EQUIL - BUG FIX - Integrate equilibrium 0D flux integrals out to the boundary

[logan-nc]

Bug

The 0D flux-surface integrals behind betat, betap*, betaj, li1/2/3, and volume integrated only up to the last grid point (psihigh) and dropped the edge slice out to the plasma boundary ψ=1 that Fortran equil_out_global includes. On the Fortran-matched DIII-D grid (psihigh=0.993) this made volume/li2/li3 ~1.5–1.8% low and betat/betan ~1.5% high.

Fix (minimal)

Replace the four trapezoidal sums with the native FastInterpolations cubic integral (already used elsewhere in this file) plus Fortran's linear edge slice to ψ=1 — a small local fsi closure, no new module-level machinery:

dpsi_b =1.0- xs[end]
functionfsi(y)
s =cubic_interp(xs, y; extrap=ExtendExtrap())
return FastInterpolations.integrate(s, xs[1], xs[end]) + (y[end] +deriv1(s)(xs[end]) * dpsi_b /2) * dpsi_b
end

Verified against the Fortran DIII-D KEFIT dcon.out (matched hamada/ldp grid, psilow=1e-4, psihigh=0.993, mpsi=128, mtheta=256): volume 0.02%, betat 0.015%, li1 0.09%, li2/li3 ~0.5%, betan 0.34%. All remaining residual is a separate, pre-existing ~0.3% crnt edge-geometry difference (every leftover is a crnt/bwall-dependent quantity), out of scope.

Regression

Tracks the previously-untracked 0D params (li1/2/3, betap1/2/3, betaj, volume, crnt, bt0, bwall, aratio, kappa) in diiid_n1. regress --cases diiid_n1 --refs develop,local: 10 changed, 36 unchanged — only the flux-integration params move (≤0.12% on the shipped example, whose psihigh=0.9995 shrinks the edge slice); all energies, q/pressure checksums, singular surfaces, Δ', and PE outputs are bit-identical.

Also

Adds a short "Minimal-change discipline" note to CLAUDE.md (reuse native ops/existing utilities before writing new ones; size the change to the problem; no throwaway benchmark/agent-memory artifacts for minor fixes).

🤖 Generated with Claude Code

[logan-nc]

@parkjk who found funny li values? I was only able to see ~1% differences with fortran and here I fix them to be <1%. I don't see any way the minor differences would have made "crazy" li unless the psilow or psihigh was 0.5 or something unusual like that. If your group is still or is ever finding odd results - be sure to post an issue so we can fix them!

[jmlmir369]

@logan-nc
I was the one who previously observed the odd li3 values. Upon checking, it seems the issue was due to my using an older version of GPEC. When I pulled the develop branch for ballooning on April 23, the li3 calculation was indeed incorrect, but it has since been resolved in the main branch. Thank you for benchmarking this