Skip to content

Route 2 (PARKED): CT/staggered B at the 2:1 AMR seam — activates only if a workload needs it #30

Description

@szaghi

Route 2 (PARKED): CT/staggered B at the 2:1 AMR seam — activates only if a workload needs it

Status: PARKED. Do not implement speculatively. This issue activates ONLY if a production workload exceeds the [IO].seam_divB_tol guard-rail (issue #29) on a run where the O(h^p) seam div(B) floor actually corrupts the physics.

Why parked

Issue #29 established (with five fix classes ruled out) that PRISM's collocated high-order FD path has no local div(B) fix at a 2:1 AMR seam — the mimetic div_h(curl_h)≡0 identity breaks at the resolution jump (Ranocha commutation), and no value-transfer restores operator commutation (the R∘P≠I wall). The resolution adopted is accept-truncation: the seam source is O(h^p) refinement-convergent, unbounded only in t at fixed h, and monitored by the seam_divB_tol guard-rail.

Route 2 is the only remaining route to an exact seam div(B) bound, but it is a large structural change against PRISM's collocated-FD core design, so it is not justified unless a real workload demands it.

What Route 2 would entail (if activated)

Re-collocate B to a constrained-transport (CT) / staggered representation at the seam (interior stays collocated-FD):

Activation trigger

Open/prioritize this issue when: a production AMR run trips [IO].seam_divB_error (or the warn-only monitor shows the floor corrupting a physics observable) AND the mitigations are insufficient — refining h is too costly, shortening the run defeats the science, and the GLM delay (hyperbolic + [physics].c_r, ~5.5× margin) doesn't buy enough runway.

References

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions