commons-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Gilles Sadowski <gil...@harfang.homelinux.org>
Subject Re: [math] Eigen decomposition of real asymmetric matrices
Date Sun, 22 Jul 2012 20:20:41 GMT
On Sun, Jul 22, 2012 at 06:39:02PM +0200, Luc Maisonobe wrote:
> On 22/07/2012 17:09, Thomas Neidhart wrote:
> > On 07/18/2012 10:55 PM, Thomas Neidhart wrote:
> >> Hi,
> 
> Hi Thomas,
> 
> >>
> >> I finalized now this feature request (MATH-235) by adding more random
> >> data test which success for an epsilon of 1e-12, which imho is pretty
> >> good for now.
> >>
> >> For the problem with the internal DecompositionSolver, I now throw an
> >> MathUnsupportedOperationException when trying to get a solver for a
> >> decomposition with complex eigenvalues. Feedback / input to solve also
> >> for the case of complex eigenvalues is very welcome.
> >>
> >> I will resolve the issue for now, as there was no feedback since my last
> >> mail a few weeks ago. If somebody detects an issue with the current
> >> implementation, please open a new issue.
> > 
> > it looks like this topic doesn't seem to catch lots of interest ;-).
> > 
> > Anyway, there is an issue MATH-651 that was about the formerly unused
> > field imagEigenvalues. As we can now handle all kinds of matrices and
> > the eigen decomposition can return real or complex eigen values, we have
> > to provide a way for users to distinguish what kind of eigen values we
> > have calculated. There are different options (with the current interface
> > which we have to keep for backwards compatibility):
> > 
> >  - always create an imaginary value array and set it to the zero vector
> >    in case of real values: requires the users to check the array for
> >    zero values to see if the result is real or complex
> > 
> >  - in case of real values, have a null array for the imaginary part
> >    getImagEigenvalue(int i) would throw an exception or return 0 in case
> >    of real values
> > 
> >  - leave the way it is now (option 1) and provide an additional method
> >    something like: boolean isResultComplex()

+1
But I preferred the name in the earlier proposal (something like
"hasComplexEigenValues").


Gilles

> > 
> > Any opinions?
> 
> I am not a user of this algorithm, so take my opinion only as a random
> thought, not as something worth considering. I prefer to always have the
> imaginary part, even if it is zero.
> 
> Luc
> 

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org


Mime
View raw message