[matt-pharr]

Closes #613

What this does

The plasma response P = Λ·L⁻¹ took Λ from the run's δW, with its wall, and the surface inductance L from free space, so a run with a wall silently mixed two vacuum models. Now:

• plasma_response errors on a force-free-states solve with a wall;
• parse_deck stops on a deck with both [PerturbedEquilibrium] and [Wall].

Both error messages give the fix: solve without a wall.

The no-wall equilibrium

The Solovev decks relied on a conformal wall, because their n = 1 kink is unstable without one, and they cannot be re-tuned. A scan over q₀ ∈ [1.9, 5] and κ ∈ [1.0, 1.6] found et[1] < 0 everywhere. Solovev's pressure is linear in ψ, so the edge current is finite and drives a peeling/external kink.

Perturbed equilibrium now runs on eq_type = "tj_analytic", with no wall: R₀ = 10, a = 1, B₀ = 1, q₀ = 1.6, q_a = 3.2, p₀ = 0.1. It is stable (et[1] = 0.435) and keeps the 2/1 and 3/1 surfaces. psihigh = 0.99 is pinned, because et[1] moves by +0.502 at 0.995. The 3/1 surface sits at ψ_N = 0.955, outside the 0.9 core window, so the tests that use that window account for it.

• New: TJ_ideal_example, and tj_n1 for it in the harness.
• Moved: Solovev_kinetic_{NTV,multiion,calculated}_example become TJ_kinetic_*, with kinetic.dat and forcing.dat rebuilt for the new equilibrium. Their harness cases are renamed to tj_kinetic_*.
• Stability only: Solovev_ideal_example keeps its wall and drops PE. solovev_n1 is unchanged.
• Tests: the PE tests run on two new fixtures, regression_tj_ideal_example and regression_tj_kinetic_calculated.

Two harness cases also needed fixing:

• solovev_kinetic_nuzero reused a deck that moved. It now runs the nuzero test fixture, widened by overrides to the same band and ion mix, and reproduces the old deck's values exactly.
• lar_ggj_tearing_n1 never produced a result, because its deck writes gpec_lar_ideal.h5. It now overrides the output name.

Release note

• Audience: users
• Numerical impact: the PE and kinetic examples move to a new equilibrium, so their values are new under tj_*; every case that kept its deck is unchanged (harness @ f388c79)
• Migration: a perturbed equilibrium needs a solve without a wall. Remove [Wall] from any deck with [PerturbedEquilibrium], or pass wall=nothing to solve. Solovev_kinetic_*_example are now TJ_kinetic_*_example, and Solovev_ideal_example no longer runs PE.

A perturbed-equilibrium calculation no longer silently combines a with-wall plasma energy with a free-space surface inductance. A run with a wall is now an error.

Regression report

Run against abc12d8eb, the head of #602 at the time (#602 has since merged into develop, adding only a case-TOML change that this branch also carries), on feynman: all cases at e375075bf, then the two repaired cases at f388c7910. That last commit changes only case TOMLs.

27 cases
22 cases, 325 quantities   unchanged, 0 changed
 1 case                    gal_resistive_pe: all N/A on both refs, as its TOML expects
 4 cases                   new, no baseline at abc12d8eb

tj_n1                  et[1] 0.4350277   beta_n 0.4156626   q0 1.599966   q95 2.980037   2 surfaces
tj_kinetic_ntv         fcgl 5.446400e-04 J   et[1] 0.4350277
tj_kinetic_multiion    NTV total 8.956676e-04 N·m (D 4.759306e-04, T 4.134019e-04, e 6.335161e-06)
tj_kinetic_calculated  et[1] 0.4380341   Im et[1] -1.957846e-03   NTV fgar 2.036734e-03 N·m

---

🤖 Generated with Claude Code

https://claude.ai/code/session_014wKUyAaxKdTJ5ch1boMFE1

[d-burg]

@matt-pharr I've pushed the Solovev-to-TJ move for #613 onto this PR, so that swap is taken. The perturbed equilibrium examples, fixtures and harness cases now run on a no-wall tj_analytic equilibrium, stable with et[1] = 0.435 and keeping the 2/1 and 3/1 surfaces. The body has the details.

Harness against abc12d8eb: 22 of 27 cases unchanged, 325 quantities, nothing moved. Four cases are new and gal_resistive_pe is N/A as expected. Tests: 58 of 59 files pass on feynman. runtests_equil.jl:305, the exact-equality check on the re-formed i-file, fails there, but it fails the same way without this work.

Three things touch #602:

1. lar_ggj_tearing_n1 doesn't run on Repo - API! - Consolidate the v2 deck sections, driver and HDF5 writers #602. Its deck writes gpec_lar_ideal.h5, so the harness reports "gpec.h5 not produced" on every ref and GGJ γ still isn't tracked. Adding "Run.output_file" = "gpec.h5" to [overrides] fixes it, and it runs now. It's in f388c7910 here; you may want it on Repo - API! - Consolidate the v2 deck sections, driver and HDF5 writers #602 too.

2. The Z_eff range check can reject valid data. _layer_mass_per_electron checks the spline at each surface, and a spline can undershoot its nodes. Measured on the 514-node DIII-D-like file:

   Z_eff profile	Node range	Spline minimum	ψ rejected
   As shipped	[1.756, 1.924]	1.75603	0%
   Smooth, edge-peaked	[1, 2.5]	1.0	0%
   Pedestal tanh	[1, 1.993]	1.0	0%
   Flat 1, ramp from ψ = 0.8	[1, 2]	0.99959	0.82%
   Flat 1, steep ramp from ψ = 0.95	[1, 3]	0.99631	0.96%

   So every node can be valid and a surface near the kink still stops the run, with an error saying no ion split exists. The shipped decks are safe. Suggested fix: check the node values once instead of the spline at each surface. That keeps [1, Z] and doesn't widen it.

3. _layer_ion_mass rejects density-specified species. It sums s.fraction, which is NaN for a species given by density, so total > 0 fails and even a single such species stops with "needs main-ion species with positive fractions". kinetic_profile_splines accepts those decks. It also averages mixed charges with no check. Weighting by each species' density would cover both spellings.

[logan-nc]

Should this be moved from draft to ready?

[logan-nc]

@matt-pharr @d-burg I took this over the line since #602 landed: merged develop (only conflict was the identical lar_ggj output-file line), added two tests for the new errors (plasma_response on a walled solve, and parse_deck with both [PerturbedEquilibrium] and [Wall]), dropped the stacked-on-#602 banner, and marked it ready with @jhalpern30 as reviewer. runtests_solve_api.jl (167) and runtests_deck_roundtrip.jl (113 + 12) pass. No src/ changes, so the harness stamp holds.