Use of parallel computing (PETSc) in plastic model (MFront)

[Laxman]

Hi everyone,
I am trying to simulate rainfall-induced landslide using the Richards mechanics process and successful doing so in the 2D model and now I am trying to replicate all the procedures that I have performed on the 2D model to the 3D using high-performance computing (PETSc). But I have encountered a problem with constitutive material properties’ definition. AFIK, there are two ways to define material properties in OpenGeoSys.

1. The first one is by defining material ID in the constitutive equation definition stage at the initial part of the project file. But this does not work with MFront if two or more material layers present, even on the 2D model and non-parallel model. Actually, this is a known issue raised in GitLab several months ago but it is not solved yet. (Issue: https://gitlab.opengeosys.org/ogs/ogs/-/issues/3063)
   

   

2. The second method is by defining constitutive parameter by using group_id_properties (which I have used in the 2D model). This works pretty well in 2D and non_parallel models until you are using the same constitutive equation for all layers but have a problem with the parallel model. Actually, it does not give any error while running just the simulation keeps showing running but nothing happens. This problem happens not only while using MFront but also with the constitutive equation native to the OpenGeoSys.
   

   

Does anyone has encountered this problem before or know the way around it? I know parallel FEM scheme with PETSc cannot be used by all processes using linear and quadratic function spaces simultaneously. I am using a linear approximation for both displacement and pressure in this case.

Way to reproduce the problem

1. Compile OGS enabling MFront and PETSc.
2. Run OGS with Trial_1.prj to reproduce problems related to point 1 and Trial_2.prj to reproduce problems related to point 2.

Input files:
3D_Trial_Model.zip (147.8 KB)

Thanks

[Rui_Feng]

Dear Laxman,

Have you solved this problem? I would like to use PETSc to deal with a THM problem, but I noticed the parallel FEM scheme with PETSc cannot be used by the all processes using linear and quadratic function spaces simultaneously including ThermoHydroMechanics. Do you have some methods to overcome this limit?

Best,
Rui

[joergbuchwald]

Hi @Rui_Feng it is not a general process thing. As Laxman wrote, he used linear elements for displacement, pressure and temperature. That is the workaround. The main thing that is affected is the convergence, which is worse in low permeable media. Possible accuracy issues can be overcome by better spatial discretizations.
There is already an MR (Enable parallel computing in TRM with Taylor-Hood elements (!3920) · Merge requests · ogs / ogs · GitLab) that you could try out as well if you really need a mixed-element scheme which is dealing with Taylor-Hood elements and PETSC at least for TRM.