Journal of Elasticity 34: 97-110, 1994.

0 1994 Kluwer Academic Publishers. Printed in the Netherlands.

On the polynomial invariants of the elasticity tensor

J.P. B O E H L E R
Institute de Mdcanique de Grenoble, France

A.A. K I R I L L O V , JR. and E.T. O N A T

Yale UniversiO,, New Haven, CT 06520, USA
Received 8 June 1992; in revised form 21 June 1993

Abstract. We consider the old problem of finding a basis of polynomial invariants of the fourth
rank tensor C of elastic moduli of an anisotropic material. Decomposing C into its irreducible
components we reduce this problem to finding joint invariants of a triplet (a, b, D), where a and b

are traceless symmetric second rank tensors, and D is completely symmetricand traceless fourth
rank tensor (D e T~). We obtain by reinterpreting the results of classical invariant theory a
polynomial basis of invariants for D which consists of 9 invariants of degrees 2 to 10 in components
of D. Finally we use this result together with a well-known description of joint invariants of a
number of second-rank symmetric tensors to obtain joint invariants of the triplet (a, b, D) for a
generic D.

O. Introduction. Motivation for the work

The main purpose of the work is to construct a classification of linear
anisotropic elastic materials. In colloquial terms we ask the question: How do
we give distinct names to distinct anisotropic elastic materials? Clearly a
designation based on the 21 components of the tensor C of elastic moduli in a
fixed reference frame is not good for this purpose because it provides,
generically, different names for different orientations of a given material. As the
material is rotated the tensor C moves on its orbit in the space T~, of elasticity
tensors. Thus what is needed is a parametrization of distinct orbits. The set O c
of the distinct orbits of elasticity moduli is a manifold of (21-3) = 18 dimensions that has a fairly complicated boundary. The problem of naming the
distinct orbits would be solved if Or, the manifold of distinct orbits, can be
mapped in a one-to-one and continuous manner into the linear space Rn; the
coordinates of an image point would then serve as the name of the associated
orbit. It is of interest to know what minimal dimension n is needed for this
purpose. The following examples may be of help in thinking about this idea:
If the manifold of interest is a circle which is one dimensional then the
minimal dimension n = 2. If the manifold of interest is the group SO(3) of rigid

98

J. P. Boehler et al.

body rotations (which is three dimensional) then n = 5. One needs at most a

37 dimensional space to map a smooth manifold of dimension 18 in the above
described manner.
In the present paper we show that 39 polynomial invariants of C can be used
to designate a certain generic set of materials.

1. Decomposition of the elasticity tensor

Let C be the tensor of elastic moduli of a homogeneous material at some
reference temperature 0o. Fix some orthonormal basis in R 3 and let Cijkl stand
for the components of C in this basis. Then for infinitesimal isothermal
deformations and rotations Hooke's law for the material has the form:

%=C~**eu

or tr=Ce.

Thus, C can be regarded as a linear map C: T~ ~ T$, where T~ is the space

of symmetric rank two tensors. This map is self-adjoint. From this follows that
C has the following symmetries: Cou = C iiu = Cotk = Ckui. We will denote the
space of all rank 4 tensors satisfying these conditions by T$; it has dimension 21.
If we rotate the material by g e SO(3) then the elasticity moduli of the rotated
material are given by the tensor which we denote gC; its components are given
by
(gC) ijkl = glpg jqgkrgls C pqrs.

This defines the action of SO(3) on the space T$ and renders it a finitedimensional representation of SO(3). For the calculation of invariants of C it
will be convenient to decompose T~, into simpler pieces. Indeed, it can be
shown (cf. [2]) that under the action of SO(3) the space T~, decomposes into
direct sum of spaces of dimensions 1, 1, 5, 5, 9:

where T]s stands for the space of completely symmetric traceless tensors of
rank n, which has the dimension 2n + 1; in other words, we have the following
lemma:
LEMMA. Each Cokl~ T~, can be represented uniquely in the form:

Polynomial invariants

99

C~j~ = 26~jJ~t + 1~(Si~tijl + 6i~Sjk)

-~- (~ij.akl -~ (~klalj

q- 5ikbjt .q- t~jtbik + t~itbjk q- (~jkbil
+ Dijk!
where 2,/~ e R, a, b ~ TS2s and D ~ TS4s are completely symmetric traceless tensors
(thus, for instance, for every permutation p, D p(ijkt) = D ijkt, and D iikl = 0), and this
decomposition is conserved by the action of SO(3). Furthermore, 2, l~ and a u, bu,
Dijkt are linear combinations of components Ciju; for example, 2 =

1-1~5(2 C iipp -- C ippi).

REMARK. The tensors a, b, D inherit the symmetries of C. For example, for
isotropic materials we have gC = C for all g e SO(3), which implies a = b = 0,
D = 0 and we have only 2,/~ which are exactly the Lam6 constants.
Now we want to formulate some general results about finite-dimensional
representations of SO(3); later we will apply them to TL
Let V be a finite-dimensional representation of SO(3). A polynomial invariant
of V is a polynomial function p on V (i.e., if we choose a basis in V then p(v)
can be written as a polynomial in coordinates of v) which is invariant under
the action of SO(3), namely
p(gv) = p(v) for any g e S0(3).

REMARK. We can omit the condition of polynomiality and obtain a definition of an invariant function on V; for example, one may consider continuous
invariants. However, in this paper we only consider polynomial invariants.
The important property of polynomial invariants is that they "separate the
orbits". This means that if we know the values p(v) for all polynomial
invariants p we can recover v uniquely up to the action of SO(3) (cf. [1]).
This property is not very useful in itself, since there exists infinite number of
polynomial invariants. Our main goal is to find a finite set of polynomial
invariants which would separate the orbits. For this purpose we introduce the
following definition:
DEFINITION. A finite set p~ . . . . , Pk of polynomial invariants of V is called
an integrity basis if every polynomial invariant of V can be written as a
polynomial in p~ . . . . . Pk"
This is one of the basic notions of the classical invariant theory. One of the
main results of this theory claims existence of such a set (cf. [1], I-3]):

100

J. P. Boehler et al.

THEOREM. For any finite-dimensional representation V of SO(3) there exists

a (finite) integrity basis.
Now we can apply this result to the problem of separating the orbits by
invariants. Let us introduce one more definition:
DEFINITION. A finite set Pt . . . . . p, of invariants of V is called a functional
basis if

P,(vl) = Pi(V2) for all i = 1. . . . , n

implies v~ = gv2 for some g ~SO(3).
Note that in this definition we do not require that all these invariants be
polynomial; nevertheless all functional bases considered here will consist of
polynomial invariants. Finally we mention the following simple theorems:
THEOREM. Any integrity basis is a functional basis.
This theorem follows immediately from the definitions and the fact that the
set of all polynomial invariants separates orbits.
Note that the converse is false: for example, for the representation of SO(3)
in the space of symmetric traceless matrices T~~ it is known that the integrity
basis is given by the invariants 12(a) = tra 2, 13(a) = tra 3. Since tra 2 >/0, one
easily sees that the invariants 12, 13 form a functional basis which is not an
integrity basis.
So we have proved the following theorem:
THEOREM. For any finite-dimensional representation V of SO(3) there exists
a (finite)functional basis consisting of polynomial invariants.
Now let us return to our problem of finding a functional basis of polynomial
invariants of T~.. Obviously, 2 and p are invariants and the functional basis of
invariants of the elasticity tensor can be obtained by adding 2 and/~ to the
invariants of the triple (a, b,D). We can construct the functional basis of
invariants of this triplet by first finding the invariants of each element of the
triplet and then adding to this list the joint invariants. It is well known that
the functional basis of invariants of a t T~2s is given by tra 2, tra 3 (similar result
for b); what is not so well known is the basis of invariants of D, which we
discuss in Section 2. In Section 3 we use this result together with a well known
description of joint invariants of a number of symmetric second rank tensors
to obtain the joint invariants of the triplet (a, b, D) for a generic D.

Polynomial invariants

101

2. Invariants of symmetric traceless tensor of rank 4

In this section we obtain the integrity basis (and therefore, the functional basis)
of polynomial invariants of D e T~ using the results of classical invariant
theory. For this purpose we first show (cf. Appendix 1) that this problem
reduces to finding invariants of homogeneous polynomials of degree 8 in x, y
under the action of SL(2, C). This last problem was considered in the 1880's by
Sylvester and von Gall ([4-1, [5]). Their result when interpreted in the language
of tensors states that the minimal integrity basis of invariants of D ~ T~s consists
of 9 invariants of degrees 2, 3 . . . . . 10 in Dijk1.1
To give the explicit expressions for these invariants we adopt the following.
conventions: Small letters a,b . . . . are used for second rank tensors, while
capital letters denote fourth rank tensors. Given two tensors a = aij, b = b u we
define their product to be the second rank tensor given by (ab), = aobj~ (we
assume summation over repeating indices). Similarly, we define the product of
two fourth rank tensors to be given by (AB)ijk I = AijpqBpqkt , and the product of
two tensors of ranks two and four as (Ab)~j = Aok,b u. We introduce the
following traces: t r a = a,, tr A = Aij~j. We also need the second rank tensor d 2
defined as

(d2)ij : Dipq~Djpqr.
Now we can formulate our main result:
T H E O R E M 1. The nine invariants dE,..., J~o listed below, which are of
deorees 2 . . . . . 10 respectively form an integrity (and, therefore functional) basis

of polynomial invariants of O ~ T~s.

I nvariant

degree

J2 = tr d 2

J3 = tr D 3

diagram

tThese 9 invariants are not independent (one easily checks that dim(T~./SO(3)) = 6, and hence
only six of them are independent); in other words, there are polynomial relations between them;
however, these relations do not enable one to express one of these invariants as a polynomial of
the remaining ones. The complete list of relations among these invariants can be found in the
dissertation [61.

102

J. P. Boehler et al.

lnvariant

degree

J4 = trd2

diagram

NO

Js = tr d2Dd 2

J 6 = tr d23

J7 = tr

d2Dd2

Js = trd2D2d2

J9 = tr d2D(d 2)

J lo = tr d~D2(d2)

10

Polynomial invariants

103

Note that the third column of the list contains a diagrammatical description
of the invariants Ji, where each circle of a diagram represents D and a line
connecting two circles corresponds to a summation over one pair of indices. 2
For example, (d2)ij = DipqrDjpq, is represented by the following diagram:

It should be pointed out immediately that this set is different from the
integrity basis of invariants given in I-5], but these sets are equivalent in the
following sense: We call two invariants X and Y of the same degree equivalent
(i.e. X ~ Y) if X - Y can be expressed as a polynomial of invariants of lower
degrees. It should be also clear that if we substitute each invariant in a
functional basis by an equivalent one, we still get a functional basis. The above
list contains two examples of equivalent invariants. We also note that
J6 '7L tr D 6.
Thus, to prove the theorem it suffices to show the equivalence of our
invariants with those given in [5]. We start the proof by showing such
equivalence for degree 8. Except for degree 6 (to which we shall return below),
equivalence of remaining invariants in the two sets can be proved by using
similar arguments.
The invariant of degree 8 given in [5], when rewritten in terms of tensors (cf.
Appendix 1) is defined as
i~k = {A}ij{D2},~k,{dz}u,

where curly brackets stand for the symmetric and traceless part of the tensor
they bracket, and A = {d2}2. Now we claim that J8 = iAk" Since {D 2} is
completely symmetric and traceless, it is obvious that i~k = tr{d2}2{D2}dz .
Next, since for a second rank symmetric tensor {a}~j = a l i - 60tra, one easily
sees that
iak = tr d 2 { D 2 } d 2 + tr dz(some invariant of degree 6) ,-~ tr d2{D2}d z.

Now let us return to Js = trd2D2d2 It is easy to see that O2 T~ and hence

the decomposition of Section 1 holds. When we substitute this decomposition
for 0 2 in the expression for J8, we see that J8 "~ trd2{D2}d2 + c . t r d ~ . But it
follows from Hamilton-Cayley theorem applied to d 2 that trd24 can be
expressed via invariants of lower degrees, namely trd2, trd 2, trd23. Thus
J8 ~ tr d2{D2}d2 "~ iak.

2One notices that these diagrams resemblethe structural formulasin chemistry;but as was noted
a long time ago (cf. [3, Appendix 1]), this resemblanceis fortuituous.

104

J. P. Boehler et al.

For degree 6 we need the following non-trivial relation

,~

c.

which can be deduced from the Gordon's identities for invariants (cf. [31, [6]).
This completes the proof of the theorem.
[]

3. Invariants of the elasticity tensor C

As we have seen in the Section 1, the problem of finding the invariants of the
elasticity tensor C reduces to the problem of finding the invariants of the triple
(a, b, D) ~ T~~@ T~~@ T~. It is known from the classical invariant theory (cf.
[3], ch. VIII) that a set composed of a number of tensors possesses a finite
integrity basis of polynomial invariants and there exists an algorithm for
finding such a set. However this algorithm involves consideration of certain
diagrams of increasing complexity. Moreover, as the number and the ranks of
the tensors forming the set increase the number of diagrams and relations
between them to be considered grows very fast. We can show that even in our
case of the triplet (a, b,D) which is relatively simple the execution of the
algorithm by hand is prohibitively long. This may be the reason why an
integrity basis of polynomial invariants of (a, b, D) seems not to be available in
the literature. On the other hand we are convinced that the problem can now
be solved with the help of a computer.
Nevertheless an integrity basis for invariants (a, b,D) is likely to be large
because of the presence of exceptional cases where a, b and D have various
kinds of symmetries. Therefore it is reasonable to look for a set of polynomial
invariants which would determine "non-exceptional" or "generic" triplets
(a, b, D) uniquely up to rotations. Indeed we shall show below that this more
modest problem can be solved without any calculations and the resulting set
consists of 39 invariants.
To define rigorously the notion of "genericity" we need the second rank
symmetric traceless tensor d~ given by (d~)ij = Dok,DkpqrD~pqr; recall also that
(d2)ij = DipqrDjpqr.
DEFINITION. A tensor D ~ T~ is called generic if d2, ds do not have a common
principal axis.

Polynomial invariants

105

To justify the use of the word "generic" in the above definition we should
show that the set of all non-generic D's is "small". For this purpose we prove
first (see Appendix 2) that this set is defined by a system of 6 polynomial
equations each of degree 15 in components of D. A well known result in
algebraic geometry implies that the set of solutions of this system can be either
the whole space T~' (this means that the system holds for any D E T~) or some
algebraic geometry implies that the set of solutions of this system can be either
computer) an example D which does not satisfy this system (and hence is
generic). This in turn implies that the non-generic set is an algebraic subvariety
of dimension at most 8 (in fact, we believe the dimension is 7) in the
9-dimensional space T~. From this it follows that the set of non-generic D's
has measure zero. Informally speaking, this means that the probability of a
randomly chosen D ~ T~i~ being generic is 1.
We hasten to note that many interesting examples are non-generic in our
sense: for example, if D has any non-trivial symmetries (i.e., gD = D for some
g SO(3), g # I) then d 2 and d, inherit these symmetries and this means that
they have a common principal axis; therefore D is not generic.
We will also call C e T~ generic if the corresponding D is generic.
Now we can give the following definition:
DEFINITION. A finite set p~,..., p, of invariants of T~, is called a weak
functional basis if

p,(Cl) = p,(C2) for all i = 1. . . . . n

implies C~

= gC 2

for some g~SO(3) provided that Cl, C 2 are generic.

Once more, we note that a weak functional basis of invariants of T~, can be
obtained by adding )~,~ to a weak functional basis of invariants of the triplet
(a, b, D). In constructing such a basis for (a, b, D) we prove first the following
theorem:
T H E O R E M 2. A generic D is uniquely determined by d2, ds, J3, J8, J9, Jlo"
Proof. First observe that having d 2 and d~ we can calculate the invariants J2,
J4, Js, J6, J7" This implies that the knowledge of d2, ds, J3, J8, J9, Jlo provides
us with all the invariants of D and therefore determines D uniquely up to
rotations; in other words, the data set of the theorem gives us the orbit
O(D)(= {gD, g~SO(3)}) of D.
Now we prove that two distinct points on the orbit D,D' = 9D give rise to
distinct pairs (d2,d~). Indeed suppose that d2(D) = d2(D' ). Hence d2(D ) =
d2(gD ) = gd2(D). But one easily sees that d 2 = 9d 2 is possible only if g is a
rotation about a principal axis of d2(D) (of 180 degrees if eigenvalues of d 2 are

106

J. P. Boehler et al.

distinct) or g = I. The same argument applies to d, if ds(D' ) = d,(D). But by

definition D is generic implies that d 2 and d S do not have a common principal
axis, therefore d2(D' ) = d2(D), ds(D') = ds(D) is possible only when g = I, or
equivalently when D ' = D. Therefore there is a one-to-one correspondence
between the orbit of D and (d2, ds) for generic D. (Incidentally, for generic D
both of these orbits are isomorphic to SO(3)).
Thus we see that the pair dE, d, determines the position of a (generic) D on
its orbit.
[]
It may be useful to give an example when the data set of the theorem does
not determine D uniquely. Such an example is given by any tensor D ~ T~
which has the symmetries of a cube (i.e., gD = D if and only if g ~ $4, the group
of symmetries of a cube). It is easily seen from the symmetry considerations
that in this example d~ = 0 and d 2 = 2I, where 2 is an invariant of D. We see
that the orbit of (d 2, d~) is a single point and hence (d2, d~) gives no information
about the position of D on its orbit. Of course in this example D is not generic,
so there is no contradiction with our theorem.
Now recalling the decomposition of T~ given in Section 1, we can formulate
the main result of this paper:
T H E O R E M 3. A weak functional basis of invariants of C is given by the
following list:

J2,.--, Jlo
tr a 2, tr a a, tr b 2, tr b 3
tr ab, tr a2b, tr ab 2, tr a2b 2 and similar invariants
for each of the pairs (a, d2), (b, d2), (a, ds), (b ds),
tr abd2, tr abds, tr ad2ds, tr bd2d ,
Proof. First, a weak functional basis of invariants of C is obtained by adding
2, # to a weak functional basis of invariants of (a, b, D). Next from the previous
theorem it follows that a weak functional basis of invariants of (a, b, D) can be
obtained by adding J3, Js, J9, JlO to the functional basis of invariants of the
quadruple (a, b, d2, ds). For the latter quadruple the functional basis is known
(cf. [7]): it is composed by the traces
tr x, tr x 2, tr x3;

tr xy, tr x2y, tr xy 2, tr x2y2;

where x, y, z run over the set {a, b, d2, d~}.

tr xyz

Polynomial invariants

107

The functional basis of invariants of C thus obtained contains J3, ds, Jg, dlo
plus 9 invariants of the pair (d2, d~). But these 13 invariants can be expressed
as polynomials of the generating invariants J2 . . . . . Jlo by Theorem 1. With
this observation we arrive at the set of 39 invariants given in the statement of
the theorem.
[]
Note that this set does not contain any invariant of degree higher than 10.
REMARK. The set ofinvariants given in the theorem is not a functional basis
but only a weak functional basis. For example, as remarked before, in the case
where D has symmetries of a cube the pair (d2,d,) does not give any
information about the orientation of D and therefore the invariants of the
theorem which were constructed for the set (a, b, d2,d ~) do not allow us to
recover (a, b, D) if a, b are non-zero tensors.

Appendix 1: Reduction to the Classical Case

We now show that the problem of finding polynomial invariants of irreducible
representation T~ of the group SO(3) is equivalent to a similar problem for the
group SL(2,C). Indeed, we will show below that the algebra of polynomial
invariants of T~~ is isomorphic to the algebra of polynomial invariants of the
space V8 of homogeneous polynomials of degree 8 under the action of SL(2, C).
On the other hand, structure of the latter algebra was studied thoroughly by
Sylvester and von Gall ([4], [5]). They found that
(i) The algebra of invariants is generated by 9 invariants of degrees 2, 3. . . . .
10 respectively.
(ii) They gave explicit expressions for these invariants.
Thus in view of the above mentioned isomorphism a minimal functional
basis of polynomial invariants of D e T~* consists of 9 invariants of degrees
2,3,..., 10.
To give the explicit formulas for these invariants we need one more notion
from the classical invariant theory. Namely if we are given two vectors a e V2,,
be V2m where Vzk stands for the representation of SL(2, C) in the space of
homogeneous polynomials of degree 2k (see below) then there is a canonical
way to construct a new vector c = (a, b)ie V2~m+,_0. These operations are
usually called transvectants (German Uberschiebung). In modern terms they
can be defined with the help of the following theorem:
THEOREM. (Clebsch-Gordon) For each triplet of non-negative integers m, n, i
such that 0 <~ i <~m + n there exists a unique up to a constant factor linear map

108

J. P. Boehler et al.

commuting with the action of S0(3).

Now we can easily give the expressions for all the generating invariants of
Sylvester and yon Gall. For example, the invariant of degree 8 looks as follows:
iAk(f) = (((f, f)4, k)4, A)4,

where f ~ Vs, k = (f, f)6, A = (k, k) 2. One can easily check that i~.~ V0 = R;
from the definition of the operations (a, b)i it follows that ia,k is an invariant.
To find the formulas for corresponding invariants of D ~ T~ we note that for
even i we have an analog of the operations (a, b)i for the tensors: given
a = ai,...~,, ~ T~, b = bj,...j, ~ T~,s there is a canonical way to construct a tensor
$$
c = (a,b)Zt~ Tn+m-2t,
in terms of components c can be written as follows:
Cil...i._dr.j,_' = {ait...i._,p~...pbj,..j,_,pr..p,} (here curly brackets stand for the symmetric and traceless part of the tensor they bracket); in particular,
(D, D)4 = {D2}. Thus it is clear that the generating invariant of degree 8 of a
tensor D ~ T~ is given by

iAk ~ {A}ij{D2}ijkl{d2}kl,
where A = {d2} 2. We have used this expression in Section 2.
We do not give here the expressions for the invariants of other degrees: they
can be found in [5].
We now return, as promised above, to the relationship between the representations of the groups S0(3) and SL(2, C). We first note that there exists an
embedding S0(3)--, SL(2,C)/+ 1, given by the well-known isomorphism
S0(3) ~- SU(2)/ +_ 1. In some sense, SL(2, C ) / _ 1 can be considered as the
complexification of S0(3). This can be defined rigorously using the notion of
Lie algebras. In particular, we have the following lemma (cf. [1]):
LEMMA. Any action of S0(3) in a real space T can be uniquely extended by
complex analyticity to the action of SL(2, C)/_+ 1 (and hence, SL(2, C)) in the
space T ~ iT. If T is irreducible, so is T @ iT.
In particular, we can apply this lemma to our representation T~s. It is known
(cf. [1]) that this representation is irreducible. Then we obtain an irreducible
representation T~ ~ iT~s of SL(2, C) of dimension 2n + 1 over C. But we know
that there is a unique (up to isomorphism) irreducible representation of
SL(2, C) of dimension 2n + 1. This representation can be realized as the action
of SL(2, C) in the space V2, of all homogeneous polynomials with complex
coefficients of degree 2n in two variables. The action of SL(2, C) in this space
is given by the formula af(x,y) = f(x',y') where x' = g l t x + g12Y, Y'
----- g21 x -1- g22Y" Thus, as representations of SL(2, C), T~s ~9 iT~~ ~- V2,.
We next consider the algebras of all polynomial complex-valued functions

Polynomial invariants

109

on T,~s and on V2n, which we denote by S(T~ ~) and S(V2n) respectively. One
easily sees that since V2n "-~ T~~ @ tTn
" ~ the algebras of polynomial functions are
isomorphic: S(T~) ~- S(V2, ).
Now the following theorem appears naturally (once more, we refer to [1] for
the proof of this theorem)
T H E O R E M . Let us denote by S(Vzn) sL~2"c~ the algebra of those polynomials
from S(V2n) which are invariant under the action of SL(2, C), and by S(TSS)sOl3)
the algebra of those polynomials from S(T~ ~) which are invariant under the action
of SO(3). Then these algebras are isomorphic:
S( V2n)s142, cJ ~ S( TSS)sO13~.
This theorem allowed us to borrow results from the classical invariant theory.

Appendix 2: Genericity
In this appendix we show that the set of all non-generic D's can be described
by a system of polynomial equations. To do this we first need a criterion for
two symmetric 3 3 matrices a, b to have a common principal axis which
could be expressed in terms of polynomials in components of a, b. We produce
such a criterion as follows: first we take the commutator ab-ba. Next we can
associate to this matrix a vector v in a standard way: v~ =e~k(abba)ik = e~jkajtbtk. NOW we have the following lemma:
LEMMA. Let a, b be symmetric 3 3 matrices with real entries. Then a, b have
a common principal axis if and only if the vector o defined above is a common
eigenvector for a and b or zero (in other words, av = )w, by = pv for some 2,

/~ER).
Proof of this lemma is simple. The only subtlety arises when a or b have
multiple eigenvalues but one easily discovers that the lemma remains valid.
Next one observes that the condition av = )w for some 2 6 R is equivalent to
the following system:
(av) ~vz = v~(av)z,
(av) ,vs = v~(av)~,
(av) zv~ = vz(av) ~,

where (av) i stands for the i-th component of the vector av. This can be rewritten
as ~ijk(aV)~Vk = O. Similarly, by = #v ~ eijk(bV)jVR = O. Substituting ~kpra~sbs, for
Vk, we obtain that a and b have a common principal axis if and only if the

110

J. P. Boehler et al.

following system of polynomial equations holds:

eijkemnZekp,ajmantapsbabs, = 0,

i = 1,2,3.
tijkemlekprbjmbabsra~taps = 0

This is a system of 6 polynomial equations of degree 5 in the components of

a and b.
Now substituting d2 and ds for a and b we obtain that d2 and dS have a
common principal axis if and only if a system of 6 polynomial equations of

degree 15 in Dijkl holds.

Acknowledgement
This work was sponsored by the National Science Foundation under a
Material Research Group Award (DMR-9001378).

References
1. H. Weyl, The Classical Groups, their Invariants and Representations. Princeton Univ. Press
(1946).
2. E.T. Onat, Effective properties of elastic materials that contain penny shaped voids, Int. J. Eng.
Sci. 22 (1984) 1013-I021.
3. J. H. Grace and A. Young, The Algebra of lnvariants, Cambridge Univ. Press (1903) (reprinted
by Chelsea Publ. Co. in 1965).
4. J. Sylvester, Tables of the generating functions and .qroundformsfi;r the binary quantics of the first
ten orders, Collected mathematical papers of J. J. Sylvester, vol. Ill, Cambridge Univ. Press
(1909) pp. 283-311.
J. Sylvester, Sur les covariants irr~ductibles du quantic binaire de huiti~me order, ibid., 481-488.
5. yon Gall, Das vollstdndige Formensystem einer biniiren Form achter Ordnung. Matematische
Annalen 17 0880) 31-51, 139-152, 456.
6. Shioda Tetsuji, On the graded ring of the invariants of binary octavic. Am. J. Math 89 0967)
1022-1046.
7. J. P. Boehler, On irreducible representations for isotropic scalar functions. Zeitschrift fiir
An.qewandte Mathematik und Mechanik 57 (1977) 323-327.