MODERN THEORY OF VECTORS AND TENSORS IN

MECHANICS AND ENGINEERING

Heng Xiao, Otto T. Bruhns and Albert Meyers

2
Contents

1 Vectors and Vector Spaces 7

1.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.1 Motivation and examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.2 Definition as geometric entities . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.1 Null vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.2 Addition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.3 Multiplication by number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.4 Scalar Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.5 Vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.6 Vector product and mixed product . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3 Algebraic Expressions: Standard Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.1 Unit vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.2 Plane case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.3 Standard expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.4 Indicial notations and conventions . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3.5 Transformation rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.4 Algebraic Expression: Arbitrary Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.4.1 Co- and contravariant expressions . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.4.2 Transformation rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5 Remarks: vectors as basis-free entities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2 Second-Order Tensors 25
2.1 Motivation and Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.1 Deformation state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.2 Stressed state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Second-Order Tensors as Linear Transformations . . . . . . . . . . . . . . . . . . . . . . 28
2.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2.2 Multiplication by number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.3 Addition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.4 2nd-order tensor spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.5 Composite product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2.6 Transposition, non-singularity and singularity . . . . . . . . . . . . . . . . . . . . 30
2.3 Algebraic Expressions: Standard Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3.1 Dyadic Product of two vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3.2 Standard expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3.3 Transformation rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.4 Algebraic Expressions: Arbitrary Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.4.1 Co-, contra- and mixed-variant expressions . . . . . . . . . . . . . . . . . . . . . 33
2.4.2 Transformation rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.5 Characteristic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5.1 Eigenvalues and eigenvectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3
4 CONTENTS

2.5.2 Characteristic equation and invariants . . . . . . . . . . . . . . . . . . . . . . . . 35

2.5.3 Cayley-Hamilton theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.6 Symmetric 2nd-order Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.6.1 Explicit formula for eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.6.2 Characteristic expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.6.3 Eigenprojections and Sylvester’s formula . . . . . . . . . . . . . . . . . . . . . . 41
2.7 Orthogonal Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.7.1 The unit axial vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.7.2 Canonical expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.7.3 Inversion, π-rotation and reflection . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.8 Skew-symmetric 2nd-Order Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.8.1 The associated axial vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.8.2 Dual and canonical expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.9 Decompositions of 2nd-Order Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.9.1 Additive decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.9.2 Polar decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

3 Tensors of Higher Order 51

3.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2 Tensor Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3 Algebraic Expressions: Standard Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.4 Algebraic Expressions: Arbitrary Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.5 Dot Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.6 Orthogonal Transformations of Vectors and Tensors . . . . . . . . . . . . . . . . . . . . . 56
3.7 Third-Order Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.8 Fourth-Order Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.9 Component Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.10 On Vectors and Tensors in General Sense . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.10.1 Abstract vectors and tensors in general sense . . . . . . . . . . . . . . . . . . . . 60
3.10.2 Characteristic expressions of 4th-order tensors . . . . . . . . . . . . . . . . . . . 61

4 Scalar, Vector and Tensor Fields 65

4.1 Motivation and Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2 Continuity and Gradients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.3 Curly, Divergent and Laplacian Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.4 Differentiation in Cartesian Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.5 Differentiation in Curvilinear Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.5.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.5.2 Curvilinear coordinate systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.5.3 Orthogonal curvilinear coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.5.4 Cylindrical polar coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.5.5 Spherical polar coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.6 Integrations of Scalar and Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.6.1 Gauss’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.6.2 Stokes’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

5 Tensor Functions 83
5.1 Motivation: constitutive functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.2 Material Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.2.1 Structural elements and their symmetries . . . . . . . . . . . . . . . . . . . . . . 84
5.2.2 Orthogonal groups and cylindrical groups . . . . . . . . . . . . . . . . . . . . . . 85
5.2.3 Crystal classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.2.4 Quasicrystal classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.2.5 Isotropic and anisotropic functions of vectors and 2nd-order tensors . . . . . . . . 90
CONTENTS 5

5.3 Scalar-Valued Functions of Symmetric Tensor . . . . . . . . . . . . . . . . . . . . . . . . 91

5.3.1 Isotropic invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.3.2 Anisotropic invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.3.3 Continuity and Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.4 Symmetric Tensor-Valued Functions of Symmetric Tensor . . . . . . . . . . . . . . . . . 97
5.4.1 Isotropic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.4.2 Anisotropic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.4.3 Continuity and Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.5 Eigenprojection Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.5.1 Characteristic expression of perturbation tensor . . . . . . . . . . . . . . . . . . . 105
5.5.2 Derivative in terms of eigenbasis . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.5.3 Derivatives in terms of eigenprojections . . . . . . . . . . . . . . . . . . . . . . . 107
5.5.4 Summary for the main ideas and procedures . . . . . . . . . . . . . . . . . . . . . 108
5.6 Continuously Differentiable Representation . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.6.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.6.2 A new notion of representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.6.3 An orthogonalized representation . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.7 On General Isotropic and Anisotropic Functions . . . . . . . . . . . . . . . . . . . . . . . 112
5.7.1 Functional bases, generating sets and irreducibility . . . . . . . . . . . . . . . . . 112
5.7.2 Isotropic extension method via structural tensors . . . . . . . . . . . . . . . . . . 114
5.7.3 New isotropic extension method and unified theory . . . . . . . . . . . . . . . . . 115
References
6 CONTENTS

Suggestions for Learning Things New to You

1) You know where they come from

2) You know what they are and what they mean
3) You know why they should be introduced
4) You know how to describe and express them
5) You know how to do various operations concerning them
6) You know what these operations mean and why they are really useful
7) You know how to use them to understand and solve meaningful questions
and eventually
8) You become expert at using them and understanding them

Above all, please always ask the following questions:

WHAT IS THE PURPOSE? WHY THIS PURPOSE? WHAT IS THE IDEA?

when studying concepts and results. Do not try hard to simply bear them in mind before you know the
answers for the above questions!
Chapter 1

Vectors and Vector Spaces

The notion of vector itself characterizes and represents various kinds of physical and geometrical quan-
tities, as will be indicated below. Besides, as will be seen, vectors serve as building blocks in defining,
constructing and understanding more sophisticated objects, i.e., tensors. Here, we focus attention mainly
on vectors defined over a 3-dimensional Eulclidean space.
There are different ways in introducing and defining vectors and their operations. In a 3-dimensional
Euclidean space, however, it is possible from the beginning to introduce vectors and their algebraic oper-
ations in a perhaps more straightforward and accessible manner. In our treatment, bases and components
will not be involved at all at the first stage, and they will be used in subsequent stages to facilitate repre-
sentations and operations of vectors from an algebraic standpoint.

1.1 Vectors
When we are invited to learn things new to us, naturally we may ask: What are they? Why are they
introduced and needed? How can we understand and use them? and so on. In what follows we attempt to
answer these questions concerning the basic objects discussed here, i.e., vectors and then tensors.

1.1.1 Motivation and examples

In everyday life, we hear of, talk about and calculate with many different kinds of quantities. In a shopping
mall or a supermarket, one says “the bananas here look fresh and wonderful. I need 4 kilograms. How
much should I pay? Last year it was 3.50 DM/kg, and now it costs 3.50 Euro/kg. So I have to pay 14 Euro
now, instead of 14 DM.” In TVs, weather reports supply us with detailed information about future weather
conditions at home and abroad, such as “It is twenty hours now. The weather forecast is as follows: Berlin
-6◦ /3◦ , cloudy, humidity 83; Beijing -10◦ /-2◦ , overcast, humidity, 75; Sydney 15◦ /23◦ , sunny, humidity 60;
· · ·.” (Okay, in these days it would be very nice to spend a holiday in Sydney)
Several quantities are mentioned in the above examples, i.e., currency, weight, time, temperature, and
humidity. Each of them can be fully characterized by a number with or without a suitable unit. They
are called scalars. In particular, scalars without any unit, i.e., pure numbers, are perhaps among the most
familiar to us. Of course, we are already familiar with many kinds of scalars. We know what they are and
what they mean, we know how to do operations with them, and we know how to handle and use them.
Also even in everyday life, however, we encounter other quantities. A weather report may say “ to-
morrow there are going to be 80km/h northwest winds in Berlin, 60 km/h northeast winds in München, 35
km/h west winds in Bochum, · · ·”, as schematically shown in Fig. 1.1.
On the other hand, when you take a long-distance travel by an airplane, at different hours you may
notice that the plane is flying in different directions (e.g., to Frankfurt, Budapest, Prag, Warsaw, Paris,
London, etc.) at a constant velocity (e.g., 800km/h) or at different velocities, as schematically shown in
Fig. 1.2.

7
8 CHAPTER 1. VECTORS AND VECTOR SPACES

DK
Arkona

5 3 5


Sylt




Helgoland


5


Kiel




Rostock


5
4 Schwerin 

Stettin

 

Emden

 

Hamburg
 

4
 

 
4 4
4
 

Bremen

Berlin



PL
 

 

 

5
 
 

NL Hannover
3 

4
=



<=

2 Magdeburg 5
<

Münster


 

 

 

 

Cottbus
:;

 
 

4 Leipzig
:

Kassel 4 4




89

Bonn Erfurt Dresden

B 3 3
3 CZ
3
7

67

L Frankfurt
5 6

4 45

Trier Nürnberg
$%

4 4

Würzburg
" "
# #

" "
# #

4
2 2
3 3

3
2 2
3 3

Saarbrücken 4 '

&'

4 ,
-
,
-

&

Regensburg
5
, ,
- -

Stuttgart
F München )

3
Freiburg
()

2
/

2
./

A
Konstanz
1

* *

+ +

3
01

* *
+ +

Basel 4

CH

Figure 1.1: Possible wind velocity distributions across Germany

Stockholm

Moskva

London
Berlin Warszawa

Paris

Madrid Roma

Figure 1.2: Flight status of an airplane along its course

1.1. VECTORS 9

A more familiar case is when you meet your friends at Uni-Center. Perhaps your friends and you may
ask “where are you going?” among yourselves. Possibly, the answers may be like, “I am going from here
to Plus”,“I am going from here to Globus”, “I am going from here to the library”, etc. This case is shown
in Fig. 1.3.

Shopping Mart
    
Cinema
              
         
       
    
              
              
              
 
 
  
  
  
  
 
                       
Library

                                
                      
                                 
                                
         
                        
       
                        
                                

  

    

  
      

Figure 1.3: Displacements starting from the same point

These examples involve quantities essentially different from scalar quantities. Yes, each of them has a
magnitude, like a scalar quantity, but it includes more than a magnitude, i.e., it simultaneously refers to a
direction. Because such quantities even with the same magnitude do not have the same meaning at all, they
cannot be represented and characterized by scalars. In mechanics and engineering and other fields, there
are a great number of basic physical and geometrical quantities which are endowed with both magnitudes
and directions. As a result, a new notion for their characterizations and representations would become
necessary.
This new notion, i.e. vectors, has been invented by mathematicians, physicists and engineers from their
different backgrounds. In a number of scientific and technological fields, e.g., mechanics and engineering,
basic geometrical and physical features of the investigated objects and their behaviour are characterized and
represented by vectors and tensors, together with the usually known scalar quantities. To study and answer
meaningful and challenging topics in mechanics and various engineering fields, as one of the coherent and
constituent steps it would be instrumental and necessary to understand and use vectors and tensors and
operations concerning them.

1.1.2 Definition as geometric entities

By a vector we mean an entity (a quantity) that is endowed with both an assigned magnitude of scalar and
an assigned direction.
Usually, scalars are designated by light-face letters, such as α, β, x, y, etc. To distinguish these nota-
tions, vectors are often represented by bold-face letters, such as a, b, c, u, v, etc.
It should be emphasized that a vector is an entity formed by the inseparable combination of an assigned
magnitude and an assigned direction. We hope to have a further, more direct understanding of this entity
with two essentially different features: magnitude and direction. It seems better to embody this notion in
visual geometric objects. It is indeed possible and proves quite simple and fruitful.
In a 3-dimensional Euclidean point space, we draw a line segment with an arrowhead at one end, as
shown in Fig. 1.4. Then it may become clear that such an arrowed segment exactly possesses all the
features of a vector. Indeed, its length may be used to represent the magnitude, while its arrowhead with
the spatial position of this line segment may be used to indicate the direction. With this understanding, the
magnitude and the direction are, in a natural and inseparable manner, combined into a geometric entity:
arrowed segment. Thus, we achieve a natural geometric embodiment or representation of the notion of
vector.
Now we may say that a vector is an arrowed segment with its length representing the magnitude and
with its arrowhead indicating the direction.
10 CHAPTER 1. VECTORS AND VECTOR SPACES

Direction

Symbol
v

Length

Figure 1.4: Vectors as geometric entities

1.2 Vector Spaces

According to their physical backgrounds etc., various different kinds of vectors may be introduced and
they are given different corresponding names. The examples given before provide velocity vectors (see
Figs. 1.1 and 1.2) and displacement vectors (Fig. 1.3). Other examples are in abundance, such as force
vectors, acceleration vectors, angular velocity vectors, heat flux vectors, electric field vectors, magnetic
field vectors, and so on. In accordance with their different physical backgrounds, their magnitudes have
different physical dimensions.
With various different kinds of vectors, we like to collect all those from a common background (physical
dimension and physical feature), respectively. This idea naturally generates different collections of vectors.
More essentially, then for each collection we further interrelate the vectors in it with each other by means
of certain simple operations resulting from geometric and physical considerations. In so doing, each such
collection of vectors with these simple operations turns out to become a creation with a both simple and
elegant structure, called a vector space. This will become clear in the course of subsequent development.
A further explanation will be given in §1.2.5.

1.2.1 Null vector

Let us imagine a vector with its length invariably becoming smaller and smaller. Then it may readily be
understood that the limit of this process will finally produce a vector with zero length, as shown in Fig. 1.5.

zero length

Figure 1.5: Null vector with vanishing length

This vector is called null (zero) vector and will be designated by a particular notation 0 . It is a particular
yet important vector and will play a similar role in basic vector operations as the number zero does in basic
number operations, as will be seen below.

1.2.2 Addition
We know how to add up several numbers to produce a sum. We know this operation is basic and necessary
(imagine what if we do not know it at all) and we have been familiar with its simple rules since we learned
them in primary schools. For vectors, from physical considerations and others it is also basic and necessary
to have an operation called addition. A somewhat sweeping reason is that, without this and other basic
operations, one could not make any progress at all in using vectors and tensors as powerful notions and
1.2. VECTOR SPACES 11

tools to understand and solve meaningful questions in mechanics and engineering and other related fields.
Another reason is that these basic operations provide us with simple means to gain insight into certain
essential features of vectors.
There are physical backgrounds of introducing addition operation for vectors. Yes, let us consider a
body subjected to two or more forces (vectors). For example, the moon attracted by the universal gravita-
tions from the sun and the earth, as shown in Fig. 1.7. To know how it will move, Newton’s second law
tells us that we have to know the composition (sum) of these force vectors (here two but possibly more) in
Fig. 1.6.

The earth

The sun

The moon

Figure 1.6: The moon attracted by the sun and the earth

Another example is a small boat crossing a river. The river is constantly flowing at a velocity along its
course, and somebody sitting on the boat is always paddling it to the opposite bank at a constant velocity
in the direction normal to the water flow, as shown in Fig. 1.7.

Figure 1.7: A boat crossing the river

To know in what direction the boat will move, we must find out the composition (sum) of the velocity
vectors (here two but possibly more) in Fig. 1.7.
Now we proceed to study the composition or sum of two vectors and its rules. To this goal, a natural
way is to consider displacements. Now let us imagine a displacement from point A to point B, and then an
immediately following displacement from point B to point C. In this process, the displacements from A to
B and from B to C may be represented by vectors u and v , respectively. It may be clear that the composition
(sum) of the two displacements from A to B and then from B to C are equivalent to the single displacement
directly from A to C, represented by a . As a result, the vector a may be called the sum of the vectors u and
v and the latter is written as u + v . Then we have a = u + v . This is shown in Fig. 1.8.
Fig. 1.8 clearly shows how to adding up two vectors u and v . First, we draw vector u starting from
point A and setting its arrowhead at B; then, starting from the arrowhead B of vector u , we draw vector v
12 CHAPTER 1. VECTORS AND VECTOR SPACES

v
a
v
B

u
A

Figure 1.8: Triangle and parallelogram rules for vector addition

with its arrowhead at point C; and finally we draw another vector a by connecting points A and C and by
setting the arrowhead at point C. This process generates a triangle composed of vectors u , v and a , where
vector a is just the sum of vectors u and v , i.e., a = u + v .
The above process is called the triangle rule for vector addition. Another equivalent process is also
shown in Fig. 1.8 and generates a parallelogram with vector a one of its diagonals. This process is also
used and called the parallelogram rule for vector addition.
Either of the foregoing two rules defines the vector addition. It is well defined from both geometrical
and physical viewpoint. Indeed, it is also in full agreement with the observation on the composition of
force vectors and velocity vectors.
With either of the two aforementioned rules, it may be easy to confirm the following basic operation
rules for vector addition:
u + v = v + u (commutative rule) ; (1.1)
(uu + v ) + w = u + (vv + w ) = (uu + w ) + v ≡ u + v + w (associative rule) ; (1.2)
u +0 = u; (1.3)
for any given vectors u , v and w .
On the other hand, given a vector u , we may obtain a new vector by simply reversing the arrowhead
to the opposite end. This new vector is called the opposed vector of u and denoted −uu. A vector and its
opposed vector have the same length but their directions are opposed to each other. The following rule
holds:
−(−uu) = u , u + (−uu) = 0 . (1.4)
For scalars, we have subtraction operation. Then what about vectors? The answer is simple: it is a straight-
forward manner to define and do subtraction operation by means of addition operation for vectors. Indeed,
we can define
u − v ≡ u + (−vv) (1.5)
for any two vectors u and v . Namely, subtracting vector v from u is simply adding up the vector u and the
opposed vector, −vv, of vector v . The following rule holds:

u +v = w ⇐⇒ w −u = v. (1.6)

To understand (1.3) and (1.4)2 , we may imagine two limiting processes, respectively. One is that the length
of vector v in Fig. 1.8 is constantly becoming smaller and smaller, and, accordingly, the sum a = u + v
is becoming closer and closer to vector u . The other is that vector v in Fig. 1.8 is assigned to have the
same length as vector u and then is constantly rotating about point B inside the triangle, and, accordingly,
the length of the sum a = u + v is becoming smaller and smaller. With these limiting processes, (1.3) is
produced as the length of vector v finally tends to vanish, while (1.4)2 is generated as point C finally arrives
at point A.

1.2.3 Multiplication by number

Let α be a number and u be a vector. Then we like to know how to multiply the vector u by the number
a. It appears that the simplest and natural procedure is to produce a vector which is collinear with u and
the magnitude of which is |α| times the magnitude of u . Moreover, the direction of this new vector is
1.2. VECTOR SPACES 13

determined simply by the sign of the number α. Namely, a positive α gives the same direction as u , while
a negative α reverses the direction of u , as indicated in Fig. 1.9.

u
αu αu
( α <0 ) ( α >0 )

Figure 1.9: Multiplication by number α

The above procedures tell us how to multiply a vector u by a number α. The result is a vector, denoted
αuu and called the product of the number α and the vector u . The following rules may be obvious:

0uu = 0 , 1uu = u ; (−1)uu = −uu ; (1.7)

(αβ)uu = α(βuu) = β(αuu) (associative rule) ; (1.8)

for any given vector u and for any given numbers α and β.
The two operations may be combined. The following distributive rules hold:

(α + β)uu = αuu + βuu ; (1.9)

α(uu + v) = αuu + αvv . (1.10)

1.2.4 Scalar Product

Is it possible to multiply two vectors to yield a reasonable product? If yes, how can we do and what is
the motivation? It seems that such questions would be more profound than those for vector additions and
multiplication by number introduced before.
For vectors emerging in mechanics and engineering, two kinds of product operations may be intro-
duced. One produces scalars and is called scalar product, while the other generates vectors and is called
vector (cross) product.
Here we take the scalar product into account. We postpone the study of the vector product till §1.2.6.
The physical motivation of the scalar product is the work done by a force. Let a body undergo a displace-
ment (vector) u along a straight path, as shown in Fig. 1.10. Then a force (vector) v acting on it does work
during this displacement. This is one of the basic notions in mechanics (see, e.g., Bruhns and Lehmann
1993 for details). If the force vector v happens to be in the direction of the displacement vector u , the work
is simply given by the product of the lengths (magnitudes) of u and v . If the force v is perpendicular to the
displcament u , then no work is done. However, generally the force v may neither be parallel nor be normal
to the displacement u . Then, for any given two vectors u and v , we like to know how to define the product
of u and v that yields the work.
v v
u

Figure 1.10: Work done by a force

Motivated by the notion of work and its calculation, now we introduce the scalar product of two vectors.
Let |uu| and |vv| be the lengths (magnitudes) of vectors u and v , respectively, and θ be the angle between
vectors u and v , 0 ≤ θ ≤ π, which is formed by allowing the starting ends of u and v to meet together (see
1.8). Then the scalar product of vectors u and v , denoted u · v , is a scalar given below:

u · v ≡ |uu| · |vv| cos θ . (1.11)

14 CHAPTER 1. VECTORS AND VECTOR SPACES

Usually, we refer to |uu| cos θ (resp., |vv| cos θ) as the projection component of vector u (resp., v ) in the
direction of vector v (resp., u ). Then, a geometric explanation of the scalar product (1.11) is quite clear and
simple: it is the product of the length (magnitude) of vector u (resp., v ) and the projection component of
vector v (resp., u ) in the direction of vector u (resp., v ).
Let u and v be non-zero vectors, i.e., |uu| · |vv| 6= 0. Then we have
u ·v
cos θ = . (1.12)
|u | · |vv|
u

This indicates a geometric feature of the scalar product of two vectors: it also characterizes the angle
between two vectors. As a result, two vectors are said to be orthogonal whenever their scalar product
vanishes.
The following properties (rules) hold true for any number α and for any vectors u , v and w .
√
|uu| = u · u ; (1.13)

u ·u = 0 ⇐⇒ u = 0; (1.14)
u ·v = v ·u (commutative rule) ; (1.15)
(αuu) · v = u · (αvv) = α(uu · v ) (associative rule) ; (1.16)
(uu + v ) · w = u · w + v · w (distributive rule) . (1.17)
The above rules show that the scalar product of vectors is similar to the usual product of two scalars in
certain respects. However, there is an essential difference between them. There exist continued products
of three and even more scalars, whereas the same is not true for the scalar product of vectors. The main
reason is that the scalar product of two vectors is already a scalar by definition.

1.2.5 Vector spaces

An Euclidean vector space is a collection of vectors in which the addition and the multiplication by num-
ber and the scalar product, governed by rules (1.1)–(1.17), can be defined and in which the vectors are
interrelated to each other by these operations in a self-contained manner.
Note the expressions “interrelated” and “self-contained” in the above definition. With two or more
vectors, we can generate a vector by means of any operation of finite steps involving addition and multi-
plication by number. Moreover, the scalar product characterizes the main feature of the geometric relation
among vectors. With these, we may say that the vectors are interrelated to each other. By “self-contained”
we mean that the result (a vector) generated by any operation of finite steps should be incorporated.
Why is the notion of vector space introduced and how to understand this notion? This question has
been briefly explained at the outset of this section. Further explanations are as follows.
As mentioned earlier, there are various kinds of vectors with different geometrical and physical back-
grounds. One can collect them into groups according to many different ways. According to the above
definition, however, we collect vectors into groups by means of basic operations directly resulting from
geometrical and physical features of vectors and in so doing classify vectors into well-defined collections,
i.e., vector spaces. This classification not only enables us to understand simple and elegant algebraic struc-
ture of those vectors from a common background, as will be seen later, but lays foundation on further
developments of sophisticated quantities, i.e., tensors.
Roughly speaking, the construction of a vector space is as follows. From geometrical or physical
backgrounds, we introduce and define various kinds of vectors, such as position vectors, displacement
vectors, velocity vectors, force vectors, etc. Then, by means of the basic operations governed by rules
(1.1)–(1.17) we collect all the vectors with the same background together, separately. In so doing we obtain
collections of vectors and then we introduce basic algebraic operations such as addition and multiplication
by number, etc. We call each of them with these operations a vector space. As a result, we have various
kinds of vector spaces. From different backgrounds we may give different names to them, such as position
vector space, displacement vector space, velocity vector space, force vector space, etc.
1.2. VECTOR SPACES 15

Physically, a common feature of vector spaces is that all the vectors in each vector space must have
the same physical dimension. In fact, the sum of two vectors with different physical dimensions, e.g., a
velocity vector and a force vector, can not be defined. However, two vector spaces with the same physical
dimension may be quite different in physical background and feature and, therefore, must be regarded as
different vector spaces.
In summary, we may say that a vector space is a collection of vectors in which the vectors are closely
connected and integrated by the basic operations into a self-contained structure. This will be clearly seen
by the structural formula (1.40) for vector spaces that will be given in §1.3.3. A perhaps vivid description
is as follows. Suppose that we have a collection of many assorted beads in different colours, each of which
is made of pearl or even precious stone. Although each individual bead looks very beautiful, a random
collection of them may look disordered and not so nice. Imagine we connect those of the same colour
together through golden threads, separately. In so doing we finally turn a disordered collection of assorted
coloured beads into bunches of necklaces each of which looks orderly, simple and elegant from an artistic
viewpoint. Now we may make a comparison of the foregoing process in constructing vector spaces with the
just-mentioned process. We may regard vectors from different backgrounds as assorted beads in different
colours, and the basic vector operations as golden threads. Then, a vector space (a bunch of necklace)
is a collection of vectors with the same background (beads with the same colour) connected by the basic
operations (golden threads). How the latter as “golden threads” connect all vectors into simple, elegant
structures, i.e., vector spaces, will be seen in §1.3.
It might be seen that the basic operations play a central role in defining vector spaces. A vector space
with the basic operations removed is no longer a vector space but only a loose collection of vectors without
the integrated connection and simple and elegant structure maintained by the basic operations, just as a
necklace with the golden thread removed is no longer a necklace but only a loose collection of beads
without the integrated connection and simple and elegant structure maintained by the golden thread.
In §1.3 we shall reveal the simple and elegant structure of Euclidean vector spaces resulting from the
basic algebraic operations.
It should be pointed out that a more general notion of vector space may be introduced from a math-
ematical standpoint. In this general case, vectors are regarded as abstract mathematical objects, and the
vector addition and multiplication by number are, in a formal and abstract manner, introduced as algebraic
operations obeying the rules (1.1)–(1.10) without any motivation. Then, a vector space is defined as a
self-contained collection of vectors in which the vector addition and multiplication by number obeying the
rules (1.1)–(1.10) are defined. Here, the scalar product may be or may not be involved. For some abstract
vector space, the scalar product obeying (1.14)- (1.17) could not be defined. If such a scalar product can
be defined, then follows a vector space with additional geometrical features. Here we are not concerned
with this general viewpoint. Details in this general respect can be found in many monographs, e.g., Halmos
(1974).

1.2.6 Vector product and mixed product

It is possible to multiply two vectors to yield a vector as product? The answer is negative in general. The
answer is positive only for the vectors treated here, i.e., vectors in a 3-dimensional space. Its physical
motivation is the moment of a force about a given point (see, e.g., Bruhns and Lehmann 1993 for details).
As shown in Fig. 1.11, there are two points A and B in 3-dimensional space, and there is a force vector v
acting at point B. By connecting points A and B we obtain a position vector u with its arrowhead at point
B and with its length exactly the distance between A and B. To describe the rotational effect caused by the
force v about the point A, we introduce a vector quantity called the moment of the force v about the point
A, denoted m . This vector is normal to both u and v and hence to the plane containing these two vectors,
and its magnitude (length) is given by |uu| · |vv| sin θ, where θ is the angle between u and v . Moreover, to
specify its arrowhead, we assume that the triplet (uu, v , m ) form a right-handed system, as indicated in Fig.
1.11.
Mathematically, we say that the moment vector m defined above is the vector (cross) product of vectors
u and v , denoted u × v . Then we have
m = u ×v. (1.18)
16 CHAPTER 1. VECTORS AND VECTOR SPACES

v
B
u
A
m= u x v

Figure 1.11: Moment of a force about a point

A geometrical meaning of the vector product u × v is that its magnitude |uu × v | is just the area of the
parallelogram with the two vectors u and v its two neighbouring sides. By choosing various pairs of two
neighbouring vectors u and v in 3-dimensional space we may construct many different parallelograms,
which have different areas and lie on planes with different normals. A useful, significant fact is that each
such parallelogram spanned by a pair of two neighbouring vectors u and v may be exactly represented by
the vector product u × v , since the latter is just an entity combining the two main geometric features of the
former: its area and its orientation in space.
Besides, two vectors are said to be collinear or parallel, whenever their vector product is the null vector.
For vector product hold the following rules.

u ×u = 0; (1.19)

u × v = −vv × u (anti-commutative rule) ; (1.20)

(αuu) × v = u × (αvv) ≡ αuu × v (associative rule) ; (1.21)
(uu + v ) × w = u × w + v × w (distributive rule) . (1.22)
Furthermore, since the vector product of two vectors is a vector, continued vector products and mixed
products of three and more vectors may be introduced and calculated. In particular, for three vectors u , v
and w , we have the mixed product (uu × v ) · w , which yields the volume of the parallelepiped with u and v
and w its three neighbouring sides (see Fig. 1.12). Indeed, we have

(uu × v ) · w = |uu × v | · |w
w| cos φ , (1.23)

where φ is the angle between u × v and w . Note that |uu × v | is the area of the parallelogram as the base
spanned by u and v and that |w
w| cos φ is just the height of the parallelepiped shown in Fig. 1.12.

v u

Figure 1.12: Parallelepiped spanned by three vectors

A significant property of the mixed product (uu × v ) · w is: Vectors u , v and w are non-coplanar (resp.,
co-planar) if and only if their mixed product is non-zero (resp., zero).
Below are some interesting properties and results involving both scalar and vector products..

|uu × v |2 + (uu · v )2 = |uu|2 |vv|2 ; (1.24)

(uu × v ) · u = (uu × v ) · v = 0 ; (1.25)

(uu × v ) · w = (vv × w ) · u = (w
w × u) · v ; (1.26)
u × (vv × w ) = (uu · w )vv − (uu · v )w
w; (1.27)
1.3. ALGEBRAIC EXPRESSIONS: STANDARD BASES 17

(uu × v ) · (w
w × r ) = (uu · w )(vv · r ) − (uu · r )(vv · w ) . (1.28)
In the above, (1.24)–(1.26) may be evident by definitions of scalar and vector products. However, (1.27)
and (1.28) may not be so evident. Their proofs are non-trivial. Eq. (1.28) is sometimes called Lagrange
identity. The importance of (1.27) and (1.28) lies in the fact that they reduce the continued products
involving vector products to scalar products. Usually, it may be much easier to perform the operation of
scalar product than vector product, as will be indicated in §1.2.6.

1.3 Algebraic Expressions: Standard Bases

We have introduced vectors and their basic operations from a geometric standpoint. The main advantage
of this treatment is that these somewhat elusive objects and notions become straightforward, accessible by
virtue of visual geometric entities. From elementary analytic geometry we know that an algebraic approach
by virtue of coordinate systems is very instrumental and powerful in describing and investigating geometric
objects and their geometric relations as well as relevant operations. Here and later, similar situation will
be expected for vectors and tensors and their operations. We shall show that an algebraic approach by
virtue of bases enables us to reveal simple and elegant structure of vector spaces and facilitate basic and
sophisticated operations.

1.3.1 Unit vector

A unit vector e is a vector with a unit length, i.e.,
e ·e = 1. (1.29)
It may be evident that, for any given non-zero vector, ±uu/|uu| yield the unit vectors collinear with u .
Let e be a given unit vector. In a vector space including e we collect all the vectors which are collinear
with or parallel to e . Then we obtain a very simple collection of vectors. It may be readily shown that each
such collection fulfills the definition for vector space and is called one-dimensional subspace. For each
vector u in it, we have
u = uee , u = u · e . (1.30)
We have u = ±|uu|, where the plus sign “+” and the minus sign “-” are taken, respectively, when the
directions of u and e are the same and opposed to each other, respectively.
It may be evident that the expression (1.30) in terms of the specified unit vector e generates all the
vectors collinear with the unit vector e , whenever the scalar u therein runs over all the scalars.

1.3.2 Plane case

Let m be a non-zero vector. In a vector space including m we collect all the vectors which are orthogonal
to m . Then we obtain a collection of vectors in which each vector lies in a plane with m its normal. It
may be shown that each such collection fulfills the definition of vector space and is called two-dimensional
subspace.
In the above plane, we choose two orthonormal vectors, (ee1 , e 2 ), i.e.,
e1 · e1 = e2 · e2 = 1 , e1 · e2 = 0 . (1.31)
Now we consider any given vector u in the foregoing plane. As shown in Fig. 1.13, we construct a vector
rectangle with the vector u a diagonal and with two sides (vectors) u 1 and u 2 collinear with e 1 and e 2 ,
respectively.
Then, from the parallelogram rule shown in Fig. 1.8 we have
u = u1 + u2 . (1.32)
Moreover, since u 1 and u 2 are collinear with the unit vectors e 1 and e 2 , respectively, from (1.30)1 we
deduce
u 1 = u1 e 1 , u 2 = u2 e 2 . (1.33)
18 CHAPTER 1. VECTORS AND VECTOR SPACES

u2
u
e2
e1 u1

Figure 1.13: Vector rectangle

Further, since u 1 and u 2 are orthogonal, we have u 1 · e 2 = u 2 · e 1 = 0. Hence, from these and the distributive
rule (1.17) as well as (1.30)2 and (1.32) and (1.33), we infer

u1 = u1 · e1 = (uu1 + u2 ) · e1 = u · e1 ,
(1.34)
u2 = u2 · e2 = (uu1 + u2 ) · e2 = u · e2 .

These indicate that u1 and u2 are exactly the projection components of vector u in the directions of e 1 and
e 2 , respectively. Combining (1.32)–(1.34), finally we arrive at

u = u1 e 1 + u2 e 2 ,
(1.35)
u1 = u · e 1 , u2 = u · e 2 .

When the scalars u1 and u2 take all possible values, the above expression in terms of two specified or-
thonormal vectors e 1 and e 2 produces all the vectors normal to both e 1 and e 2 .

1.3.3 Standard expressions

For a general case, we choose a triplet (ee1 , e 2 , e 3 ) formed by three orthonormal vectors, i.e.,

e1 · e1 = e2 · e2 = e3 · e3 = 1 ,
(1.36)
e1 · e2 = e2 · e3 = e3 · e1 = 0 .

Now we consider any given vector u in a three-dimensional space. As shown in Fig. 1.14, we first construct
a vector rectangle in the plane including e 3 and u , with u its diagonal and with two sides (vectors) u 3 and
u 03 being collinear with e 3 and lying in the plane including e 1 and e 2 , respectively, and then we construct
another vector rectangle in the plane including e1 and e2 , with u03 its diagonal and with two sides (vectors)
u1 and u2 collinear with e1 and e2 , respectively.

u3

e3 u
u2
e1
e2

u1
u3’

Figure 1.14: Twin vector rectangles (parallelograms)

For the twin vector rectangles indicated above, by using the result given in §1.2.2 we have

u = u 3 + u 03 ,

(1.37)
u 3 = u3 e 3 , u3 = u · e 3 ,
1.3. ALGEBRAIC EXPRESSIONS: STANDARD BASES 19

for the first vector rectangle, and  0

 u3 = u1 + u2 ,
u 1 = u1 e 1 , u1 = u 03 · e 1 , (1.38)
u 2 = u2 e 2 , u2 = u 03 · e 3 ,


for the second vector rectangle. Then, utilizing the equalities

u 03 = u − u 3 , u3 · e1 = u3 · e2 = 0 ,

we derive
u1 = u 03 · e 1 = (uu − u 3 ) · e 1 = u · e 1 ,
u2 = u 03 · e 2 = (uu − u 3 ) · e 2 = u · e 2 .
Thus, combining the above results, finally we obtain

u = u1 e 1 + u2 e 2 + u3 e 3 ,
(1.39)
ui = u · e i , i = 1, 2, 3 .

To explain what the expression (1.39) implies, we make the following observation and remarks.
(i) Choose and fix three orthonormal vectors e i meeting (1.36). Then, for any given vector u , we can
reduced u to its three projection components through the twin vector rectangles in Fig. 1.14. Hence, it
turns out that any given vector can be represented and expressed by its three projection components
through (1.39) by a simple, unified form of summation. These facts suggest a simple procedure
of expressing any given vector from an algebraic viewpoint: Choose three orthonormal vectors e i
meeting (1.36) and determine three ordered scalars ui by (1.39)2 , and then use (1.39)1 .
(i)) Fix the three e i . Now we give three scalars ui instead of a vector. Then the right hand side of
expression (1.39)1 generates a vector u through the twin vector rectangles in Fig. 1.14. Hence,
taking all possible ui produces all the vectors. A consequence is that vector spaces can be generated
through (1.39)1 by choosing a common triplet (ee1 , e 2 , e 3 ) meeting (1.36) and taking all possible
values of each scalar ui , separately. Thus, we have1

Vector spaces = {u1 e 1 + u2 e 2 + u3 e 3 | −∞ < ui < +∞, i = 1, 2, 3} . (1.40)

(iii) Is it possible to express all the vectors by a simpler expression like (1.35) by simply choosing a
common pair ( e 1 , e 2 ) meeting (1.31) instead of a common triplet (ee1 , e 2 , e 3 ) meeting (1.36)? The
answer is definitely no. An evident fact is that a vector u normal to both e 1 and e 2 , such as u = e 3 ,
could not be expressed by (1.35).
In view of the above facts, we say that each vector space given by (1.40) is three-dimensional and we
refer to (ee1 , e 2 , e 3 ) meeting (1.36) as a standard basis of vector spaces. The three scalars ui in (1.39) are
called the standard components of vector u relative to the standard basis (ee1 , e 2 , e 3 ). Moreover, we say
that expression (1.39) is the algebraic expression of vector u relative to the standard basis (ee1 , e 2 , e 3 ), or,
simply, as a standard expression of u for brevity.
Expressions (1.39) and (1.40) reveal the simple, elegant structure of vector spaces. This simplicity will
essentially facilitates the implementation of basic and sophisticated operations and analyses for vectors and
tensors, as will be seen next subsection and other places.

1.3.4 Indicial notations and conventions

We like to show how to perform the basic operations for vectors from an algebraic viewpoint in terms of
expression (1.39). This will be extended to treat operations for tensors later on.
First we like to explain an important idea. In theory of vectors and tensors etc., we frequently use
various collections of finite number of scalars or vectors or tensors. The members in each such collection
1 Note here that the u may be scalar quantities with various physical dimensions or without dimension. Even for the same physical
i
dimension, the vectors generated by (1.39)1 may represent different kinds of physical quantities.
20 CHAPTER 1. VECTORS AND VECTOR SPACES

have a common feature and are associated in certain order with a vector basis (ee1 , e 2 , e 3 ) as given by (1.35).
Various kinds of summations involving mixed operations of addition and multiplication for them will be
carried out. In this case, it will prove very efficient and convenient to introduce symbols attached with one
or more indices, subscripts or superscripts, to represent these members and then express various kinds of
relevant operations in terms of these indexed symbols.
As the beginning of application, the above idea has been used in presenting a standard basis meeting
(1.36) and the related components. Here we denote three orthonormal vectors and related components by
the indexed symbols e i and ui . Of course, we can replace the index i with any other, e.g., j, k, etc., as we
wish. We assume that each index takes the values 1, 2, 3. For instance, in so doing, either e i or e j , etc.,
represents e 1 , e 2 and e 3 .
From now on, if otherwise indicated, each index will take values 1, 2, 3 and this fact will no longer
be mentioned. In many cases, we write down expressions with two or more juxtaposing indexed symbols.
For such expressions, we adopt Einstein’s summation convention: each index emerging only once, called a
free index, can freely take values 1, 2, 3, and each index repeatedly emerging twice and only twice, called
a dummy index, means the summation when this index runs over 1, 2, 3. Except for few particular cases,
we can avoid any index repeatedly emerging more than twice by adjusting the symbol of dummy indices.
This is exemplified in (1.49) below.
With the above conventions, we present the following results.
e i · e j = δi j ; (1.41)
u = ui e i , v = vi e i , w = wi e i ; (1.42)
0 = 0ee1 + 0ee2 + 0ee3 ; (1.43)
−uu = (−ui )eei ; (1.44)
u ± v = (ui ± vi )eei ; (1.45)
αuu = (αui )eei ; (1.46)
√
|uu| = ui ui ; (1.47)
u · v = ui vi ; (1.48)
ui vi
cos θ = √ √ ; (1.49)
u j u j vk vk
e1 × e2 = e3 , e2 × e3 = e1 , e3 × e1 = e2 ; (1.50)
u × v = (u2 v3 − u3 v2 )ee1 + (u3 v1 − u1 v3 )ee2 + (u1 v2 − u2 v1 )ee3 ; (1.51)
u1 u2 u3
(uu × v ) · w =
v1 v2 v3 ≡ u1 v2 w3 + u2 v3 w1 + u3 v1 w2 − u1 v3 w2 − u2 v1 w3 − u3 v1 w2 . (1.52)
w1 w2 w3
The last two expressions suggest that operations involving vector product are usually much complicated.
That is why formulas (1.26) and (1.27), in particular, the Lagrange formula (1.27), are useful and signifi-
cant.
In (1.41) and henceforward, we introduce the notation

1 for i = j ,
δi j ≡ (1.53)
0 for i 6= j .
This useful notation is usually called Kronecker delta (symbol). Evidently, we have
δi j = δ ji .
An obvious yet very useful identiuty is as follows:
δi j H··· j··· = H···i··· , (1.54)
where “ · · ·” may represent several indices. For instance, we have
δi j y j = yi , δi j T jk = Tik , δi j Tr j = Tri ,
and many others.
In passing, by (1.52)2 we introduce 3 × 3 determinant, which will be used again later on.
1.4. ALGEBRAIC EXPRESSION: ARBITRARY BASES 21

1.3.5 Transformation rule

We like to emphasize that there are many (actually infinite) standard bases meeting (1.36) or (1.41). An
obvious fact is that, for a given vector u , we can derive many different standard components u · e i by
changing the standard basis e i . Now a question is: Is there any relation between these components relative
to various standard bases? If yes, what is it?
We proceed to answer this question. Let e i and ēei be any two different standard bases, i.e.,

e i · e j = δi j , ēei · ēe j = δi j . (1.55)

Relative to these two bases, vector u has the following expressions:

u = ui e i ; u = ūi ēei , (1.56)

with their respective standard components given by

ui = u · e i , ūi = u · ēei . (1.57)

Changing the dummy index “i” in (1.56)1 to “j” and then substituting the resulted expression into (1.57)2 ,
we derive
ūi = (u j e j ) · ēei = (ēei · e j )u j . (1.58)
Since each e i and each ēei are unit vectors, we infer that each e r · ēes (= ēes · e r ) is simply the cosine of the
angle between e r and ēes , call directional cosine.
Now it may become clear that the two sets of standard components of any given vector u relative to
any two given standard bases are interrelated to each other through (1.58). Fixing a standard basis e i and
allowing the basis ēei to run over all possible standard bases, from (1.58) we derive the following significant
fact.
Although the three standard components of any given vector u are changing with the changing of
standard basis ēei , its three components ūi relative to every standard basis ēei can be determined through
(1.58) by its three components ui relative to a given standard basis e i .

1.4 Algebraic Expression: Arbitrary Bases

1.4.1 Co- and contravariant expressions
In last section we know each vector is expressible as a simple form as given by (1.39) by choosing a
standard basis e i meeting (1.36) (see also (1.41)). In some case we need to choose three vectors g i that
need not be either unit vectors or mutually orthogonal, but are non-coplanar, i.e.,

(gg1 × g 2 ) · g 3 6= 0 . (1.59)

In Fig. 1.14, we replace each e i with each g i . Then, in this general case, the twin vector rectangles in Fig.
1.14 become twin vector parallelograms. Then by using the parallelogram rule for vector addition twice
we can derive
u = u 1 + u 2 + u 3 = ui g i , (1.60)
where u i is collinear with g i . Here by using the symbol ui instead of ui we intend to distinguish the
corresponding case that will be treated below.
The above expression is of the same form as (1.39)1 , but its three coefficients ui are no longer the
projection components as given by (1.39)2 . Because the three g i need not be orthonormalized, each ui is
related to u and the three g i in a complicated manner. An idea to overcome this difficulty is to introduce
another three vectors g i given by
g2 × g3 g3 × g1 g1 × g2
g1 = , g2 = , g3 = . (1.61)
(g 1 × g2 ) · g3
g (g 1 × g2 ) · g3
g (g 1 × g2 ) · g3
g
22 CHAPTER 1. VECTORS AND VECTOR SPACES

By using (1.25) and (1.26) it may be readily shown that the following so-called reciprocity conditions hold:

g i · g j = δi j . (1.62)

Hence, using (1.60) and (1.62) we derive

ui = u · gi . (1.63)
Similar to (1.60), we have the expression
u = ui g i . (1.64)
Again by using (1.62) we deduce
ui = u · g i . (1.65)
Moreover, replacing the dummy index “i” in (1.64) and (1.60) by “ j” and substituting the two resulted ex-
pressions into (1.63) and (1.65), respectively, we derive the relationship between the co- and contravariant
components as follows:
ui = (ggi · g i )u j , ui = (ggi · g j )u j . (1.66)
We refer to expressions (1.60)2 with (1.63) as well as expression (1.64) with (1.65) the co- and contravariant
expressions of vector u relative to bases g i and g i , respectively. In addition, we refer to the ui and the ui as
contra- and covariant components relative to the bases g i and g i , respectively.
For these algebraic expressions in a general sense, similar remarks as in §1.3.3 may be made. We have

Vector spaces = {ui g i | −∞ < ui < +∞} , (1.67)

Vector spaces = {ui g i | −∞ < ui < +∞} . (1.68)

Now it may be observed that there exist reciprocal relations in the above results. Usually, we say that the
three g i and the three g i form a reciprocal pair of bases, and we call the basis g i is the reciprocal basis of
the basis g i , and vice versa.

1.4.2 Transformation rules

Just as in the case of standard bases, here we can change the basis g i . In what follows we like to derive
the transformation rule for co- and contravariant components under the change of basis. Let ḡgi be another
basis meeting (1.59), and its reciprocal basis be ḡgi , given by (1.61) by replacing the “gg” therein by “ḡg”.
Then we have expressions
u = ūi ḡgi , u = ūi ḡgi , (1.69)
with
ūi = u · ḡgi , ūi = u · ḡgi . (1.70)
Replacing the dummy index “i” in (1.60) and (1.64) and substituting the two resulted expressions into
(1.70)1 and (1.70)2 , respectively, we arrive at

ūi = (ḡgi · g j )u j ,

(1.71)
ūi = (ḡgi · g j )u j ;

and
ūi = (ḡgi · g j )u j ,

(1.72)
ūi = (ḡgi · g j )u j .
Comparing the results in this section and last section, it may be seen that use of a standard basis e i meeting
(1.41) is much simpler and much more convenient than use of a general basis g i meeting (1.59) only.
Another perhaps essential fact is that a standard component is really a physical component, whereas that
is not the case for either a co- or a contravariant component. These contrasts will become sharper when
tensor quantities are treated.
Thus, if possible, for the sake of simplicity and convenience, a standard basis should be chosen and
used.
1.5. REMARKS: VECTORS AS BASIS-FREE ENTITIES 23

1.5 Remarks: vectors as basis-free entities

In §1.1, a vector is introduced as a geometric entity. In the last two sections, we derive the algebraic
expressions of vector in terms of either orthonormal bases or arbitrary bases, as given by (1.39) and (1.60)
and (1.64). In each of these expressions, the basis may be freely chosen, and, therefore, the associated
components are changing with the changing of basis. By definition, however, a vector should be a basis-
free entity. To have a further understanding of this, we like to make further observation.
Let u be a given vector. Then, for any two bases g i and ḡgi , we can determine the components ui and
ū relative to these two bases, as shown in Fig. 1.14. Of course, the components ui relative to basis g i are
i

different from the components ūi relative to basis ḡi . With the two bases g i and ḡgi and the corresponding
components ui and ūi , we form expressions ui g i and ūi ḡgi , which yield two vectors. Now a question arises
as to whether these two expressions really yield the same vector? They should be the same according to
the algebraic expression derived. But why and how?
Now we answer this question. Since u i and ūi are related to each other by the transformation rule
(1.72)1 , we have
ūi ḡgi = ḡgi · g j u j ḡgi = u j g j · ḡgi ḡgi .

On the other hand, we can express each g j in terms of the basis ḡgi . Setting u = g j in (1.69)1 and (1.70)2 ,
we have
g j = (gg j · ḡgi )ḡgi .
Hence, from the last two expressions we infer

ūi ḡgi = u j g j .

This confirms that each algebraic expression for vector in §1.3 and §1.4 always yields the same vector for
all possible bases. From this we conclude:
Although the components of a vector relative to different bases are different from each other, these
components are related to each other by a transformation rule and as such its algebraic expression always
expresses the same vector.
Sometimes, for every basis g i , we may give three scalars yi in a determined manner. For instance, let u
and v be two given vectors, and their components be ui and vi relative to each basis g i . Then, we give three
scalars yi = ui vi (no summation here) for each basis g i .
For each basis g i and three associated scalars yi , we form an expression yi g i . Evidently, this expression
produces a vector for each choice of basis g i . Only when the scalars yi given for all bases g i are related to
each other by the transformation rule (1.72)1 , the foregoing expression always yields the same vector for
all bases g i . In this case, the scalars yi given for all bases g i are indeed the components of the same vector
relative to all bases g i . Otherwise, the foregoing expression generates different vectors for different choices
of basis g i , and, hence, we could not say that the scalars yi associated with different choices of basis g i are
the components of the same vector relative to different bases.
When three ordered scalars yi associated with bases g i are treated, it would be helpful to keep the above
fact in mind.
Sometimes it is said that each triplet (y1 , y2 , y3 ) determines a vector and hence may be regarded as a
vector. From expression (1.69)1 , this is reasonable only when a basis is specified. Clearly, for the same
triplet xi , generally expression (1.69)1 supplies different vectors for different bases g i .
Finally, we like to remark that once a vector as a basis-free entity is defined, its components relative
to various bases, no matter whether standard or arbitrary, are simply derived consequences of its algebraic
expressions relative to these bases. Hence, it might not be reasonable to say that different kinds of com-
ponents, i.e., standard, co- and contravariant components, would generate different kinds of vectors. A
meaningful question is that if we introduce and define a triplet of ordered numbers, (y1 , y2 , y3 ), associated
with every basis g i in a certain manner, we like to know whether or not the expression yi g i produce the
same vector for all choices of basis g i , i.e., whether or not the introduced triplet is the components of the
same vector relative to all bases, as elaborated before.
24 CHAPTER 1. VECTORS AND VECTOR SPACES
Chapter 2

Second-Order Tensors

Starting from the notion of vector spaces, more sophisticated quantities called tensors may be introduced.
Among them, second order tensors, including stress and strain tensors, play a prominent role in mechanics
and engineering, which supply us with unified and powerful tools and means to understand and resolve
certain basic and difficult issues. This chapter is devoted to second order tensors. Tensors of higher order
will be discussed elsewhere.
It seems helpful to always keep the notion of linear mapping or transformation in mind. The simplest
example is a linear function F which transforms each scalar in a scalar set S1 into a scalar in another scalar
set S2 . Symbolically, we may signify this by

F : x ∈ S1 −→ y ∈ S2 ,
y = F[x] ,
with a linear rule
F[αxx] = αF[x], F[x1 + x2 ] = F[x1 ] + F[x2 ] ,
for any number α and for any scalars x, x1 , x2 ∈ S1 .
Here, important matter is to regard F as a transformation-like quantity whose function or role is to
transform each object in a set into an object in another set. To make this always evident, we introduce
the particular notation F[x], which means that F is a transformation-like quantity and that this quantity
transforms each object x into another object F[x] according to a linear rule.
In this and next two chapters, this simple notion will be essential to understanding various cases for
tensors, in which we interpret the scalars x ∈ S1 and y ∈ S2 here as various objects involved.

2.1 Motivation and Examples

In addition to scalar and vector quantities, more sophisticated quantities are needed in mechanics and
engineering and other related fields. The main motivation in mechanics and engineering comes from two
basic questions: how to describe and characterize the deformation state and the stressed state at each
material point in a deforming body. These will be elaborated below, separately.

2.1.1 Deformation state

Materials in nature are deformable, such as solids (metals, woods, glasses, rubbers, soils, fruits, ices, · · ·),
fluids (oils, water, drinks, soups, wines, · · ·) and gases (airs, winds, hydrogen, oxygen, cyclones, typhoons,
· · ·), etc. Some are always ready for any change of shape and deform effortlessly. Some are soft and
deform easily. Some are very hard and usually deform very slightly without any appreciable change of
shape. To describe and understand various modes of deformations, the basic question (idea) is to know the
deformation state at each material point in a material body.
What is the deformation state at a material point? At point A in an undeformed material body, we
can draw infinitesimal line segments in different directions, each of which can be represented exactly by a

25
26 CHAPTER 2. SECOND-ORDER TENSORS

vector with the arrowhead at the end opposite to point A. We call each such vector a linear element at point
A. Each linear element at point A will experience rotation and length change in a course of deformation of
the material body at issue. Such rotations and length changes for all linear elements at point A constitute
the deformation state at material point A. It may be evident that the deformation state of a deforming body
is determined by the deformation state at all points inside it.
Then, how can we describe and determine the deformation state at a material point? Let point A in an
undeformed material body move to point Ā after the material body experiences deformation. Then, a linear
element l at point A in the undeformed material body will become and correspond to a linear element l̄l at
point Ā in the deformed material body, as shown in Fig. 2.1.

l
A
l
A

Figure 2.1: Correspondence between linear elements

Now we come to an important idea that the deformation state at point A can be determined, whenever
we can find out a transformation-like quantity L which establishes a correspondence relationship between
all the linear elements l at point A, denoted VA , and all the linear elements l¯ at point Ā, denoted VĀ .
Symbolically, we signify this transformation-like quantity as follows:

L : l ∈ VA −→ l̄l ∈ VĀ ,
(2.1)
l̄l = L [ll ] .

The following properties hold:

L [αll ] = α (L
L[ll ]) (2.2)
for each number α and for each linear element l ∈ VA , and

L [ll 1 + l 2 ] = L [ll 1 ] + L [ll 2 ] (2.3)

for any two linear elements l 1 , l 2 ∈ VA .

The above properties may be evident. The former means that α times linear element l at point A
becomes and corresponds to α times linear element l̄l at point Ā, while the latter implies that via deformation
the vector parallelogram formed by any two linear elements l 1 and l 2 at point A becomes and corresponds to
the vector parallelogram formed by the two corresponding linear elements l̄l 1 and l̄l 2 at point Ā, as indicated
in Fig. 2.2.
Thus, determination of the deformation state at point A is equivalent to determination of the transformation-
like quantity L obeying (2.1)–(2.3).

l1 l

l2
l1 l2
A

Figure 2.2: Correspondence between vector parallelograms

2.1. MOTIVATION AND EXAMPLES 27

2.1.2 Stressed state

Inside a deforming material body, there emerge interacting contact area forces due to resistance of the
material body to deformations accompanying shape changes. Let Ā be any given point in a deforming
material body. Imagine that the material body is separated into two parts by a plane through point Ā. Take
a very small plane region, e.g., a small rectangle, around point Ā on the two cutting planes, separately.
These two small regions may be represented by a vector s with its magnitude the area of this region and
with its direction along the outward normal of this region, called area element at point Ā. Then, as indicated
in Fig. 2.3, there is a contact force acting on each such area element, denoted f s . All such contact forces
corresponding to all the area elements at point Ā constitute the stressed state at point Ā. The basic question
in analyzing and understanding the stressed state of a deforming material body is to determine the stressed
state at each point in this body.

f− s

s
−s

fs

Figure 2.3: Separation of a body through a plane

How can we characterize and determine the stressed state at point Ā? A basic idea is that we find out
also a transformation-like quantity Γ which establishes a correspondence relationship between all the area
elements at point Ā, denoted SĀ , and all possible area forces at point Ā, denoted FĀ . Symbolically, we
signify this below. 
Γ : s ∈ SĀ −→ f s ∈ FĀ ,
(2.4)
f s = Γ [ss] .
The following properties hold:
Γ [αss] = α (Γ
Γ[ss]) (2.5)
for every area element s ∈ SĀ , and
Γ [ss1 + s 2 ] = Γ [ss1 ] + Γ [ss2 ] (2.6)
for any two area elements s 1 , s 2 ∈ SĀ .
The property (2.5) may be evident. It means that α times area element s corresponds to α times contact
force f s . In particular, from Newton’s action-reaction principle (see, e.g., Bruhns and Lehmann 1993) we
have
f −ss = − f s , i.e., Γ [−ss] = −Γ
Γ[ss] ,
as indicated in Fig. 2.3. To understand property (2.6), out of the body we cut a small prismatic body around
point Ā with its base a triangle, as shown in Fig. 2.4.

fs
f−s −s
1 2

f−s f−s
2
1

f−s

Figure 2.4: Equilibrium of forces acting on a small prismatic body

Of the three upright surfaces of this prism, the area elements of two are just s 1 and s 2 , while the area
element of the other is just −(ss1 + s 2 ). There is a force acting on each of the five surfaces. Observing that
28 CHAPTER 2. SECOND-ORDER TENSORS

the forces acting on the upper and lower surfaces cancel each other out, from the principle of equilibrium
for force system (see, e.g., Bruhns and Lehmann 1993), we derive

f s 1 + f s 2 + f −(ss1 +ss2 ) = 0 .

The last two expressions above produce (2.6).

Thus, determination of the stressed state at point Ā is equivalent to determination of the transformation-
like quantity T obeying (2.4)–(2.6).

2.2 Second-Order Tensors as Linear Transformations

In either of the above two examples, it is necessary to find a transformation-like quantity which transforms
each vector into a vector in a certain manner, namely, which enables us to derive a desirable vector from any
given vector. Evidently, such a quantity could not be a vector, let alone a scalar. It is a more sophisticated
quantity than a vector.

2.2.1 Definition
A common feature of (2.1)–(2.3) and (2.4)–(2.6) may be summarized as follows: Find out a transformation-
like quantity T which establishes a correspondence relationship between two vector spaces V and V̄ , i.e.,

T : u∈V −→ ūu ∈ V̄ ,
(2.7)
ūu = T [uu] ,

which obeys the following properties:


T [αuu] = α (T T [uu]) for each u ∈ V ,
(2.8)
T [uu1 + u 2 ] = T [uu1 ] + T [uu2 ] for any two u 1 , u 2 ∈ V .

Mathematically, we refer to a transformation-like quantity T obeying (2.7)–(2.8) as a linear transformation

from vector space V to vector space V̄ . More commonly and briefly, we call such a linear transformation
T a second-order tensor. Two examples are the L and the Γ discussed in Examples 1 and 2. They are the
deformation (gradient) tensor and Cauchy stress tensor commonly known in mechanics. Other examples
are the thermal conductivity tensor and the electrical conductivity tensor, which characterize linear rela-
tionships between the heat flux vector and the temperature gradient vector and between the electric current
density vector and the electric field intensity vector, respectively. These linear relationships are represented
also by (2.7)-(2.8) and known as generalized forms of Fourier’s law for heat conduction and Ohm’s law for
electric conduction. For details, see Nye (1985).
At this moment it appears that the notion of tensor is somewhat abstract. Here, what matters is to
keep only one fact in mind: A tensor is a transformation-like quantity which transforms each vector into
a vector in a certain manner according to the rule (2.8). Because it enables us to understand and resolve
certain basic yet elusive issues such as the characterizations of the deformation state and stressed state etc.,
the notion of second-order tensor is very important and already results in a clearer, further understanding
of and a considerable simplification of the foregoing basic yet difficult issues. Indeed, so long as we can
understand and determine the deformation tensor L and the stressed tensor Γ , the deformation state and the
stressed state in a deforming body can accordingly be determined once and for all. As long as we know
how to deal with the second-order tensors L and Γ , we can in a straightforward, unified manner calculate
the rotation and length change of any given linear element as well as the contact force acting on any given
area element. However, now the essential point may be: Is it possible to have a further, more accessible
understanding of and a simple, easy-for-use representation for 2nd-order tensors?
Fortunately, the answer is yes. We shall show that a 2nd-order tensor may be represented by a simple
algebraic entity, and the related operations may indeed be simple. From the algebraic expression for vectors
2.2. SECOND-ORDER TENSORS AS LINEAR TRANSFORMATIONS 29

and the linearity property (2.8) we shall derive a unified, simple algebraic expression for second-order ten-
sors and show how to perform relevant operations in a simple manner. Toward this goal, we first introduce
basic algebraic operations for second-order tensors below.
In the subsequent account of this section, for the sake of simplicity by tensor(s) we mean 2nd-order
tensor(s), if not indicated otherwise.

2.2.2 Multiplication by number

For any given number α and for any given tensor T , the multiplication of T by α yields a tensor αT
T defined
by
(αT T [uu]) for each v ∈ V .
T )[uu] ≡ α (T (2.9)
There are two particular yet important tensors: null tensor and identity (unit) tensor, denoted O and I ,
respectively. The former transforms each vector into a null vector, while the latter transforms each vector
u into the vector u itself. Namely,

O [uu] ≡ 0 ; I [uu] = u for each u ∈ V . (2.10)

Moreover, we denote
T ≡ −T
(−1)T T. (2.11)
The following rules may be evident:

T = O,
0T T =T;
1T (2.12)

T = α(βT
(αβ)T T ) = β(αT
T ) (associative rule) . (2.13)

2.2.3 Addition
Let T 1 and T 2 be any two given tensors. Their addition is denoted T 1 + T 2 and defined by

T 1 + T 2 )[uu] ≡ T 1 [uu] + T 2 [uu] for each u ∈ V .

(T (2.14)

The subtraction of T 2 from T 1 is designated by T 1 − T 2 and defined by

T 1 − T 2 = T 1 + (−T
T 2) . (2.15)

The following rules hold:

T +O = O +T = T ; (2.16)
T1 +T2 = T2 +T1 (commutative rule) ; (2.17)
T = αT
(α + β)T T + βT
T, T 1 + T 2 ) = αT
α(T T 1 + αT
T2 (distributive rules) ; (2.18)
T 1 + T 2 = T ⇐⇒ T − T 1 = T 2 . (2.19)

2.2.4 2nd-order tensor spaces

As in the case of vectors, there are various kinds of tensors having different physical dimensions and
coming from different physical backgrounds. For instances, we have deformation tensors, strain tensors,
stress tensors, etc. A first step toward understanding them is to collect them according to their backgrounds,
e.g., the same physical dimensions etc. This procedure results in various collections of tensors in each of
which the tensors have the same physical background in common and basic operations introduced last
section can be carried out. Naturally, this lead to the notion of tensor space.
A 2nd-order tensor space is a collection of 2nd-order tensors in which addition and multiplication by
number can be performed in a self-contained manner.
In the remaining subsections, we introduce other operations and notions that will be useful.
30 CHAPTER 2. SECOND-ORDER TENSORS

2.2.5 Composite product

Let S and T be any given two tensors. Their composite product is a tensor, denoted S T and defined by
(SS T )[uu] ≡ S [T
T [uu]] for every vector u . (2.20)
The following rules hold:
SI = I S = S ; (2.21)
α(SS T ) = (αSS )T T ) ≡ αSS T
T = S (αT (distributive rule) ; (2.22)
T +U
S (T U ) = S T + SU (distributive rule) ; (2.23)
S (T U ≡ ST U
T U ) = (SS T )U (associative rule) ; (2.24)
for any number α and for any tensors S , T and U .
However, the composite product is not commutative. In fact, S T and T S generally yield two different
tensors. If S T = T S , then we say that tensors S and T are commutable.
Continued composite products of tensor T itself are called powers of tensor T and designated by T r ,
i.e.,
T r ≡ T 1T 2 · · · T r , T 1 = T 2 = · · · = T r = T . (2.25)
In particular, we have
T1 =T, T2 =TT, T 3 = T T T = T T 2 = T 2T . (2.26)
Tr Ts
It may be seen that any two powers T and T are commutable.

2.2.6 Transposition, non-singularity and singularity

The transposition of tensor S is a tensor, denoted S T and defined by
u · S T [vv] ≡ v · (SS [uu])


(2.27)
for any two vectors u and v . We have
ST

T
= S; (2.28)
(αSS )T = αSS T ; (2.29)
T T T
(SS +U
U ) = S +U
U ; (2.30)
T T T
(SSU ) = U S (anti-commutative rule) ; (2.31)
for any scalar α and for any tensors S and U .
On the other hand, a tensor S is said to be nonsingular, if there is a tensor T such that
ST = T S = I . (2.32)
−1
Such a tensor T is called the inverse tensor of S and will be signified by T = S . The following facts hold.
S −1 −1 = S ;


(2.33)

(αSS )−1 = α−1 S −1 ; (2.34)

−1 −1 −1
(SSU ) =U S (anti-commutative rule) , ; (2.35)
for any scalar α 6= 0 and for any two nonsingular tensors S and U .
A useful property for nonsingular tensors is: Any given nonsingular tensor S transforms each non-zero
vector u into a non-zero vector. Otherwise, suppose S [uu] = 0 for some u 6= 0 . Then we deduce
(SS −1 S )[uu] = S −1 (SS [uu]) = 0 .
However, we have S −1 S = I according to the definition of S −1 (see (2.32)). This gives (SS −1 S )[uu] = I [uu] =
u 6= 0 .
In view of the above fact, we say that a tensor is singular, whenever it transforms a non-zero vector
into zero vector.
Explicit criteria for singular and nonsingular tensors will be given by (2.57) in the next section.
2.3. ALGEBRAIC EXPRESSIONS: STANDARD BASES 31

2.3 Algebraic Expressions: Standard Bases

In this section, we derive simple algebraic expressions for tensors using vectors as building blocks.

2.3.1 Dyadic Product of two vectors

Let a and b be two given non-zero vectors. Then we construct a particular tensor which transforms each
vector u into a vector simply given by multiplying the vector a by the scalar b · u . This tensor is determined
by vectors a and b and will be signified by the particular symbol a ⊗ b . Then we have
(aa ⊗ b )[vv] ≡ (bb · v )aa . (2.36)
It may be easily proved that the mapping a ⊗ b defined above obeys (2.8) and, hence, indeed a tensor.
We refer to a ⊗ b as the dyadic product of vectors a and b . The notation “⊗00 is exclusively used to
represent the fact that two vectors a and b generates a tensor through (2.36).
The dyadic product a ⊗ b provides the simplest tensors. The best way of catching it appears to be
as follows: simply keeping the transformation property (2.36) in mind and always regarding it only as a
representation of this simple property.
The motivation of introducing the dyadic product (2.8) is two-fold. One is to provide the simplest
examples of tensors. The other is more essential: it constitutes the necessary foundation to construct and
express general tensors, as will be shown later on.

2.3.2 Standard expression

We choose a standard basis e i of vector spaces. Then we have the expression (1.39) for any given vector
u . In this section, we shall demonstrate that each tensor is simply a linear combination of the nine dyadic
products e i ⊗ e j with simple structure.
Indeed, utilizing the expression (1.39) and the linearity property (2.8), we deduce
T [uu] = T [ui ei ] = (uu · ei )(T
T [eei ]) (2.37)
for any given tensor T . Since tensor T transforms each vector e i into vector T [eei ], again by using expression
(1.39) we infer
T [eei ] = T ji e j , (2.38)
with
T ji = e j · (T
T [eei ]) . (2.39)
For each e i , (2.39) gives three coefficients Ti j . Totally, we have 9 such coefficients, each of which
comes from and is determined by tensor T for a chosen basis e i . Substituting (2.38) into (2.37), we derive
T [uu] = (uu · e i ) (T ji e j ) = T ji ((uu · e i ) e j ) . (2.40)
Now a crucial observation is: (uu · e i )ee j can be written exactly as the form of dyadic product (see (2.36)).
Thus, we obtain
T [u] = T ji (ee j ⊗ e i ) [uu] = (Ti j e i ⊗ e j ) [uu] ,
i.e.,
T − Ti j e i ⊗ e j ) [uu] = 0 .
(T
From this and (2.10) we arrive at
T = Ti j e i ⊗ e j , (2.41)
together with (2.39).
It turns out that each tensor T is expressible simply as a linear combination of the nine dyadic products
e i ⊗ e j with its coefficients Ti j given by (2.39). We refer to expression (2.41) with (2.39) the algebraic
expression of tensor T relative to the standard basis e i , or, briefly, as a standard expression of tensor T .
Besides, we say that the scalar coefficients Ti j are the standard components of tensor T relative to the
standard basis e i .
Below are some observations.
32 CHAPTER 2. SECOND-ORDER TENSORS

(i) Choose and fix a vector basis e i . Then the nine dyadic products e i ⊗ e j are given. Hence, from
expression (2.41) it turns out that a tensor T can be characterized and represented simply by the
nine scalar coefficients Ti j , each of which is related to the tensor T and the chosen basis e i by a
simple formula (2.39).
(ii) Given a basis e i and nine scalars Ti j , (2.41) generates a tensor. For all possible Ti j , (2.41) generates
all possible tensors. Then we have (cf. footnote 1)

2nd-order tensor spaces = {Ti j e i ⊗ e j | −∞ < Ti j < +∞} . (2.42)

(iii) The expression (2.41) could not be reduced further. Namely, any given eight members of the nine
dyadic products e i ⊗ e j are no longer adequate to express all tensors given in (2.42).
Consequently, each tensor space given by (2.42) is said to be 9-dimensional. Accordingly, we say that the
9 dyadic product e i ⊗ e j is a standard basis of tensor spaces T .
Now it may be clear that, with a chosen standard basis e i and the standard expression (2.41), any given
tensor T is reduced to and determined by its nine standard components Ti j . Now we say that the abstract
notion of tensor might become simple in the following sense.

(i) Choose a standard basis ei . Form the dyadic products ei ⊗ e j and keep the simple transformation
property (2.36) in mind.
(ii) Determine the nine components Ti j relative to the chosen basis e i . Then the standard expression
(2.41) gives tensor T .
(iii) Recall the main motivation and the main purpose of introducing the notion of tensor. As a transform-
like quantity, either a stress tensor or a deformation tensor provides us with a unified and powerful
tool in determining the stressed state or the deformation state. In so doing, by means of standard
expression (2.41) we are able to transform in a simple and unified manner each initial linear element
or each area element into a deformed linear element or a force vector acting on the area element
under consideration. With (2.36) and (2.41) as well as the rules for basic operations, the relevant
calculations now become accessible, efficient and standard. Indeed, we have the following formulas:

u · (T
T [uu]) = Ti j ui u j ; (2.43)

u · (T
T [vv]) = Ti j ui v j ; (2.44)
for any two vectors u and v . In the above, the ui and the vi are the standard components of vectors u
and v relative to the chosen standard basis e i .

In addition to (2.43)–(2.44), the standard expression (2.41) may facilitate the implementation of the other
operations introduced before. The results are summarized as follows.

I = δi j e i ⊗ e j = e i ⊗ e i ; (2.45)

αSS = αSi j e i ⊗ e j ; (2.46)

u ⊗ v = ui v j e i ⊗ e j ; (2.47)
S ± T = (Si j ± Ti j )eei ⊗ e j ; (2.48)
T [uu] = Ti j u j e i ; (2.49)
S T = Sik Tk j e i ⊗ e j ; (2.50)
T
S = S ji e i ⊗ e j . (2.51)
Moreover, two tensors S and T are commutable, whenever Si j T jk = Ti j S jk .
On the other hand, for a characterization of singularity and non-singularity and for other purposes, we
define the determinant of tensor S by (cf. (1.52))
2.4. ALGEBRAIC EXPRESSIONS: ARBITRARY BASES 33

S11 S12 S13

det S ≡ S21 S22 S23 = εi jk S1i S2 j S3k . (2.52)
S31 S32 S33
In the above, εi jk is given by 
 +1, i jk = 123, 231, 312 ,
εi jk = −1, i jk = 132, 213, 321 , (2.53)
0, otherwise ,


and usually called permutation symbol. Evidently, we have

ε jik = εik j = −εi jk , ε jki = εki j = εi jk . (2.54)

The permutation symbol εi jk and the Kronecker symbol δi j introduced earlier are very useful in vector- and
tensor-calculus. These two symbols are related to each other by the following useful identity

εi jk εirs = δ jr δks − δ js δkr . (2.55)

More general relations and detailed account may be found, e.g., in Ogden (1984) and Betten (1987). The
foregoing identity leads to the following useful formula:

det(SS T ) = (det S )(det T ) (2.56)

for any two tensors S and T .

Finally, a tensor S is singular or nonsingular, whenever its determinant detSS is zero or non-zero, i.e.,

det S = 0 (singular) ; det S 6= 0 (nonsingular) . (2.57)

2.3.3 Transformation rule

The components Ti j of a tensor T will change under the change of basis. Just as in the case of vectors, the
components of tensor T relative to different bases should be related to each other in a certain manner. This
will be discussed below.
Let e i and ēei be any given two standard bases obeying (1.53). Then we have

T = Ti j e i ⊗ e j , T = T̄kl ēek ēel , (2.58)

with
Ti j = e i · (T
T [ee j ]), T̄kl = ēek · (T
T [ēel ]) . (2.59)
Substituting (2.58)1 into (2.59)2 and then using (2.36) we deduce

T̄kl = ēek · ((Ti j e i ⊗ e j ) [ēel ) = ēek · (Ti j (ee j · ēel ) e i ) ,

i.e.,
T̄kl = (ēek · ei ) (ēel · e j ) Ti j . (2.60)
From the rule given by (2.60) we conclude that the 9 components of a tensor relative to any standard
basis ēei may be determined by its 9 components relative to a chosen standard basis e i as well as the 9
direction cosines ēei · e j .

2.4 Algebraic Expressions: Arbitrary Bases

2.4.1 Co-, contra- and mixed-variant expressions
Instead of a standard basis, we may choose an arbitrary basis g i merely meeting (1.59). The main purpose
of this section is to derive algebraic expressions for each tensor in terms of an arbitrary basis g i and its
reciprocal basis g i given by (1.61).
34 CHAPTER 2. SECOND-ORDER TENSORS

In terms of an arbitrary basis g i and its reciprocal basis g i , we have co- and contravariant expressions
(1.60) and (1.64) for each vector u . Hence we have

T [uu] = T [ui g i ] = ui T [ggi ] ,

T [uu] = T [ui gi ] = ui T [ggi ] .

Since each T [ggi ] and each T [ggi ] are vectors, either of them is also expressible in terms of either of the two
bases g i and g i . Then we have four cases below:

T [ggi ] = T j i g j , T [ggi ] = T ji g j ,

T [ggi ] = T ji g j , T [ggi ] = T j i g j ,
where
T j i = g j · T [ggi ], T ji = g j · T [ggi ], T ji = g j · T [ggi ], T j i = g j · T [ggi ] . (2.61)
Substituting these four expressions into the foregoing two and then utilizing (1.63) and (1.65) as well as
(2.36), we deduce
T [uu] = ui T j i g j = T j i g i · u g j = T j i g j ⊗ g i [uu] ,

T [uu] = ui T ji g j = T ji g i · u g j = T ji g j ⊗ g i [uu] ,

T [uu] = ui T j j g j = T ji (ggi · u ) g j = T ji g j ⊗ g i [uu] ,

T [uu] = ui T j i g j = T j i (ggi · u ) g j = T j i g j ⊗ g i [uu] .

From these we derive four expressions for T as follows:

T = T j i g j ⊗ gi; T = T ji g j ⊗ g i ; T = T ji g j ⊗ g i ; T = Tj i g j ⊗ gi . (2.62)

Expressions (2.62)2 and (2.62)3 will be called co- and contravariant expressions for tensor T , respectively,
and expressions (2.62)1 and (2.65)4 mixed-variant expressions for tensor T . Accordingly, T ji and T ji are
referred to as co- and contrainvariant components of tensor T , and T j i and T j i as mixed-variant components
of tensor T . These four kinds of components are not independent. Any kind of them can determine the
other three kinds. This can be done by substituting the four expressions of (2.62) into each expression of
(2.61). The results are as follows:

T j i = (gg j · g k )(ggi · g l )Tkl = (gg j · g k )(ggi · g l )T kl = (gg j · g k )(ggi · g l )Tk l ; (2.63)

T ji = (gg j · g k )(ggi · g l )Tkl = (gg j · g k )(ggi · g l )T kl = (gg j · g k )(ggi · g l )Tk l ; (2.64)

T ji = (gg j · g k )(ggi · g l )Tkl = (gg j · g k )(ggi · g l )T kl = (gg j · g k )(ggi · g l )Tk l ; (2.65)

T j i = (gg j · g k )(ggi · g l )Tkl = (gg j · g k )(ggi · g l )T kl = (gg j · g k )(ggi · g l )Tk l . (2.66)

2.4.2 Transformation rules

Let g i and ḡgi be any two given bases, and their reciprocal bases be gi and ḡgi , respectively. The components
of tensor T relative to these two pairs of mutually reciprocal bases are given by (2.61) and
j
T̄i = ḡg j · T [ḡgi ] , T̄ ji = ḡg j · T [ḡgi ] , T̄ ji = ḡg j · T [ḡgi ] , T̄ ji = ḡg j · T [ḡgi ] .

Now we establish relationships between these components and those given by (2.61). Since we have four
expressions for T (see (2.62)), for each kind of components we can derive four transformation rules. This
is done by substituting (2.62) into each of the four expressions given the last expressions in the above,
respectively. The results are as follows:
j
T̄i = (ḡg j · g s )(ḡgi · g r )T s r = (ḡg j · g s )(ḡgi · g r )Tsr = (ḡg j · g s )(ḡgi · g r )T sr = (ḡg j · g s )(ḡgi · g r )Ts r ; (2.67)
2.5. CHARACTERISTIC PROPERTIES 35

T̄ ji = (ḡg j · g s )(ḡgi · g r )T s r = (ḡg j · g s )(ḡgi · g r )Tsr = (ḡg j · g s )(ḡgi · g r )T sr = (ḡg j · g s )(ḡgi · g r )Ts r ; (2.68)
T̄ ji = (ḡg j · g s )(ḡgi · g r )T s r = (ḡg j · g s )(ḡgi · g r )Tsr = (ḡg j · g s )(ḡgi · g r )T sr = (ḡg j · g s )(ḡgi · g r )Ts r ; (2.69)
T̄ ji = (ḡg j · g s )(ḡgi · g r )T s r = (ḡg j · g s )(ḡgi · g r )Tsr = (ḡg j · g s )(ḡgi · g r )T sr = (ḡg j · g s )(ḡgi · g r )Ts r . (2.70)
The four transformation rules represent the feature of a tensor as a basis-free entity and ensure that the
four expressions given by (2.62) always yield the same tensor T for different sets of components relative
to different choices of the basis g i .
As in the case of vectors, expressions for tensors in terms of an arbitrary basis g i are much more
complicated than the standard expression for tensors in terms of a standard basis e i . As remarked earlier,
if possible, it is recommended that a standard basis should be chosen and used for the sake of simplicity.

2.5 Characteristic Properties

In this section, we shall study certain important properties of 2nd-order tensors, i.e., characteristic proper-
ties. In the succeeding sections this study will be utilized to uncover essential features of certain kinds of
important tensors.

2.5.1 Eigenvalues and eigenvectors

As a transformation-like quantity, a tensor S transforms a vector into another vector S [uu]. Now we con-
sider the following instructive question: Can we find a real scalar λ and a non-zero vector v such that its
transformed vector S [vv] is exactly λvv? Namely,

S [vv] = λvv . (2.71)

We call such a scalar λ an eigenvalue of tensor S , and such a vector v an eigenvector of S subordinate to
the eigenvalue λ.
The physical meaning of the above question may be clear for stress tensor S . In this case, (2.71) means
that we like to find an area element v such that the force vector acting on this area element takes the simplest
form, i.e., collinear with v. In this case, usually we call an eigenvalue λ of stress tensor S a principal stress.
By means of the eigenvalues and eigenvectors we may gain insight into essential features of 2nd-order
tensors, as will seen in the next sections.

2.5.2 Characteristic equation and invariants

We show how to calculate all the eigenvalues of tensor T and their subordinate eigenvectors. To this end,
we recast (2.71) as
(SS − λII )[vv] = 0 . (2.72)
This means that an eigenvalue λ of tensor S is a particular scalar that makes the tensor (SS − λII ) a singular
tensor, and that this singular tensor transforms each eigenvector v subordinate to λ into a null vector.
Let e i be any given standard basis and Si j be the standard components relative to this basis. Then, from
(2.57)2 we derive the condition for determining the eigenvalues of tensor S as follows:

S11 − λ S12 S13

det(SS − λII ) = S21 S22 − λ S23 = 0. (2.73)
S31 S32 S33 − λ

Before deriving a simplified form of the above condition, we introduce a useful and important quantity.
We refer to
trSS ≡ Skk = e k · S [eek ] (2.74)
as the trace of tensor S , with the property

tr(αSS + βT
T ) = αtrSS + βtrT
T (2.75)
36 CHAPTER 2. SECOND-ORDER TENSORS

for any two scalars α and β and for any two tensors S and T .
By means of the above definition, we may further generate many other similar quantities, such as
trSS 1 S 2 · · · S r with any given r 2nd-order tensors S 1 , · · ·, S r . In particular, S 1 = · · · = S r = S results in

Ir = trSS r . (2.76)

It may be noted that the above definition would produce different results for different bases. However,
by using the transformation formula (2.60) we infer

S̄kk = ēek · S [ēek ] = (ēek · e i )(ēek · e j )Si j

for any other basis ēek . From (1.48) and (1.39)2 we know that (ēek · e i )(ēek · e j ) just yields the scalar product
of vectors e i and e j . Hence we have

(ēek · e i )(ēek · e j ) = e i · e j = δi j .

Thus we conclude
S̄kk = Skk
for any two bases e k and ēek .
The above fact means that each of the trace trSS defined by (2.74) and the derived traces mentioned
before is independent of the choice of basis. We call such quantities invariants of tensor(s). A detailed
account in this respect will be given in Chapter 5.
Now, expanding (2.73) according to (1.51)2 , we arrive at

λ3 − J1 λ2 + J2 λ − J3 = 0 , (2.77)

where 
 J1 = I1 ,
J2 = 21 I12 − I2 ,


(2.78)
J3 = detSS = 16 I13 − 3I1 I2 + 2I3 .


Since the three traces Ir are invariants of S , so are the three Jr given above. Usually, the former and
latter are called basic and principal invariants of S , respectively. Because these invariants supply the same
values for all the bases, equation (2.76) with (2.77) is the same for all the choices of bases and is called the
characteristic equation of tensor S .
A real or complex number satisfying the characteristic equation of tensor S is called a characteristic
root of S . Generally, for a tensor S there are three characteristic roots. Of them, at least one is real. It
should be noted that only a real characteristic root of S can be an eigenvalue of S . Hence, a 2nd-order
tensor S has at least one (real) eigenvalue. However, generally the other two characteristic roots may be a
pair of conjugate complex numbers and, in this case, they are not (real) eigenvalues of S .
The three roots of the characteristic equation (2.76) can be calculated by a direct formula. Then, the
eigenvalues S can accordingly determined.
However, the determination of eigenvectors is not so easy and direct as the characteristic roots. Ac-
cording to (2.72), the three components v j of eigenvector v relative to a standard basis e j is determined by
the following system of three linear equations in v j :

(Si j − λδi j )v j = 0 (2.79)

for each eigenvalue λ of S . Since S − λII is singular, at most two of the above three equations are indepen-
dent.

2.5.3 Cayley-Hamilton theorem

Repeatedly applying S to equation (2.71) r-th times, we deduce

S r [vv] = λr [vv] (2.80)

2.6. SYMMETRIC 2ND-ORDER TENSORS 37

for an eigenvector v of S subordinate to eigenvalue λ. Then, multiplying v by the characteristic equation

(2.76) and using (2.80), we infer

(SS 3 − J1 S 2 + J2 S − J3 I )[vv] = 0 .

Now suppose we have three eigenvectors v =gg1 , g 2 , g 3 , which meet the condition (1.59). Then each vector
u may be expressed in terms of these three eigenvectors, as given by (1.60). From this fact and the last
equality in the above, we derive

(SS 3 − J1 S 2 + J2 S − J3 I )[uu] = 0 for each vector u .

Thus, we obtain
S 3 − J1 S 2 + J2 S − J3 I = O . (2.81)
This implies that every tensor satisfies its own characteristic equation. This intriguing yet very important
fact is known as Cayley-Hamilton theorem.
A direct consequence of Cayley-Hamilton theorem is that each power S r for r ≥ 3 is expressible as a
linear combination of I , S and S 2 with each coefficient an invariant of S . Besides, each trace trSS r for r ≥ 4
is determinable by the three basic invariants of S .

2.6 Symmetric 2nd-order Tensors

In this section, we discuss symmetric 2nd-order tensors, which are frequently met and used in mechan-
ics and engineering and other fields. A 2nd-order tensor S is said to be symmetric, whenever S and its
transposition S T are the same, i.e.,
ST = S . (2.82)
In terms of standard basis e i , the above condition is simply Si j = S ji . Hence, of the 9 standard components
of symmetric tensor S , only six are independent. We have expression

1
S = Si j e i ⊗ e j = Si j (eei ⊗ e j + e j ⊗ e i ) (2.83)
2
for a symmetric tensor S .
Symmetric tensors have nice characteristic properties and hence simpler structures, as shown below.

2.6.1 Explicit formula for eigenvalues

As mentioned before, not every characteristic root of tensor S is an eigenvalue of S , in general. For
symmetric tensors, however, the symmetry condition (2.81) implies that every characteristic root of tensor
S is an eigenvalue of S . A constructive proof is as follows. First, we set

1
χ = λ − J1 . (2.84)
3
Then we may recast equation (2.76) in the form

χ3 − pχ − q = 0 , (2.85)

where
1 1
p = (3I2 − I12 ), q = (2I13 − 9I1 I2 + 9I3 ) . (2.86)
6 27
Utilizing the symmetry condition Si j = S ji , we have

1
(S11 − S22 )2 + (S22 − S33 )2 + (S33 − S11 )2 + S12
2 2 2



p= + S23 + S31 ≥ 0. (2.87)
6
38 CHAPTER 2. SECOND-ORDER TENSORS

Hence, we may introduce the transformation

r
4p
χ= cos ψ .
3
Substituting this into equation (2.85) and using the trigonometric formula
3 1
cos3 ψ = cos ψ + cos 3ψ ,
4 4
we derive √
3 3q
cos 3ψ = 3/2 . (2.88)
2p
Thus, from the above we obtain
 
1 φ − 2rπ
q
λr = 2
I1 + 6I2 − 2I1 cos , r = 1, 2, 3 , (2.89)
3 3
 3 
−1 8I1 − 36I1 I2 + 36I3
φ = cos . (2.90)
(6I2 − 2I12 )3/2
Expression (2.89) with (2.90) supplies an explicit, direct formula for calculating the characteristic roots
of symmetric tensor S . From this formula we infer that, for each symmetric tensor S , every characteristic
root of S is real and hence an eigenvalue of S .

2.6.2 Characteristic expression

We shall show further characteristic properties of symmetric tensors. Let λ and λ0 be any two distinct
eigenvalues of symmetric tensor S , and v and v 0 be two eigenvectors subordinate to λ and λ0 , respectively.
Then we have
S [vv] = λvv; S [vv0 ] = λ0 v 0 .
Since S is symmetric, we have (see (2.82) and (2.27))

v 0 · S [vv] = v · S [vv0 ] .

From these we derive

λvv0 · v = λ0 v · v 0 ,
i.e.,
(λ − λ0 )vv · v 0 = 0 .
This and λ 6= λ0 lead to
v0 · v = 0 , (2.91)
i.e., any two eigenvectors of a symmetric tensor subordinate to any two distinct eigenvalues are orthogonal
to each other.
Consequently, let symmetric tensor S have three distinct eigenvalues λr . Then we can take three or-
thonormal eigenvectors of S subordinate to them, say, l r . Then we have

S [ll r ] = λr l r (no summation) , (2.92)

with
l r · l s = δrs , (2.93)
Hence, the three eigenvectors of S form a standard basis of vector space, called eigenbasis of S . Utilizing
(2.21), (2.45) and (2.92), as well as the identity

S (aa ⊗ b ) = (SS [aa]) ⊗ b (2.94)

2.6. SYMMETRIC 2ND-ORDER TENSORS 39

for any tensor S and for any two vectors a and b , we derive

S = S I = S (ll r ⊗ l r ) = (SS [ll r ]) ⊗ l r .

Thus, we arrive at
3
S= ∑ λr l r ⊗ l r . (2.95)
r=1
It turns out that a symmetric tensor is endowed with the noticiably simple structure resulting from the
symmetry condition (2.82). It assumes an appealingly simple and elegant form in terms of its eigenbasis.
We refer to (2.95) as the characteristic expression of symmetric tensor S .
Some important properties of the characteristic expression (2.95) for symmetric tensor S are as follows:

(i) Each symmetric tensor S may be determined and represented by its three eigenvalues and its eigen-
basis. All the symmetric tensors can be generated through the characteristic expression (2.95) by
allowing the eigenvalues λr and its eigenbasis l r to run over all possible cases.
(ii) When u runs over all possible unit vectors, the maximum magnitude and the minimum magnitude of
the transformed vector S [uu] are just the greatest and the smallest of the eigenvalues of S , and attained
by setting u to be their subordinate eigenvectors, respectively. When S represents strain tensor or
stress tensor, this fact represents significant physical meanings.
(iii) S is nonsingular, whenever each eigenvalue λr of S is non-vanishing. For a nonsingular S , we have
3
S −1 = ∑ λ−1
r lr ⊗ lr . (2.96)
r=1

Moreover, we have the power

3
Sα = ∑ λαr l r ⊗ l r , (2.97)
r=1

for every non-negative integer α. In particular, we have S 0 = I . For a positive definite S , i.e., λr > 0,
we may extend the α in the above to cover all the numbers. This may go even further. More generally,
we may introduce the following symmetric 2nd-order tensor
3
g (SS ) ≡ ∑ g(λr )ll r ⊗ l r (2.98)
r=1

through a real function g(λ), e.g., g(λ) = λα for any number α.

In finite strain analysis, the tensors defined by (2.98), where S is the left or right Cauchy-Green
deformation tensor, supply a general class of finite strain tensors, known as Hill’s strain class (see,
e.g., Hill 1968, 1978, and Ogden 1984). In this case, the tensor S represents the left or right Cauchy-
Green deformation tensor and the function g(λ) is a monotonically increasing function meeting a
certain normalized condition at λ = 1, i.e.,

g(1) = 2g0 (1) − 1 = 0,

known as the scale function for strain measure. The scale function
λα − 1
g(λ) =
2α
produces
1 α
S (α) =
(SS − 1)
2α
which is a well-known subclass of strain tensors, known as Doyle-Ericksen or Seth-Hill strain class
(see, e.g., Doyle and Ericksen 1956, Seth 1964, Hill 1968, 1978, Ogden 1984, Morman 1986, Basa̧r
40 CHAPTER 2. SECOND-ORDER TENSORS

and Weichert 2000). When α = −0.5, −1, 0.5, 1, the last expression generate the known strain
tensors

1 1
(II − S −1/2 ),, (II − S −1 ), (SS 1/2 − I ), , (SS − I ),
2 2
which are often associated with the names Cauchy, St. Venant, Green, Finger, Hamel, Swainger, etc.
In particular, the natural logarithmic scale function g(λ) = 12 ln λ for α = 0 yields the well-known
Hencky’s logarithmic strain tensor h = 12 ln S .

Of course, the function g(λ) in (2.98) may be taken as any constinuous function of λ. Below are some
familiar examples:

g(λ) = eλ , e−λ , sin λ, cos λ, tgλ, ctgλ, sin−1 λ, cos−1 λ, tg−1 λ, ctg−1 λ.

These familiar functions and (2.98) define the following symmetric tensors:

eS , sin S , cos S , tgSS , ctgSS , sin−1 S , cos−1 S , tg−1 S , ctg−1 S .

Sometimes, infinite tensor power serieses may be introduced below:

1 2 1 3
eS = I + S + S + S +···,
2! 3!

1 3 1 5
sin S = S − S + S +···,
3! 5!

1 2 1 4
cos S = I − S + S +···,
2! 4!

1 1
ln S = (SS − I ) − (SS − I )2 + (SS − I )3 + · · · ,
2 3
and many others. Although these tensor serieses may be instructive in relation to well-known power series
expansions of elementary real functions, it seems that the characteristic expression (2.98) through the
function g(λ) may be more appropriate and convenient from standpoints of application and computation.
It should be noted that the expression (2.98) is valid over the whole range of λ, whereas usually an infinite
power series is well-defined only within a restrictive range. That is the case for the last expression for ln S .
Moreover, it may be difficult to cope with an infinite tensor power series, whereas it may be much easier
to treat the expression (2.98) by utilizing the characteristic properties of S , as will be shown below and in
§5.5.
In expression (2.95), we may set two or the whole of the three eigenvalues λr coalescent. The two
degenerate cases yield
S = λ1 l 1 ⊗ l 1 + λ2 (ll 2 ⊗ l 2 + l 3 ⊗ l 3 ) ; (2.99)
S = λll i ⊗ l i = λII ; (2.100)
respectively.
From the characteristic expression (2.95), we can derive simple expressions for the basic and principal
invariants of S . The results are as follows:

Ir = λr1 + λr2 + λr3 . (2.101)


 J1 = I1 = λ1 + λ2 + λ3 ,
J2 = λ1 λ2 + λ2 λ3 + λ3 λ1 , (2.102)
J3 = λ1 λ2 λ3 .

2.6. SYMMETRIC 2ND-ORDER TENSORS 41

2.6.3 Eigenprojections and Sylvester’s formula

The usual characteristic expression (2.95) and related expression (2.96)–(2.98) entail the calculation of the
eigenbasis l r , which does not seem to be easy and direct (see (2.79)). In mechanics and engineering, defor-
mation analysis and stress analysis are usually complicated field problems. This means that the eigenbasis
of either stress tensor or strain tensor will be unknown at each material point and different from point to
point, and, therefore, need to be calculated for each point. Usually, this kind of calculation has to be carried
out millions of times in large-scale non-linear finite element computations for field problems. In this case,
computations of eigenbasis related to (2.95) would become unrealistic. Is there any direct approach to
bypass this?
In recent years, the above question has been studied from different contexts; refer to, e.g., Miehe
(1993, 1994, 1997), Xiao (1995a) and Xiao, Bruhns and Meyers (1997, 1998a, b). The crucial idea is to
use eigenprojections and Sylvester’s formula. Here we touch only on some essential aspects. Detailed and
further accounts will be presented in §5.5.
Let λ1 , · · ·, λm be all the distinct eigenvalues of symmetric tensor S . Here, m =3, 2, 1. Then we may
rewrite (2.95) in the form
m
S= ∑ λσ S σ . (2.103)
σ=1

In the above, each S σ is called the eigenprojection subordinate to λσ , given by

mσ
S σ = ∑ l t0 ⊗ l t0 , (2.104)
t=1

where mσ is the algebraic multiplicity of eigenvalue λσ , i.e., the multiplicity as a repeated root of the
characteristic equation (2.77). Hence, each mσ is either 1 (simple root) or 2 (twice repeated) or 3 (triply
repeated); and, moreover, l 01 , · · ·, l 0mσ are mσ orthonormal eigenvectors subordinate to λ0σ .
For a repeated eigenvalue λ0σ , the above set of mσ orthonormal eigenvectors is non-unique. Actually
there exit infinitely many such sets. Nevertheless, each eigenprojection S σ given by (2.104) is uniquely
determined by S , as will be shown slightly later.
The eigenprojections have simple manipulable properties, as given below:

S S σ = S σ S = λσ S σ (no summation) ; (2.105)


O for σ 6= τ ,
SσSτ = SτSσ = (2.106)
S σ for σ = τ ;
m
∑ Sσ = S1 + · · · + Sm = I . (2.107)
σ=1

Moreover, we may recast (2.96)–(2.98) as follows:

m
Sα = ∑ λα S σ ; (2.108)
σ=1

m
g (SS ) = ∑ g(λσ )SSα . (2.109)
σ=1

Now we aim to derive an explicit expression for each eigenprojection S θ . In what follows we choose
and fix a θ from 1, 2, · · ·, m. Removing the factor (SS − λθ I ) from the continued product (SS − λ1 I )(SS −
λ2 I ) · · · (SS m − λm I ), we obtain a continued product of the (m − 1) tensor (SS − λτ I ) with τ = 1,, 2, · · ·, m but
τ 6= θ. We designate this continued product by
m
∏(SS − λτ I ) .
τ6=θ
τ=1
42 CHAPTER 2. SECOND-ORDER TENSORS

Utilizing (2.103) and (2.106), we have

m
(SS − λτ I ) = ∑ (λσ − λτ )SSσ . (2.110)
σ=1

Using this and (2.105) and noting τ 6= θ, we derive

 
m m
∏(SS − λτ I ) = ∏(λθ − λτ ) S θ .
 
τ6=θ τ6=θ
τ=1 τ=1

Thus, observing S 1 = I for m = 1 and assuming that the above continued product vanishes for m = 1, we
obtain the following Sylvester’s formula in a unified form:
m
S − λτ I
S θ = δ1m I + ∏ . (2.111)
τ6=θ
λθ − λτ
τ=1

Now it may be clear that the characteristic expressions (2.107)–(2.109) can be calculated and deter-
mined in a straightforward manner by means of Sylvester’s formula (2.111) without involving computation
of the eigenbasis l r of S . Further discussion will be given in §5.5 of chapter 5.
The above results may be valid for n-dimensional symmetric tensors. Here, n = 3 and hence only three
cases for m need to be treated, as shown below.

For m = 3:
λ2 − λ3
S1 = − (SS − λ2 I )(SS − λ3 I ) , (2.112)
∆
λ3 − λ1
S2 = − (SS − λ3 I )(SS − λ1 I ) , (2.113)
∆
λ1 − λ2
S3 = − (SS − λ1 I )(SS − λ2 I ) , (2.114)
∆
where
∆ = (λ1 − λ2 )(λ2 − λ3 )(λ3 − λ1 ) . (2.115)

For m = 2:
S − λ2 I
S1 = , (2.116)
λ1 − λ2
S − λ1 I
S2 = . (2.117)
λ2 − λ1

For m = 1:
S1 = I . (2.118)

2.7 Orthogonal Tensor

An orthogonal tensor Q is a nonsingular tensor of the property

Q T = Q −1 . (2.119)

Namely, the transposition of an orthogonal tensor yields its inverse. Hence, we have

QQT = QTQ = I . (2.120)

2.7. ORTHOGONAL TENSOR 43

For any two vectors u and v , by using (2.27) we deduce

Qu ) · (Q
(Q QT (Q
Qv ) = v · (Q QT Q )[uu]) .
Q[uu])) = v · ((Q

Then, from this and (2.119) we derive

Q[uu]) · (Q
(Q Q[vv]) = u · v . (2.121)

In particular, we have
Q[uu]|2 = (Q
|Q Q[uu]) = u · u = |uu|2 .
Q[uu]) · (Q (2.122)
The last two expressions indicates the geometric feature of orthogonal tensor Q as a transformation
quantity: Both the length (magnitude) of any vector and the angle between any two vectors will preserve
after the transformation performed by Q .
In mechanics of rigid bodies, we know that a rigid body is regarded as an imaginary or ideal material
body in which all its linear elements undergo no length changes and angle changes in course of motion.
such ideal kinematic behaviour can be characterized and represented exactly by an orthogonal tensor, in
conjunction with a displacement vector for any chosen point, such as the mass center, etc. In addition,
the importance of orthogonal tensors consists in the fact that they are essential to and necessary for the
characterization of geometric symmetry properties of internal structural elements of solid materials, such
as crystalline and quasi-crystalline materials, etc., as will be seen in Chapter 5.

2.7.1 The unit axial vector

To further understand orthogonal tensors, we examine the characteristic property of orthogonal tensor Q .
As has been indicated, Q has an eigenvalue λ. Let l be a unit eigenvector subordinate to λ, then we have
(
Q [ll ] = λll ,
(2.123)
λ2 = 1 .

In the above, that λ2 = 1 is derived from (2.122). Then we infer that each eigenvalue of Q is either 1 or −1.
The unit eigenvector l in (2.123) is called the axial vector of orthogonal tensor Q . Its importance will
be clear in the subsequence development.

2.7.2 Canonical expression

Now we take three orthonormal vectors l , e 1 and e 2 . From (2.121) and (2.122) we know the three trans-
formed vectors Q [ll ], Q [ee1 ] and Q [ee2 ] are also orthonormal. Noting that Q [uu] is just l or −ll , we deduce
that vector l is orthogonal to both Q [ee1 ] and Q [ee2 ]. From these we infer that (ee1 , e 2 ) and (Q
Q[ee1 ], Q [ee2 ]) are
orthonormal and lie in a plane normal to l , as indicated in Fig. 2.5.

Q [ e2 ] e2

θ Q [ e1 ]

θ
e1

Figure 2.5: A pair of orthonormal vectors in the same plane

Hence, we derive that there exists an angle ϕ such that


Q [ee1 ] = e 1 cos ϕ + e 2 sin ϕ ,
(2.124)
Q [ee2 ] = −ee1 sin ϕ + e 2 cos ϕ .
44 CHAPTER 2. SECOND-ORDER TENSORS

Using (2.123) and (2.124), we can calculate the standard components of Q relative to the standard basis
(ll , e 1 , e 2 ) and then apply the standard expression (2.41) with (2.39) we obtain

Q = λll ⊗ l + (ee1 ⊗ e 1 + e 2 ⊗ e 2 ) cos ϕ + (ee1 ⊗ e 2 − e 2 ⊗ e 1 ) sin ϕ . (2.125)

Expression (2.125) is referred to as the canonical expression for orthogonal tensor Q , and the angle ϕ
therein as the rotation angle of Q . Using e 1 × e 2 = l and (3.41) given later and observing

e1 ⊗ e1 + e2 ⊗ e2 = I − l ⊗ l

for any two orthonormal vectors e 1 and e 2 that are orthogonal to l , we know that the canonical expression
(2.126) is an explicit expression in terms of the unit axial vector l and the rotation angle ϕ. Such explicit
expressions may be found, e.g., in Guo (1981) and Başar and Weichert (2002).
The powers of Q is given by

Q r = λr l ⊗ l + (ee1 ⊗ e 1 + e 2 ⊗ e 2 ) cos rϕ + (ee1 ⊗ e 2 − e 2 ⊗ e 1 ) sin rϕ . (2.126)

Moreover, from the canonical expression of Q we can calculate the basic and principal invariants. The
results are as follows:

I1 = λ + 2 cos ϕ, I2 = 1 + 2 cos ϕ, I3 = λ + 2 cos ϕ ,
(2.127)
J1 = λ + 2 cos ϕ, J2 = 1 + 2λ cos ϕ, J3 = λ .

Q = −1,
Q = 1, we call such a Q a proper orthogonal tensor. Otherwise, we have J3 = detQ
If J3 = detQ
Q and we call such a Q an improper orthogonal tensor.
For any vector u , we have

Q [uu] = λ(uu · l )ll + (uu0 cos ϕ + (ll × u ) sin ϕ) ,


(2.128)
u 0 = u − (uu · l )ll .

In the above, u 0 is the projection of the vector u in a plane normal to l . From this it may be clear that the
transformation performed by a proper orthogonal tensor Q is a rotation through the angle ϕ about an axis
in the direction of its axial vector l and may be thus referred to as a rotation straightforwardly. Besides, it
may be evident that any improper orthogonal tensor can be written as −Q Q with Q a rotation. As a result,
the transformation performed by an improper orthogonal tensor is the composition of a rotation and an
inversion that will be explained slightly later. These facts are indicated in Fig. 2.6.
ϕ ϕ=π
(u . l ) l
u^
u Q[ u ]
u
l
l
lx u ’
ϕ
u’ Q[ u ’]

Figure 2.6: (a) Transformation by proper orthogonal tensor (b) π-rotation and reflection

2.7.3 Inversion, π-rotation and reflection

From the canonical expression (2.126) we know that a proper (an improper) orthogonal tensor Q is de-
termined by its unit axial vector l and its rotation angle ϕ. This suggests a more suitable notation for an
orthogonal tensor. A rotation through the angle ϕ about an axis in the direction of the unit vector l will be
2.8. SKEW-SYMMETRIC 2ND-ORDER TENSOR 45

ϕ
designated by R l . Hence, the composition of the just-mentioned rotation and the inversion will be written
as −R Rθl .
There are four simple yet important orthogonal tensors. The first is the simplest, i.e., the identity tensor
I , which results in no transformation at all. The second is −II , called inversion, by which every vector u
and its opposed vector −uu are transformed into each other. The third is

R πl = 2ll ⊗ l − I , (2.129)

called π-rotation about axis l , by which each vector u and its symmetric counterpart ûu with respect to axis
l are transformed into each other, as shown in Fig. 2.6. Finally, the fourth is

Rπl = I − 2ll ⊗ l ,
−R (2.130)

called reflection with respect to a plane normal to l (ll -plane), by which each vector u and its symmetric
counterpart ūu with respect to an l -plane are transformed into each other, as shown also in Fig. 2.6.
The above four particular orthogonal tensors play basic roles in describing the geometric symmetry
properties of internal basic structural elements of solid materials, as will be seen in §4.2.

2.8 Skew-symmetric 2nd-Order Tensor

Skew-symmetric tensors are used to characterize spins of continuously time-dependent rotations. More-
over, they may represent some physical quantities, such as the angular velocity of a rotating body, the
changing rate of continuing rotation of the linear elements at each point in a deforming material body, etc.

2.8.1 The associated axial vector

A 2nd-order tensor Ω is said to be skew-symmetric, whenever

Ω T = −Ω
Ω. (2.131)

Hence, relative to a standard basis e i , we have Ωi j = −Ω ji . This gives

Ω11 = Ω22 = Ω33 = 0 , Ωi j = −Ω ji , j 6= i,

and hence we have the expression

1
Ω = Ωi j (eei ⊗ e j − e j ⊗ e i ) (2.132)
2
for any skew-symmetric tensor Ω . Hence we have

I1 = 0, I2 = −2(Ω212 + Ω223 + Ω231 ), I3 = 0 ; (2.133)

J1 = 0, J2 = Ω212 + Ω223 + Ω231 , J3 = 0 . (2.134)

Hence the characteristic equation of Ω is as follows:

λ3 + J2 λ = 0 .

Since J2 > 0 for non-zero Ω , the only real root is λ = 0. This means that each skew-symmetric tensor Ω
has only one eigenvalue, given by 0. Now we introduce a vector ω by

ω ≡ Ω23 e 1 + Ω31 e 2 + Ω12 e 3 . (2.135)

Then we have
Ωω = 0 . (2.136)
Thus, the vector ω given by (2.135) is an eigenvector subordinate to the only eigenvalue of Ω . We refer to
ω as the associated axial vector of skew-symmetric tensor Ω .
46 CHAPTER 2. SECOND-ORDER TENSORS

2.8.2 Dual and canonical expressions

Noting (1.50), we may reformulate (2.135) as
1
ω = Ωi j e i × e j . (2.137)
2
On the other hand, with the aid of the permutation symbol εi jk given by (2.53) we may recast (2.132) as
1
Ω = ωk εi jk (eei ⊗ e j − e j ⊗ e i ) . (2.138)
2
It may be seen that (2.137) and (2.138) constitute dual expressions of skew-symmetric tensor Ω and
its associated axial vector ω . Thus, an skew-symmetric tensor and its associated axial vector may be
expressed in terms of each other.
ω = ω /|ω
On the other hand, let e 1 , e 2 and the unit axial vector ω̄ ω| form a standard basis. Then, by using
expression (2.132) and expression (2.41) with (2.39) we derive the canonical expression of skew-symmetric
tensor Ω as follows:
Ω = |ω ω|(ee1 ⊗ e 2 − e 2 ⊗ e 1 ) . (2.139)
From this we obtain
Ω 2 = ω ⊗ ω − |ω
ω|2 I , Ω 3 = −|ω
ω|2 Ω , (2.140)
Ω [uu] = ω × u for any vector u . (2.141)
From the canonical expression (2.139) we know that an skew-symmetric tensor Ω is determined by its
associated axial vector ω .

2.9 Decompositions of 2nd-Order Tensor

For a general 2nd-order tensor, we may reduce it to the composition of two or more simple tensors as
discussed before. In this section, two kinds of decompositions will be discussed, which prove to be useful
in non-linear continuum mechanics.

2.9.1 Additive decompositions

Let L be a general 2nd-order tensor that need not be symmetric or skew-symmetric. Its symmetric and
skew-symmetric parts are given by
1 1
L + LT , L − LT .



L≡
symL L≡
skwL (2.142)
2 2
Evidently, we have the following unique additive decomposition of tensor L :
L + skwL
L = symL L. (2.143)

Although the decomposition (2.143) is quite simple and straightforward, it can lead to important results.
Indeed, in kinematics of finite deformations of continuous bodies, the additive decomposition (2.143) for
the velocity gradient tensor L generates the (symmetric) stretching tensor (also: Eulerian strain rate, defor-
mation rate, rate of deformation, etc.) and the (skew-symmetric) vorticity tensor. It is commonly accepted
that the stretching tensor is a natural, far-reaching fundamental kinematic quantities in both non-linear
solid mechanics and fluid dynamics. In addition, the vorticity tensor is basic in understanding the rotatory
motion of vortex flows of fluids, etc.
From (2.143) we deduce that a general 2nd-order tensor may be represented equivalently by its sym-
metric and antisymmetric parts. Since the latter two may be treated relatively easily, the just-mentioned
fact may be useful in some cases. Application will be made in §5.7.2.
For a symmetric tensor A , a simple yet useful decompositions is as follows:
1
A)II + Ā
A = (trA A, (2.144)
3
2.9. DECOMPOSITIONS OF 2ND-ORDER TENSOR 47

where the first term is known as the spherical or uniform part of A and the second as the deviatoric part of
A , given by
1
A = A − (trA
à A)II . (2.145)
3
A is traceless, i.e.,
Note that the deviatoric part Ā

A = 0.
trĀ (2.146)

The decomposition (2.144)-(2.145) is also quite simple. But it can also lead to important results.
Indeed, if A represents stress tensor, then Ā A is the deviatoric stress, which is significant in describing
yield behaviour of solids, in particular, metals etc.
On the other hand, if A represents a finite strain tensor g (SS ) as defined by (2.98), where S is the
symmetric, positive definite tensor F F T or F T F (left and right Cauchy-Green deformation tensors) that
will be discussed below, then a relevant, significant question in kinematics of large deformation is whether
the decomposition (2.144) can achieve or realize a fully uncoupled separation of the volumetric deformation
from the total deformation. This requires that the uniform and deviatoric parts of g (SS ) depend merely on
the volumetric deformation detSS and the isochoric deformation (detSS )−1/3 S , respectively. It is known (see,
e.g., Richter 1949 and Lehmann 1960) that this is possible for g(λ) = ln λ, i.e., for Hencky’s logarithmic
strain 12 ln S . In fact, of all the strain measures given by (2.98), the latter is the only one having the just-
mentioned favourable property. A proof is as follows.
The spherical part of the tensor g (SS ) is given by

1 1
trgg(SS ) = (g(λ1 ) + g(λ1 ) + g(λ1 )).
3 3
Now suppose that this part relies only on the volumetric deformation J3 = detSS , i.e.,
1
(g(λ1 ) + g(λ1 ) + g(λ1 )) = f (J3 ).
3
Differentiating this expression with respect to λi and using (2.102)3 , i.e.,

J3 = λ1 λ2 λ3

and noting
∂J3 J3
= ,
∂λi λi
we deduce
1 0 J3
g (λi ) = f 0 (J3 ) ,
3 λi
i.e.,
3J3 f 0 (J3 ) = λi g0 (λi ).
Observing that each λi and the Jacobian J3 are independent of each other for general triaxial deformations,
we infer that
3J3 f 0 (J3 ) = λi g0 (λi ) = constant.
for every J3 and for every λi . From these and the normalized condition following (2.98), by setting λi = 1
we infer
1
3J3 f 0 (J3 ) = λi g0 (λi ) = (λi g0 (λi ))|λi =1 = .
2
This leads to
1 1
3J3 f 0 (J3 ) = , λi g0 (λi ) = .
2 2
Thus, we arrive at
1 1
g(λ) = ln λ, f (J3 ) = ln J3 .
2 6
48 CHAPTER 2. SECOND-ORDER TENSORS

These give g (SS ) = 21 ln S and

1 1 1
h= ln S = (ln J3 )II + ln S̃S , (2.147)
2 6 2
where S̃S is the isochoric part of S given by
−1/3
S̃S = J3 S. (2.148)

Th above decomposition may easily derived by utilizing (2.98) with g(λ) = 12 ln λ and oberserving the fact
that S and its isochoric part S̃S have their eigenvectors in common and their eigenvalues are given by λi and
J3−1 λi , respectively.
Note that
−1/3
detS̃S = (J3 )3 detSS = 1, tr(ln S̃S ) = 0.
It may be evident that the decomposition (2.147)-(2.148) indeed realizes a fully uncoupled additive splitting
of the total deformation into the volumetric and the isochroic part.

2.9.2 Polar decompositions

In non-linear continuum mechanics, the basic deformation tensor is the so-called deformation gradient.
It is a non-singular tensor, denoted F . The continuity of deformation requires that the determinant of F
should be positive, i.e.,
F > 0.
detF (2.149)
It may be evident from (2.31) that the tensor F F T is symmetric. Let λr be three eigenvalues of F F T ,
and l r be three subordinate orthonormal eigenvectors. Then we have the characteristic expression
3
FFT = ∑ λr l r ⊗ l r . (2.150)
r=1

Since
F F T )[ll r ] = (F
λr = l r · (F F T l r ) · (F
F Tlr)
and F is non-singular, we infer that each eigenvalue λr of F F T is positive. As a result, we can define

12 3 1
V = FFT = ∑ λr2 l r ⊗ l r . (2.151)
r=1

Hence we have
V 2 = FFT. (2.152)
Moreover, let

− 12
R = V −1 F = F F T F. (2.153)
Then, using (2.31) and (2.24) and (2.149) we have

R R T = V −1 F F TV −1 = V −1 F F T V −1 = V −1V 2V −1 = I .

This means that the tensor R given by (2.152) is an orthogonal tensor.

From the above facts, we infer that there are a symmetric, positive definite tensor V given by (2.151)
and an orthogonal tensor R given by (2.153) such that

F =VR. (2.154)

On the other hand, let

( 1

12
U = R TV R = F T F = ∑3r=1 λr2 l̄l r ⊗ l̄l r , (2.155)
l̄l r = R T l r .
2.9. DECOMPOSITIONS OF 2ND-ORDER TENSOR 49

Then, the symmetric tensor U is also positive definite, its eigenvalues are still given by λr , and its eigen-
vectors are given by l̄l r . From (2.21), (2.120), (2.24) and (2.153) we derive

F = I F = R R T (V V R ) = R R TV R .

From this we deduce that there are a symmetric, positive definite tensor U given by (2.154) and an orthog-
onal tensor given by (2.152) such that
F = RU . (2.156)
√
Besides, using det R = ±1 and detU U = λ1 λ2 λ3 > 0, from the condition (2.146) and (2.156) and (2.56)
we deduce that detR R = 1, i.e., R is a proper orthogonal tensor. This may be understood by setting V = I in
(2.153).
Expressions (2.153) and (2.155) are known as the left and right polar decompositions of F . They
indicate that a 2nd-order tensor of the property (2.146) is the composite product of a proper orthogonal
tensor and a symmetric, positive definite tensor. If F represents the deformation (gradient) tensor, the
left polar decomposition (2.153) means that the deformation state at each material point may be carried
out by the composition of two relatively simple procedures: first a rigid-body rotation (R R) and then a
subsequent pure strain (V V ) without rigid-body rotation, whereas the right polar decomposition (2.155)
implies a converse composition: first a pure strain (U U ) without rigid-body rotation and then a subsequent
rigid-body rotation (RR).
From either (2.151) and (2.154) or (2.155)–(2.156) we derive the following useful result
3 1
F= ∑ λ 2 l r ⊗ l̄l r , (2.157)
r=1

3
R= ∑ l r ⊗ l̄l r , (2.158)
r=1

where λr are the three common eigenvalues of the symmetric, positive tensors F F T and F T F , called the left
and right Cauchy-Green deformation tensors, and l r and l̄l r are their respective subordinate orthonormal
eigenvectors.
The left and right polar decompositions are basic in kinematic analysis and constitutive formulations in
non-linear continuum mechanics.
50 CHAPTER 2. SECOND-ORDER TENSORS
Chapter 3

Tensors of Higher Order

As a linear mapping transforming each vector into a vector, the notion of 2nd-order tensors supplies us
with a powerful, unified tool to resolve some basic, difficult problems, such as characterization and deter-
mination of the deformation state and the stressed state at each point in a deforming material body, etc. For
various purposes, however, we need more sophisticated quantities to resolve more involved problems. For
instance, we need quantities that can transform each vector into a 2nd-order tensor, each 2nd-order tensor
into a vector, and each 2nd-oder tensor into a 2nd-order tensor, etc. In what follows, a general formulation
of this consideration will first be presented below, and then follows a further account of frequently used
3rd- and 4th-order tensors.

3.1 Definitions
First, a 2nd-order tensor T is a linear mapping transforming each vector into a vector; refer to (2.7)–(2.8).
Let V and T2 are a vector space and a 2nd-order tensor space. Then, a 3rd-order tensor M is a linear
mapping transforming each vector into a 2nd-order tensor , i.e.,

M : u ∈ V −→ S ∈ T2 ,
(3.1)
S = M [uu] ,

with
M [α1 u 1 + α2 u 2 ] = α1 M [uu1 ] + α2 M [uu2 ] for any vectors u 1 , u 2 ∈ V . (3.2)
As in the case of 2nd-order tensors, we may define null 3rd-order tensor, addition, and multiplication by
number for 3rd-order tensors. Then, a 3rd-order tensor space is a collection of 3rd-order tensors in which
these two basic operations can be defined and performed in a self-contained manner.
Let T3 be a 3rd-order tensor space. Then, a 4th-order tensor L is a linear mapping transforming each
vector into a 3rd-order tensor, i.e.,

L : u ∈ V −→ M ∈ T3 ,
(3.3)
M = L[uu] ,

with
L[α1 u 1 + α2 u 2 ] = α1 L [uu1 ] + α2 L[uu2 ] for any vectors u 1 , u 2 ∈ V . (3.4)
Introducing null 4th-order tensor, addition, and multiplication by number in a similar manner as in the case
of 2nd-order tensors, we obtain 4th-order tensor spaces, denoted T4 .
Following and continuing the above recurrence procedure, we can define nth-order tensors, then null
nth-order tensor, addition, and multiplication by number for them and finally nth-order tensor spaces,
denoted by Tn . Here, n = 2, 3, 4, · · ·. In particular, for the sake of completeness, sometimes we call each
scalar and each vector zeroth- and first-order tensors, respectively.
Thus, we have defined nth-order tensors and nth-order tensor spaces Tn for each non-negative integer n.
However, it should be noted that vectors, i.e., 1st-order tensors, play a fundamental role in defining tensors

51
52 CHAPTER 3. TENSORS OF HIGHER ORDER

of different order. Before introducing and defining tensors, first of all, we have to know how to introduce
and define vectors and their basic operations. To stress this fact, we may call each Tn for each n ≥ 2 an
nth-order tensor space over a 3-dimensional base vector space. Each such tensor is called a 3-dimensional
tensor. We like to mention in passing that m-dimensional tensors in a general sense may be defined by
choosing an m-dimensional vector space as a base vector space. This respect will be touched on in §3.10
of this chapter.
Another fact is that a tensor of higer order possesses richer transformation functions as the order in-
creases. This will be explained later.

3.2 Tensor Products

To further understand and express nth-order tensors, a crucial idea is the notion of tensor product of a
tensor and a vector, which is a generalization of the dyadic product of two vectors.
Let T be a pth-order tensor and e be a vector. The tensor product of tensor T and vector e is a particular
(p + 1)th-order tensor, denoted T ⊗ e and defined by

T ⊗ e )[uu] ≡ (uu · e )T
(T T for any vector u . (3.5)

Let vectors e1 , e2 and e3 form a standard basis. Then, by repeatedly using the notion of tensor product
we obtain the simplest qth-order tensors:

e i1 ⊗ · · · ⊗ e iq , 1 ≤ i1 , · · · , iq ≤ 3 . (3.6)

Note here that each index it runs over 1, 2 and 3. Hence, the total number of such tensors are 3q .
Let H be a qth-order tensor and u 1 , · · · , u q be q vectors. Then we may introduce q tensors H 1 , · · ·, H q
by the following recurrence procedure:

H0 = H ,
(3.7)
H t = H t−1 [uuq−t+1 ], t = 1, · · · , q ,

Note in the above that, for any given 1 ≤ t ≤ q, the tensor H t−1 is a (q−t +1)th-order tensor and transforms
vector u q−t+1 into the (q − t)th-order tensor H t . Finally, the above procedure gives a scalar till t = q,
designated by
H q ≡ H [uu1 , · · · , u q ] . (3.8)
Let H be the tensor given by (3.6) in the above procedure. Then we have



e i1 ⊗ · · · ⊗ e iq [uu1 , · · · , u q ] = (uu1 · e i1 ) · · · u q · e iq . (3.9)

In particular, we have

e i1 ⊗ · · · ⊗ e iq [eek1 , · · · , e kq ] = δi1 k1 · · · δiq kq . (3.10)
It will be seen that the notion of tensor product enables us to construct tensors of arbitrary order by
using a vector basis of vector spaces as building blocks. An extension of tensor product will be given in
§3.4.

3.3 Algebraic Expressions: Standard Basis

In this section, our main objective is to demonstrate that each qth-order tensor is simply expressible as a
linear combination of the 3q simplest tensors given by (3.6).
For a 3rd-order tensor M , using (1.39) we deduce

M [uu] = (uu · e i )M
M [eei ] .

Since M [eei ] is a 2nd-order tensor, using (2.41) we infer

M [eei3 ] = Mi1 i2 i3 e i1 ⊗ e i2 .
3.4. ALGEBRAIC EXPRESSIONS: ARBITRARY BASES 53

Then, from these two expressions we derive

M [uu] = Mi1 i2 i3 (uu · e i3 )eei1 ⊗ e i2 .

Hence, using (3.5) we have

M [uu] = Mi1 i2 i3 (eei1 ⊗ e i2 ⊗ e i3 ) [uu] ,
i.e.,
M − Mi1 i2 i3 e i1 ⊗ e i2 ⊗ e i3 ) [uu] = O .
(M
Thus, we arrive at
M = Mi1 i2 i3 e i1 ⊗ e i2 ⊗ e i3 , (3.11)
where
Mi1 i2 i3 = M [eei1 , e i2 , e i3 ] . (3.12)
This may be derived by using (3.9).
Repeatedly following the above procedure, for a qth-order tensor H we may obtain

H = Hi1 ···iq e i1 ⊗ · · · ⊗ e iq , (3.13)

where
Hi1 ···iq = H [eei1 , · · · , e iq ] . (3.14)
We refer to (3.13)–(3.14) as the standard expression of qth-order tensor H relative to the standard basis
ei , and the 3q scalars Hi1 ···iq as the standard components of H relative to the standard basis ei .
Of course, we can use another standard basis ēei to express H and we have

H = H̄i1 ···iq ēei1 ⊗ · · · ⊗ ēeiq ,
(3.15)
H̄i1 ···iq = H [ēei1 , · · · , ēeiq ] .

From (3.9) and (3.10) we deduce

H [ēek1 , · · · , ēekq ] = (ēek1 · e i1 ) · · · ēekq · e iq Hi1 ···iq .

This and (3.15)2 result in the transformation rule

H̄i1 ···iq = (ēek1 · e i1 ) · · · ēekq · e iq Hi1 ···iq . (3.16)

In some cases, tensors of higher order with certain index symmetric or skew-symmetric properties are
taken into account. Symmetry or skew-symmetry of qth-order tensor H (here q = 3) with respect to any
pair of indices in its standard components may be defined by following the corresponding definitions for a
2nd-order tensor. Thus, H is said to be symmetric (+) or skew-symmetric (-) with respect to its rth and sth
indices whenever
Hi1 ···ir ···is ···iq = ±Hi1 ···is ···ir ···iq . (3.17)
These definitions may be extended to cover several pairs of indices. A tensor may be partially symmetric
and/or partially skew-symmetric. Two particular cases are completely symmetric and completely skew-
symmetric. In these cases, tensor H is symmetric (skew-symmetric) with respect to any pair of its indices.

3.4 Algebraic Expressions: Arbitrary Bases

In general, taking an arbitrary basis g i meeting (1.59) and its reciprocal basis g i as given by (1.61), we may
obtain general algebraic expressions for a qth-order tensor. In this case, each e it , 1 ≤ t ≤ q, in the simple
qth-order tensors given by (3.6) may be replaced by either of g it and g it . Accordingly, this will generate so
many possibilities, as exemplified by the case of 2nd-order tensors. Here we only present the results and
do not pursue the details.
54 CHAPTER 3. TENSORS OF HIGHER ORDER

In general, a qth-order tensor H is expressible as follows:

H = Hdi1 c · · · diq c g bi1 e ⊗ · · · ⊗ g biq e , (3.18)

where
Hdi1 c · · · diq c = H [ggbi1 e, · · · , gbiq e] . (3.19)
In the above, a particular treatment is introduced for indices. First, each bit e and each dit c represent a
conjugate pair of subscript and superscript. The correspondence relationship between each such pair bit e
and each dit c is specified as follows: Each dit c can be freely taken as either of a superscript and a subscript,
and, corresponding with these two choices, each bit e is taken as a superscript and a subscript, respectively,
when bit e is taken as a subscript and a superscript, respectively. These are summarized below:

superscript if bit e is taken as subscript ,
dit c = (3.20)
subscript if bit e is taken as superscript .

With the above treatment in mind, (3.18)–(3.19) actually represent 2q different expressions and accord-
ingly we have 2q different kinds of components. For instance, we have 22 (= 4) different expressions for
2nd-order tensor (see (2.62)) and four different kinds of components (see (2.61)). For 3rd- and 4th-order
tensors, we have 23 (= 8) and 24 (= 16) different expressions and 8 and 16 different kinds of components.
Any of the 2q different kinds of components can determines all the others, as exemplified by (2.63)–(2.66)
for 2nd-order tensors. Here we no longer present a large variety of such results.
In (3.18)–(3.19) we can replace basis g i and its reciprocal basis g i with any other basis ḡgi and its
reciprocal basis ḡgi . Then the components of tensor H become

H̄di1 c · · · diq c = H [ḡgdi1 c, · · · , ḡgdiq c] .

We also have 2q different expressions relative to the new pair of bases, ḡgi and ḡgi , and 2q corresponding
different kinds of components. It may be clear that a transformation rule may be derived for the same
kind of components and for any two kinds of components. Such transformation rules are in abundance,
as exemplified in the case of 2nd-order tensors (see (2.67)–(2.70)). Here we no longer supply such results
which increase so rapidly as the order q become larger.
It may be seen that the standard expression of a tensor in terms of a standard basis of vector spaces
is much simpler than its expression in terms of a general non-standard basis. Hence, it may be expedient
to use a standard basis. We would like to point out that, for any given tensor, a standard expression is
never less general than a so-called general expression in terms of an arbitrary basis. Here the essence
is that if either a vector or a tensor is already well-defined or introduced in a basis-free manner, then its
expression in terms of any chosen basis determines the expressions in terms of all the other bases. Of
course, if a tensor would be defined or introduced in terms of bases, then it should be demonstrated that
this definition is applicable to all possible choices of bases. That is why we define and introduce algebraic
operations for vectors and tensors, which yield vectors and tensors, from an abstract standpoint without
involving bases. So long as suitable definitions are ready, algebraic expressions in terms of standard basis
or arbitrary basis may considerably facilicate their understanding and related operations, as has been seen
in the developments till now and in future.

3.5 Dot Products

Let p, q and t be three positive integers and t not exceed the smallest of p and q. Let G and H be pth- and
qth-order tensor, respectively. For a standard basis e i , we have standard components Gi1 ···i p and Hk1 ···kq of
tensors G and h relative to this basis. Then we construct a (p + q − 2t)th-order tensor by

T ≡ Gi1 ···i p−t j1 ··· jt H j1 ··· jt k1 ···kq−t e i1 ⊗ · · · ⊗ e i p−t ⊗ e k1 ⊗ · · · ⊗ e kq−t . (3.21)

For different choices of standard basis, the standard components of tensors G and H are different. Now we
examine how the tensor defined by (3.21) changes with the change of basis.
3.5. DOT PRODUCTS 55

To this end, we choose any other standard basis ēei . The standard components of tensors G and H
relative to this basis are given by Ḡi1 ···i p and H̄k1 ···kq . According to (3.21), we construct the (p + q − 2t)th-
order tensor
T ≡ Ḡi1 ···i p−t j1 ··· jt H̄ j1 ··· jt k1 ···kq−t ēei1 ⊗ · · · ⊗ ēei p−t ⊗ ēek1 ⊗ · · · ⊗ ēekq−t .
T̄ (3.22)
Are the above two (p − q − 2t)th-order tensors, constructed by the same rule but corresponding to two
different bases, the same or different?
To answer this question, applying the transformation rule (3.16) to (3.22) and using the equalities (see
(1.39) and (1.48))
(ēea · e b )(ēea · e c ) = e b · e c = δbc , (ēea · e b )ēea = e b ,
we deduce



T
T̄ = (ēei1 · e l1 ) · · · ēei p−t · e l p−t (ēe j1 · e m1 ) · · · (ēe jt · e mt ) Gl1 ···l p−t m1 ···mt ×



(ēe j1 · e r1 ) · · · (ēe jt · e rt ) (ēek1 · e s1 ) · · · ēekq−t · e sq−t Hr1 ···rt s1 ···sq−t ×
ēei1 ⊗ · · · ⊗ ēei p−t ⊗ ēek1 ⊗ · · · ⊗ ēekq−t
= δm1 r1 · · · δmt rt Gl1 ···l p−t m1 ···mt Hr1 ···rt s1 ···sq−t ēel1 ⊗ · · · ⊗ e l p−t ⊗ e s1 ⊗ · · · ⊗ e sq−t
= Gi1 ···i p−t j1 ··· jt H j1 ··· jt k1 ···kq−t e i1 ⊗ · · · ⊗ e p−t ⊗ e k1 ⊗ · · · ⊗ e kq−t = T .
It turns out that expression (3.21) always yields the same (p + q − 2t)th-order tensor for all the choices
t
of basis e i . We designate this tensor by G H , namely,
t
G H ≡ Gi1 ···i p−t j1 ··· jt H j1 ··· jt k1 ···kq−t e i1 ⊗ · · · ⊗ e i p−t ⊗ e k1 ⊗ · · · ⊗ e kq−t . (3.23)
Generally, we say that (3.23) is the t-dot product of pth- and qth-order tensors G and H . It may be easily
understood that this dot product is bilinear, i.e.,
( t t t
G (α1 H 1 + α2 H 2 ) = α1 G H 1 + α2 G H 2 ,
t t t (3.24)
(α1 G 1 + α2 G 2 ) H = α1 G 1 H + α2 G 2 H ,
for any numbers α1 and α2 and for any pth- and qth-order tensors G 1 , G 2 and H 1 , H 2 .
A number of useful operations for tensors may be regarded as examples of the above definition. In
particular, we set
0
G⊗H ≡ G H, (3.25)
and we refer to this zero-dot product as the tensor product of tensors G and H . It may be clear that the
dyadic product of two vectors and the tensor product of a tensor and a vector, introduced by (2.36) and
(3.5), are its two particular cases.
For p ≥ q = t or q ≥ p = t we set
t
H] ≡ G
G [H H ≡G H (3.26)
for any pth-order tensor G and qth-order tensor H . In the case p > q = t, we say that tensor G transforms
linearly each qth-order tensor into a (p − q)th-order tensor through (3.26). Thus, a tensor of higher order is
endowed with multiple transformation functions: For each positive integer p < q, a qth-order tensor may
be used to linearly transform each pth-order tensor into a (q − p)th-order tensor. Clearly, (2.8), (3.1) and
(3.3) etc. are particular cases of this general notion.
When G and H are pth-order tensors, we have
p
G H ≡G H, (3.27)
which yields a scalar. We call this the scalar product of tensors G and H of the same order. Evidently, we
have
G|2 ≡ G G > 0 for each non-zero tensor G .
|G (3.28)
The scalar |G
G| is called the magnitude or norm of tensor G . In particular, we have
u·v = u v
for vectors u and v .
56 CHAPTER 3. TENSORS OF HIGHER ORDER

3.6 Orthogonal Transformations of Vectors and Tensors

In many cases, such as in the formulation and study of two basic principles in continuum mechanics,
i.e., material objectivity and material symmetry, we are concerned with the transformation of vectors and
tensors under the isometric change of observing frame or referential configuration. Such a transformation
preserves the magnitudes of vectors and tensors, and hence preserves essential geometrical features of
vectors and tensors, except for orientational and directional properties. Orthogonal tensors introduced in
§2.7 are just what we need in performing this transformation. A detailed describtion is below.
Let Q and G be an orthogonal tensor and an pth-order tensor, respectively. We introduce a pth-order
tensor Q ? G by

(Q QT [uu1 ], · · · , Q T [uu p ]] for any p vectors u 1 , · · · , u p .

Q ? G )[uu1 , · · · , u p ] ≡ G [Q (3.29)

In terms of a standard basis, this is equivalent to

Q ? G )i1 ···i p = Qi1 k1 · · · Qi p k p Gk1 ···k p .

(Q (3.30)

We say that the above defines the orthogonal transformation of tensor G performed by Q , and Q ? G is the
Q -trasformed tensor of G . The following properties of the above orthogonal transformation may be useful:

(±II ) ? G = (±1) p G , (3.31)

Q ? (α1 G 1 + α2 G 2 ) = α1 Q ? G 1 + α2 Q ? G 2 , (3.32)
Q1 Q 2 ) ? G = Q 1 ? (Q
(Q Q2 ? G ) , (3.33)

t t
G
Q ? (G Q ? G)
H ) = (Q Q ?H),
(Q (3.34)
|Q
Q ? G | = |G
G| , (3.35)
for any two numbers α1 , α2 , for any orthogonal tensors Q , Q 1 , Q 2 , for any pth-order tensors G , G 1 , G 2 , for
any qth-order tensor H , and for any integer falling within 0 and the smaller of p and q. The last expression
above just indicates that the orthogonal transformation of tensor preserves the magnitude.
In particular, the orthogonal transformation of vector u and 2nd-order tensor S performed by Q are
given by

Q ? u = Q[uu], Q ? S = Q S QT . (3.36)
In the succeeding sections we shall study certain kinds of 3rd- and 4th-order tensors that are useful in
mechanics and engineering and other related fields.

3.7 Third-Order Tensors

Well-known examples of applying 3rd-order tensors are characterizations of direct and converse piezoelec-
tricity effects in some fascinating crystals. According to experimental observations, in some crystals, such
as quartz crystals, etc., application of stress induces electric field, and, conversely, application of electric
field generates stress and strain. The exact mathematical formulations of these effects are two kinds of
linear tensor transformation relations. One is to transform linearly each applied symmetric 2nd-order stress
tensor into an induced electric field vector and, conversely, the other is to transform linearly each applied
electric field vector into an induced symmetric 2nd-order stress tensor. These two kinds of linear tensor
transformation relations are performed and hence represented by 3rd-order tensors. Details may be found
in Nye (1985).
According to the definition introduced before, a 3rd-order tensor M transforms each vector u into a
2nd-order tensor S , i.e.,
S = M [uu] ≡ M · u . (3.37)
3.8. FOURTH-ORDER TENSORS 57

M can also linearly transform each 2nd-

On the other hand, from (3.26) we know that a 3rd-order tensor M̂
order tensor S into a vector u , i.e.,
M [SS ] = M̂
u = M̂ M : S. (3.38)
In crystal physics, the aforementioned direct and converse piezoelectricity effects are formulated by these
two linear tensor relations. For details, refer to Nye (1985).
A useful 3rd-order tensor is given as follows:

ε ≡ εi jk e i ⊗ e j ⊗ e k , (3.39)

where ε is the permutation symbol given by (2.53). The tensor ε given above is completely skew-symmetric
and known as 3rd-order permutation tensor.
Some applications of the permutation tensor ε are as follows:
1
u × v = (εε · v ) · u = ε [uu ⊗ v − v ⊗ u ] , (3.40)
2
ε [uu × v ] = u ⊗ v − v ⊗ u ; (3.41)
1 1
ω = ε [ΩΩ] = ε : Ω , (3.42)
2 2
ω] = ε · ω ;
Ω = ε [ω (3.43)
for any vectors u and v and for any skew-symmetric 2nd-order tensor Ω and its associated axial vector ω
(see (2.137)–(2.138)).

3.8 Fourth-Order Tensors

Fourth-order tensors are used to characterize classical and micropolar linear elasticity properties and pho-
toelasticity properties of solids, such as crystals, etc. The mathematical formulations of these properties are
linear tensor transformation relations which transform each symmetric or non-symmetric 2nd-order tensor
into a symmetric or a non-symmetric tensor. For details, refer to Nye (1985).
A 4th-order tensor L is endowed with richer transformation functions: it can transform each vector u ,
each 2nd-order tensor A , and each 3rd-order tensor M into a 3rd-order tensor M̂ M , a 2nd-order tensor S and
a vector v , respectively. Namely
M = L[uu] ≡ L · u ;
M̂ (3.44)
A] ≡ L : A ;
S = L[A (3.45)
.
M ] ≡ L..M
v = L[M M. (3.46)
The usually used 4th-order tensors are of certain index symmetry properties. In what follows we focus
attention on these cases.
A 4th-order tensor L is said to be of minor symmetry, whenever

Li jkl = L jikl = Li jlk . (3.47)

On the other hand, L is said to be of major symmetry, whenever

Li jkl = Lkli j . (3.48)

The major symmetry is equivalent to

S A]) = A
(L[A (L[SS ]) (3.49)

for any two 2nd-order tensors S and A .

Examples of 4th-order tensors with minor symmetry are supplied by photoelasticity tensors and clas-
sical Cauchy elasticity tensors. Classical Green elasticity tensor derived from a quadratic strain-energy
function provide examples of 4th-order tensors with both minor and major symmetry.
58 CHAPTER 3. TENSORS OF HIGHER ORDER

Four particular yet useful 4th-order tensors are given by

I = ei ⊗ e j ⊗ ei ⊗ e j , (3.50)

1
Isym =
(eei ⊗ e j ⊗ e i ⊗ e j + e i ⊗ e j ⊗ e j ⊗ e i ) , (3.51)
2
1
Iskw = (eei ⊗ e j ⊗ e i ⊗ e j − e i ⊗ e j ⊗ e j ⊗ e i ) , (3.52)
2
1
Īsym = Isym − I ⊗ I . (3.53)
3
Evidently, we have the decomposition
I = Isym + Iskw , (3.54)
1
Isym = I ⊗ I + Īsym , (3.55)
3
which form a correspondence with the additive decompositions (2.142)–(2.143) and (2.144), respectively.
The standard components of the foregoing four tensors are given by

Ii jkl = δik δ jl , (3.56)

sym 1

Ii jkl = δik δ jl + δil δ jk , (3.57)
2
1
Iskw


i jkl = δik δ jl − δil δ jk , (3.58)
2
1
1
Īsym = δik δ jl + δil δ jk − δi j δkl , (3.59)
2 3
relative to any standard basis ei . A noticeable fact is that these components assume the same forms for all
the bases.
Moreover, the following properties hold:

I [SS ] = S ; (3.60)

1
Isym [SS ] =S + S T = symSS ;


(3.61)
2
1
Iskw [SS ] = S − S T = skwSS ;


(3.62)
2
for any 2nd-order tensor S . In particular, we have

Isym [A
A] = A, Isym [Ω
Ω] = O ; (3.63)

Iskw [A
A] = O , Iskw [Ω
Ω] = Ω ; (3.64)
Īsym [A
A] = Ā
A; (3.65)
for any symmetric and antisymmetric tensors A and Ω .
A useful operation for two 4th-order tensors T and L is called composite product of C and L, which
yields a 4th-order tensor and is denoted T ◦ L and defined by

(T ◦ L) [SS ] ≡ (T[L[SS ]) for any 2nd-order tensor S . (3.66)

We have
T ◦ L = T : L = Ti jrs Lrskl e i ⊗ e j ⊗ e k ⊗ e l . (3.67)
The following rules hold true:
L◦I = I◦L = L, (3.68)
3.9. COMPONENT REPRESENTATIONS 59

for each 4th-order tensor L, and

C ◦ Isym = Isym ◦ C = C (3.69)
for each 4th-order tensor C with minor symmetry. Moreover, we have

L1 ◦ (L2 ◦ L3 ) = (L1 ◦ L2 ) ◦ L3 ≡ L1 ◦ L2 ◦ L3 (3.70)

for any three 4th-order tensors Li .

A 4th-order tensor tensor L is said to be nonsingular if there is a 4th-order tensor L̂ such that

L ◦ L̂ = L̂ ◦ L = I . (3.71)

Such a L̂ is called the inverse tensor of 4th-order tensor L, denoted L−1 .

In some cases, the 2nd-order tensor S and A in the transformation relation (3.45) are restricted to the
case of symmetry, as in Hooke’s law for classical linear elasticity. In this restrictive sense, the tensor L
is at least of minor symmetry (3.47), and is further of major symmetry (3.48) whenever it can be derived
from a quadratic strain-energy function. We say that a 4th-order tensor tensor L with minor symmetry is
nonsingular if there is a 4th-order tensor L̂ with minor symmetry such that

L ◦ L̂ = L̂ ◦ L = Isym . (3.72)

Such a L̂ is called the inverse tensor of 4th-order tensor L with minor symmetry, also denoted L−1 .
With the above definitions of inverse tensors and (3.66), we may invert (3.45) for a nonsingular L.
Indeed, we have
L−1 [SS ] = L−1 [L[A
A]) = L−1 ◦ L [A


A] .
From this and either (3.68) (non-symmetric S and E ) or (3.69) (symmetric S and E ), we derive

A = L−1 [SS ] . (3.73)

A further account of 4th-order tensors will be presented in §3.10 by extending the notions of vector and
2nd-order tensors.

3.9 Component Representations

Until now we have expressed vectors and tensors and their operations and relations etc. in direct notation.
They are known as direct or absolute representations. Since vectors and tensors are basis-free or basis-
independent entities, these direct representations would be indeed direct to intrinsic properties of vectors
and tensors and their operations and relations, etc. However, it may be helpful to have another way per-
forming concrete and specific derivations and calculations. For this purpose, component representations
may be used. In this case, we express vectors and tensors and their operations and relations etc. in terms
of their components relative to any chosen basis e i . First, a vector u is represented by ui , and 2nd-, 3rd-,
4th- and, generally, qth-order tensor by Si j , Mi jk , Li jkl and Hi1 ,···iq , respectively. Then, tensor operations
and relations are expressed in terms of these components. Some examples are given as follows:

Multiplication by number:

αuu −→ αui ; αSS −→ αSi j ; M −→ αMi jk ,

αM αL −→ αLi jkl ; (3.74)

Addition and subtraction:

u ± v −→ ui ± vi ; S ± T −→ Si j ± Ti j ; M ± N −→ Mi jk ± Ni jk ; L ± C −→ Li jkl ± Ci jkl ; (3.75)

Vector product and mixed product:

u × v −→ εi jk u j vk ; (3.76)
(uu × v ) · w −→ εi jk ui v j wk ; (3.77)
60 CHAPTER 3. TENSORS OF HIGHER ORDER

Composite products:
S T −→ Sir Tr j ; L ◦ C −→ Li jrs Crskl ; (3.78)
Tensor products:

u ⊗ v −→ ui v j ; S ⊗ u −→ Si j uk ; S ⊗ T −→ Si j Tkl ; ··· ; G ⊗ H −→ Gi1 ···iq H j1 ··· jq ; (3.79)

Scalar products:

u · v −→ ui vi ; S T −→ Si j Ti j ; ··· ; G H −→ Gi1 ···iq Hi1 ···iq ; (3.80)

u · S [vv] −→ ui Si j v j ; A B] −→ Ai j Li jkl Bkl ;

L[B (3.81)
Linear transformation relations:
v = S [uu] −→ vi = Si j u j ; (3.82)
S = M [uu] −→ Si j = Mi jk uk ; (3.83)
M [SS ] −→ ui = M̂i jk S jk ;
u = M̂ (3.84)
A] −→ Si j = Li jkl Akl .
S = L[A (3.85)
In principle, one can convert any tensor operation and relation in direct notation to a form in terms of
component representation. In some cases, the latter might assume an awkward form. For instance, that is
the case for the characteristic expression of symmetric tensor and the higher powers of 2nd-order tensor,
etc.
In most cases, component representation is perhaps very helpful, accessible, useful and even indispens-
able as intermediate steps in checking, proving and deriving tensor operations and relations, etc.

3.10 On Vectors and Tensors in General Sense

3.10.1 Abstract vectors and tensors in general sense
It may be seen that vectors and vector spaces serve as a basic starting point in introducing and defining
2nd-order tensors and tensors of higher order. A basis of vector spaces may be used as building blocks
to construct tensors by means of the notion of tensor products. On account of these facts, we may regard
vector spaces as base spaces for qth-order tensor spaces.
However, vector spaces and accordingly tensor spaces may be introduced in a general yet abstract
sense. Then, we have the notion of n-dimensional vector spaces in an abstract sense: Each of them is
regarded as a collection of certain members formally called vectors, for which formal, abstract addition
and multiplication by number, obeying the rules (1.1)–(1.10), are defined and performed in a self-contained
manner. Refer to, e.g., Halmos (1974) for details. Then, using these abstract vector spaces as base spaces
and formally following the same procedures used in the previous sections, we may introduce and define
2nd-order tensors and tensors of higher order in an abstract sense.
For example, we have introduced qth-order tensors using 3-dimensional Euclidean vector spaces as
base spaces and we define addition and multiplication by number for these tensors. Then we may regard
each such qth-order tensor space as a 3q -dimensional vector space. Hence, we can define pth-order tensor
spaces based on these 3q -dimensional vector spaces.
An even more abstract and general notion of vectors and tensors may be introduced by completely
abandoning the scalar product. First, we introduce a vector space V as a collection of members formally
called vectors in which addition and multiplication by number are defined and can be carried out in a self-
contained manner. Then we consider the collection V ∗ formed by all the linear mappings u ∗ from V to the
reals R , i.e.,
u ∗ : v ∈ V 7→ ξ = u ∗ [vv] ∈ R ,
u ∗ [α1 v 1 + α2 v 2 ] = α1 u ∗ [vv1 ] + α2 u ∗ [vv2 ],
u ∗ [αvv] = αuu∗ [vv],
3.10. ON VECTORS AND TENSORS IN GENERAL SENSE 61

for any v 1 , v 2 , v ∈ V and for any α1 , α2 , α ∈ R . In V ∗ , we define addition of any two linear mappings
u ∗1 , u ∗2 ∈ V ∗ and multiplication of each linear mapping u ∗ by number α ∈ R by

(uu∗1 + u ∗2 )[vv] ≡ u ∗1 [vv] + u ∗2 [vv] for any v ∈ V ;

(αuu∗ )[vv] ≡ αuu∗ [vv] for any v ∈ V .

In so doing, the collection V ∗ also becomes a vector space in an abstract sense, called the dual space of
vector space V .
Let V̄1 , · · ·, V̄r be r vector spaces, each of which is either the space V or its dual space V ∗ . We refer
to each r-fold linear mapping

T : (xx1 , · · · , x r ) ∈ V̄1 × · · · × V̄r −→ T [xx1 , · · · , x r ] ∈ R ,

with the linearity property

T [xx1 , · · · , axxα + a0 x 0α , · · · , x r ] = aT
T [xx1 , · · · , x α , · · · , x r ] + a0 T [xx1 , · · · , x 0α , · · · , x r ],

as a rth-order tensor.
There are so many possibilities for the rth-order tensors defined above. Indeed, since there are two
choices for each V̄α , i.e., either V or V ∗ , totally we have 2r different Cartesian products V̄1 × · · · × V̄r of
spaces, and, accordingly, the foregoing definition actually defines 2r different types of rth-order tensors. In
particular, we have four different types of 2nd-order tensor, they are linear mappings from V × V to R ,
from V ∗ × V to R , from V × V ∗ to R , from V ∗ × V ∗ to R , separately. They can be also regarded as
linear transformations from V to V ∗ , from V to V , from V ∗ to V ∗ , from V ∗ to V , respectively.
From an abstract and general viewpoint, a vector space and its dual space may be essentially different,
and hence different types of vectors and tensors should be distinguished, as shown above. However, a
Euclidean vector space, in which scalar product can be defined, and its dual space coincide, the above
difference thus disappears.
Here we do not pursue a detailed account of general tensors along the above abstract line. For details,
refer to, e.g., Bowen and Wang (1976). This general notion may be used to treat tensor algebra and analysis
on manifolds related to deformable continua. Systematic and fruitful applications in nonlinear elasticity
and in nonlinear continuum mechanics are presented, e.g., by Marsden and Hughes (1983) and Bertram
(1989), et al. Some recent results are surveyed by Stumpf and Hoppe (1997).
The above general notion may lead to new understanding of the tensors introduced before. In the next
subsection we touch only on 4th-order tensors which are viewed as 2nd-order tensors over usual 2nd-order
tensor spaces. Their physical backgrounds are classical and micropolar elasticity tensors either in Cauchy’s
or in Green’s sense. Besides, photoelasticity is also characterized by 4th-order tensor. In this respect, refer
to Nye (1985) for details.

3.10.2 Characteristic expressions of 4th-order tensors

We may regard each 4th-order tensor L discussed in §3.7 as a 2nd-order tensor over a 9-dimensional vector
space, the latter just being 2nd-order tensor spaces introduced in §2.4. In particular, we may regard each
4th-order tensor L with both minor and major symmetry properties (3.47)–(3.48) as a symmetric 2nd-
order tensor over a 6-dimensional vector space, the latter being formed just by symmetric 2nd-order tensor
discussed in §3.8. In this sense, we may discuss the characteristic properties of usual 4th-order tensors as
symmetric 2nd-order tensors over 6- or 9-dimensional vector spaces. Paralleling the discussion in §2.6, we
say a real number λ is an eigenvalue of the 4th-order tensor T, if there exits a non-zero 2nd-order tensor K
such that
K ] = λK
T[K K. (3.86)
We call K an eigentensor of T subordinate to λ. Similarly, we may derive the characteristic equation of
T. Its real roots supply the eigenvalues of T. If T has major symmetry, i.e., it is a symmetric as a 2nd-
order tensor over usual 2nd-order tensor space, then each of its characteristic roots is real and hence an
eigenvalue.
62 CHAPTER 3. TENSORS OF HIGHER ORDER

Following the same procedure in §2.6.2, we may derive the characteristic expression

C = λ1 A 1 ⊗ A 1 + · · · + λ6 A 6 ⊗ A 6 (3.87)

for each 4th-order tensor C with both minor and major symmetry properties, which is regarded to be a
symmetric 2nd-order tensor over a 6-dimensional vector space formed by symmetric 2nd-order tensors
over a 3-dimensional base vector space. The characteristic equation of the 4th-order tensor C as a 6-
dimensional 2nd-order tensor determines its six eigenvalues λ1 , · · ·, λ6 (see, e.g., Rychlewski 1984; Betten
1987, 1993). Now, each “eigenvector” A σ in the above, i.e., eigentensor, is a symmetric 2nd-order tensor
over a 3-dimensional vector space and the following orthonormalization condition holds:

Aσ A τ = δστ , σ, τ = 1, · · · , 6 , (3.88)

i.e., the six eigentensors A σ form a standard basis of usual symmetric 2nd-order tensor spaces as 6-
dimensional vector spaces. We have

A 1 ⊗ A 1 + · · · + A 6 ⊗ A 6 = Isym . (3.89)

For each 4th-order tensor L with major symmetry property, similarly we have the characteristic expres-
sion
L = λ1 E 1 ⊗ E 1 + · · · + λ9 E 9 ⊗ E 9 . (3.90)
Now each “eigenvector” E r in the above, i.e., eigentensor, is a 2nd-order tensor over a 3-dimensional vector
space, which need not be symmetric. The following properies hold:

Er E t = δrt , r, t = 1, · · · , 9 , (3.91)

E1 ⊗E1 +···+E9 ⊗E9 = I, (3.92)

i.e., the nine eigentensors E σ form a standard basis of usual 2nd-order tensor spaces as 9-dimensional
vector spaces.
Assuming that both the tensor C and the tensor L are non-singular, i.e., each of their respective eigen-
values is non-vanishing, then from (3.87)-(3.89) and from (3.90)-(3.92) we derive the following simple
expressions for the inverse tensors of C and L:

C−1 = λ−1 −1
1 A 1 ⊗ A 1 + · · · + λ6 A 6 ⊗ A 6 , (3.93)

L−1 = λ1−1 E 1 ⊗ E 1 + · · · + λ−1

9 E9 ⊗E9 . (3.94)
Moreover, let the symmetric tensors A and S represent the strain tensor and the stress tensor, and let (3.53)
represent Hooke’s law for classical linear elasticity and its inverse be given by (3.81). Then, we have
1
W≡ S A=A A]) = λ1 (A
(C[A A A 1 )2 + · · · + λ6 (A
A A 6 )2 , (3.95)
2

W =S (C−1 [SS ]) = λ−1 S

1 (S A 1 )2 + · · · + λ−1 S
6 (S A 6 )2 . (3.96)
These two expressions provide simple expressions for uncoupled orthogonal decompositions of the elastic
strain energy density W in terms of the elastic stiffness tensor C and the elastic compliance tensor C−1 ,
respectively. For details, refer to Rychlewski (1984a, b, 1985, 1995, 2000).
Let λ1 , · · ·, λr be all the distinct eigenvalues of a 4th-order tensor T with major symmetry property.
Then (3.87) and (3.90) may be recast as

T = λ1 P1 + · · · + λr Pθ . (3.97)

In the above, Pθ is the eigenprojection of T subordinate to λθ . The following simple properties hold:

T ◦ Pθ = Pθ ◦ T = λθ Pθ (no summation), (3.98)

3.10. ON VECTORS AND TENSORS IN GENERAL SENSE 63
(
Isym , T = C,
P1 + · · · + Pθ = J = (3.99)
I, T = L,

O, σ 6= τ ,
Pσ ◦ Pτ = (3.100)
Pσ , σ = τ.
Moreover, each eigenprojection may be uniquely determined also by Sylvester’s formula:
r
T − λσ J
Pθ = ∏ λθ − λσ . (3.101)
σ6=θ
σ=1

Note that the multiplication here is the composite product signified by “◦”. Accordingly, the expressions
(3.93)-(3.96) become
T−1 = λ−1 −1
1 P1 + · · · + λr Pr , (3.102)

A]|2 + · · · + λr |Pr [A
W = λ1 |P1 [A A]|2 , (3.103)

W = λ1−1 |P1 [SS ]|2 + · · · + λ−1 S ]|2 .

r |Pr [S (3.104)
The idea of studying the eigentensors of the 4th-order tensors, in particular, the elasticity tensor, as
shown above, dates back to W. Thomson (Lord Kelvin) (1856, 1878). Its systematic developments in
modern sense were made by Rychlewski (1983, 1984a, b, c, 1995, 2000) in studying the intrinsic structure
of Hooke’s law for classical linear elasticity and in formulating a unified limit criterion for the yielding of
plastic solids, etc. Studies were also made by Betten (1987, 1993, 1998), Mehrabadi and Cowin (1990),
Bertram and Olschewski (1991, 1993), Cowin and Mehrabadi (1992, 1995), Sutcliffe (1992), et al., in
the context of classical Green elasticity tensor, etc. A treatment covering both classical and micropolar
elasticity tensors was given by Xiao (1995c). Details in these respects may be found in these references
and related references therein. On the other hand, Betten (1987, 1993) and Betten and Helisch (1992)
derived a number of irreducible isotropic invariants of the 4th-order tensor C by studying the characteristic
properties of C. Results were also given later by Ting (1987), Boehler, Kirillov and Onat (1994), et al. A
further study was given in Xiao (1997c).
Finally, we like to remark that the characteristic expression (3.87) with (3.88)-(3.89) will emerge nat-
urally in studying the derivative of a general class of finite strain tensors and in dealing with some linear
tensor equations related to kinematics of finite deformations of continua, as shown in Xiao (1995a), Xiao,
Bruhns and Meyers (1997, 1998d), Bruhns, Meyers and Xiao (2002), and as will be indicated in 5.5.3.
64 CHAPTER 3. TENSORS OF HIGHER ORDER
Chapter 4

Scalar, Vector and Tensor Fields

In mechanics and engineering, we treat various shapes of material bodies. In courses of motions and
deformations of these bodies, we introduce scalar, vector and tensor quantities at each material particle to
characterize and represent the kinematic and physical state at this particle, such as the deformation state
and stressed state, etc. Accordingly, we are concerned with scalar, vector and tensor quantities each of
which distributes and changes over a spatial domain. A basic question in applications is to study how these
quantities change with the spatial positions, which leads to various relevant differentiation and integarion
operations. These will be discussed in this chapter.

4.1 Motivation and Definition

A continuous material body occupies a domain in the conventional 3-dimensional space. The latter is
composed of points. Then, each particle in a material body will occupy a spatial point at each instant and
move from one point to another. Accordingly, various kinds of scalar, vector and tensor quantities, such
as mass density, displacement vector, stress and strain tensors, etc., may be defined at each point in the
spatial domain occupied by the material body and on its bounding surface. Hence, we have scalar, vector
and tensor quantities depending on the spatial points occupied by the material body. We refer to them as
scalar, vector and tensor fields, separately. Fig. 1.1 schematically shows an example of vector field.
The conventional 3-dimensional space in which physical events happen is a 3-dimensional Euclidean
point space. Here we would not like to go into details of this object as an abstract mathematical notion.
Here, we keep one thing in mind: there is a natural distance between any two points. For the sake of
simplicity and clarity, we may identify each point as a vector in the usual manner, as shown below.
We choose and fix a point o, called an origin. Then we represent any point x by a vector connecting
points o and x with the arrowhead at point p, as has been done in §1.1.1 in chapter 1. The length (magnitude)
of this vector is just the distance between points o and x. We call this vector the position vector of point x
relative to the origin o and designate it by x , with the understanding that an origin o is chosen and fixed.
In this manner, each point x may be represented by and identified with its position vector x . From now on,
point x is simply said to be point x as position vector. Thus, a scalar (vector, 2nd-order tensor) field will
specify a scalar (vector,2nd-order tensor, etc.) quantity depending on the point ( position vector) x , namely,
ξ = ξ(xx); v = v (xx); T = T (xx) , (4.1)
where the point x varies in the spatial domain occupied by a material body. In this sense, scalar, vector
and tensor fields are scalar-, vector- and tensor-valued functions ξ(xx), v (xx) and T (xx) of the point x as the
position vector, respectively. Generally, we may speak of a qth-order tensor field as shown below:
ϕ = ϕ(xx) . (4.2)
We may further choose a common standard basis e i of vector spaces at all spatial points. Then each
point x has the standard expression
x = xi e i . (4.3)

65
66 CHAPTER 4. SCALAR, VECTOR AND TENSOR FIELDS

The three standard components xi are referred to as the (rectangular) Cartesian coordinates of point x
relative to the origin. The (o; x1 , x2 , x3 ) with the origin o and the Cartesian coordinates xi is said to form a
(rectangular) Cartesian coordinate system.
Certain geometrical background and features of Cartesian coordinate systems will be elaborated on in
§4.5. At this stage, we remark that there are infinite possibilities of choosing standard or arbitrary bases,
since the choice may be different at different spatial points. Here we make the simplest choice: the same
standard basis at all points. Other useful choices will be discussed later on.
Relative to a Cartesian coordinate system, the foregoing scalar, vector and tensor fields as scalar-,
vector- and tensor-valued functions of the position vector x have the following standard representations

ξ = ξ(x1 , x2 , x3 ) ; (4.4)

v = vi (x1 , x2 , x3 )eei ; (4.5)

T = Ti j (x1 , x2 , x3 )eei ⊗ e j . (4.6)
Generally, we have
ϕ(xx) = ϕi1 ···iq (x1 , x2 , x3 )eei1 ⊗ · · · ⊗ eiq . (4.7)
Hence, relative to a common Cartesian basis, a scalar field, each standard component of either a vector
field or a tensor field is a real function of the three Cartesian coordinates xi of the point x .

4.2 Continuity and Gradients

We have known that a scalar, vector, tensor field represent a quantity relying on and changing with the
spatial points in a domain occupied by a continuous material body. Vital to further studies and analyses
of these field quantities is to know how these quantities change with the spatial points. This leads to the
notion of continuity and gradient of field quantities.
Let ϕ(xx) represent a qth-order tensor field, where q=0, 1, 2 correspond to scalar, vector, tensor field,
respectively, e.g., those as shown by (4.1). We say that the field quantity ϕ is continuous at point x , if, for
any given small δ > 0, there exists a small ε > 0 such that

|ϕ(xx0 ) − ϕ(xx)| < δ (4.8)

for any two points x and x 0 meeting

|xx0 − x | < ε . (4.9)
From the continuity at point x , we know that the values of a field quantity at any two points x and x 0
that are close to each other should be also close to each other. To have a further, better understanding of
the values of a field quantity at two neighbouring points, we need the notion of gradients.
We say that the qth-order tensor field ϕ(xx) is differentiable at point x , if there exists a given (q + 1)th-
order tensor, called the gradient of the field quantity ϕ(xx) at point x and denoted ∂ϕ/∂xx, such that
∂ϕ 0
ϕ(xx0 ) = ϕ(xx) + [xx − x ] + o (|xx0 − x |) , (4.10)
∂xx
for any two neighbouring points x and x 0 . Here and henceforward, we use o(x) to represent a small vector
or tensor quantity that relies on a small scalar variable x and goes to vanish more rapidly whenvever x goes
to vanish, as indicated by (5.153) given later.
Suppose that the field quantity ϕ(xx) is differentiable at point x . Replacing x 0 in (4.10) by x + εaa, where
ε is a small number and a is any given vector, we obtain the following useful formula:
∂ϕ ϕ(xx + εaa) − ϕ(xx)
[aa] = lim . (4.11)
∂xx ε→0 ε
It should be noted that this is a consequence of the differentiability property at point x . The converse need
not be true in general. In particular, for a given unit vector a , (4.11) is known as the directional or Gâteaux
derivative of the field quantity ϕ(xx) in the direction a at x .
4.3. CURLY, DIVERGENT AND LAPLACIAN DERIVATIVES 67

We say that the field quantity ϕ(xx) is a continuous (differentiable) field , if it is continuous (differen-
tiable) at all points in the related spatial domain. A differentiable field quantity is continuous, whereas the
converse need not be true.
Some useful rules for gradients are as follows:

∂(ϕ1 + ϕ2 ) ∂ϕ1 ∂ϕ2 ∂(ξϕ) ∂ϕ ∂ξ

= + , = ξ +ϕ⊗ , (4.12)
∂xx ∂xx ∂xx ∂xx ∂xx ∂xx
for any differential scalar field ξ and for any differentiable qth-order tensor fields ϕ, ϕ1 and ϕ2 . Note that
each tensor field emerging in the above is defined point by point.
For a differentiable qth-order tensor field ϕ(xx), the gradient ∂ϕ/∂xx can be defined at each point x and
is a (q + 1)th-order tensor field. Specifically, if ϕ(xx) represents differentiable scalar, vector, tensor fields
as shown in (4.1), then the corresponding gradients ∂ξ/∂xx, ∂vv/∂xx, ∂T T /∂xx are vector, 2nd-order tensor,
3rd-order tensor fields, respectively. From them, several useful differential operators may be derived. They
will be discussed below, separately.

4.3 Curly, Divergent and Laplacian Derivatives

Starting from the gradients of differentiable scalar field, vector field and tensor field, several associated dif-
ferential operations may be introduced, as will be done below. These differential operations are frequently
used in deriving basic equations governing field quantities in continuous deformable bodies, such as equa-
tion for the conservation of mass, Cauchy’s equations of equilibrium and motion in continuum mechanics,
Navier-Stokes equations in fluid dynamics, etc.
First, for a differentiable vector field v = v (xx), its gradient ∂vv/∂xx is a 2nd-order tensor field. The
divergent derivative of vector field v = v (xx) is a scalar field, designated by divvv and defined by

∂vv ∂vv
divvv ≡ tr = I. (4.13)
∂xx ∂xx
On the other hand, the curly or rotatory derivative of vector field v = v (xx) is a vector field, designated by
curlvv and defined by
 
∂vv
curl v ≡ ε . (4.14)
∂xx
In the above, I and ε are the 2nd-order identity tensor and the 3rd-order permutation tensor given by (3.39).
Sometimes, curlvv is also denoted rotvv.
In particular, if the tensor field v = v (xx) in (4.13) is replaced by the gradient of a scalar field ξ = ξ(xx),
then we treat the divergent derivative of the gradient of the scalar field ξ = ξ(xx), which is referred to as the
Laplacian derivative of this scalar field, or, simply, the Laplacian of ξ, and usually denoted ∇2 ξ. We have

∂ξ
∇2 ξ ≡ div . (4.15)
∂xx

Next, for a differentiable 2nd-order tensor field T = T (xx), its gradient ∂T

T /∂xx is a 3rd-order tensor field.
The divergent derivative of tensor field T = T (xx) is a vector field, designated by divT T and defined by

TT
∂T
T≡
divT I. (4.16)
∂xx

In particular, if the tensor field T = T (xx) in (4.16) is given by the gradient of a vector field v = v (xx), then
we treat the divergent derivative of the gradient of the vector field v = v (xx), which is called the Laplacian
derivative of the vector field v = v (xx), or, simply, Laplacian of v = v (xx) and denoted

∂vv
∇2 v ≡ div . (4.17)
∂xx
68 CHAPTER 4. SCALAR, VECTOR AND TENSOR FIELDS

It may be evident each of the above derivatives is linear. Some useful rules for the derivatives introduced
above are recorded as follows.
∂ξ
div(ξvv) = ξdivvv + v · ; (4.18)
∂xx
∂ξ
T ) = ξdivT
div(ξT T+ ·T ; (4.19)
∂xx
∂ξ
T · v ) = (divT
div(T T)·v + ·T ; (4.20)
∂xx
∂ξ
curl(ξvv) = ξcurlvv + ×v; (4.21)
∂xx
∂(divvv)
curl(curlvv) = − ∇2 v ; (4.22)
∂xx
curl(divvv) = 0 , div(curlvv) = 0 ; (4.23)
for any differentiable scalar, vector, 2nd-order tensor fields as shown by (4.1).

4.4 Differentiation in Cartesian Coordinates

Relative to a Cartesian coordinate system, scalar, vector and tensor fields as shown by (4.1) have the
standard expressions given by (4.3)–(4.5). In this section, we shall derive explicit expressions for the
forgoing gradients and curly, divergent, Laplacian derivatives in terms of partial derivatives of the standard
components in the Cartesian coordinates. With a Cartesian coordinate system, the results are quite simple
and concise, and they are sufficiently general for use in most of application in mechanics and engineering.
To derive the desired results, we evaluate (4.11) for each a = e i . Utilizing (4.11) and the standard
expressions (4.2), (1.39), (4.7) for the point x , the vector a and the qth-order tensor field ϕ relative to the
chosen common basis e i , we infer

∂ϕ ϕ(xx + εeei ) − ϕ(xx)

[eei ]= lim
∂xx ε→0 ε
ϕi1 ···iq (x1 + εδ1i , · · · , x3 + εδ3i ) − ϕi1 ···iq (x1 , · · · , x3 )
 
= lim e i1 ⊗ · · · ⊗ e iq .
ε→0 ε

It may be clear that the limit in the last equality above is just the partial derivative of the standard component
ϕi1 ···iq (x1 , x2 , x3 ) in the ith Cartesian coordinate xi . Thus, relative to a common Cartesian coordinate system,
we obtain the simple expression
∂ϕ ∂ϕi1 ···iq
= e i1 ⊗ · · · ⊗ e iq ⊗ e i (4.24)
∂xx ∂xi
for the gradient of qth-order tensor field as shown by (4.2) and (4.7). In particular, this expression produces

∂ξ ∂ξ
= ei ; (4.25)
∂xx ∂xi

∂vv ∂v j
= e j ⊗ ei ; (4.26)
∂xx ∂xi
T
∂T ∂T jk
= e j ⊗ ek ⊗ ei ; (4.27)
∂xx ∂xi
for scalar, vector, 2nd-order tensor field as shown by (4.1) and (4.4)–(4.6), respectively.
Moreover, from the last three expressions and the definitions (4.13)–(4.16) we derive

∂2 ξ ∂2 ξ ∂2 ξ ∂2 ξ
∇2 ξ = = 2+ 2+ 2; (4.28)
∂xi ∂xi ∂x1 ∂x2 ∂x3
4.5. DIFFERENTIATION IN CURVILINEAR COORDINATES 69

∂vi ∂v1 ∂v2 ∂v3

divvv = = + + ; (4.29)
∂xi ∂x1 ∂x2 ∂x3
     
∂vi ∂v2 ∂v3 ∂v3 ∂v1 ∂v1 ∂v2
curlvv = εki j ek = − e1 + − e2 + − e3 ; (4.30)
∂x j ∂x3 ∂x2 ∂x1 ∂x3 ∂x2 ∂x1
∂2 vi
 2
∂ vi ∂2 vi ∂2 vi

2
∇ v= ei = + 2 + 2 ei ; (4.31)
∂x j ∂x j ∂x12 ∂x2 ∂x3
 
∂T ji ∂T1i ∂T2i ∂T3i
T=
divT ei = + + ei . (4.32)
∂x j ∂x1 ∂x2 ∂x3
In literature, a particular symbol ∇ is often introduced to represent a formal differential operator,
namely,
∂
∇ ≡ ei .
∂xi
Then we may reformulate the gradient and the curly, divergent, Laplacian derivatives as follows:

∂ξ
= ∇ξ, ∇2 ξ = (∇ · ∇)ξ,
∂xx

∂vv
= v ⊗ ∇, divvv = v · ∇, curlvv = v × ∇, ∇2 v = (∇ · ∇)vv,
∂xx

T
∂T
= T ⊗ ∇, T = ∇·T,
divT
∂xx
and the like. In the above, we formally regard the ∇ as a vector and each partial differential symbol ∂/∂xi
as its standard component. With this understanding and the relations

∂ ∂ξ ∂2 ∂2 ξ ∂ ∂v j ∂2 ∂2 v j ∂ ∂T jk
ξ= , 2
ξ= 2, vj = , 2
vj = 2 , T jk = ,
∂xi ∂xi ∂xi ∂xi ∂xi ∂xi ∂xi ∂xi ∂xi ∂xi

in mind, it may be clear that the foregoing expressions in terms of a single formal vector ∇ may facilitate
the memory of the related differential operations. However, without additional treatment, this advantage
will no longer be maintained in the case of using curvilinear coordinate systems.

4.5 Differentiation in Curvilinear Coordinates

As mentioned before, we may freely choose a vector basis at each point x in a spatial domain occupied
by a material body. The choice of a common standard basis leads to simple and concise expressions for
gradients and curly, divergent, Laplacian derivatives, as shown in the last section. However, in some cases
we need to take other possibilities into consideration, as indicated below.

4.5.1 Motivation
Various kinds of structural elements are designed and used for different purposes in engineering. Some of
them are shaped to assume regular and smooth curved boundary surfaces. For instance, that is the case
for shell-like structures of varied shapes and, in particular, for solid and hollow cylindrical and spherical
bodies, etc. Regular and smooth curved boundary shapes and surfaces of such bodies and structures leave
three problems outstanding. First, it might be awkward and inconvenient to use a Cartesian coordinate
system for treatment of regular, smooth curved surfaces and relevant integrations, but treatment in terms
of a “curvilinear coordinate system” defined directly through certain specific families of regular surfaces
would become quite simple and straightforward. Second, the boundary conditions in stress and strain
analyses would be most naturally and suitably formulated by means of the just-mentioned “curvilinear
70 CHAPTER 4. SCALAR, VECTOR AND TENSOR FIELDS

coordinate system”, whereas otherwise a formulation in terms of a Cartesian coordinate system would be
rather complicated and unnatural. Third, stress and strain distributions on the foregoing specific families of
regular surfaces may manifest themselves in a most simple and symmetric manner. Accordingly, it might
be most pertinent, efficient and convenient to treat the components of stress and strain relative to a suitable
“curvilinear coordinate system”. On the contrary, stress and strain components relative to any Cartesian
coordinate system would lose their features of simplicity and symmetry relative to a suitable “curvilinear
coordinate system”. In this case, it might not be so pertinent, efficient and convenient to deal with stress
and strain components relative to any Cartesian coordinate system.
The above problems and features may be clearly seen from a cylindrical body subjected to axially
symmetric loadings (see Fig. 5.1 given later). In this case, the aforementioned specific families of regular
surfaces are the cylindrical surfaces, the planes normal to or through the symmetric axis. The stress distri-
butions on these surfaces are axially symmetric on the just-mentioned three families of surfaces, whereas
that is not the case relative to a Cartesian coordinate system. In view of this, it may be natural and better
to treat the stress components on the foregoing surfaces and formulate them as functions of the so-called
cylindrical polar coordinates (r, θ, z), instead of the usual Cartesian stress components relative to any Carte-
sian coordinate system. The governing equations for the former and the related boundary conditions will
assume simple, natural forms, which would be most suitable for analyzing and solving the unknown stress
and strain in a cylindrical body.
From the above it may be clear that, for material bodies with regular curved boundary surfaces, such
as cylindrical and spherical bodies etc., we should consider three suitable specific families of surfaces
and introduce “curvilinear coordinate systems” generated by these families of surfaces, instead of the
conventional Cartesian coordinate systems. The main notion and idea will be detailed below.

4.5.2 Curvilinear coordinate systems

Note that one of the main objectives of introducing a coordinate system is to enable us to represent each
spatial point by three ordered numbers called coordinates. Here, how to explain and understand these three
ordered numbers might be essential. There are many ways attaining this objective. Cartesian coordinate
systems are among the simplest in geometrical features.
Let us first examine certain geometrical features of Cartesian coordinate systems. We choose three
mutually orthogonal families of planes. Each of these three families is quite simple: it is formed by all
the planes that are parallel to a given plane and spreads through the whole space, as indicated in Fig. 4.1
below. Besides, any two planes pertaining to any two of these three families are orthogonal to each other.
For each of these three families, it is a simple matter to specify any given plane pertaining to it. Yes, we
choose and fix a plane and call it a base plane. Then, relative to the latter, it suffices to simply associate
each plane with a single number. Each such number has a clear geometrical meaning, i.e., it is exactly the
distance between each plane and the chosen base plane parallel to it, as indicated in Fig.4.1. In particular,
the chosen base plane is associated with number 0. In this way, any three ordered numbers (x1 , x2 , x3 )
specify three mutually orthogonal planes pertaining to three ordered families of planes, separately, and
these three planes specify a spatial point x , i.e., the point intersected by them. Here the geometrical essence
is that three planes generally meet at one and only one point and hence the latter may be specified by the
former. Then we come to an important conclusion that three ordered numbers (x1 , x2 , x3 ) introduced to
specify three planes can in a natural and simple manner specify and represent a spatial point x . The former
are just the Cartesian coordinates of the latter. All the triplets (x1 , x2 , x3 ) specify and represent all the points
in the whole space. In particular, (0, 0, 0) gives the origin o, which is just the point intersected by the three
chosen base planes.
The above idea is usually embodied in and summarized by the notion of Cartesian coordinate system
(o; x1 , x2 , x3 ). In the above, we have explained the geometrical background and features of Cartesian coor-
dinates. In addition, we point out that a standard basis of vector spaces may be chosen at each spatial point
by assigning a unit vector in the direction of each of the three straight lines intersected by the three chosen
mutually orthogonal planes at the point at issue. Since such three straight lines at all spatial points are
parallel to one another, we obtain a common standard basis of vector spaces, say, e i , for all spatial points.
With the above explanations and the foregoing motivation, we are ready to extend the idea and notion
embodied in Cartesian coordinate systems and Cartesian coordinates to a general case, as will be done
4.5. DIFFERENTIATION IN CURVILINEAR COORDINATES 71

below.
From the foregoing explanation, a basic idea of using the Cartesian coordinates to determine and repre-
sent the relative positions of spatial points is to introduce three suitable, simple ordered families of planes
and then determine the relative spatial positions of these families of planes by using three ordered numbers.
Accordingly, each spatial point is specified by the intersecting point of three intersecting planes pertaining
to the chosen three families of planes and hence by the three order numbers specifying the latter. Now an
idea for extension is clear: in the foregoing account we merely replace each simple family of planes with a
simple family of surface, as shown in Fig. 4.1. This is elaborated below.

any plane xi
any surface γi

base plane x i=0 base surface γ i =0

Figure 4.1: From a simple family of planes to a simple family of surfaces

(i) We replace three chosen families of planes in constructing a Cartesian coordinate system with three
chosen simple families of surfaces, (S1 , S2 , S3 ). Here by a simple family of surfaces, Si , we mean
that all the surfaces pertaining to Si are the same in geometrical feature, spread the whole space in
a specified, simple manner, and satisfy certain analytical properties. These are specified by (4.33)–
(4.36) given slightly later;

(ii) we choose and fix a surface Si0 in each family Si , called a base surface. Then, with a specified
simple manner for the spatial distribution of each of the three chosen family of surfaces, as will be
given by (4.33), the spatial position of each surface pertaining to each family Si may be determined
and represented by assigning to it a single number, say, γi , which need not have so straightforward a
geometrical meaning as a Cartesian coordinate, but should be related to certain geometric features for
the chosen family of surfaces and specified by the form of the function defining the family of surfaces,
as will be explained soon, e.g., the radius of a cylindrical or spherical surface, the intersecting angle
of two planes, etc. Accordingly, the spatial positions of any three surfaces pertaining to the chosen
three families of surfaces, separately, may be determined and represented by a triplet of ordered
numbers, (γ1 , γ2 , γ3 ), and the latter, at the same time, also determines and represents the position
of point x . Note that each spatial point is the intersecting point of three intersecting surfaces at
this point. Algebraically, this means that from system (5.34) we can solve one and only one triplet
(x1 , x2 , x3 ), i.e., the Cartesian coordinates. We refer to such a triplet of ordered numbers, (γ1 , γ2 , γ3 ),
as the curvilinear coordinates of the point x , and the (o; γ1 , γ2 , γ3 ) with an origin o and the curvilinear
coordinates γi as a curvilinear coordinate system. Note here that we use γi to denote curvilinear
coordinates, in order to avoid possible confusion with Cartesian coordinates;

(iii) There are three curves intersected by the triplet of intersecting surfaces at each point x . Hence, in a
certain manner we may choose, as a basis of vector spaces, three vectors g i (xx) in the directions of
the three tangents of these three intersecting curves at point x . Such a basis g i (xx) is called a local
basis at point x . Note that generally the local basis g i (xx) will changes from point to point. As a
consequence, a common basis is no longer generated, except for the particular case of three families
of parallel planes.
72 CHAPTER 4. SCALAR, VECTOR AND TENSOR FIELDS

Now we present an exact mathematical formulation of curvilinear coordinates. We choose three smooth
functions of the three Cartesian coordinate variables x i as follows:

γ1 (x1 , x2 , x3 ) , γ2 (x1 , x2 , x3 ) , γ3 (x1 , x2 , x3 ) .

For any given number γi , each such function γi (x1 , x2 , x3 ) determines a surface in 3-dimensional space
through the following equation in the three Cartesian coordinate variables xi :

γi (x1 , x2 , x3 ) = γi . (4.33)

For all numbers γi , the above equation for each i generates a family of surfaces. Then we obtain three
families of surfaces through equation (4.33) for i = 1, 2, 3. Since they are generated by the same function
γi (x1 , x2 , x3 ) through equation (4.33), all the surfaces in each such family are the same in geometric feature,
and, hence, the only difference in geometrcal feature between any two such surfaces are their relative spatial
positions, possibly except for their sizes.
For any given triplet of numbers, (γ1 , γ2 , γ3 ), we have
 1
 γ = γ1 (x1 , x2 , x3 ) ,

γ2 = γ2 (x1 , x2 , x3 ) , (4.34)

 3
γ = γ3 (x1 , x2 , x3 ) .

We assume that from the above three equations we can work out a unique solution (x1 , x2 , x3 ) for any given
(γ1 , γ2 , γ3 ). Geometrically, this means that any three surfaces given by (4.33) will intersect at one and only
one spatial point and thus determines this point. Namely, any given triplet (γ1 , γ2 , γ3 ) will determine a
unique triplet (x1 , x2 , x3 ) through the above three equations. Specifically, we express this fact by

x = γ̄1 (γ1 , γ2 , γ3 ) ,

 1

x2 = γ̄2 (γ1 , γ2 , γ3 ) , (4.35)

x3 = γ̄3 (γ1 , γ2 , γ3 ) .


Note that the two pairs of triplet of three functions in (4.34) and (4.35) can be derived from each other.
In fact, they are the inverses of each other. This fact is called a curvilinear coordinate transformation and
may be signified as follows:

γi (x1 , x2 , x3 ) , i = 1, 2, 3 ⇐⇒ γ̄i (γ1 , γ2 , γ3 ) , i = 1, 2, 3.

In what follows we further assume that each function in the above pairs is continuously differentiable. In
addition, without loss of generality we may assume that the solution of (4.34) is xi = 0 for γi = 0, i.e.,

γ̄i (0, 0, 0) = 0 , i = 1, 2, 3 . (4.36)

Each number γi determines a surface through equation (4.33) and, of course, the spatial position of
this surface. A triplet (γ1 , γ2 , γ3 ) determines a point (x1 , x2 , x3 ) and the spatial positions of the triplet
of intersecting surfaces at this point. Hence, they are just the curvilinear coordinates mentioned before.
Besides, (o; γ1 , γ2 , γ3 ) is a curvilinear coordinate system with the origin o coincident with the origin of the
Cartesian coordinate system at issue.
It may be essential to bear the following fact in mind: The curvilinear coordinate γi is derived from
the Cartesian coordinates x1 , x2 , x3 through the chosen function γi (x1 , x2 , x3 ), and the geometric meaning
the curvilinear coordinate γi should have is fully specified by the chosen functional form of γi (x1 , x2 , x3 ).
For example, the three functional forms given later by (4.65) in §4.5.4 specify the following geometrical
meanings: γ1 is the radius of the cylindrical surface S1 ; γ2 is the angle between the plane S2 and the
base plane S20 ; and γ3 is just a Cartesian coordinate, as indicated in Fig. 4.2 in §4.5.4, whereas the three
functional forms given later by (4.65) in §4.5.5 specify the following different geometrical meanings: γ1 is
the radius of the spherical surface S1 ; γ2 is the angle between the semicircle S2 and the base semicircle S20
meeting at the common axis S30 ; and γ3 is the angle of the cone S3 with S30 its symmetry axis; as indicated
4.5. DIFFERENTIATION IN CURVILINEAR COORDINATES 73

in Fig. 4.3 in §4.5.5. Note here that the common axis S30 itself is a base surface for the family of conic
surfaces, S3 .
The three functions γi (x1 , x2 , x3 ) generating the three simple families of surfaces determine the local
basis g i (xx) at point x . Indeed, the gradient of the function γi (x1 , x2 , x3 ), i.e., ∂γi /∂xx, is just the outer normal
to the surface (4.33) at point x . The tangent of the intersecting curve of any two surfaces at x is normal to
their outward normals. Observing the equalities

∂γi ∂xx ∂γi ∂xk ∂γi

· j = j
= j = δi j ,
∂xx ∂γ ∂xk ∂γ ∂γ

we deduce that the local basis gi (xx) may be given by

∂xx
g i (xx) = , (4.37)
∂γi

and its reciprocal basis gi (xx) by

∂γi
g i (xx) = . (4.38)
∂xx
Namely, the normals g i (xx) of three intersecting surfaces at point x form a reciprocal basis of the local basis
g i (xx). The local basis g i (xx) given above is known as a local natural basis generated by the three functions
γi (x1 , x2 , x3 ) at point x .
The widely used curvilinear coordinate systems are orthogonal curvilinear coordinate systems, which
supply orthogonal local bases at all points x . In this case, the three functions γi (x1 , x2 , x3 ) are said to be
mutually orthogonal in the sense
∂γi ∂γ j
· = 0 , i 6= j , (4.39)
∂xx ∂xx
and we refer to the γi as orthogonal curvilinear coordinates. Hence, the normalization of the local natural
basis g i (xx) and its reciprocal basis g i (xx) at each point x will produce a common standard basis, denoted
ẽei = ẽei (xx) and called the local standard basis generated by the three functions γi (x1 , x2 , x3 ) at point x .
Namely, the orthogonalization conditions (4.39) result in

∂xx
 g (xx) ∂γi
ẽei = ẽei (xx) = i


 = (no summation) ,
|ggi x )| ∂xx



∂γi


∂γi (4.40)
g i (xx) ∂xx

 ei = ẽei (xx) = i
ẽ = (no summation) ,


 |gg (xx)| ∂γi


 ∂xx
ẽei (xx) · ẽe j (xx) = δi j .


In summary, we may say that the essence of curvilinear coordinate system is to specify three contin-
uously differentiable functions γi (x1 , x2 , x3 ) which render the system (4.34) to have a unique solution for
any given triplet (γ1 , γ2 , γ3 ) and meet (4.36). Of them, the condition (4.36) is not necessary but only a
convenient requirement. We mention in passing that a Cartesian coordinate system is the trivial case when
the three functions γi (x1 , x2 , x3 ) are chosen as the simplest, i.e.

γi (x1 , x2 , x3 ) = xi .

To conclude this subsection, we may ask a seemingly naive question: why exactly three families of
surfaces are needed, instead of less or more than three? The answer is perhaps as follows. Three suitably
chosen surfaces meet at one and only one spatial point and hence specify this point, and more than three
surfaces are thus unnecessary. On the other hand, two surces usually meet at a curve and thus determine
not only one but so many spatial points.
74 CHAPTER 4. SCALAR, VECTOR AND TENSOR FIELDS

4.5.3 Orthogonal curvilinear coordinates

The main objective of introducing a curvilinear coordinate system is to study the components of vector
and tensor quantities relative to a local basis g i (xx) and its reciprocal basis g i (xx) and to regard each such
components as a function of three curvilinear coordinates γi . In so doing, all the equations governing basic
scalar, vector and tensor fields will be formulated in terms of these components relative to a curvilinear
coordinate system. Although these governing equations assume the simplest forms relative to a Cartesian
coordinate system, their forms relative to a curvilinear coordinate system would be suitable for structures
and bodies with certain regular curved boundary surfaces, e.g., cylindrical and spherical bodies, etc.
If a non-orthogonal curvilinear coordinate system is chosen, i.e., the orthogonalization conditions (5.39)
are not met, we need to deal with a non-orthogonal local basis g i (xx) and its reciprocal basis g i (xx) at each
point x . The consequence is that we have to deal with the complicated situation of various kinds of co-,
contra- and mixed-variant components relative to these two kinds of non-orthogonal bases, each of which
usually has no direct and simple geometrical and physical meanings. In particular, relative to a non-
orthogonal curvilinear coordinate system, the treatment for the gradients and various associated derivatives
of scalar, vector and tensor fields are usually rather tedious and cumbersome. However, an orthogonal
curvilinear coordinate system enables us to use a local standard basis ẽei (xx) at each point x as given by
(4.40), and the forgoing complexity thus disappears. In the subsequent treatment, we shall not pursue the
general case covering both orthogonal and non-orthogonal curvilinear coordinate systems, and we shall
take only orthogonal curvilinear coordinate systems into consideration. Results for general curvilinear
coordinate systems may be derived by following the idea and procedures that will be used below. They
may be found in many related monographs, e.g., Klingbeil (1966), Boer (1982), Ogden (1984), Betten
(1987), et al.
For an orthogonal curvilinear coordinate system, we choose three functions γi (x1 , x2 , x3 ) such that the
orthogonalization conditions (4.39) are met. Then, we can determine a local standard basis ẽei (xx) at each
point x by using (4.40). In this case, the difference between sub- and superscripts are no longer substantial,
and hence we can always use subscripts, as already done for standard bases. In accordance with this
consideration, the orthogonal curvilinear coordinates γi will be replaced by γi .
In subsequent account, a scalar with a superimposed tilde “∼”, e.g., ξ, ˜ is used to mean a scalar field
formulated as a function of the curvilinear coordinates γi . Moreover, each vector or tensor component
relative to a local standard basis is also signified by a superimposed tilde “∼” and regarded as a function of
the curvilinear coordinates γi , such as ṽi , T̃i j , etc. Namely,

ξ˜ = ξ(γ
˜ 1 , γ2 , γ2 ) ; ṽi = ṽi (γ1 , γ2 , γ2 ) ; T̃i j = T̃i j (γ1 , γ2 , γ2 ) .

Following the foregoing objective, we first express a qth-order tensor field ϕ = ϕ(xx) relative to the local
standard basis ẽei = ẽei (xx) at each point x . Hence, we have

ϕ = ϕ(xx) = ϕ̃i1 ···iq ẽei1 ⊗ · · · ⊗ ẽeiq , (4.41)

ϕ = ϕ(xx) = ϕ̃i1 ···iq ẽei1 ⊗ · · · ⊗ ẽeiq , (4.42)

where each local standard components ϕ̃i1 ···iq is a functions of the curvilinear coordinates γi , i.e.,

ϕ̃i1 ···iq = ϕ̃i1 ···iq (γ1 , γ2 , γ3 ) . (4.43)

Now we aim to derive explicit expressions for the gradient and the curly, divergent and Laplacian
derivatives of the tensor field ϕ = ϕ(xx). Toward this goal, the main idea is to use the simple expression
(4.24) relative to a common Cartesian coordinate system, and the main procedure is to establish the re-
lationship between the local standard components given in (4.42) and the Cartesian components in (4.7).
Utilizing the transformation rule (3.16) we have

ϕi1 ···iq = (eei1 · ẽek1 ) · · · (eeiq · ẽekq )ϕ̃k1 ···kq (4.44)

at each point x . According to (4.42) and (4.35), each local standard component ϕ̃k1 ·kq is first a function
of the curvilinear coordinates γ1 , γ2 , γ3 and then each γi is in turn a function of the Cartesian coordinates
4.5. DIFFERENTIATION IN CURVILINEAR COORDINATES 75

x1 , x2 , x3 . This fact just means a composition of two functions and the consequence is that each local
standard component ϕ̃i1 ·iq may be regarded as a function of the Cartesian coordinates x1 , x2 , x3 . With this
understanding and Leibniz’s chain rule for composite functions, by using (4.43) we infer

∂ϕi1 ···iq ∂

= (eei1 · ẽek1 ) · · · (eeiq · ẽekq )ϕ̃k1 ···kq
∂xi ∂xi

∂ϕ̃k1 ···kq
=(eei1 · ẽek1 ) · · · e iq · ẽekq +
∂xi
∂ẽekq
   
∂ẽek1


e i1 · · · · e iq · ẽekq + · · · + e iq · ẽekq · · · e iq · ϕ̃k1 ···kq (4.45)
∂xi ∂xi

∂ϕ̃k1 ···kq ∂γ j
=(eei1 · ẽek1 ) · · · eiq · ẽekq +
∂γ j ∂xi
∂ẽekq
   
∂ẽek1

∂γ j
e i1 · · · · e iq · ẽekq + · · · + e iq · ẽekq · · · e iq · ϕ̃k ···k .
∂γ j ∂γ j ∂xi 1 q

From the above we deduce that if the partial derivatives of the local standard basis ẽer with respect
to each curvilinear coordinate γs are known, then the gradient of the qth-order tensor field ϕ(xx) may be
determined. We set
∂ẽea
= Γ̃abc ẽec , (4.46)
∂γb
where the coefficients
Γ̃abc = Γ̃abc (γ1 , γ2 , γ3 ) (4.47)
are known as the Christoffel symbols associated with the orthogonal curvilinear coordinate system (o; γi )
and calculated by
∂ẽea
Γ̃abc = · ẽec . (4.48)
∂γb
On the other hand, (4.25) yields
∂γ j ∂γ j
= ei . (4.49)
∂xx ∂xi
This and (4.40)1 and the relation
−1
∂γ j ∂xx
=
∂xx ∂γ j
result in
−1
∂γ j ∂γ j ∂xx
ei = ẽe j = ẽe j (no summation for j) . (4.50)
∂xi ∂xx ∂γ j
Using this and the expression
ẽek = (eea · ẽek )eea ,
from (4.44)3 and (4.47) we derive an expression for the gradient of qth.order tensor field ϕ = ϕ(xx) as
follows: !
∂ϕ ∂xx −1 ∂ϕ̃k1 ···kq q
= + ∑ Γ̃kα jt ϕ̃k1 ···t···kq ẽek1 ⊗ · · · ⊗ ẽekq ⊗ ẽe j . (4.51)
∂xx ∂γ j ∂γ j α=1

In the above, summation convention for each twice repeated subscript kα is taken granted, and here
summation for the triply repeated subscript j is also meant. The latter is an exception.
In particular, for q =0, 1, 2, i.e., for scalar, vector, 2nd-order tensor fields as shown by (4.1), we have
−1
∂ξ ∂xx ∂ξ˜
= ẽe j , (4.52)
∂xx ∂γ j ∂γ j
−1  
∂vv ∂xx ∂ṽi
= + Γ̃i jk ṽk ẽei ⊗ ẽe j , (4.53)
∂xx ∂γ j ∂γ j
76 CHAPTER 4. SCALAR, VECTOR AND TENSOR FIELDS

−1  
T
∂T ∂xx ∂T̃rs
= + Γ̃r jt T̃ts + Γ̃s jt T̃rt ẽer ⊗ ẽes ⊗ ẽe j , (4.54)
∂xx ∂γ j ∂γ j
for the gradients, and
! !
−1
2 ∂xx ∂ ∂γ j ∂ξ˜ ∂γk ∂ξ˜
∇ ξ= + Γ̃ j jk , (4.55)
∂γ j ∂γ j ∂xx ∂γ j ∂xx ∂γk

−1  
∂xx ∂ṽ j
divvv = + Γ̃ j jk ṽk , (4.56)
∂γ j ∂γ j
−1  
∂xx ∂ṽi
curlvv = + Γ̃i jk ṽk εi jk ẽek , (4.57)
∂γ j ∂γ j
!
−1
∂xx ∂2 ṽr ∂ṽt ∂ṽr
∇2 v = + Γ̃r jt + Γ̃ j jt ẽer , (4.58)
∂γ j ∂γ2j ∂γ j ∂γt

−1  
∂xx ∂T̃r j
T=
divT + Γ̃r jt T̃t j + Γ̃ j jt T̃rt ẽer , (4.59)
∂γ j ∂γ j
for the curly and divergent derivatives of scalar field ξ = ξ(xx), vector field v = v (xx), and tensor field
T = T (xx). In the above, summation is meant for each repeated (twice or triply) subscript.
Now it may become clear, expressions for the gradient of qth-order tensor field ϕ = ϕ(xx) and other
associated differential operations are available, whenever we can work out the gradients ∂xx/∂γ j and the
Christoffel symbols Γ̃abc determined by (4.48). Now we explain how to calculate these quantities. The
main procedures are as follows:

(i) For a chosen orthogonal curvilinear coordinate system (o; γ1 , γ2 , γ3 ), we establish the curvilinear
coordinate transformation relationship (4.34) and (4.35), which will be the starting point of the sub-
sequent analyses and computations;

(ii) Utilizing the equality

∂xx ∂xi
= ei , (4.60)
∂γ j ∂γ j
we derive
 2  2  2 ! 21
∂xx ∂x1 ∂x2 ∂x3
= + + . (4.61)
∂γ j ∂γ j ∂γ j ∂γ j

At this step, we work out the nine partial derivatives ∂xi /∂γ j ; then we work out the local standard
basis ẽe j by (cf., (4.60)1 and (4.49))
−1
∂xx ∂xi
ẽe j = ei (no summation for j) ; (4.62)
∂γ j ∂γ j

(iii) Using (4.40)1 and (4.48), we obtain an explicit expression for the Christoffel symbols as follows:
!
∂ ∂xx −1 ∂xx ∂xi ∂xi −1
Γ̃abc = . (4.63)
∂γb ∂γa ∂γa ∂γc ∂γc

In the above, no summation is meant for either of the twice repeated subscripts a and c. The Christof-
fel symbols may be calculated either by using the above formula or by differentiating the expression
(4.62) for the local standard basis.
4.5. DIFFERENTIATION IN CURVILINEAR COORDINATES 77

Generally, the Christoffel symbols consist of 27 ordered numbers and we need to compute them. Their
computations are a bit tedious. For an orthogonal curvilinear coordinate system, however, the conditions
(4.39) and hence (4.40)3 hold. From them we derive

Γ̃abc = −Γ̃cba .

These lead to considerable simplifications as shown below:


Γ̃ = Γ̃222 = Γ̃333 = Γ̃121 = Γ̃131 = Γ̃212 = Γ̃232 = Γ̃313 = Γ̃323 = 0 ,
 111

Γ̃112 = −Γ̃211 , Γ̃122 = −Γ̃221 , Γ̃132 = −Γ̃231 , Γ̃113 = −Γ̃311 , (4.64)

Γ̃123 = −Γ̃321 , Γ̃133 = −Γ̃331 , Γ̃213 = −Γ̃312 , Γ̃223 = −Γ̃322 , Γ̃233 = −Γ̃332 .


Thus, it suffices to calculate only nine of the 18 non-vanishing Christoffel symbols, e.g.,

Γ̃112 , Γ̃122 , Γ̃132 , Γ̃113 , Γ̃123 , Γ̃133 , Γ̃213 , Γ̃223 , Γ̃233 .

In the next two subsections, the above procedures will be used to treat two widely used orthogonal
curvilinear coordinate systems.

4.5.4 Cylindrical polar coordinates

In the subsection, we treat a curvilinear coordinate system specified by the three functions

γ = γ1 (x1 , x2 , x3 ) = (x12 + x22 )1/2 ,
 1


 
γ = γ (x , x , x ) = tg −1 x2 , (4.65)
 2 2 1 2 3 x1

γ3 = γ3 (x1 , x2 , x3 ) = x3 ,


which is known as a cylindrical polar coordinate system. The γi given above are called cylindrical polar
coordinates. As indicated in Fig. 4.2, the geometrical meanings of these coordinates are as follows: the γ1
is the radius of a cylindrical surface, the γ2 is the angle between the base plane x2 = 0 and a plane through
the e 3 -axis, and the γ3 is just the Cartesian coordinate x3 .

~
e3 = ez
S3
γ1 ~
e2 = eθ
γ2
~
e1 = er
S1
z γ3

S2

S 30
S 20
S 10

Figure 4.2: Cylindrical polar coordinate system

The curvilinear coordinate transformation of (4.65) can be easily established and given below:

x = γ1 cos γ2 ,
 1

x2 = γ1 sin γ2 , (4.66)


x3 = γ3 .
78 CHAPTER 4. SCALAR, VECTOR AND TENSOR FIELDS

The nine partial derivatives ∂xi /∂γ j are given by


∂x1 ∂x1 ∂x1

 = cos γ2 , = −γ1 sin γ2 , = 0,



 ∂x∂γ1 ∂γ2 ∂γ3
2 ∂x2 ∂x2
= sin γ2 , = γ1 cos γ2 , = 0, (4.67)

 ∂γ1 ∂γ2 ∂γ3
 ∂x3
 ∂x3 ∂x3

 = 0, = 0, = 1.
∂γ1 ∂γ2 ∂γ3
Hence, (4.61) yields
∂xx ∂xx ∂xx
= = 1, = γ1 . (4.68)
∂γ1 ∂γ3 ∂γ2
Then (4.62) produces the local standard basis

ẽe = e 1 cos γ2 + e 2 sin γ2 ,
 1

ẽe2 = −ee1 sin γ2 + e 2 cos γ2 , (4.69)


ẽe3 = e 3 .
Moreover, either using (4.63) and (4.67)-(4.68) or directly differentiating the local standard basis given
by (4.69) we obtain the non-zero Christoffel symbols as follows:
Γ̃122 = −Γ̃221 = 1 . (4.70)
Thus, relative to a cylindrical polar coordinate system, the expressions (4.52)–(4.59) for the gradients
and for the curly, divergent and Laplacian derivatives reduce to (see, e.g., Chou and Pagano 1992)
∂ξ ∂ξ˜ 1 ∂ξ˜ ∂ξ˜
= ẽe1 + ẽe2 + ẽe3 , (4.71)
∂xx ∂γ1 γ1 ∂γ2 ∂γ3
 
∂vv ∂ṽ1 1 ∂ṽ2 ∂ṽ3
∂xx = ẽe1 ⊗ ẽe1 + + ṽ1 ẽe2 ⊗ ẽe2 + ẽe3 ⊗ ẽe3 +
∂γ1 γ1 ∂γ2 ∂γ3
∂ṽ2 1 ∂ṽ3 ∂ṽ1
ẽe1 ⊗ ẽe2 + ẽe2 ⊗ ẽe3 + ẽe3 ⊗ ẽe1 + (4.72)
∂γ1 γ1 ∂γ2
 ∂γ3
1 ∂ṽ1 ∂ṽ3 ∂ṽ2
− ṽ2 ẽe2 ⊗ ẽe1 + ẽe1 ⊗ ẽe3 + ẽe3 ⊗ ẽe2 ,
γ1 ∂γ2 ∂γ1 ∂γ3
∂2 ξ˜ 1 ∂ξ˜ 1 ∂2 ξ˜ ∂2 ξ˜
∇2 ξ = + + + , (4.73)
∂γ21 γ1 ∂γ1 γ21 ∂γ22 ∂γ23
∂ṽ1 1 ∂ṽ2 ∂ṽ3 ṽ1
divvv = + + + , (4.74)
∂γ1 γ1 ∂γ2 ∂x3 γ1
     
∂ṽ2 1 ∂ṽ3 ∂ṽ3 ∂ṽ1 1 ∂ṽ1 ∂ṽ2 ṽ2
curlvv = − ẽe1 + − ẽe2 + − − ẽe3 , (4.75)
∂x3 γ1 ∂γ2 ∂γ1 ∂x3 γ1 ∂γ2 ∂γ1 γ1
   
2 ∂ṽ2 ṽ1 2 ∂ṽ1 ṽ2
∇2 v = ∇2 ṽ1 − 2 − 2 ẽe1 + ∇2 ṽ2 + 2 − ẽe2 + ∇2 ṽ3 ẽe3 . (4.76)
γ1 ∂γ2 γ1 γ1 ∂γ2 γ21
 
∂T̃11 1 ∂T̃21 ∂T̃31 T̃11 − T̃22
T=
divT + + + ẽe1 +
 ∂γ1 γ1 ∂γ2 ∂γ3 γ1 
∂T̃12 1 ∂T̃22 ∂T̃32 T̃12 + T̃21
+ + + ẽe2 + (4.77)
 1 ∂γ γ 1 ∂γ2 ∂γ3 γ1
∂T̃13 1 ∂T̃23 ∂T̃33 T̃13
+ + + ẽe3 .
∂γ1 γ1 ∂γ2 ∂γ3 γ1
Usually, the cylindrical polar coordinates γi specified by (4.65) are designated by
γ1 = r , γ2 = θ , γ3 = z ,
and the subscripts 1, 2, 3 in the above may be replaced by r, θ, z, respectively.
4.5. DIFFERENTIATION IN CURVILINEAR COORDINATES 79

4.5.5 Spherical polar coordinates

In this subsection, we study a curvilinear coordinate system specified by


 γ1 = γ1 (x1 , x2 , x3 ) = (x12 + x22 + x32 )1/2 ,

  
−1 x2


γ2 = γ2 (x1 , x2 , x3 ) = tg ,

x1 ! (4.78)
2 + x2 )1/2


 −1 (x1 2
 γ3 = γ3 (x1 , x2 , x3 ) = tg ,


x3


which is known as a spherical polar coordinate system. The γi given above are called spherical polar
coordinates. As indicated in Fig. 4.3, the geometrical meanings of these coordinates are as follows: the γ1
is the radius of a spherical surface, the γ2 is the angle between the base plane x2 = 0 and a plane through
the e 3 -axis, and the γ3 is the angle between x and the e 3 -axis.

~e = e
1 r

S3
~e = e γ 3
2 ϕ
θ γ 1 = r ~e3 = eθ

γ2 = ϕ

S2
S 20

S 30

Figure 4.3: Spherical polar coordinate system

The curvilinear coordinate transformation of (4.78) is given by


x = γ1 cos γ2 sin γ3 ,
 1

x2 = γ1 sin γ2 sin γ3 , (4.79)


x3 = γ1 cos γ3 .

The 9 partial derivatives ∂xi /∂γ j for the spherical polar coordinates are given by

∂x1 ∂x1 ∂x1

 = cos γ2 sin γ3 , = −γ1 sin γ2 sin γ3 , = γ1 cos γ2 cos γ3 ,
 ∂γ
 ∂x1

 ∂γ 2 ∂γ3
2 ∂x2 ∂x2
= sin γ2 sin γ3 , = γ1 cos γ2 sin γ3 , = γ1 sin γ2 cos γ3 , (4.80)

 ∂γ 1 ∂γ2 ∂γ3
 ∂x3 = cos γ3 , ∂x3 = 0 , ∂x3 = −γ1 sin γ3 .



∂γ1 ∂γ2 ∂γ3
Hence, (4.61) yields
∂xx ∂xx ∂xx
= 1; = γ1 sin γ3 , = γ1 . (4.81)
∂γ1 ∂γ2 ∂γ3
Then (4.62) produces the local standard basis

ẽe = e 1 cos γ2 sin γ3 + e 2 sin γ2 sin γ3 + e 3 cos γ3 ,
 1

ẽe2 = −ee1 sin γ2 + e 2 cos γ2 , (4.82)

ẽe3 = e 1 cos γ2 cos γ3 + e 2 sin γ2 cos γ3 − e 3 sin γ3 .

80 CHAPTER 4. SCALAR, VECTOR AND TENSOR FIELDS

Moreover, either using (4.63) and (4.80)-(4.81) or differentiating the local standard basis given by
(4.82), we obtain the non-zero Christoffel symbols as follows:

Γ̃122 = −Γ̃221 = sin γ3 , Γ̃133 = −Γ̃331 = 1 , Γ̃223 = −Γ̃322 = − cos γ3 . (4.83)

Thus, relative to a spherical coordinate system, the expressions (4.52)–(4.59) for the gradients and for
the curly, divergent and Laplacian derivatives reduce to (see, e.g., Chou and Pagano 1992)

∂ξ ∂ξ˜ 1 ∂ξ˜ 1 ∂ξ˜

= ẽe1 + ẽe2 + ẽe3 , (4.84)
∂xx ∂γ1 γ1 sin γ3 ∂γ2 γ1 ∂γ3
   
∂vv ∂ṽ1 1 ∂ṽ2 1 ∂ṽ3
∂xx = ẽe1 ⊗ ẽe1 + + ṽ1 sin γ3 + ṽ3 cos γ3 ẽe2 ⊗ ẽe2 + + ṽ1 ẽe3 ⊗ ẽe3 +
∂γ1 γ1 sin γ3  ∂γ2   γ1 ∂γ3
∂ṽ2 1 ∂ṽ3 1 ∂ṽ1
ẽe1 ⊗ ẽe2 + − ṽ2 cos γ3 ẽe2 ⊗ ẽe3 + − ṽ3 ẽe3 ⊗ ẽe1 + (4.85)
∂γ1  γ 1 sin γ 3 ∂γ
2 γ 1 ∂γ3
1 ∂ṽ1 ∂ṽ3 1 ∂ṽ2
− ṽ2 sin γ3 ẽe2 ⊗ ẽe1 + ẽe1 ⊗ ẽe3 + ẽe3 ⊗ ẽe2 ,
γ1 sin γ3 ∂γ2 ∂γ1 γ1 ∂γ3

∂2 ξ˜ 2 ∂ξ˜ 1 ∂2 ξ˜ 1 ∂2 ξ˜ ctgγ3 ∂ξ˜

∇2 ξ = + + + + 2 , (4.86)
∂γ21 γ1 ∂γ1 γ21 sin2 γ3 ∂γ22 γ23 ∂γ23 γ1 ∂γ3

∂ṽ1 1 ∂ṽ2 1 ∂ṽ3 2ṽ1 ṽ3 ctgγ3

divvv = + + + + , (4.87)
∂γ1 γ1 sin γ3 ∂γ2 γ1 ∂x3 γ1 γ1
 
1 ∂ṽ2 ∂ṽ3
curlvv = sin γ3 − + ṽ2 cos γ3 ẽe1 +
γ1 sin
 γ3 ∂x3  ∂γ2
1 ∂ṽ3 ∂ṽ1
γ1 − + ṽ3 ẽe2 + (4.88)
γ1  ∂γ1 ∂x3 
1 1 ∂ṽ1 ∂ṽ2
− − ṽ2 e 3 ,
γ1 sin γ3 ∂γ2 ∂γ1
 
2 ∂ṽ2 2 ∂ṽ3 2
∇2 v = ∇2 ṽ1 − 2 − 2 − 2 (ṽ1 + ṽ3 ctgγ3 ) ẽe1 +
γ1 sin γ3 ∂γ2 γ1 ∂γ3 γ1 !
2 2 cos γ3 ∂ṽ3 2 ∂ṽ1 ṽ2
∇ ṽ2 + 2 2 + − ẽe2 + (4.89)
γ1 sin γ3 ∂γ2 γ21 sin γ3 ∂γ2 γ2 sin2 γ3
!
2 cos γ3 ∂ṽ2 2 ∂ṽ1 ṽ3
∇2 ṽ3 − 2 2 + − ẽe3
γ1 sin γ3 ∂γ2 γ21 ∂γ3 γ2 sin2 γ3
 
∂T̃11 1 ∂T̃21 1 ∂T̃31 ctgγ3 2T̃11 − T̃22 − T̃33
T=
divT + + + T̃31 + ẽe1 +
 ∂γ1 γ1 sin γ3 ∂γ2 γ1 ∂γ3 γ1 γ1 
∂T̃12 1 ∂T̃22 1 ∂T̃32 2T̃12 + T̃21 ctgγ3

+ + + + T̃32 + T̃23 ẽe2 + (4.90)
 ∂γ1 γ1 sin γ3 ∂γ2 γ1 ∂γ3 γ1 γ1 
∂T̃13 1 ∂T̃23 1 ∂T̃33 2T̃13 + T̃31 ctgγ3

+ + + + T̃33 − T̃22 e 3 .
∂γ1 γ1 sin γ3 ∂γ2 γ1 ∂γ3 γ1 γ1
Usually, the spherical polar coordinates γi specified by (4.78) are designated by

γ1 = r , γ2 = φ , γ3 = θ ,

and the subscripts 1, 2, 3 in the above may be replaced by r, φ, θ, respectively.

Finally, we would like to mention in passing that, from (4.71) and (4.77) as well as (4.84) and (4.90), it
may be straightforward to derive the well-known expressions for the strain tensor and the forms of Cauchy’s
equations of equilibrium relative to cylindrical and spherical polar systems, respectively. These results are
presented in many monographs, e.g., in Bruhns (2003).
4.6. INTEGRATIONS OF SCALAR AND VECTOR FIELDS 81

4.6 Integrations of Scalar and Vector Fields

In mechanics and engineering and related fields, integrations of scalar fields and vector fields serve as basic
tools in formulating certain basic notions and in studying basic equations, such as the total entropy and the
total energy stored in the whole body, the total body force acting over the body, the total area force acting
on the whole boundary, and the work done by the area force in a course of displacement, etc.
In the following, we consider a spatial domain D , an oriented closed surface S , and a smooth curve s.
Let dV and dA and dl, respectively, represent the volume element at each point x in D , the area element at
each point x on A and the line element at each point x on s. Let ϕ = ϕ(xx) be a continuous qth-order tensor
field over D or on S or on s. The integration of ϕ over the domain D is a qth-order tensor quantity defined
by:  
Z Z
ϕ(xx)dV ≡ ϕi1 ···iq (x1 , x2 , x3 )dV e i1 ⊗ · · · ⊗ e iq ; (4.91)
D D
the integration of ϕ on the oriented surface S is a qth-order tensor quantity given by:
Z Z 
ϕ(xx)dA ≡ ϕi1 ···iq (x1 , x2 , x3 )dA e i1 ⊗ · · · ⊗ e iq ; (4.92)
S S
and the integration of ϕ along the curve s is a qth-order tensor quantity specified by:
Z Z 
ϕ(xx)dl ≡ ϕi1 ···iq (x1 , x2 , x3 )dl e i1 ⊗ · · · ⊗ e iq . (4.93)
s s

In the above definitions, a Cartesian coordinate system (o; x1 , x2 , x3 ) is chosen. Note that the respective
right-hand sides of (4.91)–(4.93) are just the conventional integrations over D or on S or on s.

4.6.1 Gauss’s theorem

Gauss’s theorem establishes relationship between the volume integration of the gradient of a continuously
differentiable tensor field ϕ = ϕ(xx) over the spatial domain D and the area integration of this field on the
oriented closed bounding surface S of D . A general formulation of this theorem is below:
I I
∂ϕ
dV = ϕ ⊗ n dA , (4.94)
x
D ∂x S
where n is the outward unit normal to the oriented bounding surface S . In particular, from this general
form and the definition (4.16) we derive the usual form of Gauss’s theorem as follows:
I I
T dV =
divT T T [nn]dA (4.95)
D S
for a 2nd-order tensor field T = T (xx).
Relative to a Cartesian coordinate system, Gauss’s theorem (4.95) is of the form
I I
∂Ti j
dV = Ti j ni dA . (4.96)
D ∂xi S
Gauss’s theorem plays a basic role in establishing various types of variational formulations in mechan-
ics and engineering and related fields. Let T be Cauchy stress tensor. Then, Gauss’s theorem simply says
that the integration of the stress distribution on the bounding surface of a material body may be given by
the integration of the divergent derivative of the stress field over this body.

4.6.2 Stokes’s theorem

The usual form of Stokes’s theorem relates the area integration of the curly derivative of a continuously
differentiable vector field v = v (xx) on an oriented but not closed surface S to the line integration of this
field along a closed bounding curve, s, of the surface S . We have
Z I
(curlvv) · n dA = v · t dl . (4.97)
S s
82 CHAPTER 4. SCALAR, VECTOR AND TENSOR FIELDS

In the above, n is the outward unit normal to the oriented surface S , as indicated before. Besides, t is the
unit tangent vector along the closed bounding edge curve s, and dl is the linear element.
Relative to a Cartesian coordinate system, Stokes’s theorem (4.97) is of the form
Z I
∂vi
εi jk nk dA = viti dl . (4.98)
S ∂x j s

Stokes’s theorem may be extended to arrive at a general formulation, which is not discussed here.
Chapter 5

Tensor Functions

One of the central topics in continuum mechanics and theories of materials and in other related fields is to
establish rational, consistent mathematic models characterizing varied complicated behaviour of materials.
These models, known as constitutive relations of materials, tell us how certain basic physical quantities
(scalars, vectors, tensors) rely on certain other basic physical quantities (scalars, vectors and tensors).
Mathematic forms of these models are called constitutive tensor functions, each of which specifies how a
basic scalar or a basic vector or a basic tensor quantity depends on or is determined by certain basic scalar,
vector and tensor quantities. In Chapters 2-3 we are already concerned with particular forms of tensor
functions, namely, linear tensor functions, each of which specifies linear dependence of a tensor quantity
on another tensor quantity, e.g., (2.7)-(2.8), (3.37)–(3.38), and (3.44)–(3.46), etc. In this chapter, we shall
discuss some aspects of tensor functions from a general viewpoint.

5.1 Motivation: constitutive functions

Materials in nature are rich and varied. They may manifest themselves in extremely different manners in
response to deformation and other actions. Some are ready to take any possible shape (e.g., airs), some
flow (e.g., water); some are soft and deform easily (e.g., sponges, rubbers, ice creams), some are very
hard and deform very slightly (e.g., bones, metals, stones, rocks, woods); some recover its original shapes
with the applied loading removed, some maintain permanent deformed shapes without loading applied;
some are brittle, some are ductile; some respond to different loading rates in the same manner, some are
sensitive to different loading rates; some possess capability of shape memory, some do not; some exhibit
piezoelectricity effects, some display photo-elasticity phenomena; · · ·; and so on.
It may be said that each material is endowed with its own distinct “individuality” (material behaviour
and property) in response to deformation and other actions. How can we characterize and represent these
“individualities” for different materials?
In continuum physics and theories of materials, the macroscopic physical state and internal state at each
point in material bodies are represented by certain basic physical quantities, which are usually scalars, vec-
tors and 2nd-order tensors, etc.1 . In order to characterize “individualities” of various materials in response
to deformation and other actions etc., the basic idea is to study, examine and establish corresponding re-
lations among these well-chosen and well-defined basic physical quantities. For example, we may ask the
following questions: how does the stress rely on the strain in an elastic body? how is the yield condition of
a solid formulated in terms of stress? how is the stress dependent on the deformation rate in a fluid? how
does the heat flux vector changes with the temperature gradient? how does the applied stress (the applied
electric field) generates the electric field (the stress)? how is the strain increment determined by the stress
increment and the current state quantities in an inelastic body? and so on.
1 How to introduce these basic scalar, vector and tensor quantities in a consistent, rational, economic manner is a fundamental

issue in constitutive theories of various material behaviour and may be at deeper and more sophisticated level from both theoretical
and experimental viewpoints. Here, we assume that these basic quantities have been well defined and well chosen, and our task is to
study some aspects of functional relations among them.

83
84 CHAPTER 5. TENSOR FUNCTIONS

The above idea and related questions lead to constitutive relations and constitutive functions. Their
mathematical forms formulate functional relations between certain physical quantities, such as elastic
strain-energy, heat flux vector, stress tensor, stress increment (rate), etc., and a set of basic state quanti-
ties, such as electric field vector, strain tensor, strain increment (rate), etc. Then we have various kinds
of constitutive tensor functions for various material behaviour, e.g., strain-energy function; yield function;
stress-strain relation, in particular, generalized Hooke’s law; direct and converse piezoelectric relations;
stress-deformation rate relations for Newtonian and non-Newtonian fluids; stress rate-strain rate relations
for elastoplastic solids; and so many others. These constitutive functions, which represent “individualities”
of materials in response to deformation and other actions, must obey certain universal principles and their
final forms should be determined by adequate and suitable experimental data. A comprehensive account in
this respect may be found in, e.g., Truesdell and Noll (1965) and Haupt (2002).
In the subsequent sections, we shall focus attention on scalar- and symmetric tensor-valued functions
depending on one symmetric tensor, which are adequate in many cases of application. An introduction to
tensor functions in a general sense will be presented only in a brief manner.

5.2 Material Symmetry

A basic, far-reaching fact in constitutive theory of materials is the discovery that if a material is observed
to possess a regular internal structure in a certain scale, such as crystal lattices, then the geometrical sym-
metry of this internal structure should impose invariance restrictions on the tensor function form of any
constitutive function of this material. Such invariance restrictions furnish universal conditions which lead
to considerable simplifications of constitutive functions in most cases. Prior to a detailed study of this
respect, it may be instrumental to have an understanding of how to describe the symmetry of internal struc-
tures of various materials. Here it is impossible and perhaps irrelevant to present a comprehensive account
of material symmetries, which may be found in, e.g., Hahn (1987) and Vainshtein (1994). To our purposes,
it seems adequate to have some relevant knowledge of this broad field, as will be introduced below.

5.2.1 Structural elements and their symmetries

If we observe the internal structures of materials in a certain scale, we may find various different cases.
Some of them look random and disorderly, some of them look regular and orderly. Typical examples of
the latter case are crystal lattices observed in various types of crystals. The basic feature of an ideal crystal
is that a basic structural element, in a regular and periodical manner, repeats itself in the three directions
of space to form the whole material body. Details about such basic structural elements can be found in
monographs on crystallography, e.g., the aforementioned references. In Fig. 5.1 are some examples.

Figure 5.1: Ball, cones, cube, cylinder, prism, and pyramid

5.2. MATERIAL SYMMETRY 85

Each basic structural element itself has certain rotation and/or rotation-reflection symmetric properties
about a center. In addition, the repeated periodicity of basic structural elements has translational symmetric
properties. The combination of these two kinds of symmetry leads to the space symmetry in a general sense.
For our purpose, we direct attention to the former, i.e., the symmetry of basic structural elements.
For a basic structural element, we can find a center in it. Then we may perform various kinds of
orthogonal transformations: inversion about the center, reflection about a plane through the center, rotations
about an axis through the center, and their combinations. If this structural element looks the same as before
after an orthogonal transformation, we say that this orthogonal transformation is a symmetry transformation
of this structural element. For instance, it may be evident that the symmetry transformations of a cube
shown in Fig. 5.1 include the inversion about the center O, the reflections about the four medium planes
through the center O, and the rotations through the angles k × 90◦ about each of the four axes through the
two centers of each pair of surfaces, and other not so obvious orthogonal transformations.
It may be clear that orthogonal tensors, introduced in §3.9, provide a natural tool to describe and
represent the symmetry transformations mentioned above. For the basic structural element of a material,
we put all its symmetric transformations, represented by certain orthogonal tensors, together to form a
collection, known as the symmetry group of this material. Each such group have the following properties
in common:

(i) The identity tensor I is included;

(ii) If a symmetry transformation Q is included, then its inverse Q −1 (= Q T ) is also included;

(iii) If two symmetry transformations Q and R are included, then their composition Q R is also included.

All the symmetry groups in the above sense have been determined and classified for all solid materials,
which are ready for various purposes of application. Here it is not our intention to know how to determine
and classify them. What matters to our purpose is to keep two points in mind: One is that the internal
structure of each solid material in a certain scale may be formed in a regular and periodic manner by
certain basic structural elements and the latter are endowed with certain symmetry properties. The other
is that these symmetry properties may be represented by a collection of orthogonal tensors known as the
(point) symmetry group of this material. Another perhaps essential point is that these symmetry properties
resulting from pure geometric symmetry of basic structural elements of materials in a certain scale will lead
to considerable simplifications of complicated tensor function forms of constitutive functions and relations
of materials, as will be seen in the subsequent development.
In the succeeding subsections, we record all classes of material symmetry groups for future reference.
In literature there have been several ways understanding and describing these classes. However, they might
be rather detailed and technical to our purpose. Here, we shall present them in a suitable, perhaps simple
manner for our purpose.
In the following, n and l 0 are two chosen orthonormal vectors, and n 1 , n 2 and n 3 are three given
orthonormal vectors.

5.2.2 Orthogonal groups and cylindrical groups

The simplest yet commonly-considered solid materials may be regarded to possess the highest symmetry:
every orthogonal tensor is a symmetry transformation. The construction and structure of such materials
may be regarded as the same in every spatial direction in an exact or approximate sense. Such materials are
usually called isotropic materials. Its material symmetry group is given by the following full orthogonal
group
Orth = {all the orthogonal tensors} . (5.1)
Its basic structural element may be represented by a ball. Moreover, possibly the symmetry of some
materials includes every rotation but excludes the inversion −II . Such materials are sometimes called
hemitropic materials. Its material symmetry group is given by the proper orthogonal group

Orth+ = {Q
Q | detQ
Q = 1, Q ∈ Orth} . (5.2)
86 CHAPTER 5. TENSOR FUNCTIONS

Another kind of frequently met materials include the rotations and/or rotation-reflections about a fixed
axis. Their material symmetry groups are given by the following five classes:

Rϕn , ±R
D∞h = {±R Rπl , | l = R ϕn l 0 ; +∞ < ϕ < −∞} , (5.3)

Rϕn , R πl , | l = R ϕn l 0 ; +∞ < ϕ < −∞} ,

D∞ = {R (5.4)
Rϕn , −R
C∞v = {R Rπl , | l = R ϕn l 0 ; +∞ < ϕ < −∞} , (5.5)
Rϕn
C∞h = {±R | +∞ < ϕ < −∞} , (5.6)
Rϕn
C∞ = {R | +∞ < ϕ < −∞} . (5.7)
The basic structural elements may be represented by cylindrical cylinders and cones. Accordingly, the
above classes are known as cylindrical groups. Materials with the above material symmetry are frequently
met and known as transversely isotropic materials, e.g., singly fiber-reinforced composite materials with
isotropic matrix materials, laminates, layered media, etc.

5.2.3 Crystal classes

Crystalline solids are classified into 7 crystal systems composed of 32 crystal classes, as given as follows.

(i) Triclinic system

S2 = {±II } , (5.8)
C1 = {II } . (5.9)
These two classes have the lowest symmetry.

(ii) Monoclinic system

Rπn } ,
C2h = {±II , ±R (5.10)
Rπn } ,
C1h = {II , −R (5.11)
C2 = {II , R πn } . (5.12)

(iii) Rhombic system

Rπn 1 , ±R
D2h = {±II , ±R Rπn 2 , ±R
Rπn 3 } , (5.13)
D2 = {II , R πn 1 , R πn 2 , R πn 3 } , (5.14)
C2v = {II , R πn 3 , −R Rπn 2 } ,
Rπn 1 , −R (5.15)

Some fiber-reinforced composite materials may be characterized by the above rhombic groups and
are usually known as orthotropic materials.

(iv) Trigonal system

2kπ/3 2kπ/3
D3d = {±R
Rn Rπl k | l k = R n
, ±R l 0 , k = 0, 1, 2} , (5.16)
2kπ/3 2kπ/3
C3v = {R
Rn Rπl k | l k = R n
, −R l 0 , k = 0, 1, 2} , (5.17)

2kπ/3 2kπ/3
D3 = {R
Rn , R πl k | l k = R n l 0 , k = 0, 1, 2} , (5.18)

2kπ/3
S6 = {±R
Rn | k = 0, 1, 2} , (5.19)
2kπ/3
C3 = {R
Rn | k = 0, 1, 2} . (5.20)
5.2. MATERIAL SYMMETRY 87

(v) Tetragonal system

2kπ/4 2kπ/4
D4h = {±R
Rn Rπl k | l k = R n
, ±R l 0 , k = 0, 1, 2, 3} , (5.21)

2kπ/4 2kπ/4
C4v = {R
Rn Rπl k | l k = R n
, −R l 0 , k = 0, 1, 2, 3} , (5.22)
2kπ/4 2kπ/4
D2d = {(−1)k R n , ((−1)k R πl k | l k = R n l 0 , k = 0, 1, 2, 3} , (5.23)
2kπ/4 2kπ/4
D4 = {R
Rn , R πl k | l k = R n l 0 , k = 0, 1, 2, 3} , (5.24)
2kπ/4
C4h = {±R
Rn | k = 0, 1, 2, 3} , (5.25)
2kπ/4
S4 = {(−1)k R n | k = 0, 1, 2, 3} , (5.26)
2kπ/4
C4 = {R
Rn | k = 0, 1, 2, 3} . (5.27)

(vi) Hexagonal system

2kπ/6 2kπ/6
D6h = {±R
Rn Rπl k | l k = R n
, ±R l 0 , k = 0, 1, · · · , 5} , (5.28)

2kπ/6 2kπ/6
C6v = {R
Rn Rπl k | l k = R n
, −R l 0 , k = 0, 1, · · · , 5} , (5.29)
2kπ/6 2kπ/4
D3h = {(−1)k R n , ((−1)k R πl k | l k = R n l 0 , k = 0, 1, · · · , 5} , (5.30)
2kπ/6 2kπ/6
D6 = {R
Rn , R πl k | l k = R n l 0 , k = 0, 1, · · · , 5} , (5.31)
2kπ/6
C6h = {±R
Rn | k = 0, 1, · · · , 5} , (5.32)
2kπ/6
C3h = {(−1)k R n | k = 0, 1, · · · , 5} , (5.33)
2kπ/6
C6 = {R
Rn | k = 0, 1, · · · , 5} . (5.34)

(vii) Cubic system

2kπ/4 2tπ/6
Oh = {±R
Rn i Rπn i ±nn j ; ±R
, ±R Rr k | k = 0, 1, 2, 3; j > i = 1, 2, 3} , (5.35)

2kπ/4 2tπ/6
Td = {(−1)k R ni Rπn i ±nn j ; R r k
, −R | k = 0, 1, 2, 3; j > i = 1, 2, 3} , (5.36)
2kπ/4 2tπ/6
O = {R
Rn i , R πn i ±nn j ; R r k | k = 0, 1, 2, 3; j > i = 1, 2, 3} , (5.37)
2tπ/6
Rπn i ±nn j ; ±R
Th = {±R Rr k | k = 0, 1, 2, 3; j > i = 1, 2, 3} , (5.38)
2tπ/6
Rπn i ±nn j ; R r k
T = {R | k = 0, 1, 2, 3; j > i = 1, 2, 3} , (5.39)

In the above, (nn1 , n 2 , n 3 ) are three given orthonormal vectors, as indicated before, and

r 0 = n1 + n2 + n3 , r 1 = n1 − n2 − n3 , r 2 = n2 − n3 − n1 , r 3 = n3 − n1 − n2 . (5.40)

5.2.4 Quasicrystal classes

In addition to the above 32 classes of crystalline materials, there are infinitely many classes of quasi-
crystalline materials, which have internal structures different from those of traditional crystal lattices. They
may be given as follows.
88 CHAPTER 5. TENSOR FUNCTIONS

Subgroups as cylindrical groups

2kπ/2m 2kπ/2m
D2mh = {±R
Rn Rπl k | l k = R n
, ±R l 0 , k = 0, 1, · · · , 2m − 1} ; (5.41)
2kπ/2m 2kπ/2m
C2mv = {R
Rn Rπl k | l k = Rn
, −R l 0 , k = 0, 1, · · · , 2m − 1} , (5.42)
2kπ/2m 2kπ/2m
D2m = {R
Rn , R πl k | l k = R n l 0 , k = 0, 1, · · · , 2m} ; (5.43)
2kπ/2m+1 2kπ/2m+1
D2m+1d = {±R
Rn Rπl k | l k = R n
, ±R l 0 , k = 0, 1, · · · , 2m} , (5.44)
2kπ/2m+1 2kπ/2m+1
C2m+1v = {R
Rn Rπl k | l k = R n
, −R l 0 , k = 0, 1, · · · , 2m} , (5.45)
2kπ/2m+1 2kπ/2m+1
D2m+1 = {R
Rn , R πl k | l k = R n l 0 , k = 0, 1, · · · , 2m} ; (5.46)
2kπ/4m+2 2kπ/4m+2
D2m+1h = {(−1)k R n , (−1)k R πl k | l k = R n l 0 , k = 0, 1, · · · , 4m + 1} ; (5.47)
2kπ/4m 2kπ/4m
D2md = {(−1)k R n , (−1)k R πl k | l k = R n l 0 , k = 0, 1, · · · , 4m − 1} ; (5.48)
2kπ/2m
C2mh = {±R
Rn | k = 0, 1, · · · , 2m − 1} , (5.49)
2kπ/2m
C2m = {R
Rn | k = 0, 1, · · · , 2m − 1} ; (5.50)
2kπ/4m
S4m = {(−1)k R n | k = 0, 1, · · · , 4m − 1} ; (5.51)
2kπ/(4m+2)
C2m+1h = {(−1)k R n | k = 0, 1, · · · , 4m + 1} ; (5.52)
2kπ/2m+1
S4m+2 = {±R
Rn | k = 0, 1, · · · , 2m} , (5.53)
2kπ/2m+1
C2m+1 = {R
Rn | k = 0, 1, · · · , 2m} . (5.54)
In the above, m ≥ 1. In particular, setting m = 1 in (5.44)–(5.48), (5.51)–(5.52), (5.53)–(5.54) and setting
m = 1, 2 in (5.41)–(5.43) and (5.49)–(5.50), we obtain the crystal classes D3d , C3v , D3 , D3h , D2d , S4 , C3h
and D2h , C2v , C2h , C2 and D4h , C4v , C4h , C4 , separately. Other choices of the integer m generate quasicrystal
classes.
The vectors n and l k in (5.41)–(5.54) represent the material symmetry axes. Their geometric and
group-theoretic features are as follows: (i) the axis n is said to be 2m-fold or 4m-fold or (4m + 2)-fold or
2π/n 2π/n
(2m + 1)-fold, for a basic symmetry transformation about this axis, i.e., R n or R n , where n = 2m or
4m or 4m + 2 or 2m + 1, will become the identity transformation I by repeating it n times; (ii) each l k is
2kπ/n
two-fold; and (iii) all the l k equipartition a unit circle in a plane normal to n . Note that l k = R n l 0 means
that the unit vector l 0 normal to n is rotated through k times angle 2π/n about the axis n .

2k π
n
n
l4

l3

l0 l2
l1

Figure 5.2: Rotating vector l 0 through angles 2kπ/n about axis n

With these in mind, the above presentations of crystal and quasi-crystal classes would not only in a
clear and simple manner specify the distributions of all relevant material symmetry axes, but also at the
same time determine what symmetry transformations should be associated with these axes. Note that the
introduction of the factor (−1)k in (5.47-(5.48) and (5.51)-(5.52), in particular in (5.23), (5.26), (5.30),
5.2. MATERIAL SYMMETRY 89

(5.33), and (5.36), naturally indicates whether the 2wo-fold axis l k is associated with a π-rotation or a
refelection, without introducing any additional particular notations.
With the given presentations and the above explanations for geometric and group-theoretic features
of two kinds of axial vectors, it may be felt that the structures of crystal and quasi-crystal classes would
become simple and clear.

Icosahedral groups

There are two classes of icosahedral groups. These groups, characterizing the symmetry of icosahedra, are
among the most complicated yet fascinating material symmetry groups. Although Klein (1884) presented
a comprehensive account of icosahedra and icosahedral groups earlier in the nineteenth century, it has been
believed for a long time that there would be no solid materials with icosahedral symmetry. Icosahedral
quasicrystals, which have been discovered only in recent years, broaden the scope of crystallography and
possess intriguing internal geometric structures and cystallographic properties which have not been known
before. For details, refer to, e.g., Vainshtein (1994) and Senechel (1995). Here we confine ourselves to
some relevant respects of the icosahedral groups.
Let n 5 and n 6 be two given unit vectors meeting at a point and with their angle determined by

1
n5 · n6 = √ . (5.55)
5

Then we may generate five unit vectors n k by rotating the vector n 5 through the angle 2kπ/5 about an axis
in the direction of n 6 , i.e.,
2kπ/5
n k = R n6 n5, k = 1, 2, 3, 4, 5 . (5.56)

It may be clear that the five n k are located on a cone with its angle determined by (5.55) and with its axis
in the direction of n 6 . Hence we have
( √
n 1 · n 2 = n 2 · n 3 = n 3 · n 4 = n 4 · n 5 = n 5 · n 1 = 1/ 5 ,
√ (5.57)
n 1 · n 3 = n 3 · n 5 = n 5 · n 2 = n 2 · n 4 = n 4 · n 1 = −1/ 5 .

Moreover, we introduce vectors

(nn j · n k )nni + (nni · n k )nn j + (nni · n j )nnk , i 6= j 6= k 6= i; i, j, k = 1, · · · , 6 . (5.58)

These produce 20 vectors and a half of them are the opposed vectors of the other half. Retaining one of
these two halves, we obtain ten such vectors and will be denoted r σ with σ = 1, · · · , 10.
The two classes of icosahedral groups are given as follows:

2kπ/5 2kπ/3
Ih = {±R
Rn k , ±R
Rr σ Rπn j ±nnk | j, k = 1, · · · , 6; σ = 1, · · · , 10} ,
, ±R (5.59)

2kπ/5 2kπ/3
Ih = {R
Rn k , Rrσ , R πn j ±nnk | j, k = 1, · · · , 6; σ = 1, · · · , 10} . (5.60)

In the above, each n k , each r σ and each n j ± n k represent a 5-fold, a 3-fold and 2-fold axis of the groups Ih
and I. It should be noted that all the vectors n j ± n k produce only 15 two-fold axes.
The symmetry groups of solid materials are exhausted by the above crystal and quasi-crystal classes,
the full and proper orthogonal groups, and the cylindrical groups.
Detailed geometric features of each class above may be given, some of which have been indicated
above. However, for our purpose here it may be adequate to know the fact that there are many classes of
material symmetry groups and they distinguish themselves by their respective orthogonal transformations.
The latter are listed in the above subsections, separately.
90 CHAPTER 5. TENSOR FUNCTIONS

5.2.5 Isotropic and anisotropic functions of vectors and 2nd-order tensors

What does the material symmetry imply? Here the main fact is the material symmetry principle: the
material symmetry representing the geometric symmetry properties of the basic structural element of a
solid material implies the invariance or the form-invariance of constitutive tensor functions characterizing
physical properties of this materials. A mathematical formulation of this principle may be presented below
in a broad sense.
Let u1 , · · · , ua and A1 , · · · , Ab be a vector variables, b 2nd-order tensor variables, respectively. We
assume that they are well chosen physical and geometrical quantities characterizing and representing the
physical state and the internal state at each point in a deforming material body. We consider scalar-, vector-,
2nd-order tensor-valued functions below:

ξ = f (uu1 , · · · , u a ; A 1 , · · · , A b ) ,

v = ρ(uu1 , · · · , u a ; A 1 , · · · , A b ) ,

S = Φ (uu1 , · · · , u a ; A 1 , · · · , A b ) ,
which specify a scalar quantity ξ, a vector quantity v and a tensor quantity S relying on or determined by the
vector and tensor variables u 1 , · · · , u a and A 1 , · · · , A c . As constitutive tensor functions modeling material
behaviour, each of them should obey the objectivity principle of material behaviour. This principle is
formulated in monographs on continuum mechanics, e.g., Truesdell and Noll (1992), Bertram (1989),
Haupt (2002), and many others. A state-of-the-art study and formulation may be found in, e.g., Svendsen
and Bertram (1999) and Bertram and Svendsen (2001).
From the objectivity principle of material behaviour, reduced forms of constitutive tensor functions may
be derived. Here we assume that the above forms are already such reduced forms obeying the objectivity
principle. Now the material symmetry principle further requires that

Qu 1 , · · · , Q u a ; Q A 1 Q T , · · · , Q A b Q T ) = f (uu1 , · · · , u a ; A 1 , · · · , A b ) for every Q ∈ G ,

f (Q (5.61)

Qu 1 , · · · , Q u a ; Q A 1 Q T , · · · , Q A b Q T ) = Q ρ (uu1 , · · · , u a ; A 1 , · · · , A b ) for every Q ∈ G ,

ρ (Q (5.62)

Qu 1 , · · · , Q u a ; Q A 1 Q T , · · · , Q A b Q T ) = Q Φ (uu1 , · · · , u a ; A 1 , · · · , A b )Q
Φ (Q QT for every Q ∈ G , (5.63)
where G is the material symmetry group of the material under consideration. These conditions simply
mean that if each variable (vector or tensor) is rotated through a symmetric transformation Q of the material
in question, then the corresponding functional value should be invariant (scalar-valued) or form-invariant
(vector- and tensor-valued). Usually, when G = Orth, Orth+ , the above tensor functions obeying the
material symmetry restriction are known as isotropic or hemitropic functions. Otherwise, they are referred
to as anisotropic functions.
The above material symmetry principle places invariance restrictions on the form of constitutive tensor
functions. For all materials characterized by the same class of material symmetry groups, the invariance
restrictions resulting from the material symmetry supply universal conditions for studying and reducing
the form of various kinds of constitutive tensor functions. Then, it is possible from them to derive a
general reduced form of complicated constitutive tensor functions from a purely mathematical procedure,
which thus would substantially facilitate both further theoretical study and experimental determination of
constitutive tensor functions.
According to the material symmetry principle, the invariance conditions for constitutive tensor func-
tions will differ from one another for materials characterized by different classes of material symmetry
groups. It may be important to derive general reduced forms of various kinds of constitutive tensor func-
tions for various types of material symmetry groups, which automatically satisfy the invariance condition
resulting from the material symmetry. This has been the main topic in the theory of representations for
tensor functions. Here it is impossible to present a comprehensive account of this theory in a general sense.
In the subsequent development, we shall study some particular yet commonly-considered cases. Some
general remarks will be given at the end of this chapter.
5.3. SCALAR-VALUED FUNCTIONS OF SYMMETRIC TENSOR 91

5.3 Scalar-Valued Functions of Symmetric Tensor

Let A be a symmetric 2nd-order tensor variable freely changing in a symmetric 2nd-order tensor space or
a suitable subset of this space. A scalar-valued function of tensor variable A specifies a scalar quantity ξ
relying on A . We have
A) .
ξ = f (A (5.64)
Such scalar-valued functions are used to represent the strain energy of elastic solids and the yield
function of elastoplastic solids, etc. In these cases, the tensor variable A may be a finite strain tensor or a
stress tensor, etc.
Let G be the material symmetry group of the solid material under consideration. Then, the above
scalar-valued function as a constitutive function of this material should obey the invariance restriction

QA Q T ) = f (A
f (Q A) for each Q ∈ G . (5.65)

For each material symmetry group G , the above invariance restriction supplies conditions by means of
which the form of the tensor function f (A A) may be reduced. Such conditions are different for materials
characterized by different classes of material symmetry groups. In what follows, we shall present a general
A) for each class of material symmetry groups listed before, which
reduced form of the scalar function f (A
automatically obeys the invariance restriction (5.65).

5.3.1 Isotropic invariants

The simplest yet commonly-considered case is isotropic materials, for which G = Orth. In this case, (5.65)
becomes
QA Q T ) = f (A
f (Q A) for every orthogonal tensor Q . (5.66)
Such a scalar-valued function f (AA) is known as an isotropic invariant of A .
The three basic and principal invariants of tensor A (cf. (2.76) and (2.78)) provide examples of isotropic
invariants. Below we demonstrate that these simple examples can result in general reduced forms of the
isotropic invariant of A .
To this goal, we apply the characteristic expression of A (cf. (2.95)). Without loss of generality,
we assume that A has three distinct eigenvalues. Since a symmetric 2nd-order tensor A can uniquely
determined by its three eigenvalues λr (A A) and its three eigenprojections A r and since the former can be
determined by its three basic invariants Ir (AA) (cf. (2.77)–(2.78)), we know that a scalar-valued function
A) may be reformulated as a function of Ir (A
f (A A) and its three eigenprojections A r , i.e.,

A) = g(I1 (A
f (A A), I2 (A
A), I3 (A
A); A 1 , A 2 , A 3 ) .

For each orthogonal tensor Q , we have

QA Q T ) = Ir (A
Ir (Q A) ,

and, besides, it may be evident that the eigenprojections of Q A Q T are given by Q A r Q T . Hence we have

QA Q T ) = g(I1 (A
f (Q A), I2 (A A); Q A 1 Q T , Q A 2 Q T , Q A 3 Q T ) .
A), I3 (A

Now the conditions (5.54) requires

A), I2 (A
g(I1 (A A); Q A 1 Q T , Q A 2 Q T , Q A 3 Q T ) = g(I1 (A
A), I3 (A A), I2 (A
A), I3 (A
A); A 1 , A 2 , A 3 )

for every orthogonal tensor Q .

Observing that Q A r Q T generates all possible eigenprojections when Q runs over all orthogonal tensors,
we conclude that the last expression above means that the function

A), I2 (A
g(I1 (A A), I3 (A
A); A 1 , A 2 , A 3 )
92 CHAPTER 5. TENSOR FUNCTIONS

always yields the same value for all possible choices of its last three variables A r , and, therefore, it should
be independent of A r . Thus, we arrive at

A) = h(I1 (A
f (A A), I2 (A
A), I3 (A
A)) = h(I1 , I2 , I3 ) (5.67)

A) meeting (5.66).
for each isotropic invariant f (A
From (5.67) follows that every isotropic invariant of A is expressible as a single-valued function of the
three basic invariants of A . Since the basic invariants of A can be determined either by the three principal
A) or by the three eigenvalues λr , we have the following two alternative reduced forms for
invariants Jr (A
isotropic invariants:
A) = h̄(J1 (A
f (A A), J2 (A
A), J3 (A
A)) = h̄(J1 , J2 , J3 ) , (5.68)
A) = h̄(λ1 (A
f (A A), λ2 (A
A), λ3 (A
A)) = ĥ(λ1 , λ2 , λ3 ) . (5.69)
In the last expression, the function f (λ1 , λ2 , λ3 ) should be symmetric with respect to its any two variables.

5.3.2 Anisotropic invariants

Since
QA Q T ) = f (A
f (Q A) , Q = −II , (5.70)
holds for each scalar function f (AA), It suffices to treat the material symmetry groups including the inversion
−II .
When the material symmetry group G in (5.65) is neither the full nor the proper orthogonal group, i.e.,
it is one of the groups listed in §5.2, a scalar function f (A A) meeting (5.65) is referred to as an anisotropic
invariant of A relative to the group G . As hinted by the result (5.67) for isotropic invariants, the main task
in the study of isotropic and anisotropic invariants is as follows: For each material symmetry group G , find
out a finite number of specified invariants of A relative to G , i.e., K1 , · · · , Kt , such that every invariant f (A
A)
obeying (5.65) is expressible as a single-valued function of these specified invariants. Namely,

f (A A), · · · , Kt (A
A) = h(K1 (A A)) . (5.71)

We call a set of specified invariants, K1 , · · · , Kt , that can attain the above goal is a functional basis of the
invariants relative to the material symmetry group G . Further, to have a sharp, compact general reduced
form, it is required that a functional basis should be irreducible in the following sense: a subset resulting
from the removal of any invariant Iα can no longer serve as a functional basis. Thus, our task is to find
an irreducible functional basis, K1 , · · · , Kt , in order to obtain an irreducible representation of the invariants
relative to each material symmetry group G .
Expressions (5.67)–(5.69) supply three irreducible representations for isotropic invariants. Accord-
ingly, (I1 , I2 , I3 ), (J1 , J2 , J3 ) and (λ1 , λ2 , λ3 ) supply three irreducible functional bases for isotropic invari-
ants. These examples show that even irreducible functional basis need not be unique.
Except for some simple cases, determination of irreducible representations of anisotropic invariants
relative to various types of material symmetry groups does not appear to be easy. The derivation is usually
too technical. After some remarks on literature, we shall simply cite the final results.
The classical results for anisotropic invariants were restricted to the case that f (A A) is a polynomial
function of the components of A , and attention was directed toward the 32 crystal classes and transverse
isotropy groups. Earlier attempt was made by von Mises (1928) and later was known to be unsuccessful.
Rivlin and Smith (1958) and Smith (1962) derived well-known results for polynomial invariants relative
to all the 32 crystal classes. These results are recorded in Green and Adkins (1960), Truesdell and Noll
(1992), Spencer (1971), Smith (1994), Haupt (2002), et al. On the other hand, an extensive, modern study
of anisotropic invariants as elastic strain-energy functions has been presented by Hackl (1999) based upon
symmetric irreducible tensors and group representation theory.
Complete and irreducible representations in a general sense covering both polynomial and non-polynomial
invariants are obtained in recent years. In Xiao (1996d), general results are presented for 32 crystal classes.
Results in unified forms are derived in Xiao (1998a) and Bruhns, Xiao and Meyers (1999a) for all cystal
classes and quasi-crystal classes. It is found that these new results in a general sense could be even sharper
and more compact than the classical results in a restrictive sense in many cases.
5.3. SCALAR-VALUED FUNCTIONS OF SYMMETRIC TENSOR 93

Below we record these new results in a general sense. The classical results may be found in the refer-
ences mentioned before.

Cylindrical groups

G = D∞h ,C∞v , D∞ (transversely isotropic)

A) = h(I1 , I2 , I3 , n · A [nn], n · A 2 [nn]) .
f (A (5.72)

G = C∞h ,C∞ (transversely isotropic)

A) = h(I1 , I2 , I3 , n · A [nn], n · A 2 [nn], (nn × A [nn]) · A 2 [nn]) .
f (A (5.73)

Crystal classes

G = C1 , S2 (triclinic)
A) = h(A11 , A22 , A33 , A12 , A23 , A31 ) .
f (A (5.74)
In the above, Ai j are the components of A relative a standard basis e i . It should be noted that the above
form depends on the choice of basis e i .
G = C2h ,C1h ,C2 (monoclinic)
A) = h(A11 , A22 , A33 , A12 , A223 , A231 , A23 A31 ) .
f (A (5.75)

In the above, Ai j are the components of A relative to a standard basis (ee1 , e 2 , e 3 ), where e 3 = n and e 1 and
e 2 are two orthonormal vectors in a plane normal to n .
G = D2h ,C2v , D2 (rhombic; orthotropic)
A) = h(A11 , A22 , A33 , A212 , A213 , A223 , A12 A23 A31 ) .
f (A (5.76)

In the above, Ai j are the components of A relative to the standard basis (nn1 , n2 , n3 ).
G = D3d ,C3v , D3 (trigonal)
A) =h(A33 , A11 + A22 , (A11 − A22 )2 + 4A212 , A223 + A231 , A23 (3C31
f (A 2 − A2 ),
23
(A11 − A22 )A23 + 2A31 A12 , (A11 − A22 )(12A212 − (A11 − A22 )2 ), (5.77)
(A11 − A22 )(A231 − A223 ) + 4A12 A23 A31 )) .

In the above and (5.79)–(5.83) given later, Ai j are the components of tensor A relative to the standard basis

e1 = l 0, e2 = n × l 0, e3 = n . (5.78)

G = S6 ,C3 (trigonal)
A) =h(A33 , A32 (3A231 − A223 ), A31 (3C23
f (A 2 − A2 ), A + A ,
31 11 22

(A11 − A22 )(12A212 − (A11 − A22 )2 ), A12 (4A212 − 3(A11 − A22 )2 ),

(5.79)
(A11 − A22 )(A231 − A223 ) + 4A12 A23 A31 ,
A12 (A231 − A223 ) − (A11 − A22 )A23 A31 )) .

G = D4h ,C4v , D2d , D4 (tetragonal)

A) =h(A33 , A11 + A22 , A212 , (A11 − A22 )2 , A223 + A231 , A223 A231 ,
f (A
(5.80)
(A11 − A22 )(A213 − A223 ), A12 A23 A31 , ) .
94 CHAPTER 5. TENSOR FUNCTIONS

The above result is derived by simplifying that given in Xiao (1998; p. 1236).
G = C4h , S4 ,C4 (tetragonal)
A) =h(A33 , A11 + A22 , (A11 − A22 )2 − 4A212 , A12 (A11 A22 ),
f (A
A413 + A423 − 6A213 A223 , A13 A23 (A213 − A223 ),
(5.81)
(A11 − A22 )(A213 − A223 ) + 4A12 A23 A31 ,
A12 (A23 A223 ) − (A11 − A22 )A13 A23 ) .

G = D6h ,C6v , D3h , D6 (hexagonal)

A) =h(A33 , A11 + A22 , (A11 − A22 )2 + 4A212 , (A11 − A22 )(12A212 − (A11 − A22 )2 ),
f (A
A223 + A231 , (A231 − A223 )(C13
4 + A4 − 14A2 A2 ),
23 13 23
(5.82)
(A11 − A22 )(A231 − A223 ) + 4A12 A23 A31 ,
(A231 − A223 )(4A212 − (A211 − A222 )2 ) + 8A12 A23 A31 (A11 − A22 )) .

G = C6h ,C3h ,C6 (hexagonal)

A) =h(A33 , A11 + A22 , (A231 − A223 )(A431 + A423 − 14A231 A223 ),
f (A
A31 A23 (3A231 − A232 )(A231 − 3A223 ), (A11 − A22 )(12A212 − (A11 − A22 )2 ),
(5.83)
A12 (4A212 − 3(A11 − A22 )2 ), (A11 − A22 )(A231 − A223 ) + 4A12 A23 A31 ,
A12 (A231 − A223 ) − (A11 − A22 )A31 A32 ) .

G = Oh , Td , O (cubic)
A) =h(A11 + A22 + A33 , A211 + A222 + A233 , A11 A22 A33 ,
f (A
A212 + A223 + A231 , A412 + A423 + A431 , A12 A23 A31 , (5.84)
A11 A223 + A22 A231 + A33 A212 , A211 A223 + A222 A231 + A233 A212 ) .

G = Th , T (cubic)
A) =h(A11 + A22 + A33 , A211 + A222 + A233 , A11 A22 A33 ,
f (A
A212 + A223 + A231 , A412 + A423 + A431 , A12 A23 A31 ,
A11 A223 + A22 A231 + A33 A212 , A211 A223 + A222 A231 + A233 A212 , (5.85)
(A11 − A22 )(A22 − A33 )(A33 − A11 ),
(A223 − A231 )(A231 − A212 )(A212 − A223 )) .

Quasi-crystal classes

In the following results for quasicrystal classes, we take a standard basis e i as given by (5.78) and
denote
1
p = A31 e 1 + A32 e 2 , q = (A11 − A22 )ee1 + A12 e 2 , (5.86)
2
φ = cos−1 (pp · e 1 ) , ψ = cos−1 (qq · e 1 ) . (5.87)
Hence we have
1
|pp|2 = A213 + A223 , |qq|2 = (A11 − A22 )2 + A212 . (5.88)
4
G = D2mh ,C2mv , D2m , D2m̄−1h , D2m̂d for m ≥ 2
5.3. SCALAR-VALUED FUNCTIONS OF SYMMETRIC TENSOR 95

An irreducible representation is given by

A) = h(K1 , · · · , K8 ) ,
f (A (5.89)

where
A − n · A [nn] , K3 = |pp|2 , K4 = |qq|2 ,
K = n · A [nn] , K2 = trA

 1

K5 = |pp|2m cos 2mφ, K6 = |qq|m cos mψ , (5.90)

K7 = |pp|2 |qq| cos(2φ − ψ) , K8 = |pp|2 |qq|m−1 cos(2φ + (m − 1)ψ) .


In the above and below, m=1, 2, 3, · · ·, and m̄ = (m − 1)/2 for an odd m, and m̂ = m/2 for an even m.
G = C2mh ,C2m ,C2m̄−1h , S4m̂ for m ≥ 2
An irreducible representation is given by (5.89) with


 K1 = n · A [nn] , K2 = trA
A − n · A [nn];

 K3 = |pp|2m cos 2mφ , K4 = |pp|2m sin 2mφ,

(5.91)


 K5 = |qq|m cos mψ , K6 = |qq|m sin mψ ,

K7 = |pp|2 q | cos(2φ − ψ) , K8 = |pp|2 |qq| sin(2φ − ψ) .


G = D2m+1d ,C2m+1v , D2m+1 for m ≥ 1

An irreducible representation is given by (5.89) with

A − n · A [nn] , K3 = |pp|2 , K4 = |qq|2 ;

K = n · A [nn] , K2 = trA

 1

K5 = |pp|2m+1 cos(2m + 1)φ , K6 = |pp|2m+1 sin(2m + 1)φ ; (5.92)

K7 = |pp|2 q | cos(2φ − ψ) , K8 = |pp|2m−1 |qq| sin((2m − 1)φ + ψ) .


G = S4m+2 ,C2m+1 for m ≥ 1

An irreducible representation is given by (5.89) with


 K1 = n · A [nn] , K2 = trA
A − n · A [nn] ;

 K3 = |pp|2m+1 cos(2m + 1)φ , K4 = |pp|2m+1 sin(2m + 1)φ ,

(5.93)
 K5 = |qq|2m+1 cos(2m + 1)ψ ; K6 = |qq|m sin(2m + 1)ψ ,



K7 = |pp|2 q | cos(2φ − ψ) , K8 = |pp|2 |qq| sin(2φ − ψ) .


The above results exhaust all the quasicrystal classes as subgroups of cylindrical groups. A result for
icosahedral classes is given in Xiao (1998a). However, this result is a bit complicated, and it is not yet
known whether it is irreducible or not. Hence it is not listed here.

5.3.3 Continuity and Derivatives

For a function of one real variable, i.e., f (x), the continuity at a point x means that the magnitude of
value difference, i.e., | f (x0 ) − f (x)|, should also be sufficiently small, whenever the magnitude of variable
difference, i.e., |x0 − x|, is sufficiently small. This concept may be extended to tensor functions. Here, the
variable x is replaced by symmetric 2nd-order tensor A , and the magnitude of variable difference by
q q
A0 − A | ≡ tr(A
|A A0 − A )T (A
A0 − A ) = tr(A A0 − A )2 (5.94)

for any two symmetric 2nd-order tensors A 0 and A .

We first consider scalar-valued functions. Tensor-valued functions will be treated later on. We say that
A) is continuous at A , if for any given small δ > 0, there exists a small ε such that
a scalar function f (A

A0 ) − f (A
| f (A A)| < δ (5.95)
96 CHAPTER 5. TENSOR FUNCTIONS

for any two symmetric 2nd-order tensors A 0 and A meeting

A0 − A | < ε .
|A (5.96)

A) is continuous at A , we can approximate f (A

Hence, if f (A A0 ) by f (A
A). The error will become smaller
0
when |AA − A | becomes smaller. To obtain a further, better approximation, we need the concept of deriva-
tive.
A suitable extension of the derivative of real functions to tensor function is not so direct. A definition
is given as follows.
Prior to the definition, we introduce a useful symbol o(ε) to represent a small quantity of higher order
of magnitude than a given small quantity ε, i.e.,
o(ε)
lim = 0. (5.97)
ε→0 ε
A) is Fréchet-differentiable at A , if there is a given symmetric 2nd-order
We say that a scalar function f (A
A, such that
tensor, denoted ∂ f /∂A
∂f
A0 ) − f (A A0 − A ) + o |A
A0 − A |


f (A A) = (A (5.98)
A
∂A
for any two symmetric 2nd-order tensors A 0 and A that are close to each other. The tensor ∂ f /∂A A, which is
either constant or dependent only on A and fulfills the just-mentioned condition at each A , is known as the
Fréchet derivative of the scalar function f (AA) at A .
We say that a scalar function f (A A) is continuous (resp., differentiable), if it is continuous (resp., differ-
entiable) at every A .
It can be proved that if f (AA) is differentiable at A , then f (A
A) is continuous at A . The converse needs
not be true, as in the case of real functions.
If we replace A 0 in (5.98) with A + εX X where the number ε may go to vanish but X is any given
symmetric 2nd-order tensor, then by using (5.97) we deduce
f (A X ) − f (A
A + εX A) ∂f
lim = X (5.99)
ε→0 ε A
∂A
for any symmetric 2nd-order tensor X .
(5.99) itself, not as a consequence of (5.98), leads to the notion of Gâteaux-differentiability property and
Gâteaux-derivative of f (A A). This notion is coincident with that introduced before under certain conditions.
Generally, the former could not guarantee the latter. Note that (5.99) may be derived as a consequence of
A by using
Fréchet-differentiability property. Usually, it may be easier to derive the Fréchet derivative ∂ f /∂A
(5.99) than using (5.98).
From now on, by differentiability and derivative we mean those in Fréchet’s sense.
Whenever a fixed standard basis e i is chosen, a scalar function f (A A) will be a function of the six
standard components Ai j relative to this basis. In this case, we have the useful expression

∂f ∂f
= ei ⊗ e j . (5.100)
A ∂Ai j
∂A
Some simple rules for derivative are as follows:
∂( f1 + f2 ) ∂ f1 ∂ f2
= + , (5.101)
∂AA A
∂A A
∂A
∂( f1 f2 ) ∂ f2 ∂ f1
= f1 + f2 (product rule) , (5.102)
∂AA ∂AA ∂AA
for any two differential scalar functions f1 = f (A A) and f2 = f2 (A
A); and

∂h ∂h ∂K1 ∂h ∂Ks
= +···+ (Leibniz’s chain rule) , (5.103)
A ∂K1 ∂A
∂A A A
∂Ks ∂A
5.4. SYMMETRIC TENSOR-VALUED FUNCTIONS OF SYMMETRIC TENSOR 97

where h = h(K1 , · · · , Ks ) is a differentiable function of scalars K1 , · · · , Ks and each of the latter is a differen-
tial scalar functions of tensor variable A .
It may be useful to have expressions for derivatives of some frequently used invariants. An example is
the stress-strain relation of a hyperelastic body, which should be derived from the strain-energy function.
In the following, we take isotropic invariants into consideration.
For the basic invariants of A, i.e., Ir = Ir (AA)(cf. (2.76))), we have

∂I1 ∂I2 ∂I3

= I, A,
= 2A A.
= 3A (5.104)
A
∂A A
∂A A
∂A
Generally, for each integer t we have
∂It At
∂trA
= At−1 .
= tA (5.105)
A
∂A ∂AA
A) (cf. (2.78)), we have
For the principal invariants of A , i.e., Jr = Jr (A

∂J1 ∂J2 ∂J3

=I, = J1 I − A , = J2 I − J1 A + A 2 . (5.106)
A
∂A A
∂A A
∂A
For each eigenvalue λσ of A , utilizing the rules (5.101)–(5.103) and differentiating the characteristic
equation of A (cf. (2.77)), we infer

∂λσ ∂J1 ∂J2 ∂J3

(3λ2σ − 2J1 λσ + J2 ) − λ2σ + λσ − = O.
A
∂A A
∂A A
∂A A
∂A
Then, using this and !
m
∂λσ ∂λσ ∂λσ
= I= ∑ Aσ ,
A
∂A A
∂A A
∂A σ=1

as well as (2.105), (2.107) and (5.106), we arrive at the interesting result

∂λσ
= Aσ , (5.107)
A
∂A
where A σ is the eigenprojection of A subordinate to the eigenvalue λσ of A . This means that the derivative
of each eigenvalue is just the eigenprojection subordinate to this eigenvalue.
Twice or higher derivative may be defined, which will be postponed till §5.4.3.

5.4 Symmetric Tensor-Valued Functions of Symmetric Tensor

A symmetric 2nd-order tensor-valued function of symmetric 2nd-order tensor variable A specifies a sym-
metric 2nd-order tensor S relying on or determined by tensor variable A , namely,

A) .
S = Φ (A (5.108)

For the sake of simplicity, (5.108) will be abbreviated to tensor-valued function. An obvious fact is

QA Q T ) = Q Φ (A
Φ (Q QT = Φ (A
A)Q A) , Q = −II , (5.109)

A). This fact implies that the conditions (5.110) are the same for any
for every tensor-valued function Φ(A
two groups G1 and G2 related by
G2 = {QQ, −QQ | Q ∈ G1 } .
If (5.108) represents a constitutive relation of a material with the material symmetry group G , then the
material symmetry principle formulated in §5.2.5 leads to the form-invariance restriction

QA Q T ) = Q Φ (A
Φ (Q QT
A)Q for each Q ∈ G . (5.110)
98 CHAPTER 5. TENSOR FUNCTIONS

A) satisfying
In what follows, we shall present a general reduced form of the tensor-valued function Φ (A
the form-invariance restriction (5.110) for each class of material symmetry group G .
A) is linear in A , then we have (3.45), i.e.,
In particular, if Φ (A

A] ,
S = L [A

where L is a constant 4th-order tensor with minor symmetry (3.47). Then, the restriction (5.110) reduces
to
Qip Q jq Qkr Qls L pqrs = Li jkl for each Q ∈ G .
Usually, tensors L obeying the above restriction are known as isotropic 4th-order tensors for G =Orth, Orth+
or anisotropic 4th-order tensors for any other G . The general forms of 4th-order isotropic and anisotropic
tensors may be found in, e.g., Nye (1985) and Gurtin (1972). The isotropic 4th-order tensor will be dis-
cussed next subsection. Further study of isotropic and anisotropic 4th-order tensors in modern sense may
be found, e.g., in the relevant references mentioned in §3.10.2 and in Huo and Del Piero (1991), Forte
and Vianello (1996, 1997), Xiao (1998c), Chadwick, Vianello and Cowin (2001), as well as the references
therein.

5.4.1 Isotropic functions

For isotropic materials, we have G = Orth in (5.110). In this case, Φ (A
A) is called an isotropic tensor-valued
function of A . The simplest examples of isotropic tensor-valued functions are given by the powers A α with
α=0, 1, 2, · · ·. We shall show that these three simplest examples A 0 (= I ), A , A 2 can actually generate all
isotropic tensor-valued functions Φ (A A).
Let a r be three orthonormal eigenvectors of A . We have

QT = (Q
Q (aar ⊗ a r )Q Qa r ) ⊗ (Q
Qa r ) = a r ⊗ a r , Q = R πa k , k = 1, 2, 3.

From these and the characteristic expression of A we deduce

Q A QT = A , Q = Rπa k , k = 1, 2, 3.

Setting Q = R πa k in (5.110), we deduce

Q Φ (A QT = Φ (A
A)Q A) , Q = R πa k , k = 1, 2, 3.

Now we apply the characteristic expression of the symmetric tensor S = Φ(A

A), i.e.,
3
A) =
S = Φ (A ∑ sr s r ⊗ s r ,
r=1

where the sr are three eigenvalues of S and the s r are three subordinate orthonormal eigenvectors of A .
Hence,
3
Q S Q T = Q Φ (A QT =
A)Q ∑ sr (QQs r ) ⊗ (QQs r ) .
r=1

Then, from the last three expressions above we deduce

3 3
∑ sr s r ⊗ s r = ∑ sr (QQs r ) ⊗ (QQs r ) , Q = R πa k , k = 1, 2, 3.
r=1 r=1

Without losing generality, we assume that the three eigenvalues sr of S are distinct. Then the last expression
means that both s r ⊗ s r (no summation) and (QQs r ) ⊗ (Q
Qs r ) (no summation) are the same eigenprojection
of S subordinate to λr . This fact results in

Q s r = −ssr , Q = R πak , k = 1, 2, 3.
5.4. SYMMETRIC TENSOR-VALUED FUNCTIONS OF SYMMETRIC TENSOR 99

This is possible only when s r = ±aar , namely, S and A have three orthonormal eigenvectors in common.
Thus, we have
A) = ∑3r=1 sr A r ,

 S = Φ (A

sr = A r Φ (A A) = sr (λ1 , λ2 , λ3 ) , (5.111)


sτ(r) (λτ(1) , λτ(2) , λτ(3) ) = sr (λ1 , λ2 , λ3 ) ,
for every isotropic function S = Φ (AA). In the above, A r = a r ⊗ a r (no summation) is the eigenprojection
subordinate to λr , and τ is any permutation transforming the three numbers 1, 2, 3 into themselves.
Applying Sylvester’s formula (cf. (2.112)–(2.115)) for eigenprojections A r to (5.111), we arrive at the
well-known Rivlin-Ericksen theorem:
A) = f0 I + f1 A + f2 A 2 ,
S = Φ (A (5.112)
where each scalar coefficient ft is an isotropic invariant of A and hence may be expressed as one of the
three forms as given by (5.67)–(5.69).
It turns out that each isotropic tensor-valued function Φ (AA) is simply a linear combination of the first
three powers of tensor variable A , i.e., A α with α =0, 1, 2, where each scalar coefficient is an isotropic
invariant. Note here that each A α is a specified isotropic tensor-valued functions of A . On account of the
general reduced form (5.118), we say that the three specified isotropic tensor-valued functions I , A , A 2 form
a generating set for all the isotropic tensor-valued functions Φ (A A), and that (5.118) is the representation
for the isotropic tensor-valued functions Φ (AA) in terms of the generating set {II , A , A 2 }.
The simplest yet most widely considered case is the linear case of the tensor function (5.108), i.e.,
S = C[A
A] ,

where C a constant 4th-order tensor with minor symmetry (cf. (3.47)). In this particular case, the form-
invariance condition (5.110) with G = Orth reduces to the requirement
Qip Q jq Qkr Qls C pqrs = Ci jkl for each Q ∈ Orth . (5.113)
Such a 4th-order tensor is known as isotropic tensor, as mentioned before. Its general form will be derived
below.
In general, each coefficient fl in (5.118) is a function of the three basic invariants Ir of the tensor A , as
shown by (5.67). For a linear tensor-valued function Φ (A A), we may set
A) ,
f0 = ΛI1 = Λ(trA f1 = 2G , f2 = 0 ,
in (5.118). In the above, Λ and G are two scalar constants. Thus, we derive a general reduced form of
A) as follows:
isotropic linear tensor-valued function Φ (A
A] = Λ(trA
S = C[A A)II + 2GA
A. (5.114)
From this we deduce that a general form of isotropic 4th-order tensor is given by
C = ΛII ⊗ I + 2G Isym . (5.115)
This and the decomposition (3.55) yield an alternative form below:
1
C = (3Λ + 2G) I ⊗ I + 2G Īsym . (5.116)
3
Since
     
1 1 1 1
I ⊗I ◦ I ⊗ I = I ⊗ I, I ⊗ I ◦ Īsym = O, Īsym ◦ Īsym = Īsym , (5.117)
3 3 3 3
we infer that (5.116) is just the characteristic expression of the isotropic 4th-order tensor C. Then we have
(cf., (3.101)
1
C−1 = (3Λ + 2G)−1 I ⊗ I + (2G)−1 Īsym . (5.118)
3
100 CHAPTER 5. TENSOR FUNCTIONS

5.4.2 Anisotropic functions

If the material symmetry group G is neither Orth nor Orth+ , then a tensor-valued function Φ (A A) obeying
the form-invariance conditions (5.110) is referred to as an anisotropic tensor-valued function of tensor
variable A relative to the material symmetry group G.
Extending the foregoing idea related to general reduced forms of isotropic tensor-valued functions, for
each material symmetry group G we may find out a finite number of specified tensor-valued functions
meeting (5.110), say, G 1 (AA), · · · , G s (A
A), such that each isotropic tensor-valued function Φ (A
A) meeting
(5.110) is expressible as a linear combination of them with each coefficient is an invariant meeting (5.65).
Namely,
Φ (A A) + · · · + fs G s (A
A) = f1 G 1 (A A) , (5.119)
for each Φ (A A) meeting (5.110), where each scalar coefficient fτ = fτ (A A) is an invariant of A meeting
(5.65). Such specified tensor-valued functions meeting (5.110) is said to form a generating set of all the
tensor-valued functions Φ (A A) meeting (5.110), and expression (5.119) is called the representation for the
tensor-valued functions Φ (A A) meeting (5.110).
Moreover, to have a compact representation, we may further require that a generating set to be used
is irreducible, in the sense that any of its subsets is no longer a generating set. If possible, even we may
require that a generating set to be used should be a minimal one, in the sense that there is no generating set
which includes a smaller number of elements than this generating set.
Just as in the case of scalar-valued functions, results for anisotropic tensor-valued functions were mainly
limited to polynomial cases. Usually, results in this restrictive sense may be derived from the corresponding
results for polynomial scalar-valued functions; refer to the relevant references mentioned in §5.3. It seems
that results for irreducible and even minimal representations in a general sense covering both polynomial
and non-polynomial cases have not been available until most recently. In Xiao (1995b, 1996a, 1997a,
1999), results for irreducible representations in a general sense are derived for all 32 crystal classes and for
all quasi-crystal classes. Each of these results is expressed in terms of a generating set including not more
than nine elements. It is proved in Xiao (1996a) that every generating set for anisotropic tensor-valued
functions has to include not less than six elements. In this reference, results for minimal representations
are given for transversely isotropic solids and certain classes of crystals and quasi-crystals.
In the following, we shall record the above new results. The notations used in the following will be the
same as the corresponding cases in §5.3. Besides, we shall use the notation
u1 ∨ u2 ≡ u1 ⊗ u2 + u2 ⊗ u1 (5.120)
for any two vectors u 1 and u 2 .

Cylindrical groups

G = D∞h ,C∞v , D∞ (transversely isotropic)

A minimal representation is given by
Φ (A A) + · · · + f6 G (A
A) = f1 G 1 (A A) , (5.121)
where
G1 = I , G2 = n ⊗ n , G3 = A , G4 = A2 ,
(
(5.122)
G 5 = n ∨ A [nn] , G 6 = n ∨ A 2 [nn] .
A simple physically-motivated representation is presented by Bischoff-Beiermann and Bruhns (1994).
G = C∞h ,C∞ (transversely isotropic)
A minimal representation is given by (5.121) with
(
G 1 = I , G 2 = n ⊗ n , G 3 = A , G 4 = A (εε[nn]) − (εε[nn])A
A,
(5.123)
G 5 = G 2 , G 6 = G 2 (εε[nn]) − (εε[nn])G
G2 ,
where
G = (nn · A e 1 )nn ∨ e 1 + (nn · A e 2 )nn ∨ e 2 . (5.124)
5.4. SYMMETRIC TENSOR-VALUED FUNCTIONS OF SYMMETRIC TENSOR 101

Crystal classes

G = S2 ,C1 (triclinic)
Φ (A A)eei ⊗ e j .
A) = Si j (A (5.125)

G = C2h ,C1h ,C2 (monoclinic)

A minimal representation is given by (5.121) with
(
G1 = n ⊗ n , G2 = e1 ⊗ e1 , G3 = e2 ⊗ e2 , G4 = e1 ∨ e2 ,
(5.126)
G 5 = n ∨ A [nn] , G 6 = n ∨ (nn × A [nn]) .

G = D2h ,C2v , D2 (rhombic; orthotropic)

A minimal representation is given by (5.121) with


 G1 = n1 ⊗ n1 , G2 = n2 ⊗ n2 , G3 = n3 ⊗ n3 ,

 G 4 = A12 n 1 ∨ n 2 − A31 n 3 ∨ n 1 ,

(5.127)
 G 5 = A23 n 2 ∨ n 3 − A12 n 1 ∨ n 2 ,



G 6 = A23 A31 n 1 ∨ n 2 + A31 A12 n 2 ∨ n 3 + A12 A23 n 3 ∨ n 1 .


G = D3d ,C3v , D3 (trigonal)

An irreducible representation is given by

Φ(A A) + · · · + f9 G9 (A
A) = f1 G1 (A A) , (5.128)

A) are given by (5.136)–(5.137) and by setting m = 1 in (5.142)–(5.146) given

where the nine G r = G r (A
later.
G = S6 ,C3 (trigonal)
A minimal representation is given by (5.121) with


 G1 = I , G2 = n ⊗ n ,


 G 3 = (A

 A D 1 )D D1 + (AA D 2 )DD2 + (A
A D3 + (A
D 3 )D A D4 ,
D 4 )D

A D 2 )D
G 4 = (A D1 − (AA D 1 )DD2 + (A
A D3 − (A
D 4 )D A D4 ,
D 3 )D (5.129)

D1 − (A D2 − (A



 A D 3 )D
G 5 = (A A D 4 )D A D3 + (A
D 1 )D A D4 ,
D 2 )D


A D 4 )D
G 6 = (A D1 + (AA D 3 )DD2 − (A
A D3 − (A
D 2 )D A D4 ,
D 1 )D


where the D α are the four mutually orthogonal shears given by

D1 = e1 ⊗ e1 − e2 ⊗ e2 , D2 = e1 ∨ e2 , D3 = n ∨ e1 , D4 = n ∨ e2 . (5.130)

G = D4h ,C4v , D2d , D4 (tetragonal)

An irreducible representation is given by

Φ (A A) + · · · + f8 G 8 (A
A) = f1 G 1 (A A) , (5.131)

A) are given by (5.136)–(5.137) and by setting m = 2 in (5.138)–(5.141) given

where the eight G r = G r (A
later.
G = C4h , S4 ,C4 (tetragonal)
A minimal representation is given by (5.121) with (5.123)–(5.124), but here the difference is that each
scalar coefficient fα should be an invariant of A relative to C4h .
G = D6h ,C6v , D3h , D6 (hexagonal)
A) are given by setting
An irreducible representation is given by (5.131), where the eight G r = G r (A
m = 3 in (5.136)–(5.141) given later.
102 CHAPTER 5. TENSOR FUNCTIONS

G = C6h ,C3h ,C6 (hexagonal)

A minimal representation is given by (5.121) with (5.123)–(5.124), but here the difference is that each
scalar coefficient fα should be an invariant of A relative to C6h .
G = Th , T (cubic)
An irreducible representation is given by (5.131) with

G1 = I , G2 = A , G3 = A2 , G4 = r ⊗ r , G5 = S ,
(
(5.132)
A , G 7 = A 2 (εε[rr ]) − (εε[rr ])A
G 6 = A (εε[rr ]) − (εε[rr ])A A2 , G 8 = S (εε[rr ]) − (εε[rr ])SS ,

where
A) = A23 n 1 + A31 n 2 + A12 n 3 ,
r = r (A (5.133)
A) = (A22 − A33 )nn1 ⊗ n 1 + (A33 − A11 )nn2 ⊗ n 2 + (A11 − A22 )nn3 ⊗ n 3 .
S = S (A (5.134)

G = Oh , Td , O (cubic)
An irreducible representation is given by (5.128) with

G 1 =II = n 1 ⊗ n 1 + n 2 ⊗ n 2 + n 3 ⊗ n 3 ,
G 2 =A11 n 1 ⊗ n 1 + A22 n 2 ⊗ n 2 + A33 n 3 ⊗ n 3 ,
G 3 =A12 n 1 ∨ n 2 + A23 n 2 ∨ n 3 + A31 n 3 ∨ n 1 ,
G 4 =A223 n 1 ⊗ n 1 + A231 n 2 ⊗ n 2 + A212 n 3 ⊗ n 3 ,
G 5 =A423 n 1 ⊗ n 1 + A431 n 2 ⊗ n 2 + A412 n 3 ⊗ n 3 ,
(5.135)
G 6 =A211 n 1 ⊗ n 1 + A222 n 2 ⊗ n 2 + A233 n 3 ⊗ n 3 +
A23 A31 n 1 ∨ n 2 + A31 A12 n 2 ∨ n 3 + A12 A23 n 3 ∨ n 1 ,
G 7 =A312 n 1 ∨ n 2 + A323 n 2 ∨ n 3 + A331 n 3 ∨ n 1 ,
G 8 =A33 A12 n 1 ∨ n 2 + A11 A23 n 2 ∨ n 3 + A22 A31 n 3 ∨ n 1 ,
G 9 =A233 A12 n 1 ∨ n 2 + A211 A23 n 2 ∨ n 3 + A222 A31 n 3 ∨ n 1 .

Quasi-crystal classes
A) and q = q (A
In the following, the vectors p = p (A A) are given by (5.86) and the angles φ = φ(AA) and
ψ = ψ(A A) by (5.87), where the standard basis e i is given by (5.78). Moreover, the four tensors D α are
given by (5.130).
G = D2mh ,C2mv , D2m , D2m̄−1h , D2m̂d for m ≥ 2
An irreducible representation is given by (5.131) with

G1 = I , G2 = n ⊗ n , G3 = A , (5.136)

G 4 = |pp|2 (D
D1 cos 2φ + D 2 sin 2φ) , (5.137)
G 5 = |pp|2m (D
D1 cos 2mφ − D 2 sin 2mφ) , (5.138)
G6 = |pp|2m+1 (D
D3 cos(2m + 1)φ − D4 sin(2m + 1)φ , (5.139)
m
G 7 = |qq| (D
D1 cos mψ − D 2 sin mψ) , (5.140)
G 8 = |pp|2m−1 |qq|(D
D3 cos((2m − 1)φ + ψ) − D 4 sin((2m − 1)φ + ψ)) . (5.141)

G = C2mh ,C2m ,C2m̄−1h , S4m̂ for m ≥ 2

A minimal representation is given by (5.121) with (5.123)–(5.124), but here the difference is that each
scalar coefficient fα should be an invariant of A relative to C2mh .
G = D2m+1d ,C2m+1v , D2m+1 for m ≥ 1
5.4. SYMMETRIC TENSOR-VALUED FUNCTIONS OF SYMMETRIC TENSOR 103

An irreducible representation is given by (5.128) with (5.136)–(5.137) and

G 5 = |pp|2m−1 (D
D1 sin(2m − 1)φ + D 2 cos(2m − 1)φ) , (5.142)

G 6 = |pp|2m (D
D3 sin 2mφ + D 4 cos 2mφ) , (5.143)
2m
G 7 = |qq| (D
D1 cos 2mψ − D 2 sin 2mψ) , (5.144)
m
G 8 = |qq| (D
D3 sin mψ + D 4 cos mψ) , (5.145)
G 9 = |qq|m+1 (D
D3 sin(m + 1)ψ − D 4 cos(m + 1)ψ) . (5.146)

G = S4m+2 ,C2m+1 for m ≥ 1

An irreducible representation is given by (5.131) with G 1 , · · ·, G 4 given by (5.136)–(5.137) and

G 5 = A (εε[nn]) − (εε[nn])A
A, (5.147)

G 6 = |pp|2 (D
D1 sin 2φ − D 2 cos 2φ) , (5.148)
G 7 = |qq|m (D
D3 sin mψ + D 4 cos mψ) , (5.149)
m
G 8 = |qq| (D
D3 cos mψ − D 4 sin mψ) . (5.150)

G = Ih , I (icosahedral)
A representation is given by (5.128) with
6
Gk = ∑ (nnα · A[nnα ])k−1 nα ⊗ nα , k = 1, 2, · · · , 6 , (5.151)
α=1

G 7 = G 22 , G8 = G2G3 + G3G2 , G 9 = G 23 . (5.152)

In the above, n α are the six 5-fold axial vectors given by (5.55)–(5.56).
Although it appears to be quite simple, it is not yet known whether or not the above representation is
irreducible.

5.4.3 Continuity and Derivatives

The notion of continuity and differentiability for scalar-valued functions may be extended to the case of
tensor-valued function in a parallel manner.
We say that a tensor-valued function Φ (AA) is continuous at A , if for any given small δ > 0 there is a
small ε > 0 such that
|Φ A0 ) − Φ (A
Φ(A A)| < δ (5.153)
whenever A 0 and A meet (5.96).
The continuity of Φ (A A) at A enables us to approximate Φ (A A0 ) by Φ (A
A) for any A 0 close to A . As
in the case of scalar functions, the notion of differentiability is needed, in order to have a further, better
estimation of Φ (AA0 ). Prior to the definition of this notion, we introduce the symbol o (x) to represent a
symmetry 2nd-order tensor which relies on a known small quantity x and has the property
|oo(x)|
lim = 0. (5.154)
x→0 x
A) is Fréchet-differentiable at A , if there is a given 4th-order
We say that a tensor-valued function Φ (A
Φ/∂A
tensor, denoted ∂Φ A, such that
Φ 0
∂Φ
A0 ) − Φ (A A0 − A |


Φ (A A) = A − A ] + o |A
[A (5.155)
A
∂A
for any two symmetric 2nd-order tensors A 0 and A that are close to each other. The tensor ∂ΦΦ/∂A A, which
is either constant or dependent only on A and fulfills the just-mentioned condition at each A , is known as
A) at A .
the Fréchet derivative of the tensor-valued function f (A
104 CHAPTER 5. TENSOR FUNCTIONS

A) is differentiable at A , then Φ (A
If Φ (A A) is continuous at A . The converse needs not be true, as in the
case of real functions.
Assume that Φ(A A) is Fréchet-differentiable at A. Using (5.154) and replacing A0 in (5.155) with A + εX
X
X
where the number ε may go to vanish but the symmetric 2nd-order tensor need not be so, we derive

Φ (A X ) − Φ (A
A + εX A) ∂ΦΦ
lim = X]
[X (5.156)
ε→0 ε A
∂A
for any symmetric 2nd-order tensor X .
Expression (5.156) in its own right leads to the notion of Gâteaux-differentiability property and Gâteaux-
derivative of the tensor-valued function Φ (A A). Generally, this notion need not imply the notion introduced
before. Note that (5.156) is derived as a consequence of Fréchet-differentiability property.
Given a fixed standard basis e i , a tensor-valued function Φ (AA) may be regarded as six component func-
tions of the six standard components Ai j relative to this basis. In this case, we have the useful expression

Φ ∂Φi j
∂Φ
= ei ⊗ e j ⊗ ek ⊗ el . (5.157)
A
∂A ∂Akl

A) and Φ 2 (A
Some useful rules for derivative are given as follows. First,for any two differentiable Φ 1 (A A),
we have
∂(ΦΦ1 + Φ 2 ) ∂Φ Φ1 ∂Φ Φ2
= + . (5.158)
A
∂A ∂AA A
∂A
A) and for any differentiable tensor-valued
In addition, f or any differentiable scalar-valued function f = f (A
function Φ (AA), we have
∂( f Φ ) Φ
∂Φ ∂f
=f +Φ ⊗ (product rule) . (5.159)
∂AA A
∂A A
∂A
Let Φ (SS ) be a differentiable tensor-valued function of tensor variable S and the latter in turn be a
A). Then, we have
differentiable tensor-valued function of tensor variable A , i.e., S = S [A

Φ(SS ) ∂Φ
∂Φ Φ ∂SS
= ◦ (Leibniz’s chain rule) . (5.160)
A
∂A ∂SS ∂A
A
On the other hand, let Φ 1 (A A) and Φ 2 (A
A) be two differentiable tensor-valued functions of A . Then, the
derivatives of their scalar product and composite product may be calculated by the rules
   
∂(ΦΦ1 Φ2 ) Φ2
∂Φ Φ1
∂Φ
X ] = Φ1
[X X ] + Φ2
[X X] ,
[X (5.161)
A
∂A A
∂A A
∂A
   
Φ1 Φ2 )
∂(Φ Φ2
∂Φ Φ1
∂Φ
X ] = Φ1
[X X] +
[X X ] Φ2 .
[X (5.162)
A
∂A A
∂A A
∂A
The derivative of scalar-valued function f (AA), ∂ f /∂A
A is a tensor-valued function of A . We refer to the
A as the twice derivative of the scalar-valued function f (A
derivative of the tensor-valued function ∂ f /∂A A)
and designate it by ∂2 f /∂A
AA , i.e.,
∂2 f
 
∂ ∂f
≡ . (5.163)
∂AA∂AA ∂A A ∂A A
Below we record some results for derivatives of frequently used isotropic tensor-valued functions.

∂II A
∂A A2
∂A
X] = O ,
[X X] = X ,
[X = AX + X A . (5.164)
A
∂A A
∂A A
∂A
In general, for each positive integer t we have

At
∂A t
= ∑ At−α X A α−1 . (5.165)
A
∂A α=1
5.5. EIGENPROJECTION METHOD 105

If A is non-singular, we have
A−1
∂A
[X A−1 X A −1 .
X ] = −A (5.166)
A
∂A
This may be derived by differentiating A A −1 = I and using (5.162). Further, utilizing Leibniz’s chain rule
(cf. (5.160)), for any positive inter t we have

A−t
∂A A−1 )t ∂A
∂(A A−1
X] =
[X [ X ]] .
[X
A
∂A A−1
∂A A
∂A
From this and (5.165)–(5.166) we derive

A−t
∂A t
X ] = − ∑ A α−t−1 X A −α
[X (5.167)
A
∂A α=1

for each positive integer t.

5.5 Eigenprojection Method

It may be readily understood that both (5.165) and (5.167) are lengthy for a large t. According to Cayley-
Hamilton theorem (cf. (2.81)), either of them may be reduced to an expression which involves the two
simplest powers A and A 2 only and the eigenvalues of A . However, it would be cumbersome for a large
t to derive such an expression directly from either (5.165) or (5.167). Further, for an arbitrary number t
that need not be an integer, either (5.165) or (5.167) no longer valid. To treat this difficult case and vari-
ous cases for general isotropic scalar- and tensor-valued functions and other relevant problems, it would be
helpful to use the so-called eigenprojection method. This method has as its origin the principal axis method
introduced and systematically used by Hill (1968, 1978). Inspired by this earlier method, the eigenprojec-
tion method is introduced by utilizing the eigenprojections instead of the eigenvectors. Here, use of the
eigenprojections would not only retain the simple characteristic properties of symmetric tensors but also
lead to desirable unique, explicit results by means of Sylvester’s formula, thus rendering calculations of
eigenbases unnecessary. Note that the latter are not uniquely determined in a case of repeated eigenvalues.

5.5.1 Characteristic expression of perturbation tensor

Now we study the derivative of Hill’s strain tensors g(AA) defined through a scalar function g(λ) (cf. (2.98)),
which include as particular cases the powers A α for all integers α.
Toward the above goal, the basic idea is to evaluate the values of (5.156) relative to the eigenbasis a i of
A , namely, we set
Φ (AA) = g (AA) , X = a i ⊗ a j + a j ⊗ a i
in (5.156), where a i and a j are any two orthonormal eigenvectors of A . In so doing, the basic procedure is
to use the characteristic expressions of the tensor A and the perturbation tensor

A + ε(aai ⊗ a j + a j ⊗ a i ) ,

as will be shown below.

If the eigenvectors a i and a j are coincident and subordinate to the eigenvalue λi , the treatment is quite
easy. In fact, observing that the tensor variable A and its perturbation A + εaai ⊗ a i have the eigenvectors in
common and using (5.116) and (5.156), we deduce

∂gg A + εaai ⊗ a i ) − g (A
g (A A)
[aai ⊗ a i ]= lim
A
∂A ε→0
 ε 
g(λi + ε) − g(λi )
= lim ai ⊗ ai (5.168)
ε→0 ε
∂g
= ai ⊗ ai .
∂λi
106 CHAPTER 5. TENSOR FUNCTIONS

On the other hand, let λi and λ j be any two eigenvalues of A and the other eigenvalue of A be λk and
let a i , a j , a k be their subordinate orthonormal eigenvectors of A . Then we have

A = λi a i ⊗ a i + λ j a j ⊗ a j + λk a k ⊗ a k , (5.169)
(
A + ε(aai ⊗ a j + a j ⊗ a i ) = A i j + λk a k ⊗ a k ,
(5.170)
A i j = λi a i ⊗ a i + λ j a j ⊗ a j + ε(aai ⊗ a j + a j ⊗ a i ) .
Then, the three eigenvalues of the perturbation tensor variable are given by λk and
¯ i = (λi + λ j + p)/2 , λ¯ j = (λi + λ j − p)/2 ,
(
λ
p (5.171)
p = (λi − λ j )2 + 4ε2 .

Let āai and āa j be any given two orthonormal eigenvectors of the perturbation tensor variable subordinate to
¯ i and λ
λ ¯ j . Then
(
āai ⊗ āai + āa j ⊗ āa j = a i ⊗ a i + a j ⊗ a j ,
(5.172)
¯ i āai ⊗ āai + λ
λ ¯ j āa j ⊗ āa j = A i j .
¯ i 6= λ
From (5.168) we know that λ ¯ j for any ε 6= 0 and then from (5.169) we derive
(
āai ⊗ āai = 1 ¯ j (aai ⊗ a i + a j ⊗ a j )) ,
Ai j − λ
¯ i −λ
λ ¯ j (A
1 ¯ i (aai ⊗ a i + a j ⊗ a j )) , (5.173)
āa j ⊗ āa j = ¯ j −λ
λ
Ai j − λ
¯ i (A

It may be readily understood that the three orthonormal vectors āai , āa j , āak = a k supply an eigenbasis of the
foregoing perturbation tensor. From the above facts and (5.116) we infer
3
A + ε(aai ⊗ a j + a j ⊗ a i ))=
g (A ∑ g(λ¯ r )āar ⊗ āar (5.174)
r=1
¯ i )āai ⊗ āai + g(λ
= g(λ ¯ j )āa j ⊗ āa j + g(λk )aak ⊗ a k .

Then using (5.168), (5.172) and (5.173) we derive

∆gg(A A + ε(aai ⊗ a j + a j ⊗ a i )) − g (A
A, ε)= g (A A)
= g(λ¯ i )āai ⊗ āai + g(λ
¯ j )āa j ⊗ āa j − g(λi )aai ⊗ a i − g(λ j )aa j ⊗ a j (5.175)
= α(aai ⊗ a i + a j ⊗ a j ) + β(aai ⊗ a i − a j ⊗ a j ) + γaai ∨ a j ,

where
¯ i ) − g(λi ) + g(λ
α = g(λ ¯ j ) − g(λ j ) , (5.176)
λi − λ j ¯ ¯ j )) − 1 (g(λi ) − g(λ j )) ,
β= (g(λi ) − g(λ (5.177)
2p 2
ε ¯ ¯ j )) .
γ = (g(λi ) − g(λ (5.178)
p

5.5.2 Derivative in terms of eigenbasis

To evaluate the following value

∂gg ∆gg(A
A, ε)
[aai ⊗ a j + a j ⊗ a i ] = lim ,
A
∂A ε→0 ε
it suffices to calculate the three limits
α β γ
lim , lim , lim .
ε→0 ε ε→0 ε ε→0ε
5.5. EIGENPROJECTION METHOD 107

The particular case when a i and a j coincide has been treated before. For orthonormal a i and a j , two cases
will be discussed below.
First, suppose λi = λ j . Then

p = 2|ε| , ¯ i = λi + |ε| ,
λ ¯ i = λi − |ε| .
λ
Hence, we have
 
α g(λi + |ε|) − g(λi ) g(λi − |ε|) − g(λi )
lim = lim − = 0;
ε→0 ε ε→0 ε ε
β g(λi + |ε|) − g(λi − |ε|)
lim = − lim = 0;
ε→0 ε ε→0 2ε
γ g(λi + |ε|) − g(λi − |ε|
lim = lim = g0 (λi ) .
ε→0 ε ε→0 2|ε|
and therefore (
∂gg 0
A [a i ⊗ a j + a j ⊗ a i ] = g (λi )(a i ⊗ a j + a j ⊗ a i ) ,
∂A
a a
(5.179)
λi = λ j , i 6= j .
Next, let λi 6= λ j . There is no harm in assuming λi > λ j . Then

p = λi − λ j + 2η + o1 (ε2 ) , ¯ i = λi + η + o2 (ε2 ) ,
λ ¯ j = λ j − η + o3 (ε2 ) ,
λ
where
ε2 ok (ε2 )
η= , lim = 0, k = 1, 2, 3.
λi − λ j ε→0 ε2

Utilizing the above facts we infer

 r
α 1 g(λi + η) − g(λi ) g(λ j − η) − g(λ j ) η
lim = lim + = 0;
ε→0 ε 2 η→0 η η λi − λ j !
β (λi − λ j )(g(λ¯ i ) − g(λ
¯ j ) − g(λi ) + g(λ j )) (λi − λ j − p)(g(λi ) − g(λ j ))
lim = lim +
ε→0 ε ε→0 2pε 2pε
 r
g(λi + η) − g(λi ) g(λi ) − g(λi − η) η
= lim + = 0;
η→0 2η 2η λi − λ j
γ limε→0 (g(λ ¯ i ) − g(λ
¯ j )) g(λi ) − g(λ j )
lim = = ,
ε→0 ε limε→0 p λi − λ j
and hence
g(λi )−g(λ j )
(
∂gg
A [a i ⊗ a j + a j ⊗ a i ] = (aai ⊗ a j + a j ⊗ a i ) ,
∂A
a λi −λ j (5.180)
λi 6= λ j , i 6= j .
Results in component form were derived by Hill (1968, 1978), and, later, rigorous treatment and further
developments were given by Carlson and Hoger (1986), Scheidler (1991a, b), and Man and Guo (1993).
Besides, results were also derived by Wang and Duan (1991) by means of alternative procedures.

5.5.3 Derivatives in terms of eigenprojections

Expressions (5.168)3 and (5.179)–(5.180) may be combined into the following unified expression
∂gg g(λi ) − g(λ j )
[aai ⊗ a j + a j ⊗ a i ] = (aai ⊗ a j + a j ⊗ a i ), i, j = 1, 2, 3 , (5.181)
A
∂A λi − λ j
g(λ )−g(λ )
with the limiting process: limλi →λ j λi i −λ j j = g0 (λi ) for λi = λ j .
˜ σ , σ = 1, · · · , m be the distinct eigenvalues of A and A 1 , · · · , A m the corresponding eigenprojections
Let λ
of A . Then, using the expression of symmetric tensor X relative to the eigenbasis a i of A and the equality
(aai ⊗ a i )X
X (aa j ⊗ a j ) = Xi j a i ⊗ a j , Xi j = X ji = a i · X [aa j ] ,
108 CHAPTER 5. TENSOR FUNCTIONS

we derive
3
∂gg g(λi ) − g(λ j )
X ]= ∑
[X (aai ⊗ a i )X
X (aa j ⊗ a j )
A
∂A i, j=1 λi − λ j
!
3 3
g(λi ) − g(λ j )
=∑ ∑ λi − λ j (aai ⊗ a i )XX (aa j ⊗ a j )
i=1 j=1 !
3 m
g(λi ) − g(λ˜ τ)
=∑ ∑ λ − λ˜ (aai ⊗ a i )XX A τ
i=1 τ=1 i τ !
m 3
g(λi ) − g(λ˜ τ) m ˜ σ ) − g(λ
g(λ ˜ τ)
= ∑ ∑ λ − λ˜ (aai ⊗ a i )XX A τ = ∑ λ˜ − λ˜ A σ X A τ .
τ=1 i=1 i τ σ,τ=1 σ τ

Hence, we arrive at the following result:

m
∂gg g(λσ ) − g(λτ )
X] = ∑
[X AσX Aτ . (5.182)
A
∂A σ,τ=1 λσ − λτ

In the above and below, we no longer use λ ˜ σ but use λσ to designate the distinct eigenvalues of A and we
use A 1 , · · · , A m to denote the subordinate eigenprojections of A .
Thus, substituting Sylvester’s formula (2.111) into (5.183), we may obtain an explicit expression of the
form
m−1
∂gg
X ] = ∑ ρrs A r X A s .
[X (5.183)
A
∂A r,s=0

In the above, each coefficient ρrs = ρsr is a symmetric function of the distinct eigenvalues of A . Explicit
expressions for these coefficients may be found in Xiao (1995a).
A) and Φ (A
The above procedure may be used to treat isotropic scalar- and tensor-valued function f (A A).
A), we have (2.71) and its derivative
In fact, for isotropic scalar-valued function f (A
m
∂f ∂ĥ
= ∑ Aσ , (5.184)
A σ=1 ∂λσ
∂A

as well as its twice derivative

∂ĥ ∂ĥ
∂2 f m −
X ] = ∑ ∂λσ ∂λτ A σ X A τ .
[X (5.185)
σ,τ=1 λσ − λτ
A∂A
∂A A

Results for the derivative of isotropic tensor-valued function may be derived by the same procedure,
but not as simple as above.

5.5.4 Summary for the main ideas and procedures

To summarize, the main idea of the procedure elaborated above is as follows:

(i) The eigenbasis of a basic symmetric tensor variable A , such as the Cauchy-Green deformation tensor
etc., is chosen as a standard basis of vector spaces;

(ii) the results for various relevant problems associated with A , such as the derivatives of isotropic scalar-
and tensor-valued functions of A etc., are derived and presented by means of the eigenbasis and the
simple characteristic properties of A , in which the characteristic expressions of A and A + εaai ∨ a j ,
i.e., (5.169)–(5.173) may play a basic role;

(iii) the final results are expressed in terms of the eigenprojections of A and then Sylvester’s formula
(2.111) is meant;
5.6. CONTINUOUSLY DIFFERENTIABLE REPRESENTATION 109

(iv) In some cases, e.g., in studying issues concerning finite strains and large rotations, such as work-
conjugacy notions either in a classical sense or in an extended sense, it is required to work out the
inverse tensors of certain relevant 4th-order tensors, such as the derivative ∂gg/∂A
A of the finite strain
A), etc. Here, eigenprojection method proves simple and powerful. Indeed, the expression
tensor g (A
(5.183) in terms of eigenprojections is just the characteristic expression of the derivative ∂gg/∂A
A. We
have

∂gg
[aai ⊗ a j + a j ⊗ a i ] = ρi j (aai ⊗ a j + a j ⊗ a i ) (no summation) (5.186)
A
∂A
for any eigenvectors a i and a j of A . Thus, its inverse tensor may be obtained directly by applying
(3.94) in §3.10.2. Details in this respect may be found in Xiao (1995a), Xiao, Bruhns and Meyers
(1997, 1998d), and Bruhns, Meyers and Xiao (2002).
As indicated before, earlier Hill (1968, 1978) (see also, e.g., Ogden 1984) introduced and systematically
and successfully used the well-known principal axis method. Later, Carlson and Hoger (1986), Simo and
Taylor (1991), Miehe (1993, 1994, 1997), and Šilhaý (1997), et al. used the characteristic expression of
the tensor variable A to treat particular or general isotropic tensor-valued function Φ (A A). On the other
hand, eigenprojections play a basic role in the study of the characteristic structure of the elasticity tensor
by Rychlewski (1984a, b, 1995, 2000).
Most recently, the foregoing ideas and procedures have been developed and systematically used to
treat various problems in finite strains and large rotations of continuous deformable bodies and in rational,
consistent modelling of inelastic behaviour at finite deformations, including the derivative of Hill’s strain
tensors and their work-conjugate stresses; a natural extension of Hill’s work-conjugate notion; the finding
and the proof of the unique, intrinsic relationship between Hencky’s logarithmic strain tensor and the
stretching tensor via corotational frames and the introduction of the logarithmic rate; a general study of
corotational rates and their defining spin tensors; a general study of non-corotational rates of Oldroyd’s
type; the finding and the uniqueness proof of the exactly integrable Eulerian elastic rate equations in linear
form and in general form , etc. For details, refer to Xiao (1995a), Xiao, Bruhns and Meyers (1997, 1998a,
b, d, 1999a, 2000d, 2002), Bruhns, Xiao and Meyers (1999b, 2002), Meyers (1999a, b), Meyers, Schiesse
and Bruhns (2000), Bruhns, Xiao and Meyers (1999), and Bruhns, Meyers and Xiao (2002).
The eigenprojection method described above may be extended to cover non-symmetric 2nd order ten-
sor and related isotropic functions. Systematic development and detailed account in a general sense are
presented by Meyers (1999a). Some recent results for tensor power series of a non-symmetric tensor are
derived by Itskov (2002) and Itskov and Aksel (2002).

5.6 Continuously Differentiable Representation

5.6.1 Motivation
Although the representation formula (5.118) has been established for nearly five decades, until very recently
some basic issues concerning it remain under investigation, as explained below.
First, given sufficient experimental data for the argument tensor A and the value tensor T , we would
like to determine the unknown representative coefficients αi in (5.118). In this case, we need to relate each
of the latter with the former. This could be done by considering the forms of Eq. (5.118) in the three
eigendirections of the argument tensor A, i.e.,

tk = α0 + α1 ak + α2 a2k , k = 1, 2, 3,

where tk and ak are eigenvalues of the value tensor T and the argument tensor A , respectively. The above
three equations provide explicit expressions for the three αi in terms of the eigenvalues tk and ak , i.e.,

αi = gi (a1 , a2 , a3 ,t1 ,t2 ,t3 ), i = 0, 1, 2 . (5.187)

Such expressions and other equivalent forms were derived and discussed by many researchers from differ-
ent contexts, e.g., Fitzjerald (1981), Betten (1984, 1987, 1993, 1998), Ting (1985), and Man (1994, 1995),
110 CHAPTER 5. TENSOR FUNCTIONS

et al. Another approach based upon the spectral decomposition of the argument tensor A was proposed
and used by Simo and Taylor (1991), Miehe (1993, 1994), and Šilhavý (1997), et al., as mentioned be-
fore. These expressions, however, require special treatment for the singly and twice coalescent cases of the
eigenvalues (a1 , a2 , a3 ) of the argument tensor A . In these cases, one has to either resort to a limiting pro-
cess resulting from the assumption of the continuous differentiability properties (see the relevant account
below) of these expressions or use the alternative expressions

A) = β0 I + β1 A
S = Φ (A

for the singly coalescent case and

A ) = γ0 I
S = Φ (A
for the twice coalescent case. In the above, the coefficients γ0 , β1 and β0 assume forms that are different
from those of their counterparts in (5.118). Moreover, it is evident that formulas like (5.187) entail the
computations of the eigenvalues of both the argument and value tensors A and T .
Second, given the smoothness (continuity and differentiability) property of the isotropic tensor func-
tion Φ (AA), we would like to know what kind of smoothness property each representative coefficient αi
possesses. This issue was studied first by Serrin (1959). The result was summarized later in the monograph
by Truesdell and Noll (1992), and further investigated by Ball (1984), et al. An extensive and systematic
treatment has not been achieved until the recent work by Man (1994, 1995). The main result given by Man
A) is of class Cr+2 (r = 0, 1, 2, · · ·), then each αi given by (5.187) is of class Cr . In particular,
is that if Φ (A
A) is twice continuously differentiable, then each αi given by (5.187) is continuous. Note that the
if Φ (A
representation (5.118) accompanies the loss of smoothness from Cr+2 to Cr .
Third, according to the representation formula (5.118), it has been known that even the continuous
differentiability property of Φ (AA) can not ensure the continuity property of each αi and hence the repre-
sentation (5.118), let alone the differentiability property of the latter. It seems that this is an undesirable
property. In fact, it is desirable that the value tensor S = Φ (A A) as the stress tensor be differentiable, since
the differentiation of the stress tensor is inevitable in the Cauchy’s equations of motion and equilibrium in
terms of the displacement vector. Although the C∞ smoothness of the isotropic tensor function Φ (A A) ac-
companies the C∞ smoothness of each representative coefficient αi , the twice continuous differentiability
property of Φ (A A) can ensure only the continuity property of each αi . According to Theorem 1.1 estab-
lished by Man (1995), as a sufficient condition the isotropic tensor function Φ (A A) should be at least triply
continuously differentiable in order that each αi and hence the representation (5.118) is continuously dif-
ferentiable. In many cases, the assumption of triply continuous differentiability property and even stronger
differentiability property for S = Φ (A A) might not be necessary and realistic. The representation (5.118)
may result in the following inconsistency issue: it may represent a continuous (continuously differentiable)
isotropic response function S = Φ (A A) as a discontinuous (non-differentiable) function. In these cases, the
representation (5.118) could not represent the continuity and differentiability properties of the response
function Φ (A A), and hence it may not be regarded as a reasonable, faithful representation of Φ (A A) in a full
sense. It seems that a natural, desired representation for the isotropic tensor function Φ (A A) should be like
this: each representative term inherits the continuity and differentiability properties of the isotropic tensor
function Φ (A A). This requirement will be specified slightly later.
Since the standard representation (5.118) fails to fulfill the above-mentioned requirement and, more-
over, it is unsatisfactory due to the other issues indicated before, it may be necessary to search for a new
representation. In the succeeding sections, we shall show that a satisfactory representation can indeed be
found, which may supply new insights into the above basic issues.

5.6.2 A new notion of representation

The usual notion of representation for form-invariant tensor functions is defined in the sense of algebraic
point of view, as given before, which has nothing to do with the continuity and differentiability properties
of the tensor functions of interest. This situation results in some undesirable consequences, as discussed
in the last section. A further development toward more satisfactory representations may need a represen-
tation notion taking continuity and differentiability into account, which will be given below. Here, we are
concerned with a general scope covering both isotropic and anisotropic tensor functions.
5.6. CONTINUOUSLY DIFFERENTIABLE REPRESENTATION 111

Let G 1 (A A), · · · , G s (A
A), G 2 (A A) form a generating set for the tensor functions Φ (A
A) that are form-invariant
under a given material symmetry group G (see (5.110)). Then we have the representation (5.119), where
A) is an invariant relative to the group G (see (5.65)). We introduce the following new notion of
each fi (A
representation for isotropic and anisotropic tensor-valued functions.
We say that a representation (5.119) for the tensor functions Φ (A A) that are form-invariant under a
given material symmetry group G is said to be a continuous representation, if for every form-invariant,
continuous tensor function Φ (A A), each term fi (A
A)GGi (A
A) in the representation (5.119) is continuous.
Further, way say that a representation (5.119) for the tensor functions Φ (A A) that are form-invariant
under a given material symmetry group G is said to be a continuously differentiable representation, if it
is a continuous representation and, moreover, for every form-invariant, continuously differentiable tensor
function Φ (AA), each term fi (A A)GGi (A
A) in the representation (5.119) is also continuously differentiable.
It should be pointed out that a representation with a generating set composed of C∞ smooth generators,
e.g., polynomial generators, need not be a continuous representation, let alone a continuously differentiable
representation. In fact, the standard representation formula (5.118) for isotropic tensor functions is not a
continuous representation, as indicated before. In classical invariant theory, “polynomial” representations
derived from integrity bases were exclusively taken into account, e.g., see, Spencer (1971), Smith (1994),
et al. Each such representation is generated by a set of polynomial tensor generators and expresses each
form-invariant polynomial tensor functions as a linear combination of the polynomial tensor generators
of this set with each representative coefficient a polynomial invariant. However, generally such a poly-
nomial representation need not be a continuous representation. The main reason is that, in contrast with
the foregoing definitions, the definition of a “polynomial representation” is concerned with the polynomial
tensor functions only, the latter being the simplest class of analytic (C∞ ) tensor functions. Indeed, the
representation (5.118) may be regarded to be a “polynomial” representation that can be derived from the
integrity basis {trA A2 , trA
A, trA A3 }, but even it is not a continuous representation for isotropic tensor functions,
as mentioned before.
Thus, if a classical result for representations of isotropic and anisotropic functions fails the new require-
ments, it seems that we need to derive an alternative representation in the sense of the new representation
notion introduced above. Here, our attention is directed to the isotropic tensor-valued functions.

5.6.3 An orthogonalized representation

A) we present the following alternative representation
For the isotropic tensor-valued function S = Φ (A

A) = c0 I + c1 Ā
S = Φ (A A + c2 G (A
A) , (5.188)

where
A|2 A 2 − Ā A 2 Ā


A) = |Ā
G (A A A (5.189)
and each ci is an isotropic invariant of A .
A, G (A
The three specified isotropic tensor-valued functions I , Ã A) given above form an orthogonalized
A), which may be derived from the usual generating
generating set for isotropic tensor-valued functions Φ (A
set formed by I , A and A 2 by means of Schmidt’s orthogonalization procedure. We have

I A = 0,
Ā

A|2 (II A 2 ) − Ā A 2 (II

I A) = |Ā
G (A A A) = 0 ,
Ā

A|2 (Ā A 2 ) − Ā A 2 (II A2 ) = 0 .

A
Ā A) = |Ā
G (A A A Ā
The above orthogonality property enables us to derive in a straightforward manner the following tenso-
rial interpolation expressions for the representation coefficients ci in (5.188).
S I
c0 = , (5.190)
3
S Ā A
c1 = 2
, (5.191)
|Ā |
A
112 CHAPTER 5. TENSOR FUNCTIONS

S G (AA)
c2 = . (5.192)
|G A)|2
G(A
Accordingly, from these and Eq. (5.188) we obtain

S I S Ā A S G (AA)
A) =
S = Φ (A I+ A+
Ā A) .
G (A (5.193)
3 A|2
|Ā |G A)|2
G(A

As a result, if (5.188) describe a stress-deformation relation, then its representative coefficients ci may be
determined by means of experimental data for the deformation (deformation rate) tensor A and the stress
tensor S through expressions (5.190)–(5.192). We would like to point out that, unlike the expressions of
the form (5.187), expressions (5.190)–(5.192) do not involve the eigenvalues of both A and S . Given data
for both A and S in any chosen coordinate system, expressions (5.190)–(5.192) involve only multiplicative
calculations concerning the tensors ĀA, A 2 and S .
The representation formula (5.193) assumes a unified form for the different cases of coalescence of the
eigenvalues of A. Indeed, we have (5.193) for the case when the three eigenvalues of A are distinct to each
other, and
S I S Ā A
S = Φ(A A) = I+ A
Ā (5.194)
3 A|2
|Ā
for the case when the three eigenvalues of A are singly coalescent, say, a1 6= a2 = a3 , and

S I
A) =
S = Φ (A I (5.195)
3
for the case when the three eigenvalues of A are twice coalescent, i.e., a1 = a2 = a3 . Note that the coeffi-
cients in the last two expressions have the same forms as those of their counterparts in Eq. (5.193). Eqs.
(5.194) and (5.195) may be obtained by setting the last term and the last two terms in Eq. (5.193) to be
zero, respectively. This is exactly in accordance with the respective continuity properties of the last two
terms. It is demonstrate in Xiao, Bruhns and Meyers (2002) that the representation (5.188) is a continu-
ous representation and that the last term and the last two terms in Eq. (5.188) or (5.193) are indeed zero,
respectively, whenever the three eigenvalues of A are singly and twice coalescent, respectively. Evidently,
A vanishes whenever the eigenvalues of A becomes twice coalescent. Moreover, we have
the tensor Ā

1 2
|G A)|2 = |Ā
G(A A| (a1 − a2 )2 (a2 − a3 )2 (a3 − a1 )2 .
3

Hence, the tensor G (AA) vanishes whenever the eigenvalues ai of A becomes singly and twice coalescent,
separately.
In addition to the above remarkable features, it is further demonstrated that the representation (5.188)
or (5.193) is a continuously differentiable representation for isotropic tensor-valued tensors Φ (A A). For
details, refer to Xiao, Bruhns and Meyers (2002). In this reference, the above results is used to derive a
A).
direct, unified formula for calculating any given Hill’s strain tensor g (A

5.7 On General Isotropic and Anisotropic Functions

5.7.1 Functional bases, generating sets and irreducibility
In the preceding sections, we have discussed two particular yet important types of isotropic and anisotropic
constitutive tensor functions: scalar- and symmetric tensor-valued functions of one symmetric 2nd-order
tensor variable. According to the material symmetry principle, however, isotropic and anisotropic consti-
tutive tensor functions of any finite number of vectors and 2nd-order tensors should be treated in a general
sense, as introduced in §5.2.5. Extending the notion of functional basis and generating set introduced in
§§5.3-5.4, the main issue in this general respect may be formulated as follows.
5.7. ON GENERAL ISOTROPIC AND ANISOTROPIC FUNCTIONS 113

(i) For the general scalar-valued function ξ(uu1 , · · · , A c ) of vectors and 2nd-order tensors that
is invariant under the invariance condition (5.61), find out a finite number of specified scalar
functions, say,

I1 = I1 (uu1 , · · · , u a ; A 1 , · · · , A b ), · · · , I p = I p (uu1 , · · · , u a ; A 1 , · · · , A b ),

each of which meets the invariance condition (5.61), such that

ξ(uu1 , · · · , u a ; A 1 , · · · , A b ) = f (I1 , · · · , I p ) . (5.196)

We say that such p specified scalar functions meeting (5.61) and the above requirement form
a functional basis of the general scalar-valued function ξ(uu1 , · · · , A c ) of vectors and 2nd-order
tensors that is invariant under the invariance condition (5.61).
(ii) For the general vector- or tensor-valued function φ (uu1 , · · · , A c ) of vectors and 2nd-order
tensors that is form-invariant under the invariance condition (5.62 and 5.63), find out a finite
number of specified vector- or tensor-valued functions, say,

G1 = G1 (uu1 , · · · , ua ; A1 , · · · , Ab ), · · · , Gq = Gq (uu1 , · · · , ua ; A1 , · · · , Ab ),

each of which meets the form-invariance condition (5.62) or (5.63), such that

φ (uu1 , · · · , u a ; A 1 , · · · , A b ) = f1 G 1 + · · · + fq G q , (5.197)

where each coefficient fα = fα (uu1 , · · · , u a ; A 1 , · · · , A b ) is invariant relative to G , i.e., meets

the condition (5.61). We say that such q specified vector- or tensor-valued functions meeting
(5.62) or (5.63) and the above requirement form a generating set of the general vector- or
tensor-valued function φ (uu1 , · · · , u a ; A 1 , · · · , A b ) of vectors and 2nd-order tensors that is invari-
ant under the form-invariance condition (5.62) or (5.63).
(iii) To either the representation (5.196) or the representation (5.197) as compact as possible,
we require that both the functional basis and the generating set therein should be irreducible
in the following sense: Any subset of either of them is non longer a functional basis or a
generating set. Further, if possible, we would find irreducible functional bases and generating
sets each of which includes as fewer as possible elements.

The above general respects are concerned with a large variety of topics for isotropic and anisotropic
functions of vectors and 2nd-order tensors relative to various kinds of material symmetry groups as listed
in §§5.2.2-5.2.4. These topics have been extensively investigated in the past decades and many results have
been obtained. A biefe summary is given below.
Earlier, attention was focused on irreducible polynomial representations mainly for scalar-valued func-
tions and for vector-valued and tensor-valued functions in some cases; see, e.g., the monographs by Trues-
dell and Noll (1992), Spencer (1971), Kiral and Eringen (1990), Betten (1993) and Smith (1994) for many
related results. These results were closely related to the classical theory of algebraic invariants intensively
studied by nineteenth century mathematicians and derived in a restrictive sense merely for polynomial
functions of components of vector variables and tensor variables.
Irreducible non-polynomial representations in a general sense were considered later for isotropic func-
tions of vectors and 2nd-order tensors by Wang (1970) and Smith (1971). These results were sharpened by
Boehler (1977), and their irreducibility was proved by Pennisi and Trovato (1987).
Isotropic functions are related to particular yet widely considered solid materials of the highest sym-
metry. It has played a basic role in the subsequent development in studying anisotropic functions, as will
be explained below.
Unlike isotropic functions, anisotropic functions are distinguished by many different kinds of material
symmetry groups, such as crystal and quasicrystal classes, etc. Usually, they should be treated one by one
by means of different procedures. In so doing, the mathematical problems to be treated might be formidably
enormous. Indeed, there are 5 clasees of transversely isotropic solids, 32 classes of crystalline solids, and
infinitely many classes of quasicrystalline solids, as shown in §§5.2.2-5.2.4.
114 CHAPTER 5. TENSOR FUNCTIONS

5.7.2 Isotropic extension method via structural tensors

There emerged a seminal idea of treating various types of anisotropic functions. This idea was originated
from Lokhin and Sedov (1963). Later, it was rediscovered and developed by Boehler (1978, 1979) and
precised by Liu (1982). Further developments have been made, e.g., by Betten (1987, 1988, 1993, 1998),
Boehler (1987), Zhang (1991), Zhang and Rychlewski (1990), Zheng and Spencer (1993), et al. Extension
of this idea has been made to cover non-orthogonal material symmetry groups by Rychlewski (1988) and
Svendsen (1994). Most recent applications may be found, e.g., in Menzel and Steinmann (2003) and
Svendsen and Reese (2003).
According to the aforementioned idea, one or more specific tensors called structural tensors, say,
M 1 , · · · , M t , are introduced to characterize and represent the material symmetry group G . Namely, for
a material symmetry group G , the tensors M 1 , · · ·, M t are chosen such that

Q ∈ G ⇐⇒ (Q
Q ? M 1 = M 1 , · · · , Q ? M t = M t , Q ∈ Orth) , (5.198)

namely,
Q ∈ Orth | Q ? M 1 = M 1 , · · · , Q ? M t = M t } = G .
{Q (5.199)
We say that the subset of the full orthogonal group Orth given by the left-hand side of (5.202) is the
symmetry group of tensors M 1 , · · ·, M t . In particular, when only one tensor M is involved, this gives
the symmetry group of tensor M . Either of the above two expressions means that tensors M 1 , · · ·, M t are
structural tensors of a given material symmetry group G , whenever the intersection of the symmetry groups
of these tensors exactly coincides with G .
Hence, if M 1 , · · ·, M t are structural tensors of the material symmetry group G , then an anisotropic
function ϕ(uu1 , · · · , A c ) of vectors and 2nd-order tensors relative to the group G , as defined in §5.2.5, may
be given by
ϕ(uu1 , · · · , u a ; A 1 , · · · , A b ) = ϕ̄(uu1 , · · · , u a ; A 1 , · · · , A b ; M 1 , · · · , M t ) . (5.200)
In the above, the latter tensor function ϕ̄ is an extended isotropic function with the structural tensors M 1 ,
· · ·, M t as additional variables. Thus, once the structural tensors are determined, the relationship (5.200)
reduces an anisotropic function to an extended isotropic function with more vector and tensor variables.
It is expected that representations for the extended isotropic function may be derived by using the much
well-known results for isotropic functions by Wang (1970) and Smith (1971). Here, a perhaps essential
point is brought to attention: Till now results are available only for conventional isotropic functions rely-
ing merely on vector variables and 2nd-order tensor variables. Little has been known for unconventional
isotropic functions involving at least one tensor variable of order higher than two. Even if the latter may be
shown to be tractable in future, it may be expected that complete functional bases and generating sets for
them would be extremly large and unduly complicated. Indeed, it appears that that is the case even for one
of the simplest examples, i.e., isotropic invariants of the 4th-order elastic tensor. In this respect, refer to,
e.g., Betten and Helisch (192), Boehler and Onat (1994), and Xiao (1997c) for details.
Nevertheless, for certain simple types of material symmetry groups G , including the cylindrical groups
and the triclinic, monoclinic, rhombic crystal classes, each structural tensor M α may chosen to be either
a vector or a 2nd-order tensor. Then, the foregoing extended isotropic function with additional tensor
variables, as shown by (5.200), is indeed an isotropic function of vectors and 2nd-order tensors, which
have been treated by Wang (1970) and Smith (1971). For any other material symmetry group G , however,
it is known that at least one structural tensor has to be of order higher than two. The main reason is as
follows.
For a vector v 6= 0, we have
{QQ ∈ Orth | Q ? v = v} = C∞v .
In the above, C∞v is the cylindrical group given by (5.5) and here the unit vector n is in the direction of v .
On the other hand, for a 2nd-order symmetric tensor S , we have

D2h , S = ∑3i=1 λi n i ⊗ n i , λ1 6= λ2 6= λ3 6= λ1 ,




{Q
Q ∈ Orth | Q ? S = S } = D∞h , S = λ1 I + (λ2 − λ1 )nn ⊗ n , λ1 6= λ2 ,


Orth, S = λII ,

5.7. ON GENERAL ISOTROPIC AND ANISOTROPIC FUNCTIONS 115

Note that the above three cases correspond with the cases when the three eigenvalues of S are distinct,
singly coalescent, and doubly coalescent. In the above, D2h is the maximal rhombic (orthotropy) group
given by (5.13), and D∞h is the maximal cylindrical group given by (5.3). Moreover, we have

{Q
Q ∈ Orth | Q ?W
W = W } = C∞h

for a 2nd-order skewsymmetric tensor W 6= O . In the above, C∞h is the cylindrical group given by (4.6)
and here the unit vector is in the direction of the associated axial vector ω of the skewsymmetric tensor W .
The above results provide the symmetry group of vector v , symmetric 2nd-order tensor tensor S ,
skewsymmetric 2nd-order tensor W , respectively.
M 1 , · · ·, M t ) be any finite set of vectors and 2nd-order tensors given by2
Now let (M

M 1 , · · · , M t ) = (vv1 , · · · , v n , S 1 , · · · , S m ,W
(M W 1 , · · · ,W
W l ),

where each v α is a non-zero vector, each S β is a non-trivial symmetric 2nd-order tensor, and each W θ is a
non-zero skewsymmetric 2nd-order tensor, i.e. 3 ,

v α 6= 0 , S β 6= λβ I , W θ 6= O , α = 1, · · · , n, β = 1, · · · , m, θ = 1, · · · , l.

Then, from the foregoing results for the symmetry groups of vector and symmetric and skewsymmetric
2nd-order tensor, we infer that the symmetry group of any finite number of vectors and 2nd-order tensors is
nothing else but one of the cylindrical groups and the triclinic, monoclinic, rhombic groups. Note that the
former is simply intersection of some rhombic (orthotropy) groups D2h and some cylindrical groups C∞v ,
C∞h , D∞h with different 2-fold symmetry axes n i and different ∞-fold symmetry axes n . Thus, any other
material symmetry group G except the cylindrical groups and the triclinic, monoclinic, rhombic groups
could not be characterized and represented by a set of vectors and 2nd-order tensors, no matter how many
they may be. Namely, for each just-mentioned not so simple material symmetry group G , we could not
find out a set of vectors and 2nd-order tensors, (M M 1 , · · · , M t ), such that the condition (5.198) or (5.202) is
fulfilled. As a result, at least one structural tensor M α of order higher than two has to be introduced.
Consequently, the foregoing extended isotropic function with additional tensor variables, as shown by
(5.200), is no loner an isotropic function of vectors and 2nd-order tensors, but relies on at least one tensor
variable of order higher than two. It seems that little has been known for isotropic functions with tensor
variable(s) of order higher than two, as remarked in a review article by Rychlewski and Zhang (1991).
Utilizing the idea of the structural tensors described above, results for certain simple types of anisotropic
functions of vectors and 2nd-order tensors have been derived in certain cases, including those relative to
the cylindrical groups and the triclinic, monoclinic, rhombic crystal classes; see, e.g., Boehler (1978, 1979,
1987), Liu (1982), Betten (1987, 1988, 1993), Zheng (1993), et al.
The above classical or traditional methods lead to resolutions of a number of important problems.
Here it is not our intention to reproduce the huge body of literature in this respect and other respects. For
detail, refer to the monographs mentioned before and the reviews articles by Rychlewski and Zhang (1991),
Zheng (1993), and Betten (1998), as well as the relevant references therein. However, a large variety of
topics for anisotropic functions in a general sense remain to be investigated until very recently, including
those relative to trigonal, tetragonal, hexagonal, cubic crystal classes (totally 24 crystal classes) and all the
quasicrystal classes.

5.7.3 New isotropic extension method and unified theory

Subsequent developments are made for the above-mentioned isotropic entension method. First, the new
idea is to replace each structural tensor M α in (5.196) with a simple, well-chosen form-invariant polynomial
vector- and/or 2nd-order tensor-valued function of either vector variable u i or 2nd-order tensor variable A k .
Then we obtain an isotropic extension function whose variables are indeed vectors and 2nd-order tensors,
2 Note that any given 2nd-order tensor may be represented by its symmetric part and skewsymmetric part in an equivalent and

unique manner according to the additive decomposition (2.142)-(2.143).

3 The trivial case when v = 0, S = λ I , W = O leads to the fact that the material symmetry group G in (5.198) has to be the
α β β θ
isotropy group Orth.
116 CHAPTER 5. TENSOR FUNCTIONS

and, therefore, the much well-known results by Wang (1970) and Smith (1971) can be directly used to
derive desired results for various kinds of anisotropic functions. This idea was originated from a summary
in Xiao and Guo (1993) and later shown to be applicable to all kinds of material symmetry groups.
Specifically, let
X = (uu1 , · · · , u a ; A 1 , · · · , A b )
for the sake of simplicity. Then, the foregoing general anisotropic tensor function will be designated by
ϕ(X). According to the new isotropic extension method, introducing a set of well-chosen vector-valued and
2nd-order tensor-valued polynomial functions that are form-invariant under the material symmetry group
G , i.e., S(X), we may establish a new isotropic extension relationship for the anisotropic function ϕ(X)
relative to G as follows:
ϕ(X) = ϕ̂(X, S(X)). (5.201)
In the above, the latter tensor function ϕ̂ is indeed an extended isotropic function of the augemented set
of vector variables and 2nd-order tensor variables, i.e., (X, S(X)). The set of additional vector variables
and 2nd-order tensor variables, i.e., S(X), has been determined and presented for each crystal class and for
each quasicrystal class. Details may be found in Xiao (1996b, 1997b).
For instance, according to the new isotropic extension method, each anisotropic function ϕ(A A) of a
symmetric tensor variable A that is invariant or form-invariant under the cubic crystal group Oh (see (5.35)
with (5.40)) may be converted to
A) = ϕ̂(A
ϕ(A A, O h [A A2 ]) ,
A], O h [A (5.202)
with
3
Oh = ∑ ni ⊗ ni ⊗ ni ⊗ ni . (5.203)
i=1
The latter tensor function in (5.202) is an isotropic function of three symmetric 2nd-order tensors. Thus,
from this extended isotropic function it may be a straightforward matter to derive a complete representation
for the aniotropic function ϕ(A A) relative to the cubic group Oh applying the results by Wang (1970) and
Smith (1971). For any other crystal or quasicrystal class, the same procedure may be used and the only
difference is that the additional 2nd-order tensor or vector variables will be different for different kinds of
material symmetry groups.
The results presented in §§5.3-5.4 are just derived by applying the above idea and procedure. Details
may be found in Bruhns, Xiao and Meyers (1999a).
Next, it is demonstrated in Xiao (1996c, 2000) that representations for arbitrary-order tensor-valued
anisotropic or isotropic functions of any finite number of vector and 2nd-order tensor variables are deriv-
able simply from combinations of those for the same type of anisotropic or isotropic functions of not more
than four variables. Thus, such decomposition theorems given in the just-mentioned references reduce rep-
resentative problems involving any finite number of vector and 2nd-order tensor variable to those involving
not more than four vector and 2nd-order tensor variables.
Furthermore, a unified procedure for constructing irreducible non-polynomial representations for scalar-
, vector-, and 2nd-order tensor-valued isotropic and anisotropic functions is designed in Xiao (1998b) and
Xiao, Bruhns and Meyers (1999c, 2000a, b). Usually, the just-mentioned three kinds of tensor functions
have to be treated by means of different procedures, separately.
With the above new ideas, systematic results for scalar-, vector-, and 2nd-order tensor-valued anisotropic
functions of vectors and 2nd-order tensors are derived in a general sense by these authors. Details may be
found in Xiao (1998b) and Xiao, Bruhns and Meyers (1999c, 2000a, b).
In recent years, 4th-order tensor-valued isotropic functions of a symmetric 2nd-order tensor or a non-
symmetric 2nd-order tensor have been taken into consideration, which are useful in formulating hypoelastic
constitutive relations in general form (see Truesdell and Noll 1965) and in simulating the texture induced
elastic anisotropy (see Böhlke 2001, Böhlke and Beteram 2001a). Results in this respect are given by
Böhlke (2001) and Böhlke and Beteram (2001a, b).
A most recent development is to consider minimal representations for vector- and tensor-valued isotropic
and anisotropic functions in a suitable sense, and, on the other hand, to study further requirements resulting
from continuity and differentiability properties. These further notions are proposed in most recent works
by these authors; see Xiao, Bruhns and Meyers (2000c, 2003). It appears that studies in these respects
5.7. ON GENERAL ISOTROPIC AND ANISOTROPIC FUNCTIONS 117

are merely at the initial stage. Results in some particular yet perhaps significant cases are presented in the
just-mentioned references and Xiao, Bruhns and Meyers (1998c). These results are concerned with mini-
nal representations for symmetric tensor-valued anisotropic constitutive functions of one symmetric tensor
relative to certain material symmetry groups, as presented in §5.4,and for isotropic symmetric tensor-
valued functions of one skewsymmetric tensor and one symmetric tensor, and for symmetric tensor-valued
transversely isotropic functions of any finite number of vectors and 2nd-order tensors relative to the five
cylindrical groups, as well as the continuouly differentiable representation for symmetric tensor-valued
isotropic constitutive functions of one symmetric tensor, as given in §5.6.
118 CHAPTER 5. TENSOR FUNCTIONS
References

Ball, J.M., 1984. Differentiability properties of symmetric and isotropic functions. Duke Math. J. 51, 699-728.

Başar, Y., Weichert, D., 2000. Nonlinear Continuum Mechanics of Solids. Springer, Berlin etc.

Berteram, A., 1989. Axiomatische Einfürung in die Kontinuumsmechanik. Wissenschaftsverlag, Mannheim, Wien,
Zürich.

Berteram, A., Olschewski, J., 1991. Formulation of anisotropic linear viscoelastic constitutive laws by a projection
method.In: Freed, A., Walker, K. (eds.), High Temperature Constitutive Modeling: Theory and Application,
ASME, MD-Vol. 26, AMD-Vol. 121, pp. 129-137.

Berteram, A., Olschewski, J., 1993. Zur Formulierung anisotroper linearen anelastischen Stoffgleischungen mit Hilfe
einer Projektion Methode. Z. Angew. Math. Mech. 73, T401-T403.

Bertram, A., Svendsen, B., 2001. On material objectivity and reduced constitutive equations. Arch. Mechanics 54.

Betten, J., 1984. Interpolation methods for tensor functions. In: X.J.R. Avula, et al. (eds.), Mathematical Modelling
in Science and Technology, pp. 52-57. Pergoman Press, New York.

Betten, J., 1987. Tensorrechnung für Ingenieure. B.G. Teubner, Stuttgart.

Betten, J., 1988. Applications of tensor functions to the formulation of yield criteria for anisotropic materials. Int. J.
Plasticity 4, 29-46.

Betten, J., 1993. Kontinuumsmechanik. Springe, Berlin.

Betten, J., 1998. Anwendungen von Tensorfunktionen in der Kontinuumsmechanik anisotroper Materialien. Z.
Angew. Math. Mech. 78, 507-521.

Betten, J., Helisch, W., 1992. Irreduzible Invarianten eines Tensors vierter Stufe. Z. Angew. Math. Mech. 72, 45-57.

Bischhoff-Beiermann, B., Bruhns, O.T., 1994. A physically motivated set of invariants and tensor generators in the
case of transverse isotropy, Int. J. Eng. Sci. 32, 1531-1552.

Boehler, J.P., 1977. On irreducible representations for isotropic scalar functions, Z. Angew. Math. Mech. 57,
323-327.

Boehler, J.P., 1978. Lois de comportement anisotrope des milieux continus, J. de Mechanique 17, 153-190.

119
120 References

Boehler J.P., 1979, A simple derivation of representations for non-polynomial constitutive equations in some cases of
anisotropy. Z. Angew. Math. Mech. 59, 157-167.

Boehler J.P. (ed.), 1987. Applications of tensor functions in solid mechanics. CISM Courses and Lectures no. 292,
Springer, Wien, etc.

Boehler, J.P., Kirillov, A.A., Onat, E.T., 1994. On the polynomial invariants of the elasticity tensor. J. Elasticity 34,
97-110.

Boer, R. de, 1982. Vector- und Tensorrechnung für Ingenieure. Springer, Berlin.

Böhlke, T., 2001. Crystallographic Texture Evolution and Elastic Anisotropy: Simulation, Modeling, and Applica-
tions. Dissertation, Otto-Von-Guerick-Universität Magdeburg, Shaker Verlag, Aachen.

Böhlke, T., Beteram, A., 2001a. The evolution of Hooke’s law due to texture development in FCC polycrystals. Int.
J. Solids Structures 38, 9437-9459.

Böhlke, T., Beteram, A., 2001b. The 4th-order isotropic tensor-valued function of a symmetric 2nd-order tensor with
application to anisotropic elasto-plasticity. Z. Angew. Math. Mech. 81, 125-128.

Bowen, R.M., Wang, C.C., 1976. Introduction to vectors and tensors. Vol. 1-2. Plenum Press, New York.

Bruhns, O.T., 2003. Advanced Mechanics of Solids. Springer, Berlin.

Bruhns, O.T., Lehmann, Th., 1993. Elemente der Mechanik I. Vieweg, Braunschweig, Wisbaden.

Bruhns, O.T., Meyers, A., Xiao, H., 2002. On non-corotational rates of Oldroyd’s type and relevant issues in rate
constitutive formulations. Preprint (to be published).

Bruhns, O.T., Xiao, H., Meyers, A., 1999a. On representations of yield functions for crystals, quasicrystals and
transversely isotropic solids. Eur. J. Mech. A/Sloids 18, 47-671.

Bruhns, O.T., Xiao, H., Meyers, A., 1999b. Self-consistent Eulerian rate type elastoplasticity models based upon the
logarithmic stress rate. Int. J. Plasticity 15, 479-520.

Bruhns, O.T., Xiao, H., Meyers, A., 2002. New results for the spin of Eulerian triad and the logarithmic spin and rate.
Acta Mechanica. 155, 95-109.

Carlson, D.E., Hoger, A., 1986. The derivative of a tensor-valued function of a tensor. Q. Appl. Math. 44, 409-423.

Chadwick, P., Vianello, M., Cowin, S.C., 2001. A new proof that the number of linear elastic symmetries is eight. J.
Phys. Mech. Solids 49, 2471-2492.

Chou, P.-C., Pagano, N.J., 1992. Elasticity: Tensor, Dyadic, and Engineering Approaches. Dover Publ., Inc., New
York.

Cowin, S.C., Mehrabadi, M.M., 1992. On the structure of the linear anisotropic elastic symmetries. J. Phys. Mech.
Solids 40, 1459-1472.
References 121

Cowin, S.C., Mehrabadi, M.M., 1995. Anisotropic symmetries of linear elasticity. Appl. Mech. Rev. 48, 247-285.

Doyle, T.C., Ericksen, J.L., 1956. Nonlinear elasticity. Adv. Appl. Mech. 4, 53-115.

Fitzjerald, J.E., 1980. A tensorial Hencky measure of strain and strain rate for finite deformations. J. Appl. Phy. 51,
5111-5115.

Forte, S., Vianello, M., 1996. Symmetry classes for elasticity tensors. J. Elasticity 43, 81-108.

Forte, S., Vianello, M., 1997. Symmetry and harmonic decomposition for photoelasticity tensors. Int. J. Engrg. Sci.
35, 1317-1326.

Green, A.E., Adkins, J.E., 1960. Large Elastic Deformations and Nonlinear Continuum Mechanics. Clarendon,
Oxford.

Guo, Z.-H., Representations of Orthogonal tensors. Solid Mechanics Archives 6, 451-466.

Gurtin, M.E., 1972. The linear theory of elasticity. In: Flügge, S. (ed.), Handbuch der Physik, Vol. VIa/2. Springer,
Berlin etc.

Hackl, K., 1999. On the representation of anisotropic elastic materials by symmetric irreducible tensors. Contin.
Mech. Thermodyn. 11, 353-369.

Hahn, T., 1987. Space-Group Symmetry. International Tables for Crystallography. vol. A, 2nd edition, Reidel,
Dordrecht.

Halmos, P.R., 1974. Finite-Dimensional Vector Spaces. Springer, Berlin, etc.

Haupt, P., 2002. Continuum Mechanics and Theories of Materials. Springer, Berlin, etc.

Hill, R., 1968. Constitutive inequalities for simple materials. J. Mech. Phy. Solids 16, 229-242.

Hill, R., 1978. Aspects of invariance in solid mechanics. Adv. Appl. Mech. 18, 1-75.

Huo, Y.Z., Del Piero, G., 1991. On the completeness of the crystallographic symmetries in the description of the
symmetries of the elastic tensor. J. Elasticity 25, 203-246.

Itskov, M., 2002. The derivative with respect to a tensor. some theoretical aspects and applications. Z. Angew. Math.
Mech. 82, 535-544.

Itskov, M., Aksel, N., 2002. A closed-form representation for the derivative of non-symmetric tensor power series.
Int. J. Solids Structures 39, 5963-5978.

Kiral, E., Eringen, A.C., 1990. Constitutive Equations of Electromagnetic-Elastic Crystals, Springer, Berlin, etc.

Klein, F., 1884. Lectures on the Icsahedron. English version, reprinted in 1957, Dover, New York.

Klingbeil, E., 1966. Tensorrechnung für Ingenieure. Hochschultaschenbücher-Verlag, Mannheim.

122 References

Lehmann, Th., 1960. Einige Betrachtungen zu den Grundlagen der Umformtechnik. Ing. Archiv 29, 316-330.

Liu, I.S., 1982. On representations of anisotropic invariants, Int. J. Eng. Sci. 20, 1099-1109.

Lokhin, V.V., SEDOV, L.I., 1963. Nonlinear tensor functions of several tensor arguments, Prikl. Mat. Mekh., 29,
393-417.

A) = αII + βA
Man, C.-S., 1994. Remarks on the continuity of the scalar coefficients in the representation H (A A2
A + γA
for isotropic tensor functions. J. Elasticity 34, 229-238.

A) = αII + βA
Man, C.-S., 1995. Smoothness of the scalar coefficients in the representation H (A A2 for isotropic
A + γA
r
tensor functions of class C . J. Elasticity 40, 165-182.

Man, C.-S., Guo, Z.-H., 1993. A basis-free formula for time rate of Hill’s strain tensors. Int. J. Solids Struct. 30,
2819-2842.

Marsden, J.E., Hughes, T.J.R., 1983. Mathematical Foundations of Elasticity. Prentice Hall, Inc., Englewood Cliffs,
New Jersey.

Mehrabadi, M.M., Cowin, S.C., 1990. Eigentensors of linear anisotropic materials. Q. J. Mech. Appl. Math. 43,
15-41.

Mentzel, A., Steinmann, P., 2003. A view on anisotropic finite hyper-elasticity. Eur. J. Mech. A/Solids 22, 71-87.

Meyers, A., 1999a. Maschinendynamik. Lecturenots. Ruhr-Universität Bochum.

Meyers, A., 1999b. On the consistency of some Eulerian strain rates. Z. Angew. Math. Mech. 79, 171-177.

Meyers, A., Shieße, P., Bruhns, O.T., 2000. Some comments on objective rates of symmetric Eulerian tensors with
application to Eulerian strain rate. Acta Mechanica 139, 91-103.

Miehe, C., 1993. Computations of isotropic tensor functions. Commun. Numer. Methods Eng. 9, 889-896.

Miehe, C., 1994. Aspects of the formulation and FE implementation of large strain isotropic elasticity. Int. J. Numer.
Methods Eng. 37, 1981-2004.

Miehe, C., 1997. Comparison of two algorithms for the computation of fourth-order isotropic tensor functions.
Computer and Structures 66, 37-43.

Mises, R. von, 1928. Mechanik der plastichen Formänderung von Kristallen. Z. Angew. Math. Mech. 8, 161-185.

Morman, K.N., 1986. The generalized strain measure with application to non-homogeneous deformations in rubber-
like solids. J. Appl. Mech. 53, 726-728.

Nye, J.F., 1985. Physical Properties of Crystals. Clarendon Press, Oxford.

Ogden, R.W., 1984. Nonlinear Elastic Deformations. Ellis Horwood Ltd., Chichester.

Pennisi, S., Trovato, M., 1987. On irreducibility of Professor G.F. Smith’s representations for isotropic functions, Int.
References 123

J. Eng. Sci. 25, 1059-1065.

Richter, H., 1949. Verzerrungstensor, Verzerrungsdeviator und Spannungs tensor bei endlichen Formänderungen. Z.
Angew. Math. Mech. 29, 65-75.

Rychlewski, J., 1983. CEIIINOSSSTTUV: Mathematical structure of elastic materials (in Russian). Preprint No.
217, Inst. Mech. Probl. USSR Acad. Sci., Moscow.

Rychlewski, J., 1984a. On Hooke’s law. Prikl. Matem. Mekh. 48, 303-314.

Rychlewski, J., 1984b. Elastic energy decompositions and limit criteria. Advances in Mechanics 7 (3), 51-80.

Rychlewski, J., 1985. On thermoelastic constants. Archives of Mechanics 36, 77-95.

Rychlewski, J., 1991. Symmetry of Causes and Effects Wydawnictwo Naukowe PWN, Warsaw.

Rychlewski, J., 1995. Unconventional approach to linear elasticity. Archives of Mechanics 47, 149-171.

Rychlewski, J., 2000. A qualitative approach to Hooke’s tensor. Part I. Archives of Mechanics 52, 737-759.

Rychlewski, J., Zhang, J.M., 1991. On representations of tensor functions: A review. Advances in Mechanics 14 (1),
75-94.

Scheidler, M., 1991. Time rates of generalized strain tensors. Part I: Component formulas. Mech. Mater. 11,
199-210.

Scheidler, M., 1991. Time rates of generalized strain tensors. Part II: Approximate basis-free formulas. Mech. Mater.
11, 211-219.

Senechal, M., 1995. Quasicrystals and Geometry. Cambridge University Press, Cambridge.

Serrin, J.B., 1959. The derivation of stress-deformation relations for a Stokesian fluid. J. Math. Mech. 8, 459-468.

Seth, B.R., 1964. Generalized strain measures with applications to physical problems. In: Reiner, M. and Abir,
D. (eds.), Second-Order Effects in Elasticity, Plasticity and Fluid Dynamics, pp. 162-172. Pergamon Press,
Oxford.

Šilhavý, M., 1997. The Mechanics and Thermodynamics of Continuous Media. Springer, Berlin.

Simo, J.C., Taylor, R.L., 1991. Quasi-incompressible finite elasticity. Continuum basis and numerical algorithms.
Compt. Meth. Appl. Mech. Eng. 85, 273-310.

Smith, G.F., 1962. On the yield function for anisotropic materials. Q. Appl. Math. 20 (1962) 241-247.

Smith, G.F., Rivlin, R.S., 1958. The strain-energy function for anisotropic elastic materials. Trans. Am. Math. Soc.
88 175-193.

Smith, G.F., 1971. On isotropic functions of symmetric tensors, skewsymmetric tensors and vectors, Int. J. Eng. Sci.
9, 899-916.
124 References

Smith, G.F., 1994. Constitutive Equations for Anisotropic and Isotropic Materials. Elsevier, New York.

Spencer, A.J.M., 1971. Theory of invarians. In: Eringen, A.C. (ed.), Continuum Physics, vol. I. Academics Press,
New York.

Stumpf, H., Hoppe, U., 1997. The application of tensor algebra on manifolds to nonlinear continuum mechanics. Z.
Angew. Math. Mech. 77, 327-339.

Sutcliffe, S., 1992. Spectral decomposition of the elasticity tensor. J.Appl. Mech. 59, 762-773.

Svendsen, B., 1994. On the representation of constitutive relations using structural tensors. Int. J. Engng Sci. 32,
1889-1892.

Svendsen, B., Bertram, A., 1999. On frame-indifference and form-invariance in constitutive theory. Acta Mechanica
132, 195-207.

Svendsen, B., Reese, S., 2003. On the modelling of internal variables as structural tensors in anisotropic inelasticity.
Int. J. Plasticity 15.

Thomson, W. (Lord Kelvin), 1856. On six principal strains of an elastic solid. Phil. Trans. Roy. Soc. 166, 495-498.

Thomson, W. (Lord Kelvin), 1878. Elasticity. In: Encylopaedia Britannica, Adam and Charles Black, Edinburgh.

Ting, T.C.T., 1985. Determination of C 1/2 and C −1/2 and more general isotropic tensor functions. J. Elasticity 15,
319-323.

Ting, T.C.T., 1987. Invariants of anisotropic elastic constants. Q. J. Mech. Appl. Math. 40, 431-448.

Truesdell, C.A., Noll, W., 1992. The nonlinear field theories of mechanics. In: Flügge, S. (ed.), Handbuch der Physik,
vol. III/3. 2nd edition, Springer, Berlin.

Vainshtein, B.K., 1994. Modern Crystallography. 1. Fundamental of Crystals. Springer, Berlin, etc.

Wang, C.-C., 1970. A new representation theorem for isotropic functions, Part I and II, Arch. Rat. Mech. Anal. 36,
166-223, 1970;ibid, 43, 392-395.

Wang, W.B., Duan, Z.P., 1991. On the invariant representation of spin tensors with applications. Int. J. Solids Struct.
27, 329-341.

Xiao, H., 1995a. Unified explicit basis-free expressions for time rate and conjugate stress of an arbitrary Hill’s strain.
Int. J. Solids Structures 32, 3327-3340.

Xiao, H., 1995b. General irreducible representations for constitutive equations of anisotropic elastic crystals and
transversely isotropic elastic solids. J. Elasticity 39, 47-73.

Xiao, H., 1995c. Invariant characteristic representations for classical and micropolar anisotropic elasticity tensors. J.
Elasticity 40, 233-265.
References 125

Xiao, H., 1996a. On minimal representations for constitutive equations of anisotropic elastic materials. J. Elasticity
45, 13-32.

Xiao, H., 1996b. On isotropic extension of anisotropic tensor functions, Z. Angew. Math. Mech. 76, 205-214.

Xiao, H., 1996c. Two general representation theorems for arbitrary-order-tensor-valued isotropic and anisotropic
tensor functions of vectors and second order tensors, Z. Angew. Math. Mech. 76, 151-162.

Xiao, H., 1996d. On anisotropic scalar functions of a single symmetric tensor. Proc. Roy. Soc. Lonon A452,
1545-1561.

Xiao, H., 1997a. On constitutive equations of Cauchy’s elastic solids: All kinds of crystals and quasicrystals. J.
Elasticity 48, 241-283.

Xiao, H., 1997b. A unified theory of representations for scalar-, vector- and 2nd order tensor-valued anisotropic
functions of vectors and second order tensors, Archives of Mechanics 50, 275-313.

Xiao, H., 1997c. On isotropic invariants of the elasticity tensor. J. Elasticity 46, 115-149.

Xiao, H., 1998a. On anisotropic invariants of a single symmetric tensor: crystal classes, quasicrystal classes and
others. Proc. Roy. Soc. London A454, 1217-1240.

Xiao, H., 1998b. On scalar-, vector- and second-order tensor-valued anisotropic functions of vectors and second-order
tensors relative to all kinds of subgroups of the transverse isotropy group C∞h . Phil. Trans. Roy. Soc. London
A 356, 3087-3122; see also: Archives of Mechanics 50, 281-319.

Xiao, H., 1998c. On symmetries and anisotropies of classical micropolar linear elasticities. A new method based
upon a complex vector basis and some systematic results. J. Elasticity 49, 129-162.

Xiao, H., 1999. A new representation theorem for elastic constitutive equations of cubic crystals. J. Elasticity 53,
37-45.

Xiao, H., 2000. Further results on general representation theorems for arbitrary-order-tensor-valued isotropic and
anisotropic tensor functions of vectors and second order tensors, Z. Angew. Math. Mech. 80, 497-503.

Xiao, H., Bruhns, O.T., Meyers, A., 1997. Logarithmic strain, logarithmic spin and logarithmic rate. Acta Mechanica
124, 89-105.

Xiao, H., Bruhns, O.T., Meyers, A., 1998a. Strain rates and material spins. J. Elasticity 52, 1-41.

Xiao, H., Bruhns, O.T., Meyers, A., 1998b. On objective corotational rates and their defining spin tensors. Int. J.
Solids Structures 35, 4001-4014.

Xiao, H., Bruhns, O.T., Meyers, A., 1998c. On symmetric tensor-valued isotropic functions of a symmetric tensor
and a skewsymmetric tensor and related transversely isotropic functions. Archives of Mechanics 50, 731-741.

Xiao, H., O.T. Bruhns, Meyers, A.,1998d. Objective corotational rates and unified work-conjugacy relation between
Lagrangian and Eulerian stress and strain measures. Archives of Mechanics 50, 1015-1045.
126 References

Xiao, H., Bruhns, O.T., Meyers, A., 1999a. Existence and uniqueness of the integrable-exactly hypoelastic equation
o
τ ∗ = λ(trD
D)II + 2µD
D and its significance to finite inelasticity. Acta Mechanica 138, 31-50.

Xiao, H., Bruhns, O.T., Meyers, A., 1999b. On anisotropic invariants of N symmetric second-order tensors: crystal
classes and quasi-crystal classes as subgroups of the cylindrical group D∞h . Proc. Roy. Soc. London A 455,
1993-2020.

Xiao, H., Bruhns, O.T., Meyers, A., 1999c. Irreducible representations for constitutive equations of anisotropic solids
I: crystal and quasicrystal classes D2mh , D2m and C2mv . Archives of Mechanics 51, 559-603.

Xiao, H., Bruhns, O.T., Meyers, A., 2000a. Irreducible representations for constitutive equations of anisotropic solids
II: crystal and quasicrystal classes D2m+1d , D2m+1 and C2m+1v . Archives of Mechanics 52, 55-88.

Xiao, H., Bruhns, O.T., Meyers, A., 2000b. Irreducible representations for constitutive equations of anisotropic solids
III: crystal and quasicrystal classes D2m+1h and D2md . Archives of Mechanics 52, 347-395.

Xiao, H., Bruhns, O.T., Meyers, A., 2000c. Minimal generating sets for symmetric 2nd-order tensor-valued trans-
versely isotropic functions of vectors and 2nd-order tensors. Z. Angew. Math. Mech. 80, 565-569.

Xiao, H., Bruhns, O.T., Meyers, A., 2000d. A consistent finite elastoplasticty theory combining additive and multi-
plicative decomposition of the stretching and the deformation gradient. Int. J. Plasticity 16, 143-177.

Xiao, H., Bruhns, O.T., Meyers, A., 2002. Basic issues concerning finite strain measures and isotropic stress-
deformation relations. J. Elasticity 67, 1-23.

Xiao, H., Guo, Z.-H., 1993. A general representation theorem for isotropic tensor functions. In: Chien, W.Z., Guo,
Z.-H. and Guo, Y.Z. (eds.), Proceedings of 2nd Conference on Nonlinear Mechanics, pp. 206-210. Peking
University Press, Beijing.

Zhang, J.M., 1991. Material anisotropy and plasticity formulations. Eur. J. Mech. A/Solids 10, 157-174.

Zhang, J.M., Rychlewski, J., 1990. Structural tensors for anisotropic solids. Arch. Mechanics 42, 267-277.

Zheng, Q.S., 1994. On transversely isotropic, orthotropic and relatively isotropic functions of symmetric tensors,
skewsymmetric tensors and vectors. Part I-V. Int. J. Engng Sci. 31, 1399-1453.

Zheng, Q.S., 1994. Theory of representations for tensor functions: A unified invariant approach to constitutive
equations, Appl. Mech. Rev. 47, 545-587.

Zheng, Q.S., Spencer, A.J.M., 1993. Tensors which charactrize anisotropies. Int. J. Engng Sci. 31, 679-693.