Tensors, and differential forms - Lecture 2

1 Introduction
The concept of a tensor is derived from considering the properties of a function under a
transformation of the coordinate system. A description of a physical process cannot depend
on the coordinate orientation or origin, and this principle can be expanded to look for other
mathematical symmetries. To begin it is best to start the description of tensors by reviewing
the transformation of a vector function under a rotation of coordinates. Consider the vector
function;

F~ (x, y, z) = fx (x, y, z)x̂ + fy (x, y, z)ŷ + fz (x, y, z)ẑ

The above is really 3 functions associated with each of the 3 spatial coordinate directions.
Under a rotation through an angle θ, about the ẑ axis the functions transform from, fi (x, y, x)
to fi′ (x′ , y ′, z ′ ). The transformation is written;

fi′ =
P
aij fj
j
 
cos(θ) sin(θ) 0
aij →  −sin(θ) cos(θ) 0 
0 0 1
Generally one can rotate about any axis, and the transformation for other axes is an obvious
permutation. As an aside here, note that a transformation about an arbitrary axis can be
obtained by a rotation about the three Euler angles in succession α, β, γ as illustrated in
Figure 1. Thus, there is a transformation aij which takes the function, F~ , defined in the
coordinate frame x, y, z into the function F~ ′ in the coordinate frame x′ , y ′, z ′ .

z z’
y’ y"
f’z f’z"
z"
f’z’
y

fx’ f x"
x’
x

Figure 1: An example to obtain the rotation of a coordinate frame about an arbitrary axis
defined by using the Euler angles

1
   
cos(γ) sin(γ) 0 cos(β) 0 sin(β)
aij →  −sin(γ) cos(γ) 0   0 1 0 
0 0 1 −sin(β) 0 cos(β)
 
cos(α) sin(α) 0
 −sin(α) cos(α) 0 
0 0 1
The rotation matricies are unitary. That is they preserve the length of the rotated vector.
Multiply the matrix by its transpose to show this. Note that tensor operations reproduce ma-
trix algebra as the product below indicates multiplication of a row vector by a column vector.

F~ · F~ † =
PP P † †
aij fj aki fk
i j k

F~ · F~ † = fj fk† aij a†ki

P P
jk i

aij a†ki
P
= δjk
i
The matrix obtained in the above when summed over the direct product of the aij , is a
diagonal matrix with 1’s along the diagonal. This matrix has the determinant of +1. Now
consider an operation which preserves length but reflects the vector about the origin. That
is, the transformation which changes a right handed coordinate coordinate system into a left
handed one.

x → −x y → −y z → −z

This forms a matrix which is diagonal with -1’s along the diagonal. It has a determinant
of -1, and compared to the above matrix, represents the symmetry operation of parity. A
vector changes sign upon reflection, ie a vector in a right handed frame is a negative vec-
tor in a left handed frame. If the vector does not change sign under reflection then it is a
pseudo-vector or axial-vector. The cross product of two real vectors is a pseudo-vector as
is obvious since each vector changes sign upon reflection and their product does not. The
operations of rotation and reflection can be written as components of matricies, and as such,
represent linear transformations of the vector components.

Remember that a scalar function remains constant under rotations and also reflections. Since
a scalar is represented by one function, the transformation is a 1-D matrix or just a number,
in this case 1. However, if the scalar function changes sign under reflection, it is a pseudo-
scalar. Using the concepts of vector and scalar, linear transformations can be expanded to
higher dimensions.

The fundamental characteristic of a tensor is the linear connection between the

differential forms of a function in different coordinate frames.

2
Thus a tensor of rank 0 is a scalar (or pseudo-scalar) and a tensor of rank 1 is a vector (or
pseudo-vector). A tensor of higher rank, Fk,l,···, transforms the function, Fk,l,··· as;

′
P
Fi,j,··· = ak,l,··· Fk,l,···
k,l,···

2 Summation Convention
Suppose n independent variables, xi with i = 1, 2, · · · n. The set of values xi define a point
in a n-dimensional space. Note the superscript index. This does not mean the variable is to
be raised to the ith power. This notation will become obvious later. There are n independent
functions, φi (x1 , x2 , · · · , n) in this space. For these functions to be linearly independent,
the Jacobian cannot vanish and any other function in the space can be described by a linear
combination of functions which “span” the space. Therefore;

∂φ1 ∂φn
···
∂x1 ∂x1
J = ··· 6= 0
1 n
∂φ ∂φ
···
∂xn ∂xn
Let x′ i = φ′ i define another coordinate system, so in the same way which previously devel-
oped the scale factors;

∂xk = P ∂xk ∂x′ i = δ k

j
∂xj ′i
i ∂x ∂x
j

A direction at a point is;

P ∂x′ i j
dx′ i = dx
j
∂xj
Now introduce the summation convention which assumes that a repeated index (a dummy
variable) is summed. The summation sign will be dropped and summation indicated by a
repeated index, so the partial derivative equations above are written as;

∂xk ∂x′ i = δ k
∂x′ i ∂xj j

′i
dx′ i = ∂x dxj
∂xj

3
3 Contravariant and Covariant vectors
Now use both a superscript and a subscript on the variable to represent a tensor component.
This is to denote different types of tensors. Note that a distance transforms as (summation
convention);
′i
dx′ i = ∂x j dxj
∂x
Thus this form represents a true vector, with components, say Ai , and it has the linear
transform;
′i
A′ i = ∂x j Aj
∂x
To clarify, the above transformation describes how a component of a contravariant vector
transforms.

Next consider the gradient (vector) operator;

~ = ∂φ x̂i
∇φ
∂xi
This form transforms as ;

∂φ′ ∂φ ∂xj
′i =
∂x ∂xj ∂x′ i
j ′i
Note that the gradient transforms as ∂x′ i rather than ∂x j . This transforms a covariant
∂x ∂x
vector component which will be noted by a subscript.
j
A′i = ∂x′ i Aj
∂x
In a transformation between Cartesian coordinate frames, there is no difference between
contravariant and covariant forms.

∂xi = ∂x′ i = ai = aj
j
∂x′ j ∂xj i

In a general curvilinear coordinate frame, this is not the case. Therefore unless stated differ-
ently, a superscript is used to denote a contravariant component of a tensor and a subscript
to denote a covariant component. A contravariant vector determines a direction and mag-
nitude of a displacement at some point in space. It forms a vector field. A covariant vector
describes the change in a particular direction of the field at a point in space.

Let λi be any n functions of the coordinates, xj . The contravariant transformation is;

4
 ′i
λ′ i = ∂x j λj
∂x
k ′i
Then write, since ∂x′ i ∂x j = δjk ;
∂x ∂x
∂x λ′ i = λj ∂xk ∂x′ i = λk
k

∂x′ i ∂x′ i ∂xj

Thus we have found the inverse transformation for components of a covariant vector.
Rewrite the above relations for vector components λj and µk and compose the form (sum-
mation convention);
′i k
λ′ j µ′j = λj ∂x j µk ∂x′ i = λj µk δjk = λj µj
∂x ∂x
This is an invariant of the transformation. The result is a scalar which is invariant under
coordinate transformations. In the old way of combining vectors this is just the dot or scalar
product. Note that this is a summation over the product of contravariant and covariant
components of two vectors.

4 Tensors
To begin, define two contravariant vectors, λi and η i . Study the transformations from un-
primed to a primed coordinate frame. Also define two covariant vectors, µi, and ζi . The
following forms are to be used;

Aij = λi η j

Aij = µi ζj

Aij = λi µj

Then define combinations of the primed forms in a similar way. The transformation ma-
tricies between these are;
′i ′j
A′ ij = Akl ∂x k ∂x l
∂x ∂x
k l
A′ij = Akl ∂x′ i ∂x′ j
∂x ∂x
′i l
A′ji = Akl ∂x k ∂x′ j
∂x ∂x

5
These forms are tensors of 2nd order. Akl is a contravariant tensor, Akl is a covariant tensor,
and Akl is a mixed tensor. Note that there are n2 elements in each tensor. The Kronecker
delta, δjk , is a mixed tensor of 2nd order.
′k j
δlk = ∂x j ∂x′ l
∂x ∂x
Tensors of any order may be constructed in a similar way. The above construction uses the
direct product of a tensor of lower order to produce one of higher order. Remember tensors
are defined by their linear transformation properties. Thus;
′i ′k
A′ i,j,··· ,k = Am,n,··· ,p ( ∂xm ) · · · ( ∂x p )
∂x ∂x
The components of covariant and mixed tensors transform in a similar way. However, note
that the order of the indicies (not the order of the terms in parenthesis) is important. Con-
sider an arbitrary contravariant tensor Aij . Write this as;

Aij = (1/2)[Aij + Aji] + (1/2)[Aij − Aji ]

Addition/subtraction of tensors can only be done for tensors of the same order. The first
term on the right side in the above equation forms a symmetric tensor and the second an
anti-symmetric tensor. Then an arbitrary tensor is a combination of symmetric and anti-
symmetric tensors.

The concept of higher dimensional tensors is developed by directly multiplying two tensors
of lower order to obtain a tensor of higher order. This is the outer (direct) product of the
tensors. The inner (dot or scalar) product of two tensors forms a tensor of lower order. Thus
consider the inner product of the tensors Aij and B jkl . The inner product sums over the
repeated index, in this case, j, to get a tensor of rank 3.

A′ij B ′ jkl = Γkl

i

The direct product is

Γrkl
uv = Auv B
rkl

Note that;
j ′w
( ∂x′ w )( ∂x t ) = δtj
∂x ∂x
This results in ;

6
 r ′k ′l
A′ij B ′ jkl = Γruv ( ∂x′ i )( ∂x′ u )( ∂x′ v )
∂x ∂x ∂x
The above transforms as a mixed tensor of rank 3. The process of reducing the order of a
tensor by the inner product is called contraction.

5 The metric
The metric is used in determining the differential length in a coordinate frame.

ds2 = gij dη i dη j
P ∂xi ∂xi
gjk =
i
∂ηj ∂ηk
Perhaps a more insightful way to defind the metric is to consider a length element, ds, along
the ith coordinate, xi , which has a basis unit vector direction, êi .

dsi = êi dxi

Here dsi is a length in the ith direction. Now suppose a recriprocal unit vector basis, êi such
that êi êj = δij . Then contract the length vectors and assume the summation convention.

d~s · d~s = ds2 = dxi dxj êi êj = dxi dxj g ij

In the above, g ij is the contravarient metric matrix. It is symmetric and depends on the
coordinate position. Note that ds2 is a scalar length produced by a contraction of the two
tensors. Now the recriprocal set of unit basis vectors can be expanded in terms of the unit
basis vectors of the space.

êi = cij êj (Summation convention)

Then êj êi = cij = gij , and ei = gij êj and ei = g ij ej . The metric is used to raise or
lower an index in a tensor. Also, the metric is restricted so that |gij | =
6 0, but otherwise
2 2
ds can be < 0 so we must take |ds | to be the measure of length. The metric is still defined as;

ds2 = gij dxi dxj

An arc of length between a and b is;

r
Rb i j
s = dt α gij ∂x ∂x
a
∂t ∂t

7
The factor, α, is taken as ±1 so that each element in the sum is > 0, and t is the parametric
variable.

6 Conjugate Tensors
The metric gij is a symmetric covariant tensor of order 2. Thus gij = gji . Further we have
that;

g11 · · · g1n
g = ···
gn1 · · · gnn
Then let g ij be the cofactor of the element gij divided by g. The cofactor of a tensor element,
Aij , is given by;

Cof actor(Aij ) = (−1)i+j Mij

where Mij is the minor of the element Aij . The minor is obtained from the determinant of
the tensor after deleting the row and column containing the element. This means that the
determinant of the tensor A is;
n
Aij (−1)i+j Mij
P
|A| =
k=1

This then gives the relation;

g ij gkj = δki

The g ij are elements of a contravariant tensor of order 2. This tensor is the conjugate of
gij . Using this symmetric tensor, one may obtain a tensor of the same order but of different
character (raise or lower the index - ie change from covariant to contravariant or contravari-
ant to covariant).

Aljk = g liAijk

Almp = g lig mj g pk Aijk

Note that the process is reversible. Also;

g ij gkj = (1/g) (−1)i+j Mij gkj

P P
j

Unless i = k the product of terms is obtained by using one term of one row and cofactors of

8
terms in another row. In the case of i = k the determinant results. As a result;

ai = gij aj

aj = g ij ai

7 Levi-Civita symbol
Define the following tensor of rank 3.

ǫ123 = ǫ231 = ǫ321 = 1

ǫ132 = ǫ213 = ǫ312 = −1

All other ǫ = 0

This is a 3-D Levi-Civita tensor and is obviously a pseudo-tensor as odd permutations of

the indicies are −1 times the even permutations. We can also define the conjugate tensor
ǫijk . Suppose a set of vectors λi η i ζ i with i = 1, 2, 3. Contract a tensor with the Levi-Civita
tensor to produce the pseudo-scalar, φ.

φ = ǫijk λi η j ζ k

φ = λ1 η 2 ζ 3 − λ1 η 3 ζ 2 + λ2 η 3 ζ 1 − λ2 η 1 ζ 3 +
λ3 η 1 ζ 2 − λ3 η 2 ζ 1

The Levi-Civita tensor may be expanded to rank 4, ǫijkl , by a similar definition. If we allow
the vectors λi η i ζ i to be differential vectors lying along each of the Cartesian coordinate
axes, ie λ → (dx, 0, 0), η → (0, dy, 0), ζ → (0, 0, dz) The value of φ = dx dy dz = dτ . This
is a differential volume element, which we identify as a pseudo-scalar. A pseudotensor of
odd rank does not change sign under parity inversion while one of even rank does.

8 Dual tensors
Any anti-symmetric tensor of rank greater than 2 can have a dual representation. This is
understood by contracting the tensor with a Levi-Civita tensor. Thus suppose a tensor of
rank 3. To be specific, assume the angular momentum obtained by;

9
  
0 x1 p2 − x2 p1 −x1 p3 + p3 x1
Lij → (1/2)  −x1 p2 + x2 p1 0 x2 p3 − p2 x3 
x1 p3 − x3 p1 −x2 p3 + p3 x2 0
 
0 C12 −C31
 −C21 0 C23 
C31 −C32 0
where to be anti-symmetric Cij = −Cji . Then the contraction with the 3-D Levi-Civita
tensor yields a vector.

c′i = (1/2)ǫijk Cjk

This produces, for the angular momentum example;

~ = [ypz − zpy ]x̂ − [xpz − zpx ]ŷ + [ypx − xpy ]ẑ

L

Note C~ ′ transforms as a vector as previously proved by tensor contraction, but this vector is
a pseudo-vector because of its symmetry properties. Note that the cross product is a special
anti-symmetric tensor with a dual which has the rotation properties of a vector, but the
symmetry under inversion is a tensor of rank 2. Tensor properties may also be developed
from the concept of the increasing order of surfaces. The properties of a point are equivalent
to a scalar, the properties of a line are vectors, the properties of a surface are pseudo-vectors,
etc.

9 Covariant derivative
The covariant derivative assures that a vector is independent of its description in an arbi-
trary coordinate frame. That is, the vector has a magnitude and points in a specific direction
independent of the frame of reference. The covariant derivative removes the change in the
vector due the curvature of the coordinate frame. Therefore the components of a covariant
~ with respect to the η i axis are;
vector representing the rate of change of an ordinary vector, A,
j
Aj, i = ∂A +
P k j
A Γki
∂ηi k

The symbol Γjki is a Christoffel symbol, and represents the change in the direction of the unit
vectors. It will be discussed in more detail, but for the moment represents the projection of
the change in the unit vectors of the space as a vector is displaced a distance, dq j .

dêi = Γkij êk dq j

The comma indicates that Aj,i is the covariant derivative. The Ai, j are the components
of a mixed tensor, covariant with respect to the index j and contravariant with respect to

10
the index i. Covariant differentiation can also be extended to tensors of higher order.

The contracted tensor,

P ~
An,n , represents the Div of the vector A;
n
P ∂An
An,n
P P n m
= + A Γnm =
n n
∂ηn n,m
P 1 ∂ (An h h h )
1 2 3
n
h1 h2 h3 ∂ηn

~
Identify An = F n /hn . Substitution in the above yields Div A.

1 P ∂ h1 h2 h3 An
~ =
Div A [ ]
h1 h2 h3 n ∂ηn hn
~ can also be developed for orthogonal coordinates by consideration of
The curl of vector B,
the component Ai = −( 1 )[Bj ,k − Bk ,j ]. The scalar Laplacian can also be written as
h1 h2 h3
P 1 ∂φ
a covariant derivative, ( 2 ) Thus the use of the covariant derivative allows one to
n hn
∂ηn , n
express an equation in the same form for any coordinate system.

10 Geodesic lines
In a general curvilinear coordinate system (Riemann space) there is a unique shortest line
which connects two points. As you know, in a spherical system this line is a great circle.
These lines are called geodesics. The length is given by;

Rb Rb i j
s = ds = dt [gij dx dx ]1/2
a a
dt dt
In the above t is the parametric variable. To find the minimum length, apply the calculus of
variations. This will be discussed later in the course. Basically one varies all paths between
a and b which go through the end points a and b, choosing the one which is stationary, ie
this is similar to using the derivative of a function to determine an extremum. The operation
reduces to the Euler-Lagrange equations and finally the differential equation;

d2 xl + Γl dxi dxk = 0
ik ds ds
ds2
In a Cartesian system the Christoffel symbol vanishes which leads to a linear relation be-
tween the length and the coordinates.

Although this will be discussed in much more detail in future lectures, an example is the
equations of general relativity. We live in a 4-dimenaional Riemann (curvilinear) space as
shown in Figure 2. This figure illustrates that the world line is a geodesic in 4-D space. In

11
the figure, τ , is the proper time. A line element is written, ds = âi dη i, so that the length
squared is;

ds2 = gij dηi dη j

Figure 2: A 4-D Riemann space showing a world line with proper time, τ along the path

In a free, flat space, ie a vacuum where there is no matter to introduce curvature, the metric
is given by;

g11 · · · g1n
g = ···
gn1 · · · gnn
At each point in the space, the structure is given by the Einstein tensor equation;

Guv = κT uv

In the above, Guv is derived from the curvature tensor of space, and T uv is the energy-
momentum tensor. The path of a world line in this space is given by a geodesic. The
curvature tensor is written in more detail as;

Guv = Ruv − 1/2 Rg uv R = Ruu Ruv = Ruwv

w

w ∂Γw w
uv + Γw Γs ] − [ ∂Γuv + Γw Γs ]
Rupv = [ p
∂η ps uv ∂η v vs up

with the affine connection representing the differential change in the basis vectors with re-
spect to a variation of the coordinates;
∂gsv ∂gsu ∂guv
Γw ws
uv = (1/2)g [ + − ]
∂η u ∂η v ∂η s

12
The solution for a flat space is easily obtained as the Christoffel symbols vanish. One can
also look for perurbative solutions where the flat metric is slightly perturbed. An analytic
solution of the general form is however, not possible.

13