Journal of the Mechanics and Physics of Solids

49 (2001) 2471 – 2492

www.elsevier.com/locate/jmps

A new proof that the number of linear elastic

symmetries is eight
Peter Chadwicka; ∗ , Maurizio Vianellob , Stephen C. Cowinc
a Schoolof Mathematics, University of East Anglia, Norwich NR4 7TJ, UK
b Dipartimento di Matematica, Politecnico di Milano, Piazza Leonardo da Vinci 32, 20133 Milano, Italy
c The Center for Biomedical Engineering and The Department of Mechanical Engineering, The School of

Engineering of the City College and The Graduate School of the City University of
New York, New York, NY10031, USA

Abstract
It is shown here that there are exactly eight di.erent sets of symmetry planes that are admissi-
ble for an elasticity tensor. Each set can be seen as the generator of an associated group character-
izing one of the traditional symmetry classes. ? 2001 Elsevier Science Ltd. All rights reserved.

Keywords: B. Anisotropic material; B. Elastic material; Classi4cation of symmetries

1. Introduction

In this paper we classify the types of material symmetry which arise in linear
anisotropic elasticity and determine the associated limitations on the elasticity tensor
c. The treatment of material symmetries in anisotropic elasticity has traditionally been
based on crystallographic considerations (Voigt, 1910; Nye, 1957; Smith and Rivlin,
1958; Hearmon, 1961; Fedorov, 1968), and the dominance of the crystallographic view-
point was total until quite recently when the “textured” elastic anisotropic symmetries
began to receive attention. (For references to the literature and a discussion of this de-
velopment see Forte and Vianello (1996, Section 6) and Cowin and Mehrabadi (1989)).
A modern approach which makes the treatment of symmetry in classical anisotropic
elasticity self-contained and independent of crystallography was introduced by Huo and
Del Piero (1991) and by Forte and Vianello (1996) who investigated, albeit using dif-
ferent de4nitions, the sets of symmetry groups which are possible for elasticity tensors

∗Corresponding author’s present address; 8 Stratford Crescent, Cringleford, Norwich NR4 7SF, UK. Tel.:
+44-1603-451655.
E-mail address: [email protected] (P. Chadwick).

0022-5096/01/$ - see front matter ? 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 2 2 - 5 0 9 6 ( 0 1 ) 0 0 0 6 4 - 3
2472 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492

without further restrictions. Here we follow a third and apparently di.erent route, 4rst
explored by Cowin and Mehrabadi (1987), in which elasticity tensors are classi4ed
according to the set of symmetry planes which they admit. We show that elastic ma-
terials fall into eight classes, each class being uniquely characterized by a precise set
of planes of mirror symmetry for the given elasticity tensor. Further, we prove that the
eight classes can be put into correspondence with the eight classical symmetry groups
as described by Forte and Vianello (1996). An account of elastic symmetry similar in
some respects to that given in Sections 2–7 below has been presented by Ting (1996,
Sections 2:5, 2:6). In Ting’s treatment it is stated, however, not proved that the number
of linear elastic symmetries is eight.
An objective of this paper is to reconcile the “symmetry planes” approach with the
“symmetry groups” approach as these two di.erent points of view presently coexist
in the literature. In the penultimate section, we show that a classi4cation of elasticity
tensors based on the simpler concepts used in the former approach leads to the same
result as a classi4cation based on the more complex notions involved in the latter. We
point out, however, that this agreement may be coincidental since in a more general
situation (concerning, for example, tensors with fewer index symmetries) the “symmetry
plane” classi4cation should naturally lead to fewer classes than the “symmetry group”
alternative.
This work is basically an e.ort of the 4rst author (Chadwick, 1995, 1997). The
second and third authors independently checked the results and prepared the work for
publication in the present form. All the equations in Sections 2–7 are taken from Chad-
wick (1995) and those in Section 8 from Chadwick (1997). Only minor modi4cations
to the original text have been made, the changes being mainly in this Introduction and
the last two sections.

2. Notations

2.1. Hooke’s Law

If, in relation to some orthonormal basis, the components of the stress and strain
tensors are ij and ij , respectively, Hooke’s law takes the form ij = cijrs rs where cijkl
are the components of the fourth-order elasticity tensor c. Here, and henceforth, all
lower case Latin subscripts take the values 1, 2, 3, and summation over these values is
implicit whenever a repeated suLx occurs. The components of c satisfy the symmetry
relations cijkl = cjikl = cijlk = cklij , which arise from the symmetry of the stress and
strain tensors and the requirement that no net work be done by an elastic material in
a closed loading cycle.

2.2. Symmetry transformations

The starting point for an analysis of the inherent symmetry of elastic materials is
the notion of a symmetry transformation. This is the name given to a transformation
e → e
between orthonormal bases e = {e1 ; e2 ; e3 } and e
= {e1
; e2
; e3
} under which
 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492 2473

the components of c are invariant. The transformation is represented by the orthogonal

matrix ‘ = [‘ij ] where
‘ij = ei
· ej : (1)
Let A be an arbitrary fourth-order tensor with components Aijkl relative to e, and T an
arbitrary second-order tensor with components Tij . Then A[T] denotes the second-order
tensor with components Aijrs Trs . In this notation the components of c relative to e are
cijkl = tr{(ei ⊗ ej )c[ek ⊗ el ]} (2)


and e → e is a symmetry transformation of the material with linear elasticity tensor c
if
tr{(ei ⊗ ej )c[ek ⊗ el ]} = tr{(ei
⊗ ej
)c[ek
⊗ el
]}: (3)
From Eq. (1),
ei
= ‘ip ep : (4)
In view of Eq. (2), we can therefore express condition (3) in the alternative form
cijkl = ‘ip ‘jq ‘kr ‘ls cpqrs : (5)
It is a simple matter to verify from Eq. (5) that if ‘ and m represent symmetry
transformations, so do ‘−1 and ‘m. The set G of the matrix representations of all
symmetry transformations of an elastic material is therefore a linear group; G is called
the symmetry group of the material.

2.3. Re;ections

Let U denote the set of all unit vectors. If a ∈ U has components ai relative to an
arbitrary orthonormal basis e, the orthogonal matrix R(a) with elements
Rij (a) = ij − 2ai aj (6)
transforms coordinates xi relative to e into xi − 2(ap xp )ai and thus brings about
reOection of the points of R3 in the plane P(a) orthogonal to a. Accordingly, a symme-
try transformation e → e
represented by R(a) is called a re;ection and P(a) a plane
of symmetry. Since R(−a) = R(a), there is no need to distinguish between the normal
to P(a) and a unit vector directed along it. From now on “normal” means “normal to
a plane of symmetry” and “the normal a” means “the normal spanned by a ∈ U”.

2.4. The reduced su<x notation

In this convenient scheme a single index is assigned to the 4rst two and the last
two subscripts of cijkl according to the rule
11 → 1; 22 → 2; 33 → 3; 23 → 4; 31 → 5; 12 → 6: (7)
Thus c1123 = c14 ; c3112 = c56 . On account of the symmetry relations cijkl = cjikl = cijlk
=cklij , the pairs 32, 13 and 21 have the same indices as 23, 31 and 12, respectively,
2474 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492

and the two suLxes in the reduced notation are interchangeable; we take the 4rst to
be the smaller whenever they are unequal. With this convention, and relative to an
arbitrary orthonormal basis, the 21 distinct components of c can be set out as a 6 × 6
symmetric matrix:
 
c11 c12 c13 c14 c15 c16
 · c22 c23 c24 c25 c26 
 
 · · c33 c34 c35 c36 
c=  ; (8)

 · · · c44 c45 c46 
 · · · · c55 c56 
· · · · · c66
the missing entries being implied by the symmetry of c.

3. Triclinic materials

It is obvious from Eq. (5) that the identity matrix I = [ij ] and its opposite −I
always represent symmetry transformations. Thus {I; −I } is invariably a subgroup of
G. An elastic material for which
G = {I; −I } (9)
is said to be triclinic. In the case of a triclinic elastic material, Eq. (5) places no
restriction on the components cijkl additional to the symmetry relations cijkl =cjikl =cijlk =
cklij . Consequently, c has 21 distinct components in relation to e and they are displayed
in matrix (8). However, e is an arbitrary orthonormal basis and, since an orthogonal
matrix can be generated from three numbers, Euler angles for example, a basis can,
in principle, be found in which the number of distinct components of c is reduced
by three to 18. For a triclinic elastic material, 18 of the 21 distinct components of c
are thus essential and an orthonormal basis in which there are 18 distinct components
is intrinsic to the material (see Fedorov (1968, Section 19) for a speci4c choice and
Cowin and Mehrabadi (1995, Section 9) for further discussion). Notwithstanding this
reduction, a triclinic elastic material has no symmetry transformation other than I and
−I , and therefore no inherent symmetry.

4. Monoclinic materials

An elastic material for which G contains one and only one reOection is said to be
monoclinic. Suppose that the basis e is chosen so that the reOection is R(e2 ); this
means that the normal e2 is intrinsic to the material. Then Eqs. (4) and (6) give
e1
= e1 ; e2
= −e2 ; e3
= e3 , and it follows from Eqs. (3) and (2) that the components
of c in which either one or three suLxes are equal to 2 are zero. In reduced suLx
notation,
c14 = c16 = c24 = c26 = c34 = c36 = c45 = c56 = 0: (10)
 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492 2475

Similarly, for a monoclinic elastic material in which the normal is spanned by e3 ,

c14 = c15 = c24 = c25 = c34 = c35 = c46 = c56 = 0 (11)
and matrix (8) simpli4es to
 
c11 c12 c13 0 0 c16
 · c22 c23 0 0 c26 
 
 · · c33 0 0 c36 
c=  ·
: (12)
 · · c44 c45 0 
 · · · · c55 0 
· · · · · c66
It is seen from Eq. (12) that, for a monoclinic elastic material, c has 13 distinct
components relative to a basis e with one member directed along the normal. A rotation
of e about the normal has no e.ect on the number of distinct components of c, and, by
a suitable choice of the angle of rotation, one further component can be made to vanish
(see Fedorov (1968, Section 19), Cowin (1995, Section 3) and Cowin and Mehrabadi
(1995, Section 10) for additional details). Thus only 12 of the distinct components of
c are essential and a basis in which there are 12 distinct components is intrinsic.

5. Conditions satised by a normal

In the study of symmetry groups containing more than one reOection, it is useful to
have a means of recognizing when a given unit vector spans the normal to a plane
of symmetry. To this end we consider three second-order tensors derived from the
linear elasticity tensor c. If a ∈ U, the acoustical tensor Q(a) is de4ned, through its
components relative to e, by
Qij (a) = cpirj ap ar : (13)
This tensor is of central importance in the theory of elastic wave propagation. The
other tensors, C and V , are de4ned componentwise by
Cij = cijrr ; Vij = cirjr : (14)
We deduce from de4nition (13) that
Q(e3 )e3 = c3q33 eq = c35 e1 + c34 e2 + c33 e3 (15)
and
Q(cos e1 + sin e2 )e3 = cpqr3 (cos p1 + sin p2 )(cos r1 + sin r2 )eq

= {c46 sin2  + (c14 + c56 ) sin  cos  + c15 cos2 }e1

+ {c24 sin2  + (c25 + c46 ) sin  cos  + c56 cos2 }e2

+ (c44 sin2  + 2c45 sin  cos  + c55 cos2 )e3 : (16)

2476 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492

If e3 spans a normal, Eqs. (11) hold and inspection of Eqs. (15) and (16) shows that e3
is an eigenvector of Q(e3 ) and of Q(cos e1 + sin e2 ) for all values of . Conversely,
if e3 has these properties, Eqs. (11) are a consequence of Eqs. (15) and (16). We have
therefore established the following
Test: a ∈ U spans a normal if and only if a is an eigenvector of Q(a) and Q(b)
for all b ∈ U orthogonal to a.
From de4nitions (14),
Ce3 = cp3rr ep = (c15 + c25 + c35 )e1 + (c14 + c24 + c34 )e2 + (c13 + c23 + c33 )e3 ;
(17a)

Ve3 = cpr3r ep = (c15 + c35 + c46 )e1 + (c24 + c34 + c56 )e2 + (c33 + c44 + c55 )e3 :
(17b)
It is immediately clear from Eqs. (11) that when e3 spans a normal it is an eigenvector
of both C and V . We thus have
The CV condition: If a ∈ U spans a normal, a is an eigenvector of C and V .
The above results, due to Cowin and Mehrabadi (1987) and Cowin (1989) (see
also Norris (1989) and Hayes and Norris (1992)), go somewhat beyond our present
needs by providing a method of 4nding all the normals possessed by an elastic mate-
rial when the components of c are known in relation to a speci4ed basis. The CV
Condition supplies a set of candidates, namely the unit vectors which are simul-
taneous eigenvectors of C and V , and from these the Test singles out the actual
normals.
We conclude these preliminaries to our discussion of the higher symmetries by in-
troducing the notation
ri () = sin ej + cos ek ; (18)
{i; j; k} being a cyclic permutation of {1; 2; 3}. The identities
R(ej )R(ri ())R(ej ) = R(ek )R(ri ())R(ek ) = R(ri ( − )) (19)
and
R(ri ())R(ri ( + 12 )) = −R(ei ) (20)
follow from de4nitions (6) and (18).

6. Tetragonal, trigonal, orthotropic and transversely isotropic materials

6.1. Restrictions arising from the presence of two normals

We turn now to the situation in which the symmetry group G contains two reOections
with normals enclosing an angle . We choose bases e and e
with e2 and e2
spanning
the normals and
e1
= r3 ( + 12 ); e2
= r3 (); e3
= e3 : (21)
 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492 2477

The presence of the normal e2 gives rise to Eqs. (10), and the eight components









c14 ; c16 ; c24 ; c26 ; c34 ; c36 ; c45 ; c56

of c relative to e
are likewise zero due to the existence of the normal e2
. With
reference to Eqs. (2), (21) and (18), and taking account of Eqs. (10), we have


c14 = tr{(e1
⊗ e1
)c[e2
⊗ e3
]} = {c25 sin2  + (c15 − 2c46 ) cos2 } sin ; (22)

c16 = tr{(e1
⊗ e1
)c[e1
⊗ e2
]}

= −{(c22 − c12 − 2c66 ) sin2  − (c11 − c12 − 2c66 ) cos2 } sin  cos ; (23)

c24 = tr{(e2
⊗ e2
)c[e2
⊗ e3
]} = {c15 sin2  + (c25 + 2c46 ) cos2 } sin ; (24)

c26 = tr{(e2
⊗ e2
)c[e1
⊗ e2
]}

= {(c11 − c12 − 2c66 ) sin2  − (c22 − c12 − 2c66 ) cos2 } sin  cos ; (25)

c34 = tr{(e3
⊗ e3
)c[e2
⊗ e3
]} = c35 sin ; (26)

c36 = tr{(e3
⊗ e3
)c[e1
⊗ e2
]} = (c13 − c23 ) sin  cos ; (27)

c45 = tr{(e2
⊗ e3
)c[e1
⊗ e3
]} = −(c44 − c55 ) sin  cos; (28)

c56 = tr{(e1
⊗ e3
)c[e1
⊗ e2
]} = {c46 sin2  + (c15 − c25 − c46 ) cos2 } sin : (29)

Since the normals are distinct, sin  = 0 and Eqs. (22) – (29) lead to the conditions

c35 = 0; (30)

(c13 − c23 ) cos  = 0; (c44 − c55 ) cos  = 0; (31)

{(c11 − c12 − 2c66 ) sin2  − (c22 − c12 − 2c66 ) cos2 } cos  = 0; (32a)

{(c22 − c12 − 2c66 ) sin2  − (c11 − c12 − 2c66 ) cos2 } cos  = 0; (32b)

c15 sin2  + (c25 + 2c46 ) cos2  = 0; (33a)

c25 sin2  + (c15 − 2c46 ) cos2  = 0; (33b)

c46 sin2  + (c15 − c25 − c46 ) cos2  = 0: (33c)

2478 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492

The alternatives o.ered by Eqs. (31) – (33) are

(i)  = 12 ; (ii)  = 12 ; c13 = c23 ; c44 = c55 [Eq. (31)],
(i)  = 12 ; (ii)  = 14  or 34 ; c11 = c22
(iii)  = 14 ; 12 ; 34 ; c11 = c22 ; c66 = 12 (c11 − c12 ). [Eqs. (32)],
(i)  = 13  or 23 ; c25 = c46 = −c15 ; (ii)  = 13 ; 23 ; c15 = c25 = c46 = 0 [Eqs. (33)].
Overall, therefore, Eqs. (30) – (33) have the following solutions.
 = 14  or 34 ; c11 = c22 ; c13 = c23 ; c44 = c55 ;

c15 = c25 = c35 = c46 = 0: (34)

 = 13  or 23 ; c11 = c22 ; c13 = c23 ; c44 = c55 ; c66 = 12 (c11 − c12 );

c35 = 0; c25 = c46 = −c15 : (35)

 = 12 ; c15 = c25 = c35 = c46 = 0: (36)

 = { 14 ; 13 ; 12 ; 23 ; 34 };

c11 = c22 ; c13 = c23 ; c44 = c55 ; c66 = 12 (c11 − c12 );

c15 = c25 = c35 = c46 = 0: (37)

In each case Eqs. (10) also apply. We discuss these four possibilities in turn.

6.2. Tetragonal elastic materials

When G contains reOections with normals e2 and r3 ( 14 ), or r3 ( 43 ), the second of
identities (19), with i = 3; k = 2 and  = 14 , or 34 , implies that the reOection with
normal r3 ( 34 ), or r3 ( 14 ), also belongs to G. Remembering that −I ∈ G, we then
deduce from (20), with i = 3;  = 14  and i = 1;  = 0, that the reOections with normals
e3 and e1 are in G. In this connection, we notice that Eqs. (10) and (34) imply Eqs.
(11) and that the conditions on the c in Eqs. (34) include those in Eqs. (36). In
addition to the pair of reOections assumed initially, the material thus has three induced
reOections.
We 4nd from Eqs. (14), (10) and (34) that, relative to e,
Cij = diag(c11 + c12 + c13 ; c11 + c12 + c13 ; 2c13 + c33 ); (38a)

Vij = diag(c11 + c44 + c66 ; c11 + c44 + c66 ; c33 + 2c44 ): (38b)
The simultaneous unit eigenvectors of C and V are e3 and every unit vector or-
thogonal to e3 , and, by the CV Condition, this subset of U contains all the normals.
 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492 2479

The procedure described in Section 6.1 yields all the normals orthogonal to e3 , so the
4ve reOections speci4ed above are the only ones.
An elastic material with 4ve normals, four of them coplanar and spaced at angles
of 14  and the 4fth orthogonal to each of the others, is said to be tetragonal. In the
preceding discussion the set of normals is
N = {r3 (0) = e2 ; r3 ( 14 ); r3 ( 12 ) = e1 ; r3 ( 34 ); e3 } (39)
and the associated restrictions (10) and (34) on the components of c reduce matrix
(8) to
 
c11 c12 c13 0 0 0
 : c11 c13 0 0 0 
 
 : : c33 0 0 0 
c=
 : :
: (40)
 : c44 0 0 
 : : : : c44 0 
: : : : : c66
The reOections with normals in N form a subset, but not a subgroup, of G.
The inclusion of all the base vectors of e in N means that e is an intrinsic basis
and the six distinct components of c in Eq. (40) are all essential. The components of
c relative to an intrinsic basis are referred to hereafter as elastic moduli.

6.3. Trigonal elastic materials

When G contains reOections with normals e2 and r3 ( 13 ), or r3 ( 32 ), the second of
identities (19), with i = 3; k = 2 and  = 13 , or 23 , induces a third reOection, with
normal r3 ( 23 ), or r3 ( 13 ). Since no two of these normals are orthogonal, identity (20)
is inapplicable.
As in Section 6.2, the CV Condition con4nes the possible normals to e3 and the
unit vectors orthogonal to e3 , and all the normals orthogonal to e3 have been found.
Returning to Eq. (16), we see that restrictions (10) and (35) do not ensure that e3 is
an eigenvector of Q(cos e1 + sin e2 ) for all values of . The Test thus disquali4es
e3 and
N = {r3 (0) = e2 ; r3 ( 13 ); r3 ( 23 )} (41)
is the complete set of normals.
An elastic material with three coplanar normals, spaced at angles of 13 , is said to
be trigonal. When the normals are given by Eq. (41), Eqs. (10) and (35) apply and
matrix (8) becomes
 
c11 c12 c13 0 c15 0
 : c11 c13 0 −c15 0 
 
 : : c33 0 0 0 
c= : (42)
: : : c 0 −c 
 44 15 
 : : : : c44 0 
: : : : : 12 (c11 − c12 )
2480 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492

Since the members of N determine e, this is an intrinsic basis and the six distinct
components of c in Eq. (42) are elastic moduli.

6.4. Orthotropic elastic materials

When G contains reOections with normals e2 and r3 ( 12 ) = e1 , identity (20), with

i = 3;  = 0; induces a third reOection with normal spanned by e3 . In con4rmation,
Eqs. (10) and (36) entail Eqs. (11). Relative to e,

Cij = diag(c11 + c12 + c13 ; c12 + c22 + c23 ; c13 + c23 + c33 ); (43a)

Vij = diag(c11 + c55 + c66 ; c22 + c44 + c66 ; c33 + c44 + c55 ) (43b)

and we infer that the simultaneous eigenvectors of C and V are e1 ; e2 ; e3 . The CV

Condition thus guarantees the completeness of the set of normals

N = {e1 ; e2 ; e3 }: (44)

An elastic material with three mutually orthogonal normals is said to be orthotropic.

When the normals are spanned by the members of e, Eqs. (10) and (36) hold and,
from Eq. (8),
 
c11 c12 c13 0 0 0
 : c22 c23 0 0 0 
 
 : : c33 0 0 0 
c=  : : : c44 0 0  :
 (45)
 
 : : : : c55 0 
: : : : : c66

In view of Eq. (44), the nine distinct components of c evidenced by Eq. (45) are
elastic moduli.

6.5. Transversely isotropic elastic materials

Restrictions (10) and (37) on the components of c reduce matrix (8) to

 
c11 c12 c13 0 0 0
 : c11 c13 0 0 0 
 
 : : c33 0 0 0 
c=  : (46)
: : : c 0 0 
 44 
 : : : : c44 0 
: : : : : 12 (c11 − c12 )
 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492 2481

It may readily be veri4ed that the formula

cijkl = c12 ij kl + 12 (c11 − c12 )(ik jl + il jk ) − (c12 − c13 )(ij k3 l3 + kl i3 j3 )

− 12 (c11 − c12 − 2c44 )(ik j3 l3 + il j3 k3 + jl i3 k3 + jk i3 l3 )

+ (c11 + c33 − 2c13 − 4c44 )i3 j3 k3 l3 (47)

reproduces Eq. (46), and there follows from Eqs. (13) and (47) the expression

Q(a) = 12 {c11 − c12 − (c11 − c12 − 2c44 )(e3 :a)2 }I + 12 (c11 + c12 )a ⊗ a

− 12 (c11 + c12 − 2c13 − 2c44 )(e3 :a)(e3 ⊗ a + a ⊗ e3 )

− { 12 (c11 − c12 − 2c44 ) − (c11 + c33 − 2c13 − 4c44 )(e3 :a)2 }e3 ⊗ e3 (48)

for the acoustical tensor. Matrices (40) and (46) di.er only in the 66 entry and
Eqs. (38a) and (38b) therefore hold in the present case with c66 = 12 (c11 − c12 ). Again,
the unit eigenvectors common to C and V are e3 and the unit vectors orthogonal to
e3 and, by the CV Condition, this set contains all the normals. Examination of Eq.
(48) shows that e3 is an eigenvector of Q(e3 ) and of Q(b) for all b ∈ U orthogonal
to e3 . Further, for any a ∈ U orthogonal to e3 , a is an eigenvector of Q(a) and of
Q(cos e3 + sin e3 ∧ a) for all . Thus, by the Test, e3 and every unit vector ortho-
gonal to e3 span normals. Although e is determined by the normals only to within an
arbitrary rotation about e3 , such a rotation has no e.ect on the components of c since
it is a symmetry transformation. The basis e is therefore intrinsic and the 4ve distinct
components of c in Eq. (46) are elastic moduli.
An elastic material for which a unit vector e and all the unit vectors orthogonal to
e span normals is said to be transversely isotropic. Both the set of normals and the
symmetry group of a transversely isotropic material are in4nite. The material is rota-
tionally symmetric about the axis of transverse isotropy spanned by e and reOectionally
symmetric about the basal plane P(e) and each of the zonal planes {P(d ): d · e = 0}.

7. Isotropic and cubic materials

7.1. Preliminary remarks

The analysis in Section 6 shows that the higher forms of symmetry proceeding from
the introduction of a second reOection into the symmetry group of a monoclinic elastic
material are of four types: tetragonal, trigonal, orthotropic and transversely isotropic.
For each type the insertion of a second normal induces further normals, there being
4ve in all for a tetragonal material and three for trigonal and orthotropic materials.
Most strikingly, when the angle enclosed by the 4rst and second normals does not
2482 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492

belong to the set

P = { 14 ; 13 ; 12 ; 23 ; 34 }; (49)
the outcome is transverse isotropy in which every unit vector coplanar with these
normals and a unit vector orthogonal to both of them span additional normals.
The next step in the investigation of material symmetry in linear anisotropic elasticity
is to determine the consequences of bringing a further normal into each of the higher
forms. It is convenient to begin with transverse isotropy.

7.2. Isotropic elastic materials

Suppose 4rst that an elastic material has two axes of transverse isotropy which are
orthogonal to one another. Let the basis e be chosen so that e3 and e1 span the two axes.
Then the matrix of elastic moduli is given by Eq. (46) and also by the modi4cation
of Eq. (46) produced by the change of indices
1 → 2; 2 → 3; 3 → 1; 4 → 5; 5 → 6; 6 → 4: (50)
Hence
c11 = c22 = c33 ; c12 = c13 = c23 ; c44 = c55 = c66 = 12 (c11 − c12 ) (51)
and the matrix of moduli is
 
c11 c12 c12 0 0 0
 : c11 c12 0 0 0 
 
 : : c11 0 0 0 
c= : (52)
: : : 1
(c − c ) 0 0 
 2 11 12 
 : : : : 1
(c − c ) 0 
2 11 12
1
: : : : : (c
2 11 − c 12 )
In view of Eqs. (51), formulae (47) and (48) reduce to
cijkl = c12 ij kl + 12 (c11 − c12 )(ik jl + il jk ) (53)
and
Q(a) = 12 (c11 − c12 )I + 12 (c11 + c12 )a ⊗ a: (54)
Inspection of Eq. (54) reveals that, for arbitrary a ∈ U, a is an eigenvector of Q(a)
and Q(b) for all b ∈ U orthogonal to a. By the Test, each member of U therefore
spans a normal and every plane is a plane of symmetry. The material has no intrinsic
directionality and is said to be isotropic. This is obviously the highest possible form
of symmetry. The two distinct moduli appearing in Eq. (52) are essential.
The requirement that the two axes of transverse isotropy be orthogonal can be re-
laxed. Suppose that unit vectors along the axes are e3 and r1 () (see Eq. (18)). If
 ∈ P, the plane containing the axes is a basal plane, as shown in Section 6.5, and
the normal e1 to this plane spans an axis of transverse isotropy. By the earlier result,
 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492 2483

the material is therefore isotropic. The argument remains valid when  = 13  or 23 ,

since the angle between r1 () and e2 , which, being orthogonal to e3 , also spans a
normal, is 16  ∈ P. If  = 14  or 34 , the angle between r1 () and 3−1=2 (e1 + 21=2 e2 );
also directed along a normal, is arccos 3−1=2 ∈ P. The plane spanned by these vectors
is hence a basal plane and the inclination to e3 of the associated axis of transverse
isotropy is 13  or 23 . Once again the material is isotropic and we have proved that an
elastic material with two distinct axes of transverse isotropy is isotropic.
The only non-trivial orientation of an additional normal in a transversely isotropic
elastic material is away from the axis of transverse isotropy and the basal plane. Let
n ∈ U span the new normal and e ∈ U the axis of transverse isotropy, and let " =
arccos(e:n). For all values of # the unit vector q=(1+#2 )−1=2 cosec "(n−cos "e+#e∧n)
is orthogonal to e and thus spans a normal. For any value of " in the interval (0; 12 ),
we can choose # so that arccos(n:q)=arccos{(1+#2 )−1=2 sin "} ∈ P. The span of n and
q is therefore a basal plane and, by the conclusion reached in the previous paragraph,
the material is isotropic.

7.3. Cubic elastic materials

Proceeding next to a tetragonal material, suppose that the set of normals (39) is
supplemented by a 4fth normal spanned by n ∈ U. If e3 :n = 0; n makes an angle not
in P with at least three of the other normals and the material becomes transversely
isotropic, with the axis of transverse isotropy spanned by e3 . If e3 :n = 0; ±1 and n is
not equal to any of the unit vectors
r1 ( 14 ); r1 ( 34 ); r2 ( 14 ); r2 ( 34 ) (55)
and
d1 = 12 (e1 + e2 ) + 2−1=2 e3 ; d2 = 12 (e1 + e2 ) − 2−1=2 e3 ; (56a)

d3 = 12 (e1 − e2 ) + 2−1=2 e3 ; d4 = 12 (e1 − e2 ) − 2−1=2 e3 ; (56b)

it may easily be shown, 4rst, that n is not orthogonal to any of the normals in Eq.
(39) and, second, that the inclinations of n to at least two of these normals are outside
P. The material is then isotropic, by the italicized result in Section 7.2.
If n = r1 ( 14 ); the angles between n and the members of N are 14 ; 13 ; 12 ; 23 ; 14 ;
all in P. To complete the tetragonal and trigonal systems implied by these angles, ad-
ditional normals, spanned by the remaining vectors in (55), are induced, as in Sections
6.2 and 6.3, giving the set
N = {e1 ; e2 ; e3 ; r1 ( 14 ); r1 ( 34 ); r2 ( 14 ); r2 ( 34 ); r3 ( 14 ); r3 ( 34 )}: (57)
Equating n to the second, third or fourth unit vector in (55) likewise leads to
Eq. (57).
An elastic material with nine normals, aligned with the edges and face diagonals of
a cube, is said to be cubic. It is evident from Eqs. (57) and (44) that a cubic material
is automatically orthotropic and that the elastic moduli of a cubic material are invariant
2484 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492

under the permutations of indices (50) and

1 → 3; 2 → 1; 3 → 2; 4 → 6; 5 → 4; 6 → 5: (58)
Matrix (45) hence becomes
 
c11 c12 c12 0 0 0
 : c11 c12 0 0 0 
 
 : : c11 0 0 0 
c= : : : c44 0
 (59)
 0 
 : : : : c44 0 
: : : : : c44
and the three distinct moduli displayed here are all essential.
In relation to Eq. (57), the normals e1 ; e2 ; e3 are called cube axes and the planes
of symmetry P(e1 ); P(e2 ); P(e3 ) cube faces. The other six normals are face diagonals
and the corresponding planes of symmetry face bisectors. A cubic material has four
octahedral planes, equally inclined to the cube axes and intersecting in the face diag-
onals. The three face diagonals in each octahedral plane form a trigonal system with
a spacing of 13 . The octahedral planes are not planes of symmetry.
We consider lastly the vectors in (56). If n = d1 , the angles between n and the
members of N given in Eq. (39) are 13 ; 14 ; 13 ; 12 ; 14 , again indicating tetragonal
and trigonal systems the completion of which introduces further normals. The induced
normals are found to be spanned by d2 ; d3 ; d4 , and the choices n = d2 , n = d3 , n = d4
similarly induce normals spanned by the other vectors in (56). The resulting set of
nine normals once more represents cubic symmetry, the cube axes now being de4ned
by r3 ( 14 ); r3 ( 34 ); e3 and the pairs of face diagonals mutually orthogonal with them by
d3 ; d4 ; d1 ; d2 ; e1 ; e2 , in turn.

7.4. Augmentation of N for a trigonal material

Continuing the programme initiated in the 4nal paragraph of Section 7.1, we consider
next the e.ect of adding a fourth normal, spanned by n∈U, to the set in Eq. (41). If
e3 :n = 0; the angles between the new and the pre-existing normals cannot be con4ned
to the set P de4ned by Eq. (49) and transverse isotropy results, with the axis of
transverse isotropy spanned by e3 . If n = ±e3 ; the Test requires e3 to be an eigenvector
of Q(cos e1 + sin e2 ) for all  and, from Eqs. (16), (10) and (35), this necessitates
the vanishing of c15 . The matrix of elastic moduli (Eq. (42)) then simpli4es to Eq.
(46) and again the material is transversely isotropic with the axis of transverse isotropy
spanned by e3 .
As pointed out in Section 7.3, the normals in Eq. (41) have the same conformation
as the face diagonals in an octahedral plane of a cubic elastic material. If, in relation
to this cubic system, n spans another face diagonal, two further octahedral planes each
contain n and one of the vectors in Eq. (41). In each of these planes the third face
diagonal is induced, as in Section 6.3, completing the full set of six. Furthermore, each
pair of othogonal face diagonals induces the cube axis orthogonal to them, as in Section
6.4, making up the nine normals which de4ne cubic symmetry. If n spans a cube axis,
 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492 2485

two cube faces each contain n and one of the vectors in Section Eq. (41), enclosing an
angle of 14 . In each of these planes another face diagonal and another cube axis are
induced, as in Section 6.2. The cube axis spanned by n and the face diagonal spanned
by the remaining vector in Eq. (41), orthogonal to n, induce a sixth face diagonal,
as in Section 6.4, and again we arrive at the set of nine normals characterizing cubic
symmetry.
Lastly, if e3 :n = 0; ±1 and n does not span a face diagonal or a cube axis, the angles
between n and at least two of the vectors in Eq. (41) are not in P. By the italicized
result in Section 7.2, the material is then isotropic.

7.5. Augmentation of N for an orthotropic material

When the set (44) is supplemented by a fourth normal, spanned by n ∈ U, the

following possibilities arise.
First, the normal n bisects two of the orthotropic normals. The second bisector is
induced, as in Section 6.2, and the material becomes tetragonal.
Second, the normal n is coplanar with two of the orthotropic normals, but does
not bisect them. A basal plane is induced, as in Section 6.5, and the material be-
comes transversely isotropic, the axis of transverse isotropy coinciding with the third
orthotropic normal.
Third, n is not coplanar with any pair of e1 ; e2 ; e3 , but makes angles 31 ; 13 ; 14  with
these base vectors, in some order. The remaining 4ve normals of a cubic system are
induced by applications of the procedures described in Sections 6.2 and 6.3. Suppose,
for example, that n = d1 , de4ned in (56a), so that the inclinations of n to e1 ; e2 ; e3 are
1 1 1
3 ; 3 ; 4 , respectively. This means that the pair d1 ; e3 originates a tetragonal system
and the pairs d1 ; e1 and d1 ; e2 trigonal systems, e3 spanning a cube axis and d1 ; e1 ; e2
face diagonals. The completion of the tetragonal system induces the remaining cube
axes, spanned by 2−1=2 (e1 ± e2 ) (i.e. r3 ( 14 ); r3 ( 34 )), and the face diagonal spanned
by d2 . The completion of the trigonal systems induces the last pair of face diagonals,
spanned by d3 and d4 (see (56b)), and we recover the cubic system described in the
4nal paragraph of Section 7.3.
Fourth, and last, n is not coplanar with any pair of e1 ; e2 ; e3 and neither n nor −n
is equal to one of d1 ; d2 ; d3 ; d4 . At least two of the angles arccos(ei :n) are outside P
and at least two basal planes are induced. The material becomes isotropic, as shown
in Section 7.2.

7.6. Concluding remarks

It has been found in Sections 7.3–7.5 that, for tetragonal, trigonal and orthotropic
elastic materials, the insertion of an additional normal has four possible consequences:
the material becomes tetragonal (if it was initially orthotropic), transversely isotropic,
cubic or isotropic. For transversely isotropic materials, the only possible change is
to isotropy, as proved in Section 7.2. The programme followed in this section has
thus produced two types of symmetry, isotropic and cubic, higher than the four types
evolved in Section 6.
2486 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492

Fig. 1. Generation of the eight types of linear elastic symmetry by the successive introduction of planes of
symmetry.

Can the process be continued? Since the normals of an isotropic elastic material
occupy the whole of U, no further addition is possible in this case and it remains only
to consider the introduction of a tenth normal, n ∈ U, into the cubic system (57).
There is no orientation of n for which at least four of the angles between n and the
vectors in Eq. (57) are not in P. Thus at least four basal planes are induced and the
outcome is isotropy.

8. Symmetry groups

It has been shown in Sections 4, 6 and 7 that when a reOection which is a symmetry
transformation is adjoined to {I; −I } and further reOections with this property are
added to the set, the process terminates after the introduction of a third reOection. A
diagrammatic representation of the process and its results is shown in Fig. 1. Exactly
eight symmetry classes are produced and a summary of their main properties is provided
in Table 1. We list there, for each class, the set of normals to the planes of symmetry
(or the equation number at which they may be found), the location of the matrix of
components of the linear elasticity tensor c, the number of distinct components of c,
the number of these components which are essential and thus qualify as elastic moduli
(see Sections 3, 4 and 6.2) and the number of planes of symmetry.
 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492 2487

Table 1
Numbers of distinct components of c, elastic moduli and planes of symmetry for the symmetry classes of
linear elastic materials

Class Normals to Matrix Distinct Elastic Planes of

the planes of c = [c ] c moduli symmetry
symmetry
Triclinic — (8) 21 18 0
Monoclinic e3 (12) 13 12 1
Trigonal (41) (42) 6 6 3
Orthotropic (44) (45) 9 9 3
Tetragonal (39) (40) 6 6 5
Cubic (57) (59) 3 3 9
Transversely isotropic e3 , all n⊥e3 (46) 5 5 ∞1 + 1
Isotropic all n ∈ U (52) 2 2 ∞3

Table 2
The sets of reOections

Class The associated reOection set

Triclinic S1 = {I; −I }
Monoclinic S2 = S1 ∪ {R(e2 )}
Tetragonal S3 = S2 ∪ {R(2−1=2 (e1 + e2 ))}
Trigonal S4 = S2 ∪ {R( 12 (31=2 e1 + e2 ))}
Orthotropic S5 = S2 ∪ {R(e1 )}
Transversely isotropic S6 = S2 ∪ {R(sin e1 + cos e2 ))};  ∈ P = { 14 ; 1
3
; 12 ; 23 ; 43 }
Cubic S7 = S3 ∪ {R(2−1=2 (e2 + e3 ))}
Isotropic S8 = S6 ∪ {R(n)}; e3 :n = 0; ±1

To each of the symmetry classes there corresponds a subset of the material symmetry
group G, de4ned in Section 2.2, which consists of reOections and is therefore also a
subset of the orthogonal group O(3). These subsets, denoted by S ;  = 1; : : : ; 8, are
speci4ed in Table 2. The numbers of reOections in the entries are unique, but the
normals are not. As 4rst noted in Section 3, −I as well as I always belongs to G,
simply because c is a tensor of even order. Similarly, an orthogonal matrix belongs to
G if and only if its opposite is also in G. For each symmetry class the symmetry group
is hence uniquely determined by the set of proper orthogonal matrices (i.e. rotations)
which belongs to it; in other words, by its intersection with SO(3), the group of
rotations (see Forte and Vianello, 1996, Section 2.1). Thus, without loss of generality,
we can de4ne the symmetry group as the set of all rotations satisfying Eq. (5).

8.1. The relationship of S to the symmetry group

A reOection is a symmetry transformation for c if and only if its opposite, the rotation
of  (or half-turn) about the normal, is a member of the symmetry group G, regarded
now as a subgroup of SO(3). It follows, from the existence of exactly eight di.erent
sets of reOections which are symmetry transformations for c, that there are at least
eight di.erent types of symmetry groups. Furthermore, the half-turns about the normals
of the reOections in each set are elements of possible symmetry groups, and each of the
2488 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492

sets S1 ; : : : ; S8 assembled in Table 2 yields a collection of rotations which necessarily

belongs to the symmetry group of c. A natural idea now is to construct the subgroup
of SO(3) generated by these rotations because, in view of the group properties, this
subgroup must be contained in the symmetry group, whatever it might be; we denote
the subgroup corresponding to S by G . By de4nition, G is the (unique) minimal
group containing all rotations through  associated with the symmetry planes of S .
It is important to observe, however, that the subgroup generated in this way cannot
be guaranteed to coincide with the symmetry group de4ned through Eq. (5); we only
know that the former is contained in the latter and this condition is not, in general,
suLcient to uniquely determine G from G . It might seem appropriate at 4rst sight
to de4ne as the “material symmetry group” the subgroup of SO(3) generated by the
half-turns associated with reOections which are symmetry transformations for c. This
is not a suitable choice, however, because it is generally to be expected that a criterion
based on the set of rotations of  which satisfy Eq. (5) will lead to a less re4ned
classi4cation of elasticity tensors than a criterion based on the set of rotations, with no
additional restrictions, obeying the same condition. There are strong clues suggesting
that this is exactly what happens for tensors with fewer index symmetries than c.
It is clear then that, a priori, we should not necessarily expect to be able to de-
duce the material symmetry group G itself from the computation of G , but only the
useful information derived from the knowledge that the latter must be a subgroup of
the former. In the case of elasticity tensors, however, something extremely interesting
occurs: contrary to expectation, the subgroups G do indeed coincide with the possi-
ble symmetry groups. Such groups, eight in number, have been determined explicitly
by Forte and Vianello (1996) and, as we shall now prove, they turn out to coincide
exactly with the subgroups G generated from the half-turns associated with all reOec-
tions present in each set S . This surprising result is interesting, and useful, because
it gives a rationale to the idea that, at least in linear elasticity, the possible types of
symmetry can be decided solely on the basis of the number and relative orientation of
the symmetry planes, along a line of thought 4rst proposed by Cowin and Mehrabadi
(1995).

8.2. Construction of the groups G

For  = 1; : : : ; 8 the group G de4ned in Section 8.1 can be constructed uniquely as

follows.

(i) Form the set A from S by deleting −I , replacing each reOection by its negative
and including the negative of any reOection generated from reOections in S by the
identities (19) and the negative of any reOection generated similarly by identity
(20).
(ii) Form the set B consisting of all the distinct products of reOections in A which
do not coincide with an element of A .

Both A and B are subsets of SO(3). Their union is G .

The calculation of G is now outlined for each type of symmetry.
 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492 2489

1. A1 = {I }; B1 = ∅. Hence G1 = 1, the subgroup of SO(3) consisting of the identity

matrix.
2. A2 = {I; R1 } with R1 = −R(e2 ). Since R21 = I; B2 = ∅. Hence G2 = {I; R1 } which is
isomorphic to the cyclic group Z2 . Note that R1 = Q(e2 ; ) where Q(n; ) denotes
the right-handed rotation through the angle  about the direction de4ned by the unit
vector n.
3. Reading o. the normals of the induced reOections from Eq. (39) we have A3 ={I; Rj ;
j = 1; : : : ; 5} with
Ri = −R(ei ); i = 1; 2; 3; R4 ; R5 = −R(2−1=2 (e1 ± e2 )):
R1 ; : : : ; R5 have only two distinct products not in A3 and we make the choice
R6 = R2 R4 ; R7 = R4 R2 :
Then B3 = {R6 ; R7 } and G3 = {I; Rj ; j = 1; : : : ; 7}. In view of the relations
R6 = Q(e3 ; 12 ); R3 = Q(e3 ; ); R7 = Q(e3 ; 32 );

{R1 ; R4 ; R5 } = R2 {R3 ; R6 ; R7 } = Q(e2 ; ){R3 ; R6 ; R7 };

G3 is isomorphic to the dihedral group D4 .
4. With reference to Eq. (41), A4 = {I; R1 ; R2 ; R3 } with
R1 = −R(e2 ); R2 ; R3 = −R( 12 (31=2 e1 ± e2 )):
There are two distinct products of R1 ; R2 ; R3 not in A4 and we choose
R4 = R1 R2 ; R5 = R2 R1 :
Then B4 = {R4 ; R5 } and G4 = {I; Rj ; j = 1; : : : ; 5}. By virtue of the relations
R4 = Q(e3 ; 23 ); R5 = Q(e3 ; 43 );

{R2 ; R3 } = R1 {R4 ; R5 } = Q(e2 ; ){R4 ; R5 };

G4 is isomorphic to the dihedral group D3 .
5. From Eq. (44), A5 = {I; R1 ; R2 ; R3 } with
Ri = −R(ei ); i = 1; 2; 3:
All the products of R1 ; R2 ; R3 are in A5 so B5 = ∅; G5 = A5 . Since
Ri = Q(ei ; ); R3 = Q(e2 ; )R1 ;
G5 is isomorphic to the dihedral group D2 .
6. From Table 1, A6 = {I; −R(e3 ); −R(sin e1 + cos e2 ) ∀ ∈ [0; 2)}. Inspection of
the identities
R(e3 )R(sin e1 + cos e2 ) = −R(sin ( − 12 )e1 + cos ( − 12 )e2 );

R(sin 2 e1 + cos 2 e2 )R(sin 1 e1 + cos 1 e2 ) = Q(e3 ; 2(1 − 2 ));

−R(e3 ) = Q(e3 ; );

−R(sin e1 + cos e2 ) = −R(e1 )Q(e3 ; 2 − ));

2490 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492

shows that the union of A6 with the set of products of the reOections in A6 consists
of the rotations about e3 together with the products of such rotations with the
half-turn −R(e1 )=Q(e1 ; ). This means that G6 is isomorphic to the two-dimensional
orthogonal group O(2).
7. Eq. (57) gives A7 = {I; Rj ; j = 1; : : : ; 9} with
Ri = −R(ei ); i = 1; 2; 3; R4 ; R5 = −R(2−1=2 (e2 ± e3 ));

R6 ; R7 = −R(2−1=2 (e3 ± e1 )); R8 ; R9 = −R(2−1=2 (e1 ± e2 )):

The three subsets {I; Rj } of A7 with j = 1; 2; 3; 4; 5; j = 1; 2; 3; 6; 7 and j = 1; 2; 3; 8; 9
are each isomorphic to A3 and thus represent tetragonal symmetry. Proceeding as
in the computation of G3 , we obtain the 3 × 2 distinct products
R2 R8 ; R8 R2 ; R3 R4 ; R4 R3 ; R1 R6 ; R6 R1 : (60)
The 4 subsets {I; Rj } of A7 with j = 4; 8; 7; j = 6; 4; 9; j = 5; 6; 8 and j = 7; 9; 5 are
each isomorphic to A4 and so represent trigonal symmetry. The procedure followed
for G4 delivers the 4 × 2 distinct products
R4 R8 ; R8 R4 ; R6 R4 ; R4 R6 ; R5 R6 ; R6 R5 ; R7 R9 ; R9 R7 : (61)
It can be veri4ed straightforwardly that the products in (60) and (61) are distinct
and not in A7 and that every product of two reOections in A7 coincides with a
member of A7 , (60) or (61). B7 therefore consists of the 14 products in (60) and
(61), implying that G7 is a group of order 10 + 14 = 24. A subgroup of SO(3)
of this order containing three copies of D4 and four copies of D3 is necessarily
isomorphic to the octahedral group O (Golubitsky et al., 1985, p. 105).
8. From Table 1, A8 = {I; −R(n) ∀n ∈ U}. An arbitrary rotation Q(m; ) is the product
of reOections R(n2 )R(n1 ) with normals n1 ; n2 such that n1 · n2 = cos 12 ; n1 ∧ n2 =
sin 12 m (see, for example, Pars, 1965, Section 7:7). The elements of A8 therefore
generate the set of all rotations, leading to the conclusion that G8 is isomorphic to
SO(3).
The group isomorphisms obtained above con4rm that the classi4cation of elastic
symmetries resulting from the successive introduction of reOections is the same as
that established by Forte and Vianello (1996: see, in particular, Section 2) by more
advanced algebraic methods using rotations.

9. Summary and conclusions

The systematic and exhaustive search for planes of symmetry carried through in
this paper has established that, in linear elasticity, every material is either isotropic
or anisotropic, and that an anisotropic material is either triclinic, monoclinic, trigonal,
orthotropic, tetragonal, cubic or transversely isotropic. It has been shown in Section
8:2 that each of the eight sets of symmetry planes is associated with a subgroup of
SO(3) and that the eight groups thus obtained coincide with those identi4ed in Forte
and Vianello (1996) as the symmetry groups of elasticity tensors.
 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492 2491

The contents of this paper and the results of Forte and Vianello (1996) show that
there are two criteria for de4ning and classifying symmetry classes in linear anisotropic
elasticity. They can be described as follows.
(1) Elasticity tensors are classi4ed according to the set of symmetry planes allowed.
(2) Elasticity tensors are classi4ed according to the symmetry group they possess.
It has been shown by Cowin and Mehrabadi (1995) that the reOections (orthogonal
transformations) associated with normals to the planes of symmetry for each set are the
generators of the associated group, suggesting the equivalence of the criteria for linear
elasticity which has been veri4ed in Section 8. It is clear, however, that in general
the criterion (1) above is weaker than (2) and will give fewer symmetry classes.
We suggest that it may be a coincidence that, in the case of the linear elasticity
tensor, the two criteria yield the same number of symmetry classes. There are results
indicating that, for tensors of higher order, or even elasticity tensors which are not
hyperelastic, the two criteria would give di.erent results (Forte and Vianello, 1998).
Criterion (1), indeed, would fail to distinguish between classes which are di.erent
according to criterion (2). A similar situation with respect to a di.erence in criteria
is evident when the work of Forte and Vianello (1996) is compared with that of Huo
and Del Piero (1991).

Acknowledgements

Fig. 1 has been taken and modi4ed, with the author’s kind permission, from the
thesis of Y.P. Arramon (1997). The authors are grateful to Professor T.C.T. Ting for
a number of helpful comments.

References

Arramon, Y.P., 1997. A multidimensional anisotropic strength criterion based on Kelvin modes. Ph.D. Thesis,
The City University of New York.
Chadwick, P., 1995, 1997. Unpublished work.
Cowin, S.C., 1989. Properties of the anisotropic elasticity tensor. Quart. J. Mech. Appl. Math. 42, 249–266.
Cowin, S.C., 1995. On the number of distinct elastic constants associated with certain anisotropic elastic
symmetries. In: Casey, J., Crochet, M.J. (Eds.), Theoretical, Experimental, and Numerical Contributions
to the Mechanics of Fluids and Solids. Zeit. Angew. Math. Phys. 46, S210 –S224.
Cowin, S.C., Mehrabadi, M.M., 1987. On the identi4cation of material symmetry for anisotropic elastic
materials. Quart. J. Mech. Appl. Math. 40, 451–476.
Cowin, S.C., Mehrabadi, M.M., 1989. Identi4cation of the elastic symmetry of bone and other materials. J.
Biomech. 22, 503–515.
Cowin, S.C., Mehrabadi, M.M., 1995. Anisotropic symmetries of linear elasticity. Appl. Mech. Rev. 48,
247–285.
Fedorov, F.I., 1968. Theory of Elastic Waves in Crystals. Plenum Press, New York.
Forte, S., Vianello, M., 1996. Symmetry classes for elasticity tensors. J. Elasticity 43, 81–108.
Forte, S., Vianello, M., 1998. Functional bases for transversely isotropic and transversely hemitropic
invariants of elasticity tensors. Quart. J. Mech. Appl. Math. 51, 543–552.
Golubitsky, M., Stewart, I., Schae.er, D.G., 1985. Singularities and Groups in Bifurcation Theory, Vol. 2.
Springer, New York.
2492 P. Chadwick et al. / J. Mech. Phys. Solids 49 (2001) 2471 – 2492

Hayes, M.A., Norris, A.N., 1992. Static implications of the existence of a plane of symmetry in an anisotropic
elastic solid. Quart. J. Mech. Appl. Math. 45, 141–147.
Hearmon, R.F.S., 1961. An Introduction to Applied Anisotropic Elasticity. Oxford University Press, Oxford.
Huo, Y.Z., Del Piero, G., 1991. On the completeness of the crystallographic symmetries in the description
of the symmetries of the elasticity tensor. J. Elasticity 25, 203–246.
Norris, A.N., 1989. 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, 413–426.
Nye, J.F., 1957. Physical Properties of Crystals. Oxford University Press, Oxford.
Pars, L.A., 1965. A Treatise on Analytical Dynamics. Heinemann, London.
Smith, G.F., Rivlin, R.S., 1958. The strain-energy function for anisotropic elastic materials. Trans. Amer.
Math. Soc. 88, 175–193.
Ting, T.C.T., 1996. Anisotropic Elasticity. Theory and Applications. Oxford University Press, New York.
Voigt, W., 1910. Lehrbuch der Krystallphysik. Teubner, Leipzig.