AE 6104 : Nonlinear Elasticity

Contents
1 Index notation 4
1.1 Summation convention . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Kronecker delta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Dot product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Permutation symbol . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Cross product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6 Differentiation notation . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Algebraic structures 10
2.1 Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Vector space or Linear space 13

3.1 Linear combination . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Linearly independent set . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Linear span . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4 Basis, components and dimension . . . . . . . . . . . . . . . . . . . 19
3.4.1 Standard basis . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.5 Vector subspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.6 Linear sum of two subspaces . . . . . . . . . . . . . . . . . . . . . . 22
3.7 Direct sum of two subspaces . . . . . . . . . . . . . . . . . . . . . . 23

4 Inner product space 24

4.1 Linear functional or linear form . . . . . . . . . . . . . . . . . . . . 24
4.2 Bilinear functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 Inner product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.4 Reciprocal basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.4.1 Length of a vector . . . . . . . . . . . . . . . . . . . . . . . 27
4.4.2 Angle between two vectors . . . . . . . . . . . . . . . . . . . 28
4.4.3 Differential arc length . . . . . . . . . . . . . . . . . . . . . 28
4.5 Physical interpretation of gij . . . . . . . . . . . . . . . . . . . . . . 28

1
CONTENTS

4.5.1 Physical components of vectors . . . . . . . . . . . . . . . . 34

4.6 Orthonormal set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.7 Orthogonal complement . . . . . . . . . . . . . . . . . . . . . . . . 37

5 Homomorphism of vector spaces or linear transformation 39

5.1 Zero transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.2 Range space R(T ) . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.3 Null space N (T ) or ker(T ) . . . . . . . . . . . . . . . . . . . . . . . 41
5.4 Rank and nullity of a homomorphism . . . . . . . . . . . . . . . . . 42
5.5 Singular homomorphism . . . . . . . . . . . . . . . . . . . . . . . . 42
5.6 Nonsingular homomorphism . . . . . . . . . . . . . . . . . . . . . . 43
5.7 Isomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.8 Invertible T : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.9 Linear operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.10 Identity operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.11 Endomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.12 Representation of homomorphism or LT by a matrix . . . . . . . . 44

6 Homomorphism as vector space 45

6.1 Product of homomorphism . . . . . . . . . . . . . . . . . . . . . . 46
6.2 Algebra of linear operators . . . . . . . . . . . . . . . . . . . . . . . 47

7 Tensor product 48
7.1 Matrix representation of tensor products . . . . . . . . . . . . . . . 50
7.2 Tensor spaces: a general multilinear function . . . . . . . . . . . . 50
7.3 Tensor multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.4 Contraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
7.4.1 Double contraction as a bilinear functional . . . . . . . . . . 55
7.4.2 Physical components of tensors . . . . . . . . . . . . . . . . 56
7.5 Transposition of a tensor . . . . . . . . . . . . . . . . . . . . . . . . 56
7.5.1 Symmetry and skew-symmetry of second order tensor . . . . 57
7.6 Invariants in 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.6.1 Cofactor tensor . . . . . . . . . . . . . . . . . . . . . . . . . 58

8 Orthogonal tensors 58

9 Skew-symmetric tensors 60

10 Spectral decomposition of a second order tensor 61

10.1 Spectral decomposition of a second order tensor . . . . . . . . . . . 64
10.1.1 Two repeated eigenvalues . . . . . . . . . . . . . . . . . . . 65
10.1.2 Three repeated eigenvalues . . . . . . . . . . . . . . . . . . . 66

AE 6104 2 Krishnendu Haldar

CONTENTS

10.2 Diagonalization of a tensor . . . . . . . . . . . . . . . . . . . . . . . 66

10.2.1 Power and polynomials of tensor . . . . . . . . . . . . . . . 67
10.2.2 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
10.2.3 Exponential map . . . . . . . . . . . . . . . . . . . . . . . . 69
10.3 Symmetric positive definite . . . . . . . . . . . . . . . . . . . . . . . 69
10.4 Polar decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . 70

11 Fourth order Tensors 71

12 Tensor calculus 72
12.1 Normed space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
12.2 Affine space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
12.3 Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
12.4 Partial derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
12.5 Product rule for bilinear mapping . . . . . . . . . . . . . . . . . . . 77
12.6 Product rule for any mapping . . . . . . . . . . . . . . . . . . . . . 79
12.7 Chain rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
12.8 Affine space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
12.9 Gradient and divergence . . . . . . . . . . . . . . . . . . . . . . . . 81
12.9.1 Gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
12.9.2 Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
12.9.3 Curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
12.10Divergence theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

13 Covariant derivative in Riemannian space 84

13.0.1 Christoffel symbols of 2nd kind . . . . . . . . . . . . . . . . . 84
13.0.2 Christoffel symbols of 1st kind . . . . . . . . . . . . . . . . . 85
13.0.3 Symmetry properties of Christoffel symbols . . . . . . . . . 85
13.1 Ricci identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
13.2 Gradient and divergence in Riemannian space . . . . . . . . . . . . 86
13.2.1 Gradient of vector-valued scalar function . . . . . . . . . . . 86
13.2.2 Gradient of vector-valued vector mapping . . . . . . . . . . 87
13.2.3 Gradient of vector-valued tensor mapping . . . . . . . . . . 89
13.2.4 Divergence of vector-valued vector mapping . . . . . . . . . 90
13.2.5 Divergence of vector-valued tensor mapping . . . . . . . . . 91

AE 6104 3 Krishnendu Haldar

1 Index notation
Consider the following sets of numbers:

a11 , a22 , a12 , a21

or

a11 , a22 , a12 , a21

or

a11 , a22 , a12 , a22

In the index notation, they are represented by: aij , aij and aij respectively, where
each of the indices (i, j) takes values 1 and 2.

1.1 Summation convention

Let n quantities be denoted by x1 , x2 , x3 , ..., xn . Consider the expression:
n
X
ai xi = a1 x1 + a2 x2 + ... + an xn
i

This expression will be written ai xi by introducing the summation convention of

EINSTEIN. When an index appears twice in a term, a summation is implied. The
range of the summation is known from the context. Note carefully that:

1. The convention does not apply to numerical indices. For instant a2 x2 stands
for single term.

2. The repeated index may be replaced by any other index. For instance ai xi =
at xt . For this reason, the repeated index is called a dummy index.

3. If an index is not dummy, it is called free index. Thus in the expression aik xi ,
k is the free index.

Some examples:

1. If n = 3, ak xk = a1 x1 + a2 x2 + a3 x3 .
∂ϕ ∂ϕ ∂ϕ
2. If ϕ is a function of x1 , x2 , x3 , ..., xn , then dϕ = ∂x1
dx1 + ∂x2
dx2 +...+ ∂xn
dxn =
∂ϕ
∂xi
dxi

AE 6104 4 Krishnendu Haldar

1.2 Kronecker delta

3. Let yi = αit xt and zi = βit yt . We can rewrite yt = αtk xk and now we can
substitute back as zi = βit (αtk xk ) = [βit αtk ]xk = γik xk . Here γik = βit αtk ,
where t is the dummy index and i, k are the free index.
4. Akmn Bkn
l
= (A1mn B1n
l
) + (A2mn B2n
l
) + ... + (AN l 1 l 1 l
mn BN n )=(Am1 B11 + Am2 B12 +
..) + (A2m1 B21
l
+ A2m2 B22
l
+ ...) + ...
5. Consider the summation (a1j xj )2 + (a2j xj )2 + ... + (anj xj )2 . This expression
is written as (ais xs )(ait xt ) or ais ait xs xt .

1.2 Kronecker delta

This is defied as:
(
1 if i = j
δij = (1)
0 if i ̸= j

Note that,

δ1j xj = δ11 x1 + δ12 x2 + ... + δ1n xn = x1

δ2j xj = δ21 x1 + δ22 x2 + ... + δ2n xn = x2
..
.
δij xj = δi1 x1 + δi2 x2 + ... + δin xn = xi
(2)

Thus if we multiply xk by δik , we simply replace the index k of xk by i. Moreover,

δii = δ11 + δ22 + δ33 = 3 (3)

1.3 Dot product

If the set {ê1 , ê2 , ê3 } contains a set of orthonormal unit vectors, then

êi · êj = δij (4)

For a Cartesian system, you can think as

     
1 0 0
ê1 = 0
  ê1 = 1
 ê3 = 0

0 0 1
Let a = ai êi and b = bj êj

a · b = (ai êi ) · (bj êj ) = ai bj (êi · êj ) = ai bj δij = ai bi (5)

AE 6104 5 Krishnendu Haldar

1.4 Permutation symbol

Figure 1: Even(left) and odd(right) permutations.

1.4 Permutation symbol


1
 if i, j, k form an even permutation
ϵijk = −1 if i, j, k form an odd permutation (6)

0 if two or more indices are equal


and we can write

ei · (ej × ek ) = ϵijk

1.5 Cross product

ê1 ê2 ê3

a×b = a1 a2 a3
b1 b 2 b3
= (a2 b3 − a3 b2 )ê1 + (a3 b1 − a1 b3 )ê2 + (a1 b2 − a2 b1 )ê3
= ϵijk aj bk êi (7)

Basically we represents a = aj êj and b = bk êk . So we can also write the compo-
nent form as:

(a × b)i = ϵijk aj bk (8)

If we replace a and b, then

b × a = ϵikj bk aj êi = −ϵijk aj bk êi = −a × b

AE 6104 6 Krishnendu Haldar

1.5 Cross product

Let c = ar êr , then

c · (a × b) = ϵijk cr aj bk (êi · êr ) = ϵijk cr aj bk δir = ϵijk ci aj bk

 
c1 c2 c3
= det a1 a2
 a3 
b1 b2 b3
 
c · e1 c · e2 c · e3
= det a · e1
 a · e2 a · e3 
b · e1 b · e2 b · e3

If we put c = ei , b = ek , and a = ej , then

   
ei · e1 ei · e2 ei · e3 δi1 δi2 δi3
ei · (ej × ek ) = ϵijk = det  ej · e1 ej · e2 ej · e3  = det  δj1 δj2 δj3  .
ek · e1 ek · e2 ek · e3 δk1 δk2 δk3

So we can write,
   
δi1 δi2 δi3 δp1 δp2 δp3
ϵijk ϵpqr = det δj1 δj2
 δj3 det δq1
  δq2 δq3 
δk1 δk2 δk3 δr1 δr2 δr3
   
δi1 δi2 δi3 δp1 δq1 δr1
= det  δj1 δj2 δj3  det δp2 δq2 δr2  ∵ det[A] det[B] = det[A] det[B]t
δk1 δk2 δk3 δp3 δq3 δr3
  
δi1 δi2 δi3 δp1 δq1 δr1
= det  δj1 δj2 δj3  δp2 δq2 δr2  ∵ det[A] det[B] = det([A][B]t )
δk1 δk2 δk3 δp3 δq3 δr3
   
δim δpm δim δqm δim δrm δip δiq δir
= det δjm δpm
 δjm δqm δjm δrm  = det  δjp δjq δjr 
δkm δpm δkm δqm δkm δrm δkp δkq δkr

We are now interested in

   
δii δiq δir 3 δiq δir
ϵijk ϵiqr = det  δji δjq δjr  = det  δij δjq δjr 
δki δkq δkr δik δkq δkr
= 3(δjq δkr − δjr δkq ) + δiq (δjr δik − δij δkr ) + δir (δij δkq − δjq δik )
= 3(δjq δkr − δjr δkq ) + (δjr δkq − δjq δkr ) + (δjr δkq − δjq δkr )
= (δjq δkr − δjr δkq )

Therefore, the Epsilon-delta identity reads

AE 6104 7 Krishnendu Haldar

1.5 Cross product

Figure 2: Even(left) and odd(right) permutations with i, j, and k.

ϵijk ϵiqr = δjq δkr − δjr δkq (9)

Example 1 Show that

a × b × c = (a · c)b − (a · b)c
Ans: Consider
a = ai ei , b = bj ej , c = ck ek
Now
b × c = ϵmjk bj ck em = Am em .
Then,
a×b×c = ϵrim ai Am er
= ϵrim ai (ϵmjk bj ck )er (see Fig. 2 for permuting r, i and m for ϵrim .)
= (ϵmri ϵmjk )ai bj ck er (bringing m at front to apply epsilon-delta identity (9))
= (δrj δik − δrk δij )ai bj ck er (from equation (9))
= δrj δik ai bj ck er − δrk δij ai bj ck er
= (δik ai )(δrj bj )ck er − (δij ai )bj ck (δrk er )
= (ak ck )(br er ) − (aj bj )(cr er )
= (a · c)b − (a · b)c
Example 2 Show that
(a × b) · (a × b) = (a · a)(b · b) − (a · b)2

AE 6104 8 Krishnendu Haldar

1.6 Differentiation notation

Consider,

a × b = ϵmij ai bj em , a × b = ϵpqr aq br ep .

Then

(a × b) · (a × b) = (ϵmij ai bj em ) · (ϵpqr aq br ep )
= ϵmij ϵpqr ai bj aq br (em · ep )
= ϵmij ϵpqr ai bj aq br (δmp )
= ϵmij ϵmqr ai bj aq br
= (δiq δjr − δir δjq ) ai bj aq br
= δiq δjr ai bj aq br − δir δjq ai bj aq br
= aq br aq br − ar bq aq br
= (aq aq )(br br ) − (ar br )(aq bq )
= (a · a)(b · b) − (a · b)2

1.6 Differentiation notation

Let

a = a(x1 , x2 , x3 )
ai = ai (x1 , x2 , x3 )
aij = aij (x1 , x2 , x3 )

Then
∂a
a, i = = ∂i a
∂xi
∂ai
ai, j = = ∂ j ai
∂xj
∂aij
aij, k = = ∂k aij (10)
∂xk
Example 3 Show that

∇ × ϕu = ∇ϕ × u + ϕ∇ × u

where ∇ = e1 ∂x∂ 1 + e2 ∂x∂ 2 + e3 ∂x∂ 3 = ei ∂i , ϕ(x1 , x2 , x3 ) is a scalar function, and

u(x1 , x2 , x3 ) is a vector function.

AE 6104 9 Krishnendu Haldar

 ∇ × ϕu = ej ∂j × ϕuk ek = ϵijk (∂j (ϕuk ))ei
= ϵijk ((∂j ϕ)uk )ei + (ϕ(∂j uk ))ei
= ϵijk (∂j ϕ)uk ei + ϕϵijk (∂j uk )ei
= ∇ϕ × u + ϕ∇ × u (11)

2 Algebraic structures
Consider the set X , equipped with some binary operations. A binary operation,
∗, takes two elements of the set and combine them to produce one element of the
same set. One can write

∗ : X × X → X,

A set with some binary operations and some rules creates an algebraic structure.

2.1 Group
A group is a set G together with a binary operation ∗ and satisfy the following set
of rules/axioms:

1. Closer: a, b ∈ G, then a ∗ b ∈ G

2. Associativity: If a, b, c ∈ G, then a ∗ (b ∗ c) = (a ∗ b) ∗ c.

3. Unique identity element: There is a distinguished element e ∈ G, called the

identity element, which has the following property e ∗ a = a = a ∗ e, ∀a ∈ G.

4. Inverses: For each a in G, there is an inverse, denoted by a−1 , with the

property a ∗ a−1 = e = a−1 ∗ a.

Example 4 Consider the set of integers Z with ∗ = +. If If a, b, c ∈ Z, then

1. Closer: a + b ∈ Z

2. Associativity: a + (b + c) = (a + b) + c.

3. Identity element: 0 ∈ Z such that 0 + a = a = a + 0, ∀a ∈ Z.

4. Inverses: For each a in G, there is an inverse, denoted by a−1 = −a, sucg

that a + (−a) = 0 = (−a) + a.

If we consider ∗ = · (multiplication), then the identity element will be 1. But the

inverse will not be an integer anymore. So, it will not be a group.

AE 6104 10 Krishnendu Haldar

2.2 Field

Example 5 Consider the rotation of a circle.

 
cos θ − sin θ
R(θ) = : (0 ≤ θ < 2π),
sin θ cos θ

where θ is the angle of rotation. The binary operation ∗ is the matrix multiplica-
tion.

1. Closer: R(θ1 ) · R(θ2 ) = R(θ1 + θ2 ) ∈ Z

2. Associativity: R(θ1 ) · (R(θ2 ) · R(θ3 )) = (R(θ1 ) · R(θ2 )) · R(θ3 ).

3. Identity element:
 
1 0
R(0) =
0 1

4. Inverses: R(θ)−1 = R(−θ)

Note that all the matrices are orthogonal. Moreover all of them have determinant
+1. This group is commonly called as Special Orthogonal Group of 2 × 2 matrices
or SO(2). The general form is SO(n). If we remove the constraint that the
determinant is +1, then we call it Orthogonal Group O(n).

Example 6 Set of all invertible n × n matrices with matrix multiplication forms

a group. It is called General Linear group GL(n).

2.2 Field
Suppose F is equipped with two binary operations called addition and multiplica-
tion and denoted by + and ·, respectively. This means ∀a, b ∈ F, we have a+b ∈ F
and a · b ∈ F. This algebraic structure (F, +, ·) is called a field if the following
postulates are satisfied :

1. Closed under addition: a + b ∈ F, ∀a, b ∈ F

2. Addition is commutative: a + b = b + a, ∀a, b ∈ F.

3. Addition is associative: (a + b) + c = a + (b + c), ∀a, b, c ∈ F.

4. ∃ an element 0 ∈ F (called zero) such that: a + 0 = a ∀a ∈ F.

5. To each element a ∈ F, ∃ −a ∈ F such that: a + (−a) = 0 ∀a ∈ F, and

a ̸= 0.

AE 6104 11 Krishnendu Haldar

2.2 Field

Figure 3: Essential elements of a field and the closer properties under addition and
multiplication. The boundary of the set in the diagram is just to schematically
represent a set.

6. Closed under multiplication: a · b ∈ F, ∀a, b ∈ F

7. Multiplication is commutative: a · b = b · a ∀a, b ∈ F.

8. Multiplication is associative: (a · b) · c = a · (b · c) ∀a, b, c ∈ F.

9. ∃ an non-zero element 1 ∈ F (called one/unity) such that: a · 1 = a ∀a ∈ F,

and a ̸= 0.

10. To each element a ∈ F, ∃ a−1 ∈ F such that: a · a−1 = 1 ∀a ∈ F, and a ̸= 0.

11. Multiplication is distributive with respect to addition: a · (b + c) = a · b + a · c

∀a, b, c ∈ F.

We simply denote (F, +, ·) as F. The elements of F are called scalars.

Example 7 The set of natural numbers N, integers I, and rational numbers Q

are examples of fields as they obey all the eleven rules for a field. However, in the
course, we mostly consider field as the set of real numbers R.

AE 6104 12 Krishnendu Haldar

3 Vector space or Linear space
Suppose V is a non-empty set equipped with a field F, is called a linear or vector
space if the following conditions are fulfilled:
(A1) Closed under addition: x + y ∈ V, ∀x, y ∈ V

(A2) Addition is commutative: x + y = y + x, ∀x, y ∈ V.

(A3) Addition is associative: (x + y) + z = x + (y + z), ∀x, y, z ∈ V.

(A4) ∃ an element 0 ∈ V (called zero) such that: x + 0 = x ∀x ∈ V.

(A5) To each element x ∈ V, ∃ −x ∈ V such that: x + (−x) = 0 ∀x ∈ V, and

x ̸= 0.
An operation is called vector addition if all the above operations hold. Now an
operation called scalar multiplication if a scalar (or a number) a ∈ F can be
combined with every element x ∈ V to give an element ax ∈ V such that the
following rules hold:
(B1) Closed under scalar multiplication: ax ∈ V, ∀a ∈ F and x ∈ V

(B2) Multiplication by scalar is associative: a(bx) = (ab)x ∀a, b ∈ F and x ∈ V.

(B3) ∃ an non-zero element 1 ∈ F (called one/unity) such that: 1x = x ∀x ∈ V.

(B4) Multiplication by scalar is distributive with respect to vector addition: a(x+

y) = ax + ay ∀a ∈ F and ∀x, y ∈ V.

(B5) Multiplication by scalar is distributive with respect to scalar addition: (a +

b)x = ax + bx ∀a, b ∈ F and ∀x ∈ V.
So a vector space is constructed with the above 5 + 5 = 10 rules.1

Example 8 Let V = Rm×n , the set of m × n matrices. Then V is a vector space

with the field F = R. Let, x = [A]mn , y = [B]mn , and z = [C]mn ∈ V. So,

1. Closed under addition: [A]mn + [B]mn = [A + B]mn ∈ V Let

   
a11 a12 · · · a1n b11 b12 · · · b1n
[A]mn =  ... .. ..  , [B]mn =  ... .. ..  .
 
. .  . . 
am1 am2 · · · amn m×n bm1 bm2 · · · bmn m×n
1
There should be no confusion about the use of the word vector. Here by vector we do not
mean the vector quantity which we have defined in vector algebra as a directed line segment.

AE 6104 13 Krishnendu Haldar

 where aij , bij ∈ R. Then
 
a11 + b11 a12 + b12 · · · a1n + a1n
[A + B]mn = 
 .. .. .. 
. . . 
am1 + bm1 am2 + bm2 · · · amn + bmn m×n
 
c11 c12 · · · c1n
=  ... .. ..  ∈ V.

. . 
cm1 cm2 · · · cmn m×n

since cij = aij + bij ∈ F as aij , bij ∈ F.

2. Addition is commutative: [A]mn + [B]mn = [B]mn + [A]mn Note that,

 
a11 + b11 a12 + b12 · · · a1n + a1n
[A]mn + [B]mn = [A + B]mn = 
 .. .. .. 
. . . 
am1 + bm1 am2 + bm2 · · · amn + bmn m×n
 
b11 + a11 b12 + a12 · · · b1n + a1n
= 
 .. .. .. 
. . . 
bm1 + am1 bm2 + am2 · · · bmn + amn m×n
= [B + A]mn = [B]mn + [A]mn

since aij + bij = bij + aij ∈ F as aij , bij ∈ F. Similarly, you can verify the
rest of the following relations.

3. Addition is associative: ([A]mn + [B]mn ) + [C]mn = [A]mn + ([B]mn + [C]mn )

4. Existence of a zero element [0]mn ∈ V such that: [A]mn + [0]mn = [A]mn

5. Additive inverse: [A]mn + (−[A]mn ) = [0]mn

6. Closed under scalar multiplication: a[A]mn ∈ V, ∀a ∈ F

7. Multiplication by scalar is associative: a(b[A]mn ) = (ab)[A]mn ∀a, b ∈ F

8. ∃ an non-zero element 1 ∈ F (called one/unity) such that: 1[A]mn = [A]mn

9. Multiplication by scalar is distributive with respect to vector addition: a([A]mn +

[B]mn ) = a[A]mn + a[B]mn , ∀a ∈ F

10. Multiplication by scalar is distributive with respect to scalar addition: (a +

b)[A]mn = a[A]mn + b[A]mn ∀a, b ∈ F

AE 6104 14 Krishnendu Haldar

Observe that we cannot multiply two matrices in the vector space! We certainly
need more tools to do so, and we discover those as we progress.

Example 9 Set of all n-tuples of real numbers Rn constitutes a vector space.

Substitute m = 0 in the previous example and immediately get the result as a
special case of Rm×n . These n-tuples of real numbers Rn are the vectors for the
physicists!

Example 10 All polynomials of degree n, Fn , over a field constitute a vector

space.

Let Fn (x) denote the set of all polynomials of x of degree n over a field F. Let us
consider the following polynomials:

f (x) = a0 + a1 x + a2 x2 + · · · + an xn , f (x) ∈ Fn (x), ai ∈ F

g(x) = b0 + b1 x + b2 x2 + · · · + bn xn , g(x) ∈ Fn (x), bi ∈ F
h(x) = c0 + c1 x + c2 x2 + · · · + cn xn , h(x) ∈ Fn (x), ci ∈ F

Moreover consider the zero polynomial as

0̂(x) = 0 + 0x + 0x2 + · · · + 0xn , 0̂(x) ∈ Fn (x), 0 ∈ F

Then,

1. f (x) + g(x) ∈ Fn (x). Note that

f (x) + g(x) = (a0 + b0 ) + (a1 + b1 )x + (a2 + b2 )x2 + · · · + (an + bn )xn

= c0 + c1 x + c2 x2 + · · · + cn xn ∈ Fn (x)

since ai + bi = ci ∈ F as ai , bi ∈ F.

2. f (x) + g(x) = g(x) + f (x) as

f (x) + g(x) = (a0 + b0 ) + (a1 + b1 )x + (a2 + b2 )x2 + · · · + (an + bn )xn

= (b0 + a0 ) + (b1 + a1 )x + (b2 + a2 )x2 + · · · + (bn + an )xn
= (b0 + b1 x + b2 x2 + · · · + bn xn ) + (a0 + a1 x + a2 x2 + · · · + an xn )
= g(x) + f (x)

since ai + bi = bi + ai ∈ F as ai , bi ∈ F. Similarly check the remaining points.

3. f (x) + (g(x) + h(x)) = (f (x) + g(x)) + h(x).

4. f (x) + 0̂(x) = f (x)

AE 6104 15 Krishnendu Haldar

3.1 Linear combination

5. f (x) + (−f (x)) = 0̂(x)

6. αf (x) ∈ Fn (x), where α ∈ F

7. k1 (k2 f (x)) = (k1 k2 )f (x), k1 , k2 ∈ F.

8. 1f (x) = f (x)

9. k1 (f (x) + g(x)) = k1 f (x) + k1 g(x), k1 ∈ F.

10. (k1 + k2 )f (x) = k1 f (x) + k2 f (x), k1 , k2 ∈ F.

Note that the polynomial are nonlinear with respect to x but the polynomial space
is linear!

3.1 Linear combination

Let x ∈ V and x1 , x2 , · · · , xn ∈ V. Then x is said to be linear combination of
{xi ∈ V, i = 1, n} such that:

x = a1 x 1 + a2 x 2 + · · · + an x n = ai x i

where {ai ∈ F, i = 1, n}.

3.2 Linearly independent set

Let V be a vector space over the field F. A finite set of vectors {x1 , x2 , · · · , xn } ∈ V
is said to be linearly independent if every relation of the form

a1 x1 + a2 x2 + · · · + an xn = 0, ai ∈ F, 1 ≤ i ≤ n

implies each ai = 0. Otherwise (i.e., if at least one ai ̸= 0) it is called linearly

dependent.

Example 11 Consider the following n vectors in V = Rn , where,

     
1 0 0
0 1 0
e1 =  ..  , e2 =  ..  , · · · , en =  ..  .
     
. . .
0 0 1

If ai ∈ F for i = 1, n are n arbitrary scalars, then

a1 e1 + a2 e2 + · · · + an en = 0.

AE 6104 16 Krishnendu Haldar

3.2 Linearly independent set

So,
           
a1 0 0 0 a1 0
 0  a2   0  0  a2  0
 ..  +  ..  , · · ·  ..  =  ..  =⇒  ..  =  .. 
           
. .  .  .  .  .
0 0 an 0 an 0

So a1 = a2 = . . . an = 0 and the vectors are linearly independent.

Figure 4: (a) Two mutually perpendicular vectors and (b) two oblique vectors.

Example 12 Consider Fig. 4. Which one of them is linearly independent?

The first figure is obvious. Let’s examine the second case. We write,
          
a1 b1 0 a1 b 1 α 0
α +β = , i.e, = . (12)
a2 b2 0 a2 b 2 β 0

If the determinant of the coefficient matrix is not equal to zero, then we always get
α = β = 0. Then a and b are linearly independent. However, if the determinant
is zero, then
a2 b2
a1 b2 − a2 b1 = 0, or = .
a1 b1
This means a and b have the same direction (Fig. 5). We can express any of them
with a scalar multiplication of the other vector, say b = λa, where λ belongs to
the field and represents a scaling.

AE 6104 17 Krishnendu Haldar

3.2 Linearly independent set

Figure 5: Two linearly dependent vectors.

Example 13 Let
     
 1 0 −1 
S=  2 , 3 , 0  ⊂ V = R3 .
   
0 1 1
 

Check if they are linearly independent or not

Let a, b, c ∈ F. Then, by definition,
       
1 0 −1 0
a 2 + b 3 + c  0  = 0 .
0 1 1 0

This gives,

a−c = 0
2a + 3b = 0
b + c = 0.

The coefficient matrix

 
1 0 −1
A = 2 3 0  .
0 1 1

and det A = 1 ̸= 0. So the only solutions are a = b = c = 0 and the set S is

linearly independent.

AE 6104 18 Krishnendu Haldar

3.3 Linear span

Example 14 Show that the set {1, x, 1 + x + x2 }, where x ∈ F is linearly inde-

pendent in the space of all polynomials.
Let a, b, c ∈ F. Then,

a(1) + b(x) + c(1 + x + x2 ) = 0

(a + c) + (b + c)x + c(x2 ) = 0
=⇒ (a + c) = 0, (b + c) = 0, c = 0 for all x.

So a = b = c = 0, and the set is linearly independent.

3.3 Linear span

Let V be a vector space and B ⊂ V. Then linear span of B or Span(B) is the set
of all linear combinations of the elements of B. Thus we have

Span (B) = a1 g 1 + a2 g 2 + · · · + ap g p

where B := {g 1 , g 2 , · · · , g p } and a1 , a2 , · · · , ap are from F.

3.4 Basis, components and dimension

A set of vectors B = {g 1 , g 2 , · · · , g n } (g i ∈ V) is said to be a basis of V if

1. B is linearly independent

2. Span(B) = V

So any x ∈ V can be expressed as x = xi g i , where xi s are called components of V

with respect to the basis {g i }. If x represents position vector, then those scalars
are called coordinates.
A vector space can have infinite many different basis, but all of them will have
the same number of elements. The number of elements in a basis is called the
dimension of the space V. In this case dim V = n.

Example 15 Show that the vectors

      
 1 2 1 
2 , 1 , −1
1 0 2
 

form a basis of V = R3 .

AE 6104 19 Krishnendu Haldar

3.4 Basis, components and dimension

Let a, b, c ∈ F. We start with

       
1 2 1 0
a 2 + b 1 + c −1 = 0 .
1 0 2 0

We further write,
    
1 2 1 a 0
2 1 −1  b  = 0
1 0 2 c 0

Now det A = −11 ̸= 0, and the only solutions are a = b = c = 0. So the vectors
are linearly independent. Next we need to show that they also span the space.
Consider an arbitrary vector
 
k1
k2  ∈ V = R3 .
k3

Then,
    
1 2 1 a k1
2 1 −1  b  = k2  .
1 0 2 c k2

As detA ̸= 0, for any arbitrary k1 , k2 , k3 , the system is solvable uniquely. So, the
considered vectors span R3 , and they are a basis.

3.4.1 Standard basis

Consider the set Bs = {e1 , e2 , · · · , en }, where,
     
1 0 0
0 1 0
e1 =  ..  , e2 =  ..  , · · · , en =  ..  . (13)
     
. . .
0 0 1

Note that they are linearly independent. This particular basis is known as standard
basis in Rn .

Example 16 What is the dimension of the space V = {0}, i.e., space only with
the zero element?

AE 6104 20 Krishnendu Haldar

3.5 Vector subspaces

Dimension ≡ Number of basis elements in V,

First check that (by yourself) V = {0} is a vector space. Since {0} is the only
element, the only possible basis is B = {0}. Then,

α0 = 0

always satisfies for any α ̸= 0, i.e., the considered space can not have any basis.
So dim V = 0.

3.5 Vector subspaces

Definition 3.1 Let V be a vector space over the field F and W ⊂ V. Then W
is called a subspace of V if W itself is a vector space over F with respect to the
operations of vector addition and scalar multiplication in V.

Theorem 3.1 The necessary and sufficient condition for a nonempty subset W
of a vector space V over field F to be a subspace is

a, b ∈ F, and x, y ∈ W =⇒ ax + by ∈ W. (14)

Necessary condition
Consider W is a subspace of V. Then W must be closed under scalar multiplication
and vector addition. This means

a ∈ F, x ∈ W =⇒ ax ∈ W
b ∈ F, y ∈ W =⇒ by ∈ W

Now as ax ∈ W and by ∈ W, then ax+by ∈ W. Hence the condition is necessary.

Sufficient condition
Now suppose W is a non-empty subset of V satisfying the given condition ax+by ∈
W for all a, b ∈ F, and x, y ∈ W. One can show that the proposed conditions
satisfy all the requirements for the vector space. As ax + by ∈ W,
1. let a = 1, b = 1, then 1.x + 1.y = x + y ∈ W . Thus W is closed under
vector addition.

2. Let a = 0, b = 0, then 0.x + 0.y = 0 ∈ W . Thus the zero vector 0 ∈ W.

3. Let a = −1, b = 0, then −1.x + 0.y = −x ∈ W . Thus the additive inverse

−x ∈ W.

AE 6104 21 Krishnendu Haldar

3.6 Linear sum of two subspaces

4. Let y = 0, then ax + b0 = ax ∈ W . Thus W is closed under scalar

multiplication.

5. Let a = 1, and y = 0, then 1x + b0 = x ∈ W .

This proves that W is itself a vector space and subspace of V. Note that any
subset of V will follow operations of vector additions and scalar multiplications,
i.e., A2, A3, B2, B4, B5. We do not need prove again those.

Figure 6: Representation of example 17.

Example 17 Let us consider a vector space V = R3 , and W := (w1 , w2 , 0) ⊂ V,

where w1 , w2 ∈ F. Show that W is a subspace of V.

Let x = (x1 , x2 , 0), and y = (y1 , y2 , 0) ∈ W, and x1 , x2 , y1 , y2 ∈ F. Then

αx + βy = α(x1 , x2 , 0) + β(y1 , y2 , 0) (α, β ∈ F)

= (αx1 , αx2 , 0) + (βy1 , βy2 , 0)
= (αx1 + βy1 , αx2 + βy2 , 0)

Now, αx1 + βy1 ∈ F and αx2 + βy2 ∈ F. So, αx + βy ∈ W, and W is a subspace

of V.

3.6 Linear sum of two subspaces

Definition 3.2 Let V be a vector space over F, and W1 , W2 be the two subspaces
of V. Then the linear sum of the subspaces W1 , and W2 , denoted by W1 + W2 , is
the set of all sums w1 + w2 such that w1 ∈ W1 and w2 ∈ W2 .

AE 6104 22 Krishnendu Haldar

3.7 Direct sum of two subspaces

Thus,

W1 + W2 := {w1 + w2 ; w1 ∈ W1 , w2 ∈ W2 }

Lemma 3.2 W1 + W2 is a subspace.

Let w′1 , w′′1 ∈ W1 , and w′2 , w′′2 ∈ W2 . Consider, v ′ = w′1 + w′2 ∈ W1 + W2 , and
v ′′ = w′′1 + w′′2 ∈ W1 + W2 . Then,

αv ′ + βv ′′ = α(w′1 + w′2 ) + β(w′′1 + w′′2 )

= (αw′1 + βw′′1 ) + (αw′2 + βw′′2 )
| {z } | {z }
∈W1 ∈W2
∈ W1 + W2

3.7 Direct sum of two subspaces

Definition 3.3 Let V be a vector space over F. Then V is said to be the direct
sum of W1 , W2 if

1. W1 + W2 = V

2. W1 ∩ W2 = {0}

We then write

V = W1 ⊕ W2 , and, dim V = dim W1 + dim W2 (15)

Example 18 Let V = R2 over the field F. Then W1 := {(a, 0) : a ∈ F} and

W2 := {(0, b) : b ∈ F} are the subspaces of R2 . Now any element (x, y) ∈ R2 can
be uniquely expressed as (x, y) = (x, 0) + (0, y). Then R2 = W1 ⊕ W2 .

AE 6104 23 Krishnendu Haldar

4 Inner product space
4.1 Linear functional or linear form
Linear functional or linear form is defined by the mapping

ϕ : V → R, such that ϕ⟨v⟩ ∈ R, , v ∈ V

Many researchers use ϕ(v) as well. Note that

ϕ⟨α1 v 1 + α2 v 2 ⟩ = α1 ϕ⟨v 1 ⟩ + α2 ϕ⟨v 2 ⟩, v 1 , v 2 ∈ V, and α1 , α2 ∈ F.

Example 19 Trace of a square matrix tr : V → R is a linear functional as we

know tr A ∈ R where V ∈ R. Also note that

tr (α1 A + α2 B) = α1 tr A + α2 tr B, A, B ∈ V, α1 , α2 ∈ F.

Example 20 A typical example of a linear functional is integration in the study

of vector spaces of functions. Let
Z b
I(f ) = f (x)dx.
a

Then,
Z b Z b Z b
I(α1 f + α2 g) = [α1 f (x) + α2 g(x)]dx = α1 f (x)dx + α2 g(x)dx
a a a
= α1 I(f ) + α2 I(g).

4.2 Bilinear functionals

Let V, U are two vector spaces. A map L : V × U → R is called bilinear if for any
v, v 1 , v 2 ∈ V, u, u1 , u2 ∈ U, and α1 , α2 ∈ F, it is linear in each variable; that is,

L⟨α1 v 1 + α2 v 2 , u⟩ = α1 L⟨v 1 , u⟩ + α2 L⟨v 2 , u⟩,

L⟨v, α1 u1 + α2 u2 ⟩ = α1 L⟨v, u1 ⟩ + α2 L⟨v, u2 ⟩.

The extension of this definition to functionals of more than two variables are
simple, and such functions are called multilinear functionals.

AE 6104 24 Krishnendu Haldar

4.3 Inner product

4.3 Inner product

We may think of a vector as a geometric object. Then it makes sense to talk about
the length of a vector and angle between two vectors. We will now study a certain
type of scalar valued bilinear functional on pair of vectors, known as inner product.

Definition 4.1 An inner product is a map

g :V ×V →R

with the following properties: For u, v ∈ V, and α ∈ R

1. g⟨u, v⟩ is bilinear,

2. g⟨u, v⟩ = g⟨v, u⟩,

3. g⟨u, u⟩ > 0, if u ̸= 0.

We call g⟨u, v⟩ the inner product of u and v and use the notation of ‘dot product
as

g⟨u, v⟩ = u · v

Similarly, we call g⟨g i , g j ⟩ the inner product of g i and g j and write

g⟨g i , g j ⟩ = g i · g j = gij

and

gi · gj = gj · gi, =⇒ gij = gji .

A vector space V equipped with the inner product is called an inner product space
and from now our vector spaces are all inner product space.

4.4 Reciprocal basis

We note that the dual space dissolves in the actual vector space when equipped
with the inner product. However, if {g 1 , g 2 , · · · , g n } be a basis of V, we can
construct a set of vectors {g 1 , g 2 , · · · , g n } such that

g i · g j = δji

where δji is called the Kronecker delta. The set {g 1 , g 2 , · · · , g n } becomes another
basis of the same vector space V again. Such a basis is known as reciprocal basis.

AE 6104 25 Krishnendu Haldar

4.4 Reciprocal basis

Theorem 4.1 Let {g 1 , . . . , g n } be a basis of V. Consider the set {g 1 , . . . , g n } of

V ∗ such that

g i · g j = δji . (16)

Then {g 1 , . . . , g n } is a basis of V.
Proof: Consider the following linear combination

α1 g 1 + α2 g 2 + · · · + αn g n = 0

Now,

(α1 g 1 + α2 g 2 + · · · + αn g n ) · g j = 0 · g j = 0
αi g i · g j = 0
αi δji = 0
=⇒ αj = 0 for j = 1, · · · , n

As the set {g 1 , g 2 , · · · , g n } is linearly independent, and dim V = n, then {g 1 , g 2 , · · · , g n }

is a basis of V. This basis, is called the reciprocal basis. Such a reciprocal basis is
unique (we are not proving its uniqueness).
From this construction, if v = v i g i ∈ V, then by taking the inner product with
g i we have

v · g i = (v j g j ) · g i = v j (g j · g i ) = v j δji = v i .

This gives the ith component of v in the {g 1 , g 2 , · · · , g n } basis system. Let u =

ui g i and v = v j g j belong to V. Then

u · v = (ui g i ) · (v j g j ) = ui v j (g j · g j ) = ui v j gij = ui (v · g j )gij = (ui gij g j ) · v.

This implies that u can also be expressed as u = uj g j . By comparing with the

above relation we get

uj = gij ui .

Since we can express u as u = ui g i w.r.t the basis {g i } or as u = ui g i w.r.t the

basis {g i }, then

u = ui g i = ui g i .

We call

ui : ith contravariant component of u,

ui : ith covariant component of u.

AE 6104 26 Krishnendu Haldar

4.4 Reciprocal basis

and

{g i } : is called covariant basis,

{g i } : is called contravariant basis.

Now, we define

g⟨g i , g j ⟩ = g ij = g i · g j .

and

gi · gj = gj · gi, =⇒ g ij = g ji .

Then

u · v = (ui g i ) · (vi g j ) = ui vj (g j · g j ) = ui vj g ij = (ui g ij g j ) · v.

So, we can write

uj = g ij ui .

The two operators

gij : uj → ui , g ij : uj → ui

enable us to lower and raise the component index. Moreover,

g i = gij g j , or g i = g ij g j .

Therefore, lowering or raising the index for the reciprocal basis can be made in
the same manner. Note that

g k · g i = gij g k · g j
δik = gij g kj = gij g jk .

This means [g ij ] = [gij ]−1 .

4.4.1 Length of a vector

The norm or length of a vector x ∈ V is defined as
p √
|x| := g⟨x, x⟩ = x · x

A vector space equipped with such a norm is called Euclidean vector space.

AE 6104 27 Krishnendu Haldar

4.5 Physical interpretation of gij

4.4.2 Angle between two vectors

The angle θ(x, y) between two non-zero vectors x, y ∈ V is defined as
x·y
cos θ(x, y) :=
|x||y|

4.4.3 Differential arc length

The differential arc length along a curve is given by

ds2 = dx · dx

4.5 Physical interpretation of gij

Consider ·V = R3 and position of a point with respect to the standard basis
{e1 , e2 , e3 } is denoted by

r = x1 e1 + x2 e2 + x3 e3

The correspondence between points (x1 , x2 , x3 ) and the three scalar parameters

Figure 7: A local (curvilinear) coordinate system. u1 , u2 , u3 are the coordinate

lines.

(u1 , u2 , u3 ) defines a local coordinate system in ·V. We write

x1 = ϕ1 (u1 , u2 , u3 ), x2 = ϕ2 (u1 , u2 , u3 ), x3 = ϕ3 (u1 , u2 , u3 ).

We assume that the mappings are bijective and have continuous derivatives, so
that the correspondence between (x1 , x2 , x3 ) and (u1 , u2 , u3 ) is unique.

AE 6104 28 Krishnendu Haldar

4.5 Physical interpretation of gij

If we only vary one local coordinate, say u1 , while the other two local coordi-
nates, (u2 , u3 ), are fixed, then we call the locus of the points (x1 , x2 , x3 ) a coordinate
line. So

cu1 = {ϕ1 (u1 , c2 , c3 ), ϕ2 (u1 , c2 , c3 ), ϕ3 (u1 , c2 , c3 )} → u1 coordinate line

cu2 = {ϕ1 (c1 , u2 , c3 ), ϕ2 (c1 , u2 , c3 ), ϕ3 (c1 , u2 , c2 )} → u2 coordinate line
cu3 = {ϕ1 (c1 , c2 , u3 ), ϕ2 (c1 , c2 , u3 ), ϕ3 (c1 , c2 , u3 )} → u3 coordinate line
are the three coordinate lines (Fig. 7). If the coordinate lines are straight,
we call it rectilinear coordinate system, else it is a curvilinear coordinate system.
Consider a point P ′ on the coordinate line u1 , obtained by a small change δu1 .
Then the position vector of the new point with respect to the standard basis could
be written as

OP ′ = ϕ1 (u1 + δu1 , c2 , c3 )e1 + ϕ2 (u1 + δu1 , c2 , c3 )e2 + ϕ3 (u1 + δu1 , c2 , c3 )e3 .

Figure 8: P P ′ becomes tangent as δu1 → 0.

So, as per Fig. 8,

P P ′ = OP ′ − OP
= ϕ1 (u1 + δu1 , c2 , c3 )e1 + ϕ2 (u1 + δu1 , c2 , c3 )e2 + ϕ3 (u1 + δu1 , c2 , c3 )e3
 

− ϕ1 (u1 , c2 , c3 )e1 + ϕ2 (u1 , c2 , c3 )e2 + ϕ3 (u1 , c2 , c3 )e3

 

∂ϕ1 1 ∂ϕ2 1 ∂ϕ3 1

= δu e 1 + δu e 2 + δu e3 .
∂u1 ∂u1 ∂u1
Then,

PP′ ∂x1 ∂x2 ∂x3

lim = g 1 = e 1 + e 2 + e3
δu1 →0 δu1 ∂u1 ∂u1 ∂u1
∂
= (x1 e1 + x2 e2 + x3 e3 )
∂u1
∂r
=
∂u1
AE 6104 29 Krishnendu Haldar
4.5 Physical interpretation of gij

∂r ∂r
Similarly, g 2 = ∂u 2 and g 3 = ∂u3 . These three local tangents vectors

n ∂r ∂r ∂r o
{g 1 , g 2 , g 3 } := 1
, , (17)
∂u ∂u2 ∂u3

Figure 9: Covariant or tangent basis vectors {g 1 , g 2 , g 3 }.

are known as covariant basis or natural basis (Fig. 9). As

r(u1 , u2 , u3 ) = x1 (u1 , u2 , u3 )e1 + x2 (u1 , u2 , u3 )e2 + x3 (u1 , u2 , u3 )e3 (18)

we write

∂r 1 ∂r ∂r
dr(u1 , u2 , u3 ) = 1
du + 2 du2 + 3 du3
∂u ∂u ∂u
1 2 3
= du g 1 + du g 2 + du g 3 . (19)
So,

dr · dr = (dui g i ) · (duj g j ) = (g i · g j )dui duj

or

ds2 = dr · dr = gij dui duj . (20)

Here ds is the differential arc-length distance along dr with respect to the local
coordinates {u1 , u2 , u3 }. Recall that, gij (u1 , u2 , u3 ) is known as metric tensor or
fundamental tensor. It is associated with the measure of distance in a generalized
way.

AE 6104 30 Krishnendu Haldar

4.5 Physical interpretation of gij

1. Riemannian space (gij is positive definite i.e. eigenvalues are ≥ 0)

(a) Non-Euclidean space: Gaussian curvature ̸= 0 (curved)
(b) Euclidean: Gaussian curvature = 0 (flat)
2. Semi/Pseudo Riemannian space (gij is negative definite i.e. at least one
eigenvalue is negative)
(a) Lorentz space: Only one eigenvalue of gij is negative

Figure 10: Covariant or tangent basis vectors {g 1 , g 2 } for the polar coordinates.

Example 21 Find ds2 for the polar coordinate system (Fig. 10)
We identify u1 = r and u2 = θ. Then,

x1 (r, θ) = r cos θ, x2 (r, θ) = r sin θ

The position vector is written as

r = x1 (r, θ)e1 + x2 (r, θ)e2 = r cos θe1 + r sin θe2

Therefore,
∂r
g1 = = cos θe1 + sin θe2
∂r
∂r
g2 = = −r sin θe1 + r cos θe2
∂θ
We compute

g11 = g 1 · g 1 = 1, g12 = g 1 · g 2 = 0, g21 = g 2 · g 1 = 0, g22 = g 2 · g 2 = r2

AE 6104 31 Krishnendu Haldar

4.5 Physical interpretation of gij

The arc-element with respect to the {g 1 , g 2 } system is

ds2 = gij dui duj

= g11 (du1 )2 + g22 (du2 )
= dr2 + (rdθ)2

Similarly, we can find metric to any curvilinear system!

As the correspondence between (x1 , x2 , x3 ) and (u1 , u2 , u3 ) is unique (bijective),
the inverse function could be written as

u1 = ψ 1 (x1 , x2 , x3 ), u2 = ψ 2 (x1 , x2 , x3 ), u3 = ψ 3 (x1 , x2 , x3 ).

Considering u3 = ψ 3 (x1 , x2 , x3 ) = c3 , some constant, then

Figure 11: g 1 is normal to the u3 = ψ 3 (x1 , x2 , x3 ) surface.

∂ψ 3 ∂ψ 3 ∂ψ 3
∇ψ 3 (x1 , x2 , x3 ) = e 1 + e 2 + e3
∂x1 ∂x2 ∂x3
or
3 ∂u3 ∂u3 ∂u3
g = e1 + 2 e2 + 3 e3 .
∂x1 ∂x ∂x
Then g 3 = ∇u3 (x1 , x2 , x3 ) is perpendicular to the tangent plane 2 at P (Fig. 11).
2
Recall that gradient of a constant surface is normal to the tangent plane. Suppose we have a
surface xy 3 = z + 2. To find the normal to the surface at (1, 1, −1), we consider f = xy 3 − z = 2.
Then ∇f = y 3 e1 + 3xy 2 e2 − e3 = e1 + 3e2 − e3 is the normal vector at (1, 1, −1).
(1,1,−1)

AE 6104 32 Krishnendu Haldar

4.5 Physical interpretation of gij

Now,
 ∂u3 ∂u3 ∂u3   ∂x1 ∂x2 ∂x3 
g3 · g3 = e +
1 1
e 2 + e3 · e 1 + e2 + e3 .
∂x ∂x2 ∂x3 ∂u3 ∂u3 ∂u3
∂u3 ∂x1 ∂u3 ∂x2 ∂u3 ∂x3
= + +
∂x1 ∂u3 ∂x2 ∂u3 ∂x3 ∂u3
∂u3
=
∂u3
= 1

Similarly,
 ∂u3 ∂u3 ∂u3   ∂x1 ∂x2 ∂x3 
g3 · g2 = e +
1 1
e 2 + e3 · e 1 + e2 + e3 .
∂x ∂x2 ∂x3 ∂u2 ∂u2 ∂u2
∂u3 ∂x1 ∂u3 ∂x2 ∂u3 ∂x3
= + +
∂x1 ∂u2 ∂x2 ∂u2 ∂x2 ∂u2
∂u3
=
∂u2
= 0

and g 3 · g 1 = 0. Essentially we have,

g i · g j = δji (21)

Therefore {g 1 , g 2 , g 3 } is the reciprocal basis of {g 1 , g 2 , g 3 } (Fig. 12).

Example 22 Find the reciprocal basis {g 1 , g 2 } for the polar coordinates, as shown
in Fig. 10.
We first find
p  x2 
r= (x1 )2 + (x2 )2 , θ = tan−1
x1
So,
∂r ∂r
g 1 = ∇r = e1 + e2
∂x1 ∂x2
x1 x2
= p e1 + p e2
(x1 )2 + (x2 )2 (x1 )2 + (x2 )2
= cos θe1 + sin θe2

AE 6104 33 Krishnendu Haldar

4.5 Physical interpretation of gij

Figure 12: Covariant {g 1 , g 2 , g 3 } and contravariant basis {g 1 , g 2 , g 3 } with respect

to a local coordinate system.

and
∂θ ∂θ
g 2 = ∇θ = 1
e1 + 2 e2
∂x ∂x
1
x  x2  x2  1 
= − p − 1 2 e1 + p − 1 e2
(x1 )2 + (x2 )2 (x ) (x1 )2 + (x2 )2 (x )
2 1
x x
= −p e1 + p e2
(x1 )2 + (x2 )2 (x1 )2 + (x2 )2
sin θ cos θ
= − e1 + e2
r r
1
Moreover, g 1 · g 1 = g 11 = 1, g 1 · g 2 = g 12 = g 21 = 0, g 2 · g 2 = g 22 = r2
.

4.5.1 Physical components of vectors

In general curvilinear coordinates, the components of vectors do not always have a
direct physical interpretation. The basis vector could have different dimensions and
so the components sincs their combination should be consistent with the dimension
of the given vector. For example, the velocity vector in the cylindrical polar system
can be written as

v = v r g r + v θ g θ + v z g z [LT −1 ]. (22)

AE 6104 34 Krishnendu Haldar

4.5 Physical interpretation of gij

Now,
∂r ∂r ∂r
gr = [·], g θ = [L], g z = [·].
∂r ∂θ ∂z
and so the components have the following dimensions

v r [LT −1 ], v θ [T −1 ], v z [LT −1 ].

Note that v r and v z have same units of velocity, but v θ has units of time inverse.
Moreover, if the dimensionless basis vectors are not unit vectors, the magnitude of
the components are not consistent with the observed or measured physical quan-
tities.
The physical components of a vector are the components that have physically
meaningful units and magnitudes. It is often convenient to derive the governing
equations for a problem in terms of vector components but to solve with the
physical components.
We now introduce the following nondimensional basis
gr gθ gz
ĝ r = , ĝ θ = , ĝ z =
|g r | |g θ | |g z |
or
g g g
ĝ r = √ r , ĝ θ = √ θ , ĝ z = √ z
grr gθθ gzz

So we write,

v = vr gr + vθ gθ + vz gz
√ √ √
= (v r grr )ĝ r + (v θ gθθ )ĝ θ + (v z gzz )ĝ z
= v̂ r ĝ r + v̂ θ ĝ θ + v̂ z ĝ z

Here,
√ √ √
v̂ r = v r grr , v̂ θ = v θ gθθ , v̂ z = v z gzz

are the physical components of v. We know that for cylindrical polar coordinate
system,

grr = g r · g r = 1, gθθ = g θ · g θ = r2 , gzz = g z · g z = 1

then

v̂ r = v r , v̂ θ = rv θ , v̂ z = v z

AE 6104 35 Krishnendu Haldar

4.6 Orthonormal set

You can also identify

g
ĝ r = √ r = cos θe1 + sin θe2 = êr
grr
g
ĝ θ = √ θ = − sin θe1 + cos θe2 = êθ
gθθ
gz
ĝ z = √ = 1.
gzz

In general notation we write

g √
v = v i g i = v̂ i ĝi , ĝi = √ i , v̂ i = v i g(ii)
g(ii)

Here, g(ii) means not summed over i. It simply says g(ii) = g i · g i (again, not
summed). Similarly,

gi p
v = vi g i = v̂i ĝ i , ĝ i = p , v̂i = vi g (ii)
g (ii)

Here, too, g (ii) means not summed over i. It simply says g (ii) = g i · g i .

4.6 Orthonormal set

Let V be an inner product space. S ⊂ V be an orthonormal set if any two distinct
vectors in S are orthonormal. If si , sj ∈ S, then

si · sj = δij

and |si | = 1. If the magnitude is not unity, S is called orthogonal.

Theorem 4.2 If S = {s1 , · · · , sn }, is any finite orthonormal set in V such that

n = dim V, then S forms a basis in V, known as orthonormal basis.
Proof: S ⊂ V be an orthonormal set of vectors. Consider

α1 s 1 + · · · + αn s n = 0

Taking dot product with sk , we write αk |sk |2 = 0. Note that k is not summed,
rather it is just one term. As |sk |2 = 1, we can further write αk = 0. So, Sm is
linearly independent.
Moreover, If v ∈ V, then v = αk sk = (v · sk )sk (k = 1, n). So any arbitrary
vector in V could be spanned by S.

AE 6104 36 Krishnendu Haldar

4.7 Orthogonal complement

4.7 Orthogonal complement

Let V be an inner product space and W is a subspace of V. The orthogonal
complement of W, written as W ⊥ and read as ‘W perpendicular’, is defined as
n o
W ⊥ := p ∈ V, p · s = 0, ∀s ∈ W .

Thus W ⊥ is the set of all vectors in ·V which are orthogonal to every vector in W.

Theorem 4.3 W ⊥ is a subspace of V.

Proof: Let p, q ∈ W ⊥ , and s ∈ W. Then by definition,

p · s = 0, q · s = 0.

Let α1 , α2 ∈ F. Then,

(α1 p + α2 q) · s = α1 p · s + α2 q · s = 0.

So, α1 p + α2 q ∈ W ⊥ ⊂ V. This implies that W ⊥ is a subspace.

Theorem 4.4 Projection theorem: Let W be any subspace of V. Then

W ⊕ W ⊥ = V.

Proof: Let dim ·V = n, and dim W = m. Let BW = {f 1 , · · · , f m } be an orthonor-

mal basis of W. Let v ∈ ·V, and (motivated by the Gram-Schmidt algorithm),
consider
m
X
h=v− (v · f i )f i .
i=1

Then, h · f 1 = v · f 1 − (v · f 1 ) = 0. Similarly, h · f 2 = · · · = h · f m = 0. This

implies,

h ⊥ {f 1 , · · · , f m } and h ∈ W ⊥ .

Now,
m
X m
X
v =h+ (v · f i )f i , where h ∈ W ⊥ , and (v · f i )f i ∈ W
i=1 i=1

This implies the linear sum

W + W ⊥ = V. (23)

AE 6104 37 Krishnendu Haldar

4.7 Orthogonal complement

Let u ∈ W ∪W ⊥ . This means u ∈ W, also u ∈ W ⊥ . Then by definition, u·u = 0,

implies u = 0. As u is arbitrary,

W ∩ W ⊥ = {0}. (24)

and equations (23), (24) imply

W ⊕ W ⊥ = ·V. (25)

Lemma 4.5 If W ⊕ W ⊥ = V, then every vector v ∈ V can be uniquely expressed

as

v = w + w⊥ .

where w ∈ W and w⊥ ∈ W ⊥ . w and w⊥ are then called orthogonal projections

of v on the subspaces W and W ⊥ .

AE 6104 38 Krishnendu Haldar

5 Homomorphism of vector spaces or linear trans-
formation
Definition 5.1 Let V and U be two vector spaces over the field F. Consider the
map

T :V→U

such that T v = u ∈ U, v ∈ V. The map T is called homomorphism or linear

transformation (LT) of V into U if

1. T (x + y) = T x + T y, x, y ∈ V.

2. T (ax) = aT x, a ∈ F, x ∈ V.

The above two conditions could be combined into a single condition as

T (ax + by) = aT x + bT y, a, b ∈ F x, y ∈ V (26)

This means, we first apply the operation on some elements in V and then map the
resulting element into U. It will be the same if we map those elements first into U
and then perform the operations in U.
A linear transformation T : V → U is called isomorphic if the mapping is
bijective3 , and we write V ∼
= U. T is known as a second order tensor that maps
vectors to vectors.

Example 23 Consider the relation between the instantaneous angular momentum

vector L, and angular velocity vector ω as

L = Iω.

The mass moment of inertia I is a second order that maps the vector ω to L.
These two vectors may not necessarily be in the same direction. A scalar multipli-
cation with a vector also gives a vector. However, like a tensor operation, scalar
multiplication can not change the direction of a vector. The above relation also
follows the linear relationship (26), i.e.,

I(αω1 + βω2 ) = αIω1 + βIω2

for any scalars α, β and for any vectors ω1 , ω2 .

3
Bijective or one-to-one correspondence, where each element of V maps to exactly one element
of U

AE 6104 39 Krishnendu Haldar

Example 24 The polarization vector P , and the electric field vector E are con-
nected as
P = ε0 χe E.
Here, ε0 , a scalar, is the electric permittivity of free space. χe , a scalar too, is the
electric susceptibility of the concerned medium. Note that P and E have the same
direction as ε0 χe is a scalar. However, some materials, e.g., liquid crystals, have
the capability to change the direction of the polarization from the applied electric
field. In this case electric susceptibility, χe , acts like a second order tensor, i.e.,
P = ε0 χe E.
The above relation also follows the linear relationship (26).

Example 25 The traction vector t, and the unit normal n to a plane (cut) at a
point are connected as
t = σn
where σ is the Cauchy stress and a second order tensor. The above relation also
follows the linear relationship (26).

Example 26 Show that T : V → U, where dim V = 3 and dim U = 2, defined by

 
v1  
v
T v2  = 1
v2
v3
is homomorphism of V into U.
Let,
   
x1 y1
x = x2  ,
 y = y2  .
 (27)
x3 y3
Then
     
ax1 by1 ax1 + by1  
ax 1 + by 1
T (ax + by) = T ax2  + by2  = T ax2 + by2  =
ax2 + by2
ax3 by3 ax3 + by3
   
        x1 y1
ax1 by1 x1 y1
= + =a +b = aT x2 + bT y2 
  
ax2 by2 x2 y2
x3 y3
= aT x + bT y
So, T is a homomorphism or a LT. Note that T behaves as a projection tensor
which projects a vector (v1 , v2 , v3 )t to a plane that contains (v1 , v2 )t components.

AE 6104 40 Krishnendu Haldar

5.1 Zero transformation

5.1 Zero transformation

0̂ is called a zero transformation if
0̂ : V → U, such that 0̂v = 0 ∈ U

5.2 Range space R(T )

Definition 5.2 Let V and U be two vector spaces over F and T be a homomor-
phism from V to U. Then range of T , written as R(T ), is basically the image set
I(T ) of the mapping T , i.e., I(T ) = R(T ) ⊆ U.
R(T ) := {T v = u ∈ U, v ∈ V} .
Lemma 5.1 R(T ) is a subspace of U

Figure 13: Range space, R(T ), of homomorphism T .

Proof: As V is a vector space, and v 1 , v 2 ∈ V (Fig. 13), then α1 v 1 + α2 v 2 ∈ V,

where α1 , α2 ∈ F. Now,
T (α1 v 1 + α2 v 2 ) = α1 T v 1 + α2 T v 2
= α1 u1 + α2 u2
where, u1 = T v 1 ∈ R(T ), and u2 = T v 2 ∈ R(T ). As, α1 v 1 + α2 v 2 is a vector in
V, after the mapping α1 u1 + α2 u2 belongs to R(T ). So, u1 , u2 ∈ R(T ) implies
α1 u1 + α2 u2 ∈ R(T ). This proves that R(T ) is a subspace of U (see equation
(14)).

5.3 Null space N (T ) or ker(T )

Definition 5.3 Let V and U be two vector spaces over F and T be a homomor-
phism from V to U. Then, the null space is defined as
N (T ) := {v ∈ V : T v = 0 ∈ U} ⊆ V.

AE 6104 41 Krishnendu Haldar

5.4 Rank and nullity of a homomorphism

It is also known as kernel of T or ker(T ).

Lemma 5.2 N (T ) is a subspace of V

Figure 14: Null space, N (T ) of homomorphism T .

Proof: Let v 1 , v 2 ∈ N (T ) ⊆ V (Fig. 14). Then, by definition, T v 1 = 0 and

T v 2 = 0. Let α1 , α2 ∈ F, then

T (α1 v 1 + α2 v 2 ) = α1 T v 1 + α2 T v 2
= 0+0=0

So, α1 v 1 + α2 v 2 ∈ N (T ), and N (T ) is a subspace of V (see equation (14)).

5.4 Rank and nullity of a homomorphism

Let T is a homomorphism from V to U. Then

Rank T = dim R(T ), Null T = dim N (T )

Rank T + Null T = dim V

5.5 Singular homomorphism

Let

T : V → U.

If any non zero v ∈ V gives T v = 0, i.e., v ∈ N (T ), and Null T = dim N (T ) ̸= 0,

then T is a singular homomorphism.

AE 6104 42 Krishnendu Haldar

5.6 Nonsingular homomorphism

5.6 Nonsingular homomorphism

Let

T : V → U.

If T v = 0 =⇒ v = 0, i.e., N (T ) = {0}, and Null T = dim N (T ) = 0, then T is

a nonsingular homomorphism.

5.7 Isomorphism
When T : V → U is bijective, we call it isomorphism. Moreover, T is isomorphic
if

1. T is nonsingular

2. dim V = dim U

5.8 Invertible T :
When T : V → U is an isomorphism, then T is invertible. We write,

T −1 : U → V

and if T v = u, then T −1 u = v. When T is invertible, it is also nonsingular and

dim V = dim U.

5.9 Linear operator

T becomes a linear operator if U = V, i.e.,

T : V → V.

5.10 Identity operator

I is called an identity operator if

I : V → V, such that Iv = v

5.11 Endomorphism
When T is an operator and isomorphic, i.e., T : V → V is bijective, and R(T ) = V.

AE 6104 43 Krishnendu Haldar

5.12 Representation of homomorphism or LT by a matrix

5.12 Representation of homomorphism or LT by a matrix

Let BV = {g 1 , · · · g n } be a covariant basis of V, and BU = {f 1 , · · · f m } be a
covariant basis for U. Given,

T : V → U.

If T acts on vectors, g j , then each n vectors T g j is uniquely expressible as a linear

combination of either covariant or contravariant basis of U. Let

T g j = Tij f i , then f i · T g j = Tij (28)

Similarly, if we express

T g j = T ij f i , thenf i · T g j = T ij . (29)

The remaining possibilities are

T g j = T·ji f i , then f i · T g j = T·ji (30)

T g j = Ti·j f i , then f i · T g j = Ti·j (31)

We write the matrix form of [T ]gi −f j as

 
T11 T12 T13
[T ]gi −f j = T21 T22 T23  . (32)
T31 T32 T33

AE 6104 44 Krishnendu Haldar

6 Homomorphism as vector space

Figure 15: Homomorphism as a vector space.

Consider the example of σn = t. If you keep changing the traction vector t,

say, by changing boundary conditions, the values of σ will keep changing without
changing n. So, we get a bunch of different second order tensors.
We now consider the set of all linear transformations from a vector space V to
U, and denote it Hom (V, U). We then impose addition and scalar multiplication
in Hom (V, U) over field F. Let T 1 and T 2 be two homomorphisms in Hom (V, U).
We define,

1. (T 1 + T 2 )v = T 1 v + T 1 v, where v 1 , v 2 ∈ F

2. (cT )v = c(T v), where c ∈ F

Closed under addition

We need to show that if T 1 , T 2 ∈ Hom (V, U) implies T 1 + T 2 ∈ Hom (V, U).
Consider, a vector α1 v 1 + α2 v 2 ∈ V. Then

(T 1 + T 2 )(α1 v 1 + α2 v 2 ) = T 1 (α1 v 1 + α2 v 2 ) + T 2 (α1 v 1 + α2 v 2 )

= α1 (T 1 v 1 + T 2 v 1 ) + α2 (T 1 v 2 + T 2 v 2 )
= α1 (T 1 + T 2 )v 1 + α2 (T 1 + T 2 )v 2

So, T 1 + T 2 ∈ Hom (V, U)

AE 6104 45 Krishnendu Haldar

6.1 Product of homomorphism

Closed under scalar multiplication

We need to show that if T ∈ Hom (V, U) implies cT ∈ Hom (V, U). Consider, a
vector α1 v 1 + α2 v 2 ∈ V. Then
cT (α1 v 1 + α2 v 2 ) = cT (α1 v 1 ) + cT (α2 v 2 )
= cα1 T v 1 + cα2 T v 2
= α1 (cT v 1 ) + α2 (cT v 2 )
So, cT ∈ Hom (V, U). Then Hom (V, U) becomes a vector space over field F
(Fig. 15). It satisfies all the vector space rules A1 to A5 , and B1 to B5 (closer
properties etc.).

6.1 Product of homomorphism

Figure 16: The composition mapping is called the product of homomorphism. We

simply write T 2 ◦ T 1 = T 2 T 1 .

Theorem 6.1 Let T 1 : V → U, and T 2 : U → W be the two homomorphisms.

Then the composition mapping T 2 ◦ T 1 ∈ Hom (V, W), i.e., again a homomor-
phism.
Consider a vector α1 v 1 + α2 v 2 ∈ V. Then
T 2 ◦ T 1 (α1 v 1 + α2 v 2 ) = T 2 (T 1 (α1 v 1 + α2 v 2 ))
= T 2 (α1 T 1 v 1 + α2 T 1 v 2 )
= α1 T 2 T 1 v 1 + α2 T 2 T 1 v 2
= α1 (T 2 ◦ T 1 )v 1 + α2 (T 2 ◦ T 1 )v 2

AE 6104 46 Krishnendu Haldar

6.2 Algebra of linear operators

So, T 2 ◦ T 1 ∈ Hom (V, W), and we simply write the product T 2 T 1 , i.e, (T 2 T 1 )v =
T 2 (T 1 v). Commutative diagrams are very useful to represent composition map-
pings. We can represent the above mentioned composition mapping as
T1 /
V U . (33)

T2
T 2 ◦T 1 =T 2 T 1

 
W

6.2 Algebra of linear operators

Theorem 6.2 Let us consider the vector space Hom (V, V). Such a structure is
called an ‘algebra’ as it is equipped with the multiplication of vectors. If A, B, C
∈ Hom (V, V) (linear operators), then the following additional product rules hold
true for an algebraic structure on the top of a vector space structure:
1. Closed under multiplication: AB ∈ Hom (V, V)
2. Multiplication is associative: A(BC) = (AB)C
3. Multiplication is distributive with respect to addition: A(B+C) = AB+AC
and (A + B)C = AC + BC
4. For each scalar α ∈ F, α(AB) = (αA)B = A(αB).
If there is an element I ∈ Hom (V, V) such that IT = T = T I for each T ∈
Hom (V, V), we call Hom (V, V) a (linear) algebra with identity and call I the
identity of the algebra.
Proof: We have the following three rules to prove these relations:
i (A + B)v = Av + Bv
ii (αA)v = αAv
iii (AB)v = A(Bv)
where v ∈ V. To prove that the multiplication is associative, we start with
A(BC)v = A(B(Cv)) = (AB)Cv
=⇒ A(BC) = (AB)C
Try yourself to prove the other relations!

AE 6104 47 Krishnendu Haldar

7 Tensor product
Definition 7.1 For any vectors v ∈ V and u ∈ U, the tensor product v ⊗ u is
also defined as a map

v ⊗ u : U → V.

such that

(v ⊗ u)w = (u · w)v

for w ∈ U. Then the mapping is homomorphism (try it yourself )4 , and {g i ⊗ f j }

is a basis of Hom (U, V).

Linearly independent
Consider cij (g i ⊗ f j ) = 0 ⊗ 0 ∈ Hom (U, V). Then

cij (g i ⊗ f j )f m = (0 ⊗ 0)f m
cij [f j · f m ]g i = 0
cij δjm g i = 0
cim g i = 0

Now for each m, cim = 0 as {g i } is a basis. So, g i ⊗ f j is linearly independent.

Span the space

Let v = v i g i and u = uj f j . Then

(v ⊗ u)w = [u · w]v = uj wj v = uj wj v i g i

We select w = wj f j for the convenience of dot product to avoid the metric (fij or
f ij ) of U. Moreover,

(g i ⊗ f j )u = [f j · u]g i = uj g i

Substituting back in the first expression, we write

(v ⊗ u)w = uj wj v i g i = v i wj (uj g i ) = v i wj (g i ⊗ f j )u∗ = (v i wj g i ⊗ f j )w

As w is arbitrary in the above expression, we finally write

v ⊗ u = v i uj g i ⊗ f j . (34)
4
Take an element αw1 + βw2 ∈ U and proceed

AE 6104 48 Krishnendu Haldar

and so gi ⊗ f j spans the space Hom (U, V). We also denote the same space as
V ⊗ U. We thus write
v ⊗ u ∈ V ⊗ U = Hom (U, V).
Moreover, the total number of basis elements are dim V · dim U = dim(V ⊗ U) =
dim Hom (U, V). Considering all the possible combinations of contravariant and
covariant basis of V and U, we can write a same tensor
v ⊗ u = v i uj g i ⊗ f j ∈ V ⊗ U
= vi uj g i ⊗ f j ∈ V ⊗ U
= vi uj g i ⊗ f j ∈ V ⊗ U
= v i uj g i ⊗ f j ∈ V ⊗ U
Note that not all tensors can be represented as a product of two vectors. However
if T ∈ V ⊗ U, it can be expressed with one of the basis {g i ⊗ f j } as

T = T ij g i ⊗ f j .
We can also write the same tensor T as
T = Tij g i ⊗ f j ∈ V ⊗ U
= Ti·j g i ⊗ f j ∈ V ⊗ U
= T·ji g i ⊗ f j ∈ V ⊗ U (35)

Note that Ti·j means i is the first index. A dot (·) is used to denote the second
index. T ij , Tij , Ti·j , and T·ji are the different components with respect to the
associated basis of a same tensor. Also note that Tj·i ̸= T·ji .

Example 27 Let u ⊗ v ∈ U ⊗ V, and v 1 ⊗ w ∈ V ⊗ W. Then

(u ⊗ v)(v 1 ⊗ w) = (u ⊗ w)[v · v 1 ] (36)
Ans: Note that the left hand side is the composition mapping or product of two
homomorphisms. Consider the following commutative diagram,
v 1 ⊗w /
W V (37)

u⊗v
(u⊗v )◦(v 1 ⊗w )=(u⊗v )(v 1 ⊗w )

 
U

AE 6104 49 Krishnendu Haldar

7.1 Matrix representation of tensor products

If p ∈ W, then
(u ⊗ v)(v 1 ⊗ w)p = (u ⊗ v) ◦ (v 1 ⊗ w)p = (u ⊗ v)((v 1 ⊗ w)p)
= (u ⊗ v)([w · p]v 1 )
= u[v 1 · v][w · p]
= (u[w · p])[v · v 1 ]
= ((u ⊗ w)p)[v · v 1 ]
 
= (u ⊗ w)[v · v 1 ] p

7.1 Matrix representation of tensor products

Consider the tensor
T = Tij f i ⊗ g j ∈ U ⊗ V. (38)
We assume dim U = dim V = 2. The basis {f 1 , f 2 } ∈ U, and the basis {g 1 , g 2 } ∈
V. Also note that the basis {f 1 , f 2 } ∈ U, and the basis {g 1 , g 2 } ∈ V. Tij are the
components of the tensor in (38), and for this particular case we can write
T = T11 f 1 ⊗ g 1 + T12 f 1 ⊗ g 2 + T21 f 2 ⊗ g 1 + T22 f 2 ⊗ g 2 .
So in the matrix form,
 
T11 T12
[T ]f i ⊗gj = . (39)
T21 T22
Similarly, if we consider u ⊗ v = ui vj f i ⊗ g j , then the matrix form reads
 
u1 v1 u1 v2
[u ⊗ v]f i ⊗gj = .
u2 v1 u2 v2

7.2 Tensor spaces: a general multilinear function

The vector spaces in multilinear mapping only with V are called tensor spaces. In
general a multilinear function on V × ... × V × V × ... × V forms a tensor of type
(r, s), r-contravariant and s-covariant, denoted by
Tsr = V ⊗ ... ⊗ V ⊗ V ⊗ ... ⊗ V .
| {z } | {z }
r times s times

A tensor of type (r, 0) is called contravariant tensor and one of type (0, s) is called
covariant tensor. Suppose {g i } and {g i } are the contravariant and covariant bases
of V respectively, then
g i1 ⊗ · · · ⊗ g ir ⊗ g j1 ⊗ · · · ⊗ g js

AE 6104 50 Krishnendu Haldar

7.3 Tensor multiplication

form a basis of Tsr . Therefore an (r, s)-type tensor T can be uniquely expressed as

T = T·ij11...i j1 js
...j s g i1 ⊗ · · · ⊗ g ir ⊗ g ⊗ · · · ⊗ g ∈ Ts
r r

Note that this is one particular format that is often useful. You have to start with
covariant basis and contravariant basis will come after them. You can not have a
suffling of covariant and contravariant basis.

Example 28 T ∈ T12 implies T = T·kij g i ⊗ g j ⊗ g k .

7.3 Tensor multiplication

The tensor multiplication of tensors from Tsr11 and Tsr22 is defined as a mapping

Tsr11 ⊗ Tsr22 → Tsr11+s

+r2
2

So, if T ∈ Tsr11 , and S ∈ Tsr22 , then

T ⊗ S ∈ Tsr11+s
+r2
2

This means if
i ...i
T = T· j11 ...jr1s1 g i1 ⊗ · · · ⊗ g ir1 ⊗ g j1 ⊗ · · · ⊗ g js1 ∈ Tsr11

and
m ...m
S = S· n11 ...nsr22 g m1 ⊗ · · · ⊗ g mr2 ⊗ g n1 ⊗ · · · ⊗ g ns2 ∈ Tsr22

then
i ...i m ...m
T ⊗ S = T· j11 ...jr1s1 S· n11 ...nsr22 g i1 ⊗ · · · ⊗ g ir1 ⊗ g m1 ⊗ · · · ⊗ g mr2
⊗g j1 ⊗ · · · ⊗ g js1 ⊗ g n1 ⊗ · · · ⊗ g ns2 ∈ Tsr11+s
+r2
2
.

The following relations now hold true.

1. Tensor multiplication is associative: Let A ∈ Tsr11 , B ∈ Tsr22 , and C ∈

Tsr33 . Then

A ⊗ (B ⊗ C) = (A ⊗ B) ⊗ C ∈ Tsr11+s
+r2 +r3
2 +s3

2. Tensor multiplication is distributive with respect to addition: Let

A ∈ Tsr11 , B ∈ Tsr22 , and C ∈ Tsr22 . Then

A ⊗ (B + C) = A ⊗ B + A ⊗ C ∈ Tsr11+s
+r2
2

AE 6104 51 Krishnendu Haldar

7.4 Contraction

3. Tensor multiplication is distributive with respect to addition: Let

A ∈ Tsr11 , B ∈ Tsr11 , and C ∈ Tsr22 . Then

(A + B) ⊗ C = A ⊗ C + B ⊗ C ∈ Tsr11+s
+r2
2

4. For each α ∈ F:

α(A ⊗ B) = αA ⊗ B = A ⊗ αB ∈ Tsr11+s
+r2
2

when A ∈ Tsr11 and B ∈ Tsr22 .

The vector space equipped with such an addition and multiplication is called tensor
algebra.

7.4 Contraction
Consider tensors of type Tsr or (r, s) such that r, s ≥ 1.

Definition 7.2 The contraction of a tensor is an operation which equalizes a con-

travariant index and a covariant index and sum over the repeated index.
We can also think of contraction as a mapping

[·] : Tsr → Ts−1

r−1

r−1
such that a tensor space Tsr becomes Ts−1 . For example, let us consider the tensor

T = T·kmp g m ⊗ g p ⊗ g k ∈ T12

Contracting p and k we obtain a tensor um such that

·T = T·kmp g m [g p · g k ] = T·kmp g m [δpk ] = (T·pmp )g m = um g m ∈ T01

Note that:

1. Every contraction of a tensor removes one contravariance and one covariance.

2. After q contractions a tensor of type (q, q) is reduced to a tensor (0, 0).

One needs to be specific about the contracting basis elements if multiple options
are there. Otherwise, one should start contracting from the last.
A double contraction is a mapping

[:] : Tsr → Ts−2

r−2

AE 6104 52 Krishnendu Haldar

7.4 Contraction

r−2
such that a tensor space Tsr becomes Ts−2 . The symbol ‘:’ is used when two
contractions take place. Consider,
mp
T = T·kl g m ⊗ g p ⊗ g k ⊗ g l ∈ T22

Then
mp mp m
: T = T·kl [g m · g k ][g p · g l ] = T·kl [δk ][δlp ] = T·kl
kl
∈ T00

A n-contraction is a mapping
r−n
[•] : Tsr → Ts−n
r−n
such that a tensor space Tsr becomes Ts−n . Unfortunately, the symbol ‘•’ (or
sometimes :) is used in most of the time beyond double contraction. You need to
understand it from the context.
Single contraction between two tensors are defined in the following way. First
we will consider for simplicity a pure contravariant (r1 , 0) and a pure covariant
tensor (0, s2 ) such that
−1
[·] : T0r1 × Ts02 → Tsr21−1

A single dot ‘·’ is used for a single contraction. For example, if T = T ijk g i ⊗g j ⊗g k
and S = Sqr g q ⊗g r , then, a single contraction between r and k (start from extreme
right) is given by

T · S = (T ijk g i ⊗ g j ⊗ g k ) · (Sqr g q ⊗ g r )
= T ijk Sqr g i ⊗ g j ⊗ g q [g k · g r ]
= T ijk Sqr g i ⊗ g j ⊗ g q [δrk ]
= T ijk Sqk g i ⊗ g j ⊗ g q

If nothing is mentioned, it is a standard practice to start from the last member.

A double contraction between two pure contravariant (r1 , 0) and a pure covariant
tensor (0, s2 ) such that
−2
[:] : T0r1 × Ts02 → Tsr21−2

Continuing the previous example we write

T : S = (T ijk g i ⊗ g j ⊗ g k ) : (Sqr g q ⊗ g r )
= T ijk Sqr g i [g j · g q ][g k · g r ]
= T ijk Sqr g i [δjq ][δrk ]
= T ijk Sjk g i

AE 6104 53 Krishnendu Haldar

7.4 Contraction

A generalized full contraction is given by

[•] : T0r1 × Ts02 → T0r1 −s2 if r1 > s2 or Ts02 −r1 if s2 > r1 .

If T = T ijkl g i ⊗ g j ⊗ g k ⊗ g l and S = Sqr g q ⊗ g r , then

T • S = (T ijkl g i ⊗ g j ⊗ g k ⊗ g l ) : (Sqr g q ⊗ g r )
= T ijkl Sqr g i ⊗ g j [g k · g q ][g l · g r ]
= T ijkl Sqr g i ⊗ g j [δkq ][δlr ]
= T ijkl Skl g i ⊗ g j

For example, if T = u ⊗ v ⊗ w ⊗ x ⊗ y and S = a ⊗ b ⊗ c, where a, b, c ∈ V and

also u, v, w, x, y ∈ V, then a 3-contraction (generalized) could be written as

(u ⊗ v ⊗ w ⊗ x ⊗ y) • (a ⊗ b ⊗ c) = u ⊗ v[w · a][x · b][y · c]

However, as explained in Example (27),

(u ⊗ v ⊗ w ⊗ x ⊗ y)(a ⊗ b ⊗ c) = u ⊗ v ⊗ w ⊗ x ⊗ b ⊗ c[y · a]

For a, b, c, d, x, y ∈ V we write
1. (a ⊗ b)c = (b · c)a
2. (a ⊗ b) : (c ⊗ d) = (a · c)(b · d)
3. (a ⊗ b)(c ⊗ d) = (a ⊗ d)(b · c)
4. (a ⊗ b ⊗ c)d = (a ⊗ b)(c · d)
5. (a ⊗ b ⊗ c) : x ⊗ y = a(b · x)(c · y)
We already know from (35), T ∈ V ⊗ V, can be written in four possible ways.
Among those four, we consider elements for the tensor spaces in the form T02 , T20
and T11 . Then,

T = T ij g i ⊗ g j ∈ T02
= Tij g i ⊗ g j ∈ T20
= T·ji g i ⊗ g j ∈ T11 (40)

Note that Ti·j g i ⊗ g j does not belong to any tensor space. T ij and T·ji are also
called associate tensor components of Tij .If we consider T = T ij g i ⊗ g j , then

T g k = T ij (g i ⊗ g j )g k = T ij (g j · g k )g i = T ij δjk g i = T ik g i
=⇒ g m · T g k = g m · T ik g i = T ik g m · g i = T ik δim = T mk

AE 6104 54 Krishnendu Haldar

7.4 Contraction

In summary,

g i · T g j = T ij , g i · T g j = Tij , g i · T g j = T·ji (41)

Similarly,

g i · T g j = T ij , (g im g m ) · T (g jn g n ) = T ij , g im g jn (g m · T g n ) = T ij
=⇒ g im g jn Tmn = T ij .

So, knowing the components with respect to a basis, one can directly calculate
components with respect to other basis through gij and g ij . In summary,

g im g jn Tmn = T ij , gim gjn T mn = T ij , gim T·ji = Tmj , · · ·

Find the rest of the relations.

7.4.1 Double contraction as a bilinear functional

Consider the following double contraction of two second order tensors.

u ⊗ v : a ⊗ b = [u · a][v · b].

and we write it, to be consistent with our notation at Section 4.2,

u ⊗ v⟨a, b⟩ = [u · a][v · b].

where L = u ⊗ v and the double contraction := ⟨, ⟩. We now write,

g⟨a, b⟩ = g⟨ai g i , bj g j ⟩ = ai bj g⟨g i , g j ⟩ = ai bj gij = gij [a · g i ][b · g j ]

= gij g i ⊗ g j : a ⊗ b
= gij g i ⊗ g j ⟨a, b⟩

As a and b are arbitrary, we can write

g = gij g i ⊗ g j .

In a similar way, we also have the other three versions of g s:

g = g ij g i ⊗ g j
= δ·ji g i ⊗ g j = g i ⊗ g i

You operate g on any vector v i g i or vi g i , you will get that vector back, i.e., gv = v.
So g is indeed an identity tensor!

AE 6104 55 Krishnendu Haldar

7.5 Transposition of a tensor

7.4.2 Physical components of tensors

The motivation for physical components is discussed in the section of ‘Physical
components of tensors’. We write

g gi
ĝ i = √ i , ĝ i = p .
g(ii) g (ii)
Then the physical component of tensors could be identified as follows:
√ √
T = T ij g i ⊗ g j = (T ij g(ii) g(jj) )ĝ i ⊗ ĝ j = T̂ ij ĝ i ⊗ ĝ j
p p
= Tij g i ⊗ g j = (Tij g (ii) g (jj) )ĝ i ⊗ ĝ j = T̂ij ĝ i ⊗ ĝ j
√ p
= T·ji g i ⊗ g j = (T·ji g(ii) g (jj) )ĝ i ⊗ ĝ j = T̂·ji ĝ i ⊗ ĝ j

7.5 Transposition of a tensor

Figure 17: Transpose of a linear operator in inner product space.

The notion of transpose is similar to what we studied in the dual space. How-
ever, we have only one vector space V. Consider,

T :V→V

such that v and T v belong to V. Let a and u be the any other two vectors such
that v · a = T v · u. Then the transpose is the mapping T t u = a, and satisfies the
relation (Fig. 17)

v · T t u = T v · u. (42)

Example 29 Show that (a⊗b)t = (b⊗a), where a, b ∈ V and a⊗b ∈ Hom (V, V).

AE 6104 56 Krishnendu Haldar

7.6 Invariants in 3D

We start with the definition (42)

v · (a ⊗ b)t u = (a ⊗ b)v · u
= (b · v)a · u
= (a · u)b · v
= (b ⊗ a)u · v
= v · (b ⊗ a)u (43)

As u and v are arbitrary, we conclude (a ⊗ b)t = (b ⊗ a).

Now we can write the transpose of (40).

T t = (T ij g i ⊗ g j )t = T ij g j ⊗ g i = T ji g i ⊗ g j = T t ij g i ⊗ g j , =⇒ T t ij = T ji
T t = (Tij g i ⊗ g j )t = Tij g j ⊗ g i = Tji g i ⊗ g j = Tijt g i ⊗ g j , =⇒ Tijt = Tji
T t = (T·ji g i ⊗ g j )t = T·ji g j ⊗ g i = T·ij g i ⊗ g j = Tit ·j g i ⊗ g j .

Note that the component for of the last expression is not admissible as g i ⊗ g j
does not belong to any admissible tensor space.

7.5.1 Symmetry and skew-symmetry of second order tensor

We write for the symmetry T t = T and for skew-symmetric T t = −T . When we
consider

T = T ij g i ⊗ g j ∈ T02 , then T t = T t ij g i ⊗ g j = T ji g i ⊗ g j ∈ T02

So, for this basis, T t = T =⇒ T ji = T ij . Similarly,

T = Tij g i ⊗ g j ∈ T20 , then T t = Tijt g i ⊗ g j = Tji g i ⊗ g j ∈ T20

So, for this basis, T t = T =⇒ Tji = Tij . Now consider

T = T·ji g i ⊗ g j ∈ T11 , then Tit ·j g i ⊗ g j = T·ij g i ⊗ g j ̸∈ T11

This means the symmetry is not define due to the exchange of covariant and
contravariant basis. For a mixed basis, the transpose creates elements which do
not reside in the initial tensor space!

7.6 Invariants in 3D
Consider T ∈ Hom (V, V) such that dim V = 3. We can immediately write the
three invariants as
[T u, v, w] + [u, T v, w] + [u, v, T w]
= I1 = tr T (44)
[u, v, w]

AE 6104 57 Krishnendu Haldar

 [T u, T v, w] + [u, T v, T w] + [T u, v, T w] 1
= I2 = [(tr T )2 − tr T 2 ] (45)
[u, v, w] 2

[T u, T v, T w]
= I3 = det T (46)
[u, v, w]
where [a, b, c] = a · (b × c).

7.6.1 Cofactor tensor

Definition 7.3 Let T ∈ Hom(V, V), and u, v ∈ V. Then the cofactor tensor,
denoted by Cof T is defined by5

Cof T (u × v) := T u × T v. (47)

Example 30 Show that (det T )I = (Cof T )t T

We start from (46) and write
[T u, T v, T w] T u · (T v × T w) T u · Cof T (v × w)
det T = = =
[u, v, w] u · (v × w) u · (v × w)
t
⇒ (det T )(u · (v × w)) = T u · Cof T (v × w) = (Cof T ) T u · (v × w)
h i h i
⇒ (det T )I u · (v × w) = (Cof T )t T u · (v × w)

The last line gives

(det T )I = (Cof T )t T . (48)

8 Orthogonal tensors
Let R ∈ Hom (V, V) and dim V = 3 such that Rt = R−1 or alternatively

RRt = Rt R = I. (49)

Then R is said to be orthogonal tensor. It has the following properties

5
The components are the same as the cofactor of a matrix. If
 
  A22 A23 A21 A23 A21 A22
A11 A12 A13  A32 A33 − A31 A33 A31 A32 
[Aij ] = A21 A22 A23  then [Cof A]ij =  
 ... ... ... 
A31 A32 A33
... ... ...

AE 6104 58 Krishnendu Haldar

 1. It preserves the length two vectors after transformation (Fig. 18):
Ru · Ru = u · Rt Ru = u · Iu = u · u. This means |Ru| = |u|.

2. It preserves the angle between two vectors after transformation (Fig. 18):
Ru · Rv = u · Rt Rv = u · Iv = u · v. So,
Ru · Rv u·v
cos θ = =
|Ru||Rv| |u||v|

3. As RRt = Rt R = I, det(RRt ) = det I implies (det R)2 = 1 or det R = ±1.

Qu
v
θ
Qv
θ u

(a) (b)

Figure 18: θ and the lengths of the vectors remain the same after an orthogonal
transformation.

When det R = 1, we call it proper orthogonal. It represents rotations, and we call

n o
+ t t
R ∈ Orth := R ∈ Hom (V, V), RR = R R = I, det R = 1

When det R = −1, we call it improper orthogonal. It represents reflections, and

we call
n o
R ∈ Orth− := R ∈ Hom (V, V), RRt = Rt R = I, det R = −1

Introducing rotation tensor as R = cos({e}, {e′ }) ∈ Orth+ (Rij is the cosine of

the angle between ei and e′j axes), we have the following transformations:

1. for a vector v: v ′ = Rv

AE 6104 59 Krishnendu Haldar

 2. for a second order tensor T = v ⊗ u: T ′ = v ′ ⊗ u′ = Rv ⊗ Ru = R(v ⊗
u)Rt = RT Rt .

3. for a n-th order tensor: T ′ = Rv 1 ⊗ Rv 2 ⊗ · · · ⊗ Rv n . It is difficult to write

it in a compact form like second order tensor. However, in the component
form we can write Ti′1 i2 ...in = Ri1 j1 Ri2 j2 . . . Rin jn Tj1 j2 ...jn .

9 Skew-symmetric tensors

u
Wu = w × u

Figure 19: Axial or dual vector.

Let W ∈ Skw and dim ·V = 3 such that W t = −W . Then W has the

following properties

1. u · W v = W t u · v = −v · W u.

2. If v = u, then from the above relation u · W u = 0. This means W u ⊥ u.

3. There exist a vector ω such that ω × u = W u. When ω is given, W is

fixed and vice versa. ω is known as axial or dual vector corresponding to
W (Fig. 19).

If ω = (ω1 , ω2 , ω3 )t , then
 
0 −ω3 ω2
[W ]ij =  ω3 0 −ω1  (50)
−ω2 ω1 0

AE 6104 60 Krishnendu Haldar

10 Spectral decomposition of a second order ten-
sor
Let T ∈ Hom (V, V), where dim V = 3. If n be the unit right eigenvector, by
definition we write

T n = λn or (T − λg)n = 0. (51)

where g is the identity tensor. As we know that there are three possibilities to
express T and g, i.e.,

T = T ij g i ⊗ g j , T = Tij g i ⊗ g j , T = T·ji g i ⊗ g j .
g = g ij g i ⊗ g j , g = gij g i ⊗ g j , g = δ·ji g i ⊗ g j = g i ⊗ g i ,

Note that the matrix representations of the last two forms of I are diagonal. Now
going back to the eigenvalue problem, we have to make sure that both T and I
should be expressed with respect to the same basis for the subtraction operation.
Let us consider the form T = Tij g i ⊗ g j (also Tij = Tji ), then we should consider
I = gij g i ⊗ g j . Let n = nk g k (we can also consider n = nk g k ), then
h i
(T − λg)n = (Tij − λgij )g i ⊗ g j nk g k
= (Tij − λgij )nk δkj g i
= (Tij − λgij )nj g i

Thus from (51) we get

(Tij − λgij )nj = 0

The above form is known as generalized eigenvalue problem, i.e., in the form of
An = λBv. However, solving a simple eigenvalue problem is convenient. So we
start with T = T·ji g i ⊗ g j ∈ T11 and g = δ·ji g i ⊗ g j ∈ T11 and n = nk g k such that

(T·ji g i ⊗ g j − λδ·ji g i ⊗ g j )nk g k = 0

(T·ji − λδ·ji )nj g i = 0

The condition for non-zero eigenvectors is

det(T·ji − λδ·ji ) = 0

and the characteristics polynomial is written as:

P(λ) = λ3 − I1 λ2 + I2 λ − I3 ,

AE 6104 61 Krishnendu Haldar

where

I1 = tr T = tr [T·ii ] = T·ji + T·ji + T·ji = λ1 + λ2 + λ3

1 1
I2 = [tr (T )2 − tr T 2 ] = [[T·ii ]2 − [T·ji T·ij ]] = λ1 λ2 + λ2 λ3 + λ3 λ1
2 2
I3 = det T = det[T·ji ] = λ1 λ2 λ3

The Cayley-Hamilton theorem in 3D is written as

T 3 − I1 T 2 + I2 T − I3 g = 0.

Operating on n, one can write (λ3 −I1 λ2 +I2 λ−I3 )n = 0 or λ3 −I1 λ2 +I2 λ−I3 = 0
as n ̸= 0. So, T satisfies its own characteristic equation λ3 − I1 λ2 + I2 λ − I3 = 0.
Let m = mk g k be the unit left eigenvector with an eigenvalue µ. We then
write, by definition,

m · T = µm or (T t − µg)m = 0. (52)

Then the characteristic polynomial becomes

det(T t − µg) = det(T − µg)t = det(T − µg) = 0 (53)

Moreover, as previously obtained, det(T − λg) = 0 implies λ = µ, i.e., the eigen-

values of the left and right eigenvectors (they are different) are equal.

Theorem 10.1 Let T ∈ Hom (V, V), and dim V = 3. If T has three distinct eigenvalues,
and n(i) = nk(i) g k is the right eigenvector corresponding to i-th eigenvalue, then
{n(1) , n(2) , n(3) } is a basis of V. The eigenvectors n(i) are covariant in nature and
they need not to be orthonormal.
Consider,

c1 n(1) + c2 n(2) + c3 n(3) = 0

Operating both sides with (T − λ1 g) we get

c1 (T − λ1 g)n(1) + c2 (T − λ1 g)n(2) + c3 (T − λ1 g)n(3) = 0

As n(1) is the eigenvalue of λ1 , we have (T − λ1 g)n(1) = 0. We write,

c2 (T n(2) − λ1 n(2) ) + c3 (T n(3) − λ1 n(3) ) = 0

⇒ c2 (λ2 n(2) − λ1 n(2) ) + c3 (λ3 n(3) − λ1 n(3) ) = 0
⇒ c2 (λ2 − λ1 )n(2) + c3 (λ3 − λ1 )n(3) = 0

AE 6104 62 Krishnendu Haldar

Operating the last equation with (T − λ2 g) we get

c2 (λ2 − λ1 )(T − λ2 g)n(2) + c3 (λ3 − λ1 )(T − λ2 g)n(3) = 0

As n(2) is the eigenvalue of λ2 , we have (T − λ2 g)n(2) = 0. We write,

c3 (λ3 − λ1 )(T − λ2 g)n(3) = 0

c3 (λ3 − λ1 )(T n(3) − λ2 n(3) ) = 0
c3 (λ3 − λ1 )(λ3 − λ2 )n(3) = 0

Now n(3) is not a null vector, and all the eigenvalues are distinct, i.e., λ1 ̸= λ2 ̸= λ3 .
This means c3 = 0. Similarly, starting with the operators (T −λ2 g), and (T −λ3 g)
give c1 = 0. And starting with the operators (T − λ1 g), and (T − λ3 g) give c2 = 0.
This means {n(1) , n(2) , n(3) } is linearly independent. As theh dimension of the
space is 3, we have {n(1) , n(2) , n(3) } is a basis.
Similarly, we can write

Theorem 10.2 Let T ∈ Hom (V, V), and dim V = 3. If T has three distinct eigenvalues,
(i)
and m(i) = mk g k is the right eigenvector corresponding to i-th eigenvalue, then
{m(1) , m(2) , m(3) } is a basis of V. The eigenvectors m(i) are contravariant in
nature and they need not to be orthonormal.

Theorem 10.3 Let T ∈ Hom (V, V), where dim V = 3. Then any two distinct
left and right eigenvectors are biorthonormal. If T has three distinct eigenvalues,
then

m(i) · n(j) = δji , and m(i) = n(i)

i.e., the left eigenvectors are the reciprocal basis of the right eigenvalues.
Let T n(i) = λi n(i) . Now,

m(j) · T n(i) = m(j) · λi n(i) , ⇒ T t m(j) · n(i) = λi m(j) · n(i) ,

⇒ λj m(j) · n(i) = λi m(j) · n(i) , ⇒ (λj − λi )m(j) · n(i) = 0.

We write,

λj m(j) · n(i) = λi m(j) · n(i) , ⇒ (λj − λi )m(j) · n(i) = 0.

As λi ̸= λj , we get m(j) ⊥ n(i) . For i = j, from above we can write m(i) · n(i) ̸= 0
(not summed), i.e., m(i) · n(i) can take any value. We select m(i) · n(i) = 1. This
is a choice, not autometic. So we write,

m(j) · n(i) = δij and identify m(j) = n(j)

AE 6104 63 Krishnendu Haldar

10.1 Spectral decomposition of a second order tensor

as reciprocal basis of {n(1) , n(2) , n(3) }. Note that if we already normalized the left
and right eigenvectors, to satisfy m(i) · n(i) = 1, we need to rescale the magnitude
again and after that they need not to be a unit vector again. As direction is the
only concern of the eigenvectors, such a rescaling of the magnitude does not affect
the results.

Theorem 10.4 Let T ∈ Hom (V, V), where dim V = 3 and T ∈ Sym . Then
any two distinct left and right eigenvectors are equal (covariant). Moreover, the
eigenvectors are mutually orthonormal. If T has three distinct eigenvalues, then
n(i) · n(j) = δij . That is m(i) = n(i)
We have
T n = λn and T t m = λm ⇒ T m = λm
This shows m = n, i.e, left and right eigenvectors are equal and covariant type.
Now, consider,
T n(i) = λi n(i) , and T n(j) = λj n(j) , (i ̸= j)
Now,
n(j) · T n(i) = n(j) · λi n(i) , ⇒ T t n(j) · n(i) = λi n(j) · n(i) ⇒ T n(j) · n(i) = λi n(j) · n(i) ,
⇒ λj n(j) · n(i) = λi n(j) · n(i) , ⇒ (λj − λi )n(j) · n(i) = 0.
We write,
λj n(j) · n(i) = λi n(j) · n(i) , ⇒ (λj − λi )n(j) · n(i) = 0.
As λi ̸= λj , we get n(j) ⊥ n(i) . For i = j, from above we can write n(i) · n(i) =
1 (not summed), as we already normalized the eigenvectors. Hence the result
n(i) · n(j) = δij .

10.1 Spectral decomposition of a second order tensor

For three distinct eigenvalues, when {n(1) , n(2) , n(3) } are the three eigenvectors,
we write the tensor T as
T = λ1 n(1) ⊗ n(1) + λ2 n(2) ⊗ n(2) + λ3 n(3) ⊗ n(3) . (54)
This is called spectral decomposition of T with respect to the mixed basis system
{n(i) ⊗ n(j) }6 . Note that the matrix form of T will be diagonal in this case. You
6
For a symmetric T , we write
T = λ1 n(1) ⊗ n(1) + λ2 n(2) ⊗ n(2) + λ3 n(3) ⊗ n(3) . (55)

AE 6104 64 Krishnendu Haldar

10.1 Spectral decomposition of a second order tensor

can check that T n(i) = λi n(i) and n(i) · T = λi n(i) . One of the advantages, among
many, of spectral decomposition is the following:

T 2 = (λ1 n(1) ⊗ n(1) + λ2 n(2) ⊗ n(2) + λ3 n(3) ⊗ n(3) )(λ1 n(1) ⊗ n(1)
+λ2 n(2) ⊗ n(2) + λ3 n(3) ⊗ n(3) )
= λ21 [n(1) (n(1) · n(1) ) ⊗ n(1) ] + λ22 [n(2) (n(2) · n(2) ) ⊗ n(2) ] + λ23 [n(3) (n(3) · n(3) ) ⊗ n(3) ]
= λ21 n(1) ⊗ n(1) + λ22 n(2) ⊗ n(2) + λ23 n(3) ⊗ n(3)

In general,

T n = λn1 n(1) ⊗ n(1) + λn2 n(2) ⊗ n(2) + λn3 n(3) ⊗ n(3) , (n ∈ R). (56)

We will come back to this aspect again very soon.

10.1.1 Two repeated eigenvalues

Consider λ2 = λ3 = λ ̸= λ1 . Then we have,

T n(1) = λ1 n(1) , T n(2) = λn(2) , T n(3) = λn(3) .

We also have,

n(1) ⊥ n(2) and n(1) ⊥ n(3) .

If we consider n̂ = αn(2) + βn(3) , it also satisfy T n̂ = λn̂. We can show this in

the following ways:

T (αn(2) + βn(3) ) = αT n(2) + βT n(3) = α(λn(2) ) + β(λn(3) ) = λ(αn(2) + βn(3) ).

This means that any linear combinations of the eigenvectors n(2) and n(3) will
be an eigenvector with the eigenvalue λ. Moreover, n(2) and n(3) need not to be
orthonormal to each other. Note that the choice of n(1) is unique and it has to
be perpendicular to the plane containing n(2) and n(3) (Fig. 20). The spectral
decomposition in this case becomes

T = λ1 n(1) ⊗ n(1) + λn(2) ⊗ n(2) + λn(3) ⊗ n(3)

= (λ1 − λ)n(1) ⊗ n(1) + λg,

where the identity tensor is given by

g = n(1) ⊗ n(1) + n(2) ⊗ n(2) + n(3) ⊗ n(3) .

AE 6104 65 Krishnendu Haldar

10.2 Diagonalization of a tensor

n(3)

n(1) 2

1 n(2)
Figure 20: Any vector in the 2 − 3 plane, i.e., linear combination of n(2) and n(3)
is an eigenvector.

10.1.2 Three repeated eigenvalues

In this case λ1 = λ2 = λ3 = λ, means every directions are eigenvectors. If we
select any three orthonormal vectors, then the spectral decomposition becomes
where the identity tensor is given by

T = g,

where the identity tensor is given by

g = n(1) ⊗ n(1) + n(2) ⊗ n(2) + n(3) ⊗ n(3) .

10.2 Diagonalization of a tensor

Considering stacking the three linearly independent right eigenvectors of T ∈ Sym
with a representation T = T·ji g i ⊗ g j as

[P ] = [n(1) , n(2) , n(3) ]

and we write the matrix form as

 1
n(1) n1(2) n1(3)


[P ]i·j = n2(1) n2(2) n2(3) 

n3(1) n3(2) n3(3)
g i ⊗g j

AE 6104 66 Krishnendu Haldar

10.2 Diagonalization of a tensor

The tensor representation of P is then

P = ni·(j) g i ⊗ g j = n(j) ⊗ g j
= n(1) ⊗ g 1 + n(2) ⊗ g 2 + n(3) ⊗ g 3 .

We now introduce a tensor Λ, consisting of its eigenvalues in the following ways

Λ = λ1 g 1 ⊗ g 1 + λ2 g 2 ⊗ g 2 + λ3 g 3 ⊗ g 3

and in the matrix form:

 
λ1 0 0
[Λ]i·j =  0 λ2 0 
0 0 λ3 g ⊗g j
i

We now perform

TP = T (n(1) ⊗ g 1 + n(2) ⊗ g 2 + n(3) ⊗ g 3 )

= T (n(1) ⊗ g 1 ) + T (n(2) ⊗ g 2 ) + T (n(3) ⊗ g 3 )
= T n(1) ⊗ g 1 + T n(2) ⊗ g 2 + T n(3) ⊗ g 3 , (recall T (a ⊗ b) = T a ⊗ b)
= λ1 n(1) ⊗ g 1 + λ2 n(2) ⊗ g 2 + λ3 n(3) ⊗ g 3
= (n(1) ⊗ g 1 + n(2) ⊗ g 2 + n(3) ⊗ g 3 )(λ1 g 1 ⊗ g 1 + λ2 g 2 ⊗ g 2 + λ3 g 3 ⊗ g 3 )
= PΛ

and write the diagonalization of a tensor with respect to the selected basis as

Λ = P −1 T P also T = P ΛP −1 (57)

Note that for a symmetric tensor

[P ]−1 = [n(1) , n(2) , n(3) ]

i.e., considering left eigenvectors or the reciprocal basis of {n(1) , n(2) , n(3) }

10.2.1 Power and polynomials of tensor

Consider,

(P −1 T P )(P −1 T P )...(P −1 T P ) = ΛΛ · · · Λ}
| {z
| {z }
k times k times

P −1 T k P = Λk .

AE 6104 67 Krishnendu Haldar

10.2 Diagonalization of a tensor

So we write

Λk = P −1 T k P also T k = P Λk P −1 . (58)

Then

Λk = λk1 g 1 ⊗ g 1 + λk2 g 2 ⊗ g 2 + λk3 g 3 ⊗ g 3

or
 k 
λ1 0 0
[Λk ]i·j =  0 λk2 0  .
k
0 0 λ3 g ⊗g j
i

This is true for any k (negative, fraction, etc.). For fraction value of k, we will
only consider positive definite tensors, which will be discussed next. A tensor
polynomial f (T ) is defined by
n
X
f (T ) = ak T k
k=i
n
X n
X
−1
= ak P Λ P k
= P (ak Λk )P −1
k=i k=i
Xn
= P( ak Λk )P −1
k=i
= P (f (Λ))P −1

Then
n
X n
X n
X
k
f (Λ ) = (ak λk1 )g 1 ⊗g + 1
(ak λk2 )g 2 ⊗g + 2
(ak λk3 )g 3 ⊗ g 3
k=i k=i k=i

or
Pn 
k=i ak λk1 P 0 0
n
[Λk ]i·j =  0 k=i ak λk2 Pn 0
 .
k
0 0 k=i ak λ3 g i ⊗g j

10.2.2 Series
We can express a series as
∞
X X∞
f (T ) = k
ak T = P ( ak Λk )P −1 = P (f (Λ))P −1
k=i k=i

AE 6104 68 Krishnendu Haldar

10.3 Symmetric positive definite

Then
∞
X ∞
X ∞
X
f (Λ) = (ak λk1 )g 1 ⊗g + 1
(ak λk2 )g 2 ⊗g + 2
(ak λk3 )g 3 ⊗ g 3
k=i k=i k=i

or
P∞ k

k=i a k λ1 P 0 0
∞
[Λ]i·j =  0 k
k=i ak λ2 P 0
 .
∞ k
0 0 k=i ak λ3 g ⊗g j
i

10.2.3 Exponential map

We can express a series as
∞
X Tk
e = T
= P (eΛ )P −1
k=i
k!

Then
∞ ∞ ∞
X λk 1
X λk 2
X λk 3
eΛ = g1 ⊗ g1 + g2 ⊗ g2 + g3 ⊗ g3
k=i
k! k=i
k! k=i
k!
λ1 1 λ2 2 λ3 3
= e g1 ⊗ g + e g2 ⊗ g + e g3 ⊗ g

or
λk1
P∞   λ 
k=i k! 0 0 e 1 0 0
λk2
[Λ]i·j = 
P∞
=  0 e λ2 0  .
 
0 k=i k! 0 
0 0
P∞ k
λ3 0 0 eλ3 g ⊗g j
k=i k! g i ⊗g j i

10.3 Symmetric positive definite

A second order tensor T is positive definite if for all v ∈ V,

v · T v ≥ 0, v · T v = 0 =⇒ v = 0.

If we decompose a tensor to its symmetric and skew-symmetric part, T = T s +T ss ,

then

v · T v = v · (T s + T ss )v = v · T s v + v · T ss v = v · T s v + 0.

So, even for a general tensor, it is sufficient to study the symmetric part of it.
When a symmetric tensor T is positive definite, we say T ∈ Psym .

AE 6104 69 Krishnendu Haldar

10.4 Polar decomposition

Theorem 10.5 Let T ∈ Hom (V, V), and also T ∈ Psym , where dim V = 3.
Then all the eigenvalues of T are strictly positive.
Let T n = λn. Then,
n · Tn
n · T n = λn · n, ⇒ λ = .
n·n
By definition, as T ∈ Psym , n · T n > 0. Also n · n > 0. These two facts prove
that λ > 0.

10.4 Polar decomposition

Theorem 10.6 Let T ∈ Hom (V, V), and invertible. Then T can be factorized in
a unique way

T = RU = V R

where R ∈ Orth and U , V ∈ Psym. Further

p p
t
U = T T, V = TTt

We first consider the factor T = RU and start with the fact that T t T ∈ Sym and
u · T t T u = T u · T u = |T u|2 > 0 for any nonzero vector u. Then t
√ t T T ∈ Psym
and so, all the eigenvalues are positive. We can define U = T T ∈ Psym by
considering the positive square roots of the eigenvalues of T t T . Now, R = T U −1 .
Then,

Rt R = (T U −1 )t T U −1 = U −t T t T U −1
= U −1 (T t T )U −1
= U −1 (U 2 )U −1 = U −1 (U U )U −1 = I

So R ∈ Orth. You can check √ that RRt = I. You can now establish the other
relation by starting with V = T T t .

AE 6104 70 Krishnendu Haldar

11 Fourth order Tensors
see hand notes

AE 6104 71 Krishnendu Haldar

12 Tensor calculus
12.1 Normed space
Definition 12.1 Let V is a vector space together with a function || · || : V → R,
called a norm, satisfying the following four conditions.

1. ||v|| > 0 for all v ∈ V

2. ||v|| = 0 if and only if v = 0

3. ||av|| = |a| · ||v|| for all v ∈ V and a ∈ R

4. ||v + w|| ≤ ||v|| + ||w|| for all v, w ∈ V. (triangle inequality)

Then the vector space is called normed linear space of simply normed space.
Example 1: If V = R, then ||v|| = |v|, v ∈ R.
p
Example 2: If V = En (Euclidean), then ||v|| = v12 + · · · + vn2 .
Example 3: If V = E2 (Euclidean), then we can define many norms. For example,

||v|| = |v1 | + |v2 |.

If V is an inner product space, then we can say that

√
||v|| = v · v,

the length of the vector v. Note that normed space does not always mean an inner
product space, but an inner product space can always have a norm. Norms always
give rise to a metric, which is the distance between two points. Here the points
means the members of the set. The distance function between any two points are
written as

d(v, w) = ||w − v||.

This particular norm, among many other possible norms, known as metric.

12.2 Affine space

see hand notes.

AE 6104 72 Krishnendu Haldar

12.3 Differentiation

12.3 Differentiation
Let V and U are the two inner product spaces with a metric norm. Moreover, D
is an affine space of V. Let v ∈ V and a map f (v) ∈ U. Note that f needs not
to be a linear transformation. It is a mapping with the argument v. f could be a
scalar valued, vector valued or a tensor valued functions. For example,

f (v) = v · v ∈ U = R is a vector-valued scalar mapping/function

f (v) = av ∈ U = V is a vector-valued vector mapping
f (v) = v ⊗ v ∈ U = Hom (V, V) is a vector-valued tensor mapping

and so on. Let f be defined at the neighborhood of 0 ∈ V. If f (v) approaches 0 ∈ U faster than v,
we write
||f (v)||
f (v) = o(v), defined as lim = 0.
v →0 ||v||
For example, consider f (t) = ta for a > 1, then f (t) = o(t). Let D be a subset of

Figure 21: When v approaches 0 ∈ V, F (v) approaches 0 ∈ U faster than v.

V and consider f : D → U (Fig. 21). We say f is differentiable at x ∈ D if the

difference

f (x + v) − f (x)

is equal to a linear transformation of v (say, T v ∈ U) plus a term (i.e. o(v) ∈ U)

that approaches zero faster than v. We write

f (x + v) − f (x) = D x f (x)[v] + o(v) (59)

AE 6104 73 Krishnendu Haldar

12.3 Differentiation

where T is rewritten as D x f (x), known as derivative (or Fréchet derivative) of f

with respect to x and it belongs to Hom (V, U). Alternatively we can write
||f (x + v) − f (x) − D x f (x)[v]||
lim = 0. (60)
v →0 ||v||
Note that the vector v could be any vector in V. Now, we freeze the direction of
v, and shrinks or expand v by considering the vector αv, where α ∈ R. From (59)
we can write

f (x + αv) − f (x) = D x f (x)[αv] + o(αv). (61)

Note that v is fixed and α is a variable. This means o(αv) approaches zero faster
than α, i.e., o(αv) = o(α) and we write
||f (x + αv) − f (x) − D x f (x)[αv]||
lim = 0
α→0 ||α||
f (x+αv )−f (x)
|| α
− D x f (x)[v]||
lim |α| = 0
α→0 |α|
f (x + αv) − f (x)
|| lim − D x f (x)[v]|| = 0.
α→0 α
The final form becomes
f (x + αv) − f (x)
D x f (x)[v] = lim , (62)
α→0 α
and we call D x f (x) directional derivative or Gateaux derivative. If we further
expand f (x + αv) in a Taylor series, we get
∂f (x) 1 ∂f (x) 2
f (x + αv) = f (x) + (αvi ) + (α vi vj ) . . . .
∂xi 2 ∂xi ∂xj
Now,
d ∂f (x)
f (x + αv) = (vi )
dα α=0 ∂xi
d2 ∂f (x)
f (x + αv) = (vi vj )
dα2 α=0 ∂xi ∂xj
..
. (63)

Then one can write

d 1 d2
f (x + αv) = f (x) + f (x + αv) α+ f (x + αv) α2 . . . .
dα α=0 2 dα2 α=0

AE 6104 74 Krishnendu Haldar

12.3 Differentiation

So, (62) reads

d
D x f (x)[v] = f (x + αv) (64)
dα α=0

Fréchet derivative pertains if Gateaux derivative exists at each direction. So it

may happen that Gateaux derivative exists at a particular direction but not at all
direction. In such a case Fréchet derivative does not exist.

Example 31 Find the derivative and directional derivative (along v ∈ V) of

ϕ(u) = u · u, where u ∈ D ⊂ V and ϕ : V → R (vector-valued scalar function)
First calculate

ϕ(u + v) = (u + v) · (u + v) = u · u + u · v + v · u + v · v
= ϕ(u) + 2u · v + o(v).

Then,

ϕ(u + v) − ϕ(u) = 2u · v + o(v)

and comparing it with (59) we get D u ϕ(u)[v] = 2u · v.

In this case, the derivative D u ϕ(u) is a linear functional in this case as it maps
from V to R, i.e., D u ϕ(u) ∈ V. So, D u ϕ(u) is a vector and it operates on a vector
v to produce a scalar. So we identify D u ϕ(u)[v] = D u ϕ(u) · v and the derivative
D u ϕ(u) = 2u. Note that the operation on [v] is a general expression and for this
case it becomes a dot product.
For the direction derivative we write,
d
D u ϕ(u)[v] = ϕ(u + αv)
dα α=0
d
= [(u + αv) · (u + αv)]
dα α=0
d
= (u · u + αu · v + αv · u + α2 v · v)
dα α=0

= (u · v + 2v · u + 2αv · v)
α=0
= 2u · v

So the directional derivative D u ϕ(u) = 2u.

Example 32 Find the derivative of ϕ(A) = det A, with respect to A, where

A ∈ Hom (V, V) (tensor-valued scalar function) and dim V = 3

AE 6104 75 Krishnendu Haldar

12.4 Partial derivative

Recall that
det(A − λI) = −λ3 + I1 λ2 − I2 λ + I3
For λ = −1, we can write
det(A + I) = 1 + I1 + I2 + I3 .
Now selecting V ∈ Hom (V, V)
ϕ(A + V ) = det(A + V ) = det A(I + A−1 V ) = det A det(I + A−1 V )
= det A(1 + tr (A−1 V ) + o(V ))
= det A + det A tr (A−1 V ) + o(V ).
= det A + det A tr (A−1 V ) + o(V ).
So,
D A ϕ(A)[V ] = det A tr (A−1 V )
= (det A)A−t : V
So,

D A ϕ(A) = (det A)A−t (65)

In this case, the derivative D A ϕ(A) : Hom (V, V) → R and so D A ϕ(A) ∈ V ⊗ V.

So it is a second order tensor. We identify D A ϕ(A)[V ] = D A ϕ(V ) : V , as two
second order tensors produce a scalar only by a double contraction.
Then
D A ϕ(A)[V ] = D A ϕ(A) : V = (det A)A−t : V (66)
So,

D A ϕ(A) = (det A)A−t (67)

12.4 Partial derivative

Consider f (A, u) which is a vector function with a vector (u) and a tensor (A)
as its arguments. Then, the partial derivative of f with respect to u is defined as
d
∂u f (A, u) = D u f (A, u) = f (A, u + αv) (68)
dα α=0

Similarly, the partial derivative of f with respect to A is defined as

d
∂A f (A, u) = D A f (A, u) = f (A + αV , u) (69)
dα α=0

AE 6104 76 Krishnendu Haldar

12.5 Product rule for bilinear mapping

Example 33 Find the derivative of ϕ(A, u) = u · Au, where A ∈ V ⊗ V, u ∈

D ⊂ V and ϕ : V ⊗ V × V → R (tensor- and vector valued scalar function)
We denote Dϕ(A, u)[v] = ∂u ϕ(A, u)[v] and Dϕ(A, u)[V ] = ∂A ϕ(A, u)[V ], where
v ∈ V and V ∈ V ⊗ V.
ϕ(A, u + v) = (u + v) · A(u + v) = u · Au + u · Av + v · Au + v · Av
= ϕ(A, u) + At u · v + Au · v + o(v).
So,
ϕ(A, u + v) − ϕ(A, u) = At u · v + Au · v + o(v)
= (At u + Au) · v + o(v).
=⇒ ∂u ϕ(A, u) = At u + Au
Now following the definition (68),
d
∂u ϕ(A, u)[v] = ϕ(A, u + αv)
dα α=0
d  
= (u + αv) · (A(u + αv))
dα α=0
d 
u · Au + αv · Au + αu · Av + α2 v · Av

=
dα α=0
 
= v · Au + u · Av + 2αv · Av
α=0
= v · Au + u · Av
∂u ϕ(A, u) · v = (Au + At u) · v =⇒ ∂u ϕ(A, u) = Au + At u
Similarly ∂A ϕ(A, u) = u ⊗ u. (work this out!).

12.5 Product rule for bilinear mapping

It will be frequently necessary to compute the derivative of a product π(f (x), g(x))
of two mappings f (x) and g(x), where x ∈ D ⊂ V. Moreover, π is a general bi-
linear mapping (Fig. 23)
π :F ×G →H
Depending on the vector spaces, the structure of the mapping π is determined.
For example, consider the following bilinear mappings.
π(ϕ, w) = ϕw ϕ ∈ F = R, w ∈ G, and π ∈ H = G
π(u, w) = u·w u ∈ F, w ∈ G, and π ∈ H = R
π(u, w) = u⊗w u ∈ F, w ∈ G, and π ∈ H = F ⊗ G
π(S, w) = Sw S ∈ F = G ⊗ G, say, w ∈ G, and π ∈ H = G

AE 6104 77 Krishnendu Haldar

12.5 Product rule for bilinear mapping

Figure 22: x ∈ D ⊂ V, v ∈ V and f : D → ·F, and g : D → ·G. π : ·F × ·G → ·H

is a general bilinear mapping, where h(x) ∈ ·H.

and so on. Consider h(x) = π(f (x), g(x)) and, if v ∈ V, then

h(x + v) = π(f (x + v), g(x + v))

= π(f (x) + D x f (x)[v] + o(v), g(x) + D x g(x)[v] + o(v))
≈ π(f (x) + D x f (x)[v], g(x) + D x g(x)[v])
= π(f (x), g(x) + D x g(x)[v]) + π(D x f (x)[v], g(x) + D x g(x)[v])
= π(f (x), g(x)) + π(f (x), D x g(x)[v])
+π(D x f (x)[v], g(x)) + π(D x f (x)[v], Dg(x)[v])
= h(x) + π(f (x), D x g(x)[v]) + π(D x f (x)[v], g(x)) + o(v)
h(x + v) − h(x) = π(f (x), D x g(x)[v]) + π(D x f (x)[v], g(x)) + o(v)

So,

D x h(x)[v] = π(D x f (x)[v], g(x)) + π(f (x), D x g(x)[v]), (70)

AE 6104 78 Krishnendu Haldar

12.6 Product rule for any mapping

If V = R, and t, α ∈ R then
D t h(t)[α] = π(D t f (t)[α], g(t)) + π(f (t), D t g(t)[α])
D t h(t)α = π(D t f (t)α, g(t)) + π(f (t), D t g(t)α)
= απ(D t f (t), g(t)) + απ(f (t), D t g(t)) as π is bilinear
and
d
D t h(t) = h(t) = ḣ(t) = π(ḟ (t), g(t)) + π(f (t), ġ(t)). (71)
dt
We have the following product rules for different π:
d
(ϕv) = ϕ̇v + ϕv̇
dt
d
(u · v) = u̇ · v + u · v̇
dt
d
(u ⊗ v) = u̇ ⊗ v + u ⊗ v̇
dt
d
(T S) = Ṫ S + T Ṡ
dt
d
(T : S) = Ṫ : S + T : Ṡ
dt
d
(T v) = Ṫ v + T v̇
dt

12.6 Product rule for any mapping

We now remove the restriction that h(x) = π(f (x), g(x)) is bilinear. If v ∈ V,
then
h(x + v) =
π(f (x + v), g(x + v))
=
π(f (x) + D x f (x)[v] + o(v), g(x) + D x g(x)[v] + o(v))
≈
π(f (x) + D x f (x)[v], g(x) + D x g(x)[v])
=
π(f (x), g(x)) + D f (x) π(f (x), g(x))[D x f (x)[v]]
+D g(x) π(f (x), g(x))[D x g(x)[v]] + o(v)
= h(x) + ∂f (x) π(f (x), g(x))[D x f (x)[v]]
+∂g(x) π(f (x), g(x))[D x g(x)[v]] + o(v)
h(x + v) − h(x) = ∂f (x) π(f (x), g(x))[D x f (x)[v]]
+∂g(x) π(f (x), g(x))[D x g(x)[v]] + o(v)
So,
D x h(x)[v] = ∂f (x) π(f (x), g(x))[D x f (x)[v]] + ∂g(x) π(f (x), g(x))[D x g(x)[v]], (72)

AE 6104 79 Krishnendu Haldar

12.7 Chain rule

12.7 Chain rule

Figure 23: x ∈ D ⊂ V, v ∈ V and f : D → ·F, and g : D → ·G.

Consider the composition function h = g ◦ f . Then h(x) = g(f (x)) as shown

in Fig. 23.
h(x + v) = g(f (x + v)) = g(f (x) + D x f (x)[v] + o(v)))
≈ g(f (x) + D x f (x)[v])) = g(f (x)) + D f (x) g(f (x))[Df (x)[v]]
= h(x) + D f (x) g(f (x))[D x f (x)[v]] + o(Df (x)[v]).
So we write the chain rule as
D x h(x)[v] = D f (x) g(f (x))[D x f (x)[v]]. (73)
If D ⊂ V = R, then
D t h(t)[α] = D f (t) g(f (t))[D t f (t)[α]]
D t h(t)α = D f (t) g(f (t))[D t f (t)α]
D t h(t)α = D f (t) g(f (t))[D t f (t)]α
D t h(t) = D f (t) g(f (t))[D t f (t)]
d ˙
h(t) = D f (t) g(f (t))[f (t)]. (74)
dt
Example 34 Find dtd (det A(t))

Note that V = R. Let f (t) = A(t), and g = det() such that g(f (t)) = det A(t).
So, following equation (66) we write
d ˙ = (det A)A−t : Ȧ
(det A(t)) = D A(t) (det A(t))[A(t)] (75)
dt
AE 6104 80 Krishnendu Haldar
12.8 Affine space

12.8 Affine space

Discussed in the class.

12.9 Gradient and divergence

Consider x ∈ D ⊂ V, v ∈ V and ·F and ·G are the two inner product spaces with
a metric norm.

12.9.1 Gradient
1. Consider a vector valued scalar function

h(x) : D → R.

then we write

D x h(x)[v] = grad ϕ(x) · v

Note that grad ϕ ∈ V is a vector.

2. Similarly consider a vector values vector function

h(x) : D → V

then

D x h(x)[v] = [grad h(x)]v

and here grad h(x) ∈ V ⊗ V is a second order tensor. We can keep on

continue in this notion for the higher order tensors.

3. In general, if

h(x) : D → ⊗n−1 V, i.e, h(x) ∈ ⊗n V

then

D x h(x)[v] = [grad h(x)]v

and here grad h(x) ∈ ⊗n+1 V is a (n + 1)th order tensor.7

Example 35 Show that grad (ϕw) = w⊗grad ϕ+ϕ grad w, where ϕ(x) ∈ ·F =
R, and w(x) ∈ ·G.
7
The notation ⊗n V = V ⊗ V ⊗ · · · ⊗ V means n copies of V.

AE 6104 81 Krishnendu Haldar

12.9 Gradient and divergence

Consider h = π(ϕ, w) = ϕw. From product rule

D x h(x)[v] = π(D x f (x)[v], g(x)) + π(f (x), D x g(x)[v])

we write

D x h(x)[v] = π(D x ϕ(x)[v], w(x)) + π(ϕ(x), D x w(x)[v])

= π(grad ϕ(x) · v, w(x)) + π(ϕ(x), [grad w(x)]v)
= (grad ϕ · v)w + ϕ([grad w]v)
= (w ⊗ grad ϕ)v + (ϕ grad w)v
  h i
grad h v = w ⊗ grad ϕ + ϕ grad w v

So,

grad (ϕw) = w ⊗ grad ϕ + ϕ grad w (76)

12.9.2 Divergence
Divergence will make sense when T (x) is a vector or higher order tensor. Consider
T (x) ∈ ⊗n V Then

div T (x) = grad T (x) : g

where the identity g ∈ V ⊗ V as long as the argument variable x ∈ V. Note

that grad T (x) ∈ ⊗n+1 V and so div T (x) ∈ ⊗n+1−2 V = ⊗n−1 V. If n = 1, then
without any ambiguity we can write

div T (x) = grad T (x) : g = tr (grad T (x))

Example 36 Show that div (ϕw) = w · grad ϕ + ϕ div w, where ϕ(x) ∈ ·F = R,

and w(x) ∈ ·G.
We know grad (ϕw) from (76) . Taking trace for both sides

tr (grad (ϕw)) = tr (w ⊗ grad ϕ) + tr (ϕ grad w)

=⇒ div (ϕw) = w · grad ϕ + ϕ div w (77)

Sometimes the following relation is used as the definition of the divergence of a

second order tensor.

a · div T = div (T t a) (78)

where a is a constant vector.

AE 6104 82 Krishnendu Haldar

12.10 Divergence theorem

12.9.3 Curl
Consider the map

ϕ : Rn → R3 .

Then curl is defined as

(grad ϕ(x) − (grad ϕ(x))t )v = curl ϕ(x) × v

Generalizing curl in any dimension and any order requires exterior calculus, which
is out of scope in this course.

12.10 Divergence theorem

Consider a bounded region D ⊂ V with a boundary ∂D. x is a generic point on D.
Moreover, we have an outward unit normal n at the boundary points. We denote
a differential area as dA ∈ ∂D and a differential volume element dV ∈ D. If

ϕ:D→R u : D → ·F S : D → Hom (·F, ·F)

then the divergence theorem gives us

Z Z
ϕn dA = grad ϕ dV (79)
Z ∂D ZD
u · n dA = div u dV (80)
∂D
Z ZD

T n dA = div T dV (81)
∂D D

You are already familiar with (79) and (80) from your elementary vector calculus
course. So, we will prove only (81).
Let a be a constant vector. Then
Z Z Z
a· T n dA = a · T n dA = T t a · n dA
∂D Z∂D ∂D
Z
t
= div (T a) dV = a · div (T ) dV
DZ D

= a· div (T ) dV
Z Z D
=⇒ T n dA = div T dV
∂D D
R R
Example 37 Show that ∂D
(u ⊗ n) dA = D
grad u dV

AE 6104 83 Krishnendu Haldar

Let a be a constant vector. Then
Z Z Z
a· (u ⊗ n) dA = a · (u ⊗ n) dA = (a · u)n dA
∂D Z∂D Z∂D
= grad (a · u) dV = (grad u)t a dV
DZ D

= a· grad u dV
Z Z D

=⇒ (u ⊗ n) dA = grad u dV
∂D D

13 Covariant derivative in Riemannian space

We have all the tools to consider derivative of any tensor mappings which are true
for any vector spaces and free from any basis. We will now consider a Riemannian
space V = R3 . Let x ∈ D ⊂ V, where D is a region of B. Let u(x) be a local
coordinates and {g 1 (u), g 2 (u), g 3 (u)} and {g 1 (x), g 2 (x), g 3 (x)}8 are the covariant
and contravariant basis, respectively. Then we write the metric as
ds2 = gij (u)dui duj
where gij is the metric tensor and positive definite when the space is Riemannian.
Denoting the local coordinate system at a point x + dx by u + du, the change
in covariant basis becomes {g i + dg i }. Now our task is to find dg i .

13.0.1 Christoffel symbols of 2nd kind

We can write,
∂g i j
dg i = du (82)
∂uj
∂gi
Now ∂uj
can be spanned by the basis {g k (u)}
∂g i
j
= Γkij g k
∂u
and
∂g i k
· g = Γkij (u). (83)
∂uj
We call Γkij as Christoffel symbols of 2nd kind.
8
You can always substitute back the coordinate relation between x and u to express
{g (x), g 2 (x), g 3 (x)} as {g 1 (u), g 2 (u), g 3 (u)}
1

AE 6104 84 Krishnendu Haldar

13.0.2 Christoffel symbols of 1st kind
∂g
Now ∂uji can be spanned by the basis {g k (u)} too. Then

∂g i
= Γijk g k
∂uj
and
∂g i
· g = Γijk (u). (84)
∂uj k
We call Γijk as Christoffel symbols of 1nd kind. Note that
∂g i ∂g i m ∂g i m
Γijk = j
· g k = j
· gmk g = g mk j
· g = gmk Γm
ij . (85)
∂u ∂u ∂u
Also note that

g im Γijk = g im gih Γhjk = δhm Γhjk = Γm

jk (86)

13.0.3 Symmetry properties of Christoffel symbols

∂g i k ∂ ∂x k ∂ ∂x ∂g j k
Γkij = · g = · g = · g k
= · g = Γkji
∂uj ∂uj ∂ui ∂ui ∂uj ∂ui
and
∂g i ∂ ∂x ∂ ∂x ∂g j
Γijk = j
· gk = j i
· gk = i j
· gk = · g k = Γjik
∂u ∂u ∂u ∂u ∂u ∂ui
∂gi
Example 38 Show that ∂uj
= −Γimj g m
We take advantage the relation g i · g m = δm
i
by taking derivative with respect to
j
u we write
∂g i ∂g
j
· g m + g i · mj = 0
∂u ∂u
∂g i
·g = −Γimj
∂uj m
∂g i
j
= −Γimj g m (87)
∂u

AE 6104 85 Krishnendu Haldar

13.1 Ricci identities

13.1 Ricci identities

We now discuss the relation between the Christoffel symbols with the metric gij .
We denote We start from gij = g i · g j and write

∂
∂k gij = (∂k g i ) · g j + g i · (∂k g j ) (where ∂k ≡ )
∂uk
= Γikj + Γjki . (88)

Now cyclically changing the position of i, j and k we write

∂i gjk = Γjik + Γkij (89)

∂j gki = Γkji + Γijk . (90)

Now (90)+(88)-(89) gives

1 
Γijk = − ∂k gij + ∂i gjk + ∂j gki (91)
2

Similarly,

1 mi  
Γm
jk
mi
= g Γijk = g − ∂k gij + ∂i gjk + ∂j gki (92)
2

This means that knowing the metric, we will be able to compute Christoffel sym-
bols.

13.2 Gradient and divergence in Riemannian space

This is continuation of Section 12.9. We have the space V = Rn with x ∈ D ⊂ Rn .
We consider this Rn space as Euclidean (E n ) or flat and a curvilinear coordinate
system u such that x = ψ(u). The local coordinate system has {g i } as local
(covariant) basis , where {ek } is a basis of E n . So we have a Riemannian space
Rn , with a metric tensor gij , embedded in E n .

13.2.1 Gradient of vector-valued scalar function

Consider a scalar function ϕ(u), such that

ϕ = D → R.

and we have

D u ϕ(u)[v] = grad ϕ(u) · v.

AE 6104 86 Krishnendu Haldar

13.2 Gradient and divergence in Riemannian space

for a scalar function ϕ(u). The directional derivative along g i could be written as
D u ϕ(u)[g i ] = grad ϕ(u) · g i . (93)
From (64) we further write
d
D u ϕ(u)[g i ] = ϕ(u + αg i )
dα α=0
d
= ϕ(u1 , u2 , . . . , ui + α, . . . , un )
dα α=0
∂ϕ
=
∂ui
where u = uk g k . So we write from (93)
∂ϕ
grad ϕ(u) · g i =
∂ui
which implies

∂ϕ i
grad ϕ(u) = g. (94)
∂ui
Note that grad ϕ(u) is a contravariant vector.

13.2.2 Gradient of vector-valued vector mapping

Consider a scalar function ϕ(u), such that
ϕ = D → Rn .
and we have
D u ϕ(u)[v] = [grad ϕ(u)]v
for a vector mapping ϕ(u). Now directional derivative along g i could be written
as
D u ϕ(u)[g i ] = [grad ϕ(u)]g i . (95)
From (64) we further write
d
D u ϕ(x)[g i ] = ϕ(u + αg i )
dα α=0
d
= ϕ(u1 , u2 , . . . , ui + α, . . . , un )
dα α=0
∂ϕ
=
∂ui
AE 6104 87 Krishnendu Haldar
13.2 Gradient and divergence in Riemannian space

We write from (95)

∂ϕ
[grad ϕ(u)]g i =
∂ui
which implies

∂ϕ
grad ϕ(u) = ⊗ gi. (96)
∂ui

If ϕ = ϕi g i , then

∂ϕ ∂ k ∂ϕk k ∂g k
= (ϕ g k ) = g k + ϕ
∂ui ∂ui ∂ui ∂ui
k
∂ϕ
= g + ϕk Γm ki g m
∂ui k
So, we write

∂ϕ  ∂ϕm 
i
grad ϕ(u) = i
⊗g = i
+ ϕ Γki g m ⊗ g i = ϕm g m ⊗ g i .
k m
∂u ∂u i

It is a standard practice in mechanics to denote the covariant derivative of a

contravariant component as

∂ϕm
ϕm = + ϕk Γm
ki
i ∂ui

If ϕ = ϕi g i , in a similar way we get

∂ϕ ∂ k ∂ϕk k ∂g k
= (ϕk g ) = g + ϕk
∂ui ∂ui ∂ui ∂ui
∂ϕk k
= g − ϕk Γkmi g m , from equation (87)
∂ui
So, we write

∂ϕ i
 ∂ϕ
m k

m i m i
grad ϕ(u) = i
⊗ g = i
− ϕ k mi g ⊗ g = ϕm g ⊗ g .
Γ
∂u ∂u i

It is a standard practice to denote the covariant derivative of a covariant compo-

nent as
∂ϕm
ϕm = − ϕk Γkmi
i ∂ui

AE 6104 88 Krishnendu Haldar

13.2 Gradient and divergence in Riemannian space

13.2.3 Gradient of vector-valued tensor mapping

Consider a tensor function T (u), such that T ∈ Hom (Rn , Rn )
T = D → Hom (Rn , Rn ).
We then write, in a similar procedure as explained in the previous subsections,
∂T
grad T (u) = ⊗ gk . (97)
∂uk
If T = T ij g i ⊗ g j , we write

∂T  ∂T ij 
k tj i it j k ij
grad T (u) = ⊗ g = + T Γkt + T Γkt g i ⊗ g j ⊗ g = T gi ⊗ gj ⊗ gk .
∂uk ∂uk k

where
∂T ij
T ij = k
+ T tj Γikt + T it Γjkt
k ∂u
If T = Tij g i ⊗ g j , we write

∂T k
 ∂T
ij t t

i j k ij
grad T (u) = k
⊗ g = k
− T Γ
it jk − Tjt ik g ⊗ g ⊗ g = T
Γ gi ⊗ gj ⊗ gk .
∂u ∂u k

where
∂Tij
Tij = − Tit Γtjk − Ttj Γtik (98)
k ∂uk

Similarly we have T·ij . (work this out!)

k

Example 39 Show that the covariant derivative of the metric coefficient gij is
always zero.
Replacing gij in place of Tij in equation (98)
∂gij
gij = − git Γtjk − gjt Γtik
k ∂uk
∂gij
= − Γjki − Γikj
∂uk
∂gij
= − (Γjki + Γikj )
∂uk
∂gij ∂gij
= − from equation (88)
∂uk ∂uk
= 0 (99)

AE 6104 89 Krishnendu Haldar

13.2 Gradient and divergence in Riemannian space

∂g
Example 40 Show that Γiik = 12 g ij ∂uijk .

As gij = 0, the first line of the previous example reads

k

∂gij
= git Γtjk + gjt Γtik
∂uk
∂gij
g ij k = g ij git Γtjk + g ij gjt Γtik
∂u
= δtj Γtjk + δti Γtik
1 ∂gij
= Γjjk + Γiik = 2Γiik =⇒ Γiik = g ij k (100)
2 ∂u
√
∂ g
Example 41 Show that Γiik = √1 .
g ∂uk

Recall from (75)

d
(det A(t)) = (det A)A−t : Ȧ
dt
Now replace A = g = gij g i ⊗ g j and dtd with ∂u∂ k and write
∂g ∂g
k
= gg −t : , where g = det g
∂u ∂uk
∂gij
= gg ij k
∂u
Now from (100), we write
√
1 ij ∂gij 1 ∂g 1 ∂ g
Γiik = g = =√ (101)
2 ∂uk 2g ∂uk g ∂uk

13.2.4 Divergence of vector-valued vector mapping

We have derived that
grad ϕ(u) = ϕi g i ⊗ g j .
j

Then
div ϕ(u) = grad ϕ(u) : g = tr (grad ϕ(u)) = ϕi
i
i
∂ϕ
= + ϕk Γiki
∂ui √
∂ϕi k 1 ∂ g
= +ϕ √ from (101)
∂ui g ∂uk
√
1 ∂ gϕk
= √ (102)
g ∂uk

AE 6104 90 Krishnendu Haldar

13.2 Gradient and divergence in Riemannian space

13.2.5 Divergence of vector-valued tensor mapping

We have derived for T = T ij g i ⊗ g j

grad T (u) = T ij g i ⊗ g j ⊗ g k .
k

Then

div T (u) = grad T (u) : g = T ij g i ⊗ g j ⊗ g k : g nm g n ⊗ g m

k
ij
= T g i g nm (gnj )(δm
k
)
k

= T ij g i (δjm )(δm
k
)
k
ij
= T g i δjk
k

= T ik g i
k

You can consider any form among the four possible options. You will get the same
final results. However, intermediate steps will be different.

AE 6104 91 Krishnendu Haldar

 AE 6104: Home work-2 [200 points]

Consider V = R3 where {g 1 , g 2 , g 3 }, and {g 1 , g 2 , g 3 } are the covariant, and con-

travariant basis, respectively. g = det[gij ]. Moreover, {e1 , e2 , e3 } is the natural
basis. You may use any software, e.g, MATLAB/Maple/Python/Wolfram Math-
ematica, etc. for computation and/or cross-check your hand calculated results.
Please provide the code+output if you use so.

1. Show that {g 1 , g 2 , g 3 } is a basis.

2. Show that {g 1 , g 2 , g 3 } is a basis.

3. Show that
√
g 1 · (g 2 × g 3 ) = g

4. Show that
g2 × g3 g3 × g1 g1 × g2
g1 = √ , g2 = √ , g3 = √
g g g

5. Show that

g i × g j = eijk g k
√
where eijk = gεijk , and εijk is the permutation with respect to the standard
basis.

6. Show that
√ √ √ 1
g1 = g(g 2 × g 3 ), g2 = g(g 3 × g 1 ), g3 = g(g × g 2 )

7. Show that

g i × g j = eijk g k
√
where eijk = εijk / g, and εijk is the permutation with respect to the stan-
dard basis.

1
 8. Show that the differential volume element dV of an infinitesimal paral-
lelepiped with sides g 1 du1 , g 2 du2 and g 3 du3 , where (u1 , u2 , u3 ) are the local
(curvilinear) coordinates is given by
√
dV = gdu1 du2 du3

9. Given the basis B = {g 1 , g 2 , g 3 }, where

     
1 2 3
g 1 = 1 , g 2 = 3 , g 3 = 3
     (1)
1 2 4
(a) find its reciprocal basis.
(b) If v = 3g 1 + g 2 + 2g 3 , find the covariant components of v.
(c) Evaluate gij and g ij .
10. Find {g 1 , g 2 , g 2 }, {g 1 , g 2 , g 2 }, gij , g ij , ds2 , and g for the following local
(curvilinear) systems

(a) Cylindrical (Fig. 1): (r, θ, x3 ) implies (radial, polar, axial)

Figure 1: Cylindrical coordinate system. Read (x, y, z) as (x1 , x2 , x3 ).

x1 = r cos θ,
x2 = r sin θ,
x3 = x 3 .

2
(b) Spherical (Fig. 2): (r, θ, ϕ) implies (radial, polar, azimuthal)

Figure 2: Spherical coordinate system. Read (x, y, z) as (x1 , x2 , x3 ).

x1 = r cos θ sin ϕ,
x2 = r sin θ sin ϕ,
x3 = r cos ϕ.

(c) Toroidal (Fig. 3): (r, θ, ϕ) implies (radial, polar, azimuthal)

x1 = (R − r cos θ) cos ϕ,
x2 = (R − r cos θ) sin ϕ,
x3 = r sin ϕ.

where R > 0 is the fixed distance radius OA (see Fig. 3).

3
Figure 3: Toroidal coordinate system. Read (X, Y, Z) as (x1 , x2 , x3 ).