr(x, t)xr(x, o:). (5.7)
r=l

The functions T r, 1r r are arbitrary except that they are independent of o:. The depen-
dence ofT, 1r upon o: is, however, tied to the realization of the composite through the
characteristic functions Xr.
With the polarizations T, 1r chosen as in (5.7), it follows that

H(r, 1r) = 1 ~ [t Tr * (Cr- C(O)t TrXr + t t T rXr * (Sx( TsXs) + Mx( 1rsXs))
0
dx{
r=l
1
r=ls=l
n n n
+ L 1rr * (Pr- P(O)t 11rrXr + L L 1rrXr * (St(TsXs) + Mt(1rsXs))]
r=l r=ls=l
n
- L(TrXr * e(O) + 1rrXr * u(O)) }. (5.8)
r=l
This expression has now to be ensemble averaged. To see the structure, consider two
representative terms:

(! dxTr * (Cr- C( = Jdxr,. * (Cr- C(

0 ))- 1 TrXr) r,.p,., 0 ))- 1

(! dxTrXr * Sx(TsXs)) = (! dxrr(x, t)xr(x, o:) j dx'(Sx)(x, x', t) * Ts(x', t)xs(x', rY))
= j dxr r(x, t) * j dx' (Sx) (x, x', t) * T s(x', t)p,. (X, x') (5.9) 8

(suffixes having been suppressed). Thus, reverting to less explicit notation,

(H(r, 7r)) = l
0
dx{ ~[t Tr * (Cr- c(O)t
r=l
1 TrPr + tt
r=ls=l
Tr * (SxTs + Mx1rs)Prs
n n n
+ L 1rr * (Pr- P(O)t 11rrPr + L L 1rr * (StTs + Mt1rs)Prs]
r=l r=ls=l
n
- L(T,. * e(O) + 1rr * u(O))Pr }. (5.10)
r=l
This expression is now rendered stationary with respect. to the functions r,.(x, t),
7rr(x, t), by requiring that
n
(Cr- c(O)t 1TrPr + L(SxTs + Mx1rs)Prs = e(O)Pr,
s=l
Dynamics of Composites 285

n
( Pr - P(o))-1 1r rPr + '"""'(S
~ (is + M t1r s)Pr s = U· (o) Pr · (5.11)
s=l

In contrast to the hierarchy developed in Section 4, equations (5.11) constitute a closed

system of equations for T r, 1r r. They can be investigated further by an analogue of the
method of smoothing.
Given that u has the representation (3.9) it follows that, within the subspace of
functions r, 1r defined by (5.7), the approximation (u) generated from (3.9) has the
representation
n
(u) = U(O)- L(Srs + M7rs)Ps, (5.12)
s=l

and hence that (5.10) can be expressed

n
(Cr- c(o)t 1TrPr + L(SxTs + Mx1rs)(Prs- PrPs) = (e)pr,
s=l
n
(Pr- P(O)t 1 1rrPr + L(StTs + Mt1rs)(Prs- PrPs) = (iJ.)Pr· (5.13)
s=l

This closed system of equations can (in principle) be solved for Tr, Trr, in terms of (e),
(u), to yield effective constitutive relations. In the case that the medium is infinite and
statistically uniform, this can be accomplished - in practice as well as in principle -
by Fourier transforms.
Obtaining detailed results requires working of a type already discussed briefly in re-
lation to perturbation theory : essentially, exactly the same working, processed through
this variational machinery, yields results applicable to large contrasts and finite volume
fractions. Furthermore, once approximate functions r, 1r have been established by solv-
ing equations (5.12), the representation (3.9) could be used to generate approximate
expressions for energy density and energy flux, analogous to those given in Section 3.3.
Ensemble averaging these would yield expressions in which the only correlations that
appear are the two-point functions Prs· To the best of the author's knowledge, this is
unexplored at present.
It is remarked that the quasistatic limit of the above formulation yields precisely
the Hashin-Shtrikman variational structure in the generalized form presented by Willis
[8] which allows for arbitrary correlations Wrs = Prs- PrPs· Bounds are not obtained in
the general time-dependent case. If, however, the generalized function of time 1i( T, 1r)
is Laplace transformed, and the transform variable s is real, the stationary principle is
a maximum principle if C(o)- Cr > 0 (in the sense of quadratic forms) and p(O)- Pr > 0
for all r, and a minimum principle if the opposite inequalities apply; see [5] for details.

6. A NONLINEAR PROBLEM
Waves in a composite whose response is nonlinear have hardly been studied at all, to
date. The methodology developed in the preceding section does not generalize directly,
 286 J.R. Willis

because of the use of convolutions, but a similar development based on Hamilton's

principle, generalizing (1.23), would be possible.
6.1. A matrix containing cracks
Just one example is now briefly outlined. A full account was given by Smyshlyaev
and Willis [10]. The medium consists of a uniform, linearly elastic matrix, with tensor
of elastic moduli C and density p. It contains cracks, defined by surfaces SA with
normals nA(x), x E SA, across which displacement components do not have to be
continuous. Outside the cracks, the displacement u(x, t) satisfies the equations of
motion (1.1). The relative displacement of the crack surfaces is denoted bA:

[u(x, t)] = bA across SA· (6.1)

Part of the problem is to find bA; it is analogous to the polarization introduced earlier 3 .
Application of the divergence theorem gives the representation

where S is defined by equation (3.11). It follows that

epq(x', t) = e~~(x', t) + L Epqkl(x', x, t) * bt(x, t)nf(x)dS, (6.3)

A

where
Epqkl = (Sx)pqijcijkl· (6.4)
The representation (6.3) incorporates delta-function terms arising from the jumps bA
across the crack surfaces. These can be written

efjacks(x, t) = L ~ [b1(x, t)nj(x) + bf(x, t)n1(x)] H(a 2 -Jx- xAI)8[nA.(x- XA)],

A
(6.5)
if the cracks are taken as circular, with radius a and centres at XA.
The stress is related only to the strain in the matrix. Thus, formally,

U = C(e _ ecracks),
if the representation (6.3) is employed. More explicitly,

apq =a~~)+ L { fipqkl(x',x, t) * bt(x, t)nf(x)dS- Cpqrse~:acks, (6.6)

A lsA
where
(6.7)
3 More precisely, -C(bA 0 nA) corresponds to a delta-function stress polarization concentrated at

the crack surface SA.

Dynamics of Composites 287

Now the boundary conditions on the crack surfaces must be introduced. It is

assumed here (this could be generalized) that the crack faces are smooth, so that they
transmit no shear stress while in contact, and that they are traction-free when open
(bA.nA > 0). Thus,

while
(6.9)
These conditions can also be expressed in the form of a variational inequality: Find
{bA, bA(n) ;::: 0} such that

L r dS' [c~(x', t)- b~(x', t)] nf(x')O";j(x', t) ~ 0,

A lsA
A= 1, 2, ... , for all cA, cA(n) ;::: 0. (6.10)

6.2 Effective constitutive relation

For the sake of simplicity, assume that the cracks are aligned, with normals parallel
to the x 3 -axis. The objective is to relate the ensemble averaged strain, (e), to the
ensemble averaged stress, (u). The total mean strain is obtained by averaging (6.3):

(6.11)

where (bA)A is the mean value of bA, conditional on a crack being centred at XA· The
strain can be separated into two parts:

(e) = (e)matrix + (e)cracks· (6.12)

The term (e)cracks is obtained by averaging (6.5):

(6.13)

where v = P.4 is the number density and C is defined by

C((bA)A) = ~ j dxA [(bA)A(x, t) ® nA + nA ® [(bA)A(x, t)] x
xH(a 2 - Ix- xAI 2 )6(x3- X3A)· (6.14)

The mean value of the stress is obtained by averaging (6.6):

In terms of the tensor of compliances, here denoted c-1,

(e) = c- 1 (u) + (e)cracks· (6.16)

288 J.R. Willis

If (e)cracks (or, equivalently, (bA)A) could be expressed in terms of (a), (6.6) would
provide an effective constitutive relation. The vector function (bA) A plays the role of
an "internal variable" for which an evolution equation is now developed, by applying
the boundary conditions (6.8) and (6.9).
Revert to the representation (6.6) for a, but eliminate a(o) in favour of (a) using
(6.15) - a further application of the method of smoothing. This gives (because nf =
6; 3 for all A)

a;3 = (a;3) + { dSK;k(x', x, t) * bt(x, t)

lsA
+{ L
Bf.A
j
Ss
dSK;k(x', x, t) * bf (x, t)

- j dxsPs / dxK;k(x',x, t) * (bf}sH(a 2 -lx- xAi 2 )6(x3- x3A)}

+(C(e)cracks)i3, (6.17)

where
K;k(x',x,t) = b.i3k3(x',x,t). (6.18)
The same type of problem appears as was noted in Section 5: the "multiple scattering"
terms r:ender the problem difficult. Now, however, specialize to a dilute array of cracks,
so that va 3 « 1. The difficult term in curly brackets is assumed to be of order (va 3 )
and is neglected, as is the term involving (e)cracks· Then, asymptotically,

ai3(x', t) = (a; 3)(x', t) + .lA dSK;k(x', x, t) * b~(x, t). (6.19)

This representation can now be combined directly with (6.8) and (6.9) to provide a
system of equations from which (bA) A may be determined. Smyshlyaev and Willis [10]
adopted a further simplifying approximation, to generate some explicit results.A direct
solution, using (6.19) with (6.8) and (6.9), has recently been developed by Capuani
and Willis [11].

7. CONCLUDING REMARKS
The formulation presented here was directed towards developing approximations
to effective constitutive relations, assuming a limited amount of statistical informa-
tion. Specifically, expressions were developed which incorporated one- and two-point
statistics. If more information were available, improvements could be made. For ex-
ample, substituting trial fields of the form employed in Section 5 into the "classical"
variational principles introduced in Section 2.4 would provide estimates which would
incorporate one-, two- and three-point statistics, analogous to the Beran bounds for
static problems [12]. Such a formulation was presented in [13] but it has still not been
explored in detail.
Conversely, problems may be confronted where not even two-point information is
given. For example, a matrix-inclusion composite may be presented for which the
 Dynamics of Composites 289

mechanical properties, and shapes and volume fractions of the inclusions are known
but the two-point functions, of which P8 1A introduced in Section 4 is an example, are
simply not available. In this case, it is necessary to develop some other approximation.
In elastostatics the self-consistent method is one natural candidate, and a dynamical
generalization has in fact been developed and applied for a matrix containing spherical
inclusions [14], aligned ellipsoids [15], randomly oriented ellipsoids [16] and cracks [17].
It reproduces the corresponding static estimates in the case of low frequency and at
higher frequencies provides estimates for dispersion and attenuation that depend upon
the shapes and sizes of the scatterers but accounts only approximately (through the
self-consistent embedding procedure) for effects of multiple scattering.

REFERENCES

1. Willis, J. R.: A polarization approach to the scattering of elastic waws. I.

Scattering by a single inclusion, J. Mech. Phys. Solids 28 (1980), 287-305.

2. Burridge, R.: The singularity on the plane lids of the wave surface of elastic
media with cubic symmetry, Quar-t. J. Mech. Appl. Math. XX (1967), 41-56.

3. Wang, C.- Y. and Achenbach, .J. D.: Three-dimensional time-harmonic elasto-

dynamic Green's functions for anisotropic solids, Pmc. R. Soc. A449 (1995).
441458.

4. Gurtin, M. E.: Variational principles in linear elastodynamics, Arch. Rat. Mech.

Anal. 16 (1964), 34-50.

5. Willis, .J. R.: Variational principles for dynamic problems for inhomogeneous
clastic media, Wave Motion 3 (1981), 1-11.

G. Keller, .J. B.: Stochastic equations and wave propagation in random media, Pmc.
Symp. Appl. Math. XVI (1964), 145- 170.

7. Ying, C. F. and Truell, R.: Scattering of a plane longitudinal wave by a spherical

obstacle in au isotropieally elastic solid, J. Appl. Phys. 27 (1956), 1086-1097.

8. Willis, .J. R.: Bounds and self-consistent estimates for the overall properties of
anisotropic composites, J. Mech. Phys. Solids 25 (1977), 185-202.

9. Willis, .J. R.: A polarization approach to the scattering of elastic waves. II.
Multiple scattering from inclusions, .J. Mcch. Phys. Solids 28 (1980), 307-327.

10. Smyshlyaev, V. P. and Willis, .J. R.: Effective relations for nonlinear dynamics
of cracked solids, .7. Mech. Phys. Solids 44 (1996), 49-75.

11. Capuani, D. and Willis, J. R.: Wave propagation in elastic media with cracks.
Part I. Transient nonlinear response of a single crack. Submitted for publication.
290 J.R. Willis

12. Beran, M. J.: Use of the variational approach to determine bounds for the effec-
tive permittivity of a random medium, Nuovo Cimento 38 (1965), 771~782.

13. Willis, J. R.: Variational principles for waves in random composites, in: Contin-
uum Models of Discrete Systems 4 (Eds. 0. Brulin and R.K.T. Hseih), North-
Holland, Amsterdam 1981, 471 ~478.

14. Sabina, F. J. and Willis, J. R.:A simple self-consistent analysis of wave propaga-
tion in particulate composites, Wave Motion 10 (1988), 127~142.

15. Sabina, F. J., Smyshlyaev, V. P. and Willis, J. R.: Self-consistent analysis of

waves in a matrix-inclusion composite. I. Aligned spheroidal inclusions. J. Mech.
Phys. Solids 41 (1993), 1573~1588.

16. Smyshlyaev, V. P., Sabina, F. J. and Willis, J. R.: Self-consistent analysis of

waves in a matrix-inclusion composite. II. Randomly oriented spheroidal inclu-
sions. J. Mech. Phys. Solids 41 (1993), 1589~1598.

17. Srnyshlyaev, V. P., Sabina, F. J. and Willis, J. R.: Self-consistent analysis of

waves in a matrix-inclusion composite. III. A matrix containing penny-shaped
cracks. J. Mech. Phys. Solids 41 (1993), 1809~ 1824.
 STRUCTURAL MORPHOLOGY
AND CONSTITUTIVE BEHAVIOUR
OF MICROHETEROGENEOUS MATERIALS

A. Zaoui
Ecole Polytechnique, Palaiseau, France

ABSTRACT

One of the main specific aspects of continuum micromechanics is related to the fact that one
has generally to deal with ill-defined bodies: only partial information on the statistical
distribution of the constituent phases of the considered random inhomogeneous materials is
available.
The first chapter briefly reports the main classical ways to use such an information in
the context of linear elasticity for the derivation of bounds and estimates; attention is then
focused on isotropic particulate composites and especially on Hashin's composite spheres
assemblage. New estimates are proposed for the overall shear modulus which include
Hashin's bounds as well as the three-phase model estimate : improved Hashin-Shtrikman-
type bounds are conjectured from this analysis.
The second chapter presents a proof of this guess as well as a generalization of the
method used for arbitrary "morphological representative patterns"; the classical Hashin-
Shtrikman variational procedure is adapted to such a morphological analysis. Applications to
the derivation of bounds and estimates for the overall elastic moduli are developed both for
isotropic and anisotropic distributions of the patterns.
In the third chapter, we lay the stress on behavioral aspects of this approach, with
special emphasis on the viscoelastic coupling. Whereas it is easy to study the influence of
morphological characteristics on the overall behaviour of linearly viscoelastic materials, this
is more difficult to do for nonlinear behaviours: Hill's treatment of the rate-independent
classical self-consistent model is highlighted and extended to rate-dependent behaviours but
this remains an approximation and new ways have still to be explored.
 292 A. Zaoui

1. INTRODUCTION

1.1 Continuum micromechanics and structural morphology

In what follows, we assume the usual conditions which define the framework of
continuum micromechanics to be satisfied: a random inhomogeneous material is considered
which obeys macrohomogeneity requirements, which implies that the pertinent scale lengths
of the body (referring to the size of the elementary "microscopic" inhomogeneities, of the
"mesoscopic" volume elements and of the "macroscopic" structure and mechanical loading
respectively) differ by one order of magnitude at least from each other. This separation of the
scales allows "representative volume elements" (R.V.E.) to be defined : their details differ
from one to the other but they lead to the same "homogeneous equivalent medium"
(H.E.M.).
Nevertheless, the detailed microstructure of each R.V.E. is so complicated that it
cannot be described completely: only some properties of the statistical distribution of the
constituent phases are known. These properties may be quantitative, such as the knowledge
of the volume fractions of the phases, of the two-point correlation functions of the elastic
moduli etc. They may also be qualitative, so as e.g. to specify the isotropy or the symmetries
of the anisotropy of the spatial phase distribution, the connectedness of one "matrix" phase,
the average shape of inclusions belonging to another phase ... We refer to all such
informations as the "morphological" description of the inhomogeneous material.
The question is now: how to deal with such informations in order to derive either
optimal bounds or pertinent estimates for the overall mechanical parameters? The main
classical ways to answer this question are first reported briefly in what follows (§2) in the
context of linear elasticity. The well-known solution of Eshelby's inclusion problem [ 1] has
been used to derive several classical estimates whose morphological meaning is discussed. A
general "systematic theory" for random elastic media [2] has been developed too which
allows to derive a set of bounds for the overall elastic moduli associated with correlation
functions of various orders.
These methods are shown to be unsatisfactory when materials exhibiting one
continuous phase, such as the matrix of reinforced composites, are considered. For the rest
of the chapter, we focus on this problem with special attention paid to Hashin's [3]
composite spheres (or cylinders) assemblage (C.S.A. or C.C.A.). We build up a continuous
set of estimates for the overall shear modulus of the isotropic C.S.A. which lie between the
older bounds derived by Hashin by using a Voigt-Reuss type approach adapted to the
C.S.A. whereas the three-phase model [4] is recovered as the self-consistent treatment of the
problem. We can guess that two of the obtained estimates correspond to new Hashin-
Shtrikman-type bounds: such improved bounds with respect to Hashin's ones should be
obtained by taking better advantage of the isotropic distribution of the composite spheres.
This guess is demonstrated and developed systematically in the next chapter (§3)
which combines the definition of "morphologically representative patterns" (M.R.P.) with
the use of Hashin-Shtrikman variational procedure [5]. It is shown that, for an isotropic
distribution of the patterns, explicit bounds can be derived from the solution of elementary
composite spheres problems: this solution may be approximated analytically as closely as
wanted. When. the centers of the patterns obey an "ellipsoidal distribution", the numerical
resolution of similar composite inclusions problems [6] lead to refined bounds whose
comparison with older ones is meaningful. Finally new estimates can be defined too,
especially in the context of the self-consistent procedure: M.R.P.-based generalized self-
consistent schemes [7] using finite element computations for the solution of the associated
inclusion elementary problems are suggested to yield an interesting alternative to the "unit
cell" or periodic homogenization techniques.
Structural Morphology and Constitutive Behaviour 293

1.2 Behavioral aspects

In the last chapter (§4), we lay the stress on behavioral aspects of the proposed
approach. This can be done easily in the case of linear (non ageing) viscoelasticity through
the Laplace transform technique. An illustration of the influence of morphology on the "long
range memory effect" is given by comparing the relaxation spectra of the shear modulus of
an isotropic two-phase material whose constituents obey a Maxwellian behaviour as
predicted either by the classical or by the generalized self-consistent scheme: whereas in both
cases the relaxation spectrum is a continuous one, which proves that the overall behaviour is
no more Maxwellian, this spectrum is split tip into two parts when the three-phase model is
used: this property is characteristic of the continuous morphology of the matrix phase [8].
The connection between morphology and overall behaviour is much harder to be
modelled in case of nonlinearity. For the rest of the chapter, attention is mainly focused on
the nonlinear classical self-consistent scheme which already confronts us with most of the
difficulties to be overcome. The advantages of Hill's treatment of elastoplasticity [9] with
respect to former ones are first recalled. The difficulty to extend it to rate-dependent plasticity
arises from the fact that, since stress and strain time derivatives of different orders are
simultaneously present in the constitutive equations, Green techniques cannot be used
directly. This difficulty can be solved [10] by combining the same kind of linearization
procedure along the loading path and the use of the Laplace transform technique: at each
incremental step, we are left with a linear viscoelastic behaviour with eigenstrains which can
be converted into a (symbollically) classical elastic problem with eigenstrains.
Nevertheless an important basic problem of nonlinear micromechanics has not been
solved by Hill's self-consistent treatment: it is associated to the fact that, at any stage, a
nonlinear phase which is not homogeneously deformed cannot be defined by a unique set of
instantaneous moduli (this is the case for the matrix for the self-consistent scheme). This
difficulty is still increased with the different versions of the generalized self-consistent
scheme which consider heterogeneously deformed phases. Some ways to appreciate this
difficulty in connection with the recent theories of nonlinear bounds are suggested in
conclusion.

2. CONTINUUM MICROMECHANICS AND MORPHOLOGICAL ANALYSIS

2.1 The need of a morphological analysis

It is well known that, except for quite special cases, e.g. when the assumption of a
periodic microstructure can be made, the material content of a Representative Volume
Element (R.V.E.) of a microinhomogeneous material cannot be described in a deterministic
way. Even if a statistical point of view is adopted, the statistical description of the spatial
distribution of the constituent phases cannot be performed completely. Consequently, we
cannot aim more than at bounding or estimating the overall mechanical characteristics of the
considered material. This is performed the more efficiently, the more completely the available
information on the phase distribution is used or the more pertinently this information is
obtained.
As a matter of fact, it is not easy to decide which kind of information, in addition to the
minimal one consisting in the determination of the volume fractions of the constituents and
their own mechanical characteristics, will lead to the closest bounds or to the most adequate
estimates. It will depend on the specific morphology of the considered material: in some
cases, the two-point correlation functions of some material parameters will be more efficient;
in other cases, it will be the symmetry of the overall anisotropy or the fact that one phase is
continuous and the other ones are not. ..
In any case, some morphological analysis has to be performed before any attempt to
derive bounds or estimates. In what follows, we shall briefly recall first the main current
294 A. Zaoui

ways to build up such bounds and estimates in order to illustrate the way they could take
account of such a morphological analysis. This will be done for linear elasticity in order to
concentrate on these morphological aspects: behavioral ones will be considered later. Let us
start with the most classical estimates of the overall elastic moduli which have been derived
from the classical solution of Eshelby's inclusion problem.

2.2 Inclusion-based estimates of the overall elastic moduli

The principle of any method for the derivation of estimates of the overall elastic moduli
cest or compliances sest of macrohomogeneous heterogeneous media is quite simple: since
the local values c, and s, are uniform per phase, we only have to estimate the average stress
a, or strain £, in each phase (r) for a given macroscopic stress E or strain E, i.e. to
estimate the average stress B~st or strain A~st concentration tensors, from which we derive:

r r
(2.1)
cest =J:,f,c,:A~st =<c:Aest >
r
or:

r r
(2.2)
sest =}; f,s,: B~st =< s: Best >
r

where fr is the volume fraction of phase ( r) and <· > stands for the volume average in the
considered R. V .E ..
Inclusion-based estimates may be defined by reference to Eshelby's solution [I] of the
problem of an ellipsoidal inhomogeneity H whose moduli c differ from those, C, of the
surrounding infinite matrix submitted to homogeneous strain EO at infinity. The strain tensor
E H in the inclusion is known to be uniform and given by

(2.3)

where SE is the Eshelby tensor, P = SE: c-l, &: =c - C and I is the unit fourth-order
tensor. We may now estimate A~st by considering that the mechanical state of each phase (r)
in the R.V.E. submitted to the global strain E is, in average, the same as the one of an
ellipsoidal homogeneity H, with the same moduli c, embedded in a fictitious homogeneous
medium with moduli C0 , submitted to some homogeneous strain EO at infinity. The shape
and orientation of H, can be specified from what we know of the ( r) phase domains
geometry whereas EO is determined by the average condition
< e >= L f,E, = E (2.4)
r

As for the moduli C 0 , they can be chosen at will in order to express at best the specific
morphology of the considered material; each choice of co
leads to a specific estimate C(t
through the corresponding po value.
Structural Morphology and Constitutive Behaviour 295

The associated set of equations reads:

- o s:_o -I Eo
Er =(I +Pr ."(J(;r) :
E =< E>=< (I+ Po: &o ri >:Eo
A~(/ = (I+ pro:&.:? rl: < (I+ Po: &o ri >-I (2.5)

C8SI =< c:(l + p0.-&.:0 rl >:<(I+ p0.-&.:0 rl > -]

= 2, Urcr:(l + Pr0:&? r 1 ):(2, fJI + Pt·&~ r 1 r 1

Of course, similar results could be obtained for the compliances, namely, with obvious
notation:
- 0 s: 0 -] 0
(J r = (I+ Qr: usr) :L
Q~ =Co -Co:Pro.-co

L =<a>=< (I +Qo:8so ri >: Lo

(2.6)
B~(/ =(I +Q~:8s?r 1 :<(1 +Q 0:8s 0 r 1 >-1
S8SI =< s:(I + Qo:8so rl >:<(I+ Qo:8so rl > -1

= LUrsr:(I +Q~:8s?r 1 J:(2,fs(I +Q~:8s?r 1 r 1

r s

Several classical estimates enter this framework for different C0 or S 0 values. Voigt's
(CV = <c>) and Reuss's (SR = <s>) ones correspond to vanishing values of so and co
respectively, which conforms with their extremal character and their underlying assumptions
(uniform strains, as transmitted to the inclusion by a rigid matrix, and uniform stresses, as
transmitted to the inclusion by an infinitely soft matrix, respectively).
Two other usual estimates, namely Mori-Tanaka's [11] and the self-consistent [12, 13]
ones, may be obtained too. The first one corresponds to the choice co= CMat, where CMar
are the elastic moduli of the matrix phase of a composite material. This is an approximate
way for expressing the continuity of the matrix of such reinforced composites: in this case,
the reinforcements as well as the matrix itself are supposed to be surrounded by the
continuous matrix, which corresponds obviously to an overestimation of the mechanical role
of this phase since interactions between particles or fibres as well as their strengthening
effect on the matrix response are neglected. From Walpole's interpretation [14] of Hashin-
Shtrikman [ 15] bounds in terms of matrix/inclusion situations, it is easy to show that for
spherical particles and isotropy, this estimate coincides with Hashin-Shtrikman's lower
bounds when, as usual, the matrix is softer than the reinforcements. Note that this derivation
only insures the reciprocity of the overall moduli and compliances ( SMT = ( CMT r 1) but not
their symmetry when particles with different aspect ratios and orientation are considered.
As for polycrystals, their phases (i.e. grains with identical lattice orientation and shape)
are rather disordered so that they play a similar morphological role: each grain with a given
lattice orientation and shape is surrounded by many other grains and, as a whole, the set of
similarly shaped and oriented grains is surrounded by almost all the other phases.
Consequently, a more adequate choice for the moduli co would be to take them as the
researched moduli cest themselves. This is the basic assumption of the "classical self-
296 A.Zaoui

consistent scheme" (C.S.C.S.) whose moduli ccscs and compliances scscs satisfy, from
(2.5) and (2.6), the equations
ccscs =< c:( I+ pCSCS: &:cscs r1 >: < (I+ pCSCS: &:cscs r1 >-1
or <&:cscs:(I+Pcscs:&:cscsr1 >=O
(2.7)
8 cscs =<s:(I +Qcscs .. &cscsr1 >:<(1 +Qcscs: 8scscsr1 >-1
or < &cscs:( I+ Qcscs: 8scscs r1 >= O
with pcscs, rx:cscs and Qcscs, &cscs formed from ccscs and scscs.
Many other estimates, of course, could be defined with other choices for or co so,
for
instance cv, SR, CMT etc., but without any clear morphological meaning. An illustration is
given on Fig. 1 for the estimation p_est of the overall shear modulus of a two-phase isotropic
incompressible material at some given volume fraction according to the foregoing method.
The ruling formula reads here:

est 2J1JJ12 + 3J1° < /1 >

/1 = (2.8)
2J11112 < jl-1 > +3J10
where Jlo can vary from zero to infinity. Note that Hashin-Shtrikman's bounds are obtained
=
for Jlo =111 and Jlo =112 and that the successive estimates /1(0-) ( =11 1), /1(2-) ( =11HS- JlMT),
... , J.Jf2n-) ... as well as /1(0+) (= J12), /1(2+) (= JlHS+), ... , J1(2n+) converge towards the self-
consistent estimate, as expressed by Kroner's recursive formula for this estimate [16], which
could be put in connection with his theory of "graded disorder".

11 est v

Fig. 1 An infinite set of estimates for the overall shear modulus of an isotropic
incompressible two-phase elastic material, for a given volume fraction
(the inclined straight line is the first bisector of the positive quadrant)
Structural Morphology and Constitutive Behaviour 297

2.3 A systematic bound theory

There is a systematic way to obtain bounds for the overall elastic moduli of random
media which get sharper and sharper by integrating more and more information on the spatial
distribution of the local moduli. Without entering into the details of their derivation, let us
emphasize some aspects of this "systematic theory" [2] in view of our further developments:
- the fundamental equation which is considered is the so-called "Lippman-Schwinger"
equation which derives from the application of the Green technique to the problem of a given
inhomogeneous body with moduli c( x): if G 0 ( x, x 1 ) is the Green tensor for an arbitrary
reference homogeneous medium with the same geometry, the moduli and fixed co
boundary, the strain field in the heterogeneous body satisfies the integral equation
e(x) + r 0 (x,x 1 )*& 0 (x 1 ):e(x 1 ) =e 0 (2.9)
where eO is the uniform strain field which would exist if the body was homogeneous and the
sign * indicates a convolution-type product. Here r 0 ( x, x 1
) is the "modified Green strain
operator", defined from G 0 (x,x 1 ) by:

rijkl - - I) =a a2a ,Goik (-X,X- I)(ij)(kl)

o (X,X (2.10)
'X j 'XJ

where the index (ij) stands for symmetrization with respect to i and}. Note that T 0 1
) is (x,x
x x
singular at = and behaves as a Dirac function at this point.
1

-if the distribution c(x) was known, (2.9) could be solved, at least formally, in order
to derive the effective moduli celf:
celf =<c:(i + To*Oco rl >: < (i + ro *Oco rl >-1
or (2.11)
with &elf =c - ceff
where the exponant (-1) refers to an operator inversion and i is the Dirac unit tensor. Similar
expressions are valid for the effective compliances self by changing c(x) into s(x) and
r 0 (x, x 1
) into its stress counterpart L1 °(x, x 1
) defined by:
L1o =Co -Co*ro.-Co (2.12)
-the same Green technique may be used to build up sets of kinematically (or statically)
admissible strain e x) (or stress a *(x)) fields from any polarization (symmetric) stress
1
(

field p(x) through the equations:

e (X) + T 0 (X,X
1 1
)*p(x 1 ) =e0
< el >= E (2.13)
a*= C0 : e I+ p
Such trial fields may be used in the potential and complementary energy theorems in order to
bound the effective moduli and compliances, according to the available information on the
distribution of the local quantities.
298 A. Zaoui

when such an information is given under the form of the first n-point correlation
functions of the local moduli or compliances, it can be used to yield n-th order bounds which
are the closer, n the larger.
- a special case which is worth-mentioning is concerned with materials obeying a n-
rank disorder condition [17]: if £0 and F 0 (i, x')
are the (singular) local part and the
(regularized) long-range part of the Green operator respectively, so that this operator may be
put in the form
r o(-X, X- ') -_ Eo Us:(-X, X- ') + Fo (-X, X- ') (2.14)
this condition reads:
<c':(F 0 *c'JP >=0 , V'p=lto(n-1) (2.15)
An interesting property of such a class of materials is that well-known bounds or estimates
are recovered as special cases: Voigt-Reuss's bounds correspond to n = 1, Hashin-
Shtrikman's ones to n = 2, whereas the self-consistent' estimate is obtained for n ---7=, the
so-called "condition of perfect disorder". In this case, all the contributions of the Green
operator reduce to their local part, so that (2.11) reduces to:
ccscs =< c:( I+ Ecscs: &cscs ri >: < (I+ Ecscs: &cscs rl >-I
(2.16)
or < &cscs:( I+ Ecscs: &cscs )-I>= 0
which coincides with (2.7a) for spheres and isotropy; (2.7b) could be obtained in the same
way by use of the local part of L1 °(x, x')
instead of 0 ( r x, x').
- one could think this systematic theory to yield the final solution of the problem of
taking morphology into account in homogenizing elastic random media. What makes the
situation less optimistic is that, in addition to the mathematical complexity of the practical
computation of bounds from given correlation functions of the elastic moduli, these
functions can hardly be extracted from an experimental investigation beyond n = 3 whereas
many primary morphological characteristics, such as the geometrical connectedness of one
phase or its inclusive nature could only be expressed through correlation functions of quite
high an order ...

2.4 Treatments of connectedness

This question has been for a long time a bone of contention between people interested
in composites and those concerned with polycrystals. For the former, any homogenization
model should obviously express, for instance, the fact that a porous medium can have some
mechanical strength up to a porosity of 100% and they found the self-consistent scheme to
be physically meaningless because of its prediction of vanishing elastic moduli for a porosity
larger than .5 or .6; on the contrary, metallurgists have considered for a long time Hashin-
Shtrikman's bounds as useless for their usual (polycrystalline) materials whereas they found
the self-consistent prediction of a rigid overall behaviour of a two-phase (duplex) metal as
soon as the volume fraction of the quasi-rigid second phase is larger than .4 or .5 as quite
sensible and meaningful ...
If one has in mind any care for the role of morphology in micromechanical modelling,
one may consider such a dispute as out-of-date: it only reflects the fact that composites and
polycrystals refer to basically different morphological types and cannot be satisfactorily
described by the same models, as already discussed. In this section, we focus attention on
the treatment of connectedness according to various models proposed in literature before
proposing a new method to do it in an improved manner.
Structural Morphology and Constitutive Behaviour 299

From what we have seen before, we can conclude that the question is still open.
Among the different inclusion-based estimates reported in section 2.2, the Mori-Tanaka's
one offers a rather crude and indirect way to express the geometrical continuity of the matrix
of a composite. The fact that it coincides with Hashin-Shtrikman's lower bound for an
isotropic particle-reinforced composite illustrates its overestimation of this morphological
property and we could hope that this property could be accounted for in a Jess extremal and
indirect way. At the opposite, the self-consistent estimate is clearly inadequate to this
purpose and is rather adapted to polycrystal-type disordered and dispersed morphologies.
As for the systematic theory of bounds, which includes Hashin-Shtrikman's ones as a
special case, it does not seem adequate anymore since it is centered around the idea of
morphological description through point correlation functions. As above-mentioned, such a
description would need correlation functions of a quasi-infinite order to have a chance to
express the connectedness of one particular phase, which is untractable in practice.
Two other ways have been proposed in literature in view of a better expression of the
matrix continuity of a composite. The first one refers to the so-called "differential self-
consistent scheme" (D.S.C.S.) [18, 19] and the second one to Hashin's "composite sphere
assemblage" (C.S.A.) [3]. We now briefly discuss the first method before paying sharper
attention to the second one.

2.4.1 The differential self-consistent scheme

Instead of considering a particle (or fiber)-reinforced composite in its actual state, with
the reinforcement volume fraction f, we take it as the final result of a differential process
which starts from the matrix phase alone. The second phase is added progressively, so as to
have its current volume fraction c raised from 0 to f; at each stage, the medium is re-
homogenized according to the self-consistent procedure; the latter may be defined within the
dilute approximation, since the considered second phase volume fraction is infinitesimal.
This differential process may be derived as follows: at a current stage, we start from a
homogeneous mixture with the overall moduli C( c) containing the matrix and the second
phases with the volume fractions ( 1-c) and c respectively. We add some infinitesimal volume
of the second phase with moduli c2. so that its final volume fraction be c+dc: it is easy to see
that, in this new mixture, the volume fraction of the newly added second phase is del( 1-c)
whereas the one of the previous homogeneous mixture is now ( 1 -c-dc)!( 1-c). The overall
moduli of this modified mixture, say C(c+dc), are calculated from (2.7). If (2.7a) is
specified for a two-phase material, with volume fractions j 1 and h , the overall moduli C
can be derived from the equation

(2.17)
where ~ and P2 are defined for the searched medium with moduli C. If now (2.17) is
specified for the considered situation, i.e. for/J = (1-c-dc)/(1-c), c 1 = C(c),f2 = dc/(1-c),
C = C(c+dc) = C(c) + dC and if only terms of first order inc are kept, we find the
following differential equation:

dC =~(c 2 -C):(I +P2:(c 2 -C))-1 (2.18)

1-c
=
where C C( c) and P2 may be defined from C instead of C( c+dc) . Integration of (2.18)
has to use the initial value C(O) = c 1. This is not easy to be performed in closed form in the
general case, but it can be achieved better in special cases.
For instance, for an isotropic composite with spherical particles and isotropic
constituents, we have the classical relations:
300 A. Zaoui

c =2J.LK +3k./ c 2 = 2J1 2K + 3k2J

1
JI)..kl =-
3 8--8kl
I)
(2.19)

3k
a=---
3k + 4Jl
Consequently, the differential equation (2.18) is split into two scalar equations, for the shear
and the bulk moduli J1 and k, namely:

dk-~ k2-k (2.20)

- 1-c 1 + a(k2 -k)
k
These are highly nonlinear equations and, due to the dependence of a and f3 on k and Jl,
they are coupled too. In order to illustrate simply the pos"sibilities of this scheme, let us
consider the case of incompressibility for both phases, and so for the composite too. In the
=
first equation (2.20), we can set f3 2/5, so that the equation for the shear modulus reads
now:
( 3J1 + 2J12) dJ.l = ~ (2.21)
5J1(J1 2 -J1) 1-c
After integration, we get
DSCS DSCS
_Jl__ £112-11 r5J2 = 0 _ n-5/2 (2.22)
JlJ 112- JlJ
This result can be compared, for JlJ < J12, with the (classical) self-consistent estimate Jlcscs
as well as with Hashin-Shtrikman's lower bound (which coincides here with Mori-Tanaka's
estimate JlMT). They are given by the following relations:
J1 cscs ( 5 f - 2) 11 2 3 - 5f 1
--= -+--+-
/1] 6 JlJ 6 6
(2.23)
JlMT =3(1-f)J1J+(2+3f)J12
11 1 ( 3 + 2f)J1 1 + 2(1- f)J1 2

It can be checked that, as expected, the three considered estimates are ordered according to:
JlMT </loses < Jlcscs. In order to pay special attention to the treatment of connectedness, it
is interesting to make the inclusions harder and harder (i.e. 112 ~ oo ). It is easy to get in this
case (Fig. 2):
DSCS
_Jl_ ~ (1- f)-5/2 JlMT 2+3f
--~----=-
JlJ 111 2(1- f)
(2.24)
11 cscs 2 2 11 cscs
--~-- .j~- --~00 , f ~l_
JlJ 2-5! 5 JlJ 5
 Structural Morphology and Constitutive Behaviour 301

(flJ == 1)

.5

0
.4 .5 1
Fig. 2 Comparison of the CSCS, DSCS and MT estimates of the inverse of the effective
shear modulus for an isotropic incompressible composite with rigid particles.

Whereas the three estimates behave similarly at low volume fractions f of the rigid
=
inclusions, namely as 11/11 1 1 + 5 f /2, they differ more and more whenfis increasing: the
two first ones remain bounded for any f and they tend to infinity only when/~ 1 whereas
the third one is unbounded as long asf ~ .4. These conclusions clearly express the fact that
the two first estimates take some geometrical continuity of the matrix into account whereas
the classical self-consistent scheme deals with some more disordered morphology so as to
allow a rigid skeleton to be formed beyond some percolation-like threshold.
As far as composite materials with a well-connected matrix are concerned, we now
have two models at disposal. Can we conclude that the differential self-consistent scheme,
which is more complex than Mori-Tanaka's one, is more relevant to express the matrix
connectedness? On the one hand, we can suspect that it does not present the exaggerated
character we have already underlined when discussing Mori-Tanaka's model. On the other
hand, it is not beyond any reproach: it can be developed in closed form for special cases only
so that its applications are more limited; moreover, though yielding the expected qualitative
properties for particle (or fiber)-reinforced composites, its morphological meaning is not so
clear.
As a matter of fact, the step by step re-homogenization procedure after every increase
of the particle (or fiber) content is rather artificial, despite its practical efficiency. It has been
argued that this procedure is a quite natural one when a wide range of particle sizes is to be
taken into account: the first steps would allow to deal with the smallest particles and the
302 A. Zaoui

following ones would be concerned with larger and larger ones. Nevertheless, this cannot be
understood literally since the elementary matrix/inclusion problem (with an infinite matrix)
has no absolute length scale and cannot distinguish by itself smaller and larger inclusions ...
That is why we cannot consider that the problem of matrix connectedness representation has
received a final solution through the differential self-consistent scheme and other ways to do
so have better to be explored.

2.4.2 The composite spheres assemblage

A much more direct and simple way to represent the matrix connectedness of a
composite material has been proposed by Hashin [3] as soon as 1962. Dealing first with an
isotropic particulate material, the "composite spheres assemblage" (C.S.A.), which has been
followed shortly after by the "composite cylinders assemblage" (C.C.A.) for fiber-reinforced
transversely isotropic materials, consists in an unbounded set of contiguous similar
composite spheres of all sizes, including vanishing ones, so as to be able to fill up the whole
space (see Fig. 3). Each composite sphere, with some radius b, has a spherical core, with
the radius a, made of the reinforcement material and a concentric spherical shell made of the
matrix material. The ratio (alb) equals Jll3 where f is the· particle volume fraction (for the
C.C.A., this ratio isjll2). Since the composite spheres touch each other and all the particles
are surrounded by some matrix material, the matrix is unambiguously connected, just as
wished.
The point is now to derive the overall moduli from those of the constituents and the
particle volume fraction/ Let us consider an isotropic C.S.A. and isotropic phases with the
shear and bulk moduli J.lJ, J.l2 and kJ, k2 , where the indices 1 and 2 refer to the matrix and
the particles respectively. Obviously, there is an infinity of ways to realize an isotropic
C.S.A. with the parameters J.lJ, J.L2, k1, k2 and/ Thus the overall moduli J.lc.s.A. and kC.S.A.
must be either bounded or estimated. The first approach to the problem, as proposed by
Hashin, aimed at bounding the overall moduli. More than fifteen years later, an estimate was
derived by Christensen and Lo [20] through the so-called "three-phase model".

or

Fig. 3 The composite spheres assemblage and Hashin's assumption for the derivation of
bounds for its overall moduli.
Structural Morphology and Constitutive Behaviour 303

2.4.2.1. Hashin's bounds for the C.S.A.

In order to build up admissible strain and stress fields to be used in the energy
theorems, Hash in considered homogeneous conditions at infinity of the form

(2.25)
and he imagined to transfer these conditions to the boundary of every composite sphere of
the assemblage and to solve the corresponding elastic problems. It is noticeable that, when
doing so, the required continuity conditions at the interface between any two contiguous
spheres are automatically satisfied so that, if the two elementary problems can be solved, we
are left with a kinematically admissible strain field and a statically admissible stress field
respectively, which are defined throughout the whole asssemblage.
From a practical point of view, isotropy allows to split each elementary problem into
two ones, corresponding to purely radial or shear loading conditions, so that the overall bulk
and shear moduli can be bounded separately. The solution of each problem derives from
basic Love's results [21]. The conclusion is twofold: the overall bulk modulus kC.S.A. is
unambiguously determined, since the upper and lower bounds coincide with the value

kCSA = 1+ 1 f (2.26)
kl + 3kl(l- f)
k21ki-1 3kl +4J1J
whereas the overall shear modulus Jlc.s.A. is not. For reasons which will appear later, the
corresponding Hashin's bounds JlH+ and jlH- may be conveniently written in the following
form:
H_
JlH+
!!:.____ = 1+ Jcp(O) - = 1+ Jcp(oo)
JlJ JlJ
-J

cp(x) = I
A(l- f)+ 1
m2fml-1
_
2/3 2
f(l- f )
BJ7f3+C+25(1-nl/
1
6(x -1)
with
A= 2(4- 5v1 )
15(1- VI)
(2.27)
B = 10(1- v 1 ) (7- 10v2)(7 + 5v1 )j1 1 - (7- 10v1 )(7 + 5v2)J1 2
21 4(7-10v2)J1 1 +(7+5v2)J1 2
10
C = 2 /7- 10v1 )(1- v 1 )

In these equations, VJ and v2 are the Poisson ratios of the considered phases.
One can see on Fig. 4 an illustration of these results for a particular case. Whereas it
can be checked that these bounds are closer to each other than Voigt-Reuss's ones, it may
also be noticed that they happen to be more distant than Hashin-Shtrikman's ones. This
remark may seem somewhat disappointing since Hashin-Shtrikman's bounds are known to
304 A. Zaoui

deal with unspecified (isotropic) morphologies, whereas the C.S.A. has especially been
defined in order to specify a matrix/inclusion-type morphology ... We only note this apparent
inconsistency at the moment and shall go back to this point later.
6,-----------------------.

I
o~--~--~----~--~--~
0 0.2 0.4 0.6 0,8 1
fig 4 Hashin's bounds for the overall shear modulus for the C.S.A., compared with Voigt-
Reuss's and Hashin-Shtrikman's bounds (Jt2IJ11 = 6, incompressibility)

2.4.2.2 The three-phase model

Aiming at "solving" the problem of the isotropic C.S.A. for the determination of the
overall shear modulus, Christensen and Lo actually proposed an estimate of this (not
uniquely defined) modulus. This estimate was derived from the condition that, under given
homogeneous (stress or strain) loading conditions at infinity, the strain energy density in
some homogeneous isotropic elastic infinite body remains unchanged if part of this body is
replaced by a composite sphere of the C.S.A.: the elastic moduli of such a homogeneous
body would yield an estimate of J.Lc.s.A. and kC.S.A .. In other words, this so-called "three-
phase model" (see Fig. 5) deals with a composite sphere embedded in a "third phase", the
homogeneous equivalent medium (H.E.M.) whose moduli J13PM and k3PM are derived from
the foregoing energy condition.
Here again, the elementary inclusion problem is solved from classical Love results as a
function of the unknown parameters J13PM and k3PM which are determined from the energy
condition, conveniently transformed into a quite tractable equation by use of Eshelby's
integral [22]. It happens that the resulting value of k3PM coincides with Hashin's finding of
eqn (2.26) whereas J13PM I J.LI is the positive solution of a second order equation. It can be
checked that this equation may be put in the following form, which refers to the notations of
formula (2.27):
x=l+fqJ(x) (2.28)
It is easy to verify that this estimate always lies between Hashin's lower and upper
bounds. Nevertheless, it is not quite clear whether the underlying model still respects the
Structural Morphology and Constitutive Behaviour 305

wished matrix connectedness and so reflects the corresponding morphology. As a matter of

fact, the treated configuration (Fig. 5) according to which a representative composite sphere
has been embedded in a fictitious homogeneous medium does not express any longer the
geometrical continuity of the matrix phase, at least as directly as initially done in Hashin's
assemblage. So, the morphological meaning of the three-phase model would improve on
clarification, which could be done by comparing this estimate of the C.S.A. elastic moduli to
some others to be derived. Such a derivation can be performed easily by applying to the
C.S.A. the method proposed in a simpler form in section 2.2: instead of dealing with simple
ellipsoidal inclusions, we would just have now to consider composite spheres.

·H·E·M·
0 0 • • •

Fig. 5 The three-phase model

2.4.2.3 A set of estimates for the C.S.A.

Let us consider the "phases" which constitute the C.S.A. in a new sense: instead of
still referring to the two mechanical ones, namely the matrix and the particles, this term is
applied now to composite spheres with the same diameter. Let (A) be such a "phase" defined
by the diameter dA, with the volume fraction!,r according to the proposed approach adapted
to the present situation, its average strain £A and stress (fA state when the homogeneous
strain E (or stress 1:) is applied to the C.S.A. at infinity is estimated to be the same as the
one of a composite sphere belonging to (A) embedded in an infinite elastic matrix with
moduli (Jlo, kO ), under some prescribed uniform strain £0 (or stress IfJ) at infinity. The
searched estimates C'// are derived from
<f>='L,f?..£A=E
A
(or < (f >= 'L,f?..O'A = };) (2.29)
A
< (f >= c~s/: < £ >
Obviously, fA and (fA do not depend on dA, so that fA = E (or (fA = J;). One
calculation only is needed in order to derive f (or 0') from £0 (or IfJ) for any composite
sphere with the volume fractionf This can be made easily from the basic Love's solutions.
306 A. Zaoui

e
If e, e1, e 2, 'E 0 and 8, 81> 82, 0 denote the average strain deviators and traces in the
composite inclusion, the spherical shell, the spherical core and at infinity respectively, one
can write the solution in the form
el= al'Eo, e2 = a2'Eo, e = [(1- J)al + fa2]'Eo
- 0 - 0 - 0
(2.30)
01 =b1 e. 02 =b2 e, O=f(l-J)b1 +fb2Je,

with scalar values aJ, a2, b1 and b2 depending on pP and kfl. We get formally

k(/ = 1 + f(k2fkl -1)b2fbl

kl 1- f + f b2fbl
(2.31)
Jlrt = 1 + f(J12IJ1J
-1)a2faJ.
J11 1-f+fa2fal
It is found easily that b2/b 1 does not depend on pO and kD, so that k{/ has a unique value
which coincides with Hashin's solution (2.26). For a2/a 1 , we find
a2 (1-j)qJ(Jlo/JlJ)
(2.32)
-;;; = J12IJ11 -1- f qJ(J1° /J11)
where cp(x) has been defined in (2.27). From (2.31), (2.32) and (2.27) we get finally:
est
Jlo =1+fqJ(J1°/p 1) (2.33)
JlJ

Fig. 6 A set of estimates for the shear modulus of an isotropic C.S.A. for a given volume
fraction (the inclined straight line is the first bisector of the positive quadrant)
Structural Morphology and Constitutive Behaviour 307

A graphic representation, similar to the one of Fig. 1, can be derived for a given f
I I
value: Fig: 6 shows such a plot of Jlost J1 1 as a function of J1 ° J1 1 which yields a
continuous set of estimates for the overall shear modulus of the isotropic C.S.A., lying
between Hashin's lower and upper bounds. Several comments can be made from this plot:
- Hashin's bounds J.LH+ and J.LH. for the shear modulus of the C.S.A. are two "bounds"
for the present set of estimates: they correspond to J1° = 0 and J1°-;oo respectively. Referring
to the single inclusion-based estimates of section 2.2 and to the analogy with classical Voigt-
Reuss's bounds, one could say that Hashin's bounds are the transcription of Voigt-Reuss's
ones to the specific morphology of the C.S.A.: as a matter of fact, Hashin's transfer of the
homogeneous stress or strain boundary conditions of the R.V.E. to the boundary of every
composite sphere is similar to Voigt-Reuss's procedure of prescribing uniform stress or
strain conditions throughout the R.V.E. (i.e. of transferring the macroscopic boundary
conditions to any point of the R.V.E.). In some sense to be made more precise, we could
say that Hashin's bounds are the composite sphere-based extension of Voigt-Reuss's ones
or that Voigt-Reuss's bounds are the"point" version ofHashin's ones.
- the three-phase model estimate is nothing but the "self-consistent" element of the
=
present set, corresponding to Jlo wst, as already apparent from (2.28): this is a justification
of the denomination of "generalized self-consistent scheme" (G.S.C.S.) which is given
frequently to the three-phase model. Now again, we could say that the three-phase model is
the composite sphere-based extension of the classical self-consistent scheme or that the
classical self-consistent scheme is the "point" version of the generalized one. Similarly to the
single inclusion-based estimates of section 2.2, we could define an iterative procedure and
recursive formulae converging towards the three-phase model estimate: this would allow this
model to appear as the proper treatment of a hypothetical "perfectly disordered" composite
spheres assemblage. Incidentally, it can be checked that Christensen-Lo's "energy condition"
coincides with a direct strain or stress condition (EO = E or IY = L), a coincidence which
can be proved to be valid in more general situations [23].
- going further along this speculative line of thinking, we can focus attention on two
= =
special estimates of the considered set, corresponding to Jlo J11 and Jlo 112 respectively.
Note that, when (k1- k2HJ11- J12) ~ 0, the first one is nothing but the classical Hashin-
Shtrikman's lower bound (or Mori-Tanaka's estimate) for the considered two-phase
isotropic material. Their derivation by embedding the composite sphere in the softer and the
stiffer phases looks like an extension of Walpole's interpretation of the classical Hashin-
Shtrikman's bounds adapted to the C.S.A., so that one could guess them to coincide with
new Hashin-Shtrikman-type bounds for the C.S.A.. If so, they would represent improved
bounds, with respect to Hashin's ones, for the shear modulus of the C.S.A., without any
new information except for a better use of the property of an isotropic spatial distribution of
the C.S.A.. Anticipating the following we could say that such new bounds are the composite
sphere-based extension of the classical Hashin-Shtrikman's ones or that Hashin-Shtrikman's
bounds are the"point" version of these new ones.
Before demonstrating this guess in the next chapter and generalizing the corresponding
"composite spheres approach" to arbitrary geometrical patterns, let us briefly mention that the
foregoing analysis of the C.S.A. can be generalized easily both to n-layered isotropic
spheres [24] and to transversely isotropic cylinders [25] assemblages, in view of
applications to coated particle or fiber-reinforced composites, as well as to approximate
treatments of nonlinear constitutive behaviours (see chapter 4).
308 A. Zaoui

3. MORPHOLOGICALLY REPRESENTATIVE PATTERN-BASED BOUNDING AND

ESTIMATING IN ELASTICITY

3.1 Introduction
We have already stressed the fact that most of the classical micromechanical
approaches use a description by points of the phase spatial distribution: this is obvious for
Kroner's systematic theory which uses intensively point-correlation functions of various
orders for the elastic moduli; it is also the case for single inclusion-based estimates (one
single inclusion in an infinite matrix has no internal microstructure so that it behaves like a
point) as well as for the classical Voigt and Reuss or Hashin and Shtrikman bounds. In
principle, point-correlation functions are a powerful tool for the description of complicated
microstructure but they reveal themselves rather limited in practice: their experimental
determination cannot be achieved beyond an order larger than 2 or 3 while the description of
as elementary morphological properties as the connectedness of a matrix phase looks
practically out of reach.
Hashin's composite spheres (or cylinders) assemblage and its by-products open a quite
novel method of morphological description by using finite inhomogeneous elements (instead
of points) as basic "bricks": first, the complete knowledge of the material content of such
bricks includes a much richer information that usual; in addition, something must be said on
the spatial distribution of the bricks, which still enriches the morphological description of the
material (one must keep in mind that both informations, namely the constitution of a
composite sphere and the spatial distribution of the spheres, were necessary to make the
C.S.A. express the matrix continuity).
Looking back upon Hashin's approach, we could find his definition of the "bricks"
rather sufficient for his purpose but his treatment of their spatial distribution somewhat
improvable (no explicit used was made of the implicitly assumed isotropy of this
distribution). In addition, one family of composite spheres was considered only and the need
to fill up the whole space with this spheres made necessary to use an artificial quasi-fractal
process, so that no local fluctuation of the particle volume fractions was allowed.
In what follows, we shall aim at saving the basic idea of dealing with finite composite
elements, without initial restriction to spheres or cylinders, and at improving the geometrical
description of their spatial distribution. By doing so, we want to combine a deterministic
description of small, but finite, well-chosen "composite patterns" and a statistical
representation of their distribution. Such a morphological representation will then be used to
adapt the classical Hashin-Shtrikman's variational procedure in order to yield bounds for the
overall moduli of a heterogeneous material [5]. Moreover, this "pattern approach" will be
applied to the definition of new estimates for these moduli, including a generalized
"morphologically pattern-based self-consistent scheme" [7] which uses numerical
computations for the resolution of the basic underlying composite inclusions problems.

3.2 A variational pattern-based approach to heterogeneous elasticity

3.2.1 Simplified approach

Let us first consider a multiphase E.V.R., whose extension can be considered as
infinite, submitted to homogeneous strain conditions E at infinity; the volume V of this
domain Q is supposed to have been entirely decomposed into several "morphological
phases" (A) with the volume fraction cA =VA jV, consisting of N A identical composite
domains DA centered at
I
Xk, translated from the same domain D A centered at the origin.
I
Structural Morphology and Constitutive Behaviour 309

This domain D A. will be considered as the "morphologically representative pattern" (say

M.R.P.) of the (A) phase (Fig. 7).

Fig. 7 A morphologically representative pattern

Note that any inhomogeneous material may be decribed in this manner, at least with as
many patterns, reducing to points, as mechanical phases. The classical "point approach" will
then be a special case of the present one and classical bounds have to be recovered as by-
products. In addition, such a pattern decomposition is not unique for a given material: some
descriptions may contain more informations than others and lead to more acurate bounds or
estimates. In the same way, for one given pattern description, some variational approaches
may be more powerful than others if they take more available information into account. Let
us start with the simplest one, namely the Voigt-Reuss-type approach, before adopting the
more efficient Hashin-Shtrikman variational procedure.

3.2.1.1 M.R.P.-based Voigt-Reuss bounds

Let us transfer the homogeneous strain boundary conditions at the boundary of each
pattern, namely
ii{(x)=E.x, 'tfiEdD;.,. 'riD;._ cQ (3.1)
I I

If one can compute the corresponding strain field ei. (x) inside each pattern D).., one has at
disposal, by simply collecting these fields throughout the whole body Q, a clearly
kinematically admissible strain field e '(x) which can be used as a Voigt-type trial field in the
potential energy theorem. This leads to a "M.R.P.-based Voigt bound" C~RP [6] such that

V > ceff • CV
CMRP- MRP·. <f. ' >=<c:e ' >= .L.J- J
~c). C).:£).'d(1), 'tiE. (3.2)
A. D). D;.
Similarly, a "M.R.P.-based Reuss bound" is obtained by applying homogeneous
stress conditions I* at the boundary of each pattern and by computing the associated stress
field a~ (x) inside the patterns such that
310 A. Zaoui

(3.3)

with ii;., (x) denoting the outward unit normal vector to dD;.,;. The optimal choice 1:* for a
given E leads to the bound sf:tRP such that

(S/:tRPr1 ~ ceff, f
sf:tRP: <a*>=< s: a*>= L ~ s;.,: a).dw, 'v'L (3.4)
;., D;.,DA.

It can be checked that C~RP and sftRP have the right required properties of symmetry and
positiveness. In the general case, the computation of the 21 components of these tensors
requires the (numerical) resolution of as many problems as different patterns under six
linearly independent homogeneous applied strain and stress tensors. The simplest case has
been solved in closed form by Hashin for the isotropic C.S.A. (see § 2.4.2.1.) and by
Hashin and Rosen [26] for the transversely isotropic C.C.A..

3.2.1.2 M.R.P.-based Hashin-Shtrikman bounds

The Hashin-Shtrikman variational procedure can be applied to this pattern approach
too. It makes use of trial fields derived from the application of an arbitrary polarization stress
field p(y) to a homogeneous reference medium with the elastic moduli co but with the
same geometry and boundary conditions as the considered inhomogeneous body. This
polarization field of symmetric second order tensors is defined throughout the body as
p(y) =LP;.,(YJcP;.,(y) (3.5)
).,

where t/J;.,(y) is the characteristic function of V;.,. As for P;.,(y), we choose it as translated
by X;...
I
from some P;.. ( x) field to be defined on D;.., so that we get

(3.6)

Fig. 8 Homologous points of a M.R.P.

Structural Morphology and Constitutive Behaviour 311

In other words, we prescribe identical values of the polarization stress field at "homologous
points", i.e. at points of the same morphological phase with the same relative position inside
the phase domains they belong to (Fig.8).
We start from the classical expression of the Hashin-Shtrikman functional HS 0 ( p)
which reads
2 HS 0 ( p) = E: C 0 : E + E: < p > + < e ': p > - < p: H 0 : p >
(3.7)
with H 0 = ( & 0 ;-1 = ( c - C 0 ;-1
where e' is the (kinematically admissible) strain field deriving fromp. From (3.6) we get:

(3.8)

where we have used for the strain field e' the following definition of "pattern-averaged"
fields f(x):

M - 1 NA - -
f:t (x)=-_Lf(x+Xk) (3.9)
N). i=l I

From (2.13), we can derive e' fromp and then e~M(x):

eA.,M(-) 1 ~ '(-x+ x-k

x =-L.e )
N). i=l I

1 NA
=e 0 - - . r f r 0 (x +Xk.x'):p(x')dm'
N;.. i=IV I

=e 0 - jrfM(x,x'):p(x')dm'
v
=e 0 - .r JrfM(x,x'):pp(x')dm'
)1 VII

Nil
(3.1 0)
=£ 0 -.L.L
. 1
jrfM(x,x'+Xp ):Pp(x')dm'
J
p J= Dll

=e 0 - _LNP jrf';/M(x,x'):Pp(x')dm'
p Dll

In this equation, (3.9) has been applied twice to the Green operator (once for each variable
x and x'). The auxiliary uniform strain eo has to be determined through the average
property:

<e'>=ENA. Je~M(x)dm=E.!:J._ Je~M(x)dm=E (3.11)

A. V DA A. DA. DA
so that we have:
 312 A. Zaoui

I; N;.. j ££ 0 - I; N11 f rf:M (x, x'): P11 (x ')dw' Jdw = E

). V DA J1 D11
(3.12)
NN
or £0 =E+ LL~ J Jrf:M(x,x'):Pjl(x')dw'dw
J1 ). V DA D11

Note that the difference between to and E stems from the infinite extension of the body and
from the corresponding definition of the Green operator in that case (an equivalent alternative
treatment would have been to deal with polarization fields deviations, i.e. fields with null
average).
After integration of (3.1 0) into (3.8), we get:

~ N;.. j P;..:(E+£;..,M -H;..:P;..)

2HS o(P;..)=E:C o:E+£..- 0 d OJ
). V DA

(3 .13)

We know that if co is such that Hf is positive or negative everywhere, then HS 0 ( p)

yields bounds for the overall moduli. Thus, the polarization fields P;.. have to be optimized in
order to get optimal bounds. The stationarity of the functional HSO is obtained by:

()HSO . s:'P
dp .u,
= 0' \;78P
).· (3.14)

From (3.13) and taking account of (3.12) for the dependence of to on P;.., we get, after some
elementary algebra:

8~50 :8p = LN;.. joP;,.(xJ:ft:~M(xJ-Hf(xJ:P;..(x)Jdw =0. (3.15)

ap ;..VD
A

This condition can be satisfied, due to the fact that the patterns are disjoint and the variables
P;.. are independent on each other, by
£~M (x)= Hf(x): P;..(x)
(3.16)
or P;..(x) = &f (x):E~M (x).
In other words, £ iM ( x) and P;.. ( x) are the solutions of the following integral
equations:
t:iM (x)= t: 0 - I; N11 j rf:M (x, x '): &Z (x '): t:~M (x')dw' (3 .17)
J1 Dll

with the condition (3.11 ), or

Structural Morphology and Constitutive Behaviour 313

0 - . - -
H)..(x).P)..(x)-E+ .!.!-- N}..NJl j TAJl
OMM - _,. f
(x,x ).P11 (x_, )dro ' dro ...
J1 ).. V D;. D11
(3.18)
... -L N11 j rf:M (i, i '): P11 (i ')dro'.
J1 Dl1

Referring to (2.9), we can guess from (3.17) that e;M(i) could be the solution of an
appropriate problem of interacting inhomogeneities D J1 in an infinite matrix, but such an
interpretation would need additional informations on the spatial distribution of the patterns to
be made more apparent.

3.2.1.3 The case of isotropy

We consider here an isotropic distribution of patterns in the following sense: let
'¥)..11 ( ii )drou be the number of couples of domains DA; and D11 j such that their centers XA;
and X11 j lie at a distance ii within drou. The assumption of an isotropic distribution of
patterns then implies that P}..J:L(ii) depends on u =lliill· In addition, this implies too that the
external shape of the patterns must be spherical: consider two patterns Aand J1 such that they
contain at least two subdomains Dk and D11 . in contact with each other (such a situation
I 1
happens necessarily since the domains have to map the whole space). As two subdomains
cannot lie at the same place, '¥)..11 ( u) must vanish over a finite neighbourhood of 0 (except
for 0 itself for A= J1 ). Let umin = JJiiminll be the minimal (non zero) distance between the
centers of (thus contiguous) subdomains A and J1 : due to the isotropic distribution of the
centers, this concerns couples of domains with an equiprobable orientation of iimin and it is
easy to prove that such a situation cannot occur unless the subdomains be spherical.
Thus, taking account also of the translation invariance of the Green operator in an
infinite medium, we can write
N11 j rf:M (i, i '): P11 (i')dro' = N11 j rf:M (i- i '): P11 (i')dro'
DJ.L D11

~ £.J
... = - 1 £.J ~ Jr 0 (X-A; - x-J1 . + - _,) PJ1 (-')d
X -X : X (J) ' (3.19)
N).. i=lj=l D
11
1

... =- 1-J drou jP)..J1(u)T0 (ii + i -i'): PJ:L(i')dro'.

N).. D
11

In this integral, i and i ' belong to the spheres D).., with the radius f).., and D J:L• with the
radius fi:L, respectively so that i - i' belongs to a sphere with a radius (f;. + f~. With use of
Fubini's theorem, the integration may be performed first on u over spherical concentric
shells with infinitesimal width du where P)..J:L ( u) is constant. These terms vanish for u ;?
f)..+ fJl due to the well-known property of the Green operator which is responsible for the
uniformity of the solution in the ellipsoidal inclusion of Eshelby's problem; thc;:y do so too
 314 A. Zaoui

for u ~ r;;., + r 11 when A ;r J.1 since 'P;..11 (u) vanishes in this case (the distance between the
centers of two spheres belonging to two different patterns cannot be smaller than the sum of
their radii). A non zero contribution comes only from A= J.1 and u = 0 (with 'Pu (0) = N).)
so that we are led to the result: N11 r~';!M (.X, .X')= O;.,J.l r 0 (x, .X'), \7' .XED;., . .X' E DJ.l.
We get finally
N 11 I r~';!M (.X, .X'): P11 (.X' )dm' =oAJ.l I r 0(.X -.X'): P11 (.X' )dm'
D11 D11
(3.20)
= I r 0 ( x-.X'): P;., (x')dm' ,
D:<
so that (3.10) reads now:

E~M(x)=£ 0 - l r 0 (x-x'):P;.,(x')dm', \tiED;., (3.21)

D:<
and the integral equation (3.17) becomes:
e~M(x)=t: 0 - IT 0 (x,x'):&~(x'):E~M(x')dm' (3.22)
DA

with the condition (3.11) to be used for the determination of eO. This proves that, due to the
assumption of an isotropic distributions of the patterns, the pattern-averaged strain trial fields
e ~M (.X) which can lead to optimal bounds for such an assumption can be derived as the
solution of elementary problems of spherical composite inhomogeneities in an infinite
homogeneous matrix (Fig. 9).

: c~ ::: : c? :::::: :c~ :::::::::::

Fig. 9 Interpretation of M.R.P.-based Hashin-Shtrikman bounds in terms

of composite inclusions problems

The elastic moduli co of the matrix have to be optimized too in order to make the
condition that Hf be positive or negative everywhere "just" satisfied: roughly speaking, this
Structural Morphology and Constitutive Behaviour 315

means that these moduli must be those of the softest or the stiffest mechanical phases
belonging to the considered composite. This allows finally to derive optimal expressions
HSopt of the Hashin-Shtrikman functional and then Hashin-Shtrikman-type bounds ct~P
th rou gh .. HSopt -- -1 E ··CHS
MRP··E·
2
Note that in the foregoing analysis nothing has been specified about the mechanical
anisotropy of the constituents and the material content of the patterns: the analysis is still
valid when the spherical patterns exhibit no internal spherical symmetry or when elasticity is
locally arbitrarily anisotropic so that, despite the isotropic distribution of the centers of the
patterns, the overall behaviour may be anisotropic (see examples of such situations on
Fig . 10). Note also that, in such cases, the Green operator need not be known in closed
form : as a matter of fact, we only have to solve composite inclusions I matrix elastic
problems, which can be performed by other techniques (including numerical, e.g. F.E.M.,
ones) than Green's ones.

Fig. 10 Examples of isotropic distribution of patterns

3.2. 1.4 Going back to the isotropic C.S .A.

A straightforward application of what precedes concerns the isotropic C.S .A. [27]. We
can consider the "morphological phases (A) as consisting of the composite spheres SA, with
given external and internal radii bA,and a). respectively, satisfying/= (a)/'b).,)3) . The optimal
value of the Hashin-Shtrikman functional is, from (3.16) and (3 .8):

HS 0 (e~M)=!_E:C 0 : E+-
1 I.,NA. E: j8c 0 (i):e~M(x)dw (3.23)
2 2V A. sA
with e~M (x) given by (3.22) and (3 .11) and & 0 (x) positive or negative everywhere. If
£i2J and £in are respectively the average strain tensors in the core and the shell of SA. when
 316 A. Zaoui

embedded in an infinite matrix with moduli C0 , submitted to tfJ at infinity, we can write
(3.23) as:

HSo ( E~M) = E: Co: E + E: J:. c). [f( c2 -Co): £i2! + (1- f)( c 1 -Co): £il)]. (3.24)
2 2 ).
But we know that £i J and £i1! do not depend on the size of 5)., so that we have:
2

HS 0(£1, £ 2) = .!._ E: C 0: E +!_E.· [f( c 2 - C 0 ): £2 + ( 1- f)( c 1 - C 0 ): £ 1]

2 2
with E = j£2 + (1- f)£ 1, £i2! = £2 , "Ef! =£1, \fA, (3.25)

and then HS 0(£1,£2) = .!._ E: [fc 2: £2 + (1- f)c]." Ej}.

2

6,----- 6,--------------

5 5

a;
::t

0~~~~-b~--~-~
0 0.2 0.4 0.6 0.8

Fig II Comparison between the bounds derived, for the overall shear modulus of an
isotropic C.S.A., according to Hashin's treatment (continuous lines) or to the present one
(dotted lines): phase 2 included in phase I, 11 2!11 1 = 6, (a) incompressibility and (b)
v1 =. 45 and v2 = 0. (from [27])
Structural Morphology and Constitutive Behaviour 317

With the notations of (2.30) and with 'E and e the deviator and trace of E, we get easily:
HSo =fJ12a2 +(1- f)JlJa/ 'E: 'E + ~~--....:;.._........__,_
fk2b2 +(1- f)klbl e2 (3.26)
fa 2 + (1- f)a 1 jb2 + (1- f)b 1 2
This leads directly to prove that the two associated bounds for kC.S.A. coincide, with the
same value as Hashin's one, whereas the bounds for Jlc.s.A., say J1 HS+ and J1 Hs-, are
given, with the notations of§ 2.4.2.3, by:
HS+ HS-
_Jl_ = 1 + f cp( f121 JlJ ), !:!:__ = 1 + f cp( 1). (3.27)
JlJ JlJ
This result is a demonstratation of the guess proposed in § 2.4.2.3. It allows to bound
the overall shear modulus of the C.S.A. more efficiently than initially proposed by Hashin,
as illustrated on Fig. II. We can add that, provided that the assumption of an isotropic
distribution of the pattern centers remains valid, several families of composite spheres,
representing either local fluctuations of particle volume fractions or different kinds of
particles, could be considered as well; especially, one family of spheres can be constituted
with the matrix material only. Similarly, we can deal with several patterns of multilayered
spherical inclusions. In addition, the transcription of this approach to composite (possibly
coated) cylinders with transverse isotropy is quite straightforward.
We can also guess that the transcription of Kroner's "graded disorder" theory to the
present pattern approach could allow us to prove that the three-phase model could be the
adequate treatment of a "perfectly disordered" C.S.A. in a way similar to the one which
allowed Kroner to prove, within the point approach, that the classical self-consistent scheme
is the adequate treatment of a "perfectly disordered" medium. Nevertheless, the question of a
proper definition - or even of the statistical feasibility - of such a "perfectly disordered"
C.S.A. remains an open one ...

3.2.1.5 Case of isotropic pattern distribution and cubic overall symmetry:

Fig. 12 Scheme of a composite with isotropically distributed cubic aligned inclusions

 318 A. Zaoui

The above treatment of the case of an isotropic distribution of the pattern centers can
also be applied to situations of macroscopic anisotropy, depending on the material content of
the patterns. Such a situation can be achieved for instance by a composite with cubic aligned
particles distributed isotropically (Fig. 12). Both phases are considered as elastically
isotropic so that the overall behaviour has a cubic symmetry: let eff, f.l~ff and f.l~ff the
overall bulk and shear moduli respectively, with the definition

2f.l~ff = <CJxx -(jyy >, 2f.l';l = <CJxy > (3.28)

<£XX - Eyy > < Exy >
where x, y and z are the directions of the cubic symmetry.
Of course the M.R.P.-based bounds for these moduli cannot be derived in closed form
anymore but numerical computations aiming at the solution of the basic matrix/inclusion
underlying problems can be performed as well. We report here the results of such
computations [6] for both shear moduli of (3.28) performed by a 3D F.E.M. technique using
the mesh represented on Fig. 13 (351 20-noded quadratic elements and 1528 nodes). The
M.R.P.-based Voigt-Reuss and Hashin-Shtrikman bounds have beeh computed by using
moduli of the reference medium either far higher and lower than those of the constitutive
phases for the former or equal to those of each of these phases for the latter. Meaningful
comparisons may be made especially with the general bounds derived by Milton and Kohn
[28] for cubic symmetry in order to appreciate the advantage of specifying a
matrix/inclusion-type morphology.

Fig. 13 Typical 3D mesh used for the solution of the auxiliary inclusion problem
corresponding to Fig 12 (from [6])
Structural Morphology and Constitutive Behaviour 319

In the present case (k2 > k1 and /12 > J..LJ), Milton and Kahn's bounds read:
fj( 2 + 3 )~ 5 6f2(k2 +2J12)
112- J..L~ff 112- J..Lt 112- /1] 112( 3k2 + 4J12)
(3.29)
f2( 2 + 3 )~ 5 61J(kj +2J1J)
111-J..LJ J..Lt-J..LJ 112-J..LJ J12(3kl+4J..LJ)
It can be checked on Fig. 14 that the specification of a matrix/inclusion morphology with
isotropically distributed aligned cubic inclusions leads to a drastic tightening of the bounds
for the shear moduli. Note that, in accordance with mechanical intuition, the M.R.P.-based
bounds for this particle-reinforced composite lie close to the lower Milton-Kahn's one;
nevertheless, the lower bounds do not coincide as they do in case of isotropy and "well-
ordered" materials (i.e. with k2 > k1 and /12 > J..LJ).

3.5
- Milton & Kobo

···--- MRP-based Hashln/Shtrikman

3.0 ---· MRP-based Voigt/Reuss

2.5

2.0

1.5

1.5 2.0 2.5 3.0 3.5 4.0 4.5

Fig. 14 Comparison between Milton-Kahn's and M.R.P.-based Voigt-Reuss and Hashin-

Shtrikman's bounds for the shear moduli of a composite with isotropically distributed
aligned cubic inclusions k1 =3, 11 1 = 1, k2 =20, 11 2 =10 (arbitrary units) (2) in ( 1)
(from [6])

In what follows, we shall now explore some routes of generalization of the proposed
approach, especially in view of dealing with anisotropic distributions as well as of building
up new tools for an improved estimation of the effective behaviour.

3.2.2 Generalization
The foregoing analysis has been extended according to two different directions: the
first one is concerned with the possibility to deal with more general definitions of patterns,
which would not need anymore to be identical within a given family nor to fill up the whole
space; the second one aims at taking better into account the overall anisotropy, especially
320 A. Zaoui

anisotropic distributions of patterns. These two kinds of extensions are briefly reported in
what follows within the framework of linear elasticity.

3.2.2.1 An extended definition of patterns

Instead of restricting ourselves to morphological phases both composed of identical
patterns and forced to map the whole space, we can consider more general situations without
sensible increase of complexity [5]. First we allow any morphological phase (A) to be
constituted of patterns which may be slightly different, both geometrically and mechanically.
Let SA. be the geometrical intersection of such patterns DA.k after translation of their center
X A.k to the origin: the foregoing definition of the polarization stress fields will be unchanged
inside one part of the patterns, namely the translated of SA., but modified outside. This new
definition will read:

pA (XAk + X) =PA (X) if XAk + X E (A), X E sA '

(3.30)
PA.(XA.k +x)=PA.o if XA.k +xE(A), x~SA.

where PA.o is a constant symmetrical second order tensor. In addition, the phases (A) may
n
not fill up the whole space: we can consider this part of where no pattern information is
available as an additional phase, say (v) , with Sv =0.
Referring to the expression of the Hashin-Shtrikman functional, it appears that we
need an additional definition of "pattern-average" fields f(x), namely:

Jt =-v1- f f(x)dw (3.31)

A.o VAO

where VA. 0 is that part of (A) whose points obey the conditions: X A.k + x E (A), x ~ SA..
This allows us to write a new version of (3.8), namely:

2HS 0(p) = E:C 0: E + L, NA. j(E + e~M -PA.: HfM ): PA. dw ...
A. V sA
(3.32)
~ VA.o (E 1m . om)·
··· + ~- +eA. -pA. · 8 A. ·PA.
A. V o o

Instead of a deterministic Hf (x) tensor with known identical values at homologous points
x, we have now to deal with pattern-averages HfM (x) and Hfm, which have to be
derivable from some statistical knowledge about the material content of the patterns DA.k .
Eqns (3.10) and (3.11) now read:

eA.1M(-) 0 ~NJ.l jrOMM(-

X =e -~ AJ.l
-1) P. (-l)d (J) I -~
X,X: J.l X
~v. rOMm -
J.lo AJ.l (x):pJ.lo
J.l ~ J.l
(3.33)
~N jrOmM(-1) P. (-l)d I
eA.lm =e 0
- ~ J.l AJ.l X : J.l X (J) -
~v. rOmm
~ J.lo AJ.l : p J.lo
J.l SJJ J.l
Structural Morphology and Constitutive Behaviour 321

and

~ (1-a;,j'M(-)d
<E , >=£.-CA.-- E;, X m+a;,E;,'m) =E (3 .34)
A. S;, SA
with a;, =V;, 0 jV;,. The stationarity condition on the Hashin-Shtrikman functional has the
new form:

(3.35)

which implies

E~M (x)= H~M ( x): P;, (x) or P;, ( X-) =yOM-1(_)

A.
'M(-)
X :E;, X
(3.36)
c-'m _ yOm ·p or _ 8 om-1. E mt
cA. - A. . A.o P;,o - A. . A.
Integration of (3.36) into (3.33) shows these equations to still rule (composite)
inclusions I matrix problems through pattern-averaged Green operators which have to be
derived from the available informations on the pattern distribution. Finally, optimal bounding
is obtained by choosing co
so as to make H~M ( x) and H~m (just) positive or negative. In
case of an isotropic distribution, the intersections S;, must be spherical and the pattern-
averaged Green operators can be determined in closed form: in addition to the result already
. d f or rOMM(-
denve AJ1 x.x_,) , name I y N rOMM(-
AJ1 x.x_,) =uA.
~:: - _, ). 'v'x E SA. • .X' E S •
TO (x.x
11 11 11
we find with similar arguments:

c,o: : : : : : :co::::::::::: :
... ..... ..... ' :co :: :::::: :::

. . . . . . . . . .. . .

Fig. 15 Interpretation of generalized M.R.P.-based Hashin-Shtrikman bounds in terms

of composite and homogeneous inclusions problems
322 A. Zaoui

rfffm(x) =rf;M (x') =o, 'v'x e s,1.. x' e s11

(3.37)
romm =_1_ 0 Eo
AJ.l V. AJ.l
J.lo

where EO is the local part of the Green operator T 0 ( x, x').

Now again, the problem reduces to several problems of composite spheres with
moduli C0 + HfM- 1(x) and homogeneous ones with moduli C 0 + Hfm- 1 in an infinite
homogeneous matrix, so that Fig. 9 is now changed into Fig. 15. Note that it is still more
apparent on this case that when the patterns reduce to points and the morphological phases to
the mechanical ones, i.e. when S,t =Sr =0, 'v'A, =r, the classical Hashin-Shtrikman
bounds (and their interpretation in terms of homogeneous spherical inclusions by Walpole)
=
are recovered soon with P,t 0 =P,t 0 Pr uniform per phase. Note also that the case of
composites with one phase constituting a continuous matrix and with inclusive patterns
which are identical within each morphological phase has been treated more completely by
Bornert [29].

3.2.2.2 Ellipsoidal distributions of patterns

Another way of generalization of what has been presented as a "simplified approach"
in 3.2.1 is the following: still considering morphological phases consisting in identical
patterns, we can deal with more general distributions of the centers than the isotropic one in
order to explore more completely the sources of anisotropic behaviour of heterogeneous
media. A simple way to do so refers to Willis's original idea [30] of "ellipsoidal phase
distribution", here adapted to the pattern approach. To do so, we still make use of the
distribution function 'P,tiii) of section 3.2.1.3 such that 'P,tiii)dwu is the number of
couples of domains D,1..I and D".
/""" 1
such that their centers X,1. I
and X".1
/"""
lie at the distance u
within dwu. The assumption of an ellipsoidal distribution of the pattern centers implies that
'P,1.11 ( u) can be written as
(3.38)

where lf/,t11 (x) is any positive real function of x and B any positive definite symmetric
second order (appropriately normalized) matrix.
Similarly to what has been shown for an isotropic distribution, it can be proved now
[6] that (3.38) implies that the outer surface of any pattern obeys the equation liB. xll =r
where r determines the pattern extension: this means that all the patterns must be ellipsoids
with the same principal directions (those of the matrix tB.B, with tB transposed from B)
and aspect ratios: we can imagine the ellipsoidal distribution as resulting from an isotropic
one which has been stretched along the principal directions of tB.B (see Fig. 16 as an
illustration of such an operation). Nevertheless the internal geometry of the patterns is still
arbitrary and may exhibit quite different anisotropic properties. An interesting possibility of
this approach lies in the possibility to investigate the competition between the two
corresponding sources of the overall anisotropy, associated with the pattern distribution and
their material content, respectively. Examples of such a competition are given in the
following.
Structural Morphology and Constitutive Behaviour 323

(a) (b)

Fig. 16 Ellipsoidal distribution of patterns: (a) external shape of the patterns;

(b) possible internal content (from [6])

The integration of such an information into Hashin-Shtrikman's variational procedure

is quite straightforward and makes basically use of the same fundamental property of the
Green operator as before. The pattern-averaged trial strain field (3.10) can be computed by
integration on ii over ellipsoidal shells with infinitesimal width where this property can be
used, namely
N11 jr~:M(x -x'):P11 (x')dw' ...
Dl'

... =- 1 ~ ~ Jr 0 (X- A; - x-J1 . + X- - X_,); P.J1 (-')d

£.. £.. X (J)
' (3.39)
NA i=lj=l D
I'
1

... =- 1- j dwr J1/fA.p ( r )T0 (ii + x- x '): P11 ( x ')dw'

NA. DJ1
By permuting the order of integration due to Fubini's theorem, the contributions vanish as
long as r c rA. + r11 , with rA. and r11 denoting the extension of the ellipsoids DA. and D11
(whose equations read liB. .$'
xll rA. and liB. xll.$'
r11 respectively): this result stems from the
evoked property of the Green integral for concentric similar ellipsoids. The contributions
324 A. Zaoui

vanish too when r s 'A. + rJ.l and A :;e 11 since DA. and DJ.l cannot overlap; the only non zero
contribution arises from r =0 and A =11 with If/AJ.l (0) = NA. 8AJ.l. Consequently we get:

NJ.I f rf':M (x, x'): PJ.I (x ')dw' = 8A.J.I f r 0 (x- x'): PJ.I (x ')dw'
D~ D~
(3.40)
= j r 0 (x-x'):PA.(x')dw', 'v'xEDA.
D;.
which is strictly the same result as (3.20), derived for an isotropic distribution, except that
DA. and DJ.l are now similar ellipsoids instead of spheres.
The end of the demonstration and the final results and interpretation are unchanged,
except for the replacement of spheres by ellipsoids. No closed form solution can generally be
expected, even for simple internal microstructures of the patterns. But, here again, direct
F.E.M. numerical calculations can be performed in order to solve the elementary underlying
composite ellipsoidal inclusions problems which give access to the derivation of bounds for
the overall moduli. An illustration of the method is reported now for the case of transversely
isotropic composites [6].

3.2.2.3 Case of transversely isotropic materials:

We consider identically shaped and oriented (along the symmetry axis z) hard isotropic
ellipsoidal inclusions distributed in a continuous weak isotropic matrix according to an
ellipsoidal symmetry. This case allows us to compare our M.R.P.-based Hashin-Shtrikman
bounds for the transverse 111[
and longitudinal 11~! overall moduli to explicit analytical
Hashin-Shtrikman-type bounds which have been derived recently by Ponte Castaneda and
Willis [31] for particulate composites. They assumed too an ellipsoidal distribution of the
centers of the inclusions (not of the patterns !) and used piecewise constant polarization
fields in the matrix and the inclusions; as for taking the matrix connectedness into account,
they did not make any explicit assumption and just used average relations in order to make
the matrix apparently disappear from the equations. We guessed that in fact such a procedure
implied some assumption on the matrix spatial distribution, which would be unlikely to
express the matrix connectedness and we expected our M.R.P.-based Hashin-Shtrikman
upper bounds to be lower than theirs.
The Hashin-Shtrikman-type lower bound derived by Ponte Castaneda and Willis is

c:~- =C1 + hf(Cz -c1r1 +~~~- hPJi~; r 1 (3.41)

with use of the following uniform tensors:
~~ij= Jr(iJ(x-x')dw', 'v'xE!Jinc

(3.42)
Pfifj = jr(i)(x -x')dw',
ndis

where (i) refers to the used comparison medium (moduli C 1 for the matrix and C 2 for the
inclusions) and .Qinc and .Qdis are the characteristic ellipsoids for the inclusions and for the
distribution respectively. The upper bound could be derived in the same way:

(3.43)
Structural Morphology and Constitutive Behaviour 325

Comparison was made for spheroidal inclusions and distribution with identical aspect
ratios w = b!a where b denotes the length along z: in this case, Dine and Qdis coincide and so
do lj~j and PJfj. Typical results are reported in Fig. 17 and Fig. 18 for 111[ and 111[ as a
function of F(w) (with F(w) = w for oblate inclusions and F(w) = 2 - 1/w for prolate ones in
view of a more symmetrical presentation). In both cases the lower bounds coincide whereas
the upper bounds lie far apart from each other, with definite advantage gained by the
M.R.P.-based Hashin-Shtrikman bound. This result illustrates the pertinence of generalized
Hashin's assemblages to the representation of the matrix connectedness.

4.0 ,----~--~----~----~-----,

--- Ponte-Castaneda and Willis (U and L)

><----->< MRP-based Hashin/Shtrikman (U and L)

.2
.g0 3.0
\
E \

Classical
/ Hashin!Shtrikman
I

(Oblate) (Spherical) (Prolate)

1.0 L _ __ _ _ _:-""=------:"-:--------:-'::--~---'

u u w u ~
Aspect ratio (F( w))

Fig. 17 Comparison between Ponte Castaneda and Willis's and M.R.P.-based Hashin-
Shtrikman's bounds for the transverse shear modulus of a composite with aligned ellipsoidal
inclusions k1 = 3, 11 1 = 1, k2 = 20, 11 2 = 10 (arbitrary units), f 2 =. 3, ( 2) in ( 1) (from [6])

We note that for w = 1 (isotropic distribution of spheres) Ponte Castafieda-Willis's

bounds coincide with the classical Hashin-Shtrikman's ones, which indicates that, at least in
this case, the matrix connectedness has not been fully exploited. As for the noticeable
(numerical) coincidence of the lower bounds, it can easily be justified analytically by solving
the involved problem of an ellipsoid Dine with aspect ratio Wine made of material (2)
embedded in the ellipsoidQdis with aspect ratio Wdis made of material (1) which is embedded
itself in an infinite matrix with moduli C 1 : this is nothing but the classical Eshelby problem
whose well-known solution leads to (3.41) after some simple calculations [6].
We end this chapter by reporting briefly Bornert's recent developments [7, 29]
concerning the continuation of the foregoing approach towards the definition of M.R.P.-
based self-consistent estimates.
 326 A. Zaoui

4.0 .-----..,..--~---.------.-------,

- - - Ponte-Castaneda and Willis (U and L)

MRP-based Hashin/Shtrikman (U and L)
)<----->(

"'
;:l
:;
-g 3.0
a....
Ol _!.1--------
"'
.c
-;"' /
/

Classical
c /

:a:I /
/
/

Hashin!Shtrikman
I
·@, 2.0 I
c I
I
0
.....l I
I

(Oblate) (Spherical) (Prolate)

1.11 L___ _ _ _ ~ _ _ _ ___[__ _ _ _-:"::_ _ _ _ ~

O.lJ 0.5 1.0 1.5 2.0

Aspect ratio (F( w))

Fig. 18 Comparison between Ponte Castaneda and Willis's and M.R.P.-based Hashin-
Shtrikman's bounds for the longitudinal shear modulus of the composite of Fig. 17 ([6])

3.3 A M.R.P.-based self-consistent scheme

The principle of such a generalization of the so-called "generalized self-consistent
scheme" is quite straightforward: we already noticed that the three-phase model was an
extension of the classical self-consistent scheme obtained by replacing the homogeneous
ellipsoids representative of the mechanical phases by the composite spheres (or cylinders) of
the C.S.A. (or of the C.C.A.). We also justified the possibility of dealing simultaneously
with different kinds of composite spheres (or cylinders) within an extension of the three-
phase model. The next step consists in dealing with several families of morphological
representative patterns (instead of composite spheres or cylinders) so as to derive an estimate
of the overall moduli from the numerical resolution of as many composite inclusion I matrix
problems as representative patterns: the matrix moduli have to be such that they relate the
average stress and strain tensors over the different representative patterns embedded in this
matrix (Fig. 19).
Another way of understanding such a procedure is the following [7]: when looking for
M.R.P.-based Hashin-Shtrikman bounds, we had first to make an optimal choice of the
polarization field p, say p ~pt, for given values of the moduli co of the reference medium,
which led to the relation (see 3.7, 3.13 and 3.16):
0 0 1 0
cpopr(Popt> E)= 2E: < Popt >
(3.44)
Structural Morphology and Constitutive Behaviour 327

We had then to chose the moduli co of the reference medium so as to be sure, at minimal
cost, of the sign of <P~pt( pgpt> E) in view of bounding the effective moduli. The self-
consistent point of view aims at minimizing I<P 0 ( p, E )I with co = ceff so as to have an

estimate of weff(E) through HS~%r(P~%r.E), with <p~%r >=0. Such a relation is

s:_eff : copt
equivalent to < uc; , >= 0 or <a opt
, >-- ceff.. < c ,pt >.
0

H.E.M. H.E.M.

• ••

Fig. 19 Principle of the M.R.P.-based self-consistent scheme

From a practical point of view, several difficulties have still to be overcome. We refer
to Bornert's work [7, 29] for more details. Suffice it to say here that consistency of the
foregoing definition with an energetical one forces the external shape of the patterns to be
ellipsoidal. Thus the numerical procedure developed for the derivation of M.R.P.-based
Voigt-Reuss or Hashin-Shtrikman bounds in case of en ellipsoidal distribution of patterns
can still be used within an iterative procedure which can start from any initial reference
medium C(O) to compute a first estimate C(2), then to compute Cf4) from C(2) and so on until
some convergence criterion lie( 2
n) - C( 2n- 2Jjj s c be satisfied. Different pattern geometries
have been considered up to now [7] (including non similar concentric ellipsoids) in order to
investigate different sources of anisotropy in particulate composites.
It can be appreciated that such an approach yields a useful alternative to the "unit cell"
or periodic homogenization techniques. At the price of a numerical treatment somehow
heavier due the iterative procedure, the assumption of a periodical distribution of patterns
which may happen to be embarrassing is given up; in addition several families of patterns
may be considered simultaneously, which may offer definite advantages in many cases,
especially when local strain or stress investigations are developed, such as in view of
analyzing plastic localization or damage initiation and growth processes. Nevertheless, this
approach would need to be extended to nonlinear behaviour, which will be seen not to be
quite straightforward in what follows ...
328 A. Zaoui

4. BEHAVIORAL ASPECTS

4.1 Introduction

Up to now linear elasticity only has been considered. A more complete understanding
of the influence of structural morphology on the overall mechanical properties of
inhomogeneous materials obviously needs to explore nonlinear constitutive behaviours too.
The basic difficulty one has to overcome is twofold:
- on the one hand, most of the micromechanical methods reported hereabove in case of
linear elasticity make use intensively of the Green techniques which are basically attached to
linearity. The question is: to what extent some linearization procedures could allow us to
preserve anything of these techniques when dealing with nonlinearity?
- on the other hand, nonlinearity implies inhomogeneity of the mechanical properties,
even in a homogeneous material, as soon as it is stressed or strained in an inhomogeneous
manner. This means that we have to deal not only, as we did up to now, with interphase
inhomogeneity but also with "intraphase" inhomogeneity of the material parameters. In other
words, the morphological description and integration of the phase spatial distribution must
be somehow complemented by those of the material parameters within each phase, the latter
beeing basically changing along any given load path (think of the inextricable extension of
the correlation functions of the elastic moduli to the case of nonlinearity ! ... ).
That is why the question of nonlinear continuum micromechanics is a matter of
intensive current research which will not be tried to be reported here. In order to give an
outline of the difficulties and possible issues but still concentrate on the influence of
morphology on the overall behaviour, restriction will be made to the extension of two simple
but basic models in this field, namely the C.S.C.S. and the G.S.C.S.: the first one is well-
suited to a "polycrystal-type morphology" and the second one to a "composite-type
morphology". Comparison of their predicted responses to the same loading path can allow to
draw significant conclusions on the influence of the phase connectedness.
In view of going from the simpler to the harder, we start with the easiest way to depart
from elasticity but still preserving the convenience of linearity, namely with the case of linear
nonageing viscoelasticity.

4.2 The case of linear nonageing viscoelasticity

In this case, obviously the mechanical characteristics will not be space-dependent
inside the same phase, due to the linear behaviour, but the rate-dependence is responsible for
many new phenomena and properties which are characteristic of the coupling between
elasticity and viscosity. One of the most important manifestations of this coupling in
heterogeneous media is the so-called "long-range memory effect" which results from the
complex delayed mechanical interactions between the constituent phases. This effect finds
expression, for instance, in the fact that the overall behaviour of a mixture of Maxwellian
constituents does not obey a Maxwell constitutive behaviour anymore.
Let the local constitutive equations read
£ =a:a+b:a (4.1)
with a and b fourth order constant tensors. Then the overall behaviour will obey a different
law which can be written in the form [32]
t
E = A eft: I + Beff: i + j J (t - s ): i (s )ds (4.2)
0
Structural Morphology and Constitutive Behaviour 329

where the tensorial kernel J is directly related to this "long range memory effect". Its
mathematical expression is, of course, model-dependent and then morphology-dependent.
So, it would be interesting to focus attention on the comparison between its properties as
predicted, other things being equal, by the C.S.C.S. and the G.S.C.S. respectively in order
to appreciate the influence of the microstructural morphology on this memory effect.
A convenient way to do so is to derive the relaxation spectrum of a two-phase isotropic
material according to both these models. Whereas this spectrum reduces to a single line for
each constituent phase, it is expected to be more complex for the two-phase material in a way
which could differ according to the morphology. For the sake of simplicity, we assume
incompressibility and isotropy for each phase ( i) so that (4.1) reduces to
e =ais +his i = 1,2 (4.3)
where e and s are the strain and stress deviators and ai and bi are material constants which
define the single relaxation time Ti = bifai of each phase. The question is: which are the
relaxation times of the two-phase material according to the C.S.C.S. and the G.S.C.S. ?
The answer [8] can be derived easily through the use of the Laplace transform
technique, since it is well-known, due to the "correspondence principle" [33], that this
transformation allows us to convert linear (nonageing) viscoelasticity into (symbolical)
elasticity: suffice it, in order to get the equations of the viscoelatic problem, to replace the
elastic moduli and variables of the elastic self-consistent equations by the Laplace-transform
of their viscoelastic counterpart [34]. Transformation of (4.3) yields:
L 1
(4.4)
fli (p)= 2bJp+1/Ti)

where p is the complex variable, jL( p) the Laplace transform off( t) and f.li( t) the shear
relaxation function of phase (i).
In the elastic case, both considered models yield the overall shear modulus peff as the
positive root of a quadratic equation which may be written in the following common form:

L( f3,f )X 2 +2M( {3,f )X+ N( f3,f) = 0 (4.5)

where X= W.ff lf.lJ, f3 = f.12lf.11 andf = f2. For the C.S.C.S., we have, from (2.8):
2-5! 5f-3 N =- 2{3
L= 1, M=-6-[3+-6-, (4.6)
3
and for the G.S.C.S., with phase (2) included in phase ( 1 ), from (2.28):

L({J,f) = 4[ 3({3 -l)x 3 -173 H17 1x 7 -217 2 ) -12677if3 -l)x 3( J-x 2 ) 2

10 7 129 3
M([3,f)=377 1([3-l)x +477 177 3x --17 2 ([3-l)x +...
4
3 3 2 2
... +417 2 173 + 12677if3 -l)x (1-x )
(4.7)
N({J,f) = -[~([3 -l)x 3 + 173H17Ix 7 + 1; 7] 2) -126TJif3 -l)x 3( l-x 2 )2,

x = !I/3, 17I = 19([3 -1), 17z = 19{3 + 16, 173 = }_(2[3 + 3).
2 2
 330 A. Zaoui

The solution of the viscoelastic problem is obtained by replacing in (4.5), (4.6) and
(4.7) the quantities X by XL(p) =WffL(p)IJ1i{p) and {3 by f3L(p) defined by:

f3L(p)=J1IcP) =kP+ 1fTI with k=bJ!b 2. (4.8)

flf(p) p+1/T2
The solution Wff L(p) can be written as the sum of two terms, If ( p) and If ( p ), namely:

If{ p) =- M({3L( p),c)

2bl ( p + 1/T1)L({3L( p ), c)
(4.9)
If( p) = {M2( {3L( p ), c)- L({3L( p ), c)N({3L( p ), c) l/2
2bj( p + 1/T1)L({3L( p ), c)
We have now to go back into the real space:
- the Laplace inversion of If ( p) is easy and yields directly the sum of two (for the
C.S.C.S.) or three (for the G.S.C.S.) exponential functions : in terms of spectral
representation, this corresponds to two (for the C.S.C.S.) or three (for the G.S.C.S.)
discrete lines, at times T1 and T2 for the former and TJ, 81 and 82 for the latter, with 81 and
82 lying between T1 and T2 ; note that their intensity can be negative for some values of the
volume fraction!
- the inversion of the second term needs recourse to the inverse Laplace integral
formula

(4.10)

where .1 is a vertical axis which has to leave on its left all the critical points of If ( p). This
is performed after some rewriting of (4.9b). For the C.S.C.S, we write fi L( p) as

J CL( ) _ k Pc( p/12

2 p- c(p+1jT1)(p+1jT2)
Pc(p) =(p + 1jr1)(p + 1jr2 )
(4.11)
k - ~(2-5I/k 2 +2(6-5I)(5f+1)k+(5J- 3/
c- 12bl
k =bJ!b2
where times r1 and r2 lie between T1 and T2,; for the G.S.C.S., we write I?L( p) as

IfL(p)=kc Pc(P/12 ,
~ (p+1jT 1)(p+1jfJ 1)(p+1jfJ2) (4.12)
Pc(P) =(p + 1/rf)(p + 1jr5.J(p + 1/r])(p + 1/r4J

e
where r' 1 and r'2 lie between T1 and 81 and r'3 and r'4 between 1 and 82 . Application
of (4.1 0), of the theorem of residues and Jordan's lemma and definition of adequate cuts on
Structural Morphology and Constitutive Behaviour 331

the real negative axis (see Fig. 20) lead to the following expressions (with T1<T2 to help
discussion):

ff(tJ = ±kc (-}"2 ~ Pc(x) txdx-...

7r -ljr; (x + 1jT1 )(x + 1j()1 )(x + 1j()2)
-I/r4 ~-Pc(x) (4.13)
... - j txdx-...
-Jjrj (x + 1jT1)(x + 1j()1)(x + 1j()2)
7r [~Pc(-1/()J) e-t/9 1 _ ~Pc(-1/()2) e-t/92 ]}
( 1j()2 -1/()J) ( 1jT1 -1/()J) ( 1jT1 -1j() 2)

-1/'tl -1/tl -l!fl

(a) (b)

Fig. 20 Integration paths used for the inversion of the Laplace transform (from [8]):
(a) C.S.C.S., (b) G.S.C.S. (with T1 < T2)

The involved integrals can be written, after putting -r:::: -1/x, in the form of a spectral
representation, namely
332 A. Zaoui

Jl(t) = j g('r)e-tf'rd'r (4.14)

0

which corresponds to the continuous spectrum g('r), whereas the exponential functions
outside the integrals are associated to discrete lines.
Addition of ff(t) for the C.S.C.S. or of fY(t) for the G.S.C.S. lead to the
following final results:
-the shear relaxation spectrum as predicted by the C.S.C.S. consists of a continuous
part and additional discrete lines. The continuous spectrum extends from 'ri to 'r2 (with
TI<'ri<'r2< T2) and its intensity is given by

kcTIT2 ~( 'r- 'ri )( 'r2 - 'r)

gc( 'r) = r::;:-:;- , 'r E { 'rJ> 'r2}. (4.15)
n-v 'ri'r2 'r( 'r- TI)(T2- 'r)
The discrete lines lie at 'r = TI for f ~ .6 (with the intensity (3-5f)/6b I) and at 'r = T2 for
f ~ .4 (with the intensity (5f-2)16b2).
- the shear relaxation spectrum predicted by the G.S.C.S. consists of two separate
continuous parts and additional discrete lines. The continuous spectrum lies between -r'I and
-r'2 and 'r'3 and 'r'4 respectively, with the intensity
('r)= e('r)kGTI(JI(J2 ~('r-'rj)('r-'r2)(-r-'r3)(-r,4 -'r)
gG 7r~'rf-r2-r3-r4 'r('r- TI)((JI- -r)(82- 'r)
'r E { 'rj, -r2J U { -r;, 'r,4} (4.16)
with e('r)=l if 'rE['rf,-r2J and e('r)=-1 if 'rE['r3,-r4].
The discrete lines lie at 'r = TI in a fixed range of volume fraction! and at 'r = (}I or 82.or
both according to the value off, TI, T2, k and hi.
We can see on Fig. 21 and 22 illustrative examples of these results as predicted, for the
same values of the material parameters a;, b; andf, by the C.S.C.S. and the G.S.C.S.
respectively. Though such results seem hard to be corroborated by direct experimental
investigation, they yield useful information on several micromechanical aspects:
- both models predict a non-Maxwellian overall behaviour from Maxwellian
constituents. Whether the resultant spectrum is continuous or not, it definitely does not
reduce to a single line: the "long-range memory effect" may be strong enough to result, even
for a simplistic two-phase material, in a continuous (bounded) spectrum, which means a
continuous infinity of relaxation times which are to be related to the mechanical interactions
between the phases and their viscoelatic nature. This result must be kept in mind in view of
the future treatment of nonlinear viscoelasticity.
- obviously morphology does affect the spectrum shape: one the one hand the
"symmetrical" morphology associated with the C.S.C.S. leads to a unique continuous
spectrum which reflects the underlying intricate phase distribution; on the other hand, the
unsymmetrical morphology inherent to the G.S.C.S. is responsible for a clear spectrum
splitting into two parts and for a spectrum shape which reflects the prominent mechanical
role played by the connected phase. This qualitative difference between the spectra might be
a sensible indicator of morphological changes which can occur during phase transitions,
such as the glass transition of polymers.
Structural Morphology and Constitutive Behaviour 333

The foregoing analysis could be extended to more general constitutive equations than
Maxwell's ones without too much efforts [35] or to more complex morphologies: especially
the existence of an interphase between the matrix and the particles, which can be taken into
account with a four-phase model, can be shown to be responsible for a specific spectrum
part which could be used as a "morphological signature" [36].

0 .l

0. 08

0. 06

0. 04

0. 02

(a)

Fig. 21 Shear relaxation spectrum as predicted by the C.S.C.S.

(ai = az = 1; T1 = 1; Tz = JO;f= .5) (from [8]).

11.5836
0. 08

0. 06

0. 04

0' 02

Fig. 22 Shear relaxation spectrum as predicted by the G.S.C.S.

(aJ = az = 1; T1 = 1; Tz = lO;f= .5) (from [8]).
334 A. Zaoui

4.3 Nonlinear classical self-consistent modelling

4.3.1 Introduction
Let us first stress the fact that the problem of nonlinearity in the context of classical
self-consistent modelling is still an open question. Though it was first addressed by Kroner
[37] thirty-five years ago for the case of elastic-plastic polycrystals, it can be considered as
unsolved yet despite the intensive research which has been, and is still beeing, devoted to
this topic.
Two main approaches have been developed:
- the first one has been initiated by Kroner; it is based on the concept of "stress-free
strain" or "eigenstrain" applied to the plastic or viscoplastic part of the total strain and makes
extensive use of Eshelby's solution of the inclusion problem [1]. Strictly speaking, this
problem refers to an ellipsoidal region of an infinite unloaded elastic medium, undergoing a
uniform eigenstrain, i.e. a prestrain which could be maintained without stress in an isolated
volume element. The basic idea consists in deriving the self-consistent equation from the.
solution of the straightforward extension of Eshelby's problem corresponding to the case
when the matrix itself undergoes the uniform macroscopic plastic or viscoplastic strain and is
subjected to the macroscopic strain or stress at infinity: the inclusion is representative of one
particular phase (usually a .set of identically oriented grains of a polycrystal) while the matrix
is constituted of the unknown H.E.M. which is searched for (usually the homogenized
polycrystal). This approach can be extended easily to inhomogeneous elasticity by referring
to Eshelby's solution of the problem of the "inhomogeneous inclusion".
- the second one refers to a new formulation proposed by Hill [9] for rate-independent
plasticity, starting from the criticism of some aspects of the foregoing approach: it relies
upon a linearization procedure along the prescribed macroscopic loading path and on the
flow theory of plasticity relating the strain and stress rates through the instantaneous
elastoplastic moduli. Now the self-consistent equation can be derived from the solution of
Eshelby's inhomogeneity problem, which is used within a rate (or incremental) formulation.
For proportional loadings, the deformation theory of plasticity can be used as well by dealing
with secant (instead of tangent) elastoplastic moduli. This approach has been extended to
finite strain formulations as well as to rate-dependent plasticity.
Though Hill's approach considerably improved Kroner's one, it will be shown
nevertheless to be not fully self-consistent anymore, so that the problem has still to be
considered as an open one. Before drawing this conclusion, let us go into more details about
the current state-of-the-art in this field.

4.3.2 Linear classical self-consistent modelling with eigenstrains and Kroner's model
In order to appreciate better the limits of Kroner's initial approach of elastoplastic self-
consistent modelling, it is useful to start from apparently a different problem, namely the
linear elastic classical self-consistent modelling when fixed eigenstrains eT are present in the
constituent phases. This is a classical question whose solution is well-known as soon as it as
been solved by any model when no eigenstrains exist. Let A:!! and B:ff be the average
stress concentration tensors for the elastic phase ( r) in this case, so that, according to (2. I)
and (2.2), the overall moduli and compliances are given by

ceff =L frcr: A:!! =< c: A elf>

r
(4.17)
self = L frsr: B:ff =< s: Beff > with B:ff = cr: A:ff: < c: A elf >-I.
r
Structural Morphology and Constitutive Behaviour 335

When eigenstrains are present, the local and global constitutive equations read

a=c:(e-er)
L.:=Ceff:(E-ET) (4.18)
with ET =<1Beff:eT >
or, when er is piecewise uniform per phase, say eJ:

ET =I, J/B:ff: eJ (4.19)

r

If we use the self-consistent scheme, A;csc and B;csc are given by (2.5) and (2.6) with
co = cscsc and the strain concentration relation reads
(4.20)

with eo defined from: < e >=E.

Consider now the case of uniform isotropic elasticity, isochoric eigenstrains denoted
ePr and spherical inclusions. We can see easily that &cscsr
= 0 and e 0 = E, whereas
P~scs is given by (2.19) where the deviatoric part only has to be used, so that (4.20)
becomes
er = E + {3( ef- EP) (4.21)

with EP = <f:'P>. This may be written in the equivalent form

(4.22)
This relation is nothing but Kroner's interaction equation [37] which was the first one to
have been proposed to rule the self-consistent approach to polycrystal plasticity.
This means that Kroner's approacl;l to elastoplastic self-consistent modelling actually
reduces to an elastic one with eigenstrains and does not really takes into account the (stress-
dependent) plastic flow of the polycrystal. Instead of expressing the actual elastoplastic
interactions between the phases, this model is considering far too strong elastic ones: this
stems clearly from the underlying inclusion problem which deals with a uniformly plastified
matrix whose plastic strain EP is not disturbed by the inclusion. Referring to the set of
estimates reported in section 2, we could say that instead of an elastic (i.e. quasi-rigid)
matrix, which leads to a Voigt-type overall behaviour, we should have considered a
plastically flowing matrix, corresponding to the actual polycrystal in the plastic regime. So, it
is not surprising that the stress-strain curves and textures predicted by this model are almost
identical to those derived from the Taylor model which assumes uniform strains. This can be
understood from the fact that, in (4.22), the stress deviations (a- L.:) are of the order of
magnitude of the yield stress whereas the term 2J.1( 1-/3) is almost equal to J.l, i.e. is J02 to
J03 larger, so that the strain fluctuations from grain to grain cannot exceed J0-3 to J0-2,
which coincides practically with Taylor's assumption.
The main point open to criticism in the foregoing approach lies in the fact that the
relations (4.18) have been considered (4.18) as constituve equations for a plastically flowing
inhomogeneous body. As a matter of fact, these equations rule the elastic part of the
336 A. Zaoui

constitutive behaviour only and do not express at all the stress-dependence of the (local as
well as global) plastic flow in the plastic regime. From this point of view, it could be
dangerous to consider the plastic strain as an eigenstrain. On the other hand, stress-
independent eigenstrains can be dealt with according to the foregoing approach even if they
are time-dependent, such as, for instance, thermal strains under variable temperature (we will
make use of this comment later).
Note that for such actual eigenstrains which do not depend on the purely mechanical
loading, Kroner's approach can be applied to linear viscoelasticity as well, with use of the
correspondence principle and the Laplace-transform technique. Suffice it for that to replace in
eqns (4.17) to (4. 22) the real elastic variables by their Laplace-transformed viscoelastic
counterpart. Use of such a possibility will be made later when dealing with nonlinear
viscoelasticity.

4.3.3 Elastoplasticity and viscoplasticity

An alternative treatment of elastoplastic self-consistent modelling was proposed by
Hill, making use of the flow theory of plasticity. Instead of (4.17), the constitutive equations
were written in the form
a=lep:i
(4.23)
t =L~;cs:E
Such a linearization procedure, using the local elastoplastic instantaneous moduli lep and the
overall ones L~;cs and still keeping in mind that these are multi-branched quantities, allows
us to convert the elastoplastic problem into a (pseudo)elastic one: at any step of a given
loading path, the elastoplastic phases of the inhomogeneous material may be considered as
elastic ones for the infinitesimal coming step with moduli lep. Instead of the inclusion
problem with stepwise uniform plastic strains, the corresponding elementary problem is
Eshelby's problem of an ellipsoidal inhomogeneity with uniform moduli lep embedded in an
infinite matrix with uniform moduli L~gcs submitted to the prescribed homogeneous stress
t or strain E rate at infinity (Fig. 23).
E E
H.E.M.
: t;.:~~:::

' (0)'' ' ' ' '

. . . . . . . . . . .
0

•
•

•
•

.
0

•
•

•
•

0
•

. . . . . ·C,t:P . . . . . .
..... . ....
• •
•

0
•

0
•

0
•

•
..
•
•

• • • • •
. .
•

• •

~) w
Fig. 23 Kroner's (a) and Hill's (b) reference Eshelby's problems
Structural Morphology and Constitutive Behaviour 337

The resulting self-consistent equations are nothing but the elastic ones (2. 7) where c
and ccscs are to be replaced by lep and L;;cs respectively. As for the interaction (or
concentration) rule, its reads from (2.5) or (2.6):
fr =(I+ p;scs . 8/;scs r1: < (I+ pCSCS: 8/cscs r1 > -1 . E (4.24)

where P;scs has been defined from the Green operator associated with L;;cs and with
8/;scs =lep r - L;;cs. In case of ellipsoids with the same shape and orientations, this
expression reduces to
fr =(I+ p;scs . 8/;scs r1: E (4.25)
which has been put by Hill in the equivalent form:

a=i+L*:(E-£) (4.26)

where the "constraint tensor" L *, depending on L;;cs and on the shape and orientation of
the ellipsoids, is given by
L* = Lcscs. (Scscs-1 _I)
ep · Esh
(4.27)
8 cscs =pcscs. Lcscs
Esh · ep
with help of Eshelby's tensor sffhCS.
Note that, for proportional monotonic loading, the deformation theory could be used as
well [38]: for isotropic plasticity, it yields an interaction law which only differs from
Kroner's one (4.22) by the fact that the term 2J1( 1-{3) is multiplied by a scalar "elastoplastic
secant accommodation factor" a ep which can be shown to be equal to 1 only in the elastic
regime and to rapidly decrease towards J0- 1 to J0- 2 as soon as the overall plastic flow
occurs. This allows the plastic strain deviations to be larger and this model to yield
predictions which can strongly differ from Taylor's ones, frequently in far better agreement
with experimental results.
A quite similar treatment can be used for the case of viscoplasticity [39] by simply
r~placing in (4.23) lep and L~}cs by their viscoplastic counterpart hvp and H~scs and the
first strain time derivatives Eand E by the second ones f." and E. Both approaches have
been extended to finite strain formulations and have successfully predicted the formation of
crystallographic textures in a number of metal forming situations.

4.3.4 Rate-dependent plasticity and nonlinear viscoelasticity

Any extension of Hill's formulation to rate-dependent elastoplasticity has to deal with
the difficult coupling between elasticity and viscosity: such a coupling makes stress and
strain time-derivatives of different orders appear simultaneously in the constitutive equations.
So, Hill's linearization procedure cannot be used anymore. Some authors [40] argued that
Hill's criticism of Kroner's approach should not apply anymore to rate-dependent plasticity
since the viscoplastic strain rate depends at any time on the current stress and not of the
stress rate so that, dislike the plastic strain, the viscoplastic one EYP could be treated as an
eigenstrain: as we have seen before, this question refers to the mechanical dependence or
independence of the considered strain whatever the mechanical variable may be concerned.
338 A. Zaoui

Other authors [41, 42] did not explicitly discussed this point but they used eqns (4.18)
as basic constitutive equations (with ET = EvP) in a Green analysis at finite strain: it is easy to
see [43] that, in the special case of linear viscoelasticity, this would correspond to the
presumption that a mixture of Maxwellian constituents obeys a Maxwellian overall behaviour
too, which we know to be wrong. By doing so, one is led directly to Kroner's model
(eventually extended to finite strain). For reasons similar to the previous ones, one would
expect from such models predictions hardly different from Taylor's ones: Fig 24 gives an
illustration of this quasi-coincidence ...

\ ... ....:

... ...
-- ...... ----
•••• + ... + + +. +

.... ............. .
·~

• .. • • • + ......

S.C. T.M.
7.0
...··· ··.\
6.0

5.0

0
rn "·o
...___
~ 3.0 - - TAYLOR MODEL

SELF C'JNSlSTINT
2.0

o.o+.-.-~~~.,.,...,~~~~~~~
o.oo o.2s o.so o.75 too t25 tso t75 2.oo
l:~3

Fig. 24 Comparison of Harren's (S.C.) and Taylor's (T.M.) model predictions (from [41])

On the other hand, we cannot deal with the elastic-viscous coupling as we did in the
linear case since nonlinearity prevents to use the convenient Laplace transform technique.
Nevertheless, by use of Hill's linearization procedure, this technique can still be utilized. Let
us first consider [10] a simple local nonlinear viscoelastic behaviour obeying the equation
i=s:cj+g((]) (4.28)
where s are the elastic compliances. Starting from time t = 0, we are supposed to have
determined already the local and overall responses to some prescribed macroscopic loading
path up to t = tn; now we look for these responses during the subsequent infinitesimal time
interval ftn, tn + dt]. So we can approximate (4.28) by:

E(t) = s: {j(t) + mn: (](t) +En (t, tn),

o ag
mn = - [ (](tn)J
a(j
E~ (t, tn) =g[ a(tn)l- mn: a(tn)+... (4.29)
... +{ g[ a(t)]- g[ a(tn)l -mw· [ a(t)- a(tn)l }[ 1- H(t- tn)J,
Structural Morphology and Constitutive Behaviour 339

where H(t- r) is the unit step function at time r. These equations show unambiguously that
the considered linearized behaviour is a Maxwellian one with the eigenstrain rate £~ ( t, tn).
Such a strain is actually a (variable) eigenstrain because its variation is completely known a
priori and does not depend on the external loading which is applied beyond tn; its time
derivative is constant beyond tn and variable but known on [0, tnl· Thus we may use the
method suggested hereabove for this case by using the Laplace or Laplace-Carson transform
technique. The latter is defined by:

f* (p) = p Jf (t )e- pt dt (4.30)

0

It can apply to E~ (t, tn) as well so as to convert the problem into a (symbolical) elastic one
with eigenstrains. As already proved, the whole set of equations is the following:
* 0*
E* = s * : (J +En or (J
* = c *: ( E * -En0* ),

s * (p)=s+___!l_,
m *-1
c *(p) = s ( p ),
p

(4.31)
ceff*( p) =< c*( p): A*( p) > or seff*( p) =< s*( p): B*( p) >,

B * ( p) = c * ( p ): A* ( p ): < c * ( p ): A* ( p) >- 1,
E nO*( p ) =<IB * ( p ):En0* ( p) >.

Due to the nonlinearity, we need an additional interaction relation between the local and
global variables, dependent on the chosen model, in order to determine at each step the
mechanical state of each phase. For the C.S.C.S., this can be made easily from what
precedes by applying the Laplace-Carson transformation to (2.5) and (4.20):
s: cscs·
A rcscs· ( P) = [ I + P,.cscs· ( p ): uC,. -I
( p)} : ...

... :<[I +Pcscs· (p):&cscs* (p);-I >-1.

(4.32)
* s: cscs*
cscs* (p):uc,. -1 *
E,.(p)=[I+P,. (p)] :{E 0 (p)+ ...
cscs· ( p ): [ c,.* ( p ): Eo*• ( p) - C cscs· ( p ): Eno· ( p)] },
... + P,. 111

< E*( p) >= E*( p).

When the ellipsoids have the same aspect ratios and orientation, pcscs is independent of the
phase and the "normalization term" < [I+ pcscs* ( p): &cscs* ( p )F1 > - 1 reduces to unity.
An illustrative example of the application of this method is reported on Fig. 25 and 26
for a shear relaxation test performed on the two-phase material which has been already
studied in the linear case, but now with the local constitutive equations
(4.33)
340 A. Zaoui

with aequ the von Mises equivalent stress; in this particular case, the problem reduces to a
scalar one for which we derive the global stress response S(t) to a stepwise strain loading
E(t) = Eo H(t). Note that use has been made at each step of the closed form solution of the
linear case reported hereabove but that collocation and FFT techniques are also necessary.
The results clearly show both the nonlinearity of the response (Fig. 25) and the variation of
the overall eigenstrain rate E~ (t, tn) at each step (Fig 26).
s
Eo

0.5 Eo= 10

0.0 '---~-----'---~---'---~·---'----~--c.
0.0 1.0 2.0 3.0 4.0

Fig. 25 Normalized shear stress relaxation for a two-phase material

according to the C.S.C.S. (m = 2,/= .5, b1 = b2 = 1, a1 = 1, a2 = 5) (from [44])

·o
En(t,tn)x10
2

0~~----------------------~"~=~1~~"~=~2~----~
n:3
n:L.
-.1

n:20

-.2

t
0.00 0.05 0.10 0.15 0.20

Fig. 26 Time variation of the overall eigenstrain rate at different steps tn

for the case of Fig. 25 (from [44])
Structural Morphology and Constitutive Behaviour 341

This method can be extended [45] to more general rate-dependent elastoplastic

constitutive equations such as the ones which are used frequently in crystalline plasticity,
namely

s
(4.34)
· _ ·O(_EL)v'
Ys- Ys 0 •
·0
'fs =£..
~Hst ·
Yt
'fs

where iis and ms are the unit normal and the unit vector of the slip system ( s), is is the
shear rate on this system and Hst is the so-called "hardening matrix" which may depend on
stress and strain. Such constitutive equations can be put in the form:
e=s: a+ g( a, a)
(4.35)
ci =h( a, a),
where the internal parameters a have been given the form of vectors, with components (aJ)
with I = I toN, for the sake of simplicity: they could be theN reference shear stresses -r? of
(4.34 ). The same linearization procedure as before has now to be applied to both e and a,
which yields:
e(t) =s: a(t) + g(t/1) + m(n): [ a(t)- a(t/1)] + n(n)·f a(t)- a(t/1)]+...
... +{ g(t)- g(t11 ) - m<nl: [ a(t)- a(t,)]- n<nl·[ a(t)- a(t11 )] }[ 1- H(t- t11 )] (4.36)

with:

and
ci(t) =h(t 11 ) +Pin): [ a(t)- a(t11 )] + qln!· [ a(t)- a(t11 )]+...

... +{ h(t)- h(t,)- PrnJ: [ a(t)- a(t,1 ) ] - q1111 .[ a(t)- a(t11 )] }[ 1- H(t- t11 )] (4.37)

with:

We then apply the Laplace-Carson transform to these equations and eliminate a* (p) through
the relation:

(4.38)

tn
j
... + {h(t)- h(t11 ) - P(nJ: [ a(t)- a(t11 )] -q 111 ,.[ a(t)- a(t11 )]} exp( -pt) dt.
0
where I is the second order unit tensor in a NxN space. This leads to the final relation:
342 A. Zaoui

E* (p)=s *(p):a * (p)+E(n)

0 * (p)

* 1 1 -1
with: s ( p) = s + -m(n) + -n(n)· (pi- q(n)J · P(n)
p p

efn/ ( p) = n(n)·( pi- q(n)r1.fi(ni p) + f(ni p)

(4.39)
f(n/ p) = .!_ [ g(tn)- m(n): a(tn)- n(n)· ii(tn) ]+...
p
tn
J{g(t)- g(tn)- m(n): [ a(t)- a(tn)]- n(n)·[ ii(t)- ii(tn)l} exp( -pt) dt.
... +
0

So, with another definition of s*(p) andefn/(p), the nature of the local constitutive
equations is the same as before: the auxiliary strain function efnitJ can be considered as a
true eigenstrain and the homogenization treatment is unchanged.
To illustrate the proposed method, we go back to the same simple example except that
the number of phases is arbitrary and that the local constitutive equations depend on one
scalar internal parameter [45]:

· . b((Jequ)m-1
e=as+ -- s
a
(4.40)
(J
a= h ( -~.P!.f
a
where a, b, h, m and v are material constants which may differ from phase to phase. For a
proportional loading, the problem reduces to a scalar one: we look for some macroscopic
stress relaxation function J.lSCSC(t, tn) and the macroscopic eigenstrain E~j(t, tn)to be
defined at each step tn. The overall stress relaxation function J.lcscs at time tn can be derived
by an iterative scheme with the recursive formula:

(L J(L
-1
CSCS* _
J.l(i+l) -
frJ.l;
cscs* * ·
fr
cscs* *
J
r 3J.l(i) + 2J.lr r 3Jl(i) + 2Jlr ( 4.41)

with: 2}1;( p) = p , T = ar
ar( p+ 1/Tr) r b
r
whereas the interaction law reads

*- 5J1; s* + 6J1;Jlsc * (Eo * o *)

sr- * * (n) -e(n)r
3Jlcscs + 2Jl; 3Jlcscs + 2Jl;
(4.42)
with:
Structural Morphology and Constitutive Behaviour 343

Typical results are given on Figure 27 for a shear test with the prescribed constant shear
strain rate E0 . The normalized stress response S/ E0 depends clearly on the strain rate
amplitude, as expected from the nonlinear local behaviour.
Similar computations applied to rate-dependent elastoplastic polycrystals are now in
progress. Extensions are developed too when instantaneous plasticity is simultaneously
present. We conclude by remarking that the above proposed formulation of the rate-
dependent elastoplastic self-consistent scheme reduces, when elasticity is neglegted, to the
one already proposed (at finite strain) by Molinari et al [46] for viscoplastic polycrystals;
whereas this treatment does not coincide with Hutchinson's [39] one, it is in better
agreement with Ponte Castaneda's recent one [47].
3.0 - -

DC
0
o........ o EDOT = 0.01 ['J'
G - - EJ EDOT = 0. 1
.o · o
.0 0 .13
<7- - £> EDOT = 1 . 0 0
o· D
2.0 .r:r o0
f-- 0
.on.o
no0
A/
0
0
w Do
(i5 8
0'
·"0
,.( - "" "
,., .-:;.·
(f)
w _p:0' /;>. ' ) "0 -
a: I::J '<.I . vlf
f-- ··v ..:>~>"'"
Cf)
13'13 _,, 0 '0 .
1.0 8,§ .o·"
'!:! ""
r:;Yfifoe>.
~0.<7
_'i,f'if

0.0
JJ'/
O.O G-------~----~------~------~------~----~
1.0 2.0
t
3.0
Fig. 27: normalized stress evolution at constant strain rate "edot" (time unit arbitrary)
(2 phases; mr = Vr =2; ar = a,{O) = l;f= .5; b1 =h1= I; bz = hz =5) (from [45]).

4.3 .5 Going back to Hill's assumption

Hill's treatment of the nonlinear classical self-consistent scheme, including its former
extension to viscoplasticity and its latest developments for rate-dependent elastoplasticity,
can be considered as the most advanced approach to polycrystal plasticity. Nevertheless, one
must keep in mind that it is still an approximate one which cannot be thought as fully self-
consistent. Replacing the original Kroner's assumption of piecewise uniform plastic
straining for the solution of the reference inclusion problem by the one of piecewise uniform
instantaneous moduli (or secant ones, or the correspondant Laplace-transformed viscoelastic
moduli for rate-dependent plasticity) allows Hill's treatment to take far better nonlinearity
into account, but this is not beyond reproach: when we have to solve any given boundary-
value problem in elastoplasticity, we know that the instantaneous (or secant) moduli vary
from point to point according to the local stress and strain state ...
There is some consistency in considering the ellipsoidal inclusion tangent (or secant)
moduli as uniform when the matrix ones have been considered so, since the mechanical state
 344 A. Zaoui

of the inclusion is homogeneous in that case. But the reverse is not true: even if the inclusion
is uniformly strained and stressed, the matrix cannot be so, neither can its moduli. And
conversely, if the matrix has not uniform moduli, the inclusion, even if ellipsoidal, is not
uniformly deformed and so its moduli are no more uniform.
A fully self-consistent treatment would have needed to embed the inclusion in the
Homogeneous Equivalent Medium whose constitutive behaviour, as resulting from self-
consistent modelling, should make its instantaneous (or secant) elastoplastic moduli vary
from point to point in any non trivial B.V.P. problem (and so differently according to the
inclusion under consideration· and to its current mechanical state too ... ). As we can
appreciate, such a treatment would need at each step a tremendous number of overlapping
iterative procedures which, despite the exponential increase in time of the numerical potential
and computing power, still seems disheartening.
Nevertheless, we have to make progress in this direction and imagine better and better
approximations. One practical objective could be to force an improved nonlinear self-
consistent approximation to abide by bounds in the simple problems for which it has been
proved to violate them [48]; such improved "variational estimates" could be derived from
bounding approaches by building them up from the self-consistent elastic ones [49, 47].
This suggests that nonlinear improved estimates ·would take advantadge of improved
nonlinear bounding theories (and vice versa?).

4.4 Nonlinear generalized self-consistent modelling

When dealing with nonlinear generalized self-consistent model, the foregoing problem
becomes quite ... generalized: even if Hill's approximation of uniform moduli in the matrix is
adopted, it can obviously not be so in the composite inclusion. So, the problem must be
faced. Several approximations have been proposed: many of them relie on the use of
plasticity deformation theory and secant elastoplastic moduli so as to preserve an isotropic
formulation and to avoid the additional complexity attached to anisotropy. Whether the
deformation or the flow theory is used, it can be combined with the approximation of
stepwise uniform elastoplastic moduli in phase subdomains as small as possible.
Let us quote for instance the approximate treatment of the elastoplastic three-phase
model [23, 50, 51] by decomposing the core and the shell into as many as wished concentric
(or coaxial) spherical (or cylindrical) shells inside which the secant moduli are assumed
uniform: use is then made of the solution [24, 25] of the elastic so-called "n-phase spherical
(or cylindrical) model" with different possible definitions of the secant moduli [50, 51, 52].
In such treatments, the stress or strain definition of the three-phase model [23] has rather to
be used instead of the energetical one. Finite element numerical computations may be
developed as well [50]. The main difficulty indeed is the same as in the case of the classical
self-consistent scheme and is concerned with the way to take the inhomogeneity of the matrix
into account since it is made of the (a priori unknown) H.E.M .. Iterative procedures dealing
with successive known approximations of the H.E.M. could be a possible approach to this
additional difficulty.

5. CONCLUSION

To sum up, we can appreciate that the integration of structural morphology into
micromechanical approaches is still a widely open question. Despite some progress has been
achieved in this field for the last thirty years, many questions still remain unsolved. The
future expected developments are concerned both with improved treatments of nonlinear
behaviour and with efficient practical means to quantify experimentally and to represent
conveniently the main morphological characteristics of inhomogeneous material as well as
their evolution under straining.
Structural Morphology and Constitutive Behaviour 345

What is at stake with these efforts is not so much to be able to predict better the
macroscopic behaviour of given materials as to do so at the mesoscale: several studies [e.g.
50] have shown that morphology plays a major role at this mesoscale in inducing local stress
or strains concentrations which rule damage initiation and evolution, so that one can guess
that any damage micromechanical theory will have to take to the best morphological
parameters into account. In addition, invention of new materials as well improvement of the
existing ones in view of optimized properties for a given use obviously have to get
morphological aspects under control.

Acknowledgements
Michel Bornert and Claude Stolz are friendly and gratefully acknowledged for their
essential contribution to most of the above reported results. Many thanks are especially due
too, between many others, to Eveline Herve and Yves Rougier for their kind and efficient
cooperation.

REFERENCES
1. Eshelby, J.D.: The determination of the elastic field of an ellipsoidal inclusion and
related problems, Proc. Roy. Soc. London, A 241 (1957), 376-396.
2. Kroner, E.: Bounds for effective elastic moduli of disordered materials, J. Mech.
Phys. Solids, 25 (1977), 137-155.
3. Hashin, Z.: The elastic moduli of heterogeneous materials, J. Appl. Mech, 29 (1962),
143-150.
4. Christensen, R.M.: Mechanics of composite materials, Wiley-Interscience, New-York
1979.
5. Stolz, C. and Zaoui, A.: Analyse morphologique et approches variationnelles du
comportement d'un milieu elastique heterogene, C. R. Acad. Sci. Paris, II 312 ( 1991 ),
143-150.
6. Bornert, M., Stolz C.and Zaoui A.: Morphologically representative pattern-based
bounding in elasticity, J. Mech. Phys. Solids, 44 (1996), 307-331.
7. Bornert, M.: A generalized pattern-based self-consistent scheme, Comp. Mat.
Science, 5 (1996), 17-31.
8. Rougier, Y., Stolz, C. and Zaoui, A.: Representation spectrale en viscoelasticite
lineaire des materiaux heterogenes, C. R. Acad. Sci. Paris, II 316 (1993), 1517-1522.
9. Hill, R.: Continuum micro-mechanics of elastoplastic po1ycrystals, J. Mech. Phys.
Solids, 13 (1965), 89-101.
10. Rougier, Y., Stolz C. and Zaoui, A.: Self-consistent modelling of elastic-viscoplastic
polycrystals, C. R. Acad. Sci. Paris, II 318 (1994 ), 145-151.
11. Mori, T. and Tanaka, K.: Average stress in the matrix and average elastic energy of
materials with misfitting inclusions, Acta Metall., 21 (1973), 571-574.
12. Hershey, A. V.: The elasticity of an isotropic aggregate of anisotropic cubic crystals, J.
Appl. Mech., 21 (1954), 236-240.
13. Kroner, E.: Berechnung der elastischen Konstanten des Vielkristalls aus den
Konstanten des Einskristalls, Z. Phys., 151(1958), 504-518.
14. Walpole, L.: On bounds for the overall elastic moduli of inhomogeneous systems- I,
J. Mech. Phys. Solids, 14 (1966), 151-162.
15. Hash in, Z. and Shtrikman, S.: A variational approach to the theory of the elastic
behaviour of multiphase materials, J. Mech. Phys. Solids, 11 (1963), 127-140.
16. Kroner, E.: Self-consistent scheme and graded disorder in polycrystal elasticity, J.
Phys. F., 8 (1978), 2261-2267.
346 A.Zaoui

17. Kroner, E.: Graded and perfect disorder in random media elasticity, J. Engng. Mech.
Div., 106-EM5 (1980), 889-914.
18. Zimmermann, R.W.: Elastic moduli of a solid containing spherical inclusions, Mech.
Mater., 12 (1991), 17-24.
19. Christensen, R.M.: A critical evaluation for a class of micro-mechanical models, J.
Mech. Phys. Solids, 38 (1990), 379-404.
20. Christensen, R.M. and Lo, K.H.: Solutions for effective shear properties in three
phase sphere and cylinder models, J. Mech. Phys. Solids, 27 (1979), 315-330.
21. Love, A.E.H.: A treatise on the mathematical theory of elasticity, Dover Publications,
New York 1944.
22. Eshelby, J.D.: The continuum theory of lattice defects, in: Progress in Solid State
Physics) (Eds. F. Seitz and D. Turnbull), Academic Press, New York 1956,79-144.
23. Herve, E. and Zaoui, A.: Modelling the effective behaviour of nonlinear matrix I
inclusiop composites, Eur. J. Mech. A I Solids, 9 (1990), 505-515.
24. Herve, E. and Zaoui, A.: n-layered inclusion-based micromechanical modelling, Int. J.
Engng ~ci., 31 (1993), 1-10.
25. Herve, E. and Zaoui, A.: Elastic behaviour of multiply coated fibre-reinforced
composites, Int. J. Engng Sci., 33 (1995), 1419-1433.
26. Hashin, Z. and Rosen, B.W.: The elastic moduli of fiber-reinforced materials, J.
Appl. l\1ech., 31 (1964 ), 223-232. ,
27. Herve, E., Stolz, C. and Zaoui, A.: A propos de I' assemblage de spheres composites
de Hashin, C. R. Acad. Sci. Paris, II 313 (1991), 857-862.
28. Milton, G. and Kohn, R.: Variational bounds on the effective moduli of anisotropic
composites, J. Mech. Phys. Solids, 36 (1988), 597-629.
29. Bornert, M.: Ph. D. Dissertation, E.N.P.C., Paris, 1996.
30. Willis, J.R.: Bounds and self-consistent estimatesfor the overall properties of
anisotropic composites, J. Mech. Phys. Solids, 17 (1977), 185-202.
31. Ponte Castefieda, P. and Willis, J.R.: The effect of spatial distributi<?n on the effective
behaviour of composite materials and cracked media, J. Mech. Phys. Solids, 43
(1995), 1919-1951.
32. Suquet, P.: Elements of homogenization for inelastic solid mechanics, in:
Homogenization Tecniques for composite media (Eds. E. Sanchez-Palencia and A.
Zaoui), Lecture Notes in Physics 272, Springer-Verlag, Berlin 1985, 193-278.
3 3. Mandel, 1.: Mecanique des milieux continus, Gauthier-Villars, Paris 1966.
34. Laws, N. and McLaughlin, R.: Self~consistent estimates for the viscoelastic creep
compliances of composite materials, Proc. R. Soc. Lond., A 359 (1978), 251-273.
35. Rougier, Y.: Ph. D. Dissertation, Ecole Polytechnique, Palaiseau, 1994.
36. Beurthey, S.: private communication, 1995.
37. Kroner, E.: Zur plastischen Verformung des Vielkristalls, Acta Metall, 9 (1961), 155-
161.
38. Berveiller, M. and Zaoui, A.: An extension of the self-consistent scheme to plastically
flowing po1ycrystals, J. Mech. Phys. Solids, 26 (1979), 325-344.
39. Hutchinson, J.W.: Bounds and self-consistent estimates for creep of polycrystalline
materials, Proc. Roy. Soc. London, A 348 (1976), 101-127.
40. Weng, G.J.: Self-consistent determination of time-dependent behaviour of metals, J.
Appl. Mech., 48 (1981), 41-46.
41. Harren, S.V.: The finite deformation of rate-dependent polycrystals, J. Mech. Phys.
Solids, 39 (1991), 345-383.
42. Nemat-Nasser, S. and Obata, M.: Rate-dependent finite elastoplastic deformation of
polycrystals, Proc. Roy. Soc. London, A 407 (1986), 343-375.
Structural Morphology and Constitutive Behaviour 347

43. Zaoui, A. and Raphanel, J.L.: On the nature of the intergranular accommodation in the
modelling of elastoviscoplastic behavior of poly crystalline aggregates, in: Large plastic
deformations, Fundamentals and applications to metal forming (Eds. C. Teodosiu,
J.L. Raphanel and F. Sidoroff), Balkema, Rotterdam 1993, 185-192.
44. Zaoui, A. Rougier, Y. and Stolz, C.: Micromechanical modelling based on
morphological analysis; Application to viscoelasticity, in: Microstructure-Property
Interactions in Composite Materials (Ed. R. Pyrz), Kluwer Academic Publishers,
Dordrecht 1995, 419-430.
45. Navidi, P., Rougier, Y. and Zaoui, A.: Self-consistent modelling of elastic-
viscoplastic multi phase materials, in: Micromechanics of plasticity and damage of
multiphase materials (Eds. A. Pineau and A. Zaoui), Kluwer Academic Pub.,
Dordrecht 1996, 123-130.
46. Molinari, A., Canova, G.R. and Ahzi, S.: A self-consistent approach of the large
deformation polycrystal viscoplasticity, Acta Metall. Mater., 35 (1987), 2983-2994.
4 7. Ponte-Castafieda, P.: Exact second-order estimates for the effective mechanical
properties of nonlinear composite materials, J. Mech. Phys. Solids, 44 ( 1996), 827-
862.
48. Gilormini, P.: A critical evaluation for various nonlinear extensions of the self-
consistent model, in: Micromechanics of plasticity and damage of multi phase materials
(Eds. A. Pineau and A. Zaoui), Kluwer Academic Pub., Dordrecht 1996, 67-74.
49 Suquet, P.: Overall properties of nonlinear composites; remarks on secant and
incremental formulations, in: Micromechanics of plasticity and damage of multi phase
materials (Eds. A. Pineau and A. Zaoui), Kluwer Academic Pub., Dordrecht 1996,
149-156.
50. Bornert, M., Herve, E., Stolz, C. and Zaoui, A.: Self-consistent approaches and strain
heterogeneities in two-phase elastoplastic materials, Appl. Mech. Rev., 47 (1994),
S.66-76.
51. Bornert, M.: Morphological effects at the local scale in two-phase materials; some
experimental evidence and tentative modelling, in: Micromechanics of plasticity and
damage of multiphase materials (Eds. A. Pineau and A. Zaoui), Kluwer Academic
Pub., Dordrecht 1996, 27-34.
52. Suquet P.: Overall properties of nonlinear composites: a modified secant moduli theory
and its link with Ponte Castaneda's nonlinear variational procedure, C. R. Acad. Sci.
Paris, II 320 ( 1995), 563-571.