Repository navigation
Repo - FEATURE - Finish Fortran GPEC 1.6.0 parity: supplied collision rates, overlap check and quadrature accounting #624
Description
These are the #591 items that remain after the v2.0 parity PRs.
- A9,
kinetic_format = "rates-v1"(Fortran PR 345). This is an input capability, not a fix.- In Fortran, a 9-column kinetic table supplies lnΛ, ν_i and ν_e (s⁻¹), which replace the internal collision model. The values are log-interpolated with pchip and must be positive.
- In Julia,
_read_kinetic_tablehardcodes 6 columns. The work needs akinetic_formatkey next tokinetic_groupin[Equilibrium], and a branch for supplied rates inkinetic_profile_splinesandresolve_ntv_species.
- A13, the singular-coupling overlap (Fortran PRs 330 and 185). Fortran now exports W^H·V, so that the overlap is V^H·W·b. Julia exports a dominant-mode overlap V^H·b̃ in root-area-weighted units (
ResonantCoupling.jl). Check whether the two are equivalent; Julia has no per-model or local-window SVD. - Inner quadrature accounting (analogue of Fortran PRs 312, 314 and 334). The pitch and energy integrals discard their error estimates (
PitchIntegration.jl,EnergyIntegration.jl), socheck_psi_quadrature_convergencesees only the outer ψ error. v2.0 adds a hard stop on a non-finite torque; this item adds reporting of inner-integral convergence. - Docs pass (Fortran PRs 282 and 266): mirror the PENTRC rotation and harmonic sign statements in
docs/src/conventions.md. - A15: check the FastInterpolations endpoint derivative under a large constant offset (Fortran PR 308). If it shows the same cancellation, file it upstream.
Activity
logan-nc commentedon Nov 2, 2026
Copied from #627 for the record: the #624 items that #627 closes without code changes, and why.
#624 items closed without code
C_resonant_area_weighted_field, which acts on b̃ (the root-area-weighted field), and the overlap V^H·b̃ uses that same basis.test/runtests_dominant_coupling.jl:97already checks thatU * (σ .* overlap)reproduces the solve's resonant field to 1e-8. That is the same check Fortran's deleted test script ran.forcing_overlapcarries units (T), which was PR 185's other point.cubic_interp+deriv1on x³ − x/2 plus a constant offset gives end-slope errors of 5e-8 at an offset of 1e6 and 1e-3 at 1e10. The unavoidable limit, machine epsilon × offset / grid spacing, is about 2e-4 at 1e10, and Julia stays within a few times it. There is no cancellation like the one Fortran fixed.