Tensors

Types of Tensors and their uses

Scalars represent a quantity with magnitude and no direction. Vectors are used to the same but
have one direction. By extension, when there’s a need for a quantity with some magnitude and
multiple directions, we use Tensors.

• Scalar: Tensor of Rank 0 (1 components)

• Vector: Tensor of Rank 1 (3 components)
• Dyad: Tensor of Rank 2 (32 = 9 components)
• Traid: Tensor of Rank 3 (33 = 27 components)

Dyad products
If U and V are two vectors, their dyad product is simply UV, which is neither a dot, cross, nor
a matrix product.

Assume U = u1 i + u3 j + u2 k and V = v1 i + v2 j + v3 k, then the dyad product is

UV = u1 v1 ii + u1 v2 ij + u1 v3 ik + u2 v1 ji + ...
where i, j, k are units vectors and ii, ij, ik, ji, etc. are unit dyads. Say u1 v1 = µ11 , u1 v2 = µ12 ,
etc., dyad can be rewritten as:

UV = µ11 ii + µ12 ij + µ13 ik + µ21 ji + ...

Scalar components µij can be rearranged into a 3x3 matrix.

Dyad product is not commutative, and gives rise to a tensor of rank 2 i.e a Dyad.
Just like Matrices, pre- and post-multiplication yields different results with dyads

S · UV = (S · U)V = σV
where σ = (S · U), the result being a vector with magnitude σ and direction V.

Similarly, post-multiplication leads to a vector with magnitude λ and direction U.

1
Tensor Contraction
The Rank of a tensor can be reduced (contracting it literally) by taking the dot product of any
two consecutive terms in an ”n-ad”.
For example, say you have a triad UVW, it can then be contracted into two different vectors (or)
rank 1 tensors:
U · VW = (U · V)W = αW
or
UV · W = U(V · W) = βU
Notice how the position of the dot changes the resultant tensor.
This contraction can be performed on a tensor of any rank n and the resultant tensor is always
of rank (n-2). Obviously n ≥ 2.

For example, the force-velocity dyad Fv. The dot product between F and v gives power (or)
rate of change of energy in a system of fluid dynamics
dE
=F·v
dt

This demonstrates the need for tensors, their ability to store information and the ease with which
they can be manipulated.

Invariance of tensors
Scalars and rank 0 tensors
Assume we define a scalar T, which measures the temperature at a point P in a frame of reference
called K. Now, we measure the temperature at the same point P in a frame of reference K* that
is in relative motion w.r.t K, and let this be defined by a scalar T*.

Since temperature is inherently an invariant quantity, this means that when transforming from
K to K*, our tensors are equivalent, i.e.
T = T*
Only scalars that transform between two coordinate systems like above (invariantly) are classified
as tensors of rank 0, i.e. the equation T = T* can be the definition of a rank 0 tensor.

Now consider T to be the frequency of light emitted by a monochromatic source. Say, in the
frame K, we measure T = ν0 . This means that in the frame K*, T* = ν0 ± ∆ν due to the Doppler
effect.
Obviously then,
T ̸= T*
In this case, despite T being a scalar, we cannot promote it to the status of a rank 0 tensor since
it is not invariant between coordinate systems.

2
Vectors and rank 1 tensors
A similar extension can be made onto vectors, where not all vectors are rank 1 tensors, but all
rank 1 tensors are vectors.

Now, consider a position vector V pointing towards a point P from the origin of K, and V*
extending to the same point from the origin of K*. Assume that the origins of K and K* do not
coincide. This necessarily means that V ̸= V*.

Now suppose, V1 and V2 are position vectors of points P1 and P2 in K. Similarly V2 * and
V2 * are also vectors of P1 and P2 in K*. The vector extending from P1 to P2 must be the same
vector in both systems, meaning we have
V1 - V2 = V1 * - V2 *
i.e., while the individual position vectors are themselves not a tensor, the difference vector is indeed
a rank 1 tensor.

Covariance and Contravariance

Assume a 3-dimensional Cartesian coordinate system. At point P in this space, there exist local
coordinate axes (lines passing through this point and parallel to the coordinate axes) and local
coordinate surfaces (planes passing through this point and parallel to coordinate planes)

Let there be two sets of unit vectors. Set (i, j, k) are tangent to the local axes and set (i*,
j*, k*) are perpendicular to the local surfaces.

From geometry, we can argue that these two sets comprise of the same unit vectors, i.e

i = i*
j = j*
k = k*
However, this approach does not yield the same result for different coordinate systems. Assume
a triangular coordinate system where all axes are mutually at 60◦ . Through the same process as
above, you will find that the two sets of vectors are in fact, not equivalent.

The set of unit vectors tangent to the local axes are Contravariant while the ones perpendicular
to the local surfaces are Covariant

3
Reciprocal sets of Vectors
The unit vector set (i, j, k) is Contravariant, and (i*, j*, k*) is Covariant.

The relationship that then exists among the Contravariant and Covariant unit vectors is then:
i · i∗ = i∗ · i = 1 i · j∗ = j∗ · i = 0 i · k∗ = k∗ · i = 0
∗ ∗ ∗ ∗
j·i =i ·j=0 j·j =j ·j=1 j · k∗ = k∗ · j = 0
k · i∗ = i ∗ ·k = 0 k · j∗ = j ∗ ·k = 0 k · k∗ = k ∗ ·k = 1

Let’s define ui ≡ (i, j, k) and uj ∗ ≡ (i*, j*, k*). We can then conclude that

ui · uj ∗ = δij

Where δij is a component of a second-rank tensor called Kronecker’s delta. All vectors satisfying
this relationship are reciprocal (
1 i=j
δij =
0 i ̸= j
Similarly, ui ∗ ·uj ∗ = gij is defined as a component of a second rank tensor called the fundamental
tensor

Now, extending onto a more general case, the Contravariant and Covariant sets now comprise base
vectors instead of unit vectors, i.e they are represented as < e(1) , e(2) , e(3) > and < e(1) , e(2) , e(3) >
as Contravariant and Covariant triplets respectively

Kronecker’s delta is then represented as a mixed tensor of rank 2, i.e

e(i) · e(j) = e(j) · e(i) = δij

And, the fundamental tensor also has two representations as follows,

e(i) · e(j) = gij (Covariant)
e(i) · e(j) = g ij (Contravariant)