
periodic inclusion lattice. By considering a standard cubic
periodic cell, then the existence of three planes of sym-
metry guarantees the material orthotropy as long as no
additional degree of anisotropy is induced by the material
symmetry of the individual phases in the composite. This
is true when the individual phases are geometrically ar-
ranged in the host in order to guarantee the existence of
such planes of symmetries, and they are individually at
most orthotropic. When assuming, in addition, that the
individual phases are either all isotropic or at most cubic,
and that the resulting geometry is invariant with respect to
rotation of the three orthogonal axis (this can be achieved
for example by considering either a cubic or a spherical
inclusion in three dimensions), the resulting material re-
sponse is then in general cubic. This is also shown in the
works [23] and [24], where the authors define a suitable
function which measures the deviation from isotropy for
composites in the context of asymptotic homogenisation.
Common lore suggests that, with a periodic inclusion
lattice, one cannot achieve isotropy without additional con-
straining strategies [12, 13], even with isotropic compo-
nents. Should this be the case, then it would be somewhat
unpleasant because, on the one hand, periodic arrange-
ments are by far the most convenient approach in both
additive manufacturing and computational material design
and, on the other hand, isotropic elasticity is very often
assumed in practical applications, especially in the context
of simple models used to validate experimental results.
In this work, we show that the maximal symmetry that
can be achieved for three-dimensional periodic composites
is not the cubic one. We give a rigorous and yet sim-
ple proof of the fact that a periodic arrangement on a
Face-Centered Cubic (FCC) lattice of spherical inclusions
of an isotropic solid within an isotropic matrix gives rise
to a large-scale isotropic response. In doing so, we also
show that any rhomboidal computational cell that gen-
erates such a lattice can be used to successfully design
homogenized solids in which the material symmetry is not
affected by the periodicity of the construction, since the
latter would preserve even the largest possible symmetry
group. It is significant to observe that the geometric sym-
metry group of such rhomboidal cells is strictly smaller
than the symmetry group of the lattice they generate, but
the lattice and not the cell is the geometrically relevant
structure when analyzing large-scale properties.
Incidentally, our result shows that the small lack of
isotropy computationally found for several periodic arrange-
ments based on a FCC lattice is to be ascribed to (un-
avoidable) numerical approximations rather than to real
geometric obstructions. This leads to important changes
in perspective for the interpretation of numerical results
and towards the design of isotropically elastic metamate-
rials, with important consequences on several applications.
We frame our discussion in the context of linear elas-
ticity by introducing, in Section 2, a normalized Voigt
representation of the elasticity tensor which is very con-
venient for the identification of material parameters and
symmetries. We then discuss the link between lattice sym-
metries and material symmetries for periodic composites
in Section 3 and, finally, present our main result and a
symmetry-preserving rhomboidal cell in Section 4.
2. Normalized Voigt representation
We are interested in describing the effective linear elas-
tic response of a composite that consists of two isotropi-
cally elastic phases. One is the matrix and the other one
occupies spherical inclusions with centers distributed on
a periodic lattice. Due to the spherical shape of the in-
clusions and the isotropic nature of the two materials, the
only source of anisotropy in the homogenized material re-
sponse can be the geometry of the inclusion lattice. It
is thus convenient to represent the linearized measure of
strain and the Cauchy stress tensor on a basis for the space
of symmetric tensors that is adapted to the geometry of
the inclusion lattice. This leads to the construction of a
normalized Voigt representation.
We denote by (a1,a2,a3)the generators of the lattice,
namely linearly independent vectors such that the centers
of the spherical inclusions are obtained as combinations of
a1,a2, and a3with integer coefficients. The set of lat-
tice sites is then denoted by L=ha1,a2,a3iZ. A set of
directors of the lattice can be constructed building an or-
thonormal basis for R3out of the generators. For instance,
we may choose
l1=a1
ka1k,kl2kl2=a2−(a2·l1)l1,
kl3kl3=a3−(a3·l1)l1−(a3·l2)l2.
We now introduce an orthogonal basis for the linear
space of symmetric tensors built upon the lattice directors.
The basis Z= (Z1,Z2,Z3,Z4,Z5,Z6)is given in terms of
dyadic products by
Z1=l1⊗l1,Z2=l2⊗l2,Z3=l3⊗l3,
Z4=l2⊗l3+l3⊗l2
√2,Z5=l1⊗l3+l3⊗l1
√2,
Z6=l1⊗l2+l2⊗l1
√2.
Such a basis is orthonormal with respect to the tensor
scalar product defined by A:B:= tr(ATB).
As customary in linear elasticity, we decompose the
deformation gradient tensor as F=J+∇u, with uthe
displacement field and Jthe isometry that maps spatial
vectors to material ones. The standard linearized strain
measure is then E=1
2(∇uJ+JT∇uT). In the infinitesimal-
displacement regime considered in linear elasticity, Jis
constant and homogeneous and can be taken as the iden-
tity. Eis a symmetric tensor, characterized by six degrees
of freedom. A possible choice of objective quantities to
represent them are the eigenvalues of Eand the orienta-
tion of its eigenvectors with respect to the lattice directors.
2