Add Galerkin-in-Time #104
Add Galerkin-in-Time #104
Conversation
…ct/Irksome into smaclachlan/the_other_git
good idea Co-authored-by: Pablo Brubeck <[email protected]>
pbrubeck
force-pushed
the
smaclachlan/the_other_git
branch
from
45bb847 to
5d286c9
Compare
| edofs = L_trial.entity_dofs() | ||
| trial_perm = [*edofs[0][0], *edofs[1][0], *edofs[0][1]] | ||
| trial_vals = trial_vals[trial_perm] | ||
| trial_dvals = trial_dvals[trial_perm] |
There was a problem hiding this comment.
We had purposefully put a matching permutation on the test space in earlier code in order to facilitate potential comparisons/compatibility with Rana-style preconditioners, where you'd want to field-split by stages in order by their stage times (in analogy to Gauss-Legendre RK methods). I agree that such a permutation doesn't matter (or even necessarily make sense) in many other cases, but I do think that there's some value to that. Thoughts? Is anything lost by just echoing this structure with a test_perm?
There was a problem hiding this comment.
Test space is DG, the interior dofs should be ordered already.
There was a problem hiding this comment.
FIAT orders the dofs different for continuous and discontinuous Lagrange?
There was a problem hiding this comment.
I think this is fine, since all DoFs for DG are internal, right? If so, I guess we're at the mercy of whatever FIAT gives us for their ordering, but probably we shouldn't be second-guessing that without cause
tests/test_disc_galerkin.py
Outdated
|
|
||
|
|
||
| @pytest.mark.parametrize("order", [0, 1, 2]) | ||
| @pytest.mark.parametrize("basis_type", ["Lagrange", ]) |
There was a problem hiding this comment.
We should be testing more basis types here
There was a problem hiding this comment.
I've now tested the basis_type values that we've implemented. Are there reasonable choices of variant that we should test as well?
There was a problem hiding this comment.
integral since we might be going down that road for coarsening to low-degree in time?
There was a problem hiding this comment.
integral is already there, for both DG and CG cases (this is a thread on an outdated commit before I added that)
We've worked out the UFL manipulation to implement both continuous and discontinuous Galerkin-in-time steppers. CG is, in some instances, equivalent to Gauss-Legendre (and, so symplectic), while DG is dissipative.