The reason we have taken so much effort to determine the iteration matrix is that if
![]() |
(146) |
![]() |
(147) |
![]() |
(148) |
Theorem If
has eigenvalues
, the iteration converges
as
provided
.
We call
the spectral radius of the matrix
.
We consider only a special case, where there is a full set of eigenvalues, and if
where
is a diagonal matrix of the eigenvalues,
then
and this will become the zero vector provided the moduli
of all the eignevalues are less than one.
Theorem (Gershgorin) The eigenvalues of
lie within the union of the discs
![]() |
(149) |
This can be seen by supposing eigenvalue has eigenvector
so that
![]() |
(150) |
![]() |
(151) |
Lemma Jacobi iteration will converge if
is strictly diagonally dominant,
![]() |
(152) |
In this case, using Jacobi iteration
![]() |
(153) |
![]() |
(154) |
Lemma Relaxation can converge only if
.
To show this, consider the eigenvalues of the iteration matrix,
,
![]() |
(155) |
![]() |
(156) |
We do not have time to go into the mathematics deeply, but there will be a critical
value for which
is a minimum. For most practical
matrices this optimum value of the relaxation parameter can be found from numerical
experiments.