[petsc-users] convergence problem- 3D Cahn Hillard

[Matthew Knepley]

On Tue, Jan 5, 2021 at 3:14 PM Sepideh Kavousi <skavou1 at lsu.edu> wrote:

> Dear Petsc Users
> I wrote a code in 2D solving PF equations for solidification of alloys. I
> have two variables in the equations 1-non-dimensional concentration (U) 2-
> order parameter (p). Everything is fine with this code.
> When I change the code to 3D, the code has convergence problems which I
> cannot solve it.
> I know the term which is causing problem is a term in the equation for U:
> (dp/dt)*(Pxx+Pyy+Pzz)
> where Pxx,Pyy,Pzz are the second derivative of order parameter in x,y,z,
> respectively.
>
> When I eliminate this term, the results of 2D and 3D are close to each
> other. But adding that term the time step should be reduced a lot to obtain
> convergence, otherwise the timestep does not converge. I have no idea how I
> should solve this problem.
>  I solve the equations considering (I chose based on suggestions of Moose
> for solving Cahn-hillard)
>  -ts_type bdf -ts_bdf_adapt -pc_type bjacobi -snes_linesearch_type l2
> -snes_type newtontr -ksp_type gmres -ksp_gmres_restart 1001 -sub_pc_type
> ilu -sub_ksp_type preonly
>
> Would you please help me solve this issue?
>

In order to see the solver you have we need the output of

  -ts_view

and then to see what is happening during the run we need

  -ts_monitor -snes_monitor -snes_linesearch_monitor
-ksp_monitor_true_residual

  Thanks,

     Matt


> Best,
> sepideh
>
>

-- 
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

https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20210105/5861d5f7/attachment.html>