Composite multilevel solve with single-cell direction (quasi-2D AMR)

[desjardi]

I'm running 3D AMR simulations where one direction has a single cell in the domain and the AMR refinement ratio is 1 in that direction. I would like to perform composite multilevel solves for proper coarse-fine flux coupling. Neither setHiddenDirection nor setSemicoarsening currently supports this:

• Hidden direction correctly adjusts mg_coarsen_ratio_v, stencils, and MG hierarchy for the single-cell direction, but asserts m_num_amr_levels == 1 in MLLinOp::define.
• Semi-coarsening allows multi-level, but the fine AMR level loop still uses mg_coarsen_ratio_v = (2,2,2) and hits "Uncoarsenable domain" because Box::coarsenable(IntVect(2,2,2)) fails when a direction has only 1 cell.

Since the refinement ratio is 1 in the single-cell direction, all AMR levels share the same extent there, and no inter-level coarsening is needed. Would it be possible to either:

1. Relax the hidden-direction assertion to allow composite multilevel solves, or
2. Adjust mg_coarsen_ratio_v in the semi-coarsening path for directions where the domain has only 1 cell?

Happy to test and contribute a PR if either approach is acceptable.

[WeiqunZhang]

Yes, I think it makes sense to relax the hidden-direction assertion to allow for composite solves. If you can contribute a PR, it will be greatly appreciated.

[desjardi]

Thanks! I just submitted a PR.