Journal of the Mechanics and Physics of Solids

52 (2004) 2167 – 2183

www.elsevier.com/locate/jmps

Simple algebraic approximations for the e*ective

elastic moduli of cubic arrays of spheres
Israel Cohen∗
School of Physics and Astronomy, Raymond and Beverly Sackler Faculty of Exact Sciences, Tel Aviv
University, Tel Aviv 69978, Israel

Received 1 October 2003; received in revised form 3 February 2004; accepted 7 February 2004

Abstract
Recently, Cohen and Bergman (Phys. Rev. B 68 (2003a) 24104) applied the method of
elastostatic resonances to the three-dimensional problem of nonoverlapping spherical isotropic
inclusions arranged in a cubic array in order to calculate the e*ective elastic moduli. The lead-
ing order in this systematic perturbation expansion, which is related to the Clausius–Mossotti
approximation of electrostatics, was obtained in the form of simple algebraic expressions for
the elastic moduli. Explicit expressions were derived for the case of a simple cubic array of
spheres, and comparison was made with some accurate results. Here, we present explicit expres-
sions for the e*ective elastic moduli of base-centered and face-centered cubic arrays as well,
and make a comparison with other estimates and with accurate numerical results. The simple
algebraic expressions provide accurate results at low volume fractions of the inclusions and are
good estimates at moderate volume fractions even when the contrast is high.
? 2004 Elsevier Ltd. All rights reserved.

Keywords: Elastic material; Particulate-reinforced material; Microstructures; Anisotropic material

1. Introduction

Kantor and Bergman (1982a) introduced an approach to the problem of calculating

the e*ective elastic sti*ness tensor C (e) of two-component composite materials with
a speci<ed microstructure. This approach is based on an expansion of the strained
state inside the composite material in terms of elastostatic resonances (eigenstates):
These are unphysical states where the sample is internally deformed and strained, even
though its boundaries are undeformed. The method was applied to composites made

∗ Tel.: 972-3-6405176; fax: 972-3-6422979.

E-mail address: [email protected] (I. Cohen).

0022-5096/$ - see front matter ? 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jmps.2004.02.008
2168 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183

of two isotropic constituents in the form of regular two-dimensional (2D) arrays of

circular–cylindrical inclusions and to 3D periodic arrays of di*erently oriented circular–
cylindrical inclusions with cubic symmetry (Kantor and Bergman, 1982a, b). In a recent
paper, this approach was applied to a 3D model of cubic arrays of spheres (Cohen and
Bergman, 2003a). This simple model is already a very diAcult problem to solve, mainly
because the eigenstates (i.e., elastostatic resonances) of an isolated spherical inclusion
are not known. In fact, only a few of them were computed—the dipole eigenstates.
This turns out to be a useful exercise because these states are often responsible for the
dominant part of the interaction between distortions of di*erent inclusions, in analogy
with electrostatic problems. When evaluating dipole–dipole interactions, a special care
must be taken in the summation of these interactions over all spheres. There appear
lattice sums which are only semiconvergent, and, therefore, should be treated carefully.
This is done using the concept of the local Lorentz <eld, in a way similar to the one
used to derive the Clausius–Mossotti (CM) approximation in electrostatics, taking into
account the exact microgeometry. The results obtained in this way are in the form of
simple algebraic expressions, similar both in form and in spirit to the CM expression for
macroscopic dielectric constants of such composites. For this reason, we denote these
approximations “CM-type approximations”. Corrections to these expressions, which are
the leading order in a systematic perturbation expansion, begin at order that is not less
than p11=3 (where p is the volume fraction of the spheres).
To determine exactly the e*ective elastic sti*ness tensor of 3D composite materi-
als with arbitrary microstructure is a very diAcult task because of the diAculties and
complications involved in solving the appropriate elastostatic equations in 3D systems.
Practical considerations such as computational e*orts would make it nearly an impossi-
ble task for nondilute dispersions. Most of the published work deals with 2D problems
because of these diAculties, and also because of the fact that the practical problem
of materials that are reinforced with parallel <bers is actually a 2D problem. In or-
der to simplify things, most of the theoretical research on 3D microstructures deals
with the idealized problem in which the centers of the inclusions coincide with lattice
points of a periodic array. The most simple model is that of a cubic array of (nonover-
lapping) spherical inclusions (this simple model is already an anisotropic one). The
spatial cubic symmetry simpli<es matters and enables one to solve numerically for
the e*ective moduli essentially exactly. The e*ective elastic moduli of such composite
materials were calculated by Nunan and Keller (1984) for the case of rigid inclusions,
by Nemat-Nasser and Taya (1981) and by Nemat-Nasser et al. (1982) for the case of
voids, and by Sangani and Lu (1987) and by Kushch (1987) for the case of elastic in-
clusions. For inclusion shapes other than spherical, we mention the approximate numer-
ical solution of Iwakuma and Nemat-Nasser (1983) for ellipsoidal inclusions, and the
multipole expansion method of Kushch (1997, 1998) for spheroidal inclusions, which
seems to be accurate, simple and e*ective from a computational standpoint. Another
recent advancement is the work of Torquato (1997) who developed new perturbation
expansions for the e*ective sti*ness tensor which are absolutely convergent. These
expansions involve n-point correlation functions which characterize the microstructure.
Explicit third-order expressions were derived for isotropic (random as well as
periodic) dispersions. The result for the bulk modulus of 3D dispersions applies also to
 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183 2169

composites with cubic symmetry (Torquato, 1998). A numerical method which is not
restricted to nonoverlapping inclusions was proposed by Cohen and Bergman (2003b)
to determine the e*ective elastic sti*ness tensor of arbitrary periodic composite
materials.
The purpose of this paper is not to introduce another accurate numerical method to
compute the elastic moduli, but instead to present simple but powerful approximations
as will be demonstrated below. We present explicit CM-type approximation for the
elastic moduli of three types of cubic arrays: simple cubic, base-centered cubic (BCC)
and face-centered cubic arrays (FCC) of spheres. The expression for the e*ective bulk
modulus is the same for all three types of cubic arrays and coincides with one of the
Hashin–Shtrikman (1963) bounds. The expressions for the two e*ective shear moduli
are di*erent for the three di*erent types of cubic arrays. We have compared these
approximations with the above-mentioned previous work. The simple algebraic expres-
sions seem to provide accurate results at low volume fractions of the inclusions even
when the contrast is in<nite; they continue to be good estimates at moderate volume
fractions, and, as expected, they fail to do so when the volume fraction is high, espe-
cially when p approaches its close-packing value pc . For the case of in<nite contrast,
i.e., rigid spheres embedded in a compressible matrix, the CM-type expressions seem
to be good estimates up to about 60% of pc for the bulk modulus, and up to about
50% of pc for the two shear moduli.

2. Summary of the underlying theory

We outline the main ideas of the method of elastostatic resonances and its implemen-
tation to the 3D case of a cubic array of spheres (Cohen and Bergman, 2003a). Our
model consists of a cubic array of nonoverlapping spherical inclusions made of isotropic
material C (1) (1 ; 1 ) (component 1) embedded in an isotropic host C (2) (2 ; 2 ) (com-
ponent 2) ( is the bulk modulus,  is the shear modulus). The e*ective elastic sti*ness
tensor C (e) is de<ned by the usual volume-averaged relation:
 = C (e)
, were  is
the stress tensor and is the strain tensor inside the composite material. The approach
begins by introducing a somewhat generalized form of the original problem, replacing
(1) (1)
the C (1) material by a di*erent material C
(s), where C
(s) = C (2) − (1=s) C, and
C ≡ C (2) − C (1) . This replacement also makes C (e) a function of the parameter s.
(1)
When s lies in certain ranges the tensor C
(s) becomes unphysical, i.e., it ceases
to be positive de<nite. The original problem is retrieved by setting s = 1. The strain
tensor (r) inside such a composite material, the boundaries of which undergo the
displacement ui = ij(0) xj , is the solution of the operator equation

(0) 1 ˆ
= +  ; (2.1)
s

where ij(0) is any constant symmetric tensor and ˆ is a linear integral operator that is
related to the tensor Green function of the problem. The eigenstates (n) of the operator
ˆ (ˆ (n) = sn (n) ) represent elastostatic resonances of the sample, i.e., states where the
2170 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183

sample is internally deformed and strained even though its boundaries are undeformed.
(1)
Obviously, such resonances can occur only at unphysical values of C
(s). The way
to proceed is to substitute the formal solution of Eq. (2.1) for the strain <eld in the
de<nition for the e*ective elastic tensor, and then expand in terms of the complete
set of eigenstates (n) . This way, a spectral representation of a related scalar function
F(s) is obtained (summation over tensorial indices is implied)

Fn
(0)
F(s) ≡ C (2) (0)
− (0)
C (e) (0)
= ; (2.2)
n
s − sn

where the poles sn are the eigenvalues of ,ˆ and the weights Fn are real and calculated
from knowledge of the resonance states of the composite material.
To solve the problem of many inclusions inside a host material, one uses an ap-
proach similar to the tight-binding method from solid-state crystal electronics in order
to expand the eigenstates (n) of the many inclusions problem in terms of the com-
plete set of eigenstates (a) of the individual inclusions, denoted by index a. A great
simpli<cation occurs if the inclusions are identical and form a periodic array in space.
In that case, Bloch’s theorem immediately speci<es√ the dependence of the expansion
(n) (n)
coeAcients Ba on the inclusion index a, Ba = (1= N )B(n) eik·ra , where N is the num-
ber of unit cells in the periodic sample, and ra represents the location of inclusion a.
Since only the k = 0 Bloch states can have nonzero weights, only these states need to
be considered. The eigenvalue problem for the k = 0 eigenstates is


(sn − s )B(n) = Q B(n) ; (2.3)


where


s b (b)∗ C (a) dV ∗
Q = ; Q = Q : (2.4)
a  (b)   (a) 
a=b

Note that sn is the eigenvalue of the many inclusions problem to be found, and s
is an eigenvalue of the single inclusion problem, which is assumed to be known,  
denotes the norm of a state , and b (r) equals 1 only inside inclusion b and vanishes
elsewhere. The matrix elements Q take into account the interaction between the strain
<elds of di*erent inclusions in the form of overlap integrals between isolated inclusion
eigenstates of pairs of di6erent inclusions. Having found the eigenvectors B(n) and
eigenvalues sn , the weights are given by the expression
 2 
(n) 
1 
B a (0) C (a) dV 

Fn =   |B(n) |2 ; (2.5)
Va  
 (a)  


where Va is the volume of a unit cell in the periodic composite structure. Once the
weights Fn and eigenvalues sn are obtained, we can calculate the e*ective elastic moduli
from F(s).
 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183 2171

3. Cubic array of spheres

The two-body interaction between di*erent spherical inclusions, which appears in

the form of overlap integrals in Eq. (2.4), decreases with increasing distance between
inclusions. In connection with this, the elements Q always behave like some positive
power of the volume fraction of the inclusions p, when p is small. From the 2D
elastostatic problem (Kantor and Bergman, 1982a) of circular–cylindrical inclusions,
and from the 3D electrostatic problem of spherical inclusions (Bergman, 1979), we
expect that, in order to include the e*ects of these interactions to leading order in
p, we only need to consider the dipole resonances of the individual spheres. These
states correspond to a strain <eld which is uniform inside the inclusion (a sphere) and
decreases slowest with distance outside the inclusion (decreases like 1=r 3 ). In the 3D
elastostatic case of a cubic array of spheres, these dipole resonances were chosen to
be certain uniform strains inside the sphere, and then their shape outside the sphere
and their eigenvalues were found following the procedure described in Appendix B of
Cohen and Bergman (2003a). The single inclusion eigenstates are found by solving the
(1)
equilibrium equations of elastostatics for a spherical inclusion C
(s) that is embedded
(2)
in an in<nite host C whose boundaries are undeformed. In order to form a complete
biorthogonal set, there are six dipole eigenstates (as this is the number of constant
linearly independent second rank symmetric tensors in three dimensions): a compression
dipole (1) which corresponds to a pure, uniform compression inside the sphere, and
<ve shear dipoles (2) ; : : : ; (6) . The eigenvalue of (1) is denoted s1 , and the common
eigenvalue of the <ve (degenerate) shear dipoles is denoted s2 . Having found the dipole
eigenstates of the individual inclusions, we have to calculate the matrix elements Q ,
;  = 1; : : : ; 6, which take into account the dipole–dipole interactions between di*erent
inclusions.
The matrix element Q represents the sum of interactions between inclusion b lo-
cated at the origin and all the other inclusions located at lattice points ra . A special
problem arises in the summation of the dipole interactions: there appear lattice sums of
terms of the form Pl(0) (a )=ra3 (similar in form to a dipole <eld in electrostatics), which
are only semiconvergent. In these sums, the distant contributions are as important as
the nearby ones, and the series converges only thanks to the alternating signs of the an-
gular functions. However, as was already noted, the isolated inclusion eigenstates were
calculated assuming that the isolated sphere was far away from the sample boundaries,
which is not the case for distant spheres near the sample boundaries. Therefore, we can
use the overlap integrals of these states only for the nearby lattice sites. The problem
is solved by means of the concept of the local Lorentz <eld which is used to obtain
the CM approximation in the analogous problem in electrostatics. The treatment of
dipole–dipole interactions in the “language” of resonances was originally introduced
by Bergman (1979) for the analogous problem of electrostatics, in order to obtain the
CM approximation. It should be noted that (1) this is not the standard approach to ob-
tain the CM approximation (for the standard approach, see, e.g., Ashcroft and Mermin
(1976), (2) since the CM approximation involves summation of dipole interactions, it
depends on the symmetry of the medium, i.e., whether the medium is isotropic, or has
cubic or other kind of symmetry.
2172 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183

The CM-type approximation is obtained by breaking up the sum in Eq. (2.4) into
near terms, which are evaluated exactly for a cubic array of spheres that are far away
from the boundaries, and far terms, which are evaluated by replacing the discrete dipole
array by a uniform continuous strain polarization and taking into account the correct
boundary conditions (zero strain <eld at the sample boundaries). The near terms are
obtained by considering those inclusions that lie within a sphere of radius L around
the central inclusion (inclusion b), and the far terms are obtained by considering those
inclusions that are at distance r ¿ L from the central sphere. Doing that, we obtain the
following modi<ed expression for Q (see Cohen and Bergman (2003a)):


s b (b)∗ C (a) dV
Q = − ps  : (3.1)
a  (b)   (a) 
0¡|rb −ra |¡L

In this equation, L must be large enough so that the use of average polarization density
for r ¿ L is a good approximation. In practice, one sums the series over a set of spheres
with larger and larger radius L until convergence is achieved. The second term on the
right-hand side of Eq. (3.1) represents the contribution of the distant spheres when we
take into account the correct boundary conditions.
The e*ective elastic tensor of a composite material with cubic symmetry has the
form
(e)  
Cijkl = e ij kl + 2e (Iijkl − ijkl ) + 2Me ijkl − 13 ij kl ; (3.2)

where  = (C1111 + 2C1122 )=3 is the bulk modulus,  = C1212 ; M = (C1111 − C1122 )=2
are two shear moduli which coincide in the isotropic case, and

1; i = j = k = l;
ijkl ≡ (3.3)
0 otherwise:

In order to isolate only one of the following three e*ective elastic moduli e ; e , or
Me in Eq. (2.2), we choose (0) to be one of the following:
(0) 1 (0) (0M ) 1
ij = 3 ij ; ij = Iij12 ; ij = √ (Iij11 + Iij22 − 2Iij33 ); (3.4)
12
where Iijkl = 12 ( ik jl + il jk ). The form of the eigenstates (1) ; (2) , and (6) inside
the inclusion was chosen to be proportional to (0) ; (0) , and (0M ) , respectively. This
fact, together with the orthogonality of the eigenstates implies that by choosing, e.g.,
(0)
= (0) , the only contribution to the numerator in Eq. (2.5) comes from the (1)
state. From cubic symmetry arguments it follows that Q (;  = 1; : : : ; 6) is diagonal.
As a consequence, when taking into account only dipole–dipole interactions, the dipole
eigenstates of the many inclusion problem are the same as those of the single inclusion
problem, but with shifted eigenvalues.
If both states are compressional dipoles (1) , then all the overlap integrals in Eq.
(3.1) vanish, and the shifted eigenvalue of the whole system is

s() = s1 + Q11 = (1 − p)s1 ; s1 = 3(2 − 1 )=(32 + 42 ): (3.5)

 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183 2173

Thus, taking into account only the strongest (dipole–dipole) interactions, we obtain
from Eq. (2.2) a CM-type approximation for e
1 VR 
2 −  e ∼
=
Va 1 − s()
p
⇒ e ∼
= 2 − ; (3.6)
1=(2 − 1 ) − 3(1 − p)=(32 + 42 )
where VR is the volume of an inclusion (a sphere), and p=VR =Va is the volume fraction
of the inclusions. This expression for e applies to simple cubic (SC) arrays, as well
as to BCC and FCC arrays of spheres. It is identical to one of the Hashin–Shtrikman
(HS) (1963) bounds, and was proven by Hill (1963) to be a bound on the e*ective
bulk modulus of composite materials with cubic symmetry.
Unlike the situation for e , the overlap integrals between two (2) states (for e ) or
between two (6) states (for Me ) do not vanish and must be summed numerically in
Eq. (3.1). If we denote these sums by s2 G ; s2 GM , respectively, then
Q22 = s2 (G − p); Q66 = s2 (GM − p); (3.7)
therefore, the shifted eigenvalues of the whole system are
s() = s2 + Q22 = (1 − p + G )s2 ; s(M ) = s2 + Q66 = (1 − p + GM )s2 ; (3.8)
where s2 is the common eigenvalue of the <ve shear dipole eigenstates of the single
inclusion problem.
6(2 + 22 ) 
s2 = ;  =  2 − 1 : (3.9)
5(32 + 42 )2
The CM-type approximations for the two e*ective shear moduli are
p 
e ∼
= 2 − ; (3.10)
1 − (1 − p + G )s2
p 
Me ∼
= 2 − : (3.11)
1 − (1 − p + GM )s2
The terms GM , G obey the relation 3G = −2GM . A detailed explanation of the way
to calculate GM is found in Appendix D of Cohen and Bergman (2003a). It has the
form
32 + 2
GM = (A1 p + A2 p5=3 ); (3.12)
2 + 22
where the values of the numerical coeAcients A1 and A2 depend on the microgeometry.
In order to determine the values of A1 and A2 , one has to sum over all lattice points
ra the terms P4 (cos a )=(a=a0 )3 for A1 , and the terms P4 (cos a )=(a=a0 )5 for A2 , where
ra = (a; a ; (a ) is the location of inclusion a, P4 (cos ) is a Legendre polynomial,
and a0 is the length of a unit cell in the lattice. Note that the <rst sum is only
semiconvergent, whereas the second one is absolutely convergent. These sums should
be computed numerically for each periodic array. Results for SC, BCC and FCC arrays
are presented in the <rst row in Table 1.
The expressions for the two shear moduli e and Me , Eqs. (3.10) and (3.11), respec-
tively, begin to di*er already at order p2 . The di*erence (Me − e ), which is a measure
2174 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183

Table 1
CoeAcients A1 , A2 in the expression for GM for the three types of cubic arrays (<rst row)

SC BCC FCC

A1 A2 A1 A2 A1 A2

−0.92900 1.1422 0.28640 −0.35958 0.25463 −0.27438

NK −0.930 1.143 0.287 −0.347 0.258 −0.274
SL −0.9290 0.2864 0.2546

Also shown are results of Nunan and Keller (NK), and results of Sangani and Lu (SL), which were
obtained for the special case of rigid spheres.

of the degree of anisotropy of the material, is proportional to (−GM ) at that order.

This fact and the numerical values of A1 ; A2 (see Table 1) imply that the behavior of
the shear moduli of SC arrays is qualitatively and quantitatively di*erent from that of
BCC and FCC arrays (at least in the range of volume fractions where the CM-type
approximation is a good approximation): (Me − e ) is positive for SC arrays, while it
is negative for BCC and FCC arrays, the absolute value of (Me − e ) being markedly
larger for SC arrays than for BCC and FCC arrays. We should expect that at low and
moderate volume fractions the value of e and the value of Me should be similar for
BCC and FCC arrays, and that the di*erence between e and Me should be small at
low volume fraction for both BCC and FCC arrays. These conclusions are supported
by numerical results.
Since the terms GM , G obey the relation 3G =−2GM , we can <nd a combination of
the two shear moduli, which is independent of these terms to <rst order (i.e., correction
to the following expression begin at order p3 ):
3 2 p 
e + Me ∼ = 2 −
5 5 1 − (1 − p)s2
p
= 2 − 1 6(2 +22 )(1−p)
: (3.13)
2 −1 − 5(32 +42 )2

This expression is identical to one of the Hashin–Shtrikman (1963) bounds for e in

the case of a composite with an isotropic microstructure, and provides a similar bound
for the combination (3e + 2Me )=5 in the case of a cubic microstructure (Kantor and
Bergman, 1984).
In order to <nd higher-order corrections to the CM-type approximations, Eqs. (3.6),
(3.10) and (3.11), one has to continue the perturbative treatment and calculate matrix
elements Q between the corresponding dipole eigenstates and nondipole eigenstates.
Since nondipole eigenstates decrease with distance faster than 1=r 3 , these corrections
will be of higher order in p. Using standard second-order perturbation theory, it was
argued by Cohen and Bergman (2003a) that the next correction to the CM-type ap-
proximations for a simple cubic array begin at order that is not less than p11=3 . Since
similar arguments apply to BCC and FCC arrays as well, we conclude that the next
correction to the CM-type approximations for all the three types of cubic arrays begin
at that order.
 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183 2175

4. Comparison and discussion

It is interesting to note that asymptotic formulas for low concentrations, which were
derived by Nunan and Keller (1984) for the e*ective shear moduli of cubic arrays of
rigid spheres (1 =2 = 1 =2 = ∞), have similar form to the CM-type approximations.
In the special case of rigid spheres, the CM-type expressions become
(32 + 42 )p
e ∼
= 2 + ; (4.1)
3(1 − p)
5(32 + 42 )2 p
e ∼
= 2 + ; (4.2)
6(2 + 22 ) (1 − p + G )
5(32 + 42 )2 p
Me ∼= 2 + : (4.3)
6(2 + 22 ) (1 − p + GM )
Based on results for the e*ective viscosity, Nunan and Keller (NK) guessed the form
of asymptotic formulas for low concentrations for the e*ective elastic moduli. The
parameters A1 ; A2 in Eq. (3.12) were determined by <tting the asymptotic formulas to
accurate numerical results. The results obtained for these parameters are presented in
the second row of Table 1 (in the notations used by Nunan and Keller A1 = 2h1 =3,
A2 = 2h2 =3). There is a good agreement with our results. Asymptotic formulas for
the case of rigid spheres were also derived by Sangani and Lu (SL) (1987) by the
method of successive approximations. Their asymptotic formulas consist of only the
leading order term in p in the expressions for GM and G , therefore, results for A1
only are presented in the third row in Table 1 (in the notations of Sangani and Lu
A1 = 2h1 =3). Note the excellent agreement with our results. Neither Nunan and Keller
nor Sangani and Lu derived an explicit asymptotic formula for e , but if we require
that their asymptotic formulas for the components of the elastic sti*ness tensor yield
the correct form of Eq. (4.1), at least up to order p2 , we obtain a relation between the
parameters that were used in their formulas: h3 = 4 + 5h1 (in terms of the notations
used in both these papers e =2 = 1 + (* + 2=3)2 =2 , the parameter h3 is denoted h2
in SL), which is in agreement with their results. It should be noted that h3 has a sign
error in Table 3 of Nunan and Keller (1984).
In order to the test the accuracy of Eqs. (3.6),(3.10) and (3.11) we compare them
with previous numerical work. As a <rst test, we consider the case of a simple cu-
bic array of spherical voids. In this arrangement, the spheres begin to overlap at the
close-packing volume fraction pc =+=6 0:52. The Poisson ratio ,=(3−2)=(2(3+
)) of the matrix (component 2) is ,2 = 0:3. Results for the e*ective bulk modulus are
presented in Table 2, and results for the e*ective shear moduli are presented in Table
3. Note the excellent agreement between the CM-type approximations (CM) and the
numerical results of Iwakuma and Nemat-Nasser (IN) (1983) for the three e*ective
moduli over the whole range of volume fractions. The numerical results of Iwakuma
and Nemat-Nasser involve some approximation, therefore, neither the CM-type approx-
imation nor the results of Iwakuma and Nemat-Nasser are expected to be accurate at
high volume fractions. There is a good agreement between the results of all the dif-
ferent methods up to p = 0:3, with the exception of the results of Sangani and Lu
(SL) (1987) for e (in terms of the notations used by Sangani and Lu, e =2 = 1 + ,
2176 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183

Table 2
E*ective bulk modulus e =2 of a simple cubic array of spherical voids, ,2 = 0:3

p E*ective bulk modulus e =2

CM IN Ku To SL

0.05 0.879 0.881 0.879 0.879

0.10 0.774 0.777 0.774 0.774 0.774
0.15 0.683 0.687 0.683 0.683
0.20 0.604 0.607 0.602 0.602 0.602
0.25 0.533 0.537 0.530 0.532
0.30 0.471 0.474 0.465 0.464 0.465
0.40 0.364 0.367 0.341 0.345 0.349
0.50 0.276 0.279 0.232 0.242

Comparison of the CM-type approximation (CM) with the results of Iwakuma and Nemat-Nasser (IN)
(1983), Kushch (Ku) (1987), Torquato (To) (1998), and Sangani and Lu (SL) (1987).

Table 3
E*ective shear moduli e =2 ; Me =2 of a simple cubic array of spherical voids, ,2 = 0:3

p E*ective shear modulus e =2 E*ective shear modulus Me =M2

CM IN Ku SL CM IN Ku SL

0.05 0.905 0.906 0.906 0.914 0.916 0.914

0.10 0.812 0.814 0.812 0.817 0.841 0.843 0.841 0.841
0.15 0.724 0.726 0.736 0.777 0.779 0.777
0.20 0.641 0.643 0.640 0.665 0.719 0.720 0.718 0.718
0.25 0.565 0.567 0.605 0.664 0.666 0.668
0.30 0.496 0.498 0.490 0.554 0.612 0.613 0.608 0.608
0.40 0.379 0.381 0.362 0.470 0.512 0.514 0.509 0.504
0.50 0.288 0.290 0.375 0.413 0.415 0.394

Comparison of the CM-type approximations (CM) with the results of Iwakuma and Nemat-Nasser (IN)
(1983), Kushch (Ku) (1987), and Sangani and Lu (SL) (1987).

Me =2 = 1 + ). We suspect that the results of Sangani and Lu for e are inaccurate
for two reasons: (1) they contradict the general behavior of the two shear moduli: the
di*erence (Me − e ), which is a measure of the degree of anisotropy of the mate-
rial, should increase with increasing inclusion volume fraction p, whereas it decreases
above p = 0:25 according to the results of Sangani and Lu, (2) their results for the
combination (3e + 2Me )=5 are above the upper bound for this combination (see Eq.
(3.13)) for all values of p. Note that (Me − e ) is indeed positive for a simple cu-
bic array, as predicted by the CM-type expressions. The third-order approximation of
Torquato (To) (1998) for e , which incorporates third-order correlation functions of
the microstructure, includes corrections to the HS expression (3.6), therefore, it should
be more accurate than the CM-type approximation. We expect it to be a good approx-
imation up to about p = 0:4. Of the above results, the numerical results of Kushch
(Ku) (1987), who used a multipole expansion method, are the most accurate ones,
 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183 2177

Table 4
E*ective elastic moduli of a simple cubic array of spheres, ,1 = ,2 = 0:3, 1 =2 = 3

p Bulk modulus e =2 e =2 Me =M2

CM IN To CM IN CM IN

0.05 1.046 1.046 1.046 1.051 1.052 1.054 1.054

0.10 1.095 1.095 1.095 1.103 1.104 1.115 1.116
0.15 1.146 1.147 1.146 1.157 1.157 1.181 1.182
0.20 1.201 1.202 1.201 1.212 1.213 1.254 1.255
0.25 1.259 1.260 1.260 1.270 1.271 1.331 1.333
0.30 1.321 1.323 1.323 1.332 1.333 1.413 1.415
0.40 1.459 1.461 1.464 1.467 1.469 1.588 1.590
0.50 1.618 1.620 1.634 1.627 1.629 1.771 1.774

Comparison of the CM-type approximations (CM) with the results of Iwakuma and Nemat-Nasser (IN)
(1983). For the bulk modulus we also show results of Torquato (To) (1998).

in our opinion. As is evident from Tables 2 and 3, the CM-type approximations are
accurate at low and moderate volume fractions. They can serve as good estimates for
the e*ective moduli up to about p = 0:3, at which the deviation is about 1% from the
exact result.
The resemblance between the results of Iwakuma and Nemat-Nasser (IN) (1983)
and the results by the CM-type approximation for the case of spherical voids is not
coincidental, as can be seen from Table 4. The results by the two methods for the
elastic moduli for the case ,1 = ,2 = 0:3; 1 =2 = 3 are almost identical. Since the
method of Iwakuma and Nemat-Nasser involves calculation of several in<nite series
in Fourier space, we believe that the di*erence between their results and those by the
CM-type expressions are due to truncation errors in the calculation of these sums. In
fact, Nemat-Nasser et al. (1982), who investigated the special case of a periodic cubic
array of spherical voids, have noticed that their approximation for the e*ective bulk
modulus, the so-called “simplest approximation”, overshoots the upper HS bound by
about 1%, possibly due to truncation errors in the numerical calculation. They have
further claimed that if a particular relation is to hold, then their approximation can be
proved to be identical to the upper HS bound. Their arguments (see Appendix E in
Nemat-Nasser et al. (1982)) can be easily generalized to the case of inclusions with
arbitrary elastic moduli, therefore, their results for the bulk modulus are identical to
one of the HS bounds for the general case, and are the same as the CM-type expression
for e . In view of our results, it seems that the approximation that they use, which
involves replacing a nonuniform strain <eld inside the inclusion by its average value, is
the analogue of the CM-type approximation (in the CM-type approximation we consider
only the dipole eigenstates of the single inclusion problem, and these states correspond
to a strain <eld that is uniform inside the inclusion). Also shown in Table 4 are
results of Torquato’s third-order approximation. As can be seen, for this case of small
contrast (1 =2 = 1 =2 = 3) the di*erence between the CM-type approximation and the
third-order approximation is very small even at high volume fractions. At p = 0:5 this
di*erence is less than 1%, whereas it is about 20% for the case of spherical voids (see
2178 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183

Table 5
(e)
E*ective elastic constant C1111 of a simple cubic array of spheres, ,1 = ,2 = 0:3

1 =2 p = 0:1 p = 0:3 p = 0:5

CM Kushch CM Kushch CM Kushch

0 2.799 2.799 1.836 1.818 1.148 1.048

10 4.101 4.102 5.880 5.930 8.645 9.646
1000 4.248 4.248 6.758 6.887 11.453 17.710

Comparison of the CM-type approximations (CM) with the results of Kushch (1997).

Table 3). We should expect that the CM-type expressions provide good estimations of
the elastic moduli over the whole range of volume fractions up to pc when the ratio
between the elastic moduli of the matrix and that of the inclusions is not too di*erent
from 1 (e.g. 0.5, 2).
In Table 5 we compare the CM-type approximation with accurate numerical results
obtained by Kushch (1997) for di*erent values of the elastic moduli. At small volume
fraction p = 0:1 the CM-type approximation is accurate even when the value of the
elastic moduli of the inclusions is three orders of magnitude bigger than that of the
matrix. At moderate volume fraction p = 0:3 (pc = 0:52), the CM-type approximation
is less than 1% o* the exact result when 1 =2 = 0; 10, and is 1.9% o* the exact result
for the case with the highest contrast 1 =2 = 1000. At p = 0:5 the deviations become
considerable, as expected.
As a <nal test, we compare the CM-type expressions with accurate numerical re-
sults obtained by Nunan and Keller (1984) for the extreme case of rigid spheres
(1 =2 = 1 =2 = ∞), ,2 = 0:4. In terms of the notations used by these authors,
e =2 = 1 + (* + 2=3)2 =2 , e =2 = 1 +  and Me =2 = 1 +  (since the data presented
here for the parameters ;  and * was extracted from the graphs of Nunan and Keller,
it accumulated some error). The CM-type approximation for the bulk modulus, Eq.
(3.6), is identical to one of the HS bounds, and is valid for SC, as well as BCC and
FCC arrays of spheres. In Fig. 1 we present the results of Nunan and Keller for the
three types of cubic arrays, along with the three-point approximation of Torquato (Eq.
(4.15) in Torquato (1998), where we used the tabulations of the three-point parameter
-2 obtained by McPhedran and Milton (1981)), and the CM-type approximation, which,
as was already mentioned, is a lower bound for the bulk modulus of composites with
cubic symmetry. As can be seen, the CM-type approximation is almost indistinguish-
able from the results of Nunan and Keller up to about p = 0:3 for SC arrays, and up
to about p = 0:45 for BCC and FCC arrays. If we remember that pc = 0:52; 0:68; 0:74
for SC, BCC and FCC arrays, respectively, then the CM-type approximation seem to
be a good estimate for the bulk modulus up to about 60% of pc in this extreme case
of in<nite contrast. The bulk modulus of the three types of cubic arrays is the same
at low and moderate volume fractions, and the di*erences between the di*erent arrays
become apparent as p approaches the close packing value pc of each type of cubic
array. The third-order approximations of Torquato for BCC and for FCC arrays are
 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183 2179

κ1/ κ2=∞, µ1 / µ2=∞, ν2=0.4 (µ2 /κ2=0.2143)

CM−type
4.5
Torquato (SC)
Torquato (BCC)
4 Torquato (FCC)
Nunan & Keller (SC)
3.5 Nunan & Keller (BCC)
Nunan & Keller (FCC)
κe / κ2

2.5

1.5

1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
p − inclusion volume fraction

Fig. 1. Normalized e*ective bulk modulus e =2 vs. inclusion volume fraction. Rigid spheres (component 1)
embedded in a compressible matrix (component 2, ,2 = 0:4): · · ·, CM-type approximation for SC, BCC and
FCC arrays; SC array: ×, Nunan and Keller; —, Torquato; BCC array: +, Nunan and Keller; - -, Torquato;
FCC array: ◦, Nunan and Keller; · –, Torquato.

almost indistinguishable in the <gure. The di*erence between Torquato’s third-order

approximation and the CM-type approximation becomes apparent only at high volume
fractions, as p approaches pc (we note that due to an error in Fig. 1 in Cohen and
Bergman (2003a), Torquato’s third-order approximation appeared to be a better ap-
proximation than it really is; it should not be a good approximation for e near pc in
the case of rigid spheres). In Figs. 2–4 we compare expressions (3.10) and (3.11) for
the two shear moduli with the numerical results of Nunan and Keller (1984) for the
three types of cubic arrays. Eqs. (3.10) and (3.11) seem to be good estimates for e
and Me , respectively, up to about 50% of pc . The value of (Me − e ) is positive for SC
arrays, while it is negative for BCC and FCC arrays, the absolute value of (Me − e )
being markedly larger for SC arrays than for BCC or FCC arrays. We note that the
shear moduli of BCC and FCC arrays are almost identical up to about p = 0:35.
As the volume fraction p of the spherical inclusions increases and the distance be-
tween the spheres decreases, the interaction between the distortion <elds of di*erent
inclusions becomes stronger and it no longer suAces to consider only the dipole–dipole
interaction, as is done in the CM-type approximation. In order to describe this strong
interaction more accurately, one has to consider also the nondipole eigenstates, i.e., the
higher multipole distortion <elds. As was already stated, in order to <nd the next cor-
rection to Eqs. (3.6), (3.10) and (3.11), one has to continue the perturbative treatment
and calculate matrix elements Q between the corresponding dipole eigenstates and
nondipole eigenstates. From the analogous 3D electrostatic problem of a cubic array of
2180 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183

κ1/ κ2=∞, µ1 / µ2=∞, ν2=0.4 (µ2 /κ2=0.2143)

10

9 Simple cubic array of rigid spheres

8
CM−type M
CM−type µ
7
Nunan & Keller M
µe / µ2

Nunan & Keller µ

6
Me / µ2 ,

1
0 0.1 0.2 0.3 0.4 0.5
p − inclusion volume fraction

Fig. 2. Normalized e*ective shear moduli vs. inclusion volume fraction. SC array of rigid spheres (component
1) embedded in a compressible matrix (component 2, ,2 = 0:4): · · ·, CM approximation for e =2 ; - -, CM
approximation for Me =2 ; ×, numerical data of Nunan and Keller for e =2 ; +, numerical data of Nunan
and Keller for Me =2 .

κ / κ =∞, µ / µ =∞, ν =0.4 (µ /κ =0.2143)

1 2 1 2 2 2 2
8

Base−centered cubic array of rigid spheres

7

CM−type M
6
CM−type µ
µe / µ2

Nunan & Keller M

5 Nunan & Keller µ
Me / µ2 ,

1
0 0.1 0.2 0.3 0.4 0.5 0.6
p − inclusion volume fraction

Fig. 3. Normalized e*ective shear moduli vs. inclusion volume fraction. BCC array of rigid spheres (com-
ponent 1) embedded in a compressible matrix (component 2, ,2 = 0:4): · · ·, CM approximation for e =2 ;
- -, CM approximation for Me =2 ; ×, numerical data of Nunan and Keller for e =2 ; +, numerical data of
Nunan and Keller for Me =2 .
 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183 2181

κ1/ κ2=∞, µ1 / µ2=∞, ν2=0.4 (µ2 /κ2=0.2143)

8

Face−centered cubic array of rigid spheres

7

CM−type M
6 CM−type µ
µe / µ2

Nunan & Keller M

Nunan & Keller µ
5
Me / µ2 ,

1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
p − inclusion volume fraction

Fig. 4. Normalized e*ective shear moduli vs. inclusion volume fraction. FCC array of rigid spheres (com-
ponent 1) embedded in a compressible matrix (component 2, ,2 = 0:4): · · ·, CM approximation for e =2 ;
- -, CM approximation for Me =2 ; ×, numerical data of Nunan and Keller for e =2 ; +, numerical data of
Nunan and Keller for Me =2 .

spheres (Bergman, 1979), and from the 2D elastostatic problem (Kantor and Bergman,
1982a) we know that many of the matrix elements Q of higher orders in p vanish
because (a) overlap integrals between some types of states vanish due to the symmetry
of the inclusions (e.g., due to the spherical symmetry of the inclusions, overlap inte-
grals between any two compression dipoles vanish in our 3D case); (b) many sums
of interactions vanish because of the lattice symmetry. That is why the correction to
the CM-type expressions actually starts with order p13=3 in the electrostatic case, and
with order p5 in the 2D elastostatic case of a square lattice of cylindrical inclusions.
Therefore, we expect Eqs. (3.6), (3.10) and (3.11) to be accurate up to higher orders
than O(p11=3 ), which can explain the good agreement with the numerical results even
when the volume fraction is not low.

5. Summary and conclusions

CM-types approximations for the elastic moduli of cubic arrays of spheres were
presented for the three types of cubic arrays: simple cubic, BCC and FCC arrays. These
approximations have the form of simple algebraic expressions for the bulk modulus e
and for the two shear moduli e and Me . Correction to the CM-type approximations
begin at order that is not less than p11=3 . Unlike e*ective medium or self-consistent
methods, the CM-type approximation takes into account the exact microgeometry. The
expressions for the shear moduli, Eqs. (3.10) and (3.11), are di*erent for the di*erent
2182 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183

types of cubic arrays. The expression for the bulk modulus, Eq. (3.6), is the same
for the three types of cubic arrays and coincides with one of the HS bounds. From
the analogous 3D electrostatic problem (Bergman, 1979), and from the 2D elastostatic
problem (Kantor and Bergman, 1982a) we expect Eqs. (3.6), (3.10) and (3.11) to
be accurate up to higher orders than O(p11=3 ). This expectation is supported by the
numerical results presented above. The simple algebraic expressions are in excellent
agreement with accurate numerical results at low volume fraction of the inclusions, and
continue to be good estimates of the elastic moduli at moderate volume fractions even
when the contrast is high, which makes the CM-type expressions simple but powerful
approximations.
Eqs. (3.10) and (3.11) predict correctly the behavior of the two shear moduli: (1)
the di*erence (Me −e ), which is a measure of the degree of anisotropy of the material,
is positive for SC arrays, while it is negative for BCC and FCC arrays; (2) deviations
from isotropy are more marked for SC than for BCC and FCC arrangements, the
absolute value of (Me − e ) being markedly larger for SC arrays than for BCC and
FCC arrays; (3) the shear moduli e and Me are almost the same for BCC and for
FCC arrays at low and moderate volume fractions.
It seems that the numerical method used by Iwakuma and Nemat-Nasser (1983) and
by Nemat-Nasser et al. (1982), the so-called “simplest approximation”, is the analogue
of the CM-type approximation. The CM-type approximations yield the same results
obtained by this numerical method.

References

Ashcroft, N.W., Mermin, N.D., 1976. Solid State Physics. Holt, New York.
Bergman, D.J., 1979. Dielectric constant of a two-component granular composite: a practical scheme for
calculating the pole spectrum. Phys. Rev. B 19, 2359–2368.
Cohen, I., Bergman, D.J., 2003a. Clausius–Mossotti-type approximation for elastic moduli of a cubic array
of spheres. Phys. Rev. B 68, 24104.
Cohen, I., Bergman, D.J., 2003b. E*ective elastic properties of periodic composite medium. J. Mech. Phys.
Solids 51, 1433–1457.
Hashin, Z., Shtrikman, S., 1963. A variational approach to the theory of the elastic behavior of multiphase
materials. J. Mech. Phys. Solids 11, 127–140.
Hill, R., 1963. Elastic properties of reinforced solids: some theoretical principles. J. Mech. Phys. Solids 11,
357–372.
Iwakuma, T., Nemat-Nasser, S., 1983. Composites with periodic microstructure. Comput. Struct. 16, 13–19.
Kantor, Y., Bergman, D.J., 1982a. Elastostatic resonances—a new approach to the calculation of the e*ective
elastic constants of composites. J. Mech. Phys. Solids 30, 355–376.
Kantor, Y., Bergman, D.J., 1982b. Clausius–Mossotti-type approximation for elastic moduli of a
three-dimensional, two-component composite. Appl. Phys. Lett. 41, 932–934.
Kantor, Y., Bergman, D.J., 1984. Improved rigorous bounds on the e*ective elastic moduli of a composite
material. J. Mech. Phys. Solids 32, 41–62.
Kushch, V.I., 1987. Computation of the e*ective elastic moduli of a granular composite material of regular
structure. Sov. Appl. Mech. 23, 362–365.
Kushch, V.I., 1997. Microstresses and e*ective elastic moduli of a solid reinforced by periodically distributed
spheroidal particles. Int. J. Solids Struct. 34, 1353–1366.
Kushch, V.I., 1998. Interacting cracks and inclusions in a solid by multipole expansion method. Int. J. Solids
Struct. 35, 1751–1762.
 I. Cohen / J. Mech. Phys. Solids 52 (2004) 2167 – 2183 2183

McPhedran, R.C., Milton, G.W., 1981. Bounds and exact theories for the transport properties of
inhomogeneous media. Appl. Phys. A: Solids Surf. 26, 207–220.
Nemat-Nasser, S., Iwakuma, T., Hejazi, M., 1982. On composites with periodic structure. Mech. Mater. 1,
239–267.
Nemat-Nasser, S., Taya, M., 1981. On e*ective moduli of an elastic body containing periodically distributed
voids. Q. Appl. Math. 39, 43–59.
Nunan, K.C., Keller, J.B., 1984. E*ective elasticity tensor of a periodic composite. J. Mech. Phys. Solids
32, 259–280.
Sangani, A.S., Lu, W., 1987. Elastic coeAcients of a composite containing spherical inclusions in a periodic
array. J. Mech. Phys. Solids 35, 1–21.
Torquato, S., 1997. E*ective sti*ness tensor of composite media: I. Exact series expansions. J. Mech. Phys.
Solids 45, 1421–1448.
Torquato, S., 1998. E*ective sti*ness tensor of composite media: II. Applications to isotropic dispersions. J.
Mech. Phys. Solids 46, 1411–1440.