[WIP] Magnetostatic solver refactor and bug fixes #5175
Conversation
Signed-off-by: roelof-groenewald <[email protected]>
Signed-off-by: roelof-groenewald <[email protected]>
for more information, see https://pre-commit.ci
roelof-groenewald
added
bug
Signed-off-by: roelof-groenewald <[email protected]>
roelof-groenewald
changed the title
Signed-off-by: roelof-groenewald <[email protected]>
Signed-off-by: roelof-groenewald <[email protected]>
There was a problem hiding this comment.
This looks good @roelof-groenewald thank you for shaking this out.
| if (field_boundary_hi[0] == FieldBoundaryType::PEC) { | ||
| if (adim == 0) { | ||
| hibc[adim][0] = LinOpBCType::Neumann; | ||
| dirichlet_flag[adim][1] = false; |
There was a problem hiding this comment.
This can be changed to use an STL map data structure as in the ProjectionDivCleaner. Seesm like a good time to do this refactoring.
There was a problem hiding this comment.
Great idea! I'll add that to my tasks for the upcoming hackathon (also changing it for the current ES solver).
| @@ -249,9 +269,9 @@ void | |||
| WarpX::setVectorPotentialBC ( amrex::Vector<amrex::Array<std::unique_ptr<amrex::MultiFab>,3>>& A ) const | |||
There was a problem hiding this comment.
Should this have some sort of warning about the inaccuracy of the field near an EB surface? This is now effetively going to be a staircase EB instead of cut cell correct?
| @@ -326,7 +326,7 @@ WarpX::ProjectionCleanDivB() { | |||
| } else if ( WarpX::electromagnetic_solver_id == ElectromagneticSolverAlgo::Yee | |||
| || WarpX::electromagnetic_solver_id == ElectromagneticSolverAlgo::HybridPIC | |||
| || ( (WarpX::electrostatic_solver_id == ElectrostaticSolverAlgo::LabFrame | |||
| || WarpX::electrostatic_solver_id == ElectrostaticSolverAlgo::LabFrameElectroMagnetostatic) | |||
| || WarpX::electrostatic_solver_id == ElectrostaticSolverAlgo::LabFrameSemiImplicit) | |||
There was a problem hiding this comment.
Is the explicit version of the solver going away completely? I.e. the LabFrameElectroMagnetostatic Algo?
There was a problem hiding this comment.
No, the magnetostatic solver is just changed to function as an add-on to an electrostatic solver. So instead of defining it as it's own solver (explicit ES + MS) the type of ES solver is set in the input and the extra flag do_magnetostatic sets whether the MS solve will be added on to the ES solve.
#5446) …ch is currently hard coded, and adding assert to force EB being enabled in order to avoid a seg fault in AMReX when computing the gradient solution. This PR partially addresses #5444. This PR adds semi-coarsening in 3D and then adds an assert to keep the magnetostatic solver from being run without an EB. This is required since in AMReX MLMG->getGradSolution will segfault when not using an EB. It should also be noted that #5175 will use a different scheme around the embedded boundaries to compute gradients, and will likely mitigate these issues. A work around in RZ to use the outer edge is to enable the embedded boundary and set the boundary radius larger than the outer grid radius. This works like it would without an embedded boundary and can be used until either the refactor or the bugfix in AMReX for getGradSolution. --------- Signed-off-by: S. Eric Clark <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
This PR introduces a major refactor of the magnetostatic solver primarily with three goals:
Example of non-zero B-field divergence introduced with the current magnetostatic solver implementation.
Same simulation as above but with the fixes from this PR included (note the colorbar range)
This PR builds on and requires first merging: