Re: [Libmesh-users] new libmeshdm fails with newtonrsls

[Ata M. <a.m...@gm...>]

Hi Roy,

You can simply regenerate the error if you pass:

./example-opt --verbose dim=1 N=1024 initial_state=strip initial_center=0.5 initial_width=0.1 dt=1e-10 max_time=.5e-8 -snes_type vinewtonssls --use-petsc-dm

This generates a seg fault on my mac with the PETSc and libMesh revisions I provided before.

My guess is the seg fault happens probably because the solution vector is not initiated correctly. So the VecDuplicate generates an error at

>> src/snes/impls/vi/vi.c: 383


Ata

> On May 3, 2018, at 11:29 AM, Roy Stogner <roy...@ic...> wrote:
> 
> 
> On Mon, 30 Apr 2018, gmail wrote:
> 
>> The new implementation of the libMeshDM unfortunately fails when
>> using variational inequality solvers VINEWTONRSLS or VINEWTONSSLS
>> (for example miscellaneous/ex7). This happens when PETSc tries to
>> VecDuplicate the solution vector for lower bound in
>> src/snes/impls/vi/vi.c: 383 ?!
>> 
>> I pulled the latest libMesh (48bd50d4a) this morning and compiled it
>> with PETSc 3.8 (756c7f9f8).  I’m attaching my config log for
>> libMesh.
> 
> Sorry, the config log doesn't go through on the mailing list.
> 
> Could you either try bisecting the problem, or give us instructions
> (run misc ex7 manually with different command line options?) for
> replicating it?
> 
> Thanks,
> ---
> Roy