You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Equilibrium - FEATURE - Add the opt-in profile_source integrate option that builds F and p from FF' and p' #628
Fortran GPEC 1.6.0 added profile_source = "integrate" (PrincetonUniversity/GPEC#297). It builds F and p in the profile spline by integrating FF′ and p′ when the file's F and p are inconsistent with them. Julia reads ffprime/pprime (Equilibrium/ReadEquilibrium.jl) but never uses them.
#506 tried this and was closed. It made the derivative path the default, which moved every g-file and IMAS result: Δ′ by up to 15 %, NTV by 2.8 %, SLAYER γ by 6 %. #596 instead warns when a g-file's FPOL is single precision. Fortran's option is opt-in: the default still uses the tabulated values.
Scope: an opt-in profile_source key in [Equilibrium] with the default unchanged, plus a diagnostic of the F/p inconsistency. No harness row should move.
Feature-parity item from the Fortran GPEC 1.6.0 survey in #591; filed so it is tracked once #591 closes.
Closing as a deliberate decision, not a gap. #506 investigated building F and p from FF′ and p′ (results in #504), and d-burg concluded the integration itself is unreliable: it moves surfaces near pedestal kinks by more than the Δ′ noise it removes, and the cubic, trapezoid and Akima quadrature rules disagree there by several units. The noise comes from single-precision g-files; OFT's double-precision writer removes it, and #596 warns when a g-file's FPOL is single precision. Julia keeps the tabulated F and p, which is also Fortran's default. Fortran 1.6.0's opt-in profile_source = "integrate" is intentionally not ported.
Activity
logan-nc commentedon Nov 2, 2026
Closing as a deliberate decision, not a gap. #506 investigated building F and p from FF′ and p′ (results in #504), and d-burg concluded the integration itself is unreliable: it moves surfaces near pedestal kinks by more than the Δ′ noise it removes, and the cubic, trapezoid and Akima quadrature rules disagree there by several units. The noise comes from single-precision g-files; OFT's double-precision writer removes it, and #596 warns when a g-file's FPOL is single precision. Julia keeps the tabulated F and p, which is also Fortran's default. Fortran 1.6.0's opt-in
profile_source = "integrate"is intentionally not ported.