Why repeated indices are dummy indices

Repeated indices can be relabelled anytime, that’s why they are called dummy indices. It is easy to check
this. Let’s look at the dot product of two vectors, i.e.,a · b. Now we know in the Einstein convention this
is expressed as,
a · b = ak bk
where the repeated index “runs over” i.e. takes ALL values from 1 to 3.
So,
ak bk = a1 b1 + a2 b2 + a3 b3 .
Now instead of k, if I use a different label, say, m for the repeated index, i.e.
a · b = am bm ,
then we see it makes no difference, as m being a repeated index has to run over all values, 1 to 3.

Thus we can relabel a repeated index (dummy index) anytime.

Why dummy indices cannot be repeated more that two times

If a dummy index is repeated more than twice it will lead to inconsistency/erroneous expressions. This
can be demonstrated simply, even in the case of Cartesian tensors in 2 dimensions, i.e. vectors have 2
components instead of 3, e.g. a = (a1 , a2 ). Let’s say we have an expression, (a · b)2 , which is the square
of (a · b), the scalar product and is a real (or more generally complex) number. Now we can expand,
(a · b)2 = (a1 b1 + a2 b2 )2 = a21 b21 + a22 b22 + 2a1 a2 b1 b2 . (1)
Lets try to reproduce this result from Einstein summation convention. According to this convention any
repeated index must be summed over values 1,2. E.g.a · b = ai bi = a1 b1 + a2 b2 . Then we can have for
our example, two possible choices,
(
2 (ai bi ) (ai bi ) = ai ai bi bi ,
(a · b) = (a · b) (a · b) =
(ai bi ) (aj bj ) = ai aj bi bj .

In the first choice we have repeated i, four times while in the second case we have not repeated any index
more than twice. Now let’s see which one reproduces the correct result (1).

ai ai bi bi = a1 a1 b1 b1 + a2 a2 b2 b2
= a21 b21 + a22 b22 .
Thus the expression with dummy index, i repeated more than twice misses the cross term, 2 a1 a2 b1 b2 in
(1). So this must be incorrect. On the other hand, the second choice,
ai aj bi bj = a1 aj b1 bj + a2 aj b2 bj
= a1 a1 b1 b1 + a1 a2 b1 b2 + a2 a1 b2 b1 + a2 a2 b2 b2
= a21 b21 + a22 b22 + 2a1 a2 b1 b2 .

1
This is indeed the correct expression.
Thus a repeated (dummy index) cannot appear more than 2 times.

Matrix representation of the equation OT O = I

The ij-component of a matrix A i.e. the element in the i-th row and j-th column is denoted by Aij . Thus
i takes values from1 to m, the number of rows, while the index j takes values fom 1 to n, where n is the
total number of columns. We will mostly work with square matrices, so number of rows and columns will
be same, m = n. Now consider two matrices, A and B. In terms of the components of A and B, the
ij-component of the matrix product AB is given by:
n
(AB)ij ≡
X
Aik B kj = Aik B kj (2)
k=1

Also, recall that the transpose of a matrix A, denoted by AT is obtained by turning the rows of A into
columns i.e. the first row of A is the first column of AT and so on. Thus, by definition, the ij-component
of AT is the same as the ji-component of A,
 ij
AT = Ajk .

Next we want to write out the matrix equation, OT O = I, in index/component notation. This equation
defines an orthogonal matrix, O, i.e. a square matrix, whose tranpose acts as its inverse, OT = O−1 .
Since both sides of the equation are matrices, this equation means individual matrix elements on both
sides should be equal, i.e.ij-component or ij-element of both sides are same for all possible values of i
and j,
 ij
OT O = I ij .
Now in class we introduced the Kronecker delta symbol, δ ij as the index or component notation for
the identity matrix, I. Since the identity matrix has digonal elements equal to unity while off diagonal
elements vanish, we have the definition of the Kronecker delta,
(
ij 1, when i = j
δ = .
0, when i 6= j
So we have,
 ij
OT O = δ ij
Expanding the lhs of this equation using the indexed representation of the product rule (2) with A = OT
and B = O, we get,
 ik
OT Okj = δ ij
 ik
but by very definition of transpose, OT = Oki . Thus plugging this in the above equation we get,

Oki Okj = δ ij .

(Ungraded) Exercise: Show that the index/component representation of the matrix equation OOT = I is
Oik Ojk = δ ij .