16 1 Vectors and Tensors in a Finite-Dimensional Space

Fig. 1.3 Cauchy stress

vector

other side of this surface. Let the area A tend to zero keeping P as inner point. A
basic postulate of continuum mechanics is that the limit

p
t = lim
A→0 A

exists and is final. The so-defined vector t is called Cauchy stress vector. Cauchy’s
fundamental postulate states that the vector t depends on the surface only through
the outward unit normal n. In other words, the Cauchy stress vector is the same
for all surfaces through P which have n as the normal in P. Further, according to
Cauchy’s theorem the mapping n → t is linear provided t is a continuous function
of the position vector x at P. Hence, this mapping can be described by a second-order
tensor σ called the Cauchy stress tensor so that

t = σn. (1.77)

Example 1.5. Moment of inertia tensor. Let us consider a material particle with a
mass dm and velocity vector v. The rotational momentum (moment of momentum)
of this particle is defined with respect to some origin O by

dl = r × p = r × (vdm) ,

where r denotes the position vector of this particle with respect to O and p = vdm
represents its linear momentum. Let further v ⊥ be a projection of v to the plane
orthogonal to r (see Fig. 1.4). Then, one can write

dl = dm (r × v ⊥ ) = dm [r × (ω̇ × r)] ,

where ω̇ denotes the angular velocity vector. Using identity (1.174) we further obtain

dl = dm [(r · r) ω̇ − (r · ω̇) r] = dm [(r · r) I − r ⊗ r] ω̇. (1.78)

All material particles within a rigid body (of the mass M) are characterized by
the same angular velocity ω̇. Thus,
1.6 Second-Order Tensor as a Linear Mapping 17

Fig. 1.4 Rotation of a rigid

body

ω̇

v⊥ v

r
O dm


l= [(r · r) I − r ⊗ r] dm ω̇ = Jω̇, (1.79)
M

where 
J= [(r · r) I − r ⊗ r] dm (1.80)
M

denotes the moment of inertia tensor. According to (1.79), it represents a linear

mapping of the angular velocity vector ω̇ of a body to its rotational momentum l.

On the basis of the “right” mapping (1.48) we can also define the “left” one by
the following condition

( yA) · x = y · (Ax) , ∀x ∈ En , A ∈ Linn . (1.81)

First, it should be shown that for all y ∈ En there exists a unique vector yA ∈ En
satisfying the condition (1.81) for all x ∈ En . Let G = g 1 , g 2 , . . . , g n and G  =
g 1 , g 2 , . . . , g n be dual bases in En . Then, we can represent two arbitrary vectors
x, y ∈ En , by x = xi g i and y = yi g i . Now, consider the vector
 
yA = yi g i · Ag j g j .
 
It holds: ( yA) · x = yi x j g i · Ag j . On the other hand, we obtain the same result
also by  
y · (Ax) = y · x j Ag j = yi x j g i · Ag j .
18 1 Vectors and Tensors in a Finite-Dimensional Space

Further, we show that the vector yA, satisfying condition (1.81) for all x ∈ En , is
unique. Conversely, let a, b ∈ En be two such vectors. Then, we have

a · x = b · x ⇒ (a − b) · x = 0, ∀x ∈ En ⇒ (a − b) · (a − b) = 0,

which by axiom (C.4) implies that a = b.

Since the order of mappings in (1.81) is irrelevant we can write them without
brackets and dots as follows

y · (Ax) = ( yA) · x = yAx. (1.82)

1.7 Tensor Product, Representation of a Tensor

with Respect to a Basis

The tensor product plays an important role since it enables to construct a second-
order tensor from two vectors. In order to define the tensor product we consider two
vectors a, b ∈ En . An arbitrary vector x ∈ En can be mapped into another vector
a (b · x) ∈ En . This mapping is denoted by symbol “⊗” as a ⊗ b. Thus,

(a ⊗ b) x = a (b · x) , a, b ∈ En , ∀x ∈ En . (1.83)

It can be shown that the mapping (1.83) fulfills the conditions (1.49)–(1.51) and for
this reason is linear. Indeed, by virtue of (B.1), (B.4), (C.2) and (C.3) we can write
 
(a ⊗ b) (x + y) = a b · (x + y) = a (b · x + b · y)
= (a ⊗ b) x + (a ⊗ b) y, (1.84)

(a ⊗ b) (αx) = a [b · (αx)] = α (b · x) a
= α (a ⊗ b) x, a, b ∈ En , ∀x, y ∈ En , ∀α ∈ R. (1.85)

Thus, the tensor product of two vectors represents a second-order tensor. Further, it
holds

c ⊗ (a + b) = c ⊗ a + c ⊗ b, (a + b) ⊗ c = a ⊗ c + b ⊗ c, (1.86)

(αa) ⊗ (βb) = αβ (a ⊗ b) , a, b, c ∈ En , ∀α, β ∈ R. (1.87)

Indeed, mapping an arbitrary vector x ∈ En by both sides of these relations and using
(1.52) and (1.83) we obtain

c ⊗ (a + b) x = c (a · x + b · x) = c (a · x) + c (b · x)
= (c ⊗ a) x + (c ⊗ b) x = (c ⊗ a + c ⊗ b) x,
1.7 Tensor Product, Representation of a Tensor with Respect to a Basis 19

[(a + b) ⊗ c] x = (a + b) (c · x) = a (c · x) + b (c · x)
= (a ⊗ c) x + (b ⊗ c) x = (a ⊗ c + b ⊗ c) x,

(αa) ⊗ (βb) x = (αa) (βb · x)

= αβa (b · x) = αβ (a ⊗ b) x, ∀x ∈ En .

For the “left” mapping by the tensor a ⊗ b we obtain from (1.81) (see Exercise 1.21)

y (a ⊗ b) = ( y · a) b, ∀ y ∈ En . (1.88)

We have already seen that the set of all second-order tensors Linn represents a
vector space. In the following, we show that a basis of Linn can be constructed with
the aid of the tensor product (1.83).
Theorem 1.7. Let F = f 1 , f 2 , . . . , f n and G = g 1 , g 2 , . . . , g n be two arbi-
trary bases of En . Then, the tensors f i ⊗ g j (i, j = 1, 2, . . . , n) represent a basis
of Linn . The dimension of the vector space Linn is thus n 2 .

Proof. First, we prove that every tensor in Linn represents a linear combination of the
tensors f i ⊗ g j (i, j = 1, 2, . . . , n). Indeed, let A ∈ Linn be an arbitrary second-
order tensor. Consider the following linear combination

A = f i Ag j f i ⊗ g j ,

where the vectors f i and g i (i = 1, 2, . . . , n) form the bases dual to F and G,

respectively. The tensors A and A coincide if and only if

A x = Ax, ∀x ∈ En . (1.89)

Let x = x j g j . Then

A x = f i Ag j f i ⊗ g j xk g k = f i Ag j f i xk δ kj = x j f i Ag j f i .

On the other hand, Ax = x j Ag j . By virtue of (1.27)–(1.28) we can represent the

 vec-
tors Ag j ( j = 1, 2, . . . , n) with respect to the basis F by Ag j = f i · Ag j f i =
f i Ag j f i ( j = 1, 2, . . . , n). Hence,

Ax = x j f i Ag j f i .

Thus, it is seen that condition (1.89) is satisfied for all x ∈ En . Finally, we show that
the tensors f i ⊗ g j (i, j = 1, 2, . . . , n) are linearly independent. Otherwise, there
would exist scalars αi j (i, j = 1, 2, . . . , n), not all zero, such that

αi j f i ⊗ g j = 0.
20 1 Vectors and Tensors in a Finite-Dimensional Space

Let αlk be one of the non-zero scalars. The right mapping of g k by both sides of the
above tensor equality yields: αik f i = 0. This contradicts, however, the fact that the
vectors f i (i = 1, 2, . . . , n) form a basis and are therefore linearly independent.

For the representation of second-order tensors we will in the following use primarily
the bases g i ⊗ g j , g i ⊗ g j , g i ⊗ g j or g i ⊗ g j (i, j = 1, 2, . . . , n). With respect to
these bases a tensor A ∈ Linn is written as
j
A = Ai j g i ⊗ g j = Ai j g i ⊗ g j = Ai· j g i ⊗ g j = Ai· g i ⊗ g j (1.90)

with the components (see Exercise 1.22)

Ai j = g i Ag j , Ai j = g i Ag j ,

j
Ai· j = g i Ag j , Ai· = g i Ag j , i, j = 1, 2, . . . , n. (1.91)

Note that the dot in the subscript indicates the position of the above index. For
example, for the components Ai· j , i is the first index while for the components A j·i ,
i is the second index.
Of special importance is the so-called identity tensor I. It is defined by

Ix = x, ∀x ∈ En . (1.92)

With the aid of (1.25), (1.90) and (1.91) the components of the identity tensor can
be expressed by

Ii j = g i Ig j = g i · g j = g i j , Ii j = g i Ig j = g i · g j = gi j ,

j
Ii· j = Ii· = Iij = g i Ig j = g i Ig j = g i · g j = g i · g j = δ ij , (1.93)

where i, j = 1, 2, . . . , n. Thus,

I = gi j g i ⊗ g j = g i j g i ⊗ g j = g i ⊗ g i = g i ⊗ g i . (1.94)

It is seen that the components (1.93)1,2 of the identity tensor are given by relation
(1.25). In view of (1.30) they characterize metric properties of the Euclidean space
and are referred to as metric coefficients. For this reason, the identity tensor is fre-
quently called metric tensor. With respect to an orthonormal basis relation (1.94)
reduces to
n
I= ei ⊗ ei . (1.95)
i=1