Repository navigation
ForceFreeStates — multi-n support in build_kinetic_metric_matrices #246
Description
ForceFreeStates.build_kinetic_metric_matrices (src/ForceFreeStates/Fourfit.jl:151-315) hardcodes n = intr.nlow at the inner-product loop (line 285) and allocates only mpert² storage. Fortran pentrc/dcon_interface.f:863-1020 (idcon_action_matrices) is intrinsically single-n (the toroidal mode is a module-scope nn, no outer-n loop, matrices sized mpert²).
PR #112 adds @assert intr.nlow == intr.nhigh to make this loud. Lifting the restriction would be a Julia-only extension beyond Fortran and would require:
- Allocating
(mpert·npert)²storage for each ofsmats/tmats/xmats/ymats/zmats. - Wrapping the inner-product loop in
for n in intr.nlow:intr.nhigh. - Updating
CalculatedKineticMatrices.compute_calculated_kinetic_matricesto consume the block-diagonal layout (it currently rebuilds the block diagonal manually).
Acceptance
npert > 1runs no longer assert.- DIIID multi-n regression case added under
regression-harness/cases/showing finite et[1] for at least n=1,2,3. - Reference: Logan PhD thesis (2015), Appendix C, Eqs. C.1-C.11.
Activity
logan-nc commentedon Oct 27, 2026
Checked against develop (db90e82) before deciding whether to open a PR tonight.
build_kinetic_metric_matricesis still single-n:Fourfit.jl:134-135setsn = nlowand the five matrices arempert²per surface; the assert from #112 is gone, so a multi-n deck withkinetic_factor > 0silently builds every kinetic matrix atnlow(callers:CalculatedKineticMatrices.jl:90,KineticForcesStructs.jl:388). Not opening the PR now: #569 rewritesCalculatedKineticMatrices(thecompute_kinetic_matrices_at_psi!signature and the block-diagonal consumer this would have to change), and the acceptance criterion needs a new DIII-D multi-n kinetic harness case, which is hours of compute I am keeping off the shared machine this week. Plan: after #569 lands, first commit restores a loud error fornlow != nhighon the kinetic path, second lifts it with the(mpert·npert)²layout and the harness case.