LAPACK Archives

[Lapack] accuracy and blas

fyi ( I messed up the reply to all in the previous email. Thanks for the email 
Sven, Julien. )


On Jul 1, 2011, at 11:40 PM, <jas@Domain.Removed> <jas@Domain.Removed> wrote:

Dear Julien
That's very clear -- thanks very much indeed.
And thank you for replying so quickly.
Best regards, Jonathan

Hello Jonathan,

Changing the BLAS library will not change the condition number of your
problem.
(So using an optimized BLAS library will not help, nor add any problems.)

Cheers,
J



On Jul 1, 2011, at 5:31 PM, Jonathan Sherratt wrote:

I apologise in advance that this is a rather ill-informed question:
numerical linear algebra is not my field, and I got conflicting advice
from my numerical analysis colleagues when I asked them about this.....
I am using DGEEVX to calculate eigenvalues/eigenvectors of some real
non-symmetric matrices (typical size 1000x1000). The matrices have a
large condition number and the lapack-estimated errors bounds EERRBD are
quite significant for the eigenvalues close to the origin. Currently I
am just using the generic blas routines. I realise that using an
optimized blas library would speed up the calculations, though that is
not a big issue for me. But I am wondering whether I would be likely to
get any reduction in errors by using an optimized blas?
Thanks very much
Jonathan Sherratt
Professor of Mathematics
Heriot-Watt University
Edinburgh, UK


--
Heriot-Watt University is a Scottish charity
registered under charity number SC000278.

_______________________________________________
Lapack mailing list
Lapack@Domain.Removed
http://lists.eecs.utk.edu/mailman/listinfo/lapack





-- 
Heriot-Watt University is a Scottish charity
registered under charity number SC000278.



<Prev in Thread] Current Thread [Next in Thread>


For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or