Appendix A

Brief Review of Tensors

A.1 Introductory Remarks

In the study of particle mechanics and the mechanics of solid rigid bodies vector notation provides
a convenient means for describing many physical quantities and laws.
In studying the mechanics of solid deformable media, physical quantities of a more complex
nature, such as stress and strain, assume importance.1 Mathematically such physical quantities are
represented by matrices.
In the analysis of general problems in continuum mechanics, the physical quantities encountered
can be somewhat more complex than vectors and matrices. Like vectors and matrices these physical
quantities are independent of any particular coordinate system that may be used to describe them.
At the same time, these physical quantities are very often specified most conveniently by referring to
an appropriate system of coordinates. Tensors, which are a generalization of vectors and matrices,
offer a suitable way of mathematically representing these quantities.
As an abstract mathematical entity, tensors have an existence independent of any coordinate
system or frame of reference, yet are most conveniently described by specifying their components
in an appropriate system of coordinates. Specifying the components of a tensor in one coordinate
system determines the components in any other system. Indeed, the law of transformation of tensor
components is often used as a means for defining the tensor.
The objective of this appendix is to present a brief overview of tensors. Further details pertaining
to this subject can be found in standard books on the subject such as [2, 4], or in books dealing
with Continuum Mechanics such as [1, 3].

1 We recall that in describing stresses and strains one must specify not only the magnitude of the quantity, but also

the orientation of the face upon which this quantity acts.

1
2 A. Brief Review of Tensors

A.2 General Characteristics

The following general characteristics apply to all tensors:

Tensor Rank
Tensors may be classified by rank or order according to the particular form of transformation law
they obey. This classification is also reflected in the number of components a given tensor possesses
in an N -dimensional space. Thus, a tensor of order p has N p components. For example, in a three-
dimensional Euclidean space, the number of components of a tensor is 3p . It follows therefore, that
in three-dimensional space:

• A tensor of order zero has one component and is called a scalar. Physical quantities possessing
magnitude only are represented by scalars.
• A tensor of order one has three components and is called a vector; quantities possessing both
magnitude and direction are represented by vectors. Geometrically, vectors are represented by
directed line segments that obey the Parallelogram Law of addition.
• A tensor of order two has nine components and is typically represented by a matrix.

Notation
The following symbols are used herein:

• Scalars are represented by lowercase Greek letters. For example, α.

© ª
• Vectors are represented by lowercase Latin letters. For example, a or a .
© ª
• Matrices and tensors are represented by uppercase Latin letters. For example, A or A .

Cartesian Tensors
When only transformations from one homogeneous coordinate system (e.g., a Cartesian coordinate
system) to another are considered, the tensors involved are referred to as Cartesian tensors. The
Cartesian coordinate system can be rectangular (x1 , x2 , x3 ) or curvilinear, such as cylindrical (R, θ, z)
or spherical (r, θ, φ).
A.3. Indicial Notation 3

A.3 Indicial Notation

A tensor of any order, its components, or both may be represented clearly and concisely by the
use of indicial notation . This convention was believed to have been introduced by Einstein. In
this notation, letter indices, either subscripts or superscripts, are appended to the generic or kernel
letter representing the tensor quantity of interest; e.g. Aij , Bijk , δij , akl , etc. Some benefits of using
indicial notation include: (1) economy in writing; and, (2) compatibility with computer languages
(e.g. easy correlation with “do loops”). Some rules for using indicial notation follow.

Index rule
In a given term, a letter index may occur no more than twice.

Range Convention
When an index occurs unrepeated in a term, that index is understood to take on the values
1, 2, · · · , N where N is a specified integer that, depending on the space considered, determines
the range of the index.

Summation Convention
When an index appears twice in a term, that index is understood to take on all the values of its
range, and the resulting terms are summed. For example, Akk = a11 + a22 + · · · + aN N .

Free Indices
By virtue of the range convention, unrepeated indices are free to take the values over the range, that
is, 1, 2, · · · , N . These indices are thus termed “free.” The following items apply to free indices:

• Any equation must have the same free indices in each term.

• The tensorial rank of a given term is equal to the number of free indices.

• N (no. of free indices)

= number of components represented by the symbol.

Dummy Indices
In the summation convention, repeated indices are often referred to as “dummy” indices, since their
replacement by any other letter not appearing as a free index does not change the meaning of the
term in which they occur.
In the following equations, the repeated indices are thus “dummy” indices: Akk = Amm and
aik bkl = ain bnl . In the equation Eij = eim emj i and j represent free indices and m is a dummy
index. Assuming N = 3 and using the range convention, it follows that Eij = ei1 e1j + ei2 e2j + ei3 e3j .
Care must be taken to avoid breaking grammatical rules in the indicial “language.” For example,
the expression a • b = (ak êk ) • (bk êk ) is erroneous since the summation on the dummy indices is
ambiguous. To avoid such ambiguity, a dummy index can only be paired with one other dummy
index in an expression. A good rule to follow is use separate dummy indices for each implied
summation in an expression.
4 A. Brief Review of Tensors

Contraction of Indices
Contraction refers to the process of summing over a pair of repeated indices. This reduces the order
of a tensor by two.
For example:

• Contracting the indices of Aij (a second-order tensor) leads to Akk (a zeroth-order tensor or
scalar).
• Contracting the indices of Bijk (a third-order tensor) leads to Bikk (a first-order tensor).

• Contracting the indices of Cijkl (a fourth-order tensor) leads to Cijmm (a second-order tensor).

Comma Subscript Convention

A subscript comma followed by a subscript index “i” indicates partial differentiation with respect
to each coordinate xi . Thus,
∂φ ∂ai ∂Cij
φ,m ≡ ; ai,j ≡ ; Cij,kl ≡ ; etc. (A.1)
∂xm ∂xj ∂xk ∂xl

• If i remains a free index, differentiation of a tensor with respect to i produces a tensor of order
one higher. For example

∂Aj
Aj,i = (A.2)
∂xi
• If i is a dummy index, differentiation of a tensor with respect to i produces a tensor of order
one lower. For example

∂Vm ∂V1 ∂V2 ∂VN

Vm,m = = + + ··· + (A.3)
∂xm ∂x1 ∂x2 ∂xN
A.4. Coordinate Systems 5

A.4 Coordinate Systems

The definition of geometric shapes of bodies is facilitated by the use of a coordinate system. With re-
spect to a particular coordinate system, a vector may be defined by specifying the scalar components
of that vector in that system.
A rectangular Cartesian coordinate (RCC) system is represented by three mutually perpendicular
axes in the manner shown in Figure A.1.

x
3

e^
3
e^
1

x1 ^ x
e 2
2

Figure A.1: Rectangular Cartesian Coordinate System

Any vector in the RCC system may be expressed as a linear combination of three arbitrary,
nonzero, non-coplanar vectors called the base vectors. Base vectors are, by hypothesis, linearly
independent. A set of base vectors in a given coordinate system is said to constitute a basis for that
system. The most frequent choice of base vectors for the RCC system is the set of unit vectors ê1 ,
ê2 , ê3 , directed parallel to the x1 , x2 and x3 coordinate axes, respectively.

.
Remark
1. The summation convention is very often employed in connection with the representation of
vectors and tensors by indexed base vectors written in symbolic notation. In Euclidean space any
vector is completely specified by its three components. The range on indices is thus 3 (i.e., N = 3).
A point with coordinates (q1 , q2 , q3 ) is thus located by a position vector x, where

x = q1 ê1 + q2 ê2 + q3 ê3 (A.4)

In abbreviated form this is written as

x = qi êi (A.5)
where i is a summed index (i.e., the summation convention applies even though there is no repeated
index on the same kernal).
.

The base vectors constitute a right-handed unit vector triad or right orthogonal triad that satisfy
the following relations:

êi • êj = δij (A.6)

and
6 A. Brief Review of Tensors

êi × êj = ²ijk êk (A.7)

The set of base vectors satisfying the above conditions are often called an orthonormal basis.
In equation (A.6), δij denotes the Kronecker delta (a second-order tensor typically denoted by
I), defined by
(
1 if i = j
δij = (A.8)
0 if i =
6 j

In equation (A.7), ²ijk is the permutation symbol or alternating tensor (a third-order tensor),
that is defined in the following manner:


1 if i, j, k are an even permutation of 1,2,3
²ijk = −1 if i, j, k are an odd permutation of 1,2,3 (A.9)


0 if i, j, k are not a permutation of 1,2,3

An even permutation of 1,2,3 is 1, 2, 3, 1, 2, 3, · · · ; an odd permutation of 1,2,3 is 3, 2, 1, 3, 2, 1, · · · .

The indices fail to be a permutation if two or more of them have the same value.

.
Remarks
1. The Kronecker delta is sometimes called the substitution operator since, for example,

δij bj = δi1 b1 + δi2 b2 + δi3 b3 = bi (A.10)

δij Cjk = Cik (A.11)

and so on. From its definition we note that δii = 3.

2. In light of the above discussion, the scalar or dot product of two vectors a and b is written as

a • b = (ai êi ) • (bj êj ) = ai bj δij = ai bi (A.12)

In the special case when a = b,

2 2 2
a • a = ak ak = (a1 ) + (a2 ) + (a3 ) (A.13)

The magnitude of a vector is thus computed as

1 1
|a| = (a • a) 2 = (ak ak ) 2 (A.14)

3. The vector or cross product of two vectors a and b is written as

a × b = (ai êi ) × (bj êj ) = ai bj ²ijk êk (A.15)

4. The determinant of a square matrix A is

A.4. Coordinate Systems 7

¯ ¯
¯A11 A12 A13 ¯¯
¯
det A = |A| = ¯¯A21 A22 A23 ¯¯
¯A31 A32 A33 ¯
= A11 A22 A33 + A12 A23 A31 + A13 A21 A32
− A31 A22 A13 − A32 A23 A11 − A33 A21 A12
= ²ijk A1i A2j A3k (A.16)

5. It can also be shown that

¯ ¯ ¯ ¯
¯ Ai1 Ai2 Ai3 ¯¯ ¯¯A1i A1j A1k ¯¯
¯
²ijk det A = ¯¯ Aj1 Aj2 Aj3 ¯¯ = ¯¯A2i
A2j A2k ¯¯
¯Ak1 Ak2 Ak3 ¯ ¯A3i A3j A3k ¯
¯ ¯
¯ Air Ais Ait ¯
¯ ¯
²ijk ²rst det A = ¯¯ Ajr Ajs Ajt ¯¯
¯Akr Aks Akt ¯
¯ ¯
¯ δir δis δit ¯
¯ ¯
²ijk ²rst = ¯¯ δjr δjs δjt ¯¯
¯δkr δks δkt ¯
This leads to the following relations:

²ijk ²ist = δjs δkt − δjt δks

²ijk ²ijr = 2δkr

²ijk ²ijk = 6
.
8 A. Brief Review of Tensors

A.5 Transformation Laws for Cartesian Tensors

Define a point P in space referred to two rectangular Cartesian coordinate systems. The base vectors
for one coordinate system are unprimed, while for the second one they are primed. The origins of
both coordinate systems are assumed to coincide. The position vector to this point is given by

x = xi êi = x0j ê0j (A.17)

To obtain a relation between the two coordinate systems, form the scalar product of the above
equation with either set of base vectors; viz.,
¡ ¢
ê0k • (xi êi ) = ê0k • x0j ê0j (A.18)
Upon expansion,

xi (ê0k • êi ) = x0j δkj = x0k (A.19)

Since êi and ê0k are unit vectors, it follows from the definition of the scalar product that

ê0k • êi = (1)(1) cos (ê0k , êi ) ≡ Rki (A.20)

The Rki are computed by taking (pairwise) the cosines of angles between the and xi axes.x0k
For a prescribed pair of coordinate axes, the elements of Rki are thus constants that can easily be
computed.
From equation (A.19) it follows that the coordinate transformation for first-order tensors (vec-
tors) is thus

x0k = Rki xi (A.21)

where the free index is the first one appearing on R.
We next seek the inverse transformation. Beginning again with equation (A.17), we write
¡ ¢
êk • (xi êi ) = êk • x0j ê0j (A.22)
Thus,

xi δki = x0j êk • ê0j (A.23)

or

xk = Rjk x0j (A.24)

The free index is now the second one appearing on R.

.
Remark
1. In both above transformations, the second index on R is associated with the unprimed system.
.

In order to gain insight into the direction cosines Rij , we differentiate equation (A.21) with
respect to xi giving (with due change of dummy indices),

∂x0m ∂xj
= Rmj = Rmj δji = Rmi (A.25)
∂xi ∂xi
A.5. Transformation Laws for Cartesian Tensors 9

We next differentiate equation (A.24) with respect to x0m , giving

∂xk ∂x0j
= R jk = Rjk δjm = Rmk (A.26)
∂x0m ∂x0m
Using the chain rule, it follows that

∂xk ∂xk ∂x0m

= δki = = Rmk Rmi (A.27)
∂xi ∂x0m ∂xi
In direct notation this is written as

I = RT R (A.28)
−1 T
implying that the R are orthogonal tensors (i.e, R = R ). Linear transformations such as those
given by equations (A.21) and (A.24), whose direction cosines satisfy the above equation, are thus
called orthogonal transformations.
The transformation rules for second-order Cartesian tensors are derived in the following manner.
Let S be a second-order Cartesian tensor, and let

u = Sv (A.29)
in the unprimed coordinates. Similarly, in primed coordinates let

u0 = S0 v0 (A.30)
0 0 0
Next we desire to relate S to S . Using equation (A.21), substitute for u and v to give

Ru = S0 Rv (A.31)
But from equation (A.29)

Ru = RSv (A.32)
implying that

RSv = S0 Rv (A.33)
Since v is an arbitrary vector, and since R is an orthogonal tensor, it follows that

S0 = RSRT 0
or Sij = Rik Rjl Skl (A.34)
In a similar manner,

S = RT S0 R or Sij = Rmi Rnj Smn

0
(A.35)
The transformation rules for higher-order tensors are obtained in a similar manner. For example,
for tensors of rank three,

A0ijk = Ril Rjm Rkn Almn (A.36)

and

Aijk = Rli Rmj Rnk A0lmn (A.37)

10 A. Brief Review of Tensors

Finally, the fourth-order Cartesian tensor C transforms according to the following relations:
0
Cijkl = Rip Rjq Rkr Rls Cpqrs (A.38)
and
0
Cijkl = Rpi Rqj Rrk Rsl Cpqrs (A.39)

A.6 Principal Values and Principal Directions

In the present discussion, only symmetric second-order tensors with real components are considered.
For every symmetric tensor A, defined at some point in space, there is associated with each direction
(specified by the unit normal n) at the point, a vector given by the inner product

v = An (A.40)
This is shown schematically in Figure A.2.
n
v = An

Figure A.2: Normal Direction Associated with the Vector v

.
Remark
1. A may be viewed as a linear vector operator that produces the vector v conjugate to the
direction n.
.

If v is parallel to n, the above inner product may be expressed as a scalar multiple of n; viz.,

v = An = λn or Aij nj = λni (A.41)

The direction ni is called a principal direction, principal axis or eigenvector of A.
Substituting the relationship ni = δij nj into equation (A.41) leads to the following characteristic
equation of A:

(A − λI) = 0 or (Aij − λδij ) nj = 0 (A.42)

For a non-trivial solution, the determinant of the coefficients must be zero; viz.,
¯ ¯
det (A − λI) = 0 or ¯Aij − λδij ¯ = 0 (A.43)
A.6. Principal Values and Principal Directions 11

This is called the characteristic equation of A. In light of the symmetry of A, the expansion of
equation (A.43) gives
¯ ¯
¯(A11 − λ) A12 A13 ¯¯
¯
¯ A12 (A22 − λ) A23 ¯¯ = 0 (A.44)
¯
¯ A13 A23 (A33 − λ)¯
The evaluation of this determinant leads to a cubic polynomial in λ, known as the characteristic
polynomial of A; viz.,

λ3 − I¯1 λ2 + I¯2 λ − I¯3 = 0 (A.45)

where

I¯1 = tr (A) = A11 + A22 + A33 = Akk (A.46)

1
I¯2 = (Aii Ajj − Aij Aij ) (A.47)
2

I¯3 = det (A) (A.48)

The scalar coefficients I¯1 , I¯2 and I¯3 are called the first, second and third invariants, respectively,
derived from the characteristic
¡ equation
¢ of A.
The three roots λ(i) ; i = 1, 2, 3 of the characteristic polynomial are called the principal values
or eigenvalues of A. Associated with each eigenvalue is an eigenvector n(i) . For a symmetric tensor
with real components, the principal values are real. If the three principal values are distinct, the
three principal directions are mutually orthogonal. When referred to principal axes, A assumes a
diagonal form; viz.,
 (1) 
λ 0
A= 0 λ(2) 0  (A.49)
0 0 λ(3)

.
Remark
1. Eigenvalues and eigenvectors have a useful geometric interpretation in two- and three-
dimensional space. If λ is an eigenvalue of A corresponding to v, then Av = λv, so that depending
on the value of λ, multiplication by A dilates v (if λ > 1), contracts v (if 0 < λ < 1), or reverses
the direction of v (if λ < 0).
.

Example 1: Invariants of First-Order Tensors

Consider a vector v. If the coordinate axes are rotated, the components of v will change.
However, the length (magnitude) of v remains unchanged. As such, the length is said to be invariant.
In fact a vector (first-order tensor) has only one invariant, its length.
.
Example 2: Invariants of Second-Order Tensors
12 A. Brief Review of Tensors

A second order tensor possesses three invariants. Denoting the tensor by A, its invariants are
(these differ from the ones derived from the characteristic equation of A)

I1 = tr (A) = A11 + A22 + A33 = Akk (A.50)

1 £¡ 2 ¢¤ 1
I2 = tr A = Aik Aki (A.51)
2 2
1 £¡ 3 ¢¤ 1
I3 = tr A = Aik Akj Aji (A.52)
3 3
Any function of the invariants is also an invariant. To verify that the first invariant is unchanged
under coordinate transformation, recall that

A0ij = Rik Rjl Akl (A.53)

Thus,

A0mm = Rmk Rml Akl = δkl Akl = Akk (A.54)

For the second invariant,

A0ik A0ki = (Ril Rkm Alm ) (Rkn Rip Anp )

= Ril Rip Alm Rkm Rkn Anp
= δlp Alm δmn Anp
= Apm Amp (A.55)

Finally, for the third invariant,

A0ik A0km A0mi = (Ril Rkp Alp ) (Rkn Rmq Anq ) (Rms Rit Ast )
= Ril Rit Alp Rkp Rkn Anq Rmq Rms Ast
= δlt Alp δpn Anq δqs Ast
= Atp Apq Aqt (A.56)

.
A.7. Tensor Calculus 13

A.7 Tensor Calculus

Several important differential operators are summarized below.

Gradient Operator
The linear differential operator

∂ ∂ ∂ ∂
∇= ê1 + ê2 + ê3 = êi (A.57)
∂x1 ∂x2 ∂x3 ∂xi
is called the gradient or del operator.

Gradient of a Scalar Field

Let φ(x1 , x2 , x3 ) be a scalar field. The gradient of φ is the vector ∇φ with components

∂φ
∇φ = grad φ = êi = êi φ,i (A.58)
∂xi
If n = ni êi is a unit vector, the scalar operator

∂
n • ∇ = ni (A.59)
∂xi
is called the directional derivative operator in the direction n.

Divergence of a Vector Field

Let v(x1 , x2 , x3 ) be a vector field. The scalar quantity

∂vi
∇ • v = div v = = vi,i (A.60)
∂xi
is called the divergence of v.

Curl of a Vector Field

Let u(x1 , x2 , x3 ) be a vector field. The vector quantity

∂uk
∇ × u = curl u = εijk êi = εijk uk,j êi (A.61)
∂xj
is called the curl of u.

.
Remark
1. When using uk,j for ∂uk /∂xj , the indices are reversed in order as compared to the definition
of the vector (cross) product; that is,

∇ × v = εijk vk,j êi (A.62)

whereas u × v = εkij ui vj êk .
.
14 A. Brief Review of Tensors

The Laplacian Operator

The Laplacian operator is defined by

∂2( )
∇2 ( ) = div grad ( ) = ∇ • ∇ ( ) = = ( ),ii (A.63)
∂xi ∂xi
Let φ(x1 , x2 , x3 ) be a scalar field. The Laplacian of φ is then

µ ¶
2 ∂
∇ φ= êi • (φ,j êj )
∂xi
∂2φ
= (êi • êj ) = φ,ji δij = φ,ii (A.64)
∂xj ∂xi

Let v(x1 , x2 , x3 ) be a vector field. The Laplacian of v is the following vector quantity:

∂ 2 uk
∇2 v = êk = uk,ii êk (A.65)
∂xi ∂xi

.
Remark
1. An alternate statement of the Laplacian of a vector is

∇2 v = ∇ (∇ • v) − ∇ × (∇ × v) (A.66)
.
References

[1] Fung, Y. C., A First Course in Continuum Mechanics, Second Edition. Englewood Cliffs, NJ:
Prentice Hall (1977).
[2] Joshi, A. W., Matrices and Tensors in Physics, 2nd Edition. A Halsted Press Book, New York:
J. Wiley and Sons (1984).
[3] Mase, G. E., Continuum Mechanics, Schaums Outline Series. New York: McGraw-Hill Book Co.
(1970).
[4] Sokolnikoff, I. S., Tensor Analysis, Theory and Applications. New York: J. Wiley and Sons
(1958).

15