issues with parallel
Situation seems to be with petsc in parallel. parallelTest.py
Summary: DG is probably fine, Lagrange with Dirichlet BCs don't work at all, Lagrange with weak is apparently also not quite correct. Even in the case where Lagrange does work the number of iterations in the Newton solver is not as expected, i.e., the Newton solver does not converge.
Some issues even with the internal solvers
-
why do Dirichlet bcs fail even with the dune-fem internal solvers on ALU? -
why is the Newton residual not small when using the internal petsc solvers with the Lagrange space?
grid | space | BCs | petsc-fem | petsc4py | numpy-fem |
---|---|---|---|---|---|
ALU | dg | N/A | works | works | no test |
Yasp | dg | N/A | works | works | no test |
ALU | lag | weak | fails^1 | fails^3 | works |
Yasp | lag | weak | works^1 | works | no test |
ALU | lag | strong | fails^1 | fails^2 | fails^4 |
Yasp | lag | strong | works^1 | fails^2 | no test |
- works means that the difference between petsc-fem and petsc4py was small
- Newton method does not converge. Residual is not zero in
iteration 1
as it should be for linear problem But the actual results seems on yasp nevertheless. - fails to converge
- the petsc4py solver converges but the difference between the two solution is 0.21
- constant residual in gmres iteration
Test script is attached run with mpiexec -np 2 parallelTest.py
.