At Thu, 21 Aug 2008 19:35:12 +0200,
Reinhold Bader wrote:
>     the routine fgsl_linalg_QR_update appears to be incorrectly
>     documented.  Comments in the source as well as my own testing
>     says that the update formula is
> 
>      Q'R' = Q ( R + w v^T )
> 
>     and not Q'R' = Q R + w v^T, as the 1.10 manual presently says.

Thanks, you're right, I've corrected the manual to match the comment
in the source.

>     the routine gsl_linalg_QRPT_update appears to be also
>     incorrectly documented; in this case I think the comments in the
>     source are also partially incorrect.  My own testing implies
>     that the update formula is
> 
>      Q'R' = Q ( R + w v^T ) P
> 
>     and not Q'R' = Q R + w v^T, as the 1.10 manual presently says,
>     or
>             Q'R' = Q ( R + w v^T P), as mentioned in the qrpt source
> file.

Yes, the manual is wrong. I think the comment in the source is right
though -- the test program below reproduces Q(R + w v^T P) (let me
know if I'm missing something here!).

I've added some tests for gsl_linalg_QRPT_update, since they were
missing, and also extended it to handle rectangular matrices.

Thanks for the bug report.

-- 
Brian Gough