Chapter 3

DIGRESSION INTO TENSOR THEORY

Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com
by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

1. Invariant Quantities
We have been considering the transformation of some quantities
due to some actual physical processes. We found that some quanti­
ties constructed from them were invariant under the mathematical
transformations representing these physical processes. For a more
powerful formulation of Special Relativity it is necessary to obtain
a more formal understanding of these invariant quantities. A scalar
quantity is invariant under coordinate transformations. For example,
the magnitude of a 2-dimensional vector, a, is a scalar. It remains
unchanged by translating the origin, rotating the axes through
any angle or converting to plane polar coordinates. In fact the mag­
nitude of any vector, in any space, is an invariant quantity, being a
scalar. Of course, the position vector of a point does not satisfy this
requirement as it is defined not only by the point but also by the
origin.
A vector is also an invariant quantity. This statement may
seem strange after discussing Lorentz transformations, which seem
to give the transformation of a position 4-vector, without changing
the origin. The point is that the quantity being transformed is not
the vector but its components. In usual 3-dimensional space any
vector field, i.e., a position dependent vector A(x), can be written
as
A(x) = A1(x)e1 + A 2 (x)e 2 + A 3 (x)e 3 , (3.1)
39
 40 Relativity: An Introduction to the Special Theory
Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com
by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

Fig. 13. By transforming coordinates A = PQ is not altered. However,

it has components (Ax,Ay) in XOY and {A'x,A'y) in X'O'Y'. Since the

coordinates of P and Q transform by the same matrix so does PQ = A ,

in its components.

where e x , e 2 , e 3 are the three basis vectors in the given coordinate

system and A1 (x), A 2 (x), A3 (x) are the corresponding components
of the vector. The entire set of basis vectors may be more compactly
represented by e<, bearing in mind that t can take the values 1, 2,
3. Similarly the components can be represented more compactly by
A'(x). Then Eq. (3.1) may be rewritten in the form
3
A(x) = £ A ( x K . (3.2)
«=i
Einstein introduced a further simplification of notation by
taking the summation convention, that repeated upper and lower
indices are summed over. We will generally follow this conven­
tion, explicitly stating any deviation from it. Using this convention,
Eq. (3.2) becomes
A(x) = A i (x)e < . (3.3)
 Digression into Tensor Theory 41
Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com
by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

Fig. 14. The vector A = PQ, has Cartesian components in the XOY-
frame Ax = MN, Ay = KL and in the * ' 0 ' Y ' - f r a m e , A'x = M'N', A'y -
K'L'. It can be shown that M'JV'= MAT cos 0+ KL sin 0, K1 L' = KLCOB0-
MNain 0.

As seen from Fig. 13, the vector A is invariant under transformations

of the coordinate system, but its components, A*, do vary. Clearly,
if there is a rotation in the XY-plane, A1 and A2 are transformed by
the usual rotation matrix (see Fig. 14). Thus, for a rotation through
an angle 0, the transformation is given by

a A'H =
l'3
COS 0
— sin 0
0
sin 0
cos 0
0
°\
°1
This matrix also gives the transformation of coordinates.
Al
A*
A*
(3.4)

How can we explain this transformation in terms of Eq. (3.3)?

The point is that e< are the basis vectors in one coordinate system
and will change to e^ in the transformed coordinate system. Also,
A* will change to A * in such a way that

A'(*)e1.(x) = A V K ( z ' ) (3.5)

 42 Relativity: An Introduction to the Special Theory

Now, for general coordinate transformations the generalization of

Eq. (3.4) is
Aa{x') = ~;Ai{x) . (3.6)

A vector satisfying the transformation law, given by Eq. (3.6), for its
components, is called a contravariant vector. Inserting this equation
Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com

into Eq. (3.5), we see that

by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

A'(x) (*(*)-§■£■:
A<(x)(e i(x)-^e>a{x>))=0.(*-))=(>. (3.7)

Since Eq. (3.7) holds for all A*(i), the expression inside the brackets
must be zero. Hence

ei(x)=^e'a(x') . (3.8)

Now notice that in the Einstein summation convention

dxi dxa td^_ _d_\ ,a _ dx^_ _
dx'> dx* - \dx'» dx*)X ~ dx'» 6
' [i
*}
the Kronecker delta defined by 6g — 1 if a = b, 8% = 0 if a ^ b. Thus
the Kronecker delta is the identity matrix and merely substitutes a
new index for a previous one, i.e.,

6?Ab = Aa, S^Ba = Bb . (3.10)

Multiplying Eq. (3.8) by dxi/dx'b and using Eqs. (3.9) and (3.10)
we see that
«(*) = f£*(«).
dxi
(3.11)
A covariant vector, B(x), is defined by

B(x) = Bi(x)ei{x) . (3.12)

From the foregoing discussion it is clear that

B
'«V) = J^Bd*) (3-13)
 Digression into Tensor Theory 43

and
eV) = |W(*) . (3.14)
The scalar or dot product of two vectors can only be defined
if one is covariant and the other is contravariant so that the two
transformations cancel out and leave the product invariant. Using
Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com

Eqs. (3.6), (3.9) and (3.13), we see that

by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

^WM-^^AWW
= SjAi{x)B}{x) = Ai{x)Bi(x) .
(3.15)

To be able to define a scalar product of two contravariant or two

covariant vectors we need to convert one type of vector into the
other.

2. Tensors
Generally, invariant quantities are called tensors. Their invari-
ance is assured in much the same way as the invariance of vectors,
namely the variation of one part is cancelled by the variation
of another part. Thus, for example, a scalar obtained by taking
the scalar product of two vectors is a tensor. Of course, so are the
covariant and the contravariant vectors. Another example would be
the tensor
A = A"(x)a,(*)■,(*) , (3.16)
provided its components satisfy the transformation rule

1 dxa dx'b {
A (3.17)
•V) = dxi dx' A '(*) ,
so as to cancel the variation of e\ and e, with coordinate transfor­
mations. A tensor whose components satisfy Eq. (3.17) is called a
contravariant tensor. Similarly there can be a covariant tensor

B = fl0(<c)e,(*)e»'(«) , (3.18)
 44 Relativity: An Introduction to the Special Theory

whose components satisfy the covariant transformation law

dxi dx3'
(3.19)
* . ( * > - & & * * ) •
Also, there can be a mixed tensor with one part contravariant and
the other covariant,
Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com
by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

C = C*(z)e,(x)e'(z) , (3.20)

whose components satisfy a mixed transformation law

, 8xa dx3
W-^giW. CM)
Notice that the terms in the above expressions can be put in any
order since the index label contains the information of how matrix
multiplication is to proceed. However, the order of the indices is
fixed. In general, A%1 ^ A3% for example.
It is clear that tensors can be constructed using any number of
basis vectors, e, together. The number of e's involved in a tensor is
called its rank. The nature of the tensor depends on how many of
the e's have lower indices (like e<) and how many upper indices (like
e'). If the number of lower index e's is £ and the number of upper
index e's is k, the tensor is said to be of valence I . ). Clearly the
rank of the tensor will be (k +1). If t = 0 the tensor is contravariant,
if k = 0 it is covariant and if neither is zero, it is mixed. Thus, a
scalar is a tensor of rank zero, a vector is a tensor of rank one, and
tensors of rank more than one are the new quantities defined.
Of particular interest is the metric tensor, g, which is a covari­
ant tensor that associates to every contravariant vector a unique
covariant vector, thus enabling us to define the scalar product of
two contravariant vectors. The length of a vector is the square root
of the dot product of the vector with itself. Thus, the metric tensor
enables us to define the length of a contravariant vector - hence its
name. Generally, we have

A k = flfcy A1' = gjk A3' , (3.22)

 Digression into Tensor Theory 45

where all the components may generally be position dependent. To

be able to associate back the contravariant vector with the covariant
vector, we require that gtj be invertible, i.e., g must be invertible.
Thus, there must exist g~ 1 such that

g_1g = g-g-1=I , (3-23)

Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com
by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

where I is the identity tensor, with components £*., i.e., it is the

identity matrix in any coordinate system. It is clear that g~ 1 must
be a contravariant tensor. Its components are generally written as
gi:', so that Eq. (3.23) can be rewritten as

9ik g]k = S) . (3.24)

Thus, multiplying Eq. (3.22), by gik gives, using Eq. (3.23),

gik Ak = gik gik A* = 6) A* = A' , (3.25)

giving us back the original contravariant vector.

Now, given a vector using Cartesian coordinates,

A = A*i + Ay$ = Aiei , (3.26)

we know that the square of its length is given by

A2 = (A*) 2 + (A*) 2 = AiAi= gi] A'A'

2
= 9ll (A') + 2g12 A*A" + g2i (A") 2 . (3.27)

By direct comparison we see that in this case y n = 1 = g22, ffi2 — 0.

Generally, with Cartesian coordinates gi}- = 1 if i = j and gti = 0 if
i 7^ j . Thus, it behaves much like the identity matrix, except that it
is entirely covariant. If we represent A* as a column vector and, for
consistency A,- as a row vector, then gtJ is a partitioned row matrix.
For example, in two dimensions,

9i] = (1 0 : 0 1) , (3.28)
 46 Relativity: An Introduction to the Special Theory

so that
A, = (l 0 i 0 ! ) ( £ ) = (A" A") . (3.29)

In other words gi}- simply transposes the vector. It is clear from

here why there is no difference between covariant and contravariant
vectors in Cartesian coordinates. However, this is not the case when
Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com

other coordinate systems are used.

by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

3. C o o r d i n a t e Transformations
So far we have dealt with coordinate transformations in a very
abstract way. To make the procedure more concrete we shall consider
some specific examples. To start with we consider the transformation
from Cartesian to plane polar coordinates. In this case

x 1 = x, x2 = y; x'1 = r, x'2 = 9 , (3.30)

where the two sets of coordinates are related by

x = r cos 9, y = r sin 9 , \
(3.31)
r = Vx2 + y2 , 9 = tan" 1 (y/x) . f

Now the coordinate transformation matrix is given by

dx^_(dr/dxfdr/dx dr/dy
d r/3y\
\d9/dx d9/dy) ' V-6Z)

and its inverse transformation is given by

dxi
\_fdx/dr dx/dO\
\dz'°) \dy/dr dy/d9) ' ^66'
Now, it is easily seen from Eq. (3.31) that
dr x dr y .
— = - = cos 9, —- = - = sin 9 ,
ox r ay r
89 y/x2 sin 9 39 1/x _ cos 9
dx 1 + tan 2 9 r ' dy 2
1 + tan 9 ~ r
(3.34)
 Digression into Tensor Theory 47

Also, we have

- = costf, -^ = -rmn» ,1
(3.35)
3y . 3y {
—- = sin 5, —- = r cos 5 .
dr ' d$ >
Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com

Thus, the coordinate transformation matrix and its inverse are given
by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

by
fdx'a\ _ / cos 9 sin 9 \
V az € / V - sin 0/r cos 0/r / '
(3.36)
/ 3x i \ __(cos 0 -rsin0\
V a i ' " / ~~ \ s i n 5 r cos 9 ) '
It is easy to check, from Eq. (3.36), that as required
i
/ dx
3x' \ ' a ' a\-i
/ 3 ztdx
(3.37)
Vd*W ~ \~dxT')
Consider the vector A with contravariant Cartesian components
A and Av, so that its covariant components Ax — Ax and A„ = Ay.
x

Now using Eqs. (3.6) and (3.36), we have

Ar = A1 = A1 cos 5 + A2 sin 0
= Ax cos 0 + A" sin 9 ,
(3.38)
Ae = A2 = A1 (sin 9/r) + A2(cos 9/r) [
= (A" cos 9 - A* sin 9)/r . J

This transformation is shown diagrammatically in Fig. 15.

Now let us see the transformation of the covariant components
using Eqs. (3.13) and (3.36) to give

Ar = A[ - Ai cos 9 + A2 sin 9 = Ax cos 9 + Ay sin 9 ,

A<, = A'2 = Ax (—r sin 9) + A2r cos 5 = r(Av cos 0 - Ax sin
(3.39)
so that A r = A r and Ae = r2Ae ^ A9! Here we see a difference
between contravariant and covariant components.
 48 Relativity: An Introduction to the Special Theory
Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com
by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

Fig. 15. The polar and Cartesian components of the vector A.-Ax = MN,
Ay = KL, Ar = PR. Since the polar coordinates of P are (r, 0), we see
that PS = rAg. Also Ax — Ar cos 0, Ay = Ar sin 0.

Let us also consider the transformation of the metric tensor from

Cartesian to plane polar coordinates. Remember t h a t j n = 1 = </22>
9i2 — 0 = 02i• Since g is a covariant tensor, Eq. (3.19) is applicable
to it. Hence we have
1
, , ,. / Cdxx 1 \\ *2 , 3 i 2 \\ a2 „
{Ox"

= cos 2 0 + sin 2 0 = 1 ,

«i.M= (£)(£)■!+(£) (£)■! (3.40)

= —r sin 0 cos 0 + r sin 0 cos 0 = 0 ,
/ / M /5a;1 \2 , fdx2 \2

2 • 2
+ r 2 cos 2 0 = r 2
= r sin
Thus, the metric tensor in plane polar coordinates is

•UM>-(; r °) ■ (3.41)

Notice t h a t this metric tensor correctly converts the contravariant

components to covariant components. Correctly speaking, of course,
 Digression into Tensor Theory 49

the metric tensor should be written as (1 0 : 0 r 2 ), but it

becomes quite inconvenient to use that notation for higher dimen­
sions, or to write the inverse matrix, which here becomes

<r6M)=(i i/W •
(3.42)
Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com
by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

To check that the scalar product of the vector A with

itself remains invariant, i.e., the length is the same whether it is
expressed in Cartesian or polar coordinates, we use Eqs. (3.38) and
(3.41), or equivalently Eqs. (3.38) and (3.39), to give

(Af = g>ab(x>)A«V)A»(x')
= (A' l ) a + r ' ( A ' a ) a
= (A* cos 9 + Ay sin 9f + r2(A" cos 9 - Ax sin 9)2 /r 2
= (A')2 + (A")2 = gi3ix)Ai(x)A'(x) . (3.43)

This fact is shown diagramatically in Fig. 16 for two dimensions. It

holds generally for all dimensions, of course.

Fig. 16. The polar components are Ar = PR and Ag = LPOQ. Now, PR

= PT+ TR= PScos 9+ UQ = Ax cos 9+ QSain 9 = Ax cos 9+ Ay sin 9.
Ay cos 9- Ax sin 9 = US - TS = UT.UAr<g. r, UT/r w A„ radians.
 50 Relativity: An Introduction to the Special Theory

Now let us consider the transformation of the metric tensor for

a flat three dimensional Euclidean space into spherical polar coordi­
nates. Here we have

x 1 = x = r sin 9 cos <f> = x1 sin x 2 cos x 3 ,

x2 = y = r sin 9 sin <j> = x x
sin x 2 sin x 3 , (3-44)
Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com

3 'l '2
x = z = r cos 9 = x cos x
by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

i = e = r cos 0 = x cos x
Vlso, we have in Cartesian coordinates
Also, we have in Cartesian coordinates
ffu = 922 =933 = 1, 9a = 0 otherwise. (3.45)

Thus we obtain

- i fdx/dr dx/d9 dx/d<f>\

2PJ = ay/ar 3y/30 dy/cty
\3z/3r 3z/30 dz/d<f>l

(
sin 0 cos <f> r cos 0 cos <j> — r sin 0 sin <f>
sin 0..sin ^
cos 0
cos 9
r cos 0 sin ^
—r sin 6
-r sin 0
r sin 0 cos ^
0
0
3.46)
7(3.46)
which yields

9'M)
A1 0 0

\0
O

Or2
r 2

0 ( 0
0
\

r2sm26j
0
,

| ,
(HAT)

(3.47)
lince we have
0 0 r 2 sin2 9
since we have
g[ 1 = sin2 9 cos2 <f> + sin2 9 sin2 ^ + cos2 9 = 1,
g'12 = r(sin 9 cos 0 cos2 ^ + sin 9 cos 0 sin2 <j> — sin 0 cos 0) = 0,
013 — r s m 2 ^ ( - s m 4 c o s ^ + sin ^ cos ^) = 0,
022 = ' ,2 (cos 2 0 cos2 $ + cos2 0 sin2 # + sin2 9) - r2 ,
023 = r 2 ( - sin 0 cos 0 sin ^ cos <j> + sin 0 cos 0 sin # cos <f>) = 0,
2
ff33 = r (sin 2 0 sin2 ^ + sin2 9 cos2 ^) = r 2 sin2 0.
(3.48)
and g'ab is symmetric.
 Digression into Tensor Theory 51

Exercise 3

1. Given a 3-dimensional vector with Cartesian components Ax,

Ay and Az, work out the components in a frame obtained by a
rotation through 9X in the XY-plane, then a rotation through 92
Relativity: An Introduction to the Special Theory Downloaded from www.worldscientific.com

in the YZ-plane and then a rotation through 03 in the ZX-plane.

by NANYANG TECHNOLOGICAL UNIVERSITY on 11/05/14. For personal use only.

Is the result the same if the order of rotation is reversed?

2. Prove that the Kronecker delta is unaltered by coordinate trans­

formations.

3. Prove that the position vector, in plane polar coordinates has no

polar component. Explain this fact diagramatically.

4. Given the coordinate transformation

p = xy, q = x/y ,

determine the metric tensor in the (p, q) coordinates if x and y

are the usual Cartesian coordinates.

5. Work out the metric of a four-dimensional Euclidean space in

hyperspherical coordinates, i.e., having one radial coordinate and
three angular coordinates.

6. By restricting the three-dimensional metric tensor in spherical

polar coordinates to the surface of a sphere, r = a, determine the
metric tensor for the surface of a sphere. (Remember that the
resultant space is two-dimensional and not three-dimensional.)