Ferrite-FEM/Ferrite.jl

AMR Roadmap

Open

#1,456 opened on Aug 5, 2026

 (0 comments) (0 reactions) (0 assignees)Julia (109 forks)github user discovery
AMRhelp wanted

Repository metrics

Stars
 (439 stars)
PR merge metrics
 (Avg merge 15d 5h) (17 merged PRs in 30d)

Description

I think it's good to have a AMR roadmap for the various open topics left. The ordering is driven by dependencies and API instead of importance. Happy to discuss and reorder. (Already discussed with the lazy guy @claude)

Roadmap

  1. #1413 — ForestBWG as the grid (lazy materialization facade, drop NonConformingGrid) First because every later item touches the boundary it reshapes: where conformity_info lives, what materialization emits, and the cell-numbering lifecycle / reclose!. Landing it last would mean building those against NonConformingGrid + explicit creategrid and churning them afterwards. It's also the smallest of the five.
  2. #1411 — Transient adaptivity: old-to-new cell id map and solution transfer Biggest user-facing unlock (time-dependent problems), and deliberately scoped to linear Lagrange, so it needs nothing from #1409/#1410. It composes naturally with the facade: the epoch bump in refine!/coarsen! is exactly the moment to emit the old-to-new map, and reclose! is its natural consumer.
  3. #1410 — Curved geometry (higher-order geometry interpolation of the macro mesh) Before #1409 because the degree-scaled node lattice and degree-parametric Lnodes materialization built here are shared infrastructure: once node identity lives on the [0, n·2^b] lattice, the entity-dof bookkeeping that higher-order conformity constraints need largely falls out of the same element-node matrix. (At least if we go the full Lnodes path)
  4. #1409 — Conformity constraints beyond linear Lagrange
  5. #1455 — Generalized Morton SFC (simplices, wedges, general element types) Last regardless of the rest of the ordering: it reshapes the Morton core itself, so anything in flight when it lands would need rework. Everything above is hypercube-only by design.

Notes

  • #1410 + #1409 together are what deliver sub-/iso-/super-parametric adaptivity — geometry order and field order are decoupled, so they're two halves of one capability with no cross-terms.
  • The other open issues with the AMR tag are more "convenience" issues and aren't blocking any of the above.

Contributor guide