Journal of Elasticity manuscript No.

(will be inserted by the editor)

On anisotropic polynomial relations of the elasticity tensor
N. Auray B. Kolev M. Petitot
Received: date / Accepted: date
Abstract In this paper, some new necessary conditions for an elasticity tensor to possess
spatial symmetries are provided. Our motivation is related to the problem of the identica-
tion of the symmetry class of an elasticity tensor. Contrary to classical approaches, which
are based on spectral properties of the linear operator describing the elasticity, our setting
is based on polynomial invariants and covariants of the elasticity tensor C. We provide a set
of algebraic relations in H
4

the higher irreducible component in the decomposition of Ela,

to characterize the orthotropic ([D
2
]), trigonal ([D
3
]), tetragonal ([D
4
]), transverse isotropic
([SO(2)]) and cubic ([O]) symmetry classes. A bifurcation diagram is constructed to make
explicit how to travel in H
4
from a given isotropy class to another. Finally, we study the
link between these polynomial invariants and those obtained as the coecients of the char-
acteristic or the Betten polynomials. We show, in particular, that the Betten invariants do
not separate orbits of the elasticity tensors.
Keywords Symmetry classes Invariants Elasticity
Introduction
Physical motivation
In the theory of linear elasticity, the stress tensor and the strain tensor are related, at a xed
temperature, by Hookes law

ij
= C
ijkl

kl
.
If the medium under study is homogeneous, its elastic behavior is fully characterized by a 4th-order
elasticity tensor C. The innitesimal strain tensor is dened as the symmetric displacement gradient

ij
=
1
2
(u
i,j
+ u
j,i
)
and, assuming that the material is not subjected to volumic couple, the associated stress tensor is the
classical symmetric Cauchys one. Hence, both and belong to the 6-D vector space S
2
(R
3
), where
S
2
denotes the symmetric tensor product. As a consequence, the elasticity tensor is endowed with minor
symmetries:
C
ijkl
= C
jikl
= C
ijlk
.
N. Auray
LMSME, Universite Paris-Est, Laboratoire Modelisation et Simulation Multi Echelle,MSME UMR 8208 CNRS, 5 bd
Descartes, 77454 Marne-la-Vallee, France
E-mail: Nicolas.auray@univ-mlv.fr
B. Kolev
LATP, CNRS & Universite de Provence, 39 Rue F. Joliot-Curie, 13453 Marseille Cedex 13, France
E-mail: kolev@cmi.univ-mrs.fr
M. Petitot
LIFL, Universite des Sciences et Technologies de Lille I, 59655 Villeneuve dAscq Cedex, France
E-mail: Michel.Petitot@li.fr
2 N. Auray et al.
In the case of hyperelastic materials, the stress-strain relation is furthermore assumed to derive from
an elastic potential. Therefore the elasticity tensor is the second-derivative of the potential energy with
respect to strain tensor. Thus, as a consequence of Schwarzs theorem, the hyperelasticity tensor possesses
the major symmetry:
C
ijkl
= C
klij
.
The space of (hyper) elasticity tensors is therefore the 21-D vector space Ela = S
2
S
2
(R
3
). To each
hyperelastic material corresponds an elasticity tensor C but this association is not unique, there is a
gauge group. Indeed, a designation based on the components of C in a xed reference frame is relative
to the choice of a xed orientation of the material. If this problem is invisible for isotropic materials, it
becomes more prominent as the anisotropy increases. For a totally anisotropic (triclinic) material how can
it be decided whether two sets of components represent the same material? To address this problem some
approaches have been proposed in the literature so far. Roughly speaking, those propositions are based
on the computation of the elasticity tensor spectral decomposition. But despite of being theoretically
well-founded, the symmetry class identication relies on the evaluation of the roots multiplicity of the
characteristic polynomial. Such a problem is known to be highly ill-posed [42]. Therefore this approach
is dicult to handle in practice especially when working with noise corrupted data.
In order to avoid the computation of the spectral decomposition, an alternative approach based on the
invariants (and covariants) of the elasticity tensor can be considered. Working toward this direction,
in the present paper some new necessary polynomial conditions characterizing symmetries
are provided. The approach used in the present work follows of a long series of papers [29, 30, 32, 31,
33, 9]. It is interesting to notice that the general framework introduced here is well-known by the high
energy physics community [1, 2, 38], and has stimulated mathematical researches on the subject [35, 39,
36, 37]. However, these methods do not seem to have yet been applied in elasticity. Before introducing
our results, we wish to draw a short picture of this approach.
Invariant-based identication in a nutshell
As the material is rotated by an element g = (g
j
i
) SO(3), the elasticity tensor C moves under the action
of the 3D rotation group SO(3) on the space of elasticity tensors Ela
C
ijkl
g
i
p
g
j
q
g
k
r
g
l
s
C
pqrs
.
Hence, from the point of view of linear elasticity, the classication of elastic materials can be assimilated to
the description of the orbits of this SO(3)-action on Ela. Practically, this is a dicult problem because the
orbit space is not a smooth manifold in general. Usually, the orbit space of the linear representation of a
compact Lie group is a singular space, which can be described by the means of the isotropy stratication [2]
(described in Appendix A). In the eld of elasticity, the rst rigorous classication of the dierent isotropy
classes (or symmetry classes) was established by Forte and Vianello [18],
Since the rotation group SO(3) is compact, the algebra of invariant polynomial functions on Ela is
nitely generated [41] and separates the orbits [2, Appendix C]. Therefore, it is possible to nd a nite
set of polynomial invariants J
1
, . . . , J
N
which denes a polynomial function J : Ela R
N
such that
J(C) = J(C

) i C and C

are in the same orbit.

This way, an SO(3)-orbit can be identied with a point in J(Ela) R
N
. It results from the Tarski-
Seidenberg theorem [12], that the image of a R-vector space under a polynomial map is a semi-algebraic
set, that is a subset of R
N
dened by a boolean combination of polynomial equations and inequalities
over R [12]. The orbit space can therefore be endowed by a semi-algebraic structure. The invariants
J
1
, . . . , J
N
are polynomials in 21 variables (dimEla = 21). The problem is therefore to construct
a complete basis of invariants. This problem is known to be hard, and no complete answer
have been provided yet [9].
In this paper a simpler version of the problem is considered. To that aim the space Ela
will be decomposed into SO(3)-irreducible components:
Ela := S
2
S
2
(R
3
) 2H
0
(R
3
) 2H
2
(R
3
) H
4
(R
3
)
where H
n
(R
3
) is the space of nth-order harmonic tensors on R
3
. Because of the lack of ambiguity,
this notation will be shortened H
n
. It is well known that dimH
n
= 2n + 1 for all n 0. Historically
this decomposition, also known as the harmonic decomposition, has already been used in the study of
On anisotropic polynomial relations of the elasticity tensor 3
anisotropic elasticity tensors [5, 25, 6, 18]. We therefore refer to these references for a deeper insight into
this topic.
It is known [18] that Ela is divided into eight conjugacy classes for the SO(3)-action, i.e.,
into eight strata,
[H]
, which group tensors having symmetry groups conjugate to H. The
same result holds for H
4
, therefore we will transfer the study from Ela to H
4
and provide
necessary and sucient conditions for a tensor D H
4
to possess symmetries. Pulled back
to Ela, these results give necessary invariant conditions for an elasticity tensor to be in a
given symmetry class.
To be more specic, for each isotropy class [H], we build a linear slice Ela
H
and we show that
the equivalence relation D D

restricted on
[H]
reduces to the linear action of a certain monodromy
group
H
on Ela
H
. The cardinal of
H
is equal to the number of intersection points of a generic orbit in

[H]
with the linear slice Ela
H
. The closure of each stratum
[H]
, which we denote by
[H]
and which
corresponds to the set of tensors which have at least isotropy [H], is a semialgebraic set which we describe
explicitly by a nite number of equations and inequalities on the invariant polynomials J
1
, . . . , J
N
. For
each class [H] for which
H
is nite, we give explicit relations between invariants to check whether D
belongs to this class. This procedure which is applied throughout the paper is detailed more
in depth in Appendix A.
Organization of the paper and main results
The paper is divided in two sections, in the rst one esults concerning both the stratication of Ela
and H
4
are detailed. The principal new results of this paper are gathered in the next section, where we
provide algebraic relations to identify symmetry classes in H
4
with nite monodromy group: orthotropic
([D
2
]), trigonal ([D
3
]), tetragonal ([D
4
]), transverse isotropic ([SO(2)]) and cubic ([O]). For each of these
classes, we provide a parametrization of the corresponding strata by rational expressions involving up to
six polynomial invariants. Bifurcation relations between related classes are also computed and summed-
up in a diagram (c.f. Figure 2 in subsection 2.3). Finally, in subsection 2.4, we prove, as a collateral
observation, that the invariants dened by the coecients of the Betten polynomial [7] do not separate
the orbits of Ela. This paper is concluded in section 3 and the extension of the method to a broader
class of situations is discussed. In Appendix A, the general geometric framework behind our results is
introduced. Although materials introduced in this section are not new, and are probably well-known by
most mathematicians and the high-energy physics community, it does not seem to have been yet exploited
by the mechanical community. We clarify the geometry behind the so called normal forms of elasticity
tensors (linear slices) and their ambiguity (monodromy groups). This permits us to justify, on a rigorous
mathematical basis, the results given in Table 1 and in Table 2. Calculations of the monodromy groups,
are done in Appendix C;
Notations
To conclude this introduction let us specify some typographical conventions that will be used throughout
this paper. Scalars will be noted by Greek letters; minuscule letters indicate elements of H
2
, i.e., deviators;
bold minuscule letters stand for elements of R
3
R
3
, i.e., second order tensors; elements of H
4
will be
indicated by capital letters, while elasticity tensors (elements of S
2
S
2
(R
3
)) will be noted by bold capital
letters. Some minor exceptions to these rules may occur (such as for the stress and strain tensors)
and will therefore be indicated in the text.
1 The space of elasticity tensors
1.1 The harmonic decomposition of elasticity tensors
To study a particular representation of an SO(3)-representation, such as the space of elasticity tensors,
the rst step is to decompose it into a direct sum of elementary pieces. The SO(3)-irreducible pieces are
symmetric, traceless tensors (i.e., harmonic tensors) and the decomposition is known as the harmonic
decomposition. The space of elasticity tensors admits the following decomposition [5, 6, 18, 20]:
Ela 2H
0
2H
2
H
4
. (1)
4 N. Auray et al.
Therefore, each C Ela can be written as C = (, , a, b, D) where , H
0
, a, b H
2
and D H
4
and
such that for all g SO(3)

C = (g) C

= , = , a =
2
(g) a,

b =
2
(g) b,

D =
4
(g) D.
Elements of the set (, , a, b, D) are covariants (c.f. subsection B.2) to C : (, ) are invariants, (a, b)
are
2
-covariants, and D is
4
-covariant. The explicit harmonic decomposition is well-known, given an
orthonormal frame (e
1
, e
2
, e
3
) of the Euclidean space, we get [9]:
C
ijkl
=
ij

kl
+ (
ik

jl
+
il

jk
)
+
ij
a
kl
+
kl
a
ij
+
ik
b
jl
+
jl
b
ik
+
il
b
jk
+
jk
b
il
+ D
ijkl
,
(2)
where the metric is written q
ij
:=
ij
. This formula can be inverted to obtain the 5 harmonic tensors
(, , a, b, D) from C. On Ela = S
2
S
2
(R
3
), there are only two dierent traces
1
:
d
ij
= (tr
12
C)
ij
:=
3

k=1
C
kkij
, v
ij
= (tr
13
C)
ij
:=
3

k=1
C
kikj
.
where d and v are known as, respectively, the dilatation tensor and the Voigt tensor [14, 13]. Starting
with (2) we get
d = (3 + 2)q + 3a + 4b, v = ( + 4)q + 2a + 5b. (3)
Taking the traces of each equation, one obtains:
tr(d) = 9 + 6, tr(v) = 3 + 12,
and, nally:
=
1
15
(2 tr(d) tr(v)), =
1
30
(tr(d) + 3 tr(v)),
a =
1
7
(5 dev(d) 4 dev(v)), b =
1
7
(2 dev(d) + 3 dev(v)),
where we write dev(a) := a
1
3
tr(a) q the deviatoric part (i.e traceless part) of a 2nd-order tensor.
1.2 The isotropy stratication of Ela
The isotropy classes of elasticity tensors have been computed in [18, 8]. There are exactly eight classes:
the isotropic class [SO(3)], the cubic class [O], the transversely isotropic class [O(2)], the trigonal class
[D
3
], the tetragonal class [D
4
], the orthotropic class [D
2
], the monoclinic class [Z
2
] and the the triclinic
class [1].The precise denitions of these groups will soon be detailed. This stratication is a consequence
of the index symmetries of elasticity tensors. It is not a general result that encompass all kind of 4th-order
tensors. For example, the photo-elasticity tensors space, which is a 4th-order tensor space, solely endowed
with minor symmetries is divided into twelve symmetry classes [19].
Using the framework discussed in Appendix A, we obtain the geometric characteristics
2
indicated in
Table 1, where the following objects appear.
V
H
the xed-point set of H, which is the subspace of tensors having at least H as isotropy group. Its
dimension equals the dimension of the matrix space used to represent them. The space V
H
can be
constructed for any closed subgroup H, not only for an isotropy subgroup. In elasticity, for instance,
xed-point sets have been constructed for Z
3
, Z
4
(which are not isotropy subgroups). This is why it
was believed for a long time that there exists 10 type of elasticity.
1
The notation tr
ij
indicates that the contraction should be done on the ith and jth indices.
2
The dimensions of
[H]
and
[H]
/G are obtained using formula (25) of subsection A.3, dimV
H
is computed using the
trace formula (26) and the explicit formulas provided [4].
On anisotropic polynomial relations of the elasticity tensor 5

[H]
the [H]-stratum, which is the set of tensors which symmetry group is conjugate to H. It corresponds
to a symmetry class according to Forte and Vianello [18]. We got the following dimension relation
dim
[H]
= dimV
H
+ e
H
where e
H
= dimSO(3)/N(H) is the number of Euler angles
3
needed to
dene the appropriate coordinate system in which the elastic tensor has the minimum number of
elastic constants [28]. The strata dimension corresponds to what Norris, for example, calls the number
of independent parameters characterizing a tensor.

[H]
/G the orbit space of
[H]
. Each point of this space represents the G-orbit of a tensor whose symmetry
group is conjugate to H. We have the dimensional relation
dimV
H
= dim
[H]
/G + dim
H
,
where
H
= N(H)/H is the monodromy group. In particular, when
H
is discrete,
[H]
/G and V
H
have the same dimension, this is the situation for most of elasticity symmetry classes. At the opposite,
when
H
is continuous, the dimension of the orbit space
[H]
/G is strictly smaller than the one of
the xed-point set V
H
. In elasticity, this case occurs only for classes [Z
2
] and [1].
All these dimensions are linked by the formula
dim
[H]
= dim
[H]
/G + dim
H
+ e
H
.
System : H N(H)
H
card
H
dimV
H
dim
[H]
/G dim
[H]
Triclinic 1 SO(3) SO(3) 21 18 21
Monoclinic Z
2
O(2) O(2) 13 12 15
Orthotropic D
2
O S
3
6 9 9 12
Trigonal D
3
D
6
S
2
2 6 6 9
Tetragonal D
4
D
8
S
2
2 6 6 9
Transverse Isotropic O(2) O(2) 1 1 5 5 7
Cubic O O 1 1 3 3 6
Isotropic SO(3) SO(3) 1 1 2 2 2
Table 1 Isotropy classes for Ela.
The geometric characterization of Ela and its stratication is summed-up Table 1 where
the following closed subgroups of SO(3) appear:
O(2) is the subgroup generated by all the rotations around the z-axis and the order 2
rotation : (x, y, z) (x, y, z) around the x-axis.
SO(2) is the subgroup of all the rotations around the z-axis.
Z
n
is the unique cyclic subgroup of order n of SO(2), the subgroup of rotations around
the z-axis.
D
n
is the dihedral group. It is generated by Z
n
and : (x, y, z) (x, y, z).
T is the tetrahedral group, the (orientation-preserving) symmetry group of a tetrahedron.
It has order 12.
O is the octahedral group, the (orientation-preserving) symmetry group of a cube or
octahedron. It has order 24.
I is the icosahedral group, the (orientation-preserving) symmetry group of a icosahedra
or dodecahedron. It has order 60.
1 is the trivial subgroup, containing only the unit element.
completed by S
n
the symmetric group, i.e, the group of permutations acting on n elements.
The results appearing in Table 1 are well-known in elasticity, but their constructions are
usually ad-hoc and do not stand on rigorous mathematical foundations. Therefore, the aim
of Appendix A is to provide such a rm geometric construction.
3
e
H
= dimG/N(H) is the set of distinct subgroups of G which are conjugate to H; that is [H], the conjugacy class of
H.
6 N. Auray et al.
1.3 The isotropy stratication of H
4
Let consider now the isotropy stratication of H
4
which is the higher order component of the harmonic
decomposition of Ela. The isotropy classes of the 9-D space H
4
are the same as the 8 classes of Ela. The
corresponding poset (partially ordered set) is illustrated in Figure 1. With the convention that a group
at the starting point of an arrow is conjugate to a subgroup of the group pointed by the arrow. The main
characteristics of each symmetry classes are reported in the Table 2.
Fig. 1 The poset of isotropy classes for H
4
.
System : H N(H)
H
card
H
dimV
H
dim
[H]
/G dim
[H]
Triclinic 1 SO(3) SO(3) 9 6 9
Monoclinic Z
2
O(2) O(2) 5 4 7
Orthotropic D
2
O S
3
6 3 3 6
Trigonal D
3
D
6
S
2
2 2 2 5
Tetragonal D
4
D
8
S
2
2 2 2 5
Transverse Isotropic O(2) O(2) 1 1 1 1 3
Cubic O O 1 1 1 1 4
Isotropic SO(3) SO(3) 1 1 0 0 0
Table 2 Isotropy classes for the tensorial representation of SO(3) on H
4
.
Contrary to Ela, the integrity basis for H
4
is known, as a consequence a characterization
of its symmetry classes in terms of polynomial relations between invariants is possible. This
is the goal of the next section in which attention will be focused on classes having nite
monodromy group
H
, i.e, the cubic class [O], the transversely isotropic class [O(2)], the
trigonal class [D
3
], the tetragonal class [D
4
] and the orthotropic class [D
2
]. For each of them,
explicit polynomial relations between the elementary invariant of H
4
will be given. These
set of invariants are generated in the following way:
Proposition 1.1. Let D H
4
. The 2nd-order tensors d
2
, . . . , d
10
:
d
2
= tr
13
(D
2
) d
3
= tr
13
(D
3
) d
4
= d
2
2
d
5
= d
2
Dd
2
d
6
= d
3
2
d
7
= d
2
2
Dd
2
d
8
= d
2
2
D
2
d
2
d
9
= d
2
2
Dd
2
2
d
10
= d
2
2
D
2
d
2
2
(4)
are covariant to D. The nine fundamental invariants are dened by:
J
k
:= tr(d
k
), k = 2, . . . , 10.
On anisotropic polynomial relations of the elasticity tensor 7
The rst six invariants J
2
, . . . , J
7
are algebraically independent. The last three ones J
8
, J
9
, J
10
are
linked to the formers by polynomial relations. These fundamental syzygies were computed in [40].
Remark 1.2. Notice that the rst invariant J
2
(D) = D, D is the squared norm of D, it corresponds to
the squared Frobenius norm of D. In particular, J
2
(D) = 0 if and only if D = 0.
For technical details and historical considerations concerning these fundamental invariants, we refer
to the original publication of by Boelher and al.[9] and the references therein.
2 Invariant characterization of H
4
symmetry classes
The aim of this section is to describe the strata in terms of polynomials relations between
(J
1
, . . . , J
10
). The algebraic procedure used to obtain these relations is detailed in subsec-
tion A.5. In the rst subsection the space H
4
will be parametrized. The next subsection is
devoted to identify the polynomial relations for each class of nite monodromy. In the third
subsection, a bifurcation diagram is constructed to make explicit how we travel from a
given isotropy class to another. And the last subsection some remarks are made on the
characteristic polynomial of elements of H
4
.
2.1 Global parametrization
On an euclidean vector space, any bilinear form can be represented by a symmetric linear operator. This
applies in particular to an elasticity tensor C which may be considered as a quadratic form on S
2
(R
3
),
where the Euclidean structure is induced by the standard inner product on R
3
. It is therefore customary
to represent an elasticity tensor C by its corresponding symmetric linear operator C on S
2
(R
3
) (relatively
to the canonical scalar product a, b = tr(ab)). The corresponding matrix representation [15] is given by
C =

c
11
c
12
c
13

2c
14

2c
15

2c
16
c
12
c
22
c
23

2c
24

2c
25

2c
26
c
13
c
23
c
33

2c
34

2c
35

2c
36

2c
14

2c
24

2c
34
2 c
44
2 c
45
2 c
46

2c
15

2c
25

2c
35
2 c
45
2 c
55
2 c
56

2c
16

2c
26

2c
36
2 c
46
2 c
56
2 c
66

where c
mn
are the components of the elasticity tensor in an orthonormal frame. We use the conventional
rule to recode a pair of indices (i, j) (i, j = 1, 2, 3) by a integer m = (i, j) (m = 1, 2, . . . 6), where
(i, i) := i for 1 i 3 and (i, j) := 9 (i + j) for i ,= j.
Given a general tensor D H
4
, we denote by D the corresponding symmetric linear operator. In the
canonical basis of R
3
, D has the following matrix representation, depending on nine real parameters:
D =

D
11

2 D
12

2 D
T
12
2 D
22

where
T
represents the transposition,
D
11
=

h
9
h
8
h
9
h
8
h
9
h
9
h
7
h
7
h
8
h
7
h
8
h
7

D
12
=

h
5
h
6
h
2
h
1
h
5
h
2
h
4
h
3
h
6
h
4
h
1
h
3

,
and
D
22
=

h
7
h
1
h
3
h
2
h
4
h
1
h
3
h
8
h
5
h
6
h
2
h
4
h
5
h
6
h
9

Using this parametrization, D has the following invariance properties:

1. Z
2
-invariant i h
2
= h
4
= h
5
= h
6
= 0 ,
2. D
2
-invariant i h
k
= 0 pour 1 k 6 ,
3. D
3
-invariant i h
1
= h
2
= h
3
= h
4
= h
6
= 0 et h
7
= h
8
= 4h
9
,
4. D
4
-invariant i h
k
= 0 pour 1 k 6 et h
7
= h
8
,
5. O(2)-invariant i h
k
= 0 pour 1 k 6 et h
7
= h
8
= 4h
9
,
6. O-invariant i h
k
= 0 pour 1 k 6 et h
7
= h
8
= h
9
,
7. SO(3)-invariant i h
k
= 0 pour 1 k 9 .
8 N. Auray et al.
2.2 Syzygies relations
2.2.1 Cubic symmetry ([O])
An O-invariant tensor D H
4
has the following matrix representation:
D =

8 4 4 0 0 0
4 8 4 0 0 0
4 4 8 0 0 0
0 0 0 8 0 0
0 0 0 0 8 0
0 0 0 0 0 8

(5)
where R. As indicated in the table 1, the monodromy group for the cubic system is
1, therefore linear slice corresponding to the octahedral group is of degree 1. In other
terms, there is no ambiguity in dening the normal form Equation 5. The evaluation of the
invariants (4) on this slice gives the following parametric system:
J
2
= 480
2
, J
3
= 1920
3
, J
4
= 76800
4
,
J
5
= 0, J
6
= 12288000
6
, J
7
= 0,
J
8
= 0, J
9
= 0, J
10
= 0.
We notice that the parameter 4 = J
3
/J
2
is a rational invariant.
Proposition 2.1. A harmonic tensor D H
4
is in the closed stratum
[O]
if and only if the invariants
J
2
(D) J
10
(D) satisfy the following polynomial relations
3 J
4
= J
2
2
, J
5
= 0, 30 J
3
2
= J
2
3
, 9 J
6
= J
2
3
,
J
7
= 0, J
8
= 0, J
9
= 0, J
10
= 0.
(6)
D is in the cubic class [O] if moreover J
2
(D) ,= 0. In that case, it admits the normal form (5) where
4 := J
3
(D)/J
2
(D).
We observe that J
2
, J
3
R, R. Therefore we do not need to add any inequality to the syzygy
system (6) to ensure that is real. This fact is general to any linear slice of degree 1.
Remark 2.2. All 2nd-order tensors in H
2
covariant to D vanish.
As a consequence, for the cubic case, another characterization is possible:
Corollary 2.3. A harmonic tensor D H
4
is in the closed stratum
[O]
if and only if there exists
c
2
, c
3
R

such that
d
2
= c
2
q, d
3
= c
3
q and 10 c
2
3
c
3
2
= 0. (7)
Proof. Since d
k
are covariant to an O-invariant tensor, they are multiple of the identity, i.e., d
k
= c
k
q,
for k = 2, . . . , 10. Thus J
k
= tr(d
k
) = 3 c
k
and the set of syzygies (6) can be expressed in terms of the
c
k
. The announced relations follow, using the additional fact that Dq = 0 and the denitions of the d
k
.
Conversely, suppose that these relations are veried. The fundamental D covariants, provided in
proposition (1.1), are either d
3
, powers of d
2
, or involved products of D and d
2
. If d
2
= c
2
q, powers
of d
2
are scalar multiples of the metric q, and because D is harmonic all terms involving Dd
2
vanish.
Therefore J
5
, J
7
, J
8
, J
9
, J
10
vanish and the relations between J
2
, J
4
and J
6
are automatically satised.
Because tr q = 3 and 10c
2
3
c
3
2
= 0, we have moreover 10J
2
3
J
3
2
= 0. Therefore the fundamental syzygies
of the cubic class are satised.
2.2.2 Transversely isotropic symmetry ([O(2)])
An O(2)-invariant tensor D H
4
has the following matrix representation:
D =

3 4 0 0 0
3 4 0 0 0
4 4 8 0 0 0
0 0 0 8 0 0
0 0 0 0 8 0
0 0 0 0 0 2

(8)
On anisotropic polynomial relations of the elasticity tensor 9
where R. As in the previous case, the monodromy group for the transversely isotropic
symmetry is 1, and the linear slice is of degree 1. The evaluation of the invariants (4) on this slice
gives:
J
2
= 280
2
, J
3
= 720
3
, J
4
= 32800
4
,
J
5
= 80000
5
, J
6
= 4528000
6
, J
7
= 17600000
7
,
J
8
= 211200000
8
, J
9
= 3872000000
9
, J
10
= 46464000000
10
The parameter = 7 J
3
/18 J
2
is a rational invariant. The linear slice corresponding to the group O(2) is
of degree 1, which was already known (see Appendix C).
Proposition 2.4. A harmonic tensor D H
4
is in the closed stratum
[O(2)]
if and only if the invariants
J
2
(D) J
10
(D) satisfy the following polynomial relations
98 J
4
= 41 J
2
2
, 63 J
5
= 25 J
3
J
2
, 3430 J
3
2
= 81 J
2
3
, 1372 J
6
= 283 J
2
3
,
882 J
7
= 275 J
2
2
J
3
, 4802 J
8
= 165 J
2
4
, 12348 J
9
= 3025 J
2
3
J
3
, 67228 J
10
= 1815 J
2
5
.
(9)
D is in the transversely isotropic class [O(2)] if moreover J
2
(D) ,= 0. In that case, it admits the normal
form (8) where = 7 J
3
(D)/18 J
2
(D).
As for [O] the expression of ensures that, for all J
2
, J
3
R, the solution is real.
Remark 2.5. All 2nd-order tensors in H
2
covariant to D are multiple of diag(1, 1, 2) and are eigenvectors
of D corresponding to the eigenvalue 12.
2.2.3 Trigonal symmetry ([D
3
])
A D
3
-invariant tensor D H
4
has the following matrix representation:
D =

3 4

2 0 0
3 4

2 0 0
4 4 8 0 0 0

2 0 8 0 0
0 0 0 0 8 2
0 0 0 0 2 2

(10)
where (, ) R
2
. As indicated in the table 1 for this symmetry class is no more 1 but S
2
,
and the linear slice is of degree 2. Its action is given by:
, .
Therefore, we will have to make a choice between and to dene our normal form. The
evaluation of the invariants (4) on this slice gives:
J
2
= 280
2
+ 16
2
J
3
= 144

5
2

J
4
= 32800
4
+ 2720
2

2
+ 88
4
J
5
= 32

2
+ 50
2

2
J
6
= 4528000
6
+ 436800
2

4
+ 20640
4

2
+ 496
6
J
7
= 320

22
2
+
2

2
+ 50
2

2
J
8
= 3840
2

22
2
+
2

2
+ 50
2

2
J
9
= 3200

2
+ 50
2

22
2
+
2

2
J
10
= 38400
2

2
+ 50
2

22
2
+
2

2
We notice that the parameter:
=
1
4
J
5
J
2
2
3 J
4
10 N. Auray et al.
is a rational invariant, therefore for all J
2
, J
4
, J
5
R, the parameter is real (c.f. subsection A.5).
However, the minimal equation satised by is of degree 2
J
2
= 280
2
+ 16
2
. (11)
A condition is needed for to be real. This condition, which is
2
0, is given by the following
inequality on J
2
, J
4
, J
5
:
2J
2
(J
2
2
3 J
4
)
2
35J
2
5
0 (12)
Having computed , the equation (11) has two roots with opposite signs.
Proposition 2.6. A harmonic tensor D H
4
is in the closed stratum
[D
3
]
if and only if the invariants
J
2
(D) J
10
(D) satisfy the following polynomial relations
192 J
6
= 51 J
2
3
+ 216 J
2
J
4
+ 10 J
3
2
36 J
7
= 2 J
2
2
J
3
+ 6 J
3
J
4
+ 27 J
2
J
5
768 J
4
2
= 99 J
2
4
+ 552 J
2
2
J
4
+ 10 J
2
J
3
2
+ 240 J
3
J
5
240 J
8
= 33 J
2
4
+ 96 J
2
2
J
4
+ 30 J
2
J
3
2
+ 40 J
3
J
5
576 J
4
J
5
= 41 J
2
3
J
3
+ 120 J
2
J
3
J
4
+ 216 J
2
2
J
5
+ 30 J
3
3
1152 J
9
= 99 J
2
3
J
3
+ 296 J
2
J
3
J
4
+ 648 J
2
2
J
5
+ 10 J
3
3
1440 J
5
2
= 11 J
2
5
+ 32 J
2
3
J
4
70 J
2
2
J
3
2
+ 240 J
2
J
3
J
5
+ 240 J
3
2
J
4
8640 J
10
= 891 J
2
5
+ 2592 J
2
3
J
4
+ 730 J
2
2
J
3
2
+ 2160 J
2
J
3
J
5
+ 240 J
3
2
J
4
(13)
together with inequality (12). D is in the trigonal class [D
3
] if moreover 3 J
4
J
2
2
,= 0 and 98 J
4
41 J
2
2
,= 0.
In that case, it admits the normal form (10) where
=
1
4
J
5
J
2
2
3 J
4
and where is the positive root of J
2
= 280
2
+ 16
2
.
Remark 2.7. All 2nd-order tensors in H
2
covariant to D are multiple of diag(1, 1, 2) and are eigenvectors
of D corresponding to the eigenvalue 12.
2.2.4 Tetragonal symmetry ([D
4
])
A D
4
-invariant tensor D H
4
has the following matrix representation:
D =

+ 3 + 4 0 0 0
+ + 3 4 0 0 0
4 4 8 0 0 0
0 0 0 8 0 0
0 0 0 0 8 0
0 0 0 0 0 2 + 2

(14)
where (, ) R
2
. As for the [D
3
] class, the monodromy group is S
2
, and the linear slice is of
degree 2. Therefore a choice has to be made between and to dene the normal form.
The evaluation of the invariants (4) on this slice gives:
J
2
= 8
2
+ 280
2
J
3
= 48

2
+ 15
2

J
4
= 32
4
+ 960
2

2
+ 32800
4
J
5
= 128 (5 )
2
(5 + )
2
J
6
= 128
6
+ 5760
4

2
+ 86400
2

4
+ 4528000
6
J
7
= 512

55
2
+
2

(5 )
2
(5 + )
2
J
8
= 6144
2

55
2
+
2

(5 )
2
(5 + )
2
J
9
= 2048 (5 )
2
(5 + )
2

55
2
+
2

2
J
10
= 24576
2
(5 )
2
(5 + )
2

55
2
+
2

2
On anisotropic polynomial relations of the elasticity tensor 11
We notice that the parameter
=
1
4
J
5
J
2
2
3 J
4
is a rational invariant, therefore for all (J
2
, J
4
, J
5
) R
3
, R. However, the minimal equation satised
by is of degree 2
J
2
= 8
2
+ 280
2
(15)
As for [D
3
], the condition
2J
2
(J
2
2
3 J
4
)
2
35J
2
5
0 (16)
has to be added in order to ensure to be real.
Having computed , the equation (15) has two roots with opposite signs. The linear slice corresponding
to the group D
4
is of degree 2, which was already known (see Appendix C).
Proposition 2.8. A harmonic tensor D H
4
is in the closed stratum
[D
4
]
if and only if the invariants
J
2
(D) J
10
(D) satisfy the following polynomial relations
6 J
6
= 3 J
2
3
+ 9 J
2
J
4
+ 20 J
3
2
3 J
7
= J
2
2
J
3
3 J
3
J
4
+ 3 J
2
J
5
6 J
4
2
= 3 J
2
4
+ 9 J
2
2
J
4
+ 20 J
2
J
3
2
20 J
3
J
5
5 J
8
= 3 J
2
4
+ 6 J
2
2
J
4
+ 30 J
2
J
3
2
5 J
3
J
5
3 J
4
J
5
= 7 J
2
3
J
3
15 J
2
J
3
J
4
+ 3 J
2
2
J
5
60 J
3
3
6 J
9
= 5 J
2
3
J
3
13 J
2
J
3
J
4
+ 6 J
2
2
J
5
20 J
3
3
5 J
5
2
= 2 J
2
5
+ 4 J
2
3
J
4
+ 10 J
2
2
J
3
2
20 J
2
J
3
J
5
+ 30 J
3
2
J
4
15 J
10
= 9 J
2
5
+ 18 J
2
3
J
4
+ 85 J
2
2
J
3
2
30 J
2
J
3
J
5
+ 15 J
3
2
J
4
(17)
together with inequality (16). D is in the tetragonal class [D
4
] if moreover 3 J
4
J
2
2
,= 0 and 5 J
3
2

8 J
2
J
4
70 J
2
3
,= 0. In that case, it admits the normal form (14) where
=
1
4
J
5
J
2
2
3 J
4
and where is the positive root of J
2
= 8
2
+ 280
2
.
Remark 2.9. All 2nd-order tensors in H
2
covariant to D are multiple of diag(1, 1, 2) and are eigenvectors
of D corresponding to the eigenvalue 12.
2.2.5 Orthotropic symmetry ([D
2
])
A D
2
-invariant tensor D H
4
has the following matrix representation:
D =

3

3

2
0 0 0

3

3

1

1
0 0 0

2

1

1

2
0 0 0
0 0 0 2
1
0 0
0 0 0 0 2
2
0
0 0 0 0 0 2
3

(18)
where (
1
,
2
,
3
) R
3
. The monodromy group is the group S
3
acting on (
1
,
2
,
3
) by permutation,
therefore the linear slice is of degree 6. This fact will make the following computation rather
cumbersome. As the evaluations of the invariants (4) on this slice are S
3
-invariant, they can therefore
be expressed as polynomial functions of the elementary symmetric polynomials (
1
,
2
,
3
) dened by:

1
:=
1
+
2
+
3
,
2
:=
1

2
+
1

3
+
2

3
,
3
:=
1

3
.
12 N. Auray et al.
More precisely, we have:
J
2
= 14
2
+ 8
2
1
J
3
= 6
1

2
+ 24
3
J
4
= 40
1

3
112
2
1

2
+ 68
2
2
+ 32
4
1
J
5
= 64
2
1

3
12
2

3
16
3
1

2
+ 28
1

2
2
J
6
= 344
3
2
+ 192
3
1

3
24
2
3
672
4
1

2
+ 1008
2
1

2
2
+ 128
6
1
504
1

3
J
7
= 432
2
1

3
+ 384
4
1

3
+ 104
2
2

3
96
1

2
3
64
5
1

2
+ 192
3
1

2
2
248
1

3
2
J
8
= 608
3
1

3
+ 80
4
2
768
5
1

3
+ 192
2
1

2
3
+ 72
2

2
3
+ 288
4
1

2
2
416
2
1

3
2
+ 744
1

2
2

3
J
9
= 5248
4
1

3
+ 2880
2
1

2
2

3
+ 1328
1

2
3
+ 144
3
3
+ 2304
6
1

3
1152
3
1

2
3
880
3
2

3
256
7
1

2
+ 1024
5
1

2
2
2304
3
1

3
2
+ 2160
1

4
2
J
10
= 10752
5
1

3
1280
3
1

2
2

3
5664
1

3
2

3
2688
2
1

2
3
800
5
2
4608
7
1

3
+ 2304
4
1

2
3
1344
2
2

2
3
288
1

3
3
+ 1536
6
1

2
2
4224
4
1

3
2
+ 3104
2
1

4
2
As it can be observed, the parametric system obtained for [D
2
] is rather complicated. Nevertheless, the
evaluation of J
2
, . . . , J
10
on the slice and their expressions using
1
,
2
,
3
is simple using a computer.
By computing the elimination ideal of a Groebner basis for the ideal generated by
J
k
p
k
(
1
,
2
,
3
), k = 2, . . . , 7,
we obtain a set of 6 syzygies.
1350 J
3
J
7
840 J
4
J
6
+ 465 J
2
2
J
6
+ 270 J
5
2
+ 720 J
2
J
3
J
5
+ 747 J
2
J
4
2
170 J
3
2
J
4
564 J
2
3
J
4
+ 70 J
2
2
J
3
2
+ 84 J
2
5
= 0 (19a)
1620 J
4
J
7
+ 810 J
2
2
J
7
+ 360 J
5
J
6
1110 J
2
J
3
J
6
+ 999 J
2
J
4
J
5
+ 960 J
3
2
J
5
549 J
2
3
J
5
972 J
3
J
4
2
+ 1638 J
2
2
J
3
J
4
80 J
2
J
3
3
312 J
2
4
J
3
= 0 (19b)
4050 J
5
J
7
25650 J
2
J
3
J
7
14310 J
2
J
4
J
6
+ 9600 J
3
2
J
6
+ 7965 J
2
3
J
6
+ 9450 J
3
J
4
J
5
+ 10530 J
2
2
J
3
J
5
+ 1134 J
4
3
+ 11259 J
2
2
J
4
2
12330 J
2
J
3
2
J
4
9018 J
2
4
J
4
+ 400 J
3
4
+ 3270 J
2
3
J
3
2
+ 1350 J
2
6
= 0 (19c)
12150 J
2
J
3
J
7
+ 3600 J
6
2
11610 J
2
J
4
J
6
+ 9750 J
3
2
J
6
+ 4410 J
2
3
J
6
+ 8505 J
3
J
4
J
5
+ 3645 J
2
2
J
3
J
5
+ 1458 J
4
3
+ 5670 J
2
2
J
4
2
10710 J
2
J
3
2
J
4
4104 J
2
4
J
4
+ 400 J
3
4
+ 2580 J
2
3
J
3
2
+ 576 J
2
6
= 0 (19d)
1800 J
6
J
7
10800 J
2
J
4
J
7
+ 4800 J
3
2
J
7
+ 4950 J
2
3
J
7
+ 4020 J
3
J
4
J
6
8370 J
2
2
J
3
J
6
+ 162 J
4
2
J
5
+ 7371 J
2
2
J
4
J
5
+ 2880 J
2
J
3
2
J
5
3483 J
2
4
J
5
9216 J
2
J
3
J
4
2
+ 640 J
3
3
J
4
+ 11946 J
2
3
J
3
J
4
720 J
2
2
J
3
3
2160 J
2
5
J
3
= 0 (19e)
60750 J
7
2
+ 178200 J
3
J
4
J
7
546750 J
2
2
J
3
J
7
+ 3780 J
4
2
J
6
246780 J
2
2
J
4
J
6
+ 348000 J
2
J
3
2
J
6
+ 137025 J
2
4
J
6
+ 116640 J
2
J
3
J
4
J
5
75600 J
3
3
J
5
+ 223560 J
2
3
J
3
J
5
+ 29808 J
2
J
4
3
+ 82170 J
3
2
J
4
2
+ 177660 J
2
3
J
4
2
438390 J
2
2
J
3
2
J
4
148014 J
2
5
J
4
+ 17200 J
2
J
3
4
+ 102000 J
2
4
J
3
2
+ 22221 J
2
7
= 0 (19f)
If these relations are satised, the elementary symmetric polynomials
1
,
2
and
3
can be expressed,
on the open stratum dened by
6 J
6
9 J
2
J
4
20 J
3
2
+ 3 J
2
3
,= 0,
On anisotropic polynomial relations of the elasticity tensor 13
as rational expressions of J
2
, . . . , J
7
. More precisely, we have

1
=
9

3 J
7
3 J
2
J
5
+ 3 J
3
J
4
J
2
2
J
3

6 J
6
9 J
2
J
4
20 J
3
2
+ 3 J
2
3
.
and

2
=
4
7

1
2

1
14
J
2
,
3
=
1
24
J
3
+
1
7

1
3

1
56

1
J
2
,
Then
1
,
2
,
3
are recovered (up to their monodromy group) as the roots of the third degree polyno-
mial in
p() =
3

2
+
2

3
But, to ensure that given J
2
, . . . , J
7
R we can nd
1
,
2
,
3
R, the following inequalities have to be
satised

2
= 2
2
1
6
2
0 ,
3
=
2
1

2
2
4
3
1

3
+ 18
1

2
3
27
2
3
0
The discriminant
3
and
2
can be rewritten, using J
2
, . . . , J
7
, as

2
=
N
2
14

6 J
6
9 J
2
J
4
20 J
3
2
+ 3 J
2
3

2
;
3
=
6 J
6
9 J
2
J
4
20 J
3
2
+ 3 J
2
3
432
where
N
2
= 3645 J
7
2
+ 7290 J
2
J
5
J
7
7290 J
3
J
4
J
7
+ 2430 J
2
2
J
3
J
7
36 J
2
J
6
2
+ 108 J
2
2
J
4
J
6
+ 240 J
2
J
3
2
J
6
36 J
2
4
J
6
3645 J
2
2
J
5
2
+ 7290 J
2
J
3
J
4
J
5
2430 J
2
3
J
3
J
5
3645 J
3
2
J
4
2
81 J
2
3
J
4
2
+ 2070 J
2
2
J
3
2
J
4
+ 54 J
2
5
J
4
400 J
2
J
3
4
285 J
2
4
J
3
2
9 J
2
7
.
We will summarize these results in the following proposition.
Proposition 2.10. A harmonic tensor D H
4
is in the closed stratum
[D
2
]
if and only if the invariants
J
2
(D), , J
10
(D) satisfy the polynomial relations (19), completed by the three equations
S
k
(J
2
, . . . J
10
) = 0, k = 8, 9, 10 (20)
obtained from J
k
p
k
(
1
,
2
,
3
), where we have substituted the corresponding rational expressions for the

i
, and we have cleared the denominators. The inequalities
6 J
6
9 J
2
J
4
20 J
3
2
+ 3 J
2
3
0 (21)
as well as
3645 J
7
2
+ 7290 J
2
J
5
J
7
7290 J
3
J
4
J
7
+ 2430 J
2
2
J
3
J
7
36 J
2
J
6
2
+ 108 J
2
2
J
4
J
6
+ 240 J
2
J
3
2
J
6
36 J
2
4
J
6
3645 J
2
2
J
5
2
+ 7290 J
2
J
3
J
4
J
5
2430 J
2
3
J
3
J
5
3645 J
3
2
J
4
2
81 J
2
3
J
4
2
+ 2070 J
2
2
J
3
2
J
4
+ 54 J
2
5
J
4
400 J
2
J
3
4
285 J
2
4
J
3
2
9 J
2
7
0
(22)
are also required. D is in the orthotropic class [D
2
] if moreover inequality (21) is strict. In that case, it
admits the normal form (18) where
1
,
2
,
3
are the roots of the third degree polynomial
p() =
3

2
+
2

3
.
Remark 2.11. Even if it is the core of the method, it is worth to note that, instead of
1
,
2
,
3
, the three
numbers
1
,
2
,
3
are uniquely dened on the open stratum: they are invariants of D and can be written
as rational expressions of (J
2
, . . . , J
7
). It should be noticed that it is also possible to express
1
,
2
,
3
as rational functions of only (J
2
, . . . , J
5
) on an open subset of
[D
2
]
. We will, however, not give these
formulas here.
Remark 2.12. All second rank tensors in H
2
covariant to D commute.
14 N. Auray et al.
2.3 Bifurcation conditions for tensors in H
4
An important observation has to be made. Mathematical results [40] tell us that the invariant algebra
of H
4
is generated by the 9 fundamental invariants (J
2
, . . . , J
10
). In this set 6 invariants (J
2
, . . . , J
7
)
are algebraically independent, meanwhile the others are linked to the formers by polynomial relations.
As discussed before, it is a well-known fact that this algebra separates the orbits. But, as demonstrated
exhaustively in this section, for any class with a nite monodromy group
4
, the rst 6 invariants (J
2
, . . . , J
7
)
are necessary and sucient to separate the orbits inside each class. This observation can be summed-up
by the following important theorem.
Theorem 2.13. For the following isotropy classes: the cubic, the transversely isotropic, the trigonal the
tetragonal (and the trivial isotropic) classes in H
4
(R
3
), the rst four invariants (J
2
, . . . , J
5
) separate the
orbits inside the class. For the orthotropic class, the rst six invariants (J
2
, . . . , J
7
) separate the orbits
inside the class.
We will now summarize the previous results by giving bifurcation conditions on the J
k
to make
explicit how we travel from a given isotropy class to another. More precisely, so far, we have given
necessary and sucient conditions to belong to a closed stratum
[H]
with nite monodromy group. As
we have seen,
[H]
is characterized by a set of polynomial relations (syzygies) involving the invariants
J
k
. These relations dene implicit equations which characterize the closed stratum
5
. The set of relations
that characterizes a given (closed) stratum generates an ideal (c.f. subsection B.4), I
[H]
in R[J
2
, . . . , J
10
],
and we have
[H
1
] _ [H
2
] I
[H
2
]
I
[H
1
]
.
The bifurcation conditions, detailed on Figure 2, correspond to relations (generators) which need to be
added to I
[H
1
]
to obtain I
[H
2
]
.
Fig. 2 Bifurcation paths for nite monodromy isotropy classes in H
4
.
4
i.e., for all symmetry classes, except the monoclinic and the triclinic classes
5
Notice that a rigorous and complete classication of these closed strata would require also some real inequalities that
we have not completely detailed here.
On anisotropic polynomial relations of the elasticity tensor 15
2.4 Relation with the characteristic polynomial
To conclude this subsection, some links that exist between the invariants of H
4
and the characteristic
polynomial of D will be investigated. The rst link is formulated by the following proposition:
Proposition 2.14. Let D H
4
. The coecients of the characteristic polynomial of D are SO(3)-
invariants of D. They can be expressed in terms of the invariants J
2
, . . . , J
5
dened above. We have

D
(z) = z
6

1
2
J
2
z
4

1
3
J
3
z
3
+
1
5

J
2
2
2 J
4

z
2
+
2
25
( J
2
J
3
3 J
5
) z. (23)
Proof. Notice rst that z = 0 is always an eigenvalue of D. Indeed, if q stands for the standard metric
on R
3
, we have Dq = tr
12
(D) = 0. The coecient of z
k
in
D
is a homogeneous invariant polynomial
of degree 6 k for 1 k 4. Therefore, it can be expressed as a polynomial in J
2
(D), J
5
(D). The
computation of this last expression reduces therefore to identify some real coecients. This can be done
by specializing on particular values or on some particular strata where it is easy to compute, the cubic
or the transverse isotropic stratum, for instance.
Besides the characteristic polynomial, Betten [7] has introduced a two variables, invariant polynomial
of the elasticity tensor. It is dened as
B
C
(, ) = det(C C
,
),
where
C
,
=

+ 2 0 0 0
+ 2 0 0 0
+ 2 0 0 0
0 0 0 2 0 0
0 0 0 0 2 0
0 0 0 0 0 2

is the totally isotropic tensor. Notice that B

D
(0, ) =
D
(2 ) and that B
C
is of degree lower than 1 in
(indeed, this can be observed by subtracting the rst line to the second line and to the third line in the
determinant). The objective of this polynomial is to obtain more invariants than the ones obtained from
the characteristic polynomial. However, it does not bring any new invariants for a tensor D H
4
(R
3
).
Indeed, we have the following result.
Lemma 2.15. The Betten polynomial of a harmonic tensor D H
4
(R
3
) is related to its characteristic
polynomial by the formula
B
D
(, ) = (3 + 2 )
r
D
(2 ),
where
D
() =
r
D
().
Proof. Let L
k
(1 k 6) denote the lines of the determinant det(D C
,
). For a harmonic tensor D,
we have
3

k=1
d
p,k
= 0, 1 p 6.
Il veut dire quoi ce d,k ? Thus, substituting L
1
L
1
+ L
2
+ L
3
in the determinant det(D C
,
),
we get
B
D
(, ) = (3 + 2 ) P(, ),
but since B
D
(, ) is of degree lower than 1 in we have P(, ) = P(0, ). Now for = 0 we have
B
D
(0, ) =
D
(2 ) = (2 ) P(0, ),
from which we get P(0, ) =
r
D
(2 ). This nishes the proof.
Corollary 2.16. The invariants dened by the coecients of the Betten polynomial do not separate the
orbits of Ela.
Proof. If this was the case, it would also separate the orbits of H
4
but then lemma 2.15 and proposi-
tion 2.14 would imply that J
2
, J
3
, J
4
, J
5
separate the orbits of H
4
, which is false.
16 N. Auray et al.
3 Conclusion
In this paper, ve sets of algebraic relations were provided to identify the following symme-
try classes fromH
4
(section 2): orthotropic ([D
2
]), trigonal ([D
3
]), tetragonal ([D
4
]), transverse
isotropic ([SO(2)]) and cubic ([O]). These relations are invariant necessary conditions for a
elasticity tensor to be in one of each discussed symmetry class. We consider this paper as a
rst step towards a more systematic use of invariant-based methods in continuum mechanics. Using the
geometrical framework introduced in this paper some extensions of the current method can be considered:
We aim at extending the current syzygy-characterization to classes having continuous monodromy
group;
To propose a complete set of normal forms of elasticity tensor constructed using this invariants ap-
proach;
To tackle the increasing degree of the algebraic relations for low symmetry group, it might be inter-
esting to characterize classes in terms of covariants.
Such an approach was conducted here in the context of classical elasticity, but the framework used
allows a direct extension of this method to other kinds of anisotropic tensorial behaviours. It can be suited
for the study of anisotropic features of piezoelasticity [21], exoelasticity [27] or strain-gradient elasticity
[26]. Furthermore, and in a more experimental view, we aim at testing our approach on experimental
situations. This would allow to compare the proposed approach with the more classical ones.
A Geometry of orbit spaces
In this section, we consider a linear representation
: G GL(V )
of a compact, real Lie group
6
G on a nite dimensional R-vector space V . The induced action will be noted
g v := (g)v, g G, v V.
Since G never acts without xed points (because it always xes 0 at least), the orbit space V/G is never a dierentiable
manifold [2]. This is the main reason why the description of an orbit space is dicult, in general. For a compact, real Lie
group, the algebra of G-invariant polynomials, is known to separate the orbits (c.f. subsection B.5). Using this fact, V/G
can be described as a semialgebraic
7
subset of R
N
. Let J
1
, J
2
, . . . , J
N
be a set of polynomial invariants which separate
the G-orbits. Then, the map
J : v

J
1
(v), J
2
(v), . . . , J
N
(v)

,
induces a homeomorphism between the orbit space V/G and J(V ) R
N
, which is a semialgebraic subset of R
N
. In
particular, such a set is a stratied space. It is the union of manifolds of various dimensions, called strata, each of them
being the union of orbits of a given isotropy type. This construction is the objective of the present section.
A.1 Orbits, symmetry groups and xed point sets
Two vectors v
1
and v
2
are said to be G-related, and we write v
1
v
2
if there exists g G such that v
2
= g v
1
. The set of
all vectors v V which are related to v by G is called the G-orbit of v and is denoted by
G v := g v [ g G .
The G-orbits are compact submanifolds of V [2, 11].
The set of transformations of G which leave the vector v xed forms a closed subgroup of G, which is called the isotropy
subgroup (or symmetry group) of v. It will be denoted by
G
v
:= g G; g v = v .
It can easily be checked that symmetry groups of G-related vectors are conjugate
G
gv
= g G
v
g
1
, v V, g G. (24)
Let H be any subgroup of G. The set of vectors v V which are xed by H,
V
H
:= v V [ h.v = v for all h H ,
6
In the following G will always represent a compact real Lie group, therefore this precision will usually be omitted.
Furthermore, we will always assume that this representation is faithful, that is ker = Id.
7
A semialgebraic set is a subset S of R
n
dened by a nite number of polynomial equations P
j
(x
1
, . . . , x
n
) = 0, and
inequalities Q
k
(x
1
, . . . , x
n
) 0.
On anisotropic polynomial relations of the elasticity tensor 17
is called the xed point set of H. It it is a linear subspace of V . Notice that, for each v V
H
, H G
v
. One can check that
if H
1
H
2
then V
H
2
V
H
1
. However, it may happen that V
H
2
= V
H
1
but H
1
,= H
2
. Furthermore, it should be pointed
out that, in general, given H
1
, H
2
in the same conjugacy class [H], we have V
H
2
,= V
H
1
. Notice also that a G-orbit belongs
to an isotropy class greater than [H] i it intersects the space V
H
.
Given a subgroup H of G, the normalizer of H, dened as
N(H) :=

g G [ gHg
1
= H

is the maximal subgroup of G, in which H is a normal subgroup. It can also be described as the isotropy group of H when
G acts on its subgroups by conjugacy. The proof of the following important lemma can be found in [22].
Lemma A.1. The space V
H
is N(H)-invariant. Moreover, if H is an isotropy subgroup, then N(H) is the maximal
subgroup of G which leaves invariant V
H
.
A.2 Isotropy stratication of a linear representation
Two vectors v
1
and v
2
are in the same isotropy class, and we write v
1
v
2
, if their isotropy subgroups are conjugate in G,
that is if there exists g G such that
G
v
2
= g G
v
1
g
1
.
Due to (24), two vectors v
1
and v
2
which are in the same G-orbit are in the same isotropy class but the converse is generally
false. Therefore, the relation to be in the same isotropy class is weaker than the relation to be in the same G-orbit. This
observation is summed up by the following implication
v
1
v
2
v
1
v
2
.
The conjugacy class of a subgroup H will be denoted by [H]. The conjugacy class of an isotropy subgroup will be called an
isotropy class. One can show that there is only a nite number of isotropy classes (see [11]). In the particular case of SO(3)
representations on tensors, this result is known to the mechanical community as a consequence of the Hermann theorem [23,
3].
On the set of conjugacy classes of closed subgroups of G, there is a partial order induced by inclusion
8
. It is dened as
follows:
[H
1
] [H
2
] if H
1
is conjugate to a subgroup of H
2
.
Endowed with this partial order, the set of isotropy classes
9
has a least element and a greatest element.
The set of all vectors v in the same isotropy class dened by [H] is denoted
[H]
and called a stratum. It is the union of
all G-orbits of points which isotropy group is conjugate to H (it is not a vector subspace in general). There is only a nite
number of (non empty) strata and each of them is a smooth submanifold of V [2, 11]. The partition
V =
[H
0
]

[H
1
]

[H
n
]
is called the isotropy stratication of (V, ).
The partial order relation on conjugacy classes induces a (reverse) partial order relation on strata

[H
2
]

[H
1
]
i [H
1
] [H
2
].
The set of strata inherits therefore the structure of a nite poset. An orbit G v is said to be generic if it belongs to the
least isotropy class. The generic stratum
[H
0
]
, which is the union of generic orbits can be shown to be a dense and open
set in V . Moreover, for each other stratum
[H]
, we have dim
[H]
< dim
[H
0
]
. On the opposite side, the stratum which
corresponds to the greatest isotropy subgroup is called the minimal stratum
10
.
Remark A.2. The closure of a stratum
[H]
, denoted by
[H]
and called shortly the closed stratum associated to [H],
corresponds to vectors v V such that v has at least isotropy [H]. Notice that
[H]
corresponds to vectors v V such that
v has exactly isotropy [H] and that a closed stratum is a union of strata.
A.3 Fixed point sets and monodromy groups
One consequence of lemma A.1 is that the linear representation : G GL(V ) induces a linear representation of N(H) on
V
H
, obtained by the restriction

/N(H)
: N(H) GL(V
H
).
This induced representation is not faithful because its kernel contains H. However, when H is an isotropy group, its kernel
is exactly H and we get a faithful linear representation

H :
H
GL(V
H
) where
H
:= N(H)/H.
8
This order relation is called containment in [22]. It is clearly reexive and transitive. The antisymmetry property is
however more dicult to check and, up to the authors knowledge, is proved only when G is a nite group or a compact
topological group (see [11]).
9
The partially ordered set of conjugacy classes of all closed subgroups of SO(3) is described in Appendix C.
10
Since the isotropy group of the null-vector is G itself, this minimal stratum is always represented by the isotropy class
[G] = G. The stratum
[G]
is always reduced to 0 if is a non-trivial irreducible representation but it may not be
otherwise.
18 N. Auray et al.
Furthermore, in that case, the equivalence relation dened by (V, ), when restricted to the set V
H

[H]
, is the
same as the equivalence relation dened by (V
H
,
N(H)
). Indeed, let v
1
, v
2
in V
H
, such that [G
v
1
] = [G
v
2
] = [H]. Then
G
v
1
= G
v
2
= H. Suppose that v
2
= g v
1
, for some g G. Then ghg
1
v
2
= v
2
, for each h H. Thus ghg
1
G
v
2
= H
for all h H and g N(H).
We have therefore reduced (locally, on the stratum
[H]
) the problem of describing the orbit space of V mod G by the
orbit space of V
H
mod
H
. This is especially meaningful when the group
H
is nite, in which case, we say that V
H
is a
linear slice. Then, each G-orbit intersects V
H
at most in a nite number of points and the natural map V
H
V
H
/
H
is a
nite ramied covering
11
of V
H
/
H
. The degree of V
H
is dened as the cardinal of
H
. It is equal to the index [N(H), H]
of H in its normalizer N(H). It is also the number of points in a generic ber of the covering V
H
V
H
/
H
. The group

H
is called the monodromy group of V
H
.
A.4 Strata dimensions
The main formula, which is the basis for all other equations relating strata dimensions, xed-point subspaces and dimension
of isotropy subgroups is explained in the following lemma.
Lemma A.3.
dim
[H]
= dimV
H
+ dimGdimN(H) (25)
Proof. The restriction of the projection map : V V/G to a stratum
[H]

/
[H]
:
[H]

[H]
/G
is a ber bundle with ber G/H, i.e., a space that locally looks like
[H]
/GG/H. Therefore, we have
dim
[H]
= dim(G/H) + dim(
[H]
/G)
where dim(G/H) = dimGdimH.
Notice that the set V
H

[H]
is the subspace of vector in V
H
which isotropy class is exactly [H]. It is an open and
dense set in V
H
, i.e., almost all elements of V
H
are in V
H

[H]
. These vectors correspond to generic elements in V
H
.
Now, since the equivalence relation dened by (V, ), when restricted to the space V
H
, is the same as the equivalence
relation dened by (V
H
,

H), the basis

[H]
/G of the bundle
[H]
is dieomorphic to (V
H

[H]
)/
H
(see [11] for a
more rigourous justication). Thus
dim
[H]
/G = dim(V
H

[H]
)/
H
= dimV
H
dim
H
dim
[H]
= dimV
H
+ dim(G/N(H)) = dimV
H
+ dimGdimN(H)
Remark A.4. The dimension of V
H
can be computed using the trace formula for the Reynolds operator (see [22]). Letting

be the character of the representation (V, ), we have

dimV
H
=
1
[H[

hH

(h), (26)
if H is a nite group (and the preceding formula has to be replaced by the Haar integral over H for an innite compact Lie
group). Explicit analytical formulas based on (26) were obtained in [22] and [4].
A.5 Implicit equations for closed strata
Let [H] be a xed isotropy class and
[H]
V be the corresponding stratum. Given a nite set of invariant polynomials
which separate the orbits of G, the closed stratum
[H]
can be characterized by a nite set of relations (equations and
inequalities) on J
1
, J
2
, . . . , J
N
. To obtain these equations, we x a subgroup H in the conjugacy class [H] and choose
linear coordinates (x
i
)
1iq
on V
H
. Then, we evaluate J
1
, J
2
, . . . , J
N
on V
H
(as polynomials in the x
i
) and we try to
obtain implicit equations on the J
k
for the parametric system
J
k
= P
k
(x
1
, . . . x
q
), k = 1, . . . , N, P
k
R[X
1
, . . . , X
q
], (27)
satised by the restriction of J
1
, J
2
, . . . , J
N
to V
H
.
This is a dicult task in general (one might consider [16] for a full discussion about the implicitization problem).
Moreover, it could happen that the algebraic variety V
I
dened by such an implicit system is bigger than the variety
V
P
, dened by the parametric system (27). Besides, we are confronted to the diculty that we work over R which is not
algebraically closed. To overcome these diculties, we observe that the restrictions of J
1
, J
2
, . . . , J
N
to V
H
are
H
-
invariant and therefore can be expressed as polynomial expressions of some generators
1
, . . . ,
r
of the invariant algebra
11
A ramied covering is a generalization of the concept of a covering map. It was introduced in the theory of Riemann
surfaces and corresponds to the situation where a nite (or discrete) group acts on a manifold but with xed points. A
typical example is given by the map z z
n
on the complex plane. Outside 0, this map is a covering but 0 is a singularity,
a point of ramication. In higher dimension, the set of ramications points may be more complicated.
On anisotropic polynomial relations of the elasticity tensor 19
of the monodromy group
H
on V
H
. This observation reduces the problem to consider rst the implicitization problem for
the parametric system
J
k
= p
k
(
1
, . . . ,
r
), k = 1, . . . , N, p
k
R[X
1
, . . . , X
r
]. (28)
To solve this problem, we use a Groebner basis [16] (see subsection B.4) or a regular chain [10]. The main observation
is that, in each case we consider, the system (28) leads to a system of relations (syzygies) which characterizes the closed
stratum
[H]
S
j
(J
1
, J
2
, . . . , J
N
) = 0, j = 1, . . . l S
j
R[X
1
, . . . , X
N
], (29)
and a system

i
= R
i
(J
1
, J
2
, . . . , J
N
), i = 1, . . . , r R
i
R(X
1
, . . . , X
N
), (30)
which express the
i
as rational
12
functions of J
1
, J
2
, . . . , J
N
on the open stratum
[H]
. Beware, however, that these
rational expressions may not be unique. However if some
i
can be written as P
1
/Q
1
as well as P
2
/Q
2
then P
1
Q
2
P
2
Q
1
belongs to the ideal generated by the S
j
.
Because the solutions are rational, the fact that the eld on which we work is real or complex does not matter at this
level. Therefore, each real solution (J
1
, J
2
, . . . , J
N
) of (29) corresponds to a unique real solution (
1
, . . . ,
r
) given by (30).
Nevertheless, we cannot take for granted that a real solution (
1
, . . . ,
r
) of (30) corresponds to a real point (x
1
, . . . , x
q
) in
V
H
. For this, we need to compute an additional system of inequalities on the
i
, or equivalently on the J
k
, which permits
to exclude complex solutions of

i
= Q
i
(x
1
, . . . x
q
), i = 1, . . . , r, Q
i
R[X
1
, . . . , X
q
].
However, in the present paper for the linear slices we consider, the only monodromy group that we encounter are 1, S
2
and
S
3
and the action is the standard one in appropriate coordinates. In these cases, and more generally when the monodromy
group is S
n
and the action is isomorphic to the standard one, r = q = n, the invariants
1
, . . . ,
n
are algebraically
independent and (in an appropriate coordinates system) x
1
, . . . , x
n
are the roots of the polynomial
p(z) = z
n

1
z
n1
+ + (1)
n

n
.
Therefore, the problem reduces to nding conditions on the
i
that ensure that all the roots of p are real. The solution is
due to Hermite, [12]. He has proved that the number of distinct real roots of a real polynomial p of degree n is equal to the
signature of the Hankel matrix B(p) := (S
i+j2
)
1i,jn
, where S
k
:=

n
i=1
(x
i
)
k
is the power sum of the roots of p. In
particular, p has real roots if and only if B(p) is non-negative.
B Geometric concepts
B.1 Normal form
A normal form is a map NF : V V such that
v
1
v
2
if and only if NF(v
1
) = NF(v
2
).
The existence of a computable normal form allows to decide whether two vectors v
1
and v
2
belong to the same G-orbit.
Remark B.1. In practice, one contents itself usually with a normal form dened up to a nite group. For instance, one
is satised with a diagonal matrix as the normal form for the action of the rotation group on the space of symmetric
matrices. However this diagonal form is not unique. It is dened up to the permutation of the diagonal entries.
B.2 Covariants
Given a second representation (V

) of the same group G, we say that a linear map : V V

is equivariant if the
following diagram
V

V

(g)

(g)
V

V

commutes for each g G. The composition of two equivariant maps is equivariant.

Given an equivariant linear map , the element (v) V

will be said to be covariant to v V . It is clear that every

element v V is covariant to itself (with = Id). However, this relation is not symmetric. Indeed, if v

is covariant to v,
the converse is not necessarily true (v is not necessarily covariant to v

) unless the linear map : V V

is bijective. In
this case, we say that the representations (V, ) and (V

) are isomorphic or equivalent.

When the representation (V

) is trivial, i.e., when

(g) = Id for all g G, the covariant element (v) will be called

an invariant. We have then
v v

(v) = (v

).
12
The fact that the solutions are rational will not be justied here. We just observe that this is the case for all classes we
have treated in this article.
20 N. Auray et al.
B.3 Polynomial invariants
The linear action of the group G on the R-vector space V extends naturally to the vector space of polynomial functions
dened on V with values in R. This extension is given by
(g P)(v) := P(g
1
v)
for every polynomial P R[V ] and every vector v V . The set of all invariant polynomials is a sub-algebra of the algebra
R[V ] of all polynomials dened on V . We denote this sub-algebra by R[V ]
G
and call it the invariant algebra.
When G is a compact Lie group (or more generally a reductive group), this algebra R[V ]
G
is of nite type [24, 34]. This
means rst that it is possible to nd a nite set of invariant polynomials J
1
, . . . , J
N
which generates R[V ]
G
(as an algebra).
But, in general, this algebra is not free, which means that it is not always possible to extract from J
1
, . . . , J
N
a subset of
generators which are algebraically independent
13
. The set of polynomial relations on J
1
, . . . , J
N
(also known as syzygies) is
an ideal of R[J
1
, . . . , J
N
] that is nitely generated. Such a set of generators is called an integrity basis for R[V ]
G
. Of course
it is not unique and moreover its cardinal may change from one basis to another.
B.4 Ideal and Groebner basis
Consider a subset V of k
n
(where k = R or C) dened implicitly by a nite set of polynomial equations p
1
, . . . p
r
. The
fundamental observation is that the set I of polynomials p k[x
1
, . . . , x
n
] such that p(x) = 0 for each point x S (where
x = (x
1
, . . . , x
n
)) satises the following properties:
(i) 0 I,
(ii) If p, q I, then p +q I,
(iii) If p I and q k[x
1
, . . . , x
n
], then pq I.
Such a subset I of k[x
1
, . . . , x
n
] (or more generally any commutative ring) is called an ideal. In the example above the ideal
I is (nitely) generated by p
1
, . . . p
r
(i.e., every p I can be written as a nite sum

s
j=1
f
j
p
j
where f
j
k[x
1
, . . . , x
n
])
and we write I =< p
1
, . . . p
r
>.
To solve a system of linear equations in k
n
, a standard algorithm is to reduce the matrix of the system to a triangular
form (a normal form, even if not unique). A Groebner basis of an ideal I is a set of generators for I which is a normal form
for an algebraic system, in a certain sense analog to a triangular system for a set of linear equations. It permits to compute
and answer eectively a lot of problems related to ideals k[x
1
, . . . , x
n
] and algebraic systems. It solves, in particular, the
implicitization problem. We refer to [16] for an excellent exposition on Groebner basis (which is very accessible for non
specialists).
B.5 Separation of orbits
Given an algebra , of invariant functions dened on V , we say that , separates the orbits if given two vectors v
1
, v
2
V
which are not on the same orbit, it is always possible to nd a function f , such that f(v
1
) ,= f(v
2
). In general, when
the group G is not compact, the algebra of polynomial invariants does not separate the orbits
14
. However, it is remarkable
that for any compact group, the algebra of real invariant polynomials always separates the orbits. This result is attributed
to Weyl [41].
Proposition B.2. Let (V, ) be a nite dimensional, linear representation of a compact group G. Then the algebra of
real invariant polynomials R[V ]
G
separates the orbits.
C Normalizers of closed subgroups of SO(3)
Every closed subgroup of SO(3) is conjugate to one of the following list [17, 22]
SO(3), O(2), SO(2), D
n
(n 2), Z
n
(n 2), T, O, I, and 1 (31)
The poset of conjugacy classes of closed subgroups of SO(3) is completely described [22] by the following inclusion of
subgroups (which generates order relations between conjugacy classes)
Z
n
D
n
O(2) (n 2),
Z
n
Z
m
and D
n
D
m
, (if n divides m),
Z
2
D
n
(n 2),
Z
n
SO(2) O(2) (n 2),
and the arrows in Figure 3 which complete the poset, taking account of the exceptional subgroups O, T, I (beware that
Figure 3 cannot be realized by an inclusion diagram between representatives of these conjugacy classes).
13
Polynomials P
1
, . . . , P
N
are said to be algebraically independent over R if the only polynomial in N variables which
satises Q(P
1
, . . . , P
N
) = 0 is the zero polynomial.
14
This can be illustrated by the example of the adjoint action of the general linear group GL(n, C) on its Lie algebra
gl(n, C). Each orbit is described by a unique Jordan form but the algebra of polynomial invariants, which is generated by
tr(M), tr(M
2
), . . . tr(M
n
) where M gl(n, C), is not able to distinguish between two dierent Jordan forms which have the
same diagonal part.
On anisotropic polynomial relations of the elasticity tensor 21
Fig. 3 Exceptional classes in the poset of conjugacy classes of closed subgroup of SO(3).
Remark C.1. Notice that O and I are maximal subgroups of SO(3) which both contained T.
Recall that the normalizer of a subgroup H of a group G, denoted by N(H) is the biggest subgroup of G in which H
is normal. It is dened by
N(H) :=

g G; gHg
1
= H

.
Proposition C.2. The normalizers of the subgroups in the list (31) are given below.
N(SO(3)) = SO(3), N(O(2)) = O(2), N(SO(2)) = O(2),
N(O) = O, N(I) = I, N(T) = O,
N(1) = SO(3), N(Z
n
) = O(2) for n 2, N(D
n
) = D
2n
for n 3,
and N(D
2
) = O. Moreover, the quotient groups are given by
N(SO(3))/SO(3) = 1, N(O(2))/O(2) = 1, N(SO(2))/SO(2) = Z
2
,
N(O)/O = 1, N(I)/I = 1, N(T)/T = Z
2
,
N(1)/1 = SO(3), N(Z
n
)/Z
n
= O(2) for n 2, N(D
n
)/D
n
= Z
2
for n 3,
and N(D
2
)/D
2
= S
3
, the symmetric group of 3 elements.
*
References
1. M. Abud and G. Sartori. The geometry of orbit-space and natural minima of Higgs potentials. Phys. Lett. B, 104(2):147
152, 1981.
2. M. Abud and G. Sartori. The geometry of spontaneous symmetry breaking. Ann. Physics, 150(2):307372, 1983.
3. N. Auray. Demonstration du theor`eme de Hermann `a partir de la methode Forte-Vianello. Comptes Rendus Mecanique,
336(5):458463, 2008.
4. N. Auray. Analytical expressions for anisotropic tensor dimension. Comptes Rendus Mecanique, 338(5):260265, 2010.
5. G. Backus. A geometrical picture of anisotropic elastic tensors. Rev. Geophys., 8(3):633671, 1970.
6. R. Baerheim. Harmonic decomposition of the anisotropic elasticity tensor. The Quarterly Journal of Mechanics and
Applied Mathematics, 46(3):391418, 1993.
7. J. Betten. Irreducible invariants of fourth-order tensors. Math. Modelling, 8:2933, 1987. Mathematical modelling in
science and technology (Berkeley, Calif., 1985).
8. A. Bona, I. Bucataru, and M. Slawinski. Characterization of Elasticity-Tensor Symmetries Using SU(2). Journal of
Elasticity, 75(3):267289, 2004.
9. J.-P. Boehler, A. A. Kirillov, Jr., and E. T. Onat. On the polynomial invariants of the elasticity tensor. J. Elasticity,
34(2):97110, 1994.
22 N. Auray et al.
10. F. Boulier, D. Lazard, F. Ollivier, and M. Petitot. Representation for the radical of a nitely generated dierential
ideal. In proc. ISSAC 95, International Symposium on Symbolic and Algebraic Computation, pages 158166, Montreal,
Canada, 1995.
11. G. E. Bredon. Introduction to compact transformation groups. Academic Press, New York, 1972. Pure and Applied
Mathematics, Vol. 46.
12. M. Coste. An Introduction to Semialgebraic Geometry. Universite de Rennes, 2002.
13. S. C. Cowin. Properties of the anisotropic elasticity tensor. Q. J. Mech. Appl. Math., 42:249266, 1989.
14. S. C. Cowin and M. Mehrabadi. On the identication of material symmetry for anisotropic elastic materials. Q. J.
Mech. Appl. Math., 40:451476, 1987.
15. S. C. Cowin and M. M. Mehrabadi. Eigentensors of linear anisotropic elastic materials. Q. J. Mech. Appl. Math.,
43:1541, 1990.
16. D. Cox, J. Little, and D. OShea. Ideals, varieties, and algorithms. Undergraduate Texts in Mathematics. Springer,
New York, third edition, 2007. An introduction to computational algebraic geometry and commutative algebra.
17. B. A. Dubrovin, A. T. Fomenko, and S. P. Novikov. Modern geometrymethods and applications. Part I, volume 93
of Graduate Texts in Mathematics. Springer-Verlag, New York, second edition, 1992. The geometry of surfaces,
transformation groups, and elds, Translated from the Russian by Robert G. Burns.
18. S. Forte and M. Vianello. Symmetry classes for elasticity tensors. J. Elasticity, 43(2):81108, 1996.
19. S. Forte and M. Vianello. Symmetry classes and harmonic decomposition for photoelasticity tensors. International
Journal of Engineering Science, 35(14):1317 1326, 1997.
20. S. Forte and M. Vianello. Restricted invariants on the space of elasticity tensors. Math. Mech. Solids, 11(1):4882,
2006.
21. G. Geymonat and T. Weller. Symmetry classes of piezoelectric solids. Comptes Rendus Mathematique, 335(10):847852,
2002.
22. M. Golubitsky, I. Stewart, and D. G. Schaeer. Singularities and groups in bifurcation theory. Vol. II, volume 69 of
Applied Mathematical Sciences. Springer-Verlag, New York, 1988.
23. B. Herman. Some theorems of the theory of anisotropic media. Comptes Rendus (Doklady) Acad. Sci. URSS, 48(2):89
92, 1945.
24. D. Hilbert. Theory of algebraic invariants. Cambridge University Press, Cambridge, 1993. Translated from the German
and with a preface by Reinhard C. Laubenbacher, Edited and with an introduction by Bernd Sturmfels.
25. J. Jerphagnon, D. Chemla, and R. Bonneville. The description of the physical properties of condensed matter using
irreducible tensors. Advances in Physics, 27(4):609680, 1978.
26. H. Le Quang, N. Auray, Q.-C. He, and G. Bonnet. Symmetry groups and symmetry classes for sixth-order tensors.
Proceedings of the Royal Society A: Mathematical, Physical and Engineering Science, page (to be submitted, 2012.
27. H. Le Quang and Q.-C. He. The number and types of all possible rotational symmetries for exoelectric tensors.
Proceedings of the Royal Society A: Mathematical, Physical and Engineering Science, 467(2132):23692386, 2011.
28. A. N. Norris. On the acoustic determination of the elastic moduli of anisotropic solids and acoustic conditions for the
existence of symmetry planes. Quart. J. Mech. Appl. Math., 42:413426, 1989.
29. P. J. Olver. Group theoretic classication of conservation laws in elasticity. In Systems of nonlinear partial dierential
equations (Oxford, 1982), volume 111 of NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., pages 323331. Reidel,
Dordrecht, 1983.
30. P. J. Olver. Conservation laws in elasticity. I. General results. Arch. Rational Mech. Anal., 85(2):111129, 1984.
31. P. J. Olver. Canonical elastic moduli. J. Elasticity, 19(3):189212, 1988.
32. P. J. Olver. Conservation laws in elasticity. III. Planar linear anisotropic elastostatics. Arch. Rational Mech. Anal.,
102(2):167181, 1988.
33. P. J. Olver. Canonical anisotropic elastic moduli. In Modern theory of anisotropic elasticity and applications (Research
Triangle Park, NC, 1990), pages 325339. SIAM, Philadelphia, PA, 1991.
34. P. J. Olver. Classical invariant theory, volume 44 of London Mathematical Society Student Texts. Cambridge University
Press, Cambridge, 1999.
35. R. S. Palais. The classication of G-spaces. Mem. Amer. Math. Soc. No. 36, 1960.
36. C. Procesi and G. Schwarz. Inequalities dening orbit spaces. Invent. Math., 81(3):539554, 1985.
37. C. Procesi and G. W. Schwarz. The geometry of orbit spaces and gauge symmetry breaking in supersymmetric gauge
theories. Phys. Lett. B, 161(1-3):117121, 1985.
38. G. Sartori and G. Valente. Tools in the orbit space approach to the study of invariant functions: rational parametrization
of strata. J. Phys. A, 36(7):19131929, 2003.
39. G. W. Schwarz. Lifting smooth homotopies of orbit spaces. Inst. Hautes

Etudes Sci. Publ. Math., (51):37135, 1980.
40. T. Shioda. On the graded ring of invariants of binary octavics. Amer. J. Math., 89:10221046, 1967.
41. H. Weyl. The classical groups. Princeton Landmarks in Mathematics. Princeton University Press, Princeton, NJ, 1997.
Their invariants and representations, Fifteenth printing, Princeton Paperbacks.
42. Z. Zeng. Computing multiple roots of inexact polynomials. Mathematics of Computation, Vol. 74, p. 869-903, 74:869
903, 2005.