[petsc-users] ill-conditioned problems in PETSc

[Hong Zhang]

Jim,

>
>     In fact, I will deal with the large sparse symmetric or unsymmetric
> (depending on the constitutive
> model used for the soil and the formulation of equations) linear system
> which usually occurs in geotechnical
> problems. As you know, In most geotechnical problems, the coefficient
> matrix can be severely
> ill-conditioned, thus calling  for the development of robust and efficient
> preconditioners and iterative methods.
>
>     The iterative method and parallel solver are my choices, actually are
> required by my professor. As I know,
> MUMPS is just an external solver like Matlab. So is it a right way to
> analyse the preconditioners and iterative
>
MUMPS and Superlu_dist are parallel sparse direct solvers for
ill-conditioned problems.
They can solve problems with 10k unknowns sequentially and scale up to
10+ or 100+ processors, i.e. solve problems with 100k unknowns.
Give them a try. Then, explore other preconditioners.

methods provided by PETSc like CG, GMRES, Jacobi, sor, etc. for my work?
>
>     Also I do not quite understand the second part of what Barry said
> ('You will not be able to use the external
> solvers with those mode'). So, what does 'those mode' mean. Using quad
> precision?
>

Yes, high precisions. Start from small problems with robust solvers, then
slowly
move to large, complex problems.

Hong

>
>
>
>
> At 2012-06-28 02:38:25,"Barry Smith" <bsmith at mcs.anl.gov> wrote:
> >
> >   Also if it the condition number is, say, larger than 10^10 you likely would benefit from using quad precision for the calculations.  http://www.mcs.anl.gov/petsc/documentation/faq.html#precision
> >
> >   You will not be able to use the external solvers with those mode but since it is 1d likely running sequentially will be fine.
> >
> >    Barry
> >
> >
> >On Jun 27, 2012, at 9:46 AM, Matthew Knepley wrote:
> >
> >> On Wed, Jun 27, 2012 at 7:55 AM, w_ang_temp <w_ang_temp at 163.com> wrote:
> >> Hello,
> >>
> >>     I want to deal with the ill-conditioned problems of Terzaghi's one-dimensional consolidation
> >>
> >> finite element equations in geotechnical engineering. So is there special support on ill-conditioned
> >>
> >> problems using ksp in PETSc.
> >>
> >>
> >> Use a direct solver, like built-in LU,  or in parallel, MUMPS (--download-mumps).
> >>
> >>   Matt
> >>
> >>     Thanks.
> >>
> >>                                                      Jim
> >>
> >>
> >>
> >>
> >>
> >>
> >> --
> >> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
> >> -- Norbert Wiener
> >
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120628/e9a5d1d5/attachment.html>