Vectors, Bases and Dual Bases: Contravariance and

Covariance

Matrices and Operators: Transformation of Vectors - II

Non-Degenerate Matrices
Earlier, in the context of transformation of vectors, we had seen in Fig. 2 that
when a matrix operates on vectors, they transform the vectors by rotating and
scaling them.
Most of such matrices when operating on some “special” vectors, only scale them
(and do not rotate). Typically there will be 2 such vectors for 2 × 2 (i.e. two
dimensional) matrices (since the negative of these vectors also get scaled, there
will be 4 vectors) forming two axial directions as shown in Fig. 3

Figure 3: A vector (pale blue) in regular coordinate system transforms to an-

other vector (pale red) when acted upon by an operator (matrix) as does the
point (marked pale red) move to another point (marked pale blue). However,
certain vectors (like the blue and red vectors) only scale when operated on by
such matrices - these vectors form certain directions in space (for 2-dimensions,
there are two such lines - blue and red lines in the figure - corresponding to two
such vectors and their negative counterparts). See text for further discussions
on this concept.

Such vectors that only scale under the transformation are called the “eigen-
vectors” of the matrix. The amount by which the vectors are scaled, is called
“eigenvalues”. A 2-dimensional matrix will have a maximum of 2 eigenvalues -
an n-dimensional matrix will have a maximum of n eigenvalues. Such matri-
ces that have their number of eigenvalues equal to their dimensions, are called
“non-degenerate” matrices. For the non-degenerate example in Fig. 3, the
eigenvectors corresponding to these eigenvalues are shown to be perpendicular
to each other - however, the directions can be in non-orthogonal (to each other)
directions too. In fact some matrices may not have as many eigenvalues as their
dimensions - such matrices are called “degenerate” matrices. If the number of
distinct eigenvalues is n − 1 (where n is the dimension of the matrix) then the
eigenvalue that “repeats” is termed 2-fold degenerate, if the number of eigenval-
ues is n − 2 then one of the eigenvalues will be 3-fold degenerate and so on (the
matrix may have different eigenvalues with different fold degeneracies) - defined
as the “order” of the degeneracy.

Eigenvalues and eigenvectors of a Matrix

Recalling our statement of eigenvectors above: an eigenvector |A⟩ of a matrix
M only scales and does not rotate. If we assume that the scaling happens by a
factor λ, then we can mathematically define |A⟩ to be related to M as:

M |A⟩ = λ |A⟩ (99)

in general λ can be a complex number and we say that |A⟩ is an eigenvector of

M.
The eqn. (99) is called an eigenvalue equation of a linear matrix M. There
will be as many values of λ as the dimension of the matrix and this forms a
“spectrum” of eigenvalues of M.
It should be noted that any scalar multiple α |A⟩ of |A⟩ will also be an eigen-
vector of M since:

M (α |A⟩) = α (M |A⟩)
= αλ |A⟩
= λ (α |A⟩)

where α is a complex number.

Hence
M (α |A⟩) = λ (α |A⟩) (100)
is an eigenvalue equation like eqn. (99).
It is thus customary to define a unit vector |Â⟩ that would be an eigenvector of
M (in order to remove the ambiguity posed by infinite scalar multiples):

⟨A|A⟩
|Â⟩ = (101)
||A||2

Degenerate Matrices
If the eigenvalue λ is g-fold degenerate, then there exist a total of g unit vectors
|B̂i ⟩ that are linearly independent having the same eigenvalues:

M |B̂i ⟩ = λ |B̂i ⟩ (102)

This implies that we can construct a vector |A⟩ that is a linear combination of
|B̂i ⟩:
Xg
|A⟩ = ci |B̂i ⟩ (103)
i=1

In such a case:
g
X
M |A⟩ = M ci |B̂i ⟩
i=1
g
X
= ci M |B̂i ⟩
i=1
g
X
= ci λ |B̂i ⟩
i=1
Xg
=λ ci |B̂i ⟩
i=1

i.e.
g g
! !
X X
i i
M ci |B̂ ⟩ =λ ci |B̂ ⟩ (104)
i=1 i=1

Hence any linear combination of all the eigenvectors corresponding to a g-fold

degenerate eigenvalue of a matrix M will also be its eigenvector with the same
eigenvalue.
We can, therefore, isolate that g linearly independent eigenvectors of M so as
to form a subspace of their own and the space will satisfy all the conditions of
a complete vector space as described in module PH7221-0100-01.

Computing the Eigenvalues and Eigenvectors of Matrices

We rearrange eqn. (99) to form the “Characteristic Equation”:

(M − λI) |A⟩ = |0⟩ (105)

where |0⟩ is a “Null Vector”.

In order for eqn. (105) to hold, we must have the determinant of the left hand
side to go to zero (since |A⟩ is assumed to not be a null vector):

Det (M − λI) = 0 (106)

The above equation will give us a nth order polynomial in λ which can be solved
to obtain n values of λ.
Each value of λ can then be put back in eqn. (105) to obtain n2 algebraic
equations; n of which have to be solved simultaneously (and this repeated n
times) to yield the n eigenvectors.
An example for 2-dimensions is given below

Example 6
Let
 
m11 m12
M=
m21 m22

and let
 
a1
|A⟩ =
a2

be an eigenvector of M.

Exercise 11: Substitute the matrices M and |A⟩ given above to

show that:
    
m11 − λ m12 a1 0
=
m21 m22 − λ a2 0

For the above matrix equation to hold, we must have the determinant of
the 2 × 2 matrix on the left hand side of the above equation to be zero:

m11 − λ m12
=0
m21 m22 − λ

Exercise 12: Compute the determinant above and show that it

leads to the quadratic equation:

λ2 − (m11 + m22 ) λ + (m11 m22 − m12 m21 ) = 0

Exercise 13: Solve the quadratic equation and show that it

yields two solutions:

m11 + m22 1
q
2
λ1 = + (m11 − m22 ) + 4m12 m21 (A)
2 2q
m11 + m22 1 2
λ2 = − (m11 − m22 ) + 4m12 m21 (B)
2 2

Exercise 14: Open the brackets of the characteristic equation of

Exercise 11 and obtain the two algebraic equations:

(m11 − λ) a1 + m12 a2 = 0 (C)

m21 a1 + (m22 − λ) a2 = 0 (D)

Exercise 15: Set λ = λ1 (from eqn. A) into eqns. (C) and (D)
and solve the resulting equations simultaneously and show that
one of the solutions will be:

a1 = 1
 
1
q
2
a2 = m22 − m11 + (m11 − m22 ) + 4m12 m21
2m12

hence, gathering a1 and a2 above as a vector, show that:

 
1  
|λ1 ⟩ =  1
q
2 
m22 − m11 + (m11 − m22 ) + 4m12 m21
2m12

Exercise 16: Set λ = λ2 (from eqn. B) into eqns. (C) and (D)
and solve the resulting equations simultaneously and show that
one of the solutions will be:

a1 = 1
 
1
q
2
a2 = m22 − m11 − (m11 − m22 ) + 4m12 m21
2m12

and gather a1 and a2 above as a vector, show that:

 
1  
|λ2 ⟩ =  1
q
2 
m22 − m11 − (m11 − m22 ) + 4m12 m21
2m12

Exercise 17: Assuming all terms to be real show that:

m12 − m21
⟨λ2 |λ1 ⟩ =
m12
This implies that if M was symmetric and m12 = m21 , then

⟨λ2 |λ1 ⟩ = 0

and the eigenvectors would be orthogonal to each other.