Re: eigenvectors/values for 60x60, bw 8 sparse symmetric posdef matrix

On Nov 19, 9:53 pm, spellu...@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
(Peter Spellucci) wrote:
In article <3c7a2540-cd69-4691-88d6-0792822e7...@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>, erik <e...@xxxxxxxxxxx> writes:

>I use the Jacobi iteration to get the eigenvalues of a 60x60 positive
>definite symmetric sparse band matrix with bandwidth 6.
>I that a good choice ?
>I must do it in PHP, which does not really do array/vector/matrix
>stuff like C and Fortran.
>In this language, arrays are sparse by design.
>If you do not assign a value to an array element, that element does
>not exist.
>Also, PHP is slooooow compared to Fortran.

much worse than this the jacobi rotation scheme will destroy the banded form
of your matrix such that you will have to work with the full 60 times 60 case ..
not a problem on a fast machine in this order of magnitude but if PHP is really
so much slow as you indicate it may be indeed not a good idea. there is
band-QR which protects the banded structure of the matrix and is very fast, but
it might be some work to translate this from the f77 form in lapack resp.
eispack to this language (I don't know this)
hth
peter

indeed: your suggestion set me on the right track:
I found QL stuff in EISPACK. Now the process only takes 1/30th or so
of the time... very acceptabel.
thmx :)
.