tolerance for diagonal entries + Slepc

[Farshid Mossaiby]

Matrix Market indices are 1-based.

Please keep this in mind when searching.

--- On Mon, 8/3/09, Umut Tabak <u.tabak at tudelft.nl> wrote:

> From: Umut Tabak <u.tabak at tudelft.nl>
> Subject: Re: tolerance for diagonal entries + Slepc
> To: "PETSc users list" <petsc-users at mcs.anl.gov>
> Date: Monday, August 3, 2009, 7:08 PM
> Hong Zhang wrote:
> > 
> > Umut,
> > The error msg
> > "Matrix is missing diagonal entry 1847!"
> > has nothing to do with the tolerance.
> > It indicates that you did not input any value
> > to the 1847-th diagonal entry.
> > Many routines required diagonal entries been
> > allocated and set.
> > If that entry is zero, you should set zero value to
> it.
> > 
> > Hong
> > 
> Hi Hong,
> 
> Thanks for the quick reply, that last part of the matrix in
> matrix market format, I read this with one of the examples
> that you can read in the matrix market real format:
> ...
> 1848 925 1.207005667e-06
> 1848 1032 1.545494254e-08
> 1848 1771 1.207005667e-06
> 1848 1777 4.828022669e-06
> 1848 1782 -3.078839811e-08
> 1848 1842 1.545494254e-08
> 1848 1848 6.181977018e-08 // this is source of the issue
> ...
> So actually very close to zero. I have this numerical
> problem on most of my matrices that I would like to
> interface with Petsc. The point which made me think this way
> is that Matlab which uses Arpack to solve eigenvalue
> problems did not experience any problems to solve the
> system. In Petsc+Slepc, I ran into this problem(a
> generalized eigenvalue problem,BTW).
> 
> Best regards,
> Umut
> 
> 
>