Convergence issues due to huge differences in stiffness tensor

[Bala-1005]

Hello everyone,

I am working on kks model for solidification with the two phases being solid and liquid. I have included the tensor mechanics module too similar to the one shown in the example "/moose/modules/combined/examples/phase_field-mechanics/kks_mechanics_VTS.i".

The stiffness of the liquid phase is several orders lower than the solid(~10^9). This has caused some convergence issues close to the boundary and when two solid parts of the domain interact. I am not sure how to alleviate this problem as this seems to be the only issue with the code. I have fiddled around with different boundary conditions but that has not helped at all.

How do I solve this issue?

Thanks,
Bala

[aeslaughter]

Have you tried using automatic_scaling? This can be enabled in the Executioner block with "automatic_scaling = true"

[Bala-1005]

I was able to slightly solve the issue by increasing the number of nonlinear iterations to 100. But only for several timesteps. The issue returned with convergence being the issue. The output file is very large and therefore I am attaching the part where it starts to cut timesteps and stop converging.
output.txt
error.txt

Thanks,
Bala

[jiangwen84]

Try to set solve_type = 'PJFNK'

Maybe you want to loose your convergence tolerance a bit, i.e., nl_rel_tol = 1e-7 and nl_abs_tol=1e-7

[Bala-1005]

Thank you @jiangwen84 and everyone. I was able to solve the issue by increasing the nonlinear iterations number, loosening the tolerance a bit, and changing the boundary conditions from a displacement boundary condition to pressure boundary conditions. The simulation runs quite well now.
One final problem, I hope :), is that there are a few memory issues. Any suggestions on how to solve these issues? Here is the error file. I used --malloc-dump while running the file.

el2b_KHS.txt

Thanks,
Bala

[jiangwen84]

I am not sure what is happening. Is this two dimension problem? Seems the error comes from PETSc, but not sure about the reason.

[Bala-1005]

Yes. It is a two dimension problem.
Thanks,
Bala