I want to decompose a 3D mesh (sphere) into multiple subdomains and export the stiffness matrices / load vectors to solve the problem with an external CG solver. Unfortunately it seems that the local matrices are non-symmetric. // ff-mpirun -np 4 sphere2.edp -wg -ffddm_overlap 2 -ffddm_schwarz_met…

You can do that with PETSc quite trivially, with -ksp_type cg -pc_type asm. The interface is much more concise as well, see this example.

Here is the script using PETSc CG + ASM. You can also export the matrix and precondition it externally using the additional command line argument -mat_view ascii:Poisson. macro dimension 3 // EOM verbosity = 0; load "msh3" load "tetgen" // load "medit" // load "iovtk" include "MeshSurface.idp" …

Thank you for your script. I have run it with “ff-mpirun -n 4 sphere3.edp -mat_view ascii:Poisson” and I get a “Poisson” textfile containing a matrix description. But it seems it is only one partition? At least it says “Mat Object: 1 MPI processes”. How can I get all the submatrices? Is there a …

Is there a way to export the local matrices from the ASM as well? I would like to implement the whole AS+CG algorithm and then later compare it to the solution of FreeFem. I just need some program to do the meshing/partitioning/assembly.

Here is what you can do: replace sym = 1 by sym = 0 replace the current sparams by: set(A, sparams = "-pc_type gasm -pc_gasm_overlap 2 -pc_gasm_type basic -sub_pc_type cholesky -ksp_type cg -ksp_monitor -ksp_view -pc_gasm_view_subdomains -ksp_view_mat ascii:Poisson:ascii_matlab -ksp_view_rhs asc…

Thank you very much, that was really helpful. Can I export the restriction matrices and the decomposition of the one as well?

The restriction is the list of indices in the Outer subdomain: that you get with -ksp_view. You can define in your external solver a matrix of dimension (size of local matrix, that you get in the -ksp_view as well) x (size of global matrix, in your case 5654). Then, you just fill with ones for all l…

Exporting Matrices of AS method for Poisson Equation

General Discussion

prj July 31, 2020, 8:26am 5

You are correct, it’s the global matrix. You are free to partition this matrix how you seem appropriate in your external solver if you don’t want to use PETSc CG.

Topic		Replies	Views
Partitioning the stiffness matrix without partitioning the mesh for PETSc General Discussion	10	1289	March 27, 2021
Saving and loading global/distibuted arrays General Discussion	39	2810	August 5, 2020
Composite FE spaces + Parallel assembly with PETSc General Discussion	6	241	June 23, 2024
Parallel implementation of a problem with multiple right hand sides General Discussion	7	603	January 17, 2022
Partition matrices for a coupled problem General Discussion	9	1073	May 14, 2020

Exporting Matrices of AS method for Poisson Equation

Related topics