/  2
 
Using SVD to generalise Sylvester’s Law
of Inertia to non-square matrices
Let
 
be a Hermitian matrix
1
. All the Eigenvalues of 
 
are hence real. Define

as the number of positive Eigenvalues,

as the number of negative Eigenvalues and

as the number of zeroEigenvalues.Define the inertia of 
 
,

, as the tuple

 
 
 
 Suppose
is invertible (and hence square). Then
Sylvestia’s law of Inertia
states that
2
:

 
 It is possible to generalise this for non-square matrices
.Suppose
 

is still Hermitian and that

with
.
can be written as

, where:
 

and

 
 

and

 
 

and

where
 This is the
Singular Value Decomposition
of 
.It hence follows that:
 

 Define


so that:

 
 
Note
, and so
is Hermitian.
is invertible
by definition, and thus it follows from Sylvester’s
Law that:

 

 The problem is hence reduced to finding the inertia of 
.As
only has non-zero values on its diagonal, and
, it follows that

when either
 or
 
. Denote the principle
submatrix of 
by
. It follows that:

 
1
i.e.
 


, where the bar denotes complex conjugation.
2
This can be interpreted as saying that for a sesquilinear form
 
, any basis
where
 

,
 

has the same set of values
 
(up to permutation).

Share & Embed

More from this user

Add a Comment

Characters: ...