Re: regression in eigs

[c.]

On 4 Nov 2013, at 19:42, Rik <address@hidden> wrote:

> On 11/03/2013 11:56 PM, c. wrote:
>> On 4 Nov 2013, at 08:55, c. <address@hidden> wrote:
>> 
>>> Hi,
>>> 
>>> I just found a nasty regression in eigs when a numerical value is provided
>>> and the size of the matrix / number of eigenvalues cause a fallback to eig
>>> 
>>> I think the fix is trivial so it is worth doing before the release, but I 
>>> don't have time to do it until this evening, so I just pushed a failing test
>>> to show the problem for the moment.
>>> 
>>> I'll push the fix later today if no one does it before me.
>>> 
>>> c.
>> I take it back, it seems I can't push the changeset without creating a new 
>> head, 
>> so here's the changeset as an attachment instead.
>> c.

Rik, 

Thanks for pushing this, sorry I was to busy to do it myself.

> 11/4/13
> 
> Carlo,
> 
> I pushed your changeset.  I also fixed the problem in this changeset
> (http://hg.savannah.gnu.org/hgweb/octave/rev/42453dcfa95e).  There is still
> a small difference in that eigs normally returns the k eigenvalues closest
> to sigma (when numeric sigma) in descending order.  The patch returns the k
> eigenvalues closest to sigma in the order of their closeness to sigma.  
> If
> someone wants to patch that it would be fine,

I checked on Matlab and the eigenvalues are sorted in the order 
of their closeness to sigma when the matrix is small.
so, however strange, the current approach is the compatible one.

>  but the regression is cleared
> up and all tests pass again.
> 
> --Rik

thanks,
c.