Module II
Lecture-8: Tensor analysis-II∗
1 A compendium of results of tensor analysis (continued)
• Mapping a vector to a one-form or “Lowering an index operation”: The metric
tensor, η = ηµν dxµ ⊗ dxν is a rank (0, 2) tensor i.e. it is a multilinear map from a pair of
vectors (say V, W ) to the real line,
η(V, W ) ∈ R.
Also since the metric is a symmetric tensor,
η(V, W ) = η(W, V ).
If instead of supplying two vectors if we supply only one vector to the metric η as an argument,
i.e. η(V, ), what kind of quantity do we obtain? The answer is that we have obtained a
one-form (cotangent space vector or covariant vector) because the object η(V, ) will feed a
vector, say W and spit out a real number, namely η(V, W ) and it is linear in the argument
as well. These are the defining properties of a one-form. This is evident when we use the
component-basis form,
η(V, ) = ηµν dxµ (V ) ⊗ dxν ,
= ηµν dxµ V λ ∂λ ⊗ dxν ,
= ηµν V λ dxµ (∂λ ) ⊗ dxν ,
= ηµν V λ δλµ ⊗ dxν ,
= ηµν V µ dxν ,
= (ηνµ V µ ) dxν .
Here we have use the definition of the dual basis, dxµ (∂λ ) = δλµ , and we have dropped the
⊗ sign because it is redundant (tensor product of a scalar with a tensor is same as normal
product of that of a scalar with a tensor). Thus the metric represents an isomorphism from
the tangent space to the cotangent space (in general it represents an isomorphic map from
the direct space to the dual space). The components of one-form dual to the a given vector
V = V µ ∂µ will also be denoted by the same letter but with lowered indices:
V∗ = Vµ dxµ , Vµ = ηµν V ν .
∗
Please report typos and errors to sroy@phy.iith.ac.in / abhattacharyya@iitgn.ac.in.
1
In physics parlance we call this map, a “lowering an index” operation, since it can be done
for components of a more generic tensor, say T......ν... by contracting some upper index with
ηµν
ηµν T......ν... = T...µ
... ...
.
• Inner product of two vectors: One defines the inner product (or the scalar product) of
two vectors V, W , denoted by V.W by,
V.W = η(V, W ) .
In terns of component-basis notation this works out to be
V.W = η(V, W )
= ηµν dxµ (V ) ⊗ dxν (W ) ,
= ηµν dxµ (V ) dxν (W ) ,
= ηµν dxµ (V α ∂α ) dxν W β ∂β ,
= ηµν V α W β dxµ (∂α ) dxν (∂β ) ,
= ηµν V α W β δαµ δβν ,
= ηµν V µ W ν .
Here again we have dropped the ⊗ sign since the tensor product of two scalars (real numbers)
is same as the normal product of two scalars. Clearly the inner product is symmetric in V
and W , i.e.
V.W = W.V = ηµν V µ W ν (1)
since the metric is a symmetric tensor, ηµν = ηνµ . Further using the value of the components
of the Minkowski metric, η00 = −1, η11 = η22 = η33 = 1 (rest all zero), we get,
V.W = ηµν V µ W ν = −V 0 W 0 + V 1 W 1 + V 2 W 2 + V 3 W 3 ,
= −V 0 W 0 + V · W . (2)
Note that for inner product of 4-vectors we are using a lower dot, while for the inner product
of 3-vectors we are using a center dot. Using the notion of the “lowering the index operation”,
i.e. by replacing ηµν V µ = Vν or ηµν W ν = Wµ , we get,
V.W = Vν W ν = V µ Wµ . (3)
Noting the index structure in (1) or (3), we see that all indices are contracted. Thus the
inner product is a scalar (invariant) under Lorentz transformations
V.W = ηµν V µ W ν = ηµν V ′µ W ′ν
where V ′µ and W ′ν are components of the vectors V and W in a different inertial frame.
2
• Norm (squared) of a vector & timelike, spacelike and null vectors: The inner
product of a vector V with itself gives the squared norm of the vector, denoted by ∥V ∥2 :
∥V ∥2 = V.V,
but we will simply denote the norm-squared as V 2 . Since the Minkowski metric ηµν has
indefinite signature, V 2 can be positive, negative or zero. Accordingly, vectors are spacelike
(V 2 > 0), timelike (V 2 < 0) or null (V 2 = 0) respectively. Since the norm squared is a Lorentz
invariant quantity, under a Lorentz transformation a timelike vector will remain timelike,
spacelike vector will remain spacelike and a null vector will remain null. Examples of timelike
2
−1/2
vectors are velocity 4-vector of a point particle, uµ = (γc, γv) where γ = 1 − vc2 is the
Lorentz factor. The covariant (one-form) dual is,
uµ = ηµν uν = (−γc, γv) .
So, in this case we have the norm-squared,
u2 = uµ uµ = −γ 2 c2 + γ 2 v 2
v2
2 2
= −γ c 1 − 2
c
2
= −c < 0.
In natural units, i.e. c = 1, we have u2 = −1. Another example is the the energy-momentum
4-vector, p = Ec , p of a physical system (particles or waves). For a point particle, the
energy-momentum 4-vector is defined as usual,
p=mu
where m is the invariant mass of the particle (which is referred to as the rest mass in older
terminology). For this case the norm-squared,
p2 = m2 u2 = −m2 c2 .
Next, consider the motion of a wave, for we can define the wave-propagation 4-vector, k ≡
ω
c
, k where ω is the frequency of the wave and k is the wave-vector (the normal vector to
the wave-fronts in the direction of the motion of energy/momentum). In this case, the norm
is,
ω2
k 2 = − 2 + k2 .
c
2
For a light wave or gravitational wave, k = 0, so it is a null vector and hence the wave
propagation 4-vector can be represented as, k = (|k| , k).
3
Comment: Although the Minkowski metric, η has an indefinite signature (i.e. norm squared
of a vector can be both positive, negative or zero) it does not lead to any inconsistencies
because it is nondegenerate. Usually the metric tensor in Euclidean geometry is defined
with positive signature, which guarantees that the norm squared is positive semidefinite, i.e.
g(U, U ) ≥ 0 with the equality holding only when U = 0 (zero vector). However for the
Minkowski metric (and in general for a Lorentzian signature metric) we have to relax this
condition and instead work with the nondegeneracy condition, which says that if
η(U, V ) = 0
for a given U and for all V , then and only then U = 0. This nondegeneracy condition
eliminates the possibility of the Minkowski metric (or for that matter any Lorentzian metric)
having eigenvectors with a 0 eigenvalue despite having special null vectors, i.e. non-zero
vectors U such that g(U, U ) = 0. Of course for the Minkowski metric we already know that
the eigenvalues are either +1 or −1, and so the nondegeneracy condition is automatically
satisfied.
