Comput. Methods Appl. Mech. Engrg.

189 (2000) 419438

www.elsevier.com/locate/cma

On the theory of fourth-order tensors and their applications in

computational mechanics
Mikhail Itskov *
Institut fur Statik und Dynamik, Ruhr-Universit
at Bochum, Universit
atsstrae 150, 44780 Bochum, Germany
Received 1 December 1998; received in revised form 29 November 1999
Dedicated to Professor Dr. -Ing. Yavuz Bas ar on the occasion of his 65th anniversary

Abstract
Many problems concerned with the mathematical treatment of fourth-order tensors still remain open in the literature. In the present
paper they will be considered in the framework of a complete theory involving a set of notations and denitions, a tensor operation
algebra, dierentiation rules, eigenvalue problems, applications of fourth-order tensors to isotropic tensor functions and some other
relevant aspects. A tensor is understood to be an invariant quantity with respect to any co-ordinate system transformation, which
justies the use of absolute notation preferred in this work. As a most important application eld of fourth-order tensors, elastic
moduli are formulated in a material and a spatial description and given for various hyperelastic material models. 2000 Elsevier
Science S.A. All rights reserved.
Keywords: Fourth-order tensors; Tensor algebra; Eigenvalue problems; Isotropic tensor functions; Tangent moduli

1. Introduction
Fourth-order tensors as a mathematical object have found in the last 20 years a wide use in computational mechanics and especially in the nite element method. Their well-known applications are tangent
(elastic or elasto-plastic) moduli as well as damage tensors playing an important role in the formulation of
constitutive and evolution equations. The use of fourth-order tensors remains, nevertheless, extremely
complicated because of the absence of a closed theory embracing many relevant aspects. In spite of active
eorts in recent times (see [39,11,21,22,24,2730,37,38,41,42]), there are many problems concerned with
the treatment of fourth-order tensors, which are not completely solved and still remain open in the literature.
How can a fourth-order tensor be constructed from two second-order ones?
How can algebraic operations with fourth-order tensors be given in the absolute notation?
What does the derivative of a second-order with respect to another one mean? There is no unied definition for the derivative with respect to a tensor. Is there a dierence in the denition of the derivative
with respect to a symmetric second-order tensor?
How can the eigenvalue problem be formulated for fourth-order tensors and how can principal invariants and eigenvalues of a fourth-order tensor be calculated? How many irreducible invariants does a
fourth-order possess?

Present address: Department of Applied Mechanics and Fluid Dynamics, University of Bayreuth, D-95440 Bayreuth, Germany.
E-mail address: mikhail.itskov@uni-bayreuth.de (M. Itskov).

0045-7825/00/$ - see front matter 2000 Elsevier Science S.A. All rights reserved.
PII: S 0 0 4 5 - 7 8 2 5 ( 9 9 ) 0 0 4 7 2 - 7

420

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

The goal of the present paper is to try to solve these problems in a framework of the complete theory and
to demonstrate then its application for the derivation of tangent moduli. The important characteristic
feature of the development is the absolute notation used for tensor quantities throughout the paper. This is
not only a notation manner but the basic concept of the paper. A tensor is understood to be an invariant
quantity with respect to any co-ordinate system transformation and can be neither covariant, nor contravariant, nor mixedvariant in contrast to its components. In this connection, for example, the widespread
use of the metric tensor as an argument of a function or as a dierentiation parameter seems to be
meaningless, since the metric tensor represents in the end the identity tensor. Such notations always have to
be accompanied by an indication of the component variance to be used in the present operation and will be
avoided here.
As a further crucial point of the paper it should be noted that in contrast to many other works we deal
with arbitrary non-symmetric fourth-order tensors and understand them as a linear mapping of one arbitrary second-order tensor into another one (see also [19,29]). This considerably simplies the treatment of
fourth-order tensors, which can be seen especially clearly in consideration of the eigenvalue problem. The
symmetry properties of fourth-order tensors, which are relevant, for instance, for tangent moduli, can be
enforced without loss of generality at the last stage of derivation.
The starting point of the development is the set of notations and denitions given in Section 2. To have
more freedom in construction of fourth-order tensors we propose a new tensor product of two second-order
ones. The crucial point of the presented theory lies in a new denition of the derivative of one second-order
tensor with respect to another one 2;2 . Due to this denition, the well-known product derivative rule valid
for scalar values is proved to hold also for second-order tensors. The double contraction of a fourth-order
tensor with a second-order one (4:2) is also redened to be consistent with the dierentiation rule intro_
duced, such that the important rate relation AB
A;B : B_ is satised. Further, it is shown that the derivative with respect to a symmetric tensor is not unique and requires a slight correction in the denition.
Special attention is paid in Section 2 to the transposition operations for fourth-order tensors and the resultant denition of a symmetric and a super-symmetric tensor. In this context, we introduce a special
transposition operation bringing a fourth-order tensor to the standard denition of the tensor derivative. In
the following this will permit to verify some important results by comparing them with those available in
the literature.
The indispensable part of the proposed theory is the algebra of fourth-order tensors given in Section 3.
Simple, double, quadruple contractions and other operations with fourth-order tensors constructed from
second-order ones are presented in absolute notation and can easily be proved using the denitions of the
previous section. The following important topic is the formulation and proof of the dierentiation rules
such as the product derivative and the chain rule. On this basis, we then obtain the derivative of the power
tensor function for positive as well as negative integer exponents.
Section 4 is devoted to the formulation of eigenvalue problems and spectral decomposition of fourthorder tensors. Using matrix representation the eigenvalue problem of a fourth-order tensor is reduced to
that of a matrix and can then be solved by a standard procedure. For a symmetric fourth-order tensor this
yields nine real eigenvalues and nine corresponding eigentensors. A complete analogy with the eigenvalue
problem of a second-order tensor can be observed. On the basis of the other possible denitions for the
double contraction 4:2 at whole 27 principal invariants and 27 eigenvalues of the fourth-order tensor can be
received in this way. The discussed procedure is illustrated by spectral decompositions for a number of
important fourth-order tensors like the identity, the trace projection and the transposition projection
tensor.
Application of the proposed theory to isotropic tensor functions is the topic of Section 5. Attention is
focused on the representation of the derivative of an isotropic tensor-valued tensor function with respect to
its argument, which is of major importance for the formulation and numerical calculation of tangent
moduli in large strain elasticity and elasto-plasticity. We start with the consideration of a special class of
isotropic tensor functions, which can be expanded in tensor power series. In the case of a symmetric tensor
argument we receive for the derivative in question two closed-form representations. The rst one is expressed in terms of eigenvectors of the tensor argument and corresponds to the well-known result (see
[9,27]). The second representation is given through the eigenvalue bases and is advantageous, as it avoids
the numerically expensive computation of eigenvectors. In contrast to the analogous result by Miehe [24],

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

421

this representation does not use the derivatives of the eigenvalue bases and is obtained in a simpler and
more compact form.
The tensor algebra and dierentiation rules developed are applied in Section 6 to formulate tangent
moduli in a material and a spatial description. It is proved that the tangent moduli presented are symmetric
fourth-order tensors and permit consequently the spectral decomposition. The important question to be
treated then is how the isotropy can be identied by an elastic modulus. Using the spectral decomposition
we prove that the elastic modulus corresponds to an isotropic material in the natural state if only it can be
represented in the form of the St.VenantKirchho model involving two material parameters (see also
[14,15,17,19,20,29,31] and references therein). Elastic moduli for this and some other hyperelastic material
models such as MooneyRivlin and Ogden are given in Section 7.

2. Basic notations and denitions

A second-order tensor (a bold capital letter):
A Aij gi
gj Aij gi
gj Ai
j gi
gj

2:1

A fourth-order tensor (a bold italics capital letter):

k
gi
gj
gk
gl



D Dijkl gi
gj
gk
gl Dijkl gi
gj
gk
gl Di
j
l

2:2

The simple contraction of arbitrary order tensors:

m
g i
gl
gm
gn :
AD Ai
j gi
gj Dk
l
nm gk
gl
gm
gn Ai
j Dj
l
n

2:3

Tensor products of two second-order tensors are dened as follows:

D A
B Ai
j gi
gj
Bk
l gk
gl Ai
j Bk
l gi
gj
gk
gl ;

2:4

D A  B Ai
j gi
gj  Bk
l gk
gl Ai
j Bk
l gi
gk
gl
gj :

2:5

It can easily be conrmed the tensor products satisfy the distributive rule:
A
B C A
B A
C;

A  B C A  B A  C:

2:6

Transposition operations


T and


t for fourth-order tensors:

T

k
k j
gi
gj
gk
gl Di
j
l
g
gi
gl
gk ;
DT Di
j
l

2:7

k
k
gi
gj
gk
gl t Di
j
l
gi
gk
gj
gl :
Dt Di
j
l

2:8
T

Remark. One can easily prove that the transposition operations


and


are not commutative with
each other: DTt 6 DtT .
According to the above denitions the transposition applied to the tensor products (2.4) and (2.5) yields:
T

A
B AT
BT ;

A  B B  A;

A  B A  BT :

2:9

A symmetric fourth-order tensor (cf. [19,29])

DT D:

2:10

A super-symmetric fourth-order tensor

ET E

and

E t E:

2:11

422

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

By means of the operation


0

1
E 0 E E t ;
2

2:12

a symmetric fourth-order tensor can be transformed into a super-symmetric one.

Remark. For a fourth-order tensor some other transposition operations can be dened and particularly:
k
k
gi
gj
gk
gl R Di
j
l
g i
gl
g j
gk ;



R : DR Di
j
l

2:13

which possess the important property

A  BR A
B:

2:14

Double contractions of a fourth-order tensor with a second-order one 4 : 2 and 2 : 4 are dened by
k
k
gi
gj
gk
gl : Am
n gm
gn Di
j
l
Aj
k gi
gl ;
D : A Di
j
l

2:15

k
k j
gi
gj
gk
gl Ai
l Di
j
l
g
gk :
A : D Am
n gm
gn : Di
j
l

2:16

The double contraction of a fourth-order tensor with another fourth-order one (4:4):
k
n
s
ik jr
n
l
gi
gj
gk
gl : Amr
D : A Di
j
l

n
s gm
g
gr
g D
j
l A
n
k gi
g
gr
g :

2:17

Evidently, the double contraction of fourth-order tensors is not commutative D : A 6 A : D, but can easily
be proved to full the associative rule:
A : B : C A : B : C:

2:18

In contrast to the double contraction the scalar product (quadruple contraction) of two fourth-order
tensors 4 :: 4
k
k jl
gi
gj
gk
gl :: Am
n
sr gm
gn
gr
gs Di
j
l
A
i
k
D :: A Di
j
l

2:19

obey the commutative rule:

D :: A A :: D:

2:20

According to the denitions (2.15)(2.17) and (2.19) the operations 2 : 4, 4 : 2, 4 : 4 and 4 :: 4 satisfy
the distributive rule e.g.:
A : B C A : B A : C;

A B : C A : C B : C;

2:21

A : B C A : B A : C;

A B : C A : C B : C;

2:22

A :: B C A :: B A :: C; A B :: C A :: C B :: C:

2:23

The second-order identity tensor:

I g i
gi :

2:24

The fourth-order identity tensor:

I I
I gi
gi
gj
gj :

2:25

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

423

The trace of a fourth-order tensor is dened by

tr D D :: I I :: D Di
i
jj :

2:26

Powers of fourth-order tensors:

D0 I;

D1 D;

D2 D : D; . . .

Dn D
: D{z
:


D
|
} :

2:27

An inverse fourth-order tensor:

D : D1 D1 : D I:

2:28

The matrix representation of a fourth-order tensor


is dened through its mixedvariant components by
2 11
2
3
1
2
3
1
2
3 3
D
1
1 D1
1
1
D1
1
1
D1
2
1
D1
2
1
D1
2
1
D1
3
1
D1
3
1
D1
3
1
1
2
3
1
2
3
1
2
3 7
6 D1
1
2
D1
1
2
D1
1
2
D1
2
2
D1
2
2
D1
2
2
D1
3
2
D1
3
2
D1
3
2
7
6
6 D1 1 D1 2 D1 3 D1 1 D1 2 D1 3 D1 1 D1 2 D1 3 7
6
1
3

1
3

1
3

2
3

2
3

2
3

3
3

3
3

3
3 7
6 21
2
3
1
2
3
1
2
3 7
7
6 D
1
1 D2
1
1
D2
1
1
D2
2
1
D2
2
1
D2
2
1
D2
3
1
D2
3
1
D2
3
1
7
6 21
22
23
21
22
23
21
22
23 7
6
2:29
D 6 D
1
2 D
1
2 D
1
2 D
2
2 D
2
2 D
2
2 D
3
2 D
3
2 D
3
2 7:
6 D2 1 D2 2 D2 3 D2 1 D2 2 D2 3 D2 1 D2 2 D2 3 7
6
1
3

1
3

1
3

2
3

2
3

2
3

3
3

3
3

3
3 7
6 31
2
3
1
2
3
1
2
3 7
7
6 D
1
1 D3
1
1
D3
1
1
D3
2
1
D3
2
1
D3
2
1
D3
3
1
D3
3
1
D3
3
1
7
6
4 D3 1 D3 2 D3 3 D3 1 D3 2 D3 3 D3 1 D3 2 D3 3 5

1
2

1
D3
1
3

1
2

2
D3
1
3

1
2

3
D3
1
3

2
2

1
D3
2
3

2
2

2
D3
2
3

2
2

3
D3
2
3

3
2

1
D3
3
3

3
2

2
D3
3
3

3
2

3
D3
3
3

The partial derivative of a scalar-valued tensor function aA with respect to its second-order argument A
0;2 is given by
aA;A

oa
oa i
g
gj

oAi
j gi
gj oAi
j

2:30

and possesses the following important properties:

1. The derivative is independent of the choice of the component variance used in the dierentiation. Since
Akl Ak
m gml , we have
oa i
oa oAkl i
oa jl i
oa i
g

g
gj
g g
gj
g
gl :
j
oAi
j
oAkl oAi
j
oAil
oAil
Similarly it can easily be proved also for the covariant components.
2. The derivative is independent of the choice of the co-ordinate system. In a transformed co-ordinate system we obtain:
k

A Ai
j gi
gj A
l gk
gl ) A
l gk
gi Ai
j gj
gl ;
k

oa i
oa oA
l i
oa k
oa k
g
gj
g
gi gj
gl gi
gj
g
gl :
i g
gj
k
k
k
oA
oAi
j
oA
l
oA
l
oA
l

j
These properties ensure the objectivity of the tensor derivative and justify the use of the absolute notation
aA;A .
Remark 1. Properties 1 and 2 and their proofs are not valid for the derivative with respect to the metric
tensor. As mentioned above, the derivative with respect to the metric tensor is not objective since it depends
not only on the component variance but also on the co-ordinate system used in dierentiation (see [16]).
Remark 2. It is important to note that in the dierentiation with respect to a tensor its vector base is kept
constant. Disregard of this fact may lead to an incorrect result. As a typical example we consider derivatives
with respect to the GreenLagrange and Almansi strain tensors

424

and

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

1
1
E C G gij Gij Gi
Gj Eij Gi
Gj
2
2
1
1
e g b1 gij Gij gi
gj eij gi
gj ;
2
2

respectively, which are often used for the formulation of constitutive relations. It can be seen that these
tensors have equal covariant components Eij eij related, nevertheless, to the dierent bases remaining
constant in the dierentiation. Thus
!


! ow
ow T
ow
ow
ow
ow
T
F F

Gi
Gj F F 2
Gi
Gj FT 2
gi
gj 6
g
gj
F
oE
oEij
ogij
ogij
oe oeij i
2

ow
ow
g
gj 2
g
gj ;
1 i
oGij i
obij

which evidently contradicts the constitutive equation for the Cauchy stress tensor of the form s w;e (see
e.g. [32]).
The partial derivative of a second-order tensor C Ci
j Gi
Gj with respect to another second-order one
A Ak
l gk
gl (2;2 ) is dened by
C;A

oCi
j Gi
Gj
oAk
l gk

gl

oCi
j
oAk
l

G i
gk
gl
G j

2:31

for the tensor bases Gi

Gj of C, which are independent of the tensor components Ak
l . In other respect the
choice of co-ordinate systems and the variance of components of both tensors A and C are irrelevant, which
can easily be proved similarly to the properties of the derivative 0;2 . This ensures the objectivity of the
denition (2.31) and justies the use of absolute notation for the tensor C;A .
Remark. The derivative obtained according to the denition (2.31) can be transformed to the standard one
R
by means of transposition


. Using (2.13) we have:
R

C;A

oCi
j
oAk
l

Gi
Gj
gk
gl :

The denitions (2.30) and (2.31) become non-unique if the tensor A is symmetric. In this case the dierentiation cannot still be carried out with respect to nine components of A, since only six of them are independent. This deciency can be eliminated by modifying the denitions (2.30) and (2.31). Accordingly the
derivatives with respect to the symmetric tensor A AT can be dened by
aA;A

C;A

3
3
1X
oa
1X
oa i
gi
gj gj
gi
g
gj gj
gi
2 i;j 6 i oAij
2 i;j 6 i oAij
3
1X
oa i
g
gj gj
gi ;
2 i;j 6 i oAi
j

2:32

3
3
oCi
j
oCi
j
1 X
1 X
Gi
gk
gl gl
gk
Gj
Gi
gk
gl gl
gk
Gj
2 k;l 6 k oAkl
2 k;l 6 k oAkl
3
oCi
j
1 X
Gi
gk
gl gl
gk
Gj :
2 k;l 6 k oAk
l

2:33

As can be observed the dierentiation in (2.32) and (2.33) is accomplished with respect to the six independent components of A and the resulting tensor is then symmetrized.

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

425

3. The algebra of fourth-order tensors

In the following we summarize some important tensor identities, which will be intensively used in further
derivation:
T

A
B : C ACB;
A  B : C B : CA;

C : A
B AT CBT A
B : C;
C : A  B C : AB;

3:1
3:2

A
BC A
BC;
A  BC AC  B;

CA
B CA
B;
CA  B CA  B;

3:3
3:4

A
B : C
D AC
DB;

A  B : C  D B : CA  D;
T

A
B : C  D ACB  D;
T

A
B :: C
D A : C B : D ;
A
B :: C  D ACB : D;




t
A
Bt : C
D ADT
CT B ;
T

E : F F : E ;
T

3:6

A  B :: C  D A : DC : B;

3:7

E : A A : E;

A  B :: C
D A : C BD ;
t

A
B : C  D AC B  D;
t

3:8
3:9

E : F E : F ;

3:10

C : E C : E ;

3:11

AE : CB AEB : C:

3:12

E :CE:C ;

3:5

A  B : C
D A  C BD ;

The above relations directly emanate from the corresponding denitions (2.3)(2.5), (2.7)(2.8), (2.15)
(2.17) and (2.19). For example the identity 3:51 can easily be proved using (2.17), which leads to
A
B : C
D Ai
j Bk
l gi
gj
gk
gl : Cm
n Dr
s gm
gn
gr
gs
Ai
j Bk
l Cj
n Dr
k gi
gn
gr
gl AC
DB:
By virtue of (2.31) we construct the partial derivative of a second-order tensor with respect to itself yielding
the fourth-order identity tensor:
A;A

oAi
j
oAk
l

gi
gk
gl
gj dik dlj gi
gk
gl
gj gi
gi
gj
gj I:

3:13

The properties of the fourth-order identity tensor I following from (2.25), (3.1) and 3:51 can now be
given by
I : B B : I B;

I : D D : I D:

3:14

Using (2.26), (3.7) and (3.8) the validity of the following important identities with the trace of a fourthorder tensor can also be proved:
j
kl
C
i
j
;
trC : D trD : C C : D :: I C :: D D :: C Di
k
l

3:15

trA
B A
B :: I tr Atr B; trA  B A  B :: I A : B;
tr I I :: I I : II : I 9; tr I  I I  I :: I I : I 3:

3:16
3:17

Due to the important property

I  I : A A : I  I tr AI

3:18

the tensor I  I introduced in 3:172 is referred to as the trace projection tensor.

Remark. For the standard denition of the tensor derivative the trace projection tensor has the form

R
I  IR gi
gj
gj
gi gi
gi
gj
gj I
I:

426

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

Now, we construct the partial derivative of a second-order tensor with respect to its transverse counterpart
delivering the transposition projection tensor T:
A;TA A;AT

oAi
j
oAk
l

gj
gk
gl
gi dik dlj gj
gk
gl
gi I t T;

3:19

which possesses the following properties:

T : B gj
gi
gj
gi : Bk
l gk
gl Bi
j gj
gi BT ;

B : T BT :

3:20

For the derivative with respect to the symmetric part symA 1=2A AT of a tensor A one receives
under consideration of (2.12) and (2.33)
1
A;symA A;TsymA symA;symA I T I 0 :
2

3:21

Generalizing this result to the derivative of an arbitrary tensor delivers:

B;symA B;A : A;symA B;A : I 0 B;A 0 :

3:22

Now, attention is focused on the tensor dierentiation rules. We start with the product derivative
AB;C A;C B AB;C ;

3:23

which is well known for scalar values. According to the new denition of the derivative 2;2 (2.31) the
product derivative rule (3.23) also holds for second-order tensors.
Proof.
AB;C

oAi
j Bj
k gi
gk
oCl
m gl
gm
oAi
j
oCl
m

oAi
j Bj
k
oCl
m

gi
gl
gm
gk

Bj
k gi
gl
gm
gk Ai
j

oBj
k
gi
gl
gm
gk A;C B AB;C :
oCl
m

Using this remarkable relation one can easily obtain the derivative of the tensor power function with respect to its argument B;nB n 1; 2 . . .. Considering (3.3), 3:51 and (3.15) one gets:
B;2B B
I I
B;
B;3B

3:24
2

n1
X

n1
X
n1r
r
Bn1r
Br
B
I
: I
B ;

B
I B
B I
B B
I B
I : I
B I
B ;

B;nB

r0

B;nB : I nBn1 ;

3:25
3:26

r0

3:27

where we adopt B0 I. To obtain the derivative of the inverse function B1 ;B we rst use the identity:
BB1 ;B I;B 0, which leads by virtue of (3.3) and (3.23) to
B1 ;B B1
B1 :

3:28

By analogue and exploiting (3.28) we can further write for the derivative of the power function with a
negative integer exponent:

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

Bn ;B

n1
X

Br1
Brn ;

427

3:29

r0

Bn ;B : I nBn1 :

3:30

Two other important dierentiation rules can be formulated as follows:

aA;B A  a;B aA;B ;

A : B;C A : B;C B : A;C ;

3:31

where a denotes a scalar function.

Proof.
aA;B

oaAi
j gi
gj
oBk
l gk
gl

oaAi
j
oBk
l

oAi
j
oa i
k
j
gi
g
gl
g k A
j gi
g
gl
g a k gi
gk
gl
gj
oB
l
oB
l
k

A  a;B aA;B ;
A : B;C

oAi
j Bi
j
oCk
l gk
gl

oAi
j
oB j
Ai
j i
k Bi
j k
oC
l
oC
l

!
gk
gl A : B;C B : A;C :

For the derivative of a scalar f A and a tensor-valued AC tensor function one can prove the following
chain rules:
f A;B
A;B

i
of oA
j k
g
gl f ;A : A;B ;
oAi
j oBk
l

3:32

oAi
j oCk
l
m
j
m gi
g
gn
g A;C : C;B :
oCk
l oB
n

3:33

4. Spectral decomposition, eigenvalues and principal invariants of a fourth-order tensor

The double contraction 4:2 (2.15) establishes the basis for the formulation of the eigenvalue problems
D : M kM

4:1

for a fourth-order tensor and enables then to determine its nine eigenvalues and principal invariants. To
this end, we transform (4.1) as follows
D kI : M 0:

4:2

This equation has a non-trivial solution

 11
2
 D
1
1 k
D1
1
1

 D1 1
2
D1
1
2
k


1
2
detD kI 
 ...
.
.
.

 D3 1
D3 2

1
3

1
3

M 6 0 if the corresponding characteristic equation


3

...
D1
3
1

13
D
3
2 
 0;
...
. . . 
. . . D3 3 k 

4:3

3
3

which can also be presented in the form:

k9 ID k8 IID k7 IIID k6


IXD 0;

4:4

is fullled. In this equation the coecients ID ; IID ; IIID ; . . . ; IXD denote the principal invariants of the tensor
D. On the basis of Newton formula and in view of the identity A : B A B they can be expressed in terms
of principal traces of D (see also [42]):

428

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

ID tr D k1 k2 k3


k9 ;
i
1h
2
IID tr D2 tr D k1 k2 k1 k3


k8 k9 ;
2

1
3
1
IIID tr D3 tr D2 tr D tr D3 k1 k2 k3 k1 k2 k4


k7 k8 k9 ;
3
2
2


1
4
1
1
2
2
4
IVD tr D4 tr D3 tr D tr D2 trD tr D2 tr D
4
3
2
6

4:5

k 1 k 2 k 3 k 4 k1 k2 k 3 k 5


k6 k 7 k 8 k 9 ;
..
.
IXD det D k1 k2 . . . k9 :
In such a way the eigenvalue problem of a fourth-order tensor is transformed into the eigenvalue problem
of its matrix representation and can then be solved by a conventional procedure of the linear algebra.
Remark. The double contraction 4:2 (2.15) used for the formulation of the eigenvalue problem (4.1) is only
one of three possible ones. Each of them would lead to its own matrix representation, eigenvalue problem
and, nally, to its nine principal invariants. Thus one can obtain the basis of a fourth-order tensor consisting of its 27 invariants. The question about the integrity and irreducibility of this basis requires a special
consideration (see [5]).
T

If the tensor D is symmetric, which immediately leads to symmetry of its matrix representation D D
in an orthogonal co-ordinate system, the solution of the eigenvalue problems (4.1) yields nine real eigenvalues kn with the corresponding orthogonal eigentensors Mn n 1; 2; . . . ; 9. In this case the spectral
decomposition of D is given by (the spectral theorem):
D

9
X

kr M r  M r ;

Ms : Mt dst :

4:6

r1

If the eigenvalues kn n 1; 2; . . . ; 9 of D are known and do not coincide, its eigenvalue bases Mr  Mr can
be presented by the closed formula [29,41]:
M r  Mr

9
Y
D ks I
s6r

with

kr ks

n
Y

Ai A1 : A2 :


An :

4:7

If the tensor D is super-symmetric, which is relevant for the tangent moduli, its matrix representation in an
orthogonal co-ordinate system includes three pairs of equal rows and columns and has the reduced rank 6.
This means that at least three invariants and three eigenvalues of D are identically equal to zero. For the six
eigentensors of D associated with non-vanishing eigenvalues one can obtain by virtue of 2:93 and 3:102 :
t

Mr  Mr Mr  MTr Mr  Mr

Mr MTr :

4:8

Thus, the spectral decomposition of a super-symmetric fourth-order tensor can be expressed in terms of
only six eigenvalues and corresponding symmetric eigentensors (see also [11,12,21,30,35]):
D

6
X

kr M r  M r ;

r1

Mr MTr :

4:9

The closed formula solution for the eigenvalue bases (4.8) takes in this case the form:
M r  Mr

6
Y
D ks I 0
:
kr ks
s6r

4:10

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

429

Concluding this section we present for example spectral decompositions of some important fourth-order
tensors I, I  I, T and I 0 :
I

3
X

ni
nj  ni
nj ;

k1 k2


k9 1;

4:11

k1 3;

4:12

i;j1


 

1
1
I  I 3 p I  p I ;
3
3
T

3
X

ni
ni  ni
ni

k2 k3


k9 0;

3
X




1
1
p ni
nj nj
ni  p ni
nj nj
ni
2
2
i;j<i

3
X




1
1
p ni
nj nj
ni  p ni
nj nj
ni ;
2
2
i;j>i

k1 k2


k6 1;

k7 k8 k9 1;

4:13

3
3
X
X




1
1
1
p ni
nj nj
ni  p ni
nj nj
ni ;
ni
ni  ni
ni
I 0 I T
2
2
2
i
i;j<i

k1 k2


k6 1;

4:14

k7 k8 k9 0

with ni i 1; 2; 3 being a set of orthogonal unit vectors: ni
nj dij .

5. Application to isotropic tensor functions
A tensor-valued tensor function GA is said to be isotropic if it satises the relation [39]:
GQAQT QGAQT

5:1
T

for an arbitrary orthogonal tensor Q Q .

In the further derivation we shall rst deal with tensor power series building a special class of isotropic
tensor functions:
2

GA g0 I g1 A aI g2 A aI g3 A aI


;

5:2

where gi and a are scalar-valued constants. These functions are referred here to as analytical tensor
functions. For instance, isotropic tensor functions such as exponential and logarithmic ones being important and often used in computational mechanics are dened in the form (5.2) by
expA I A

1 2
A


;
2!

5:3

1
1
2
3
lnA A I A I A I



2
3

5:4

Our purpose is to obtain the derivative of an isotropic function with respect to its argument, which is of
high importance in computational elasticity and elasto-plasticity for the construction of tangent moduli.
Using the relation (3.26) and setting in (5.2) without loss of generality a 0, we receive for the derivative of
an analytical tensor function
GA;A g1 I g2 A
I I
A


gn

rn1
X
r0

An1r
Ar

5:5

430

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

On the basis of expression (5.5) the important identity for the time derivative can be conrmed as follows
_



gn
GA;A : A_ g1 A_ g2 AA_ AA

rn1
X

_
_ r


GA:
An1r AA

5:6

r0

The representation (5.5) holds for arbitrary tensors. In the special case of a symmetric tensor argument
A AT the derivative GA;A takes the form:
GA;A g1 I 0 g2 A
I I
A0


gn

rn1
X

An1r
Ar 0

5:7

r0

and can expressed by means of the spectral decomposition of A

A k 1 n1
n1 k2 n2
n2 k3 n3
n3 ;
by
GA;A

1
X

"
gn n

3
X
a

n1

kn1
a na

5:8

na
n a
n a

rn1
X

3
X

r0

a;b6a

#
kn1r
krb na
a

na
n b
nb :

5:9

It can easily be seen that the derivative (5.9) presents a super-symmetric fourth-order tensor in the spirit of
denition (2.11):
T

GA;A GA;A ;

GA;A GA;A :

Introducing the abbreviation for the diagonal function [24] Gk

representation for G(A)

P1

n0

5:10
gn kn allowing the alternative

GA Gk1 n1
n1 Gk2 n2
n2 Gk3 n3
n3 ;

5:11

the relation (5.9) can be given in the case of distinct eigenvalues ki in terms of eigenvectors ni by
GA;A

3
X

G0 ka na
na
na
na

3
X
Gka Gkb
0
na
na
nb
nb

k
k
a
b
a;b6a

G0 ka na
na
na
na

3
1 X
Gka Gkb
na
na
nb nb
na
nb :
2 a;b6a
ka kb

3
X
a

5:12

Applying the operation


R (2.13) to the relation (5.12) we can return GA;A to the standard denition of
the tensor derivative:
GA;A R

3
X

G0 ka na
na
na
na

3
1 X
Gka Gkb
na
nb
na
nb nb
na ;
2 a;b6a
ka kb

5:13

which corresponds to the well known result [9,27].

In terms of the eigenvalue bases Mi ni
ni i 1; 2; 3, which can be obtained by means of the closedformula solution [2,26,36], the derivative (5.13) is expressible in the form:
GA;A

3
X
a

G0 ka Ma
Ma

3
X
Gka Gkb
0
Ma
Mb :
k

k
a
b
a;b6a

5:14

The last representation has the advantage, that it avoids a numerically expensive determination of eigenvectors. Comparing with the analogous solution by Miehe [22,24] it is observable that the representation
(5.14) does not include the derivatives of the eigenvalue bases and has a simpler and more compact form.
The special cases of two or three equal eigenvalues ki require a special treatment. Considering in (5.14)
the limit case ka kb D ! 0, one can obtain the following results:

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

431

1. Triple coalescence of eigenvalues: k1 k2 k3 k; A kI,

GA;A G0 kI 0 :

5:15

2. Double coalescence of eigenvalues: ka 6 kb kc k; A ka Ma kI Ma ,

Gka Gk
0
Ma
I Ma I Ma
Ma
ka k
0
G0 kI Ma
I Ma :

GA;A G0 ka Ma
Ma

5:16

According to the representation theorem [39] an arbitrary isotropic tensor function, which cannot
generally be presented via power series (5.2), is given in the form:
GA u0 AI u1 AA u2 AA2 ;

5:17

where ui A ui QAQT are scalar-valued isotropic tensor functions of A. By virtue of (3.13), (3.24) and
(3.31) we construct the derivative GA;A by
GA;A u1 AI u2 AA
I I
A I  u0 A;A A  u1 A;A A2  u2 A;A ;

5:18

where exploiting the isotropy of ui : ui A ui IA ; IIA ; IIIA their derivatives can be given by

ui A;A


oui oui
oui T
oui
IA I
A
IIIA AT :

oIA oIIA
oIIA
oIIIA

5:19

Using the result (5.18) and (5.19) and by virtue of (3.1) and (3.2) it can easily be proved that the relation:
AGA;A : lA GA;A A : Al AGA;A : I
A : l I
A : GA;A : A
I : l;

5:20

where l presents an arbitrary second-order tensor, holds for symmetric tensors A AT .

For the time derivative of an arbitrary isotropic tensor function we obtain by using (3.1), (3.2), (3.14),
(5.18) and (5.19) the well-known relation:
_
_
Iu_ 0 A Au_ 1 A A2 u_ 2 A GA:
GA;A : A_ u1 AA_ u2 AAA_ AA

5:21

To complete this section we give the derivatives of some isotropic tensor functions, which cannot be expanded in power series (5.2):
trAI;A I  trA;A I  I;


1
1
devA;A A trAI ;A I I  I P
3
3

5:22
5:23

with P being the deviatoric projection tensor: A : P P : A devA.

6. Tangent moduli
One of the most important applications fourth-order tensors nd in computational elasticity and elastoplasticity, where they appear as tangent or elastic moduli. The algebra operations and dierentiation rules
presented above now enable to formulate tangent moduli in absolute notation and discuss then their
properties. The formulation succeeds in spatial as well as material descriptions. Important notations to be
exploited in this context are:

432

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

w wC
F gi
Gi
C FT F
b FFT
E 12 C I
S
s FSFT
C; c
_ 1
l FF
d 1=2l lT
Lm a a_ la alT

energy density function (per unit volume of the undeformed body),

deformation gradient,
right Cauchy-Green tensor,
left CauchyGreen tensor (Finger tensor),
GreenLagrange strain tensor,
second PiolaKirchho stress tensor,
Kirchho stress tensor,
tangent(elastic) moduli in material and spatial description respectively,
spatial gradient of velocity,
rate of deformation tensor,
Lie derivative (Oldroyd rate) of a spatial tensor a aij gi
gj .

We start with the general denitions of the tangent moduli. Using the DoyleEricksen formula [13] and
by virtue of relations (3.1) and 3:122 the tangent moduli are constructed such that they satisfy the corresponding constitutive rate equations.
Material description:
ow ow

;
oC oE
o2 w
o2 w

C4
oCoC oEoE
oS
1
1
_
: C_ C : C_ C : E:
S_ 2
oC 2
2

S2

6:1
6:2
6:3

Spatial description:
s 2F

ow T
F ;
oC

6:4

o2 w
: FT
F cabcd FaA FbB FcC FdD CABCD ;
oCoC


o2 w _
o2 w T
T
_
: E FT 4F
F : FT dF c : d:
Lm s FSF F 4
oCoC
oCoC
c F
FT :

6:5
6:6

The expression of the tangent modulus in the spatial formulation (6.5) coincides with the well-known
representation [33]:
c2

os
o2 w
4
;
og
ogog

6:7

if only the derivation in (6.7) is carried out with respect to the covariant components of the metric tensor
g gij gi
gj in the current conguration and it is assumed that the covariant vector bases of s sij gi
gj
are independent of gij . This leads to
c

os osij
oS T
o2 w

F : FT
F F
FT :
: FT
F;
gi
gk
gl
gj F
og ogkl
oC
oCoC

6:8

which is evidently identical with (6.5). Such a derivative of a tensor, where its basis is kept constant, is
analogical to the Lee derivative. Otherwise, the dierentiation with respect to the metric tensor in the spirit
of the denition (2.31) would lead to a meaningless result, since g presents the identity tensor.
Material and spatial formulations presented are valid both for isotropic and anisotropic materials. In
literature there exists also a spatial representation of the elastic modulus in term of the left CauchyGreen
tensor b proposed by Miehe [23] for isotropic materials
c 4b

o2 w
b;
obob

Lm s c : d:

6:9

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

433

In the following we show, that the denition of the material tensor 6:91 leads to the rate relation 6:92 if
only one assumes that the tensor b is not symmetric while derivating with respect to b. To prove this we rst
transform the elastic modulus 6:91 to the proposed denition (2.31) of the tensor derivative. Then
c 4b
I :

o2 w
o2 w
: I
b 4I
b :
: b
I;
obob
obob

6:10

such that
o2 w
c 4 b brl i k gm
gk
gr
gj
obj obl
mi

!R
4bmi brl

o2 w
o2 w
b:
g
gj
gk
gr 4b
i
k m
obob
obj obl

Using the constitutive relation for isotropic materials in the form s 2w;b b 2bw;b and by virtue of
(2.33), (5.20) and (3.11) one can obtain:
h
i
i
h
0
0
Lm s s;b : b_ ls slT 2I
w;b 2bw;bb : lb 2w;b
I 2w;bb b : blT ls slT
1
1
4I
b : w;bb : b
I : d blT w;b w;b lb ls slT c : d blT w;b w;b lb ls slT :
2
2
On the contrary, assuming b is not symmetric in the dierentiation and using the corresponding denition
of the tensor derivative (2.31) we receive:
Lm s 2I
w;b 2bw;bb : lb 2w;b
I 2w;bb b : blT ls slT
4I
b : w;b : b
I : d 2lT bw;b 2w;b bl ls slT c : d:
It can easily be proved that the tangent moduli C (6.2) and c (6.5) are super-symmetric according to their
denitions. For C it directly follows from (5.10). As to c we have by using (2.9), (3.9) and (3.10)
 2 T

T
o2 w
ow
: FT
F FT
FT :
: F
FT T c;
cT F
FT :
oCoC
oCoC
"
#t
t
t
o2 w
o2 w
T
T
T
: F
F F
F :
: F
F
c F
F :
oCoC
oCoC
 2
tt
ow
T
T
: F
F c:
F
F :
oCoC

6:11

6:12

Similarly, the symmetry can also be proved for the other elastic modulus
A

o2 w
;
oFoF

A AT ;

6:13

which is used in literature, too [33].

The super-symmetry of the elastic moduli C (6.2) and c (6.5) permits their spectral decomposition in the
form (4.9). For example we get for C
C

6
X
r1

kr M r  M r ;

Mr MTr :

6:14

Now, the question is what properties the elastic modulus corresponding to an isotropic material possesses
and how the isotropy can be identied by tangent moduli. This problem can easily be solved for the natural
state, where b C I. In this case the tensor function dSdE C : dE should satisfy the isotropy condition:
QdSQT C : QdEQT

6:15

434

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

for an arbitrary strain variation dE and an arbitrary orthogonal tensor Q QT . Using the decomposition
of dE in terms of Mr :
dE

6
X

dEr Mr ! QdEQT

r1

dS

dEr QMr QT

6:16

r1

we obtain then from (6.14):

6
X

6
X

kr M r  Mr

6
X

r1

!
dEs Ms

s1

6
X

kr dEr Mr :

6:17

r1

Under consideration of (6.17) the condition (6.15) takes then the form:
6
X

kr dEr QMr QT

r1

6 X
6
X
r1



kr dEs Mr : QMs QT Mr :

6:18

s1

Contracting the left- and right-hand sides with Mt yields

6
X

6
6
X

 X




kr dEr Mt : QMr QT
kt dEr Mt : QMr QT !
kt kr dEr Mt : QMr QT 0

r1

r1

8dEr

and

r1

8Q Q :

6:19

Since eigentensors Mt are symmetric the condition (6.19) can be satised only in two following cases:
1: k1 k2


k6 2l ! C 2lI 0 ;
1
2: k1 3k; M1 p I; k2


k6 0 ! C kI  I;
3

6:20
6:21

where 2l and k denote material constants. Generally we can state that an elastic modulus corresponds to an
isotropic material in the natural conguration, if and only if C can be represented via linear combination of
the tensors I 0 and I  I in the form of the St. VenantKirchho model (see also [14,15,17,19,20,29,31]):
C 2lI 0 kI  I:

6:22

The analogous statement for the constitutive relations has been proved by Ciarlet [10].
7. Examples of elasticity tensors
Using the results (6.1), (6.2), (6.4) and (6.5) of the previous section as well as tensor algebra operations
and dierentiation rules (3.5), (3.6), (3.9), (3.10), (3.13)(3.15), (3.22), (3.23), (3.28), (3.31)(3.33), (5.21) and
(5.22), we derive here tangent tensors in material and spatial descriptions for a number of hyperelastic
material models intensively used in continuum mechanics.
1. St.VenantKirchhoff material model:
1
2
2
w jtr E l trdevE ;
2

7:1

k and l being the compression and shear moduli respectively.

Material description:
2

ow
otrdev E odev E
jtr EI l
:
jtr EI 2l dev E;
oE
odev E
oE


oS
1
jI  I 2l I 0 I  I :
C
oE
3

7:2
7:3

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

435

Spatial description:
1
s FSFT jtr b 3b lb dev b;
2



1
0
T
T
c F
F : C : F
F jb  b 2l b
b b  b :
3

7:4
7:5

2. MooneyRivlin material model [25]:

w C1 IC 3 C2 IIC 3 C1 Ib 3 C2 IIb 3;

7:6

where C1 and C2 are material constants.

Material description:
ow
2C1 I 2C2 IC I C;
oC
oS
4C2 I  I I 0 :
C2
oC
S2

7:7
7:8

Spatial description:
ow
b 2C1 b 2C2 IC b b2 ;
ob
h
i
0
c F
FT : C : FT
F 4C2 b  b b
b :
s FSFT

3. Ogden material model [27]:

X lr a


k1r ka2r ka3r 3 ;
wki
ar
r

7:9
7:10

7:11

where ki denote the eigenvalues of the right stretch tensor U and ar ; lr are material parameters.
Material description:


S 2 aI aII IC I aII C aIII IIIC C1 ;
h
C 4 cII : 2cI II IC cII II I2C bII I  I cII II C  C IIIC bIII cIII III IIIC C1  C1
cI II cII II IC I  C C  I IIIC cI III cII III IC I  C1 C1  I
i
0
IIIC cII III C  C1 C1  C bII I 0 IIIC bIII C1
C1 :
Spatial description (cf. [18]):


s 2 aI aII Ib b aII b2 aIII IIIb I ;
h
c 4 cII : 2cI II Ib cII II I2b bII b  b cII II b2  b2 IIIb bIII cIII III IIIb I  I
cI II cII II Ib b  b2 b2  b IIIb cI III cII III Ib b  I I  b
i
0
IIIb cII III b2  I I  b2 bII b
b IIIb bIII I 0 :

7:12

7:13

7:14

7:15

Using the alternative representation form of the energy function w wIC ; IIC ; IIIC the unknown coecients appearing in (7.12)(7.15)
a K bK

ow
;
oK

cKL

o2 w
;
oKoL

K; L IC ; IIC ; IIIC

7:16

can be specialized for an arbitrary isotropic material model. For the ValanisLandel hypothesis [40] and its
special case Ogden model (7.11) these coecients have been determined by Bas ar and Itskov [1] for the
problems.
cases of distinct as well as equal eigenvalues without explicit solving the eigenvalue

436

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

4. Material models with decoupled volumetricisochoric response. For these models the energy function is
assumed to be additively split into volumetric and isochoric parts:
7:17

w U J W C;
where
J IIIC 1=2 ;

C J 2=3 C:

7:18

Accordingly the corresponding stress tensors and elastic moduli permit the additive decomposition:
S Svol Siso ;
s svol siso ;

C Cvol Ciso ;
c cvol ciso

7:19
7:20

and take the following form:

Material description:
Svol 2

oU
U 0 J C1 ;
oC

7:21

Siso 2




oW
oW oC
S
 iso ;
:
2
Siso : p dev Siso C C1 C1 dev C
oC
oC oC

7:22

0


oSvol 00 2
U J U 0 J C1  C1 2U 0 J C1
C1 ;
oC

7:23

Cvol 2




0

oSiso
oSiso
2C1 P0 : C
I
Siso
: p 2C1
C1 0 dev C Siso
oC
oC
h

0 i

0

1
: p 2 C1
Siso ;
pT : Ciso : 2 C :
Siso

Ciso 2

7:24

where
Siso 2

oW
;
oC

Ciso 2

oSiso
o2 W
;
4
2
oC
oC

7:25

oC
oJ 2=3
1
J 2=3 I 0 C  C1 J 2=3 I 0 :
C
p
oC
oC
3
Spatial description:
svol FSvol FT U 0 J I;

7:26

siso FSiso F devsiso ;

7:27

00 2

cvol F
F : Cvol : F
F U J U J I  I 2U J I ;
0

ciso F
F : Ciso : F
F P : ciso 2I
siso : P 2I
siso
2
2
P0 : ciso : P0 tr siso P0 siso  I I  siso ;
3
3

7:28
0

7:29

where
T

siso F Siso F ;

ciso F
F : Ciso : F
F;

F J 1=3 F:
R

7:30

It can easily be seen that after the transformation


the expressions (7.23) and (7.24) coincide with
those given by Simo and Taylor [34]. In the spatial description we have forgone to use the denition (6.9)

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

437

for the isochoric moduli ciso . For this reason the representation (7.26)(7.29) holds in contrast to the
analogous result by Miehe [23] not only for isotropic but also for anisotropic materials.
8. Conclusion
In the paper we have presented a theory of fourth-order tensors. The theory enables to solve many
relevant problems in the treatment and application of fourth-order tensors, which are still open in the
literature.
Starting with the set of notations and denitions we have presented in absolute notation the tensor
algebra and dierentiation rules, which are applied then to isotropic tensor functions. The important result
to be emphasised is a closed-formula solution for the derivative of an isotropic tensor function, which is
given in terms of eigenvalue bases for the cases of distinct and equal eigenvalues. This solution has a simpler
and more compact form in contrast to those known in the literature.
The next important topic are the eigenvalue problems and the spectral decomposition of a fourth-order
tensor. It is shown that the eigenvalue problem of a fourth-order tensor can be reduced to that one of its
matrix representation and then easily solved. This leads to a spectral decomposition involving nine eigenvalues and nine corresponding eigentensors. In the case of super-symmetry characterizing for example
some tangent moduli, the spectral decomposition can be given through only six symmetric eigentensors.
Concerning the invariants of a fourth-order tensor we have come to their total number of 27, which can be
obtained using various contraction rules with a second-order tensor.
A most important application of the fourth-order tensors are tangent moduli. The tensor algebra and
dierentiation rules presented simplify considerably the formulation and derivation of tangent moduli,
which has been demonstrated on examples of some hyperelastic material models. The elastic moduli are
obtained in a material as well as in a spatial description and compared with the results available in the
literature.
Acknowledgements
The nancial support of the German National Science Foundation (DFG) is gratefully acknowledged.
References
[1] Y. Bas ar, M. Itskov, Finite element formulation of the Ogden material model with application to rubber-like shells, Int. J. Numer.
Engrg. 42 (1998) 12791305.
Methods
[2] J. Betten, Interpolations methods for tensor functions, in: X.J.R. Avula et al. (Eds.), Mathematical Modelling in Science and
Technology, Pergamon Press, New York, 1984.
[3] J. Betten, Invariants of fourth-order tensors, in: J.P. Boehler (Ed.), Applications of Tensor Functions in Solid Mechanics, CISM
courses and lectures No. 292, Wien, 1987, pp. 171186.
[4] J. Betten, Tensorrechnung f
ur Ingenieure, Teubner, Stuttgart, 1987.
[5] J. Betten, W. Helisch, Irreduzible Invarianten eines Tensors vierter Stufe, Z. Angew. Math. Mech. 72 (1992) 4557.
[6] J. Betten, W. Helisch, Simultaninvarianten bei Systemen zwei- und vierstuger Tensoren, Z. Angew. Math. Mech. 75 (1995)
753759.
[7] J. Betten, W. Helisch, Tensorgeneratoren bei Systemen von Tensoren zweiter und vierter Stufe, Z. Angew. Math. Mech. 76 (1996)
8792.
[8] P. Chadwick, R.W. Ogden, On the denition of the elastic moduli, Arch. Rat. Mech. Anal. 44 (1971) 4153.
[9] P. Chadwick, R.W. Ogden, A theorem of tensor calculus and its application to isotropic elasticity, Arch. Rat. Mech. Anal. 44
(1971) 5468.
[10] P.G. Ciarlet, Mathematical elasticity, Volume I: Three-dimensional elasticity, North-Holland, Amsterdam, 1988.
[11] S.C. Cowin, M.M. Mehrabadi, On the structure of the linear anisotropic elastic symmetries, J. Mech. Phys. Solids. 40 (1992)
14591472.
[12] P. Dluzewski, P. Rodzik, Elastic eigenstates in nite element modelling of large anisotropic elasticity, Comput. Methods Appl.
Mech. Engrg. 160 (1988) 325335.
[13] T.C. Doyle, J.L. Ericksen, Nonlinear elasticity, Advances in Applied Mechanics IV, Academic Press, New York, 1956.
[14] M.E. Gurtin, The linear theory of elasticity. in: S. Flugge (Ed.), Handbuch der Physik A, Vol. VI a/2, Springer, Berlin, 1972.

438
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]

M. Itskov / Comput. Methods Appl. Mech. Engrg. 189 (2000) 419438

M.E. Gurtin, A short proof of the representation theorem for isotropic, linear stress-strain relations, J. Elasticity 4 (1974) 243245.
M. Itskov, The derivative with respect to a tensor: some theoretical aspects and applications, Z. Angew. Math. Mech., in press.
J.P. Jaric, On the representation of symmetric isotropic 4-tensors, J. Elasticity 51 (1998) 7379.
M. Kaliske, H. Rothert, On the nite element implementation of rubber-like materials at nite strains, Engrg. Comput. 14 (1997)
216232.
J.K. Knowles, On the representation of the elasticity tensor for isotropic materials, J. Elasticity 39 (1995) 175180.
L.C. Martins, P. Podio-Guidugli, A new proof of the representation theorem for isotropic linear constitutive relations, J. Elasticity
8 (1978) 319322.
M.M. Mehrabadi, S.C. Cowin, Eigentensors of linear anisotropic elastic materials, Quart. J. Mech. Appl. Math. 43 (1990) 1541.
C. Miehe, Computation of isotropic tensor functions, Commun. Numer. Methods Engrg. 9 (1993) 889896.
C. Miehe, Aspects of the formulation and nite element implementation of large strain isotropic elasticity, Int. J. Num. Methods
Engrg. 37 (1994) 19812004.
C. Miehe, Comparison of two algorithms for the computation of fourth-order isotropic tensor functions, Comput. Struct. 66
(1998) 3743.
M. Mooney, A theory of elastic deformations, J. Appl. Phys. 11 (1940) 582592.
K.N. Morman, The generalised strain measure with application to non homogeneous deformations in rubber-like solids, J. Appl.
Mech. 53 (1986) 726728.
R.W. Ogden, Non-linear Elastic Deformations, Ellis Horwood Series in Mathematics and its Applications, Chichester, 1984.
J. Rychlewski, On the Hooke's law, Prikl. Matem. Mekhan. 48 (1984) 303314.
J. Rychlewski, Symmetry of tensor functions and spectrum theorem (in Russian), Uspechi Mechaniki (Advances in Mechanics)
11(3) (1988) 77124.
J. Rychlewski, Unconventional approach to linear elasticity, Arch. Mech. 47 (1995) 149171.
K.A. Pericak-Spector, S.J. Spector, On the representation theorem for linear isotropic tensor functions, J. Elasticity 39 (1995)
181185.
C. Sansour, On the spatial description in elasticity and the DoyleEricksen formula, Comput. Methods Appl. Mech. Engrg. 107
(1993) 239249.
J.C. Simo, J.E. Marsden, On the rotated stress tensor and the material version of the DoyleEricksen formula, Arch. Rat. Mech.
Anal. 86 (1984) 213231.
J.C. Simo, R.L. Taylor, Quasi-incompressible nite elasticity in principal stretches. Continuum basis and numerical algorithms,
Comput. Methods Appl. Mech. Engrg. 85 (1991) 273310.
S. Sutclie, Spectral decomposition of the elastic tensor, J. Appl. Mech. 59 (1992) 762773.
T.C.T. Ting, Determination of C1=2 ; C1=2 and more general isotropic tensor functions of C, J. Elasticity 15 (1985) 319323.
T.C.T. Ting, Invariants of anisotropic elastic constants, Quart. J. Appl. Mech. 40 (3) (1987) 431448.
R. Trostel, Vektor- und Tensoralgebra, Vieweg, Wiesbaden, 1993.
C. Truesdell, W. Noll, The nonlinear eld theories of mechanics, in: S. Fl
ugge (Ed.), Handbuch der Physik, Vol. III/3, Springer,
Berlin, 1965.
K.C. Valanis, R.F. Landel, The strain-energy function of a hyperelastic material in terms of the extension ratios, J. Appl. Phys. 38
(1967) 29973002.
L.J. Walpole, Fourth-rank tensors of the thirty-two crystal classes: multiplication tables, Proc. R. Soc. Lond. A391 (1984)
149179.
Q.-S. Zheng, J. Betten, On the tensor function representations of 2nd-order and 4th-order tensors, Part I. Z. Angew. Math. Mech.
75 (1995) 269281.