Vector e Tensor Analysis With Linear Algebra (Slides)

PART 1: VECTOR & TENSOR ANALYSIS with LINEAR ALGEBRA
Objectives
Introduce the concepts, theories, and operational
implementation of tensors (of which vectors are a subset) in
advanced engineering analysis. The emphasis is on geometric
and physical interpretations for engineering applications.
Study some of the fundamental rules of linear algebra and show
analogies with tensor analysis. We will study elementary topics
of linear algebra: Matrices, determinants, systems of linear
equations, and eigenvalues and eigenvectors.
Vectors: Definitions
Description of Physical Quantities
Scalar: A quantity described only by magnitude; described a single number, e.g.,
temperature, pressure,
Vector: A quantity described by both magnitude and direction, e.g., velocity,
displacement,
Tensor: A higher-order vector, gives information in addition to magnitude and
direction, e.g., the state of stress and strain in a continuous medium are secondorder tensors.
Vectors: Definitions
A free vector can be displaced
parallel to itself and act at any point;
requires three numbers to specify a
free vector, e.g., velocity.
A sliding vector can only be
displaced along a line through a fixed
point containing the vector; requires
five numbers to specify a sliding
vector, intersection of a line and a
coordinate plane (2) and the vector
(3), e.g., force.
A bound vector requires six numbers
(coordinates points x1, x2), e.g.,
displacement.
x2
x1
Geometric Definitions
Vectors have magnitude and direction
and satisfy the parallelogram law of
addition.
E
D
D+ E
Example: A finite rotation has

magnitude and direction, but is it a
vector . . .
. . . but the line segment (arc) that
connects D and E. Therefore, a finite
rotation is not a vector since it does
not satisfy the geometric definition.
(What about a differential rotation?)
D+ E
D
E
Invariance
Vectors are invariant under a
coordinate transformation.
Example: The position vectors U1
and U2 indicate the position of the
fixed point as the coordinate system
translates. After the coordinate
translation, U1 U2 therefore, a
position vector is really not a
vector since it is not invariant under
a coordinate translation! (What
about a pure coordinate rotation?)
U1
U2
Vector Algebra
E
Elementary Operations
D+E
Addition
D+ E E+ D commutativity
D+ E+ F (D + E) + F = D + (E + F) associativity
D + = D additive identity
Subtraction
D E D + (E)
E+F
D+ E + F
E
E
D+E
D
DE
6
Vector Algebra
Scalar Multiplication
mD= Dm
|mD| = m|D|, m > 0
|mD| = m|D|, m < 0
0D =
Division is not a defined vector operation
Unit Vector
H D =
(1/2)D
D
(1)D
2D
D D
= H D = 1
D a
A vector can always be written in terms

of pure magnitude and direction using a
unit vector
D = a H D
direction
magnitude
7
Vector Algebra
Linear Dependence
Given vectors {D1,D2,,Dn} and scalars {1,2,,n}, not all zero. If one can write,
1D1 + 2D 2 + " + nD n =
(1)
then the vectors are linearly dependent, i.e., one is a linear combination of the
others.
Example:
n = 2, D 2 =
n = 3, D3 =
1
D colinear
2 1
1
( 1D1 + 2D 2 ) coplanar
If (1) cannot be satisfied, the vectors are linearly independent.
Vector Algebra
nE + pF
Expansion of a Vector with Respect to Other Vectors
Given D and E, linearly independent
mD
(non-colinear) then, vectors F and G can
always be constructed:
2D: F = mD + nE
D
3D: G = mD + nE + pF
nE
Scalar (Dot, Inner) Product
pF
D E = D E cos(D, E ) = ab cos , 0
Rules
1. D E = E D
2. D (E + F) = (D E) + (D F)
3. If D E (orthogonal) D E = ab cos( / 2) = 0
4. If D E = 0 D E or D = or E=
Vector Algebra
D

5.
D D = D2 = aa = a2
D H = a cos = aH projection of D in direction of H
H
aH
Example: Vector representation of work:
projection of force
magnitude of
work = in direction

of displacement displacement
dW = ( f cos ) ( ds ) = I GV
Vector (Cross, Skew, Outer) Product
D E = F = ab sin H DE
The vector product obeys the right-hand rule:
Bringing D into E advances H DE in the direction
of a right-handed screw.
F
H DE
10
Vector Algebra
Other rules:
1. D E = E D anticommutativity
2. If D || E = 0 or = sin = 0 D E = .
If D E = 0 then either D || E or E = D D = .
3. (D + E) F = (D F) + (E F) distributive but order must be preserved.
Example: Moment (torque) about some point
O from a force acting at a point P.
position of point P with respect to O
U
I
force
P moment (torque)
l
perpendicular distance from O to line
through I
l
P
H P
P = U I = rf sin H P = flH P
P
11
Vector Algebra
V = SQ
This definition of plane area can be generalized

to describe a general plane area as a vector
quantity. By convention, the area is enclosed
on the left side when traversing the closed
contour in a counterclockwise direction.
C
S
Example: Determine the projected area from the

oblique cut through a circular cylinder
S = magnitude of slant area
S = magnitude of projected area
Q = unit normal to area S
V = S Q , V = SQ
S is the projection of V in direction of Q
S = V Q = SQ Q
S
Q
S
12
Vector Algebra
Rigid-Body Rotation
Determine the velocity at any point in an
arbitrarily shaped, 3-D body rotating about
some arbitrary axis.
U = position vector
Y = linear velocity
= angular velocity at point P:
Y = aH
from geometry:
a = r sin
Y = r sin H = U
a
U
H U
Y = U
13
Vector Algebra
Multiple Products
scalar triple product: D (E F)
1. D (E F) = D E F [DEF]
2. D (E F) = F D E = E F D (cyclic permutation)
3. D (E F) = D F E = F E D = E D F (noncyclic permutation)
4. If three vectors D, E, and F are coplanar,
then [DEF] = 0 (a necessary and sufficient
condition.
5. [DEF] represents the volume of a
parallelapiped
EF
D
volume = D E F
F
E
14
Vector Algebra
Multiple Products
vector triple product: D (E F)
1.
Parentheses preserve the order of the
operation and must be retained, i.e., D (E F) (D E) F
2. D (E F) is in the plane of E and F.
3. D (E F) E(D F) F(D E)
3
a H3
Vector Components and Basis

A basis in n-space contains n linearly
independent basis vectors.
{H1, H2, H3} represents a basis.
scalar component a1H
1
1
2
3
D = a H1 + a H 2 + a H 3
H3
H1
H2
a2H2
vector component
15

'XDO5HFLSURFDO%DVLV
We can construct another basis {H1, H2, H3} from {H1, H2, H3}
that enables us to obtain the scalar component of a vector.
Since H1 H2 is perpendicular to both H1 and H2, a3H3 is the only
nonzero component from the dot product, i.e.,
D (H1 H 2 ) = a1H1 (H1 H 2 ) + a 2H 2 (H1 H 2 ) + a 3H3 (H1 H2 )
or
a3 =
= a 3H3 (H1 H 2 )
D (H1 H 2 )
= D H3 ,
H3 (H1 H 2 )
where,
H H
H H
H H
H3 = 1 2 , and similarly, H1 = 2 3 , and H 2 = 3 1 .
[H1H 2H3 ]
[H1H 2H3 ]
[H1H 2H3 ]
1

Now we say {H1, H2, H3} is the dual or reciprocal basis of
{H1, H2, H3} (and vice versa) since,
H1 H1 = H 2 H 2 = H3 H3 = 1.
6XPPDWLRQ&RQYHQWLRQ(LQVWHLQRU,QGH[1RWDWLRQ
n
D = a i Hi a i Hi sum over a repeated (dummy) index

i =1
For example, in 3-space:
D = a1H1 + a 2H 2 + a 3H3

.URQHFNHU 'HOWD
With the dual basis we can now introduce a symbol called the
Kronecker delta ij defined by,
1 i = j
Hi H j = ij =
0 i j
Since a vector is invariant to a coordinate transformation, it can
be written in terms of any basis. In particular, we can
represent an arbitrary vector D using the dual basis,

D = (D H1 )H1 + (D H 2 )H 2 + (D H3 )H3
= (ai Hi H1 )H1 + (ai Hi H 2 )H2 + (ai Hi H3 )H3
= (ai1i )H1 + (ai 2i )H 2 + (ai 3i )H3
(2)
= a1H1 + a2H 2 + a3H3
Note that the scalar components for the dual basis are written
with subscripts. In general, we define,
ai = D Hi cogredient scalar components
a i = D Hi contragredient scalar components
Note that ai transform like Hi and ai transform like Hi since in
addition to (2), we can write,
D = (D H1 )H1 + (D H 2 )H 2 + (D H3 )H3
= a1H1 + a 2H 2 + a 3H3
(3)
4

For an arbitrary vector D written in terms of an arbitrary basis,
(2) and (3) can be written as
(4)
D = (D Hi )Hi and (D Hi )Hi
Examples:
Let D = a i Hi and E = b j H j , then
D E = a i b j (Hi H j ) = a i b j i j = a j b j = a1b1 + a 2b2 + a 3b3 .
A second-order tensor might be written as,
I
= ij Hi H j = 11H1H1 + 21H1H 2 + "
5

2UWKRQRUPDO%DVLV
In general, each scalar component and basis vector has
different units. For an orthonormal basis, the basis vectors are
unit vectors (dimensionless) that are mutually perpendicular.
The scalar components then have the units of the vector, i.e.,
unit + orthogonal = orthonormal
In this case the cogredient and contragredient components are
the same, so
[H1H 2H3 ] = H1 (H 2 H3 ) = H1 H1 = 1 Hi Hi = H i
and
D = a1H1 + a2H 2 + a3H 3 .
Here, the ai are physical components that have the units of the
vector.
6

Most engineering applications requiring reference to a specific
coordinate system employ an orthonormal system. The most
commonly used are the rectangular Cartesian, cylindrical, and
spherical coordinate systems. We will later examine each of
these systems in considerable detail.
*UDP6FKPLGW2UWKRQRUPDOL]DWLRQ
Purpose: Construct an orthonormal basis from an arbitrary set
of linearly independent vectors, i.e., starting with the general
basis {H1 , H 2 ," , H n } we will construct the orthonormal basis
{H1 , H 2 ," , H n }.

Why go through the trouble of creating an orthonormal basis?
Because, it is generally easier to work with an orthonormal
basis.
Procedure:
H1 H1 =
1. Given {H1, H2, H3}, normalize

2.
H1
.
| H1 |
Choose H2 and set H2 = H 2 H1.
H2
H2
H1

3. Require , H2 H1
H2
H2
H1
H1 (H 2 H1 ) = H1 H2 H1 = 0 = H1 H 2
2
4. Normalize H2
H 2 =
H2 H 2 (H1 H 2 )H1
=
H2
H2
5. For the remaining vectors, employ the recursion relation,

Hr +1 = H r +1 (H1 H r +1 )H1 (H 2 H r +1 )H 2 " (H r H r +1 )H r
Hr +1
H
H
=
6. Finally, normalize 2
r +1
Hr +1
9

Note: Graham-Schmidt orthonormalization does not
necessarily yield a right-handed system. For a left-handed
system, an appropriate renumbering of the orthonormal base
vectors will create a right-handed system.
3HUPXWDWLRQDQG.URQHFNHU 'HOWD6\PEROV
The Kronecker delta was introduced earlier where it is used in
the dot product of a basis and its dual basis, i.e.,
1 i = j
Hi H j = ij =
.
i
j
For an orthonormal system, the basis and the dual basis are
identical, H j H j . The convention is to choose the subscript
(cogredient basis) so,
10

1 i = j
H i H j = ij =
0 i j
The cross product operation can be represented in index

notation by introducing the permutation symbol (actually a
third-order tensor, often called the permutation tensor or
alternating tensor):
H i H j = ijk H k for a right-handed orthonormal system,
ijk
cyclic permutation of ijk

1
= 1 noncyclic permutation of ijk

0 repeated index
11

Index Notation Examples
Applications of ij and ijk:
D E = (ai H i ) (b j H j )
= ai b j ij
= a1b111 + a1b212 + a1b313
+ a b + a b + a b
2 1 21
2 2 22
2 3 23
+ a3b1 31 + a3b2 32 + a3b3 33

= a b + a b + a b
1 1
2 2
= ai bi
3 3
12

H1
D E = a1
b
1
H 2
a2
b
H 3
a3
b
= (a2b3 a3b2 )H1 + (a3b1 a1b3 )H 2 + (a1b2 a2b1 )H 3

= a H b H
i i
= ai b j ijk H k
7KH LGHQWLW\ ijk imn = jm kn jn km

Note: Vector relations are invariant. It is convenient to
develop relations and do proofs in an orthonormal (usually
rectangular Cartesian) coordinate system because we can
employ ij and ijk.
13

Example: Show that (D E) (F G) = (D F)(E G) (D G)(E F)
(D E) (F G) = (D F)(E G) (D G)(E F)
= a b H c d H
i
j ijk k
n mnp
= ai b j cm dn ijk mnp kp
= ai b j cm dn ijk mnk
= ai b j cm dn ( im jn in jm )
= ai ci b j d j ai di b j c j
= (D F)(E G) (D G)(E F) Q.E.D.
Note: An orthonormal system implies that the scalar

components are the physical components, we will no longer
employ the caret ^ above the physical components for an
orthonormal system.
14

%DVLV'XDODQG&RPSRQHQWV$*UDSKLFDO,OOXVWUDWLRQ
Recall from Eq. (4), D = (D Hi )Hi and D = (D Hi )Hi . Then
D Hi =| D || Hi | cos(D, Hi ) and (D Hi )Hi =| D || Hi | cos(D, Hi )

orthogonal projection of D
ai
=| D | cos(D, Hi )
| Hi |
in the direction of Hi
orthogonal projection of D
ai
=| D | cos(D, Hi )
i
| Hi |
in the direction of H
15

2-D illustration for basis {H1, H2} and dual basis {H1, H2}:
a2H
a2
| H2 |
H2
H1 H 2
j
j
H
H
i
i
H 2 H1
a2
| H2 |
a 2H 2
= a1H1 + a2H 2
a1H1
a1
| H1 |
D = a1H1 + a 2H 2
H1
a1
| H1 |
a1H1
H1
16

An important thing to note in the figure is that vector D does
not change in orientation or magnitude when represented in
either coordinate systemit is invariant. Note, however, that
in general, both the scalar and vector components are different
for different coordinate systems.
a3
6SHFLILFDWLRQRID9HFWRU
Given vector D in terms of a
general basis {H1,H2, H3}
cos =
H3
D H1
D H2
, cos =
|D || H1 |
|D || H 2 |
D H3
cos =
|D || H3 |
H1
H2
a2
a1
17

Associated with the basis {H1,H2,H3} one may write an ordered
triple, e.g., (a1,a2,a3), (a1,,), etc. The numbers are ordered in
that they are placed in the order of the base vector to which
they are associated and the specify the magnitude and direction
of D.
This leads to an analytical definition of a vector as: An
ordered set of numbers that obey certain specific vector
rules. Our task now is to develop these rules.
18

Invariance & Transformation Laws
We stated that a vector is invariant under a coordinate
transformation. This means that we may represent any vector
in terms of the basis of any coordinate system.
We started with a given general covariant basis denoted as
{e1 , e 2 , e3 } . We then introduced a method of constructing the
dual (reciprocal) basis. Again, the purpose of the dual basis
thus far is to enable a simple reciprocal scalar (dot) product
operation. In engineering applications, we often have need to
transform from one coordinate system to another; for example
in an astrodynamics application, we might transform from a
coordinate system fixed to an orbiting satellite to a geocentric
(Earth-centered) system. In fluid mechanics we might
transform from a body-fit coordinate system to a rectangular
Cartesian computational system
1

We now develop the general rules for such a coordinate
transformation. Introduce a new cogredient basis
associated with some new coordinate system, and its
contragredient dual
{e1 , e2 , e3 }
{e 1 , e 2 , e 3 }
As explained earlier, we know a given vector a can be

represented in terms of each basis and its dual, e.g.,
a = a i ei = a j e j = a i ei = a j e j
Now using this relation for the contravariant components and
the similar relation for the covariant components, we have for
the barred system,
2

a s = (a i ei ) e s = (ei e s )a i
as = (a i ei ) es = (ei es )a i
= (a j e j ) e s = (e j e s )a j
= (a j e j ) es = (e j es )a j
Once again use Eqs. (2) and (3),

a = (a e j )ei and a = (a e j )e j
to write the transformation relations for the basis vectors and
duals. Replacing the arbitrary vector a with the specific basis
vectors from the barred system gives the barred basis vectors
in terms of the unbarred basis vectors,
e s = (ei e s )ei
es = (e j es )e j
= (ei e s )ei
= ( e j es )e j

Now define the dot products with the associated transformation
laws:
Cogredient law:
es = asj e j and as = asj a j
Contragredient law:
e s = b sj e j and a s = b sj a j
es = c js e j and as = c js a j
s
is
s
is
e = d ei and a = d ai
Mixed laws:
where
asj e j es
c js e j es
bis e s ei
d is ei e s
4

Be sure to recognize that the choice of letter for the dummy
variables is arbitrary and were made for convenience.
If both systems are orthonormal, then
ei = ( ei e j )e j = ij e j
where
ij = cos( ei , e j )
The asj , etc. are not direction cosines since
a e es
j
s
asj
and cos(e , es ) = j
| e || es |
j
Before continuing, we now visit the elements of linear algebra

5
Elements of Linear Algebra

Our discussion of vector algebra is now focused on coordinate
transformations. The subscript notation results in transformation
relations described by a system of linear equations. This provides
the impetus for us to revisit the basic concepts and operations from
linear algebra.
Matrix Definitions and Basic Concepts
A matrix is a rectangular array of
a11
objects (elements that are
numbers, functions, etc.) with its A [ a ] a21

ij
size indicated by the number of
#
rows and columns, i.e., an mn

a1m
matrix A with m rows and n
columns.
a12 " a1n
a22
amn
1

For a square matrix m = n, the elements a11, a22, , amn are the
main diagonal elements.
Column matrix:
x1
x
2
x=
#

xm
Row matrix:
y = [ y1
y2 " yn ]
We will denote rectangular matrices with uppercase letters and

column and row matrices (also referred to as vectors) with bold
lowercase letters.
2

A submatrix of a matrix A is obtained by deleting certain rows
and/or columns of A .
Transpose: An interchange of the rows and columns, e.g.,
a11
a
21
A [ aij ]
#
a1m
a12 " a1n

a11
a
a22
AT [a ji ] 12
%
amn
a1n
a21 " am1
a22
amn
A diagonal matrix is a square matrix with all off-diagonal

elements zero; e.g.,
b11 0
B = 0 b22
0 0
0
0
b33

A symmetric matrix is a square matrix with [aij] = [aji], or
A = AT .
An antisymmetric (skew-symmetric) matrix is a square matrix
with aij = aji or A = AT. Note that this definition requires the
diagonal elements be zero.
Trace: sum of the diagonal elements.
Lower triangular matrix: A square matrix for which all
elements above the diagonal are zero, e.g.,
a11
L = a21
a31
0
a22
a32
0
0
a33

Upper triangular matrix: A square matrix for which all
elements below the diagonal are zero, e.g.,
a11
L= 0
a12
a22
0
a13
a23
a33
Unit matrix: A square matrix for which all diagonal elements

are one and all off-diagonal elements are zero, e.g.,
1 0 0
I = 0 1 0
0 0 1
Each element is zero in the zero matrix, designated as 0.

5

Rules for Addition and Scalar Multiplication
Addition is only defined for matrices of the same size, e.g.,
C = A + B [cij] = [aij] + [bij] .
Rules:
1. A + B = B + A
2. A + B + C = (A + B) + C = A + (B + C)
3. A + 0 = A
4. A + (A) = 0
Multiplication of a scalar and a matrix: Each element of the
matrix is multiplied by the scalar, e.g., if k is a scalar then,
kA = [kaij] .

Matrix Multiplication
One of the motivations for defining matrix multiplication is
that it allows us to write systems of linear equations in a more
compact form. It also allows us to apply certain other matrix
operations and manipulations that allow us to solve linear
systems.
The product of two matrices A and B is defined by the
operation,
n
C = AB [cij ] = aik bkj ,

k =1
Where n is the column dimension of A and the row dimension

of B, e.g.,

1 2
1 0 2
C=
2
3
1 3 4 1 3
1 2
1 0 2
1 1 + 0 2 + 2 1
2
3
1 3 4
1 3
1 2
1 0 2
3 1 2 + 0 (3) + 2 3
2
3
1 3 4
1 3

1 2
3
8
1 0 2
2
3
1 3 4
1 1 + 3 2 + 4 1
1 3
1 2
8
1 0 2
3
2
3
1 3 4 1 3 9 1 2 + 3 (3) + 4 3
3 8
C =
9 1

Note the matrix C has the same number of rows as the
prefactor A and the same number of columns as the postfactor
B.
To illustrate the importance of the order of the prefactor and
the postfactor, here is an example of the product of a row
vector and a column vector,
1
C = [3 6 1] 2 = [19],

4
but interchanging the order gives,
1
3 6 1
C = 2 [3 6 1] = 6 12 2 .

4
12 24 4
10

This illustrates one of the most important rules of matrix
multiplication: In general, matrix multiplication is not
commutative, i.e.,
1. AB BA, in general.
The prefactor A premultiplies the postfactor B on the left-hand
side of the inequality and A postmultiplies B on the right-hand
side of the inequality.
An additional consequence of the rules of matrix multiplication
is that,
2. AB = 0 does not always imply A = 0, or B = 0, or BA = 0.
Additional rules are:
3. ABC = A(BC) = (AB)C
4. (A + B)C = AC + BC and C(A + B) = CA + CB
11

5. AI = IA = A, where I is the identity (unit) matrix.
6. (AB)T = BTAT
7. If A is a square matrix, the powers of A are defined by
A2 = AA, A3 = AA2.
Matrices and Linear Transformations
The multiplication of a square matrix A and a column vector x
transforms the vector into another column vector b, i.e., matrix
A is an operator that transforms vector x into vector b,
Ax = b.
With this result, lets now return to the vector transformation
laws. In particular, the contragredient component
transformation law is written in index notation as,
12

a s = is a i .
In n = 3 space, this is a system of three linear equations,
a 1 = 11a1 + 21a 2 + 31a 3
a 2 = 12 a1 + 22 a 2 + 32 a 3
a 3 = 13 a1 + 23 a 2 + 33 a 3
Using the rules of matrix multiplication, we can write the
system as
a 1 11
2 2
a = 1
a 3 13

21
22
23
31 a1

32 a 2
33 a 3
13

This form allows us to introduce a number of special
techniques for solving general systems of linear equations.
For the general system Ax = b, the (mn) matrix A is defined
as the coefficient matrix; x and b are both column vectors. If
all the elements of b are zero, the system is called
homogeneous, otherwise, it is called nonhomogeneous.
The augmented matrix is defined by combining b and A as,
a11 " a1n b1
# b2
A = [aij | b j ] = # %
am1 " amn bm
This matrix contains all the information from A and b, and is

easier to work with using the basic row and column operations
we will discuss.
14

Classifying Linear Systems
A linear system is
1. Overdetermined if there are more equations than
unknowns (may or may not have a solution).
2. Determined if m = n (may or may not have a solution).
3. Underdetermined if there are more unknowns than
equations (always has a solution).
Elementary Row Operations
1. Interchange of two rows (columns).
2. Multiplication of a row (column) by a nonzero scalar.
3. Replace a row (column) with scalar multiple of itself and
another row (column).
15

Two linear systems are row equivalent if one is obtainable
from the other by a finite sequence of elementary operations.
This is the condition that guarantees that the solution of a
linear system modified by elementary operations is a solution
of the original system.
The rank of a matrix or augmented matrix is designated r and
is the number of nonzero rows after the matrix has been
reduced to echelon form. The echelon form is the form of the
augmented matrix after the last step of the Gauss elimination.
(Note that one can show that AT is the same rank as A.
16

Existence and Uniqueness Conditions for Linear Systems
1. The linear system of m equations has a solution if and
only if is of the same rank as A.
2. If r = n, the system has a unique solution.
3. If r < n, there are an infinite number of solutions where r
of the unknowns can be expressed in terms of the
remaining n r unknowns.
17

Gaussian Elimination
For Gaussian elimination, the
elementary row operations are
used to reduce the system of
equations to the form shown
here (for r m and a11 0,
c22 0, , krr 0).
a11 x1 + a12 x2 + " + a1n xn = b1

c22 x2 + " + c2 n xn = b2
#
krr xr + " + krn xn = b2
0 = br +1
#
This is the echelon form just

defined and the rules are
essentially those of the general
case,
0 = bm

1. No solution if r < m and one of the numbers br +1 ," , bm is
not zero.
2. One unique solution if r = n and br +1 ," , bm, if present, are
zero.
3. Infinitely many solutions if r < n.
Example: Use Gaussian elimination to show that unique to the
33 system:
3 x1 + x2 x3 = 2
x1 + 4 x2 + x3 = 12
2 x1 + x2 + 2 x3 = 10
is
1
x = 2 .

3
2

Solution:
1. Write the augmented matrix:
3 1 1 2
1 4 1 12
2 1 2 10
2. Row operations, row 1 (R1) is the pivot row and a11 = 3 the
pivot element used to create zeros in the first column,
1
2
R2 R1 R2 and R3 R1 R3
3
3
1
2
1
3
0 11/ 3 4 / 3 34 / 3
0 1/ 3 8 / 3 26 / 3

R2 is now the pivot row and a22 the pivot element,
3
R3 R2 R3
33
1
2
1
3
0 11/ 3 4 / 3
34 / 3
0
84 / 33 252 / 33
0
3. Now that the coefficient matrix is in upper triangular form
(also in echelon form), use back substitution to determine
the unknowns xi,
84
252
R3 : x3 =
x3 = 3
33
33
11
4
34
R2 : x2 + (3) =
x2 = 2
3
3
3
R1 : 3x1 + (2) (3) = 2 x1 = 1

Determinant
For a square (nn) matrix A, the determinant of A is a special
number that, according to Reddy & Rasmussen, in some sense
measures the size of A and indicates whether or not A is
singular (a term we will define later).
We introduce the computation of a general nn determinant by
showing the computation procedure for low-order
determinants. For a 11 matrix, by definition,
det A = |A| = |a11| = a11.
For a 22 matrix A, the determinant is defined according to,
det A =
a11
a12
a21
a22
= a11a22 a12 a21.

A 33 determinant is expanded in terms of the 22,
a11
a12
det A = a21
a31
a22
a32
a13
a22
a23 = a11
a32
a33
a23
a33
a12
a21
a23
a31
a33
+ a13
a21
a22
a31
a32
Higher-order determinants are defined as a sum of multiples of

determinants of (n 1)(n 1) matrices. These
(n 1)(n 1) matrices are the sum of determinants of
(n 2)(n 2) matrices and so on. Computationally, we
reduce the nn determinant until we can evaluate 33 or 22
determinants, for which we have the previous explicit
formulas.

For n 2, define the minor of aij as Mij
Mij = determinant of the (n 1)(n 1) submatrix formed
by deleting row i and column j of A.
And the cofactor of aij as Cij = (1)i + j M ij
We can now define the determinant of a nn (n 2) matrix A
as,
n
det A =| A |= (1) k + j akj M kj cofactor expansion by row k

j =1
n
det A =| A |= (1)i+ k aik M ik cofactor expansion by column k

i =1
Note that one obtains the same result whether the expansion is
by rows or columns!
7

The adjoint matrix is the transpose of the matrix obtained by
replacing each of the elements by its corresponding cofactor,
e.g., for a n = 3 case,
C11 C12
Adj( A) = C21 C22
C31 C32
C13
C23
C33
We will later make use of the adjoint matrix when computing

the inverse of a matrix.

As shown earlier, the vector (cross) product of two vectors a
and b can be expressed in terms of a determinant; the scalar
triple product of three vectors a, b, and c can also be
represented by a determinant,
e1
e 2
e 3
a1
a2
a3
a b = a1
b1
a2
b2
a3 and a (b c) = b1
b3
c1
b2
c2
b3
c3
Additional properties of determinants:

1. det (AB) = det A det B
2. det AT = det A
3. det (kA) = kn det A
9

4. If A is a matrix obtained from A by multiplying a row (or
column) of A by a scalar k, then det A= k det A.
5. If any two rows (or columns) of a determinant are
interchanged, the value of the determinant is multiplied by
1.
6. If two rows (or columns) of A are proportional, the value
of the determinant is zero.
7. The value of a determinant is unchanged by adding a
multiple of one row (or column) to another.
8. A matrix is singular if and only if (iff) its determinant is
zero. As mentioned in property 6., this indicates that two
rows (or columns) are linearly dependent.
10

Inverse of a Matrix
For the nn matrix A, the inverse of A is denoted A1 and is
defined by
AA1 = A1 A = I
If a square matrix A has an inverse, that inverse A-1 is unique
and A is nonsingular. If A does not have an inverse, it is said
to be singular.
If one can compute the inverse of the coefficient matrix of a
linear system of equations Ax = b, then the solution to the
system is obtained using,
AA1x = Ix = A1b x = A1b
11

The inverse A-1 can be computed using the formula,
1
Adj( A)
det A
Example: Compute A-1 from the previous example,
A1 =
3 1 1
A = 1 4 1
2 1 2
1. First compute cofactors:

C11 =
4 1
1 1
1 4
= 7, C12 =
= 0, C13 =
= 7,"
1 2
2 2
2 1
12

2. Construct adjoint matrix and compute determinant
(expanded by first row):
T
7 0 7
7 3
Adj( A) = 3 8 1 = 0 8
5 4 11
7 1
4 1
1 1
1
det A = (3)
(1)
+ (1)
1 2
2 2
2
5
4
11
4
= 28
1
3. Compute A-1
1/ 4 3/ 28 5 / 28
Adj( A)
A1 =
2/7
= 0
1/ 7
det A
1/ 4 1/ 28 11/ 28
13
True or False?
The identity matrix is a linear operator, I.e., it operates on a
vector to transform it into another (different) vector.
Three vectors in a plane are always linearly dependent.
If two vectors are perpendicular, their cross product is the zero
vector.
Given two row vectors a = [a1, a2, a3 ] and b = [b1, b2, b3 ] , the
inner (dot) product can be written as ab = abT.
If a free index appears on one side of an equation, then it must
also appear as a free index on the other side of the equation.
Given the set of basis vectors ei each of the dual basis vectors in
the set ei is constructed to be orthogonal to all the basis vectors.
14

Gauss-Jordan elimination for computing A-1
1. Write the augmented matrix [A|I]
3 1 1 1 0 0
[ A | I ] = 1 4 1 0 1 0
2 1 2 0 0 1
2. Row operations
1
2
R2 R1 R2 and R3 R1 R3
3
3
1
1
0 0
1
3
0 11/ 3 4 / 3 1/ 3 1 0
0 1/ 3 8 / 3 2 / 3 0 1
1

3
R2 R3
33
1
1
0
0
1
3
0 11/ 3 4 / 3
1/ 3
1
0
0
84 / 33 7 /11 1/11 1
0
R3
3. Eliminate the upper elements on the LHS of the partition

to create the identity matrix.
3
33
R2 R2 and
R3 R3
11
84
1
0
0
3 1 1
0 1 4 /11 1/11 3/11
0
1/ 4 1/ 28 11/ 28
1
2 0
2

4
R3 R2 and R1 + R3 R1
11
3 1 0 3/ 4 1/ 28 11/ 28
0 1 0
0
2 / 7 1/ 7
0 0 1 1/ 4 1/ 28 11/ 28
R2
R1 R2 R1 and R1 / 3 R1
1 0 0 1/ 4 3/ 28 5 / 28
0 1 0
0
2 / 7 1/ 7
0 0 1 1/ 4 1/ 28 11/ 28
1/ 4 3/ 28 5 / 28
A 1 = 0
1/ 7
2/7
1/ 4 1/ 28 11/ 28
3

&UDPHUV5XOH
Although not as efficient as other methods for computing the
solution of an nn system of linear equations, Cramers rule is
useful when solving differential and partial differential
equations (e.g., the method of characteristics).
If the determinant of the coefficient matrix, D = det A, of an
nn system of linear equations, A[ E, is not zero, the unique
solution is given by,
D
D
D1
D
, x2 = 2 , x3 = 3 , " , xn = n
D
D
D
D
where Di is the determinant with the i-th column of A replaced
by E.
As an example, we apply Cramers rule to the earlier
numerical example solved with Gauss elimination,
x1 =

3 x1 + x2 x3 = 2
x1 + 4 x2 + x3 = 12
2 x1 + x2 + 2 x3 = 10
1. Compute determinants:
3 1 1
A = 1 4 1
2 1 2
2
1 1
D1 = 12 4
10 1
det A = 3
1 = 28, D2 = 1 12
2
2 10
4 1
1 1
1 4
1
1
= 28
1 2
2 2
2 1
1
3 1
1 = 56, D3 = 1 4 12 = 84
2
2 1 10

2. Solution:
x1 =
D
D1
D
= 1, x2 = 2 = 2, x3 = 3 = 3
D
D
D
Note that it may seem that Cramers rule did not require
substantially more arithmetic operations than Gaussian
eliminationhowever it did. It generally requires O(n3)
basic arithmetic operations to solve a system of nequations using Gauss or Gauss-Jordan elimination, but
O(n4) basic arithmetic operations for Cramers rule. For
our case this is O(27) compared to O(81) operations. In
terms of software packages, as far as I know, Cramers
rule is never used to solve systems of linear equations.
6

(LJHQYDOXHV (LJHQYHFWRUV
As an engineer, you have undoubtedly been introduced to
eigenvalues and possibly eigenvectors. We develop
background here and will later make use of
eigenvalues/vectors in the discussion of second and higherorder tensors.
Given the linear equation A[ = E the vector [ is called the
eigenvector (characteristic vector) and the scalar is the
eigenvalue (characteristic value) of matrix A that characterizes
the length (and sense) of the eigenvector [.
The spectrum of A is the set of eigenvalues of A and the
spectral radius of A is the absolute value of the largest
eigenvalue.
7

Example: Find eigenvalues and eigenvectors of
3 0 0
5 4 0
3 6 1
Solution:
1. Compute roots of the characteristic polynomial
3
D ( ) =
5
3
4
0 = (3 )(4 )(1 ) = 0
6
1
roots: 1 = 3, 2 = 4, 3 =1.
8

These roots are the eigenvalues. They form the spectrum
with a spectral radius of 4.
2. Compute the eigenvectors:
1 = 3
0 = 0
1
2
5 x1 + x2 = 0 set x1 = 1 [ = 5 or 10
27 / 2
27
3 x1 + 6 x2 2 x3 = 0
2 = 4
x1 = 0
0
5 x1 = 0 set x2 = 1 [ = 1

2
3 x1 + 6 x2 3 x3 = 0
9

3 = 1
2 x1 = 0
0
5 x1 + 3 x2 = 0 set x3 = 1 [ = 0

1
3 x1 + 6 x2 = 0
Properties of eigenvalues and eigenvectors of an nn square

matrix A:
1. A has at least one eigenvalue and at most n numerically
different eigenvalues, but it may have fewer than n.
2. If [ is an eigenvector of a matrix A corresponding to an
eigenvalue , so is k[, for any k 0, i.e.,
A[ = [ implies k(A[) = A(k[) = (k[).
10

M is the algebraic multiplicity, the number of times the
root of the characteristic polynomial is repeated, and m
is the geometric multiplicity, the number of independent
eigenvectors corresponding to . According to property 1
above, the sum of algebraic multiplicities equals n and in
general m M.
4. A real matrix may have complex eigenvalues that occur in
conjugate pairs and complex eigenvectors.
5. The eigenvalues of a symmetric matrix (AT = A) are real.
6. The eigenvalues of a skew symmetric matrix (AT = A) are
pure imaginary or zero.
3.
11

(LJHQYHFWRUV 'LDJRQDOL]DWLRQ
Similar matrices have the same spectrum (i.e., same
eigenvalues),
A = T 1 AT : nn matrix A , is similar to A for some nn
matrix T.
This is an important property, particularly for numerical
analysis, to diagonalize (or nearly diagonalize) matrices for
computing approximations to eigenvalues and eigenvectors.
The eigenvectors corresponding to a set of distinct eigenvalues
form a linearly independent set. Thus, these eigenvectors form
a basis.
If an nn matrix A has a basis of eigenvectors, then D = X 1 AX
is diagonal with eigenvalues of A as the entries on the main
diagonal.
12
True or False?
13
Vector Calculus & General Coordinate Systems

The vector algebra included operations involving sums and
products of vectors. The definitions and operations defined in the
linear algebra provide the basis for linear transformations and
matrix operations useful in tensor analysis.
The vector calculus allows us to apply the methods of differential
and integral calculus in the general tensor analysis. We begin with
the usual basic definitions and operations.
'HULYDWLYHRID9HFWRU)XQFWLRQRID6FDODU
H s
s
D
D(t)
D(t + t)
dD
D(t + t ) D(t )
= lim
dt t 0
t
s =| D |
dD
D s ds
= lim
= H s
t
0
dt
s t dt
14

3URGXFW5XOHV
d
dD
dE
(D E) =
E + D
dt
dt
dt
d
dD
dE
(D E) =
(order preserved)
E + D
dt
dt
dt
Note that because a vector is composed of two distinct parts,
magnitude and direction, a nonzero derivative could result
from: a) a change in magnitude but not direction, b) a change
in direction but not magnitude, or c) a change in both
magnitude and direction as illustrated in the previous diagram.
15

For case b), a constant length vector,
| D |= const
d
D
=
D
dt
= const
d
dD
da
| D |
2
(D D = a ) D
= a
= 0
dt
dt
dt
| D |= const
D
d
D
D
const
dt
| D |
In general coordinates, the base vectors are not necessarily

constant in magnitude or direction,
dH
da i
D = a Hi
Hi + a i i
dt
dt
By definition, the base vectors of Cartesian systems have
constant magnitude and direction dHi / dt = .
i
16

Example: Compute the acceleration of a body in a circular
orbit.
Y(t)
U(t)
= H z
U = rH r
Y(t + t)
Y = U = vH t
U(t + t)
Y(t)
Y(t + t)
17

dY d
d
dU
= ( U ) =
U +
dt dt
dt
dt
v dH z
=
U + Y = Y
r dt
v
v2
= H z ( U ) = [H z (H z H r )]
r
r
v2
= [(H z H r )H z (H z H z )H r )]
r
v2
= H r
r
D=
18

Example: Prove
dD d 3D
d dD d 2D
D 2 = D 3
dt dt dt
dt dt
Solution:
0
d D d D dD d 3D
d dD d D dD dD d D

D
=
+ D 2 2 + 3
dt dt dt 2 dt dt dt 2
dt
dt dt
dt
dD d 3D
= D 3 Q.E.D.
dt dt
2
19
True or False?
An orthogonal matrix Q is an nn matrix that is invertible with
the property that its transpose is also its inverse, i.e., Q1 = QT.
The rows (or columns) of an orthogonal matrix form a set of
orthonormal vectors.
True
An orthogonal matrix preserves the length of a vector that it

multiplies, ||Qx|| = ||x||.
True: A rotation matrix is an example. It takes vector x and transforms it
with a pure transformation into another vector y with the same
magnitude.
A system of linear equations is simple if it has a solution and

nonsimple if there is no solution.
False: A system of linear equations is consistent if it has a solution and
inconsistent if there is no solution.
1
True or False?
If a system has more unknowns than equations it is an
overdetermined system.
False: If the system has more unknowns than equations it is
underdetermined. If the system is consistent, some of the variables can
be chosen arbitrarily and the other variables determined in terms of the
arbitrary ones.
Ax = 0 is a homogeneous linear transformation. The solution

x = 0 is the only nontrivial solution.
False: x = 0 is the trivial solution.
MATLAB uses Gaussian elimination with partial pivoting to

solve linear systems Ax = b.
True
A linear system is ill conditioned if a small relative error in the

data causes a large relative error in the computed solution. An
ill-conditioned matrix is close to singular.
True
2
True or False?
For a Cartesian system the axes are independent, when applying
a series of 3D rotations about several axes by matrix
multiplications, the order of the rotations is not important.
False: Except for special cases changing the order changes the result.
The eigenvalue equation is written as f(x) = x where f is linear

function that transforms the vector x into a scalar multiple of
itself. The matrix eigenvalue problem is written as
Ax = x or ( A I ) x = 0
True
Eigenvalues and Eigenvectors are named after the famous

German contemporary of Englishman Sir Isaac Newton.
False: eigen is the German word for own or proper
True or False?
Eigenvalues are computed from the characteristic polynomial
formed from the determinant of the nn matrix. This
polynomial has degree n.
True
If a nn matrix has n distinct eigenvalues, all the eigenvectors

are linearly independent. If a nn matrix has m distinct
eigenvalues (m n), then the matrix has at least m linearly
independent eigenvectors. If there are repeated roots, then the
number of independent eigenvectors is determined from the
rank of the matrix.
True: If is an eigenvalue of multiplicity k, N = n rank(A I) is
the number of independent eigenvectors associated with .

The vector algebra included operations involving sums and
products of vectors. The definitions and operations defined in the
linear algebra provide the basis for linear transformations and
matrix operations useful in tensor analysis.
The vector calculus allows us to apply the methods of differential
and integral calculus in the general tensor analysis. We begin with
the usual basic definitions and operations.
'HULYDWLYHRID9HFWRU)XQFWLRQRID6FDODU
H s
s
D
D(t)
D(t + t)
dD
D(t + t ) D(t )
= lim
dt t 0
t
s =| D |
dD
D s ds
= lim
= H s
t
0
dt
s t dt
1

3URGXFW5XOHV
d
dD
dE
(D E) =
E + D
dt
dt
dt
d
dD
dE
(D E) =
(order preserved)
E + D
dt
dt
dt
Note that because a vector is composed of two distinct parts,
magnitude and direction, a nonzero derivative could result
from: a) a change in magnitude but not direction, b) a change
in direction but not magnitude, or c) a change in both
magnitude and direction as illustrated in the previous diagram.

For case b), a constant length vector,
| D |= const
d
D
=
D
dt
= const
d
dD
da
| D |
2
(D D = a ) D
= a
= 0
dt
dt
dt
| D |= const
D
d
D
D
const
dt
| D |
In general coordinates, the base vectors are not necessarily

constant in magnitude or direction,
dH
da i
D = a Hi
Hi + a i i
dt
dt
By definition, the base vectors of Cartesian systems have
constant magnitude and direction dHi / dt = .
i

Example: Compute the acceleration of a body in a circular
orbit.
Y(t)
U(t)
= H z
U = rH r
Y(t + t)
Y = U = vH t
U(t + t)
Y(t)
Y(t + t)
4

dY d
d
dU
= ( U ) =
U +
dt dt
dt
dt
v dH z
=
U + Y = Y
r dt
v
v2
= H z ( U ) = [H z (H z H r )]
r
r
v2
= [(H z H r )H z (H z H z )H r )]
r
v2
= H r
r
D=

Example: Prove
dD d 3D
d dD d 2D
D 2 = D 3
dt dt dt
dt dt
Solution:
0
d D d D dD d 3D
d dD d D dD dD d D

D
=
+ D 2 2 + 3
dt dt dt 2 dt dt dt 2
dt
dt dt
dt
dD d 3D
= D 3 Q.E.D.
dt dt
2

&DUWHVLDQ&RRUGLQDWH6\VWHPV
A general Cartesian coordinate system is oblique, i.e., the
basis vectors are generally not all mutually orthogonal. As
stated earlier, however, the basis vectors of a Cartesian system
are constant in magnitude and direction.
The usual convention is to refer to the familiar orthonormal
Cartesian system as the Cartesian system, with basis vectors
usually denoted as
{L , M, N }, {H x , H y , H z }, {Li }
z
N
L
( x, y, z ) = ( x1 , x 2 , x 3 ) = ( x1 , x2 , x3 )
U = x j L j
M

In any coordinate system, the differential distance between
two points is given by the differential arclength, computed
from dUdU. In particular, for the Cartesian system,
z
dU dU = (ds ) 2 = dx i dx i
= (dx) 2 + ( dy ) 2 + (dz ) 2
ds
dy
dz
dx
y
x
9

&XUYLOLQHDU&RRUGLQDWHV
Define a coordinate system
( q1 , q 2 , q 3 )
with the coordinate transformation from the Cartesian system,
q1 = q1 ( x1 , x 2 , x 3 )
q 2 = q 2 ( x1 , x 2 , x 3 )
q 3 = q 3 ( x1 , x 2 , x 3 )
10

U
q 3
x3
q3
x2
q1 = const
q2 = const
x1
U
q 2
U
q1
q3 = const
U
q1
q2
11

If the transformation is linear, it defines a Cartesian system.
If the transformation is nonlinear, it defines a curvilinear
system.
The Jacobian of the transformation is defined by the following
determinant,
x1
q1
x1
q 2
x1
q 3
x1
q1
x 2
q1
x 3
q1
x j
x 2
= 1
J=
i
q
q
x 2
q 2
x 2
x1
=
3
q
q 2
x 2
q 2
x 3
q 2
x 3
q1
x 3
q 2
x 3
q 3
x 2
q 3
x 3
q 3
x1
q 3
Ended here
16 SEP
12

If J 0, then J-1 (inverse Jacobian) is defined and the inverse
transformation is also defined,
x1 = x1 (q1 , q 2 , q 3 )
x 2 = x 2 (q1 , q 2 , q 3 )
x 3 = x 3 (q1 , q 2 , q 3 )
The position arrow is
U = U (q i ), i = 1, 2,3
and a differential displacement is then
U
U
U
U
dU = 1 dq1 + 2 dq 2 + 3 dq 3 = i dq i
q
q
q
q
13

The vectors U / q are tangent to the coordinate curves defined
by the intersection of the coordinate surfaces (qi = const).
Using these vectors, we define a unitary basis,
i
U
, i = 1, 2,3
q i
Note, in general, the orientation and magnitude of the basis
vectors are not constant, e.g.,
Hi =
14
Oblique-Cartesian system: Basis vectors

have constant magnitude and orientation
Curvilinear system: Basis vectors

generally have non-constant
magnitude and orientation
15

The coordinate transformation was written for a general
system in terms of the original Cartesian system. We almost
always write the transformations in this manner. In terms of
the original Cartesian system, the unitary basis is given by,
U x j
Hi = i = i Li , i = 1, 2,3
q
q
This is a linear system that is easily written in matrix format.
The coefficient matrix is the Jacobian matrix,
1
1
H1 x / q
H = x1 / q 2
2
H3 x1 / q 3
x 2 / q1
x 2 / q 2
x 2 / q 3
x 3 / q1 L1

3
2
x / q L 2
x 3 / q 3 L3

16

)XQGDPHQWDO0HWULF7HQVRU
In a unitary system, the differential distance separating two
infinitesimally spaced points is,
dU dU = (ds ) 2 = (Hi H j )dq i dq j
Now define the components of the fundamental metric tensor
as,
gij Hi H j
Then,
dU dU = (ds ) 2 = gij dq i dq j
17

In matrix format, the fundamental metric tensor is,
g11
G = g 21
g 31
g12
g 22
g 32
g13
g 23
g 33
Properties of the fundamental metric tensor:

1. Symmetric, i.e.,
Hi H j = H j Hi g ij = g ji
2. The norm (magnitude) of the unitary base vectors is,
| Hi |= (Hi Hi )1/ 2 = ( gii )1/ 2 (no summation)
18

3. Describes the curvature of the space,
a) A flat space has no curvature and is called Euclidean.
In this case, all the gij components are constant.
b) A curved space is called Riemannian In this case,
the gij components are not constant. An example is
Lobachevskian space. This space has hyperbolic
curvature.
We can compare these two spaces by
Euclidean
looking at the geometry of a triangle in
each. For Euclidean geometry, we know
the sum of the interior angles of a triangle is
always 180.
+ + = 180
19

In Lobachevskian geometry, that sum is always less
than 180, the difference being proportional to the
area of the triangle. (Penrose, Roger, The Emperors
New Mind, p. 156).
Lobachevskian
+ + = 180 = const area
20

Example: Find the unitary basis vectors and components of
the fundamental metric tensor for elliptic-cylindrical,
coordinates defined by the following inverse transformation
(a = constant):
x1 = a cosh q1 cos q 2 , x 2 = a sinh q1 sin q 2 , x 3 = q 3
In terms of the Cartesian basis, the unitary basis is,
x i
H1 = 1 Li = a sinh q1 cos q 2 L1 + a cosh q1 sin q 2 L 2
q
x i
H 2 = 2 Li = a cosh q1 sin q 2 L1 + a sinh q1 cos q 2 L 2
q
x i
H 2 = 3 Li = L3
q
21

If J 0, then J1 (inverse Jacobian) is defined and the inverse
transformation is also defined,
x1 = x1 (q1 , q 2 , q 3 )
x 2 = x 2 (q1 , q 2 , q 3 )
x 3 = x 3 (q1 , q 2 , q 3 )
The position arrow is
r = r (q i ), i = 1, 2,3
and a differential displacement is then
r
r
r
r
dr = 1 dq1 + 2 dq 2 + 3 dq 3 = i dq i
q
q
q
q
1

The vectors r / q are tangent to the coordinate curves defined
by the intersection of the coordinate surfaces (qi = const).
Using these vectors, we define a unitary basis,
i
r
, i = 1, 2,3
q i
Note, in general, the orientation and magnitude of the basis
vectors are not constant, e.g.,
ei =
Oblique-Cartesian system: Basis vectors

have constant magnitude and orientation
Curvilinear system: Basis vectors

generally have non-constant
magnitude and orientation
3

The coordinate transformation was written for a general
system in terms of the original Cartesian system. We almost
always write the transformations in this manner. In terms of
the original Cartesian system, the unitary basis is given by,
r x j
corrected
ei = i = i i j , i = 1, 2,3
q
q
This is a linear system that is easily written in matrix format.
The coefficient matrix is the Jacobian matrix,
1
1
e1 x / q
e = x1 / q 2
2
e3 x1 / q 3
x 2 / q1
x 2 / q 2
x 2 / q 3
x 3 / q1 i1

3
2
x / q i 2
x 3 / q 3 i3

4

Fundamental Metric Tensor
In a unitary system, the differential distance separating two
infinitesimally spaced points is,
dr dr = (ds ) 2 = (ei e j )dq i dq j

Now define the components of the fundamental metric tensor
as,
gij ei e j
Then,
dr dr = (ds ) 2 = gij dq i dq j

In matrix format, the fundamental metric tensor is,
g11
G = g 21
g 31
g12
g 22
g 32
g13
g 23
g 33
Properties of the fundamental metric tensor:

1. Symmetric, i.e.,
ei e j = e j ei g ij = g ji
2. The norm (magnitude) of the unitary base vectors is,
| ei |= (ei ei )1/ 2 = ( gii )1/ 2 (no summation)

3. Describes the curvature of the space,
a) A flat space has no curvature and is called Euclidean.
In this case, all the gij components are constant.
b) A curved space is called Riemannian In this case,
the gij components are not constant. An example is
Lobachevskian space. This space has hyperbolic
curvature.
We can compare these two spaces by
Euclidean
looking at the geometry of a triangle in
each. For Euclidean geometry, we know
the sum of the interior angles of a triangle is
always 180.
+ + = 180
7

In Lobachevskian geometry, that sum is always less
than 180, the difference being proportional to the
area of the triangle. (Penrose, Roger, The Emperors
New Mind, p. 156).
Lobachevskian
+ + = 180 = const area

Example: Find the unitary basis vectors and components of
the fundamental metric tensor for elliptic-cylindrical,
coordinates defined by the following inverse transformation
(a = constant):
x1 = a cosh q1 cos q 2 , x 2 = a sinh q1 sin q 2 , x 3 = q 3
In terms of the Cartesian basis, the unitary basis is,
x i
e1 = 1 ii = a sinh q1 cos q 2 i1 + a cosh q1 sin q 2 i 2
q
x i
e 2 = 2 ii = a cosh q1 sin q 2 i1 + a sinh q1 cos q 2 i 2
q
x i
e 2 = 3 ii = i3
q
9

Components of the fundamental metric tensor are:
g11 = e1 e1 = a 2 [sinh 2 (q1 ) cos 2 (q 2 ) + cosh 2 (q1 )sin 2 (q 2 )
= g 22
g33 = 1, g12 = g 21 = g13 = g31 = 0
Components and Bases
Recall,
a = (a ei )ei = (a ei )ei
j
Now set a = e ,
e j = (e j ei )ei
With the definition for the components of G, we have,
10

g ij ei e j contravariant component of the
fundamental metric.
gij ei e j covariant component of the
fundamental metric.
Then according to the cogredient and contragredient
transformation laws raising and lowering of the indices is
accomplished with the following,
e j = gij ei and
a j = gij ai
ej = gij ei and
aj = gij ai .
Note that when dealing with a unitary basis, cogredient
components and vectors are referred to as covariant
components and contragredient components and vectors are
referred to as contravariant components.
11

Also, if we dot both sides of the ej transformation equation,
(ej = gij ei) ek,
then we get the neat result
(4)
jk = gij g ik
For this relation, note the sum over i, e.g.,
11 = g11 g 11 + g 21 g 21 + g31 g 31 = 1,
12 = g11 g 12 + g 21 g 22 + g31 g 32 = 0.
Now with a given unitary basis ei, both sets of fundamental
metric components can be generated via,
12

Now with a given unitary basis ei, both sets of fundamental
metric components can be generated via,
gij = ei ej
ei
e =
i
e j ek
[e1e 2e3 ]
g ij = ei e j
(ijk cyclic)
13

The cross product step is avoided by using the linear
transformation
ej = gij ei
corrected
or in matrix notation,
e1 g11
e = g
2 21
e3 g 31
g12
g 22
g 32
(5)
g13 e1

g 23 e 2
g33 e3
(6)
g13 a1

g 23 a 2
g 33 a 3
(7)
and
a1 g11
a = g
2 21
a3 g31
g12
g 22
g 32
14

To determine the e j in terms of the ei the matrix equation (6)
must be inverted. Let,
g |gij| det G ,
Mij = minor of gij ,
Cij (1)i+jMij = cofactor of gij (no summation).
Employing Cramers rule,
e1
g12
g13
e1 = e 2
e3
g 22
g 32
g 23 =
g 33
e1M 11 e 2 M 21 + e3 M 31 ei Ci1
=
g
g
We obtain similar expressions for e2 and e3. In general then

C
e j = ij ei .
g
15

Continuing in matrix format, you will probably recognize
where this is leading, from the previous section on linear
algebra. Since, gij = gji, the fundamental metric tensor is
symmetric and Cij = Cji, then,
T
1
e j = Cij [ei ] e j = G 1 [ei ]
g
so
T
1
Cij = G 1.
g
We designate the elements of G1 with superscripts, i.e.,
G 1 = g ij .
16

So what have we accomplished with all this? If G = [gij] is
known, we can use linear transformations and the rules of
linear algebra to determine the dual basis and covariant
components without formulae that involve cross products. In
fact, knowing what we now know about systems of linear
equations we could have anticipated this result from the matrix
representation of Eq. (6), i.e.,
[ei ] = G e j
e j = G 1 [ei ].
Another thing to note is the result in Eq. (4) is also anticipated

since, in matrix notation, the Kronecker delta is the unit
matrix,
17

1 0 0
ij = 0 1 0 .
0 0 1
Note that the product in Eq. (4) is just,
jk = gij g ik
I = GG 1.
18

The General Permutation Symbol
In the Cartesian system, the cross product is well defined
analytically and geometrically. What about general
coordinates?
We define the general permutation symbol by the operation
ei e j = E ijk e k (for a right-handed system)
where
Eijk = ei e j ek [e1e2e3 ].
Using det( gij ) = g = [e1e 2e3 ]2 , we then write
Eijk = g ijk and E ijk
1
ijk .
g
19

Physical Components of a Vector
Recall a physical component of a vector is defined by
a i e i = a i ei (no summation).
Then,
a i e i e i = a i ei e i
ei
g
a i = a i ii (no summation).
| ei |
g ii
Therefore, the physical component, in terms of the
contravariant and covariant components is,
a i = a i ei
a i = a i gii and similarily ai = ai g ii (no summation).
20

Orthogonal Curvilinear Coordinate Systems
Because of the many conveniences of orthogonal systems,
most space-coordinate systems used in engineering analysis are
orthogonal. Many of these systems are also curvilinear
systems, in particular, the spherical and cylindrical systems
with which you are familiar. In this section we will look at
orthogonal curvilinear systems and how they relate to our
original Cartesian system.
Scale Factors
Define the scale factors
h1 =| e1 |= g11 , h2 =| e 2 |= g 22 , h3 =| e3 |= g33 .
21

Orthogonal Curvilinear Coordinate Systems
With these definitions, then,
e1 = h1e1 , e 2 = h2e 2 , e3 = h3e 3
For a general curvilinear system, we earlier showed that a
differential displacement is written as,
dr = dq1e1 + dq 2e 2 + dq 3e3 .
Now using the scale factors,
dr = (h1dq1 )e1 + (h2 dq 2 )e 2 + (h3 dq 3 )e 3 .
So, for the arclength, the differential distances are
dr dr = (ds ) 2 = (h1dq1 ) 2 + (h2 dq 2 ) 2 + (h3 dq 3 ) 2
ds1 = h1dq1 , ds2 = h2 dq 2 , ds3 = h3 dq 3
22
q3
r + dr
r
q1
h1dq 2 = (ds ) 2
dr
h1dq 3 = (ds )3
q2
h1dq1 = ( ds )1
23

The scale factors scale the qj to the appropriate magnitude and
dimension for an orthogonal curvilinear system.
In terms of the original Cartesian system,
2
2
2
r x1 x 2 x 3
h1 1 = 1 + 1 + 1
q
q q q
1/ 2
x1 2 x 2 2 x 3 2
r
h2 2 = 2 + 2 + 2
q
q q q
x1 2 x 2 2 x 3 2
r
h3 3 = 3 + 3 + 3
q
q q q
1/ 2
1/ 2
24

Differential Volume Element
In many applications, especially finite-volume and finiteelement methods, you often must determine the volume of a
differential element. For instance, a finite-volume form of the
mass conservation equation in fluid mechanics requires a
computation of the flux of mass through the boundaries, which
must balance the creation of mass inside the volume. In most
applications, the differential cell (volume) is of some variable
shape determined by a curvilinear coordinate system. Here we
introduce a general expression for determining a differential
volume.
Recall how the scalar triple product is related to the volume of
a parallelepiped (with appropriate sign):
25

[e1e2e3] = volume of parallelepiped (with appropriate sign.
In general,
dV = ds1 ds 2 ds3
= dq1e1 dq 2e 2 dq 3e3
= dq1dq 2 dq 3 [e1e 2e3 ]
= dq1dq 2 dq 3 g = dq1dq 2 dq 3 J
For an orthogonal curvilinear system,
dV = dq1e1 dq 2e 2 dq 3e3
= h1h2 h3 dq1dq 2 dq 3 [e1e 2e 3 ]
= h1h2 h3 dq1dq 2 dq 3
26

Finally, for the Cartesian system, the familiar result
dV = dx1dx 2 dx3 = dx dy dz
Note we can gain a bit of insight into the physical meaning of
the Jacobian J. Combining the general expression for the
differential volume element with that for the Cartesian system,
we find,
J=
dV
dx dy dz
=
.
1
2
3
1
2
3
dq dq dq
dq dq dq
This shows that the Jacobian of the transformation is the ratio

of a differential volume in the Cartesian system to that of the
general system. You can also see (if you havent already
discovered this) how the Jacobian is related to the fundamental
metric, .i.e., J = g .
27

Curvilinear Systems: Spherical Coordinates
The curvilinear spherical coordinate system is probably
familiar to all of you. In engineering and physics this
coordinate system is used to take advantage of spherical
symmetry. Lets examine this coordinate system in detail.
The curvilinear transformations and inverse transformations
that define the spherical system are given by,
q1 = r = x 2 + y 2 + z 2
z
2
1
q = = cos
x2 + y2 + z 2
y
q 3 = = tan 1
x
x = r sin cos
y = r sin sin
z = r cos

and scale factors and fundamental metric components are
1/ 2
x 2 y 2 z 2
h1 = hr = + +
r r r
1/ 2
= (sin cos ) 2 + (sin cos ) 2 + cos 2
=1
h2 = h = r
h3 = h = r sin
g11 = h12 = 1
g 11 =
1
=1
2
h1
g 22 = h22 = r 2
g 22 =
1
1
=
h22 r 2
g 33 = h32 = r 2 sin 2
g 33 =
1
1
=
h32 r 2 sin 2
Since this is an
orthogonal
curvilinear
system,
gij = gij = 0, i j
2
= const
r = const
e r
= const

Spherical coordinates basis and dual basis:
ei = hi e i (no summation)
e j = g ij ei
We can now write the basis and dual basis,

e r = e r
e r = e r
e
e = re
e =
r
e
e = r sin e e =
r sin

Basis vectors in terms of the Cartesian basis:
1 x j
e i =
i (no summation in i )
i j
hi q
1 x j
1 x j
1 x j
e r =
i , e =
i , e =
i
r j
j
j
hr q
h q
h q
In matrix format, these three equations are,

cos i x

sin i y
0 i z

So this matrix equation gives spherical basis in terms of the
Cartesian basis.
e r sin cos

e = cos cos

e sin
sin sin
cos sin
cos

Now for the inverse transformation that gives the Cartesian
basis in terms of the spherical basis. We can start with the now
familiar relation,
a = (a ei )ei
ii = (ii e j )e j
Then, for example,
i = (i e )e + (i e )e + (i e )e
x
x
r
r
x
In matrix format,
i x sin cos

i y = sin sin
cos
iz

cos cos
cos sin
sin
sin e r

cos e
0 e
6

Note that if we designate the coefficient matrix of the
transformation as R, then the inverse transformation coefficient
matrix is R1 = RT Thus, R is orthogonal (AEM Sec. 3.5, 3.8).
It can be shown that any orthogonal transformation represents
a rotation (possibly combined with a reflection).
Physical components of an arbitrary vector a:
1
ai (no summation),
hi
A convenient
consequence of using
= ar = a r = ar ,
spherical coordinates
a
= a = ra = ,
is that the position
r
arrow has a single
a
component, i.e.,
= a = r sin a =
.
r sin
r = re r .
a i = ai = hi a i =
a r
a
a

Curvilinear Systems: Cylindrical Coordinates
e Z
Now we examine this

familiar curvilinear
coordinate system in
detail.
z
R
e R
r
Z
x
8

Transformation and inverse transformation:
q1 = R = x 2 + y 2
x = R cos
y = R sin
z=Z
y
q 2 = = tan 1
x
q3 = Z = z
Scale factors:
1/ 2
x 2 y 2 z 2
h1 = hR = + +
R R R
h2 = h = R
h3 = hZ = 1
=1

Fundamental metric components:
1
g11 = h12 = 1
g 11 = 2 = 1
h1
1
1
=
h22 R 2
g 22 = h22 = R 2
g 22 =
g33 = h = 1
1
g = 2 =1
h3
2
3
33
Again, since this is an

orthogonal curvilinear
system,
gij = gij = 0, i j
Basis and dual:

e R = e R
ei = hi e i (no summation)
j
ij
e = g ei
e = Re
e Z = e z
e R = e R
e
e =
R
Z
e = e z
10

In terms of the Cartesian basis:
1 x j
e i =
i j (no summation in i )
hi q i
In matrix format, these three equations are:
e R cos sin 0 i x
e = sin cos 0 i

y
e 0
0
1 i z
Z

and the inverse transformation is:
i x cos

i y = sin
0
iz

sin
cos
0
0 e R
0 e

1 e Z
11

You can again verify that the coefficient matrix R is
orthogonal, i.e., R1 = RT.
The physical components of an arbitrary vector a:
1
a i = ai = hi a i = ai (no summation),
hi
a R = a R = a R = aR ,
a = a = Ra =
R
a Z = aZ = a Z = aZ .
Finally, the position arrow is r = Re R + Ze Z .

12

General Transformation between Two Curvilinear Systems
Up to this point we have explored how to transform to and
from the Cartesian system to a curvilinear system and, in
particular, the spherical and cylindrical systems. But what of
the situation where we need a transformation between
curvilinear systems, say, cylindrical to spherical or vice versa?
We now present the rules for doing these types of
transformations. Consider two curvilinear systems, and as
before denote them as the unbarred and barred systems.
The transformations and inverse transformations are written
(for i = 1, 2, 3) as
13

q i = q i (q j ),
{e1 , e 2 , e3 },
q i = q i (q j ),
{e1 , e2 , e3 },
dr = dq i ei
dr = dq i ei
Since the vector dr must be the same, regardless of the
coordinate system,
e s (dq i ei = dq j e j )
dq i is = dq j (e s e j )
dq s = (e s e j )dq j .
Similarly, if we dot with the dual of the barred system,
e s (dq i ei = dq j e j )
( e s ei )dq i = dq s .
14

Recall from multivariable differential calculus, the chain rule
for a differential,
q s
q s i
j
s
dq =
dq and dq = j dq .
j
q
q
By comparison with the covariant and contravariant
transformation laws in the Vector Algebra section, we see that,
s
q s
q s
s
s
e ej =
j and e ei = i is .
j
q
q
s
We can now write the
15

Covariant Transformation Law:
q j
es = s e j
q
q j
as = s a j .
q
Contravariant Transformation Law:

q s i
e = ie
q
s
q s i
a = i a.
q
s
An example of the summations:

q1
q 2
q 3
e1 = 1 e1 + 1 e 2 + 1 e3 ,
q
q
q
q 1 1 q 1 2 q 1 3
e = 1e + 2e + 3e .
q
q
q
1
16

Note these transformation laws allow one to determine the
basis and dual basis of one system in terms of the other by
using the given transformation relations,
q i = q i (q 1 , q 2 , q 3 ) and q j = q j (q1 , q 2 , q 3 )
An additional relation can also be shown,
q j i
es = q s e j e
q j
i
s = s (e j e i )
q
q j q i
= s j
q q
i
s
Look familiar? In fact, everything

weve done here is similar to the
transformations we derived between
the curvilinear system and Cartesian.
In fact, we could write the present
relations in matrix format just as we
did in the previous section.
17

Actually, none of this should surprise you since the Cartesian
system is just a particular (albeit special for we humans)
curvilinear system.
Analytical Definition of a Vector
If the ordered triples,
(a1 , a2 , a3 ) (a1 , a2 , a3 )
( q1 , q 2 , q 3 ) (q 1 , q 2 , q 3 )
satisfy
q i
aj =
ai ,
q j
18

then a j and a i are the covariant components of vector a.
i
j
Note: If q and q are rectangular Cartesian coordinates, the
covariant and contravariant are identical.
A vector quantity is independent of any coordinate system, thus

is invariant to a coordinate transformation
19

x3
x
x2
a1
x
Example: Vector a
in two Cartesian
systems.
a3
a3
a2
a2
a1
x2
x1
20

Example: Noninvariance of the
position vector
x3
P
r
x3
x2
r0
x1
x2
x1
The barred coordinate system is translated from the unbarred,
r = r0 + r
21

Both are rectangular Cartesian thus,
i = i and x i = x i + x i .
0
i
i
The contravariant transformation law gives, for some a,
x i j x i j
i
a = j a = j a = ij a j = a i .
x
x
Similarly, for these Cartesian systems, the covariant
components are,
ai = ai .
Thus, the components of vector a are unchanged by the
coordinate transformation. But!
x j x j since x j = x j + x0j .
22

Because the tail of the position vector is, by definition,
located at the origin of the coordinate system, it is tied to that
origin. Therefore, the position vector is not invariant to a
coordinate translation. Consequently, the position vector r
is not a vector under a translation transformation.
Some ordered triples are vectors
for certain types of
transformations, but not others.
For instance, r transforms as a
vector for a rotation
transformation.
P
r
23

'HULYDWLYHVRIDQ2UWKRQRUPDO%DVLV
An orthonormal basis vector triad can be viewed as a rotating
rigid body, i.e., the orientation of the triad may change, but the
vectors remain fixed with respect to each other.
The H i have constant unit magnitude
but variable orientation.
H 3
Note we use instead of since

is a finite rotation, thus is not a
vector.
H1
d
H 2
1

Recall, for rigid-body rotation,
Y = U
then,
dU
=
U dU = U
dt
dt
Since |U| = const for rigid-body rotation, we can separately look

at each of the H i using,
U = H i
dH i = H i , i = 1, 2,3
(8)

Example: Cylindrical coordinates
In the cylindrical system, a change in position causes a rigidbody rotation of the basis-vector triad if it involves the angle
(a rotation about the z-axis). Based on this, we develop
expressions for the differential changes.
H Z
H Z
H R
U1
H
H R
U2

H Z
arc-length formula:
s = r for constant r,
ds = r d
= d H Z
H R
dH = d (H R )
= d H R
ds = Rd = d
dH R = d H
For this simple rotation about the z-axis, the differential

change due to the rotation is . In the cylindrical coordinate
system, we have the following for the differentials of the basis
vectors:
4

dH i = H i ,
dH R = H R = d (H H R ) = d H ,
dH = H = d (H Z H ) = d H R ,
dH Z = H Z = d (H Z H Z ) = .
H r
Example: Spherical coordinates

This case is a bit more
complex, since two angles
(,) are involved. A
change in position results in
a superposition of two
angular rotations.
U1
H r
U2

H z
d H
H r
d H z
= d H z + d H
(9)
6

The vector H z is not a member of the spherical triad, so we
write it in terms of the spherical basis,
H z = (H z H r )H r + (H z H )H + (H z H ) H = cos H r sin H .

=0
The superposition of differential rotations, Eq. (9), becomes,

= d cos H r d sin H + d H .
Employing Eq. (8), we write the differentials of the
orthonormal spherical basis vectors,
dH i = H i ,
dH r = H r = d sin H + d H ,
dH = H = d cos H d H r ,
dH = H = d cos H d sin H r .
Sign
corrected
7

We will define the curl vector-differential operator soon,
however we will introduce it now, a bit prematurely perhaps,
to state that for a general orthogonal system,
1
1
= curl(dU ) = ( dU ).
2
2
This expression is obtained by taking the curl of the equation
Y = U (how this is done will become clear later). Here we
have introduced the gradient or del operator and the curl
operation.

&XUORID9HFWRU
The curl of an arbitrary vector D is written as
curl D D
where (del, nabla) is a vector differential operator, the

general form of which we will define. For now, in an
orthogonal curvilinear system,
h1H1
h2H 2
h3H 3
1
curl(dU ) dU =
h1h2 h3
q1
q 2
q 3
h1 (h1dq1 ) h2 (h2 dq 2 ) h3 ( h3dq 3 )
9

H1 3
2
2
2
curl(dU ) =
dq
(
h
)
dq
(
h
3
2 )
h2 h3
q 2
q 3
1
2
3
2
dq
(
h
)
dq
(
h
1
3 )
3
1
q
q
H 3 2
2
1
2
dq
(
h
)
dq
(
h
+
2
1 )
h1h2
q1
q 2
H 2
h1h3
1
Now, for example: dH1 = H1 = ( dU ) H1
2
H
= 2 dq 2 1 (h22 ) dq1 2 (h12 )
2h1h2
q
q
H 3 3
2
1
2
dq
(
h
)
dq
(
h
3
1 )
1
3
h1h3
q
q
10

The differentials dH 2 and dH 3 are computed similarly.
Comparing these results to the total differential
H i
dHi = j dq j ,
q
We can identify the partial derivatives of the orthonormal base
vectors
H1
H1 h1 H 3 h1
,
=
h2 q 2 h3 q 3
q1
H1
H 2 h2
H1 H 3 h3
,
.
=
=
2
1
3
1
h1 q
h1 q
q
q
11

H 2 H1 h1
H 2
H1 h2 H 3 h2
,
,
=
=
1
2
2
1
3
h2 q
h1 q h3 q
q
q
H 2 H 3 h3
.
=
3
2
h2 q
q
H 3 H1 h1
H 3 H 2 h2
,
,
=
=
q1 h3 q 3
q 2 h3 q 3
H 3
H1 h3 H 2 h3
.
=
3
1
2
h1 q h2 q
q
12

5RWDWLQJ5HIHUHQFH)UDPHV
A primary application for the following analysis is in classical
mechanics (dynamics). For this problem of relative motion,
the unbarred frame represents a fixed (inertial) reference
frame. The rotating barred frame may also be translating
with respect to the fixed frame.
x3
x3
E
H
P
3
H3
H1
x1
x2
U0
H1
x1
H2
x2
H2
Corrected
superscripts
13

We will show how the rotation of an observer in the barred
frame affects the measurement of the time rate of change of an
arbitrary vector quantity Eand how this relates to an observer
in the fixed frame.
For simplicity, assume the {H1 , H 2 , H3 } are a constant (magnitude
and direction) basis in the nonrotating frame. Since the basis is
constant, the time derivative of a vector E is
dE dbi
=
Hi .
dt
dt
In the rotating frame, the basis vectors {H1 , H2 , H3 } have constant
magnitude also, but variable orientation. So, for an observer in
the rotating frame,
dH
dE db i
=
Hi + b i i .
dt
dt
dt
14

Since the Hi have constant magnitude, the rate of change is due
only from the rigid-body rotation of the frame, i.e.,
d Hi
= Hi .
dt
Note that because the rotating observer is rotating with the
barred coordinates, the observer does not detect a change in
orientation (to this observer, the inertial frame appears to be
rotating). We then define the time rate of change of E, as
observed by the observer in the rotating frame,
db i
dE
Hi
.
dt
dt rot
Thus we have,
15

dE dE
=
+ b i ( Hi )
dt dt rot
dE
+ ( E).
dt rot
What does this mean? An observer in the inertial frame is not

rotating so sees the absolute derivative dE/dt. The rotating
observer, however, sees the derivative (dE/dt)rot and an
additional part due to the fact that the observer is rotating.
Now define a vector differential operator:
d
d
=
dt dt
+ .
rot
(10)
16

9HORFLW\DQG$FFHOHUDWLRQLQ5RWDWLQJ)UDPHV
Velocity
We now inspect the determination of velocity and acceleration
at point P with respect to a rotating coordinate frame. We
apply the vector differential operator (10) to U = U0 + U ,
dU d
=
(U0 + U )
dt dt rot
dU d U
= 0+
+ U.
dt dt rot
17

where,
dU
= absolute velocity,
dt
dU0
= absolute velocity of the origin of the rotating frame
dt
dU
= velocity of object at P with respect to rotating frame
dt rot
U = velocity of point P in the rotating frame.
Note that when dU0/dt = and ( d U / dt ) rot = , the point P is

fixed with respect to the rotating frame and we have a rigidbody rotation with respect to the inertial frame.
18

Acceleration
Applying the operator (10) to the velocity, we determine the
acceleration,
d 2U d
=
2
dt
dt
dU d U
0 +
rot
dt dt
d 2U0 d 2 U
= 2 + 2
dt
dt
d 2U0 d 2 U
= 2 + 2
dt
dt
+
rot
+
rot
d
dt
+ U
rot
U +
rot
dU
dt
d
dU
U + 2
dt
dt
+ ( U )
rot
+ ( U )
rot
19

where,
d 2U
= absolute acceleration
dt 2
d 2U0
= absolute acceleration of the origin of the rotating frame
dt 2
tangential component of acceleration in the plane
d
U =
dt
of U and (dU / dt ) rot and perpendicular to U
2
dU
dt
= Coriolis acceleration
rot
( U ) = centripetal acceleration
20

Example:
Given: At the instant shown,
= 20 rad/s M
Y
B
OAB
dOAB
= 200 rad/s 2 M
dt
The velocity and acceleration of D,
relative to the rod, are 50 in/s and 600
in/s2 upward, respectively.
A
8
30
O
Z
OAB =
D
X
Find: The velocity and acceleration of

the collar D.
21

Solution:
Note that for this problem, U = U . First find position and
velocity at D,
U = (8 in)(sin30 L + cos30 M) = (4 in)L + (6.93 in)M
Y D = ( Y D )OAB + U
The two parts of the velocity are
( Y D )OAB = (50 in/s)(sin30 L + cos30 M)

= (25 in/s) L + (43.3in/s) M
U = (20 rad/s) M [(4 in) L + (6.93 in) M]
= 80 in/s N
Y D = (25 in/s) L + (43.3 in/s)M + 80 in/s N
22

Now the acceleration,
d 2U0 d 2 U
+ 2
DD =
dt
dt 2
+
OAB
d
U + 2 ( Y D )OAB + ( U )
dt
The individual terms are

d 2U0
dt 2
= (600 in/s 2 )(sin30 L + cos30 M)

OAB
= (300 in/s 2 ) L + (520 in/s 2 ) M

d
U = ( 200 rad/s 2 ) M [(4 in) L + (6.93 in) M]
dt
= (800 in/s 2 ) N
23

2 ( Y D )OAB = 2(20 rad/s) M [(25 in/s L + (43.3 in/s) M]
= (1000 in/s 2 ) N
( U ) = (20 rad/s) M (80 in/s) N
= (1600 in/s 2 ) L
D D = (1300 in/s 2 ) L + (520 in/s 2 ) M (1800 in/s 2 ) N
24

Derivatives of a General Unitary Basis
We have examined the computation of derivatives in an
orthonormal curvilinear system. With the exception of the
Cartesian system, these are the easiest to compute. Now we
will develop more general procedures and formulae for an
arbitrary curvilinear system with a unitary basis. Although the
development is straightforward, we will introduce symbols and
nomenclature that is probably unfamiliar. You must pay close
attention to the definitions.
Computational mechanics (fluid, structures, etc.) when the
physical domain is mapped onto a curvilinear grid is an
example where the basis vectors change with respect to
location in the grid, thus partial derivatives such as ei / q j
must be evaluated.
1

Christoffel symbols and contravariant derivatives
Recall, for an arbitrary vector a,
a = (a ei )ei
If a = ei / q j, we can write
ei ei k
=
e ek .
q j q j
(11)
For a more compact notation, lets define the Christoffel

symbol of the second kind,
ei k
e .
=
j q j
corrected
(12)

Now, the contravariant derivative of the covariant basis is,
ei k
=
j
q
i
ek .
j
(13)
In eq. (13) note the summation over the dummy k. Reference

to the dual basis (ek) is eliminated by introducing the
components of the fundamental metric,
e k = g kr e r ,
ei
k
kr
(
e
=
g
e r ),
j
q
k
kr ei
.
= g
j
j
q
(14)
3

A few more manipulations will enable us to write the
Christoffel symbol completely in terms of components of the
fundamental metric. First, rewrite eq. (14),
k
1 kr ei
ei
=
e
+
e
g
r
r .
j
j
j 2
q
q
(15)
Now use the fact that r(qi) is a continuous function then,

ei
r r
=
=
q j q j q i q i q j
e j
= q i .
(16)
This result might be unexpected. The equality of mixed

second-partial derivatives, however, should not surprise us if
we recall from the differential calculus,
4

2 f
2 f
=
xy yx
iff f (x,y) is continuous. Substituting eq. (16) into eq. (15), we
have,
k
1 kr e j
ei
=
g
e
e
r .
i r
j
j 2
(17)
For the final steps of this development, we now write the dot
products in terms of the fundamental metric components,
e j
e r
e
e
e
e
(
)
,
j
i
r
j
i
i
i
q
q
q
5

or
e j
er =
q i
Then,
e r
e
e
e
(
)
.
j
r
j
q i
q i
1 kr
e r
e r
(
)
(
)
,
=
e
+
e
j
r
j
i
r
i
i
i
j
j
j 2
q q
q
q
1 kr
(
)
(
)
(
)
=
e
e
e
e
e
e
g
q i j r q j i r q r i j ,
j 2
or finally,
k
1 kr g jr gir g ij
= g i + j r .
j 2
q
q
q
(18)
6

Thus, eq. (18) gives the scalar components of the covariant
derivative in eq. (13).
Now we develop a similar relation for the contravariant
derivative of the contravariant basis. Start with,
k
k
e
e
=
(
)
i = 0,
i
q j
q j
ei k
e k
e + ei j = 0,
j
q
q
k
ei k
e k
ei j = j e = .
q
q
i j
(19)
Again employ the relation a = (a ei )ei .

7

If a = e k / q j , we can write
i
e k e k
=
e
i e ,
q j q j
or, using eq. (19),

k i
e k
=
e .
j
q
i j
(20)
Again, eq. (20) is the contravariant derivative of the

contravariant basis.
Finally, without going through the details, one can show,
k
1 g
=
=
g
ln
.

i
i
2 g q
i k q
(21)
8

Christoffel symbol of the first kind
The Christoffel symbol of the first kind, is defined in terms of
derivatives of the transformation relating the curvilinear
system to the original Cartesian system,
[ij , m] =
ei
r
r
e
.
m
m
i
j
j
q q q
q
(22)
Now with a = (a e m )e m ,
m
ei ei
=
e
m e ,
q j q j
or
ei
= [ij , m]e m .
j
q
(23)
(24)
9

The difference between eqs. (23) and (24) and eqs. (12) and
(13) is that eqs. (12) and (13 ) give the contravariant derivative
of the covariant basis in terms of contravariant scalar
components and the covariant basis. Equations (23) and (24)
give the contravariant derivative in terms of covariant scalar
components and the contravariant (dual) basis.
To summarize and show the relation of the two Christoffel
symbols,
ei k
= ek
j
q
i j
ei
= [ij , k ]e k
j
q
Expansion in terms of e k
Expansion in terms of e k
10

The symbols can be directly related with ek = gkmem,
k
ei
m
m
=
e
=
[
,
]
ij
m
g kme .
j
q
i j
(25)
Now we can write [ij,m] in terms of the covariant components

of the fundamental metric,
k
[ij , m] = g km
i
1 gim g jm g ij
.
+
=
j 2 q j
q i q m
(26)
corrected
Example: Orthogonal curvilinear coordinates

For orthogonal curvilinear coordinates, recall,
1
gij = 0, i j
hi gii =
(no summation)
g ii
11

g12 g12 g11
1 1 11 g11 g11 g11
12
+
2
= g 1 + 1 1 + g
1
2
1
1
2
q
q
q
q
q
g13 g13 g11

+g 1 +
3
2
q
q
q
13
1 1 g11 g11 1 1 (h12 )

=
2 1 1 =
2 g11 q
q 2 h12 q1
1 h1
.
=
h1 q1
12

These results of course agree with eqs. (25) and (26).
Before concluding this discussion, it is worth mentioning that
the Christoffel symbol of second kind is most often used in the
literature. It is often denoted with the symbol, ijk .
13

Directional Derivative, Gradient, and Divergence
Directional derivative and gradient
Imagine a day in the Boulder area with a temperature
distribution (F) shown in the contour schematic. With With
Boulder at the origin of the 2-D coordinate system, the
temperature distribution is an example of a scalar field, i.e., a
scalar function of a vector (in this case, the position).
Lyons
Bldr
Ned
Lmt
Nwt
90
Laf
88
Lou
Sup
Bfd
87
89
Wmr
US-36
14

Directional Derivative, Gradient, and Divergence
In general,
= (r) = (q1, q2, q3) ( = temperature) .
Suppose we want to determine the rate of change of roughly
along US-36 towards Denver, indicated by the e direction. To
do this, we develop the directional derivative that gives the
rate of change of at a given r, in the direction of e .
Start with,
d = 1 dq1 + 2 dq 2 + 3 dq 3 = i dq i .
q
q
q
q
Recall, for a differential displacement,
dr = dq i ei .
15

Then,
d =
j
i
dq
e
ei
j
q

= 1 e1 + 2 e 2 + 3 e3 dr
q
q
q
Now with ds = |dr|, we can define the direction by the unit

vector,
dr
e = .
ds
The directional derivative is defined by
j
d
=
e e ,
j
ds
q
(27)
16

and is the rate of change of with respect to s in the direction
of e .
Define the vector,
j
e
(28)
q j
as the gradient vector that points in the direction of maximum
change of . Note that grad is defined in terms of covariant
components scalar components and the contravariant basis.
For a general representation of level surfaces of , i.e., surfaces
for which = const, the unit normal vector is the unit vector
locally perpendicular to a level surface,
grad
17

n
grad
= C1
= C2
r
For the surface, the unit normal is

grad
.
n =
|grad |
d
= e grad
ds
(29)
18

For a closed surface, it is the convention to choose +n pointing
outward.
Now define the del operator, ,

grad e1 1 + e 2 2 + e3 3 ,
q
q
q
.
i
(30)
q
Note that is a vector differential operator, i.e., it has some
properties of a vector. Since we have defined to operate
from left to right, it does not share the commutativity property
of a vector for the dot product, i.e.,
ei
a a
19

In fact, the left side of the inequality is itself an operator.
For orthogonal curvilinear coordinates,
e i
e =
hi
i
(no summation),
e1 e 2
e 3
.
=
+
+
1
2
3
h1 q h2 q
h3 q
(31)
For the Cartesian system,
= ii i .
x
The divergence of a vector a is defined as,
(32)
div a a.
(33)
20

As we will see, a is related to the net efflux of some
vector quantity a per unit volume, at a point in space.
In general,
a i j
i k j
div a = j i + a k
q
i j
i
a i
= j + ai .
q
i j
Now recall another relation you thought wed never use, eq.
(21),
k
1 g
=
=
ln
.
g

i
i
2 g q
i k q
21

Then,
a i a i ( g )
div a = i +
.
i
q
q
g
So we have two alternate forms commonly seen in the
literature,
div a =
1
i
(
),
ga
i
g q
(34)
and
div a =
1
( Ja i ).
i
J q
(35)
Where J is the Jacobian.

22

a = a i hi e i = ai e i
(a i = a i hi = ai no summation)
Then,
div a =
=
i
(
),
h
h
h
a
1
2
3
i
h1h2 h3 q
1
h1h2 h3
(36)
(
)
(
)
(
)
+
+
h
h
a
h
h
a
h
h
a
q1 2 3 1 q 2 1 3 2 q 3 1 2 3 . (37)
Note we intentionally emphasize the physical components in

the form of eq. (37). And finally, for the Cartesian system,
div a =
ai
.
xi
(39)
23

/DSODFLDQ RIDVFDODU
Consider the operation that defines the Laplacian,
div(grad ) = ( ) = 2 ,
where,
grad j H j ,
q
1
div D =
( ga i ).
j
g q
Now write grad using contravariant components with,
H j = g ij Hi
grad = g ij
Hi .

Then,
div(grad ) = 2 =
1
ij
g
g
.
i
j
q
q
(39)
In orthogonal curvilinear coordinates ,

g = ( h1h2 h3 ) 2 and g ii =
2 =
1
hi2
(no summation),
1

ij
h
h
h
g
1 2 3
q j
h1h2 h3 q j
(40)
Expanded, this becomes,

1
=
h1h2 h3
2
h2 h3 h1h3 h1h2
. (41)
+ 2
+ 3
1
1
2
3
q
h
q
q
h
q
q
h
q
Finally, for the Cartesian system we have

2
= i i
g = 1, g =
(42)
x x
The Laplacian appears in many engineering and scientific
applications. In fact Kreyszig says: Laplaces equation,
2 = 0 is the most important partial differential equation of
physics. For engineers, the Laplacian is particularly
important in wave and heat conduction problems.
ij
ij

In fluid mechanics, the velocity potential is used used to
define velocity, i.e., Y = for an irrotational flow. For a
steady, incompressible, inviscid (frictionless), irrotational
flow, the governing equation is Laplaces equation.
= const
flow
streamline
Flow about a circular cylinder showing the

velocity potential in relation to the streamlines.
4

Note the streamlines and = const curves are everywhere
orthogonal, in fact they are used to define the natural
coordinate system used in fluid mechanics.
&XUORID9HFWRU
The curl of a vector is defined by the operation
(43)
curl D D
For the general curvilinear system,
curl D = H j j (a i Hi )
q
a i
H
= H j Hi + a i ij
q
q
a i
j
i k
= H j Hi + a H k .
i j
q
j

Since i is a dummy index, we set i = k (k is also a dummy)
and write,
a k
k
curl D = j + a i (H j Hk ).
i j
q
Now introduce another shorthand notation, define the

covariant derivative of the contravariant component,
a k
i k
a = j + a .
q
i j
indicates derivative w.r.t. jth coordinate
Then we have,
k
,j
(44)
curl D = a,kj (H j H k ) = g ij a,kj (Hi H k ).

6

Recall
Hi H k = Eijk H k = g ijk H k and H i H k = ijk Hk .

Then,
(45)
curl D = g g ij a,kj ikmH m .
For orthogonal curvilinear coordinates, we have already seen
that the curl operation can be written as,
curlD D =
h1H1
h2H 2
h3H 3
h1h2 h3 q1
h1a1
q 2
h2 a 2
.
3
q
h3 a 3
(46)

And, finally for the Cartesian system,
a j
ijk H k .
(47)
xi
To give some physical significance to the curl we note that
the curl of the velocity is related to the rotation of the velocity
field. In fact, one can show that for rigid-body rotation,
curl Y = 2
In other words, the curl of the velocity field is twice the local
rotation.
A vector field D(U) is said to be irrotational or conservative if
curl D = 0. Thus, for example, a static electric field obeys the
Maxwell equation
curl ( = ( = .
curl D =

Now, let us look at a few examples of vector relations and
identities using Cartesian index notation. The tables
(handouts) are a compilation of useful relations and
identitities from Reddy & Rasmussen. These relations can
often be found scattered throughout other references, but not
so neatly organized as in the handouts. Remember, we
exercise in the Cartesian system, because the index notation
is most convenient. Any vector identity or proof that shown
in the Cartesian system must hold for any system because of
the invariance of a vector quantity.

Integral Relations (Theorem of Gauss)
The following relations from vector integral calculus relate
surface integrals to volume integrals. As we will see, these
relations can be used to give a more physical interpretation to
some of the vector differential relations defined in the previous
sections. The vector integral relations form the basis of the
integral and differential conservation laws. For development,
we will focus on the conservation laws of continuum
mechanics. For computational methods, they form the basis of
finite-volume and finite-element methods.
We begin with an arbitrary region R in space enclosed by a
surface S,

n
dS
S
n
R
r

A differential volume element is d, and dS is the magnitude of
a differential surface element whose orientation is determined
by the outward unit normal n . Now introduce three vector
integral theorems,
grad d = w
n dS
Gradient Theorem
diva d = w
n a dS
Divergence Theorem (50)
curla d = w
n a dS
Curl Theorem
(49)
(51)

Using the divergence theorem, we can create a derivative
relation similar to the directional derivative. Set a = ,
div grad d = 2 d = w
n grad dS
R
(52)
Here, we identify the normal derivative,

n
(53)
n
n
is the rate of change of
in the direction
normal to the surface S
enclosing region R.
4

In Cartesian coordinates, the normal derivative is,
= ni ii
i j = nj
n
x j
x j
(54)
Because n is a unit vector, the ni are direction cosines.

Using the integral relations, we can develop more intuitive,
alternative definitions for some of the vector differential
relations introduced earlier. (Note that in the integral approach
we make no references to any coordinate system.)
We begin by examining the limit as the region R shrinks to a
point, i.e.,
R , S S .
5

Why does 0 but S S 0? The integral relations are
now,
grad d = w
n dS grad w
n dS
S
diva d = w
n a dS diva w
n a dS
curla d = w
n a dS curla w
n a dS
Again, since we have not referred to any coordinate system,

and these relations involve vectors, the following are called the
invariant forms,

1
0
grad lim
n dS
(55)
1
0
n a dS
(56)
1
0
w
n a dS
(57)
diva lim
curla lim
Physical interpretations of integral definitions

Note that each of the invariant forms resembles the differencequotient definition of a derivative from the differential calculus
(1-D),
du
u ( x + x) u ( x)
= lim
x
dx x0
where u is some scalar function of x.
7

grad :
= C2
= C3
= C5
= C1
= C4
In eq. (55) the function appears as a weighting function for

n dS , i.e., larger values of n dS contribute more to the total
integral than smaller (in magnitude) values. As 0, the
integral is weighted most in the direction of greatest increase
of , i.e., the term in the direction of grad is the dominant
term in the integral, thus the integral definition corresponds
with the original gradient definition, eq. (28).
8

div a:
v
vn
v
dS
S
vt
R
For illustration, set a = v in eq. (56), where v is the velocity of

some quantity flowing from region R through the surface S.
Note that R and S define a bounded region in space, not
necessarily a physical boundary. Then,
9

vn dS = v n dS outflow (efflux) through dS .
As 0, the integral, eq. (56), becomes,
div v
net outflow
unit volume
(58)
10

curl a:
We use Stokes theorem to show the curl is related to the
circulation of a vector field:
curl a
capping
surface
n
dS
ds
a
11

a ds
(curla) n dS = v

circulation of
a about C
Shrink the region until the capping surface just covers a

differential plane area bounded by the curve C, i.e.,
n
|curl a|n
curl a
Cn
ds
C Cn: bounds
the infinitesimal
plane surface S.
12

n curla v a ds,
C
Let S 0,
1
Cn a ds
n
S 0 S v
circulation around C
n (curla) = n (curla)
.
n
area enclosed by C
n (curla) = n (curla) = lim
13

Example: Show that in a fluid, the bouyancy force f is equal
and opposite to the local gravitational acceleration g and equal
to the weight of the displaced fluid.
Solution:
Pressure is defined as the inward normal stress on a surface,
i.e., it is the (force)/(unit area) at a point on a surface S,
to the
oriented in direction opposite the outward unit normal n
surface. For a body immersed in a static pressure field p(r),
the total force is
f = w
pn dS .
S
The magnitude and direction of the greatest increase in

pressure is grad p = g, where is the mass density. From the
gradient theorem eq. (49),
14
grad d = w
n dS
S
Now with the scalar field set to (r) = p(r),
grad p d = w
n p dS
S
Thus,
f = w
pn dS = grad p d
R
= g d = g d = mg Q.E.D.
R
Eureka! This is Archimedes principal!
15

Example: Show that the continuity equation
+ div v = 0
t
is the local mass conservation relation. Where t is time, is
density, and v is velocity.
Solution:
Begin with a fixed region in space with, = (r,t), v = v(r,t),
and
v mass flux=
mass
.
(area)(time)
16

n
v
S
R fixed in space
The mass conservation law is, (with no mass sources)

rate of change of net flux of mass
mass in region R = across surface S
d
d = w
v n dS
R
S
dt
17

Since R is fixed in space,
R t d + w
S v n dS = 0.
Applying the divergence theorem, the surface integral is
converted to a volume integral, and

+
div
v
d = 0.
R t
Since this relation must hold for any arbitrary region R, the
only way the integral is always zero is if the integrand is zero,
thus
+ div v = 0. Q.E.D.
Stopped here 14 Oct
t
18

Reynolds Transport Theorem
The Reynolds transport theorem allows us to develop general
transport relations. For example, in some medium (gas, liquid,
solid, vacuum), we can write relations that describe the
evolution of the field distribution of some vector (tensor) or
scalar quantity due to the transport or convection of medium
quantities (e.g., mass, momentum, energy, etc.)
Recall, in 1-D integro-differential calculus, the derivative of an
integral function I(t) is given by the Liebniz rule,
I (t ) =
b (t )
a (t )
f ( x)dx,
dI db
da
=
f (b(t ))
f (a (t )).
dt dt
dt
19

For a function of 1-D space and time,
I (t ) =
b (t )
a (t )
f ( x, t )dx,
b ( t ) f
dI db
da
=
f (b(t ), t )
f ( a(t ), t ) +
dx.
a
t
(
)
dt dt
dt
t
Now generalize this to a 3-D system,
n
S (t)
vS
dS
Let Q(r,t) represent some

flow quantity, e.g., density.
In the general case, each
point on the surface S(t)
has a velocity vS.
R(t)
20

I (t ) = Q (r, t )d
the total amount of Q in

region R at time t.
dI
Q
=
d + w
Q(r, t ) v S n dS
R
t
S
t
(
)
(
)
dt
t
(59)
For the special case of a material region,

vS = v velocity of the medium,
i.e., each point on the surface S(t) moves with the local
velocity of the medium. For this case, it is standard to denote
the time derivative with a special notation,
d
D
substantial (material) derivative.

dt
Dt
21

Then,
DI
Q
=
d + w
S (t ) Q(r, t ) v n dS
R ( t ) t
Dt
(60)
Conservation Equations
Using the Reynolds transport theorem, we can develop the
physical conservation laws in the form used in study of
classical tensor, vector, scalar fields.
Mass
The mass dm in a differential volume element d is d. Then
the total mass in the region R(t) is,
R (t )
d .
22

The statement of mass conservation is that the mass contained
in a region R(t) that is moving at the medium velocity v is
conserved,
D
(r, t )d = 0.
R
(
t
)
Dt
(61)
Alternatively, using the Reynolds transport relation eq. (60),
R (t ) t d + w
S (t ) (r, t ) v n dS = 0.
(62)
23

Momentum
Newtons second law is:
sum of external time rate of change
forces
= of momentum
d
=
f
dt (mv).
Now write this for a material region,
momentum of mass
momentum in R(t ) is
dm is vd
R (t )
vd
Momentum conservation is then written as

D
v(r, t )d = f .
R
(
t
)
Dt
(63)
24

( v)
R (t ) t d + w
S (t ) vv n dS = f .
(64)
Energy
The first law of thermodynamics is
heat transfer across
change in system
+ work on the
the
boundary
into
=
total energy

system boundary
the system
E = Q + W
25

The total energy in d is
2
v
total energy in d = N
u + d
2
potential N
kinetic
where u is the specific internal energy and v is speed. We then

write the energy equation as
D
vv
Q W
u
+
d
=
+
R
(
t
)
2
Dt
dt
dt
(65)
26


v v
vv
Q W
u
d
u
v
n
dS
+
+
+
=
+
.
R (t ) t 2 w
S (t ) 2
dt
dt
(64)
Note that we have explicitly used a to designate differential
of heat Q and work W. Why?
27

Integral Relations (Theorem of Gauss)
The following relations from vector integral calculus relate
surface integrals to volume integrals. As we will see, these
relations can be used to give a more physical interpretation to
some of the vector differential relations defined in the previous
sections. The vector integral relations form the basis of the
integral and differential conservation laws. For development,
we will focus on the conservation laws of continuum
mechanics. For computational methods, they form the basis of
finite-volume and finite-element methods.
We begin with an arbitrary region R in space enclosed by a
surface S,
dW
n
dS
S
n
R
r

A differential volume element is dW and dS is the magnitude of
a differential surface element whose orientation is determined
by the outward unit normal n . Now introduce three vector
integral theorems,
gradI dW
diva dW
w
n a dS
Divergence Theorem (50)
curla dW
w
n u a dS
Curl Theorem
w
n I dS
Gradient Theorem
(49)
(51)

Using the divergence theorem, we can create a derivative
relation similar to the directional derivative. Set a = I,
div gradI dW
2I dW
w
n gradI dS
S
(52)
Here, we identify the normal derivative,

n
wI
wn
wI
(53)
{ n I
wn
I in the direction
enclosing region R.
4

In Cartesian coordinates, the normal derivative is,
wI
wn
ni ii
wI
ij
wx j
nj
wI
wx j
(54)
Because n is a unit vector, the ni are direction cosines.

Using the integral relations, we can develop more intuitive,
alternative definitions for some of the vector differential
relations introduced earlier. (Note that in the integral approach
we make no references to any coordinate system.)
We begin by examining the limit as the region R shrinks to a
point, i.e.,
R o 'W , S o 'S .
5

Why does 'W o 0 but S o 'S z 0? The integral relations are
now,
gradI dW
w
n I dS o gradI 'W # w
'S
n I dS
diva dW
w
n a dS o diva 'W # w
curla dW
w
n u a dS o curla 'W # w
n u a dS
'S
n a dS
Again, since we have not referred to any coordinate system,

and these relations involve vectors, the following are called the
invariant forms,

1
'W o0 'W
gradI # lim
'S
n I dS
(55)
1
'W o0 'W
n a dS
(56)
1
'W o0 'W
w
n u a dS
(57)
diva # lim
curla # lim
'S
Physical interpretations of integral definitions

Note that each of the invariant forms resembles the differencequotient definition of a derivative from the differential calculus
(1-D),
du
u ( x 'x) u ( x)
lim
dx 'xo0
'x
where u is some scalar function of x.
7

grad I:
I n
S
R
I = C1
I = C2
I = C3
I n
I = C5
I = C4
In eq. (55) the function I appears as a weighting function for

n dS , i.e., larger values ofI n dS contribute more to the total
integral than smaller (in magnitude) values. As 'W o 0, the
integral is weighted most in the direction of greatest increase
of I, i.e., the term in the direction of grad I is the dominant
term in the integral, thus the integral definition corresponds
with the original gradient definition, eq. (28).
8

div a:
v
vn
v
dS
S
vt
R
For illustration, set a = v in eq. (56), where v is the velocity of

some quantity flowing from region R through the surface S.
Note that R and S define a bounded region in space, not
necessarily a physical boundary. Then,
9

vn dS
v n dS { outflow (efflux) through dS .
As 'W o 0, the integral, eq. (56), becomes,

div v {
net outflow
unit volume
(58)
10

curl a:
We use Stokes theorem to show the curl is related to the
circulation of a vector field:
curl a
capping
surface
dS
ds
a
11

a ds
(curla) n dS v

circulation of
a about C
Shrink the region until the capping surface just covers a

differential plane area bounded by the curve C, i.e.,
|curl a|n
curl a
Cn
'S
ds
a
C o Cn: bounds
the infinitesimal
plane surface 'S.
12

n curla # v a ds,
C
Let 'S o 0,
n (curla)
n (curla)
1
Cn a ds
'S o0 'S v
circulation around C
n (curla) {
.
area enclosed by C
n (curla)
lim
13

Example: Show that in a fluid, the bouyancy force f is equal
and opposite to the local gravitational acceleration g and equal
to the weight of the displaced fluid.
Solution:
Pressure is defined as the inward normal stress on a surface,
i.e., it is the (force)/(unit area) at a point on a surface S,
to the
oriented in direction opposite the outward unit normal n
surface. For a body immersed in a static pressure field p(r),
the total force is
w
pn dS .
S
The magnitude and direction of the greatest increase in

pressure is grad p = Ug, where U is the mass density. From the
gradient theorem eq. (49),
14
gradI dW
w
n I dS
S
Now with the scalar field set to I(r) = p(r),
grad p dW
w
n p dS
S
Thus,
f
w
pn dS
S
grad p dW
U g dW
R
g U dW
R
mg Q.E.D.
Eureka! This is Archimedes principal!
15

Example: Show that the continuity equation
wU
divU v 0
wt
is the local mass conservation relation. Where t is time, U is
density, and v is velocity.
Solution:
Begin with a fixed region in space with, U U(r,t), v = v(r,t),
and
U v { mass flux=
mass
.
(area)(time)
16

n
Uv
S
R fixed in space
The mass conservation law is, (with no mass sources)

rate of change of
mass in region R
d
U dW
R
dt
net flux of mass

across surface S
U v n dS
17

Since R is fixed in space,
wU
R wt dW w
S U v n dS 0.
Applying the divergence theorem, the surface integral is
converted to a volume integral, and
wU
div
U
v

dW 0.
R wt
Since this relation must hold for any arbitrary region R, the
only way the integral is always zero is if the integrand is zero,
thus
wU
divU v 0. Q.E.D.
Stopped here 14 Oct
wt
18

Reynolds Transport Theorem
The Reynolds transport theorem allows us to develop general
transport relations. For example, in some medium (gas, liquid,
solid, vacuum), we can write relations that describe the
evolution of the field distribution of some vector (tensor) or
scalar quantity due to the transport or convection of medium
quantities (e.g., mass, momentum, energy, etc.)
Recall, in 1-D integro-differential calculus, the derivative of an
integral function I(t) is given by the Liebniz rule,
I (t )
dI
dt
b (t )
a (t )
f ( x)dx,
db
da
f (b(t ))
f (a (t )).
dt
dt
19

For a function of 1-D space and time,
I (t )
dI
dt
b (t )
a (t )
f ( x, t ) dx,
b ( t ) wf
db
da
f (b(t ), t )
f ( a (t ), t )
dx.
a
t
(
)
dt
dt
wt
Now generalize this to a 3-D system,
S ( t)
vS
dS
Let Q(r,t) represent some

flow quantity, e.g., density.
In the general case, each
point on the surface S(t)
has a velocity vS.
R(t)
20
Q(r, t )dW
I (t )
the total amount of Q in

region R at time t.
wQ
R (t ) wt dW w
S (t ) Q(r, t ) v S n dS
dI
dt
(59)
For the special case of a material region,

vS = v { velocity of the medium,
i.e., each point on the surface S(t) moves with the local
velocity of the medium. For this case, it is standard to denote
the time derivative with a special notation,
d
D
o
{ substantial (material) derivative.
dt
Dt
21

Then,
wQ
R (t ) wt dW w
S (t ) Q(r, t ) v n dS
DI
Dt
(60)
Conservation Equations
Using the Reynolds transport theorem, we can develop the
physical conservation laws in the form used in study of
classical tensor, vector, scalar fields.
Mass
The mass dm in a differential volume element dW is UdW. Then
the total mass in the region R(t) is,
R (t )
U dW .
22

The statement of mass conservation is that the mass contained
in a region R(t) that is moving at the medium velocity v is
conserved,
D
U (r, t )dW
Dt R ( t )
0.
(61)

wU
R (t ) wt dW w
S (t ) U (r, t ) v n dS
0.
(62)
23

Momentum
Newtons second law is:
sum of external
forces
time rate of change

of momentum
d
f
dt (mv).
Now write this for a material region,
momentum of mass
o momentum in R (t ) is
dm is U vdW
R (t )
U vdW
Momentum conservation is then written as

D
U v (r, t )dW
Dt R ( t )
f.
(63)
24

w( U v)
R (t ) wt dW w
S (t ) U vv n dS
f.
(64)
Energy
The first law of thermodynamics is
change in system
total energy
'E
heat transfer across
the boundary into work on the
system boundary
the system
G Q GW
25

The total energy in dW is
total energy in dW
2
U u U v dW
N N
2
potential
kinetic
where u is the specific internal energy and v is speed. We then

write the energy equation as
D
vv

U
u
dW
R
(
t
)
Dt
2
G Q GW
dt
dt
(65)
26

w
v v
vv

U
u
d
W
U
u
R (t ) wt 2 w
S (t ) 2 v n dS
G Q GW
dt
dt
(64)
Note that we have explicitly used a G to designate differential
of heat Q and work W. Why?
27

Dyadics and Tensors
We have discussed the general notion that a scalar represents a
zero-order tensor and a vector is a first-order tensor. Now
introduce the notion of a second-order tensor and a dyadic.
Begin with an example (Reddy & Rasmussen),
Example: Angular momentum of a rigid body.
H0
dH 0
dW
v (r ) u r
r u ( U dW ) v
= angular momentum
about O for
differential mass
element dm = U dW
r
O
28

H0
(r u v ) U dW
angular momentum vector.
Now write HO in a form to show the explicit dependence on

the angular velocity vector Z,
H0
r u ( u r) U dW .
(65)
Recall the triple vector product,
a u (b u c) b(a c) c(a b).

Then eq. (65) becomes,
H0
r
R (r )r U dW .
(66)
29

For rigid-body rotation, Z is not a function of position, i.e., Z z
Z(r), so we should be able to factor it. To do this, we
introduce the dyad rr.
A dyad is two vectors, side by side, that act as one entity. A
dyadic is a sum of dyads. The unit dyadic is
I
I
e1e1 e 2e 2 e3e3 .
(67)
Like the identity matrix of linear algebra, the unit dyadic
changes a vector into itself, e.g.,
I
I e1 (e1 ) e 2 (e 2 ) e3 (e3 )
e1Z 1 e 2Z 2 e3Z 3
30

The unit tensor is symmetric, thus,
I
I e1e1 e 2e 2 e3e3 .
(68)
and
I
I
(69)
I I .
Returning to the angular momentum example, with the aid of
the unit dyadic, we can factor Z in eq. (65),
I
2
H 0 r I rr U dW .
R
(70)
With this form, we define the moment of inertia tensor,
I
I
2
(71)
I 0 r I rr U dW .
R
and
I
(72)
H 0 I 0 .
31

I
Note, similar to the square matrix in linear algebra, I 0 is an

operator that transforms a vector (column or row matrix) into
another vector (or into itself if the square matrix is the identity
matrix).
We finish this example by developing a relation for rotational
kinetic energy E of the total mass m,
E
1 2 1
mv
v vU dW
2
2 R
1
1
u
r
v
d
(r u v ) U dW
U
W
(
)
2 R
2 R
I
1
1
(r u v ) U dW
I 0
R
2
2
H0
rotational kinetic energy in terms

of the moment of inertia tensor.
32

Note the use of the generic term tensor generally refers to a
2nd-order tensor or dyad. Before discussing detailed rules of
the tensor algebra and calculus, one more illustrative example
of how tensors are used.
stopped here 21 Oct
Example: In a continuous medium, the stress at a given
location depends on the force f at that location and the
orientation of the plane area 'S on which the force acts (stress
= force/area). The stress vector at position r is a function of
two vectors, f (n ) and n .
33

f (n )
fn
n
ft
'S
The stress tensor contains the information necessary to

determine the state of stress at a point. Consider a tetrahedron
shaped element in the continuum,
34

q3
(n )
e 2
(e 2 )
e1
2
(e1 )
1
q2
q1
e 3
(e 3 )
3
From Newtons second law,
fsurface f body
ma,
35

n
S
n c
Sc
S Sn Sc
S c S n c
Sn n c
S cn c
projn cS
Then,
'S2
(n e1 )'S ,
(n e 2 )'S ,
'S3
(n e 3 )'S .
'S1
36

Now employ the divergence theorem,
q3
'h
n
div r dW
w
r n dS
S
Shrinking R to an infinitesimal
element,
r n dS
w
'W #
S
q2
div r
'h'S
.
3
r
'W
q1
37

Newtons second law then becomes,
'S [ 1 (n e1 ) 2 (n e 2 ) 3 (n e 3 )]
U'h
3
'S f
U'h
3
'S a.
Solving for the stress vector S,

1 (n e1 ) 2 (n e 2 ) 3 (n e 3 )]
U'h
3
(a f ).
Continuing to shrink R,
lim n (e11 e 2 2 e 3 3 ).
'ho0
Using tensor notation we write this as,

I
(n ) n .
(72)
38

Lets make a careful distinction,
i V ij e j stress vector,
I
V ij e i e j stress tensor.
The stress vectors are the stresses on the area perpendicular to
the ith-coordinate, in the e j direction.
The stress vector S is on the area perpendicular to the arbitrary
unit normal vector n . The constitutive relation for the stress
tensor depends upon the properties of the medium.
39
Tensor Analysis
q3
(n )
e 2
(e 2 )
e1
2
(e1 )
1
q2
q1
e 3
(e 3 )
3
From Newtons second law,
fsurface f body
ma,
1
Tensor Analysis
n
S
n c
S Sn Sc
S c S n c
Sn n c
S cn c
projn cS
Sc
Then,
'S2
(n e1 )'S ,
(n e 2 )'S ,
'S3
(n e 3 )'S .
'S1
Tensor Analysis
Now employ the divergence theorem,
q3
'h
n
div r dW
Shrinking R to an infinitesimal
element,
r n dS
w
'W #
S
q2
w
r n dS
div r
'h'S
.
3
r
'W
q1
3
Tensor Analysis
Newtons second law then becomes,
'S [ 1 (n e1 ) 2 (n e 2 ) 3 (n e 3 )]
U'h
3
'S f
U'h
3
'S a.
Solving for the stress vector S,

1 (n e1 ) 2 (n e 2 ) 3 (n e 3 )]
U'h
3
(a f ).
Continuing to shrink R,
lim n (e11 e 2 2 e 3 3 ).
'ho0
Using tensor notation we write this as,

I
(n ) n .
(72)
4
Tensor Analysis
Lets make a careful distinction,
i V ij e j stress vector,
I
V ij e i e j stress tensor.
The stress vectors are the stresses on the area perpendicular to
the ith-coordinate, in the e j direction.
The stress vector S is on the area perpendicular to the arbitrary
unit normal vector n . The constitutive relation for the stress
tensor depends upon the properties of the medium.
Tensor Analysis
Dyadic Rules
As defined earlier, a dyad is a second-order tensor represented
by two vectors adjacent to one another, e.g., ab. A dyadic is
composed
of a linear combination of dyads, e.g.,
I
a1b1 a 2b 2 " a nb n .
(73)
I
Define the dot (scalar) product operation between a dyadic
and a vector v,
I
v a1 (b1 v ) a 2 (b 2 v ) " a n (b n v ),
I
(74)
v ( v a1 )b1 ( v a 2 )b 2 " ( v a n )b n .
Note the dot product is generally not commutative. (Can you
think of a situation where it is?)
Tensor Analysis
Define the transpose (conjugate),
IT
b1a1 b 2a 2 " b na n .
Then,
I
v v (a1b1 a 2b 2 " a nb n )
(75)
(76)
(b1a1 b 2a 2 " b na n ) v
IT
v.
Note the correspondence of these rules and results with some
of the rules of matrix and vector multiplication in linear
algebra.
Using index notation, lets explore this further. Each of the
dyads in the dyadic is composed of two vectors, i.e.,
Tensor Analysis
I
ai b i
aij e j bik e k
aij bik e j e k
(77)
I jk e j e k .
In 3-space,
I
I 11e1e1 I 12e1e 2 I 13e1e3
I 21e 2e1 I 22e 2e 2 I 23e 2e3
(78)
I 31e3e1 I 32e3e 2 I 33e3e3
or, using matrix linear algebra notation,
Tensor Analysis
I
e1
e2
I 11 I 12 I 13 e1
e3 I 21 I 22 I 23 e 2

31
32
33
I
I
I e3
(79)
eTj [I ij ]ei .
Again, there is considerable overlap between the current

analysis and the formal linear algebra. There are differences,
however. For instance, we do not distinguish between row and
column vectors when writing
I IT
v v.
Using the linear algebra notation, this would be written as
I IT
T
v v.
9
Tensor Analysis
In general, we can write
I
I jk e j e k (contravariant scalar components)
I jk e j e k (covariant scalar components)

=Ikj e j e k
(mixed scalar components)
Example: Computing stress at a point.

Given the Cartesian system (x1,x2,x3) and the stress tensor
200 400 300
[V ij ] 400 0
0 psi
300 0 100
K
Find: The stress vector at a point on a plane passing through
the point and parallel to the plane
x1 2 x2 2 x3 6 0.
10
Tensor Analysis
Find the normal and tangential components.
Solution:
First we find the normal to the plane,
P
,
n
P ( x1 , x2 , x3 ) x1 2 x2 2 x3 6
| P |
w w w
P i1
i2
i3
( x1 2 x2 2 x3 6)
wx2
wx3
wx1
i 2i 2i
2
3
o n 1
3
Now the stress vector,

I
K
(n ) n
11
Tensor Analysis
S1
S
2
S
3
200 400 300 1/ 3 1600 / 3

400 0
0 2 / 3 400 / 3 psi.

300 0 100 2 / 3 100 / 3

K I
Note that we actually computed n . We can do this
because the stress tensor is symmetric.
1
K
(n )
(1600i1 400i 2 100i3 ) psi
3
For the normal component,
K
V n (n ) n 2600 / 9 psi,
and the tangential component is
100
K
Vt
| |2 V n2
1781 469 psi.
81
corrected
12
Tensor Analysis
Symmetric and Antisymmetric Dyadics
We just showed that the stress tensor is symmetric. Here we
show that the properties of symmetry and antisymmetry (skewsymmetry) are analogous to those of matrices in linear algebra,
I IT
o symmetric.
I ij ei e j
I
I ij e j ei I ji ei e j
IT

o I ij
I ji .
o antisymmetric.
I ij ei e j
I ij e j ei I ji ei e j o I ij I ji .
For the antisymmetric case,
I ii I ii (no summation) o I ii 0.
Therefore, only the j z i components are independent, i.e.,
I 12
I 21 , I 13
I 31 , I 23
I 32 .
13
Tensor Analysis
Thus, there are only three independent components for an
antisymmetric dyadic.
Any dyadic can be separated into a symmetric and
antisymmetric part,
I 1 I IT 1 I IT
[ ] [ ],
2

2

symmetric
antisymmetric
or
I ij
1 ij
1
(I I ji ) (I ij I ji ).
2
2
14
Tensor Analysis
Example:
5 4 3
[I ij ] 2 2 2 , [I ji ]
1 3 5
[I ij ]
5 2 1
4 2 3 ,
3 2 5
10 6 4
0 2 2
1
1
6 4 5 2 0 1 .
2
2
4 5 10
2 1 0
Tensor Analysis
Transformation laws for second-order tensors
The same rules of invariance apply to second-order tensors and
dyadics that apply to vectors (first-order tensors).
Again we use the transformation between two curvilinear
systems, the barred and unbarred systems,
wq j
wq j
es
e j , es
ej ,
s
s
wq
wq
wq s j
wq s j
s
e
e , e
e .
j
j
wq
wq
The following
are possible representations of the generic
I
tensor in the two coordinate systems,
s
Tensor Analysis
I
I ij ei e j
I
I mn em en
Iij ei e j
Imn e m e n
I ij ei e j
Inm em e n
Ii j ei e j
Imn e m en
Transforming the basis, invariance requires that,

m
I
wq n
ij
ij wq
I ei e j I
em en .
wq i wq j
It then follows that we can write
I
I mn em en .
3
Tensor Analysis
So, in general,
mn
Imn
m
n
Imn
wq m wq n
I
wq i wq j
ij
wq i wq j
Iij m n
wq wq
contravariant law
covariant law
wq m wq j
I
wq i wq n
mixed laws
i
n
wq wq
Ii j m j
wq wq
(81)
(82)
i
j
(83)
Tensor Analysis
For the Cartesian system, all transformations are the same, and
wx j
i j E ij i j ,
wxi
where Eij are the direction cosines (symmetric matrix). Then,
comparing to eq. (81), we get
i
i
I
I mn im in
I ij E mi E nj ii i j
and
I mn I ij E mi E nj
or in matrix format
[I ] [ E ][I ][ E ]T .
5
Tensor Analysis
The Unit Tensor
We already introduced the unit tensor (dyadic) in eq. (68) as
I
I ei ei e j e j
or
I
I G ij ei e j G i j ei e j .
(85)
i
Note this is the invariant component form since the G j and G i j
components are either zero (i z j) or one (i = j) in any
coordinate system.
Recall,
ei (ei e m )e m gime m ,
ei
(ei e m )e m
g ime m .
Tensor Analysis
Then,
I
I g ij ei e j contravariant components
gij ei e j covariant components
(86)
(87)
G i j ei e j
(88)
invariant (mixed) components
G e e
Note, a tensor is always invariant when taken as a whole.
What we are referring to here as invariant components is the
fact that in eqs. (88) the scalar components do not change
during a coordinate transformation as they do in eqs. (86) and
(87), since the gij and gij are system dependent.
i
j i
Tensor Analysis
Invariants of a second-order tensor
There are three quantities associated with a second-order
tensor that do not change when the tensor undergoes a
coordinate transformation. They are,
I
I1 trace Iii ,
I2
I3
1 j i
(Ii I j IiiI jj ),
2
det Ii j .
Tensor Analysis
Double-dot Product
Define the operation between two second-order tensors
(89)
ab : cd { (b c)(a d).
Alternatively, you may see the definition
(90)
ab cd { (b c)(a d).
or
ab : cd { (a c)(b d).
(91)
Note that Reddy and Rasmussen mistakenly say that the form
in eq. (91) is not commutative. One can easily see that it is.
We will use the definition in eq. (89).
Tensor Analysis
I
I
For two tensor and ,
I I
: (I ij ei e j ) : (\ mne me n )
I ij\ mn g jm gin .
and
I I
:
(I ij ei e j ) : (\ mne me n )
I ij\ mnG jmG in

I ij\ ij .
10
Tensor Analysis
Tensor Gradient
We have developed the general gradient operation on a scalar
quantity, now let us look at the gradient of a vector,
grad a { a e j
w
( a i ei )
j
wq
wa i
we
e j ei a i ij
wq
wq
i
k
j wa
e j ei a i
ek .
i
j
q
w
(92)
Since i and k are dummies,

wa k
i k
a e j a
ek .
i
j
q
w
11
Tensor Analysis
Using the notation for partial derivatives, we write using the
covariant derivative of the contravariant component ak,
(94)
a a,kj e j e k ,
or using,
(95)
a ak , j e j e k ,
corrected
where,
(96)
wa k
i i
ak , j { j a
wq
k j
is the covariant derivative of the covariant component ak.
In the Cartesian system, we have
a
wa j
ii .
i i j
wq
(97)
12
Tensor Analysis
One can show that a is related to the rotational part of a
vector field, i.e.,
antisymmetric part of a

1
1
b a (a)T { (curl a) u b.
2
2

change in a over distance
b from rotation of a
(97)
rotation of a
Note that this vector identity is #14 in the table (handout).

Compare eq. (97) to the earlier result
1
(curl v ) u r
2
relating the rotation, angular velocity, and linear velocity.
v
13
Tensor Analysis
grad a { a
e i w
(a j e j ).
hi wq i
(98)
Expanding the derivative, we have,
we j
e i wa j
e j aj i
hi wq i
wq
a j we j
1 wa j
e i e j e i i .
hi wq i
hi wq
Now expanding we j / wq i (some of the details are left to the

reader),
1
Tensor Analysis
we j
wq i
w e j

wq i hi
wh j
1 k
e
e

h
k k
j
wq i
h j i j
wh j
1 k

e .
G
h
k
jk
i k
wq
h j i j
No summation on hi thus
must insert to factor e k .
Tensor Analysis
After much more work, employing eq. (18) and
gij
2
i ij
hG , g
ij
G ij
hi2
. (no summation on hi )
one obtains,
we j
wq i
G jk whi G ij wh j

j
k
h
q
h
q
w
w
j
k
e k .
(99)
with no summation with the hj and hk and finally,
ai whi
1 wak ak whi
G

i
ik
hi wq h j wq j
hk wq k
e i e k .
(100)
Again with no summation of the scale factors.

3
Tensor Analysis
Example: Tensor gradient in cylindrical and spherical
coordinates.
For each case we will first write the terms in eq. (100),
incrementing the index i then combine them for the final result.
cylindrical: q1 = R, q2 = I, q3 = z
h1 = 1, h2 = R, h3 = 1
Tensor Analysis
i = 1:
wa2
wa2
1 wa1
e e 1 e1e 2 1 e1e 3
1 1 1
wq
wq
h1 wq
0
0
a1 wh1 0 a2 wh1 a3 wh1

1
2
3
w
w
w
h
q
h
q
h
q
2
3
1
u(G11e1e1 G12e1e 2 G13e1e 3 )

0
0
a1 wh1 0
a1 wh1 0
a1 wh1 0

e e
e e
e e
1 1 1
2 1 2
3 1 3
h2 wq
h3 wq
h1 wq
waI
wa R
wa
e R e I Z e R e z
e R e R
wR
wR
wR
Tensor Analysis
i = 2:
wa1
wa3
wa2

e
e
e
e
wq 2 2 1 wq 2 2 2 wq 2 e 2e 3
a1 wh2 a2 wh20 a3 wh2 0

1
2
3
h
q
h
q
h
q
w
w
w
1
2
3
u(G 21e 2e1 G 22e 2e 2 G 23e 2e 3 )

0
0
a2 wh2
a2 wh2 0
a2 wh20

e e
e e
e e
1 2 1
2 2 2
3 2 3
h2 wq
h3 wq
h1 wq
1
h2
waI
waZ
1 wa R

a

a

e
e
e
e
e
e
e
e
e
e
R I I
I I
I z
I I R
I R
R wI
wI
wI
1 wa R
1 waI
1 wa z

a

a

e
e
e
e
e I e z
R I I
I I R
R wI
R wI
R wI
Tensor Analysis
i = 3:
wa3
1 wa1
wa2
e
e

e
e

e 3e 3
3 1
3 2
wq 3
wq 3
h3 wq 3
a1 wh3 0 a2 wh3 0 a3 wh3 0

1
2
w
w
h
q
h
q
h3 wq 3
2
1
u(G 31e 3e1 G 32e 3e 2 G 33e 3e 3 )
0
0
a3 wh3 0
a3 w0h3
a3 wh3 0

e e
e e
e e
1 3 1
2 3 2
3 3 3
h2 wq
h3 wq
h1 wq
waI
wa R
wa
e z e I z e z e z
e z e R
wz
wz
wz
Tensor Analysis
Combining, we have the cylindrical system,
waI
wa R
wa
e R e R
e R e I Z e R e z
wR
wR
wR
1 wa R
1 waI
1 wa z

aI eI e R
aR eI eI
e I e z
R wI
R wI
R wI
waI
wa R
wa

e z e R
e z e I z e z e z
wz
wz
wz
(101)
Tensor Analysis
Following this example for the cylindrical system, you can
show that for the spherical system, the tensor gradient is,
a
waI
wa
war
e r e r T e r eT
e r e I
wr
wr
wr
1 wa
1 wa
1 waI
r aT eT e r T ar eT eT
eT e I
r wT
r wT
r wI
waT
war
aI sin T
aI cosT
wI
wI

e I e r
e I eT
r sin T
r sin T
waI
ar sin T aT cosT
wI
e I e I

(102)
r sin T
9
Tensor Analysis
Divergence of a Second-Order Tensor
We define the divergence operation for a second-order tensor,
I
I
w
div { ei i (I jk e j e k )
wq
(103)
jk
I
we j
we
i wI
e i e j e k I jk i e k I jk e j ki
wq
wq
wq
jk
m
i wI
jk n

I
e i e j e k I jk
e
e
e
e
m k
j n .
i j
i k
wq
Now relabel the dummy indices and factor,
10
Tensor Analysis
I
wI rs
r
i
rk r
I
e i e r e s I js

e
e
e
e
r s
r s
i
j
i
j
q
w
wI rs
js r
rk r i
I
I

G r es .
i
i j
i j
wq
I wI rs
i rk s
i I js
I
es .
i j
i k
wq
(104)
In the Cartesian system this is
wIij
i j.
wxi
(105)
11
Tensor Analysis
Tensor Gradient Revisited
There are two tensor operations that we have not defined, but
now will because the result of these operations may shed more
light on the physical interpretation of the tensor gradient.
The contraction of a second-order tensor or dyadic is the
result of placing the dot operation between the vectors of each
dyad,
Is { I ij ei e j
I ij gij
Is { Ii j ei e j
Iii
{ Iij ei e j
Iij g ij
{ I ij ei e j
I ii
This operation reduces the order of the tensor by two.

The vector of a second-order tensor or dyadic is the result of
placing the cross product between the vectors of each dyad,
12
Tensor Analysis
V { I ij ei u e j
gI ij H ijk e k
{ Iij ei u e j
gIij g im g jnH mnk e k
{ Ii j ei u e j
gIi j g imH mjk e k
{ I ij e i u e j
gI ij g jmH imk e k .
This operation reduces the order of the tensor by one.

One can show the the following two identities,
I IT
a [ ] { V u a,
(106)
(107)
a [b (b)T ] { ( u b) u a.
13
Tensor Analysis
Equation (107) was briefly discussed as eq. (97).
Equation
I
(106) shows how the antisymmetric part of , which hasIonly
three independent components, is related to the vector of ,
)V.
So for the gradient of vector a, a,
(a) S
a,kj e j e k
a,kk
div a.
(108)
Note that div a involves the diagonal terms a thus is

associated with the symmetric part of a (since the diagonal
terms of the antisymmetric part are always zero).
The vector of a, (a)V is,
(a)V
a,kj e j u e k
g g ji a,kj H ikme m
curl a.
(109)
14
Tensor Analysis
Therefore, (a)V and the antisymmetric part of a are related
to the rotation of the vector field as previously presented.
So we can conclude that the tensor gradient contains
information about the dilitation and rotation of a vector field.
This information is extracted with the internal operations just
defined in eqs. (108) and (109) or when a is used to
transform a vector into another vector, (e.g., b in eq. (107) is
used to transform a into another vector).
IN-CLASS EXAMPLES
Stopped here 2 Nov
15
Tensor Analysis
Integral Theorems for Dyadics and Second-Order Tensors
The vector integral relations (49)-(51) are increased by one
order,
grad a dW
I
div dW
I
curl dW
I
n (curl ) dS
CS
dS
w
na
S
(110)
I
n dS
(111)
I
n u dS
(112)
I
ds
(113)
16
Tensor Analysis
The invariant forms can also be shown as before in eqs. (55)(57)
grad a
I
div
1
'W o0 'W
lim
1
lim
'W o0 'W
dS
na
(110)
I
n dS
(111)
'S
'S
I
curl
I
1
lim
S n u dS
'W o0 'W w
I
I
1
n curl lim
Cn ds
'S o0 'S v
(112)
(113)
Note that (113) and (117) are added for completeness.

17
Tensor Analysis
Eigenvalues and Eigenvectors of Dyadics
We have shown how a dyadic can be written in matrix notation
and some of the matrix operations that apply to dyadics. Let us
now extend this further by computing eigenvalues and
eigenvectors of second-order tensors.
I
The dyadic is an operator that changes a vector into another
vector, e.g.,
I
a b.
There is a special case where only the
b
magnitude and/or sense of the vector is
changed, i.e.,
Oa
I
a
a Oa
where O is the eigenvalue that characterizes the change in a.
18
Tensor
Analysis
I
a Oa 0
I
I
( O I ) a 0
This is the same relation we developed for the homogeneous
system in linear algebra. We can write this equation in matrix
form as,
I12
I13 a1
I11 O
I
I22 O
I23 a2
21
I32
I33 O a3
I31
0
0 .

0
From Cramers theorem, we know that to obtain nontrivial

solutions,
I
I
det( O I ) Iij OG ij 0.
19
Tensor Analysis
From Cramers theorem, we know that to obtain nontrivial
solutions,
I
I
det( O I ) Iij OG ij 0.
Note:
i. Each eigenvalue has an associated eigenvector.
ii. Since complex roots must appear in conjugate pairs,
at least one O is real for n = 3.
I
iii. For symmetric , all O are real.I
For most practical engineering problems is symmetric
(Reddy &
I Rasmussen), so for the remainder of the analysis we
assume is symmetric. Furthermore,
iv. The eigenvectors associated with distinct
eigenvalues of a symmetric dyadic are orthogonal.
20
Tensor Analysis
Computing eigenvectors (symmetric dyadic):
Since only the direction of the eigenvectors is of importance,
we can choose to write the set of eigenvectors as unit vectors,
i.e, if a is a given eigenvector, then the unit vector associated
with a is
a1i1 a2 i 2 a3i3
a
e* r
.
2
2
2
|a|
(119)
a1 a2 a3
Then,
a
O1 o r 1 e1*
| a1 |
O2 o r
a2
| a2 |
e *2
O3 o r
a3
| a3 |
e *3
Choose r to ensure a
right-handed system.
(120)
21
Tensor Analysis
I
The represent the principal directions of .
Axes of a coordinate system aligned with the principal
directions are the principal axes.
The orthogonal set of eigenvectors forms a basis. Thus we can
use
(i j e *j )e *j
and write,
I
I i i I e * e * .
ij i j
Then,
I
I
( O j I ) a j
mn m n
(I e e O j I ) a j
* * *
mn m n
0 (no j summation)
22
Tensor Analysis
* *
Imn
e mG nj O j e *k G kj
Imj* e *m O j e *j
e *j (Imj* e *m
O j e *j ) o Iij*
O jG ij (no j summation)
or,
Iij*
O1 0
0 O
2
0 0
0
0 .
O3
This result shows that a dyadic written in principal axis

coordinates reduces to only components along the diagonal
that are the eigenvalues.
23
Tensor Analysis
Problems with coordinate symmetry will result in a repeated
root, e.g., O1 distinct from O2 = O3. In this case, two directions
are independent and the third is arbitrary. So one can choose
the third eigenvalue such that the principal axes form a righthanded coordinate system, i.e., a3 = a1 u a2.
Example: Find the eigenvalues and eigenvectors of
4 4 0
4 0 0 .
0 0 3
24
Tensor Analysis
Taylor Series Expansion
From differential calculus, any analytic function can be
expanded as an infinite power series, i.e., for a scalar function
of a scalar, u(x),
xG
x+G
25
Tensor Analysis
du
G 2 d 2u

u ( x G ) u ( x) G
dx G 0 2! dx 2 G
"
G n d nu
n! dx
. (121)
G 0
This is called the increment form of the Taylor series.

For a scalar function of two scalar variables,
w
w
u ( x h, y k ) u ( x, y ) h k u ( x, y ) "
wy
wx
n
1 w
w
" h k u ( x, y )
wy
n! wx
(122)
26
Tensor Analysis
u(x,y)
x+h
y+h
Note, can you write aa sample term w

w
h

k
wx
u ( x, y ) ?
from the operator of eq. (122)?
w
y
27
Tensor Analysis
We now generalize these results to a scalar function of a
vector, I(r), or a vector function of a vector, a(r),
I(r +)
r
I(r)
28
Tensor Analysis
I
( I ) "
2!
a
a(r ) a(r ) a
( a) "
2!
I (r ) I (r ) I
(123)
(124)
Recall the Taylor series for the exponential function,
x2
xn
e 1 x " .
2!
n!
Noting the form of (121) - (124), we can define a convenient
operator form of the Taylor series,
x
d
dx
d G 2 d2
G n dn
1 G

"
,
dx 2! dx 2
n ! dx n
(125)
29
Tensor Analysis
Continuum Velocity Field: Relative Motion
Given the velocity at a given point in a continuum velocity
field, we can use the Taylor series expansion to determine the
velocity at a neighboring location. This motion can be
decomposed into a combination of translational, rotational and
shearing (deforming) motions,
v(r)
v (r 'r ) v (r ) 'r v h.o.t.
'r
r
The gradient of the velocity is
written in its symmetric and
antisymmetric parts,
v
r+'r
v(r+'r)
1
1
[v (v )T ] [v (v )T ],
2
2
30
Tensor Analysis
where,
1
I
v
[v (v )T ] { rate of strain tensor,
2
1
1
'r [v (v )T ]
curl v u 'r
2
2
rate of rigid-body rotation.
Thus,
v (r 'r )
I 1
v
(
r
)

'
r
curl v u 'r .
N rateN
2
of strain

translation
rate of rotation
31
Tensor Analysis
The rate of strain tensor looks like
H11 H12 H13
[H ij ] H 21 H 22 H 23
H 31 H 32 H 33
1 wv3 wv1

2 wx1 wx3
1 wv3 wv3 wv3

2 wx3 wx3 wx3

Here is an illustration of the combined straining motions.
extension
(diagonal terms)
+
shear
(off-diagonal terms)
32
Tensor Analysis
Note:
trace[H ij ] H ii
wv1 wv2 wv3

wx1 wx2 wx3
div v
rate of change of volume per unit volume (dilatation).
This concludes our discussion of tensor analysis.
33
True or False?
Let an unbarred system represent an inertial frame and a barred
system, a rotating frame. If b represents some observed vector
quantity, since a vector is invariant, the vector components of b
must remain unchanged.
True or False?
Let an unbarred system represent an inertial frame and a barred
system, a rotating frame. If b represents some observed vector
quantity, since a vector is invariant, the vector components of b
must remain unchanged.
False. The vector components of b change, however b must remain
independent of coordinate system.
x3
x3
e3
x2
r
e3
e2
r0
e1
x1
e1
x1
e2
x2
True or False?
In English, what does the following equation state?
db
dt
db
b i ( u ei )
dt rot
db
( u b).
dt rot
True or False?
In English, what does the following equation state?
db
dt
db
b i ( u ei )
dt rot
db
( u b).
dt rot
An observer in the inertial frame is not rotating so sees the absolute

derivative db/dt. The rotating observer, however, sees the derivative
(db/dt)rot and an additional part due to the fact that the observer is
rotating.
True or False?
What defines a unitary basis. How does such a basis differ from
the Cartesian basis?
True or False?
What defines a unitary basis. How does such a basis differ from
the Cartesian basis?
wr
x3
q3
wq 3
x2
q2 = const
x1
q1 = const
wr
wq 2
r
q1
wr
wq1
q3 = const
q2
6
True or False?
What is the Jacobian, the Jacobian matrix?
True or False?
What is the Jacobian, the Jacobian matrix?
J
dV
dq1dq 2 dq 3
dx dy dz
dq1dq 2 dq 3
The Jacobian of the transformation is the ratio of a differential volume in

the Cartesian system to that of the general system.
True or False?
Scale factors are used to transform an orthogonal curvilinear
basis into an orthonormal basis. The also scale the scalar
components into physical components.
True or False?
Scale factors are used to transform an orthogonal curvilinear
basis into an orthonormal basis. The also scale the scalar
components into physical components, i.e. unitless components.
True
h1 | e1 |
e1
g11 , h2 | e 2 |
h1e1 , e 2
h2e 2 , e3
g 22 , h3 | e3 |
g33 .
h3e 3
So, for the arclength, the differential distances are
dr dr
(ds ) 2
o ds1
(h1dq1 ) 2 (h2 dq 2 ) 2 (h3 dq 3 ) 2

h1dq1 , ds2
h2 dq 2 , ds3
h3 dq 3
10
True or False?
What is the fundamental metric tensor? Is it always symmetric?
Why or why not?
11
True or False?
Explain the directional derivative. How is it computed?
Lyons
Laf
Bldr
Ned
Lmt
Nwt
90
Sup
Bfd
87
89
Wmr
US-36
dr
ds
dI
ds e
88
Lou
wI j
e e
j
wq
12
True or False?
C1
I
r
gradI
C2
dI
ds
e gradI
13
True or False?
What does the Christoffel symbol of the second kind represent?
14
True or False?
What does the Christoffel symbol of the second kind represent?
wei
wq j
wei k
wq j e e k .
wei k
e .
j
wq
15
True or False?
What is the normal derivative?
wI
wn
wI
{ n I
wn
I in the direction
enclosing region R.
16
True or False?
State the integral form of the momentum conservation law in
English:
w( U v)
R (t ) wt dW w
S (t ) U vv n dS
f.
17
True or False?
The stress vector S at a point is computed using the linear
I
transformation (n ) n . what are the terms?
18
True or False?
The stress vector S at a point is computed using the linear
I
transformation (n ) n . what are the terms?
f (n )
fn
n
ft
'S
19
True or False?
A dyadic written in principal axis coordinates reduces to only
components along the diagonal that are the eigenvalues.
20
True or False?
A dyadic written in principal axis coordinates reduces to only
components along the diagonal that are the eigenvalues.
True
21
Ordinary Differential Equations

First-Order Equation:
dy
dt
f (t , y )
Geometric Meaning: Direction Fields

Many ODEs have complicated solution forms or no explicit
formula at all. We can plot the field of tangent vectors, the
direction field to give a qualitative view of the solution.
Example
y ty
0.
Rewrite:
dy
dt
ty.
1

[t,y]=meshgrid(-2:.2:2);
f = t.*y;
theta=atan(f);
u=cos(theta);
v=sin(theta);
quiver(t,y,u,v);
For the point that

passes through (1,2)
the solution is
y (t )
2e ( t
1) / 2

Example (related to van der Pol equation):
t
y 0.1(1 t 2 )
y
ERC for Cooperative Mobile Sensing Systems (CCMSS)

CCMSS Vision
STORM
WILDFIRE
POLAR
CCMSS is a university, government, and industrial partnership

dedicated to the development of sensor-integrated unmanned
vehicle systems with novel capabilities to revolutionize
volumetric, in-situ sensing, and distributed communications
through mobility, cooperation, and affordability. These
integrated systems will enable access and persistence in the
hazardous and hard-to-reach environments associated with
scientific inquiry, natural and man-made disasters, and public
safety. These capabilities will be developed and demonstrated
through three Transforming Thrusts:
1. Wildfire addresses the sensing, communications, and
safety needs to support fire-fighting operations and
increase capabilities for modeling and prediction;
2. Polar addresses heterogeneous mixes of sensorintegrated unmanned vehicles for novel, in-situ data
acquisition in volumes that range from beneath the
ocean surface into the troposphere;
3. Storm addresses the challenges of volumetric in-situ
data acquisition in severe storms, from the ground into
the cloud. It specifically addresses the challenge of
penetrating an active tornado.
The CCMSS partnership is equally dedicated to the
development of an outreach and education program that
embraces diversity to produce educational and industrial
leaders, and a workforce to maintain international leadership.
1
Project 1
In the inertial frame (X,Y,Z), the globe of radius rG (m) is

located at (X,0,Z) (X > rG, Z > rG) and it rotates with a constant
angular speed Z deg/s.
z
Z y
Z
x
Y
v
I
X
Input vector: IN = [X,T,rG,Z,lat,long,v]

Output vector: [latmax,I,G,N,'t]
2

Orbital Motion
m1
m2
d 2r
4S
dT

r
dt 2
r2
dt
d 2T
dr dT
r 2 2
0
dt
dt dt
(r (0), dr (0) / dt ,T (0), dT (0) / dt )
3

Rewrite as a nonlinear system:
x1
r , x2
x3 T , x4
dr
dt
dT
dt
dx
x2
dt
dx2
4S 2
2
x1 x4 2
dt
x1
dx3
dt
dx4
dt
x4

2 x2 x4
x1
4

function xdot=orbit(t,x)
xdot=[x(2);x(1)*x(4)^2-4.0*pi^2/x(1)^2;x(4);-2.0*x(2)*x(4)/x(1)];
initcond=[2.0 0.0 0.0 1.5; 1.0 0.0 0.0 2.0*pi; 2.0 0.0 0.0 4.0];
tspan=linspace(0,5,1000);
options=odeset('RelTol',1e-6,'AbsTol',[1e-6 1e-6 1e-6 1e-6]);
lintype=['- ' '-.' '-.'];
for i = 1:3
[t,x]=ode45('orbit',tspan,[initcond(i,:)]',options);
polar(x(:,3),x(:,1),lintype(2*(i-1)+1:2*i));
hold on
end
Unit Quiz
A piecewise continuous function is mostly continuous but with
a finite number of jump discontinuities.
Unit Quiz
A piecewise continuous function is mostly continuous but with
a finite number of jump discontinuities.
Every polynomial is continuous at each point in R.
Unit Quiz
For the ODE
d n y (t )
d n1 y (t )
dy (t )

"

a0 (t ) y (t ) f (t )
a
(
t
)
a
(
t
)
n 1
1
n
n 1
dt
dt
dt
If the forcing function f(t) has discontinuities at points in the
interval, a unique solution exists only if there are a finite
number of jump discontinuities and the magnitude of the jumps
f 0 f 0 is finite. A solution is defined as a function y(t) that
satisfies the ODE, including at the points of discontinuity in
f(t).
Unit Quiz
What are left and right limits? For a continuous function, these
limits must approach the same value.
Unit Quiz
For a numerical method, truncation error decreases with the
step size and roundoff error increases with step size.
Unit Quiz
On some interval I, the Taylor series yields the power series
expansion for a function f(x) together with an expression for
the remainder after n terms.
f c(a )
f ( n ) (a)
f ( n1) ( x1 )
n
f ( x) f (a)
( x a) "
( x a)
( x a ) n1
1!
2!
(n 1)!

Rn ( x )
for some x1 I .
Unit Quiz
The algorithm ode45 uses a 4th-order and 5th-order RK
approximation to compare the error at each point and adjust the
step size to keep within the error tolerance. This is an example
of automatic step-size control.
Unit Quiz
RK is a one-step method. It computes the point at n + 1 by
using only points from point n. This is in contrast with multistep methods that require points before n to compute n + 1.
Unit Quiz
A stable ODE is stiff when it has a decaying exponential
particular solution with a time constant which is very small
relative to the interval over which it is being solved.
Unit Quiz
ODEs with the solution specified at two different points are
called initial-value problems (IVP), where the locations of the
points generally correspond to the physical boundaries of the
problem.
10
Calculus of Variations & Variational Methods

The calculus of variations is used to obtain extrema of expressions
involving unknown functions called functionals. Applications
range from simple geometric problems to finite-element methods
to optimization theory.
References:
Ewing, G. M., Calculus of Variations with Applications,
Dover, 1985.
Hildebrand, F. B., Methods of Applied Mathematics, 2nd Ed.,
Dover, 1992.
Reddy & Rasmussen, Advanced Engineering Analysis,
Kreiger 1990.
Weinstock, R., Calculus of Variations with Applications to
Physics and Engineering, Dover.
11

Unconstrained Minimization
In preparation for an introduction to the calculus of variations,
recall maxima, minima (extrema), and inflections of functions
from the differential calculus.
x0 critical (stationary)
local maximum
f ( x)
point
df
f max
0
f max f ( x0 )
dx
critical value
x0
local minimum
f ( x)
df
dx
f max
x0
inflection
f ( x)
df
dx
0
x0
12

df
dx
df
dx
0, necessary condition for extremum

x x0
d2 f
2
dx
2
d f
0 2
dx
2
d f
dx 2
x x0
0 o local max
x0
0 o inflection
x0
! 0 o local min
x0
13

Given f ( x, y ) , a necessary condition for a minimum at ( x0 , y0 )
is
df
( x0 , y0 )
wf
wf
dx

dy
wx
wy ( x , y )
0 0
dr f
( x0 , y0 )
Since x and y are independent variables, dx and dy are independent

so,
wf
wx
0 and
wf
wy
14

Constrained Minimization
Now minimize f with a constraint, e.g.,
G ( x, y ) 0
Lagrange Multiplier Method
From (1),
wG
wG
dx
dy
wx
wy
dr G
(1)
We introduce the modified function with no constraints

FL ( x, y, O ) { f ( x, y ) O G ( x, y )
and set
wFL
wF
wF
dFL
dx L dy L d O
wx
wy
wO
(2)
(3)
15

Example:
Find the stationary (critical) point of the function
f ( x, y )
2 x2 y 2 8x y 1
with the constraint 2 x y

Soln:
FL ( x, y, O ) (2 x 2 y 2 8 x y 1) O (2 x y )
wFL
wx
wFL
wy
wFL
wO
0,
2 y 1 O 0, o x
2 x y 0.
4 x 8 2O
0.5,
y 1.0, O
3.0
16

Calculus of Variations: Functionals and Euler Equations
The focus of the calculus of variations is the determination of
maxima and minima of expressions that involve unknown
functions. Here we look at a few classical problems to introduce
some of the concepts.
The Brachistochrone (the one that started it all):

Weinstock gives the problem as it was originally stated by John
Bernoulli in 1696: Given two points A and B in a vertical plane,
to find for the moveable particle M, the path AMB, descending
along which by its own gravity, the beginning to be urged from
point A, it may in the shortest time reach the point B.
17

Reddy & Rasmussen state in engineering terms: Design a chute
between two points A: (0,0) and B: (xb,yb) in a vertical plane such
that a material particle, sliding without friction under its own
weight, travels from point A to point B along the chute in the
shortest time.
xb
A
yb
18

Vector e Tensor Analysis With Linear Algebra (Slides)

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Vector e Tensor Analysis With Linear Algebra (Slides)

Transféré par

Droits d'auteur :

Formats disponibles

PART 1: VECTOR & TENSOR ANALYSIS with LINEAR ALGEBRA

Example: A finite rotation has

A vector can always be written in terms

If (1) cannot be satisfied, the vectors are linearly independent.

Scalar (Dot, Inner) Product

Example: Vector representation of work:

This definition of plane area can be generalized

Example: Determine the projected area from the

Vector Components and Basis

Vector Components and Basis

Vector Components and Basis

D = a i Hi a i Hi sum over a repeated (dummy) index

For example, in 3-space:

Vector Components and Basis

Vector Components and Basis

= a1H1 + a2H 2 + a3H3

Vector Components and Basis

Vector Components and Basis

Vector Components and Basis

Vector Components and Basis

1. Given {H1, H2, H3}, normalize

Choose H2 and set H2 = H 2 H1.

Vector Components and Basis

5. For the remaining vectors, employ the recursion relation,

Vector Components and Basis

Vector Components and Basis

The cross product operation can be represented in index

cyclic permutation of ijk

= 1 noncyclic permutation of ijk

Vector Components and Basis

+ a3b1 31 + a3b2 32 + a3b3 33

Vector Components and Basis

= (a2b3 a3b2 )H1 + (a3b1 a1b3 )H 2 + (a1b2 a2b1 )H 3

7KH LGHQWLW\ ijk imn = jm kn jn km

Vector Components and Basis

Note: An orthonormal system implies that the scalar

Vector Components and Basis

D Hi =| D || Hi | cos(D, Hi ) and (D Hi )Hi =| D || Hi | cos(D, Hi )

Vector Components and Basis

Vector Components and Basis

Vector Components and Basis

Vector Components and Basis

Vector Components and Basis

As explained earlier, we know a given vector a can be

Vector Components and Basis

Once again use Eqs. (2) and (3),

Vector Components and Basis

es = asj e j and as = asj a j

Vector Components and Basis

Before continuing, we now visit the elements of linear algebra

Elements of Linear Algebra

numbers, functions, etc.) with its A [ a ] a21

rows and columns, i.e., an mn

a12 " a1n

Elements of Linear Algebra

We will denote rectangular matrices with uppercase letters and

Elements of Linear Algebra

a12 " a1n

a21 " am1

A diagonal matrix is a square matrix with all off-diagonal

Elements of Linear Algebra

Elements of Linear Algebra

Unit matrix: A square matrix for which all diagonal elements

Each element is zero in the zero matrix, designated as 0.

7KH LGHQWLW\ ijk imn = jm kn jn km