Re: [Libmesh-users] KSP convergence reason - DIVERGED_INDEFINITE_PC

[Renato P. <re...@gm...>]

Hi Roy,

I am still not able to get it to work.
I tried matrix ordering, varied solver and PC types.

I would think the problem is when the constrains are applied...
The final matrix is probably not nice.
How can I debug this?

Renato

On Fri, Mar 15, 2019 at 3:36 PM Renato Poli <re...@gm...> wrote:

> Thanks.
>
> Do you see any sense in adding a SCALAR variable and "add_constrain_row"
> to it instead of using an existing DOF as reference?
> If that works, it would be nicer from the coding perspective.
> Would it be numerically more stable?
>
> Renato
>
> On Thu, Mar 14, 2019 at 6:04 PM Stogner, Roy H <roy...@ic...>
> wrote:
>
>>
>> On Thu, 14 Mar 2019, Renato Poli wrote:
>>
>> > How do I change the dof ordering options?
>>
>> "-pc_factor_mat_ordering_type string" with a string from
>>
>> https://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/Mat/MatOrderingType.html#MatOrderingType
>>
>> >       Add a SCALAR variable to use as a Lagrange multiplier?
>> >
>> > This sounds like a much cleaner solution.
>> > I could not understand how to do that in libmesh so far.
>> > It is probably time to put some more effort here...
>> >
>> > I see I must add a Kp_alpha matrix. Do I need the Kalpha_p as well?
>>
>> Yes; the Kalpha_p ends up being what looks most like the actual
>> constraint equation.
>>
>> ...
>>
>> Except you don't have a constraint *equation* here, you have
>> constraint *equations*.  You wouldn't want a SCALAR lagrange
>> multiplier, you'd want to add a LAGRANGE variable subdomain-restricted
>> to boundary elements along the constrained domain boundary... this is
>> sounding like more trouble than it's worth unless you can't get the
>> directly constrained solves to converge.
>> ---
>> Roy
>>
>