Académique Documents
Professionnel Documents
Culture Documents
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
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
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
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
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
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
C
S
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
H3
H1
H2
a2H2
vector component
15
= 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
D = a1H1 + a 2H 2 + a 3H3
(2)
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
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
H1
.
| H1 |
H2
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
H
H
=
6. Finally, normalize 2
r +1
Hr +1
9
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
ijk
11
2 2 22
2 3 23
2 2
= ai bi
3 3
12
H 2
a2
b
H 3
a3
b
= ai b j ijk H k
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.
14
15
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
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
18
{e1 , e2 , e3 }
{e 1 , e 2 , e 3 }
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
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
es = (e j es )e j
= (ei e s )ei
= ( e j es )e 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
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
a22
amn
1
Column matrix:
x1
x
2
x=
#
xm
Row matrix:
y = [ y1
y2 " yn ]
a1m
a
a22
AT [a ji ] 12
%
amn
a1n
a22
amn
0 0
0
0
b33
a31
0
a22
a32
0
0
a33
a12
a22
0
a13
a23
a33
0 0 1
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
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
4
12 24 4
10
21
22
23
31 a1
32 a 2
33 a 3
13
16
17
0 = bm
is
1
x = 2 .
3
2
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
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
a11
a12
a21
a22
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
Note that one obtains the same result whether the expansion is
by rows or columns!
7
C31 C32
C13
C23
C33
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
10
11
3 1 1
A = 1 4 1
2 1 2
4 1
1 1
1 4
= 7, C12 =
= 0, C13 =
= 7,"
1 2
2 2
2 1
12
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
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
0
84 / 33 7 /11 1/11 1
0
R3
1/ 4 1/ 28 11/ 28
1
2 0
2
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
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 =
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
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
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
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
5 x1 + 3 x2 = 0 set x3 = 1 [ = 0
1
3 x1 + 6 x2 = 0
11
True or False?
13
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
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
| 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 |
16
U(t)
= H z
U = rH r
Y(t + t)
Y = U = vH t
U(t + t)
Y(t)
Y(t + t)
17
D=
18
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
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.
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.
True or False?
Eigenvalues are computed from the characteristic polynomial
formed from the determinant of the nn matrix. This
polynomial has degree n.
True
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
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.
| 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 |
U(t)
= H z
U = rH r
Y(t + t)
Y = U = vH t
U(t + t)
Y(t)
Y(t + t)
4
D=
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
{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
ds
dy
dz
dx
y
x
9
( 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
x3
q3
x2
q1 = const
q2 = const
x1
U
q 2
U
q1
q3 = const
U
q1
q2
11
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
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
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
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
17
g 31
g12
g 22
g 32
g13
g 23
g 33
18
always 180.
+ + = 180
19
Lobachevskian
20
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
r
, i = 1, 2,3
q i
Note, in general, the orientation and magnitude of the basis
vectors are not constant, e.g.,
ei =
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
g 31
g12
g 22
g 32
g13
g 23
g 33
always 180.
+ + = 180
7
Lobachevskian
10
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
e j ek
[e1e 2e3 ]
g ij = ei e j
(ijk cyclic)
13
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
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
G 1 = g ij .
16
[ei ] = G e j
e j = G 1 [ei ].
17
0 0 1
jk = gij g ik
I = GG 1.
18
Eijk = ei e j ek [e1e2e3 ].
Using det( gij ) = g = [e1e 2e3 ]2 , we then write
1
ijk .
g
19
20
21
22
q3
r + dr
r
q1
h1dq 2 = (ds ) 2
dr
h1dq 3 = (ds )3
q2
h1dq1 = ( ds )1
23
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
25
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
dV
dx dy dz
=
.
1
2
3
1
2
3
dq dq dq
dq dq dq
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
x 2 y 2 z 2
h1 = hr = + +
r r r
1/ 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
e = re
e =
r
e
e = r sin e e =
r sin
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
a = (a ei )ei
ii = (ii e j )e j
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
= 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
z
R
e R
r
Z
x
8
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
g 22 = h22 = R 2
g 22 =
g33 = h = 1
1
g = 2 =1
h3
2
3
33
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
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
R
a Z = aZ = a Z = aZ .
13
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
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
15
q j
as = s a j .
q
q s i
a = i a.
q
s
16
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
(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
19
x2
a1
x
Example: Vector a
in two Cartesian
systems.
a3
a3
a2
a2
a1
x2
x1
20
x3
P
r
x3
x2
r0
x1
x2
x1
The barred coordinate system is translated from the unbarred,
r = r0 + r
21
P
r
23
H 3
H1
d
H 2
1
Y = U
then,
dU
=
U dU = U
dt
dt
U = H i
dH i = H i , i = 1, 2,3
(8)
H Z
H Z
H R
U1
H
H R
U2
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
H r
U1
H r
U2
d H
H r
d H z
= d H z + d H
(9)
6
=0
Sign
corrected
7
curl(dU ) dU =
h1h2 h3
q1
q 2
q 3
h1 (h1dq1 ) h2 (h2 dq 2 ) h3 ( h3dq 3 )
9
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
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
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
H
P
3
H3
H1
x1
x2
U0
H1
x1
H2
x2
H2
Corrected
superscripts
13
dE
+ ( E).
dt rot
+ .
rot
(10)
16
dU d
=
(U0 + U )
dt dt rot
dU d U
= 0+
+ U.
dt dt rot
17
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
dU
dt
= Coriolis acceleration
rot
( U ) = centripetal acceleration
20
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
d 2U0 d 2 U
+ 2
DD =
dt
dt 2
+
OAB
d
U + 2 ( Y D )OAB + ( U )
dt
24
ei ei k
=
e ek .
q j q j
(11)
ei k
e .
=
j q j
corrected
(12)
ek .
j
(13)
(
e
=
g
e r ),
j
q
k
kr ei
.
= g
j
j
q
(14)
3
1 kr ei
ei
=
e
+
e
g
r
r .
j
j
j 2
q
q
(15)
e j
= q i .
(16)
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
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
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)
i
e k e k
=
e
i e ,
q j q j
e .
j
q
i j
(20)
(21)
8
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
Expansion in terms of e k
Expansion in terms of e k
10
(25)
1 gim g jm g ij
.
+
=
j 2 q j
q i q m
(26)
corrected
= g 1 + 1 1 + g
1
2
1
1
2
q
q
q
q
q
q
q
13
2 g11 q
q 2 h12 q1
1 h1
.
=
h1 q1
12
13
Ned
Lmt
Nwt
90
Laf
88
Lou
Sup
Bfd
87
89
Wmr
US-36
14
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
j
i
dq
e
ei
j
q
= 1 e1 + 2 e 2 + 3 e3 dr
q
q
q
j
ds
q
(27)
16
17
grad
= C1
= C2
r
d
= e grad
ds
(29)
18
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
(no summation),
e1 e 2
e 3
.
=
+
+
1
2
3
h1 q h2 q
h3 q
(31)
= ii i .
x
The divergence of a vector a is defined as,
(32)
div a a.
(33)
20
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
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)
(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)
ai
.
xi
(39)
23
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 .
1
ij
g
g
.
i
j
q
q
(39)
1
hi2
(no summation),
1
ij
h
h
h
g
1 2 3
q j
h1h2 h3 q j
(40)
h2 h3 h1h3 h1h2
. (41)
+ 2
+ 3
1
1
2
3
q
h
q
q
h
q
q
h
q
ij
= const
flow
streamline
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
(44)
curlD D =
h1H1
h2H 2
h3H 3
h1h2 h3 q1
h1a1
q 2
h2 a 2
.
3
q
h3 a 3
(46)
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 =
dS
S
n
R
r
grad d = w
n dS
Gradient Theorem
diva d = w
n a dS
curla d = w
n a dS
Curl Theorem
(49)
(51)
div grad d = 2 d = w
n grad dS
R
(52)
(53)
n
n
is the rate of change of
in the direction
normal to the surface S
enclosing region R.
4
= ni ii
i j = nj
n
x j
x j
(54)
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
grad lim
n dS
(55)
1
0
n a dS
(56)
1
0
w
n a dS
(57)
diva lim
curla lim
= C2
= C3
= C5
= C1
= C4
v
vn
v
dS
S
vt
R
net outflow
unit volume
(58)
10
curl a
capping
surface
n
dS
ds
a
11
circulation of
a about C
n
|curl a|n
curl a
Cn
ds
C Cn: bounds
the infinitesimal
plane surface S.
12
Let S 0,
1
Cn a ds
n
S 0 S v
circulation around C
n (curla) = n (curla)
.
n
area enclosed by C
13
grad d = w
n dS
S
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
15
+ 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
R fixed in space
d
d = w
v n dS
R
S
dt
17
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
b (t )
a (t )
f ( x)dx,
dI db
da
=
f (b(t ))
f (a (t )).
dt dt
dt
19
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
n
S (t)
vS
dS
R(t)
20
dI
Q
=
d + w
Q(r, t ) v S n dS
R
t
S
t
(
)
(
)
dt
t
(59)
(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
R
(
t
)
Dt
(61)
R (t ) t d + w
S (t ) (r, t ) v n dS = 0.
(62)
23
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
R
(
t
)
Dt
(63)
24
(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
2
v
total energy in d = N
u + d
2
potential N
kinetic
u
+
d
=
+
R
(
t
)
2
Dt
dt
dt
(65)
26
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
dW
n
dS
S
n
R
r
gradI dW
diva dW
w
n a dS
curla dW
w
n u a dS
Curl Theorem
w
n I dS
Gradient Theorem
(49)
(51)
div gradI dW
2I dW
w
n gradI dS
S
(52)
wI
wn
wI
(53)
{ n I
wn
is the rate of change of
I in the direction
normal to the surface S
enclosing region R.
4
ni ii
wI
ij
wx j
nj
wI
wx j
(54)
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
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
I n
S
R
I = C1
I = C2
I = C3
I n
I = C5
I = C4
v
vn
v
dS
S
vt
R
net outflow
unit volume
(58)
10
curl a
capping
surface
dS
ds
a
11
circulation of
a about C
|curl a|n
curl a
Cn
'S
ds
a
C o Cn: bounds
the infinitesimal
plane surface 'S.
12
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
w
pn dS .
S
gradI dW
w
n I dS
S
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.
15
U v { mass flux=
mass
.
(area)(time)
16
R fixed in space
d
U dW
R
dt
U v n dS
17
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
I (t )
dI
dt
b (t )
a (t )
f ( x)dx,
db
da
f (b(t ))
f (a (t )).
dt
dt
19
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
S ( t)
vS
dS
R(t)
20
Q(r, t )dW
I (t )
wQ
R (t ) wt dW w
S (t ) Q(r, t ) v S n dS
dI
dt
(59)
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
0.
(61)
0.
(62)
23
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
f.
(63)
24
f.
(64)
Energy
The first law of thermodynamics is
change in system
total energy
'E
system boundary
the system
G Q GW
25
2
U u U v dW
N N
2
potential
kinetic
U
u
dW
R
(
t
)
Dt
2
G Q GW
dt
dt
(65)
26
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
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
(r u v ) U dW
H0
r u ( u r) U dW .
(65)
r
R (r )r U dW .
(66)
29
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
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
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
33
fn
n
ft
'S
(n )
e 2
(e 2 )
e1
2
(e1 )
1
q2
q1
e 3
(e 3 )
3
fsurface f body
ma,
35
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
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
U'h
3
'S f
U'h
3
'S a.
U'h
3
(a f ).
Continuing to shrink R,
lim n (e11 e 2 2 e 3 3 ).
'ho0
(72)
38
39
Tensor Analysis
q3
(n )
e 2
(e 2 )
e1
2
(e1 )
1
q2
q1
e 3
(e 3 )
3
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.
U'h
3
(a f ).
Continuing to shrink R,
lim n (e11 e 2 2 e 3 3 ).
'ho0
(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)
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 .
Tensor Analysis
In general, we can write
I
I jk e j e k (contravariant scalar components)
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
Tensor Analysis
S1
S
2
S
3
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
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 )
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)
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
13
Tensor Analysis
For orthogonal curvilinear coordinates,
grad a { a
e i w
(a j e j ).
hi wq i
(98)
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
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)
ai whi
1 wak ak whi
G
i
ik
hi wq h j wq j
hk wq k
e i e k .
(100)
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
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
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
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)
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
Tensor Analysis
V { I ij ei u e j
gI ij H ijk e k
{ Iij ei u e j
{ Ii j ei u e j
{ I ij e i u e j
gI ij g jmH imk e k .
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)
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)
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
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
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
1 w
w
" h k u ( x, y )
wy
n! wx
(122)
26
Tensor Analysis
u(x,y)
x+h
y+h
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)
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
extension
(diagonal terms)
+
shear
(off-diagonal terms)
32
Tensor Analysis
Note:
trace[H ij ] H ii
div v
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
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
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
dr dr
(ds ) 2
o ds1
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
is the rate of change of
I in the direction
normal to the surface S
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
dy
dt
f (t , y )
y ty
0.
Rewrite:
dy
dt
ty.
1
2e ( t
1) / 2
STORM
WILDFIRE
POLAR
Project 1
Y
v
I
X
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
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
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.
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
x0
local minimum
f ( x)
df
dx
f max
x0
inflection
f ( x)
df
dx
0
x0
12
df
dx
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
( x0 , y0 )
wf
wf
dx
dy
wx
wy ( x , y )
0 0
dr f
( x0 , y0 )
0 and
wf
wy
14
dr G
(1)
(2)
(3)
15
2 x2 y 2 8x y 1
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
17
yb
18