ILU fill level being set incorrectly

[Gary Rebt]

Hi,

It seems like in Python, the option to set the fill level of ILU in the solver options for the PETSC solver such as in

prm = solver.parameters
prm["linear_solver"] = "gmres"
prm["preconditioner"] = "ilu"
prm["krylov_solver"]["preconditioner"]["ilu"]["fill_level"] = 1
solver.solve()

does not actually modify anything. It still uses a fill level of 0. However, when modifying directly the PETSc options, so inserting

PETScOptions.set('pc_factor_levels', 1)

the ILU fill level is indeed set to 1 as it should.

The effect can be highlighted by looking at the number of iterations performed by the solver which has to be lower when the fill level is set to 0 than for a fill level of 1.

[Anders Logg (Chalmers)]

Removing milestone: 1.7 (automated comment)

[Prof Garth Wells]

I suggest that we remove the parameter. Getting into controlling solver details from the parameter system is a recipe for inconsistencies, especially across different backends.

[Prof Garth Wells]

Best controlled directly via backend.