Repository navigation
InnerLayer.SLAYER - API - Choose one default radial label for the Fitzpatrick layer formalism #417
Description
Activity
Reconciliation of the two evidence sets, after cross-checking where each was measured — this sharpens the decision this issue should make.
The two validations sampled different regimes and do not contradict each other:
- The TJ cross-validation's "flux-type labels reproduce TJ's label to <= 1.2 percent" was measured on the circular TJ eps = 0.15 benchmark (K^(2mu) conversion factors at the q = 2 and q = 3 surfaces: halfwidth 0.38 percent, fsa 0.46 percent, volume 0.30 percent off TJ's reference, vs midplane 1.9 percent off). On a circular equilibrium all flux-type labels coincide to O(eps^2), so this certifies the family but cannot discriminate within it.
- The layer-overlap investigation's per-surface label table on the shaped DIII-D-like deck is the discriminator:
:fsatracks the Eq. (30):fluxlabel to ~1 percent at psi = 0.5, degrading to ~12 percent by psi = 0.99, while:halfwidthsits ~16 percent below:fluxin the core — on shaped equilibria:halfwidthis a genuinely different label, not an approximation to Eq. (30).
Consequence — this issue should make two decisions, not one:
- Edge overlap criterion (layer widths near the separatrix): no freedom. Eq. (100) is non-covariant and the geometric labels lose their Jacobian approaching the separatrix (they converge to the separatrix shape, so da/dpsi collapses and surfaces past psi ~ 0.99 cannot be scored at all). The criterion must be evaluated in the Eq. (30) flux label. This part is effectively settled and the overlap-cap PR implements it that way internally.
- Interior Delta-prime conversion / layer-stack default: a convention choice with real but bounded spread (labels agree to ~1-5 percent in r at interior rational surfaces on the shaped deck; ~ +/- 3 percent in the 2/1 growth rate).
:fluxremains the theory-anchored candidate for consistency with (1), but flipping it is the results-moving!-tagged change described above, and:fsais the closest geometric stand-in if a geometric label is ever preferred for diagnostic comparison.
One correction to the caveats section: the earlier statement that
:halfwidthis "the closest stand-in for the circular flux label" holds only on circular equilibria; on shaped decks:fsais the closest geometric approximation to Eq. (30) and:halfwidthshould be treated as its own convention (shift-free midplane half-chord), useful for comparison with midplane diagnostics rather than as a flux-label proxy.
Question
SLAYER's entire layer stack —
rs, the r-based shearsval_r,S(Lundquist),tau_R,tau_E,W_d, the Delta-prime reference-length factork_ref(PR #403), and the new resistive-layer-width overlap criterion (layer-overlap psihigh cap, PR to follow) — is expressed in a radial labelrselected by the programmaticrs_methodoption. The default is:midplane(outboard axis-to-edge chord), kept for continuity. Should the default move to the flux label?Evidence, from two independent investigations
Non-covariance (layer-overlap work). Fitzpatrick's diffusive-resistive width, Nucl. Fusion 2025 Eq. (100), scales as
(q/|dq/dr|)^(1/2)and picks upJ^(1/2)under a change of radial variable — the formula is anchored to the coordinate it was derived in, which is the Eq. (30) flux labelr = sqrt(2*psi_t/B0). Evaluated on the DIII-D-like deck the label choice moves the overlap cut by one rational surface (:midplane0.99752 vs:flux0.99939). The flux label reduces to the geometric minor radius on a circular LAR equilibrium (r/a = 0.99–1.05) and supplies a systematic shape correction on the shaped deck (r/a = 1.20–1.45), so it is shape-general by construction. The same non-covariance argument applies to the layer dispersion that the r_s-referenced Delta-prime feeds.TJ cross-validation (Delta-prime convention work, PR #403). In the four-arm radial-label comparison on the shaped DIII-D-like equilibrium, the flux-type labels (
:halfwidth,:fsa,:volume) reproduce TJ's circular flux label to <= 1.2 percent while:midplaneis the outlier (the Shafranov shift compresses the outboard chord). The 2/1 growth rate moves about +/- 3 percent across labels at defaults, growing to O(1) sensitivity at q >~ 4 on shaped surfaces.Both lines of evidence, obtained independently, agree: the geometric midplane chord is the odd label out for Fitzpatrick-formalism quantities, and the Eq. (30) flux label is the theoretically anchored choice.
Why this is not being flipped inside either PR
Changing the default moves
rs,sval_r,lu,tau_Rand the Delta-prime normalization everywhere at once — a results-moving change that needs the regression harness, a!-tagged PR of its own, and its own argument. Both in-flight PRs therefore keep:midplaneas the default and scope their label usage explicitly (PR #403 exposes the labels programmatically only, not via TOML; the overlap-cap work evaluates its criterion in:fluxinternally).Caveats on record
:fsaand other geometric labels lose their Jacobian near a separatrix (da/dpsicollapses as the surfaces converge to the separatrix shape); this is edge-specific — at interior rational surfaces the geometric labels are well-behaved and validated.dr/dpsi ~ qkeeps growing, which is what the width conversions need.Decision needed
After PR #403 and the layer-overlap cap PR land: pick the long-term default (
:fluxbeing the theory-anchored candidate), run the harness on the flip, and ship it as a single!-tagged change.