Rank one perturbation of a symmetric matrix:
cs.nju.edu.cn/people/zhaojinxi/miao.pdf
And the secular equation can be solved
with laed4(N,I,D,Z,rho,Delta,Dlambda,info), Delta is useful for the
calculation of the new eigenvectors. Z is not the original modifying vector
but its transformation with the original eigenvectors.
Laszlo
2006/11/15, Laszlo Sragner <sragner@Domain.Removed>:
Hi,
if X is an N x M (N>>M) matrix let C = X' * X and C = VDV'
if w is a Mx1 vector what is the diagonalization of ( C + ww' )?
I understand that the trick is to diagonalize (D+pp') where p=Vw
and that this is the heart of the divide and conquer algorithm
but are there any LAPACK function to do it?
I am also interested to do an SVD on
[ X ]
[ -- ]
[ w']
which is an analogous problem and interesting if C is rank defficient.
Cheers,
Laszlo
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.cs.utk.edu/private/lapack/attachments/20061121/daf0cc7e/attachment.html
|