GSO gives the same result as TMatrixDEigen in cases where they both work.
I could explicitly check that the vectors obtained by GSO are indeed
orthogonal, but I think that get same result indicates there's not a bug
that might make them non-orthogonal ... though I guess I can't exclude a
numerical problem in the cases where they disagree (one works, on
doesn't), i.e., TMatrixDEigen fails, but GSO just gives a wrong result.
Hopefully, using symetric matrix will resolve the problem ..
-Arthur
A.E. Snyder, The Former Group C (TFC) \!c*p?/ SLAC Mail Stop #95 ((. .)) Box 4349 | Stanford, Ca, USA, 94309 '\|/` e-mail:snyder_at_slac.stanford.edu o phone:650-926-2701 _ http://www.slac.stanford.edu/~snyder BaBar FAX:707-313-0250 Collaboration & Fermi/GLAST
On Wed, 6 Jun 2012, Edmond Offermann wrote:
> You must have a numerically challenged problem.
> Did you try your GSO results and checked hat they are
> indeed eigen values/vectors ?
>
>
> ____________________________________________________________________________
> From: Arthur E. Snyder <snyder_at_slac.stanford.edu>
> To: Edmond Offermann <edmondoffermann_at_yahoo.com>
> Cc: "roottalk_at_root.cern.ch" <roottalk_at_lxroot01.cern.ch>
> Sent: Wednesday, June 6, 2012 4:42 PM
> Subject: Re: [ROOT] Histogram mean --> eigen matrix
>
> Thanks, Edmond, I'll try explicitly sysmetrix matrix formulation.
>
> Incidentlally, while eigen finding also fails in R, it works in maple. I'm
> also baffeled by why my home brewed Gram-Schmidt orthogonalization routine
> works when apparently more sophisticated approaches don't -- though my GSO
> though does assume symmetry, so may be that has something to do with it.
>
> If sysmetrix matrix routine works that will solve the problem for all
> practical purpuoses.
>
> -Arthur
>
> A.E. Snyder, The Former Group C (TFC) \!c*p?/
> SLAC Mail Stop #95 ((. .))
> Box 4349 |
> Stanford, Ca, USA, 94309 '\|/`
> e-mail:snyder_at_slac.stanford.edu o
> phone:650-926-2701 _
> http://www.slac.stanford.edu/~snyder BaBar
> FAX:707-313-0250 Collaboration
> &
> Fermi/GLAST
>
>
>
> On Wed, 6 Jun 2012, Edmond Offermann wrote:
>
> > Hi Arthur,
> >
> > What about using the fact that a correlation matrix is positive definite
> and symmetric,
> > so use TMatrixDSymEigen.
> >
> > The TMatrixDEigen is based on a Fortran subroutine in EISPACK, so moving
> to another package
> > might not help that much because it is most likely also based on EISPACK.
> >
> > - Eddy
> >
> >___________________________________________________________________________
> ______________________________________________
> > From: Arthur E. Snyder <snyder_at_slac.stanford.edu>
> > To: Arthur E. Snyder <snyder_at_slac.stanford.edu>
> > Cc: "roottalk_at_root.cern.ch" <roottalk_at_lxroot01.cern.ch>
> > Sent: Friday, June 1, 2012 4:17 PM
> > Subject: Re: [ROOT] Histogram mean
> >
> > Hi Rooters,
> >
> > I am having trouble with |TMatrixDEigen| for larger matrices. I have made
> 200x200 matrices of correltions which I'm
> > trying to digagonalize. It mostly works, but sometimes fails with
> >
> > root [13] TMatrixDEigen emat(*scanMatrix)
> > Error in <MakeSchurr>: too many iterations
> >
> > It apparently thinks the matrix is too close to singular.
> >
> > However, I can still find diagonalilzed variables with ugly, homemade
> Gram-Schmidt (which I'm trying to replace with
> > something better).
> >
> > The matrix that gives trouble has in fact smaller correlations than some
> of the ones that work successfully.
> >
> > So what's going on here? And how do I fix it? Do I have to move beyond
> root?
> >
> > -AE
> >
> >
> >
> >
> >
> >
> >
> > A.E. Snyder, The Former Group C (TFC) \!c*p?/
> > SLAC Mail Stop #95 ((. .))
> > Box 4349 |
> > Stanford, Ca, USA, 94309 '\|/`
> > e-mail:snyder_at_slac.stanford.edu o
> > phone:650-926-2701 _
> > http://www.slac.stanford.edu/~snyder BaBar
> > FAX:707-313-0250 Collaboration
> > &
> > Fermi/GLAST
> >
> >
> >
> >
> >
> >
> >
>
>
>
Received on Thu Jun 07 2012 - 02:41:15 CEST
This archive was generated by hypermail 2.2.0 : Fri Jun 08 2012 - 05:50:02 CEST