7.

2 The covariant derivative of a vector/tensor field

the basis vectors such that they point along these coordinates (though not nec-
essarily are normalized the same as the coordinate units).
2. A local coordinate basis: The covector basis can be seen as unit steps in each of
the directions: ẽµ = dx µ . The contravariant basis vectors are then ~eµ = ∂/∂x µ .
Typically this coordinate basis is not orthonormal.
So far in the entire booklet we have implicitly assumed a local coordinate basis (re-
member that we used the words coordinate transformation on equal footing as basis
transformation). For tensor mathematics this basis is usually the easiest to use. But
in realistic applications (like in Chapter ??) a local orthonormal basis has more phys-
ical meaning. In the application of Chapter ?? we will first convert everything to the
coordinate basis in order to do the tensor math, and then, once the final answer is
there, we will convert back to the local orthonormal basis, which is physically more
intuitive. But since the tensor mathematics works best for the local coordinate basis,
we will assume this basis for the rest of this chapter.
Now let us take the example of circular coordinates in two dimensions. We can
express the usual x- and y-coordinates in terms of r and φ as: x = r cos φ and y =
r sin φ. In r-direction the coordinate basis vector ∂/∂r can act as a normal basis
vector. In the φ-direction, however, the coordinate basis vector ∂/∂φ has a different
length at different radii. However, 1r ∂/∂φ is again a normalized (unit) vector. The
set of basis vectors (∂/∂r, ∂/∂φ) is the local coordinate basis, while the set of basis
vectors (∂/∂r, 1r ∂/∂φ) is a local orthonormal basis, which happens to be parallel to
the coordinates.
Now let us use the coordinate basis. Let us take a vector field that, at every
location, when expressed in this local basis, takes the form (1, 0). This is shown in
Fig. XXXX. Expressed in cartesian p coordinatesp(for which a global basis is possible)
this would be a vector field ( x/ x2 + y2 , y/ x2 + y2 ). Clearly this vector field is
not constant. In cartesian coordinates one would write, for instance,
 p   2
y /( x2 + y2 )3/2

∂ x/ x2 + y2
= , (7.1)
− xy/( x2 + y2 )3/2
p
∂x y/ x2 + y2

which is clearly non-zero. However, if we switch to circular coordinates, and take

the derivative with respect to, for instance, φ, then we obtain zero,
   
∂ 1 0
= . (7.2)
∂φ 0 0

In the circular coordinate system it looks as if the vector does not change in space,
but in reality it does. This is the problem one encounters with derivatives of tensors
in curved coordinates. Going to the local orthonormal basis does not help. One has
to define a new, ‘covariant’ form of the derivative.

7.2. The covariant derivative of a vector/tensor field

In this section we will introduce the covariant derivative of a vector field. We will
do so without proof. Again, for a more detailed discussion we refer to the literature.
We define ∂µ to be the ordinary coordinate-derivative ∂/∂x µ , and ∇µ to be the new
covariant derivative. The covariant derivative of a vector field vµ is defined as

∇µ vα = ∂µ vα + Γαµν vν , (7.3)

where Γαµν is an object called the Christoffel symbol. The Christoffel symbol is not
a tensor because it contains all the information about the curvature of the coordi-
nate system and can therefore be transformed entirely to zero if the coordinates are

36
 7.2 The covariant derivative of a vector/tensor field

straightened. Nevertheless we treat it as any ordinary tensor in terms of the index

notation.
The Christoffel symbol can be computed from the metric gµν (and its companion
gαβ ) in the following way,

∂g βν ∂g βµ ∂gµν
 
1 1
Γαµν = g αβ + − ≡ gαβ ( g βν,µ + g βµ,ν − gµν,β ) . (7.4)
2 ∂x µ ∂x ν ∂x β 2

The covariant derivative of a covector can also be defined with this symbol,

∇µ wα = ∂µ wα − Γνµα wν . (7.5)

The covariant derivative of a tensor tαβ is then

β
∇µ tαβ = ∂µ tαβ + Γαµσ tσβ + Γµσ tασ , (7.6)

and of a tensor tα β ,
∇µ tα β = ∂µ tα β + Γαµσ tσ β − Γσµβ tα σ . (7.7)

◮NOW MAKE EXERCISES TO SHOW WHAT HAPPENS WITH THE ABOVE EX-
AMPLE

From the exercise we can see that the above recipe indeed gives the correct an-
swer even in curved coordinates. We can also prove that the covariant derivative of
the metric itself is always zero.

◮ ANOTHER EXERCISE

The covariant derivative ∇µ produces, as its name says, covariant expressions.

Therefore ∇α tµν γ is a perfectly valid tensor. We can also contract indices: ∇α tαν γ ,
or with help of the metric: g αγ ∇α tµν γ . Since ∇µ gαβ = 0 (as we saw in the above
exercise) we can always bring the gαβ and/or g αβ inside or outside the ∇µ operator.
We can therefore write

gαγ ∇α tµν γ = ∇α (tµν γ g αγ ) = ∇α tµνα (7.8)

We can also define:

∇α = gαβ ∇ β (7.9)

37
7.2 The covariant derivative of a vector/tensor field

38
 Tensors in special relativity
A
Although tensors are mostly useful in general relativity, they are also quite conve-
nient in special relativity. Moreover, when we express the formulas using tensors,
they expose the structure of the theory much better than the formulas without ten-
sors do. We will start here at the point when four-vectors are introduced into special
relativity.
From the simple thought experiments with moving trains, meant to give a visual
understanding of the Lorentz transformations, it has become clear that 3-dimensional
space and 1-dimensional time are not separate entities, but rather should be seen as
one. In matrix form one found for a pure Lorentz transformation
 ′
x0
   0
γ −γβ 0 0 x
 1′    x1 
 x  −γβ γ 0 0
 2′  =    , (1.1)
x  0 0 1 0  x 2 
x 3 ′ 0 0 0 1 x3
√
with γ = 1/ 1 − v2 /c2 and β = vc . Moreover, we found that the inner product now
satisfies the somewhat strange formula

h~a,~bi = − a0 b0 + a1 b1 + a2 b2 + a3 b3 . (1.2)

With the knowledge which we now have about the ‘metric tensor’, this formula can
be understood much better. Namely, we know that the inner product of two vectors
does not really exist (at least not in a form which is independent of the coordinate
system). Rather, we should be talking about the inner product of a vector with a
covector. We can obtain one from the other by raising or lowering an index, making
use of the metric tensor,
aα = gαβ a β . (1.3)
Eq. (1.2) should thus really be viewed as

h~a,~bi g = aµ bµ = gµν aµ bν . (1.4)

where b̃ is a covector. From the fact that inner products such as (1.2) are invariant
under Lorentz transformations we now find the components of the metric tensor,
 
−1 0 0 0
 0 1 0 0
gµν =  0 0 1 0 .
 (1.5)
0 0 0 1

39
APPENDIX A. TENSORS IN SPECIAL RELATIVITY

Minkowski space, which is the space in which the rules of special relativity hold, is
thus a space which has a metric which is different from the metric in an ordinary
Euclidean space. In order to do computations in this space, we do not have to in-
troduce any strange inner products. Rather, we simply apply our new knowledge
about inner products, which automatically leads us to the modified metric tensor.
Let us end with a small related remark. We have seen in Chapter (5) that the
metric can also be used to describe curved spaces. The Minkowski metric which we
have encountered in the present section can be used to describe curved space-time.
This concept is at the basis of the general theory of relativity. In that theory, gravity
is explained as a consequence of the curvature of space-time. A good introduction
to this theory is given in [5].

40
 Geometrical representation
B
The tensors as we have seen them in the main text may appear rather abstract. How-
ever, it is possible to visualise their meaning quite explicitly, in terms of geometrical
objects. A simple example of this is the arrow-vector. An arrow-vector is effectively
just a set of numbers that transforms in a particular way under a basis transforma-
tion. However, it is useful to think about it in terms of an arrow, so as to visualise
its meaning.
We can make a similar geometrical visualisation of a covector, as well as for all
other tensors. The present appendix contains a list of these geometrical representa-
tions. As we go along we will also describe how various manipulations with tensors
(such as additions and inner products) can be translated to this geometrical picture.
None of this is necessary in order to be able to compute with tensors, but it may
help in understanding such computations.

• arrow-vector
An arrow-vector is represented by (how else) an arrow in an n-dimensional
space. The components of the vector (i.e. the numbers vµ ) are obtained by
determining the projections of the arrow onto the axes. See figure B.1.

Figure B.1: Geometrical representation of a vector, and the determination of its com-
ponents by projection onto the axes.

• co-vector
A co-vectors is closely related to the gradient of a function f (~x), so it makes
sense to look for a geometrical representation which is somehow connected
to gradients. A useful related concept is that of contour lines (curves of equal
height, in two dimensions) or contour surfaces (in three dimensions). Now
consider two consecutive contour lines. The intersection they make with a
neighbouring x2 = constant lijne is equal to the increase of x 1 necessary to

41
APPENDIX B. GEOMETRICAL REPRESENTATION

increase f (~x ) by 1, if the other coordinate is kept constant. So in two dimen-

sions,

∂ f −1
 1  
∂x
intersection at x1 -axis = = (2.1)
∂ f x2 constant ∂x1 x2 constant
Therefore, if we take the inverse of this intersection, we again obtain the first
component of the gradient.
Given the above, it makes sense to use, for the generic representation of a
co-vector, two consecutive lines (or, in three dimensions, planes), such that
the intersections with the axes are equal to the components of the co-vector.
Note that one should also keep track of the order of the lines or planes. In the
representation sketched in figure B.2 we have drawn a flexible type of arrow.
This is not an arrow-vector as above, but only meant to indicate the order of
the surfaces. We could equally well have labelled the surfaces with different
numbers or colours.

asafsnede

Figure B.2: Geometrical representation of a co-vector, and the determination of its

components by computing the inverse of the intersection with the axes.

• anti-symmetric contravariant second-order tensor tµν

An anti-symmetric contravariant second order tensor can be made from two
vectors according to
tµν = vµ wν − wµ vν . (2.2)
This is a bit like an outer product. The order of the two vectors is clearly
of relevance. We can visualise this object as a surface with a given area and
orientation. The components of the tensor can be obtained by measuring the
projection of the surface onto the various coordinate surfaces. A negative com-
ponent is obtained when the orientation of the surface assocatiated to the ten-
sor is negative. Let us illustrate this with an example: if the surface of the
projection on the x 1 , x2 surface is equal to 2, then the t12 component equals 2
and the t21 component equals −2. See figures B.3 and B.4.

1
1

2
2

Figure B.3: Representation of an anti-symmetric contravariant 2nd-order tensor.

• anti-symmetric covariant second order tensor tµν

This object can again be seen as a kind of outer product of two vectors, or
rather two co-vectors,
tµν = vµ wν − wµ vν . (2.3)

42