Professional Documents
Culture Documents
Itskov - On The Theory of Fourth-Order Tensors and Their Applications in Computational Mechanics (CMAM)
Itskov - On The Theory of Fourth-Order Tensors and Their Applications in Computational Mechanics (CMAM)
www.elsevier.com/locate/cma
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
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],
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.
2:1
2:2
2:3
2:4
D A B Aij gi
gj Bkl gk
gl Aij Bkl 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
k
k j
gi
gj
gk
gl Dijl
g
gi
gl
gk ;
DT Dijl
2:7
k
k
gi
gj
gk
gl t Dijl
gi
gk
gj
gl :
Dt Dijl
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
2:10
and
E t E:
2:11
422
2:12
2:13
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 : Amn gm
gn Dijl
Ajk gi
gl ;
D : A Dijl
2:15
k
k j
gi
gj
gk
gl Ail Dijl
g
gk :
A : D Amn gm
gn : Dijl
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 Dijl
ns gm
g
gr
g Djl Ank 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 :: Amnsr gm
gn
gr
gs Dijl
Aik
D :: A Dijl
2:19
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
2:24
2:25
423
2:26
D1 D;
D2 D : D; . . .
Dn D
: D{z
: D
|
} :
2:27
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
D11 D111
D111
D121
D121
D121
D131
D131
D131
1
2
3
1
2
3
1
2
3 7
6 D112
D112
D112
D122
D122
D122
D132
D132
D132
7
6
6 D1 1 D1 2 D1 3 D1 1 D1 2 D1 3 D1 1 D1 2 D1 3 7
6 13
13
13
23
23
23
33
33
33 7
6 21
2
3
1
2
3
1
2
3 7
7
6 D11 D211
D211
D221
D221
D221
D231
D231
D231
7
6 21
22
23
21
22
23
21
22
23 7
6
2:29
D 6 D12 D12 D12 D22 D22 D22 D32 D32 D32 7:
6 D2 1 D2 2 D2 3 D2 1 D2 2 D2 3 D2 1 D2 2 D2 3 7
6 13
13
13
23
23
23
33
33
33 7
6 31
2
3
1
2
3
1
2
3 7
7
6 D11 D311
D311
D321
D321
D321
D331
D331
D331
7
6
4 D3 1 D3 2 D3 3 D3 1 D3 2 D3 3 D3 1 D3 2 D3 3 5
12
1
D313
12
2
D313
12
3
D313
22
1
D323
22
2
D323
22
3
D323
32
1
D333
32
2
D333
32
3
D333
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
oAij gi
gj oAij
2:30
g
gj
g g
gj
g
gl :
j
oAij
oAkl oAij
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 Aij gi
gj Al gk
gl ) Al gk gi Aij gj gl ;
k
oa i
oa oAl i
oa k
oa k
g
gj
g gi gj gl gi
gj
g
gl :
i g
gj
k
k
k
oA
oAij
oAl
oAl
oAl
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
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 Cij Gi
Gj with respect to another second-order one
A Akl gk
gl (2;2 ) is dened by
C;A
oCij Gi
Gj
oAkl gk
gl
oCij
oAkl
G i
gk
gl
G j
2:31
C;A
oCij
oAkl
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 oAij
2:32
3
3
oCij
oCij
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
oCij
1 X
Gi
gk
gl gl
gk
Gj :
2 k;l 6 k oAkl
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.
425
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 Aij Bkl gi
gj
gk
gl : Cmn Drs gm
gn
gr
gs
Aij Bkl Cjn Drk 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
oAij
oAkl
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
Cij
;
trC : D trD : C C : D :: I C :: D D :: C Dikl
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
3:18
R
I IR gi
gj
gj
gi gi
gi
gj
gj I
I:
426
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
oAij
oAkl
gj
gk
gl
gi dik dlj gj
gk
gl
gi I t T;
3:19
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
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
oAij Bjk gi
gk
oClm gl
gm
oAij
oClm
oAij Bjk
oClm
gi
gl
gm
gk
Bjk gi
gl
gm
gk Aij
oBjk
gi
gl
gm
gk A;C B AB;C :
oClm
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:
Bn ;B
n1
X
Br1
Brn ;
427
3:29
r0
Bn ;B : I nBn1 :
3:30
3:31
oaAij gi
gj
oBkl gk
gl
oaAij
oBkl
oAij
oa i
k
j
gi
g
gl
g k Aj gi
g
gl
g a k gi
gk
gl
gj
oBl
oBl
k
A a;B aA;B ;
A : B;C
oAij Bij
oCkl gk
gl
oAij
oB j
Aij ik Bij k
oCl
oCl
!
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 oAj k
g
gl f ;A : A;B ;
oAij oBkl
3:32
oAij oCkl
m
j
m gi
g
gn
g A;C : C;B :
oCkl oBn
3:33
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
13
4:3
33
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
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
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
5:1
T
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
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
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 :
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
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:
431
5:15
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
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
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
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
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
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
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
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
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
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
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
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
7:6
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
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
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
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
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)
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]