Removing DirichlteBC-dofs to compute eigenvalues

[APaganini]

Thanks to slack crowd, I've learnt that to compute eigenvalues of a differential operator it is not sufficient to incorporate DirichletBCs by just adding bcs=bc in the matrix assembly (e.g. as in assemble(m, bcs=bc).M.handle), but that instead one should also remove the dofs associated with the Dirichlet bcs. The demo https://www.firedrakeproject.org/demos/qgbasinmodes.py.html doesn't do it, though. Shall we update it? (Lawrence and Matt have recently learnt taught me how to do that).

Bonus question: In the same demo, PETS options are set to opts, but the solver includes them with the spell es.setFromOptions(), which does not take opts as an explicit input. Is this correct? (and a sort of PETSc wizardry?)

[wence-]

Thanks to slack crowd, I've learnt that to compute eigenvalues of a differential operator it is not sufficient to incorporate DirichletBCs by just adding bcs=bc in the matrix assembly (e.g. as in assemble(m, bcs=bc).M.handle), but that instead one should also remove the dofs associated with the Dirichlet bcs. The demo https://www.firedrakeproject.org/demos/qgbasinmodes.py.html doesn't do it, though. Shall we update it? (Lawrence and Matt have recently learnt taught me how to do that).

Please go ahead and do so.

Bonus question: In the same demo, PETS options are set to opts, but the solver includes them with the spell es.setFromOptions(), which does not take opts as an explicit input. Is this correct? (and a sort of PETSc wizardry?)

The opts reference the global PETSc options database. setFromOptions() inspects the options database to configure things.

[APaganini]

1. I'll write the update
2. Many thanks for claryfing that it is processed via global variables
3. Apologies for a typo. I originally wrote Lawrence and Matt have recently learnt how to do that whereas I obviously meant Lawrence and Matt have recently taught me how to do that :)

[APaganini]

The update is here: #2402