KineticForces - BUGFIX! - Drop the spurious major radius from the omega_D prefactor

[logan-nc]

Release note

• Audience: users
• Numerical impact: omega_D changes by a factor of R0, moving NTV torque and delta-W on any machine with R0 != 1 m (diiid_n1: torque +4.55%, dW +15.85%; the other 45 quantities are exactly unchanged); Solovev fixtures are unaffected only because they set r0 = 1 (harness @ 2e71e97)
• Migration: Re-baseline saved NTV torque and kinetic-energy comparisons — the precession frequency was previously too large by a factor of R0.

The general-aspect-ratio precession prefactor carried an extra factor of the major radius, leaving omega_D with units of m/s rather than a frequency. Torque and kinetic-energy results move on any machine whose major radius is not 1 m; the overall sign convention is a separate question and is not touched here.

---

What

One factor in the general-aspect-ratio precession prefactor (src/KineticForces/BounceAveraging.jl):

-    wdbar = ro^2 * bo * wdfac * wbbar * 2 * (2 - sigma) * total_wd
+    wdbar = ro * bo * wdfac * wbbar * 2 * (2 - sigma) * total_wd

Julia analog of Fortran GPEC PR #281.

Why

wbbar = ro·2π/((2−σ)·I₁) already carries one factor of ro, which its own normalization bhat = √(2T/m)/ro cancels. Reusing wbbar inside wdbar imports that ro a third time, while dhat = (T/q)/(B₀·ro²) removes only the two written explicitly.

Dimensionally: both bounce integrals carry the J·b·dθ = dl length, so I₁ is a length and I₂/I₁ is 1/Wb. With T/q in volts and V/Wb = 1/s, the prefactor 4π·wdfac·(I₂/I₁)·(T/q) is already a frequency — the surviving ro left omega_D in m/s.

Verification

Holding the physics fixed and varying only the machine size (both bounce integrals scale as the length they carry, so I₂/I₁ is ro-independent):

	omega_D ratio when ro is doubled
before (ro^2)	2.0	still carries a length
after (ro^1)	1.0	frequency, ro-invariant

omega_b, built from the same I₁, is untouched and keeps its correct v_th/L scaling (ratio 0.5) — that is the control showing the change is confined to the precession prefactor.

Upstream measured the same correction against an independently validated guiding-centre precession operator: the least-squares slope of omega_D against the reference moves from -6.494 (median over 7 surfaces; -ro = -6.411 m on that ITER equilibrium) to -1.021, with omega_b agreeing to within [0.9964, 1.0011] both before and after.

Impact — this is not a small change

omega_D changes by a factor of ro on every machine (≈6.4× on ITER scale). It feeds the resonance denominator ℓ_eff·ω_b·√x + n·(ω_E + ω_D·x), so all NTV torque and kinetic-δW results move. Every kinetic regression case will need re-baselining; that is expected, not a regression.

Only the magnitude is addressed here. The overall sign of the precession is a separate convention question (it depends on whether P_φ = ±q·ψ_p, which cannot be read off from the native chart) and is deliberately not touched.

Validation

• runtests_kinetic.jl: 277/277. Not sufficient alone — those unit tests do not exercise the pitch-integration / compute_bounce_data paths.
• Regression harness: pending. Draft until the kinetic cases are run and the new baselines are recorded with the shift explained.

🤖 Generated with Claude Code

[logan-nc]

Ready for review; not yet ready to merge.

The physics/code argument is complete and self-contained, so review can start now. The regression-harness table is still outstanding and is a merge gate per CLAUDE.md — I'll post it here before this is merged.

Note the machine currently has other regression runs in flight, so the kinetic cases are queued behind the PR #339 re-baseline rather than run in parallel.

[logan-nc]

Regression harness — develop (b3abe074) vs this branch

regress --cases diiid_n1,solovev_kinetic_ntv,solovev_kinetic_nuzero,solovev_kinetic_calculated --refs develop,bugfix/kf-omega-d-major-radius

Case	R₀	Result
diiid_n1	1.743 m	2 changed (both NTV), 46 unchanged
solovev_kinetic_ntv	1.0 m	6 unchanged
solovev_kinetic_nuzero	1.0 m	5 changed, all ~1e-7 (FP noise)
solovev_kinetic_calculated	1.0 m	6 changed, all ~1e-7 (FP noise)

The only physically meaningful movement in the whole suite:

diiid_n1                          develop         branch          Diff
NTV torque FGAR [N·m]             5.591200e-01    5.336824e-01    2.544e-02 (4.55%)
NTV kinetic energy dW FGAR [J]    6.909782e-02    7.997363e-02    1.088e-02 (15.74%)

All 46 other diiid_n1 quantities — energies, q profile, singular surfaces, Δ′, island widths, ODE steps — are exactly unchanged, which is the containment check: this patch moves omega_D and nothing else. The response is nonlinear in omega_D (it enters the resonance denominator ℓ_eff·ω_b·√x + n·(ω_E + ω_D·x)), so a 1.74× change in the frequency gives a few-percent shift in the integrated torque rather than a proportional one.

⚠️ The regression suite is nearly blind to this bug

The Solovev fixtures set r0 = 1.0 exactly (examples/Solovev_kinetic_calculated_example/gpec.toml:73). With ro = 1, ro² ≡ ro, so this fix is an algebraic no-op on every Solovev case — the ~1e-7 movements above are floating-point reassociation from writing ro * instead of ro^2 *, nothing more.

That means:

• diiid_n1 (R₀ = 1.743 m) is the only tracked case that exercises this at all, and only through its two NTV quantities.
• A unit-magnitude major radius in the primary kinetic fixtures hides this entire class of dimensional error. Upstream needed an ITER case (R₀ = 6.411 m) to see it as a 6.4× discrepancy; at R₀ = 1.74 m it is a 1.74× error in omega_D, and at R₀ = 1.0 m it is invisible.

Worth considering as a follow-up: either a Solovev kinetic variant with r0 ≠ 1 (cheap — the analytic equilibrium takes r0 directly), or a note in the case TOMLs that r0 = 1 makes them insensitive to major-radius scaling errors. I have not changed the fixtures in this PR, since re-baselining them here would obscure the containment result above.

Validation summary

• runtests_kinetic.jl 277/277.
• Dimensional argument verified independently of upstream: holding physics fixed and doubling ro, omega_D scales as 2.0 before the fix (i.e. carries a length) and 1.0 after; omega_b, built from the same bounce integral, keeps its correct v_th/L scaling as the control.

[github-actions]

This pull request is missing a reviewer.

If you are not ready to name them, mark this pull request as a draft.
docs/development/contributors.md suggests lead developers to ask.
Merging is not blocked here, but no pull request may be merged without human review.

[logan-nc]

Regression re-run after re-syncing with develop

PR #339 merged, so this branch was updated to develop@c42d558e and the harness re-run at the new head. Baseline and branch now both contain the multi-ion work.

diiid_n1	Summary: 2 changed, 45 unchanged
solovev_kinetic_ntv	Summary: 6 unchanged
solovev_kinetic_nuzero	Summary: 6 changed, 8 unchanged
solovev_kinetic_calculated	Summary: 7 changed, 7 unchanged

Every changed row on the two Solovev kinetic cases is at 0.00% relative (absolute diffs ~1e-6 to 1e-8 on O(1) eigenvalues, plus one-step ODE count moves) — they cross the harness's tight 1e-10 eigenvalue threshold but are not physically meaningful.

The one substantive movement is on diiid_n1, and it reproduces the earlier measurement almost exactly even though the baseline moved 56+ commits:

NTV torque FGAR [N·m]            5.549513e-01 -> 5.296763e-01   (4.55%)
NTV kinetic energy dW FGAR [J]   6.840635e-02 -> 7.924971e-02  (15.85%)

The other 45 diiid_n1 quantities are exactly unchanged, which is the containment check: this patch moves omega_D and nothing else.