Hey Lorenzo,
If you believe that your matrix should be invertible and
lu should work, try giving PETSc the following options,
-pc_type lu -pc_factor_mat_solver_package superlu
-pc_factor_nonzeros_along_diagonal
To my knowledge this is the most stable lu factorization routine in PETSc.
If it doesnt work, the matrix is probably not invertible.
Thanks.
On Thu, Mar 7, 2013 at 9:06 AM, Lorenzo Zanon <[email protected]>wrote:
> Hi,
>
> My problem converges using:
>
> -pre
> -snes_type ls
> -snes_type basic
>
> but I still get divergence if I add to the above options the direct linear
> solver:
>
> -ksp_type preonly -pc_type lu
>
> This looks suspicious to me, also because I get divergence if I delete the
> "-pre" option (also with iterative linear solver)...
>
> Best,
> Lorenzo
>
>
>
> On Mar 6, 2013, at 7:32 PM, Subramanya Gautam Sadasiva wrote:
>
> > I have noticed that the default line search type used by petsc fails
> quite regularly (at least for the problem that I am solving) . I would
> suggest doing "-snes_linesearch_type basic" . This works for the latest
> version of petsc. I think this option used to be "-snes_ls_type" for the
> older versions of petsc.
> >
> >
> > Subramanya Sadasiva
> >
> > "But memory eventually fades. Turbulences damp out, internal strains
> yield to plastic flow, concentration inhomogeneities diffuse to uniformity.
> Systems tend to subside to very simple states,independent of their specific
> history."
> > Herbert Callen . Thermodynamics and an Introduction to Thermostatics.
> >
> > ----- Original Message -----
> > From: "Lorenzo Zanon" <[email protected]>
> > To: "Derek Gaston" <[email protected]>
> > Cc: "Subramanya Gautam Sadasiva" <[email protected]>, "libmesh-users"
> <[email protected]>
> > Sent: Wednesday, March 6, 2013 1:26:21 PM
> > Subject: Re: [Libmesh-users] Newton iteration
> >
> > I get convergence only using the preconditioning option "-pre". If I set
> a direct solver for the linear system - as I usually do - with the option
> "-ksp_type preonly -pc_type lu", I get instead a failure message:
> >
> > Nonlinear solve did not converge due to DIVERGED_LS_FAILURE
> >
> > I think there's still a problem, because I should get convergence also
> with the linear solver, if the problem is properly set. Or am I wrong?
> >
> > Thanks a lot,
> > Lorenzo
> >
> > On Mar 6, 2013, at 4:48 PM, Derek Gaston wrote:
> >
> >> I would also recommend using -ksp_monitor so you can see how the linear
> solve is going. I suspect that it's going nowhere and therefore the solve
> is failing...
> >>
> >> Derek
> >>
> >>
> >> On Wed, Mar 6, 2013 at 7:11 AM, Subramanya Gautam Sadasiva <
> [email protected]> wrote:
> >> Hi,
> >> If you are using petsc, I'd suggest running with
> -snes_converged_reason. It looks to me that your problem is diverging for
> some reason. The reasons could be that the line-search or the solver is
> failing , which can also be changed with command line options.
> >>
> >> Subramanya Sadasiva
> >>
> >> "But memory eventually fades. Turbulences damp out, internal strains
> yield to plastic flow, concentration inhomogeneities diffuse to uniformity.
> Systems tend to subside to very simple states,independent of their specific
> history."
> >> Herbert Callen . Thermodynamics and an Introduction to Thermostatics.
> >>
> >> ----- Original Message -----
> >> From: "Lorenzo Zanon" <[email protected]>
> >> To: [email protected]
> >> Sent: Wednesday, March 6, 2013 6:53:02 AM
> >> Subject: [Libmesh-users] Newton iteration
> >>
> >> Hello,
> >>
> >> I'm modifying miscellaneous_ex3.C (2D Laplace-Young Problem Using
> Nonlinear Solvers, in my current libmesh version). After adapting residual
> and jacobian to my own problem, although the tolerance is not met, the
> Newton solver stops after one iteration with an all-zero (incorrect)
> solution:
> >>
> >> NL step 0, |residual|_2 = 1.598435e+00
> >> Laplace-Young system solved at nonlinear iteration 0 , final nonlinear
> residual norm: 1.598435e+00
> >>
> >> What could be the reason for that?
> >>
> >> Thanks a lot,
> >> Lorenzo
> >>
> >>
> ------------------------------------------------------------------------------
> >> Symantec Endpoint Protection 12 positioned as A LEADER in The Forrester
> >> Wave(TM): Endpoint Security, Q1 2013 and "remains a good choice" in the
> >> endpoint security space. For insight on selecting the right partner to
> >> tackle endpoint security challenges, access the full report.
> >> http://p.sf.net/sfu/symantec-dev2dev
> >> _______________________________________________
> >> Libmesh-users mailing list
> >> [email protected]
> >> https://lists.sourceforge.net/lists/listinfo/libmesh-users
> >>
> >>
> ------------------------------------------------------------------------------
> >> Symantec Endpoint Protection 12 positioned as A LEADER in The Forrester
> >> Wave(TM): Endpoint Security, Q1 2013 and "remains a good choice" in the
> >> endpoint security space. For insight on selecting the right partner to
> >> tackle endpoint security challenges, access the full report.
> >> http://p.sf.net/sfu/symantec-dev2dev
> >> _______________________________________________
> >> Libmesh-users mailing list
> >> [email protected]
> >> https://lists.sourceforge.net/lists/listinfo/libmesh-users
> >>
> >
>
>
>
> ------------------------------------------------------------------------------
> Symantec Endpoint Protection 12 positioned as A LEADER in The Forrester
> Wave(TM): Endpoint Security, Q1 2013 and "remains a good choice" in the
> endpoint security space. For insight on selecting the right partner to
> tackle endpoint security challenges, access the full report.
> http://p.sf.net/sfu/symantec-dev2dev
> _______________________________________________
> Libmesh-users mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/libmesh-users
>
> --
> Vikram Garg
> Postdoctoral Associate
> Center for Computational Engineering
> Massachusetts Institute of Technology
>
> <https://lists.sourceforge.net/lists/listinfo/libmesh-users>
> <https://lists.sourceforge.net/lists/listinfo/libmesh-users>
> http://www.runforindia.org/runners/vikramg
>
------------------------------------------------------------------------------
Symantec Endpoint Protection 12 positioned as A LEADER in The Forrester
Wave(TM): Endpoint Security, Q1 2013 and "remains a good choice" in the
endpoint security space. For insight on selecting the right partner to
tackle endpoint security challenges, access the full report.
http://p.sf.net/sfu/symantec-dev2dev
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users
