Subject: Re: [ublas] ublas double precision issues
From: Ungermann, Jörn (j.ungermann_at_[hidden])
Date: 2011-06-16 05:21:38
are you using double or similar as underlying type? inner_prod uses by
default the value_type of the supplied vectors. Should you use some float
type of less precision, you might try inner_prod_prec, which I found in the
source code, but which I haven't used, yet.
We use mostly double, as we esteem precision higher than performance. We
also use the numeric bindings to use the ATLAS dot product instead of the
ublas one (mostly for dense-matrix vector operations, though).
> -----Original Message-----
> From: ublas-bounces_at_[hidden] [mailto:ublas-
> bounces_at_[hidden]] On Behalf Of Umut Tabak
> Sent: Montag, 13. Juni 2011 17:47
> To: ublas mailing list
> Subject: Re: [ublas] ublas double precision issues
> On 06/13/2011 02:11 PM, Umut Tabak wrote:
> > Dear all,
> > I have implemented a lanczos solver in ublas, however there is a
> > subtle question from the point of view of accuracy related to
> > point operations.
> As a follow up to my question, the algorithm does some
> orthogonalizations, using the inner_prod function, this is where the
> accuracy is being lost. Could you comment on this also?
> If I have a thousand ideas and only one turns out to be good,
> I am satisfied.
> Alfred Nobel
> ublas mailing list
> Sent to: j.ungermann_at_[hidden]