Chapter III: Tensors
Chapter III: Tensors
Chapter III: Tensors
For every
contravariant
Definition: index there is an
A transformation
a tensor is an array of covariant and contravariant components, and for every
T kRmnp ( functions of x i ) which transforms into a "similar" covariant index
there is a B
array of components T ' k'R'm'n'p' (functions of q i ) as follows: transformation.
T k R m n p ' A k.k' A R R' A m.m' Bn n' Bp p' T ' k' R' m'n'p'
The transformation, A (and its covariant counterpart, B), can be any linear transformation on the coordinates such
as the general coordinate transformations we considered in Chapter II. The most common coordinate transformation
in three dimensional Euclidian space is a rotation from one Cartesian system to another. In relativistic problems A is
generally a Lorentz transformation from one {x,y,z,t} system to another {x',y',z',t'}. For such transformations the indices
(written as greek letters) run from 1 to 4.
Note that a tensor is defined in terms of a transformation, A. Some quantities are tensors (i.e., transform as shown
above) only when A is a rotation. Some quantities are tensors under the generalized coordinate transformation. Other
quantities are tensors under Lorentz transformations. So when stating that a quantity is a "tensor" one should add "with
respect to .......(name) transformation".
Definition: the rank (contravariant or covariant) of a tensor is equal to the number of components:
Examples
1. Strain tensor w.r.t. rotations between Cartesian systems:
jk
3. Inertia tensor, I , (w.r.t rotations between Cartesian systems):
jk
T = ½I ωj ωk; ω / angular velocity.
+
jk
I / | - mmm ρ(r) xkxj dV if j
k
|
jRm
| = ½ |ε |mmmρ(r)[(xR)²+ (xm)²] dV if j = k
.
-1
The above are all tensors under orthogonal transformations in three dimensional space, A = AT, and are called
affine tensors. Affine ==> cartesian systems.
Theorem:
Any product of covariant and contravariant vectors which are not operators is tensor under the general linear
coordinate transformation:
Fi Gk HR Rm = TikRm .
The components of T then transform as shown in the box above.
Addition of tensors:
Two tensors of identical rank can be added:
proof:
1. consider A k
i
BjR δkR = Aik Bjk
= Aik [A-1]Tjk
= [A A-1]ij
= δij
So δij transforms like a tensor under a general coordinate system.
j
Similarly, δi is a mixed tensor, with covariant rank = 1 and contravariant rank = 1.
Any tensor of rank 2 can be written as a sum of symmetric and anti-symmetric parts:
Tij =
ij ji
½ [ T + T ] +½ [ T - T ]
ij ji
Direct Products of Tensors: (generally ==> "all possible combination of component products")
Mi Wj = Tij (9 components of T)
dxidxj = Tij
EjM/Mx i
= Tji
In the last example, note the comparison with EL (an awkward notation for the direct product of two "vectors"). This
stands for 9 components and shouldn't be confused with something like E·L or ExL.
III-4
Consider the last example above once more, in the Cartesian system:
= x x^ E1L1 + ^
^ y x^ E2L1 + z^x^ E3L1 +
x^y^ E L2 + y^^
1
y E2L2 + ^z y^ E3L2 +
x^z^ E L3 + y^z^ E L3 + z^^
1 2
z E3L3
This is a complicated expression and can be much more simply written as EjLi. In this form one can carry all the
information and (for example) show that this "composite" is a tensor:
Just as any product of covariant and/or contravariant vectors not involving operators are tensors, any product of tensors
(which are not operators) are also tensors. LiE would not transform like a tensor since the first factor is an operator.
j
M ii / trace of M = 3iMii
MijWij
dxidxi
Contraction of a tensor is the operation by which one sets one covariant index (say i) equal to a contravariant
index (say j) and sums over i. The resultant tensor is said to be a contracted tensor. Note that one cannot arbitrarily
set two indices equal and sum; one must have a physical reason for doing so.
III-5
proof:
2. Thus,
Kijk Dim = BjR Akp Bms [K tR
p
Dts]' by hypotheses (a) and (b).
3. [Kijk - Bi
t
BjR Akp K'tRp] Dim = 0 for all A
Dim is arbitrary in the sense that the primed system can be any coordinate frame. Thus we
could have started this derivation in any (primed) frame and transformed to any (double primed)
frame. In this case we could find a frame in which D'im
0. Thus [...] = 0 and
Scalars (zero rank tensors) have the same value in all systems.
Example: the trace D / tr D / Dii is independent of the coordinate system.
proof:
j
1. We assume Di is a mixed tensor: DRm = BRi Amj D'ij
2. trD = DRR
= BRi ARj D'ij
= [A-1]TRi ARj D'ij
= [A-1A]ij D'ij
= δij D'ij = D'ii
......
Numerical tensor / a tensor with the same (numerical) components in all systems. δij is an example.
proof:
i
2. This holds in all systems related to the Cartesian system by dx = Aijdqj. Furthermore, ds² is a zero rank tensor
j
( = dr·dr = dxidxi and has the same value in all systems) and dq is a first rank contravariant tensor. Thus the
hypotheses of the quotient rule theorem are satisfied and gij transforms like a second rank covariant tensor.
ij R
Theorem: If a tensor, T k , has all zero components
in one coordinate system, the tensor has all zero
components in all coordinate systems.
proof:
1. Transform to the xi system from the (primed) system in which all components are zero.
2. Now transform to any system (denoted by double primes) from the xi system:
proof:
j R
1. MijW RC mn - Dimn is a tensor since M, W, C and D are tensors.
j R
2. MijW RC mn - Dimn = Timn = 0 in one coordinate system (given).
j R
3. Thus T'imn = 0 in all systems and M'ijW' RC' mn - D'imn = T'imn = 0 holds in all systems.
III-8
Rmn
x i·(FxG) =
^ εijkF G j k
i
= [/g]-1 A Rε F'mG'n
= (FxG) i
= [/g] AiR u ·(FxG)
-1 R
-1 i
= |A| A R (FxG)' R
i
It is said that (FxG) transform like a pseudo tensor (almost like a tensor!) with weight, W = -1. The inverse
transformation is:
-1 R i
= (FxG)'R = [/g]+1 [A ]ix
^ ·(FxG)
-1 R
= |A| [A ] i (FxG)
i
Example:
The differential volume element, dV is a pseudo tensor ( a pseudo scalar) with weight = +1:
1
= /g εRmn dqRdqm un · dqkuk δR δm2 δk3
= /g ε123 dq1dq2dq3
Note the use of the volume of a parallelopiped written as a vector product. In the above expression the /g appears
(rather than [/g]-1) since use is made of the contravariant components of dr in the qi system.
Example:
ijk
We circumvented the above awkward treatment of the transformation properties of ε by defining a new quantity,
Rmn
ε' //g, which does indeed transform like a 3rd rank contravariant tensor:
Often if a quantity transforms like a pseudo tensor, one can define a new quantity (depending on /g) which transforms
like a true tensor.
dxi = AiR dq R
= A'in dq'
n
2. To obtain the transformation between the general systems directly, multiply the last two terms by [A-1] j
i and
sum over i:
δjR dq = R
[A-1A']jn dq' n
dq j
= [A-1A']jn dq' n
dq j
= Mjn dq' n