[logan-nc]

Adds Tearing.IslandFlattening: static flattening of the equilibrium pressure across magnetic islands of prescribed width, with the stored energy and pressure the islands remove.

Part of #629 (critical field, vacuum islands and flattening): its third item, saturated-island pressure flattening. #665 carries the other two (the critical resonant field and the vacuum-island widths); #629 closes when both are in.

What it computes

For full island widths w (ψ_N) keyed by "m/n", every q = m/n surface of the equilibrium gets an island. Inside each island dp/dψ = 0; outside it the gradient is unchanged; overlapping islands are merged and flattened once; the pressure keeps its value at the last profile point and is integrated inward. This is the saturated-island profile of Fortran GPEC, evaluated in closed form for merged intervals [a_k, b_k]:

p_mod(ψ) = p(ψ) + Σ_{k: b_k > ψ} [p(b_k) − p(max(ψ, a_k))], ΔW = (3/2) ∫ (p − p_mod) dV,

so there is no grid dependence, and the per-surface Fortran estimate's double counting of overlaps does not arise. ΔW uses Gauss–Legendre on the profile knots and interval ends, exact for the piecewise-cubic integrands.

Outputs, in Tearing/IslandFlattening/:

• W, ΔW, ΔW/W and the on-axis pressure drop;
• each island's surface and flattened interval (with a flag when it reaches past the profile grid);
• the merged intervals with the pressure drop across each, which for an edge interval is the pedestal-top pressure change;
• p and p_mod on the profile grid.

Given a bootstrap profile (jbs_B keyword), the result also gives the part the islands remove (j_bs ∝ p′ to leading order, so it vanishes on the intervals). The deck route for that waits for a bootstrap-current source on develop.

Use

[Tearing.IslandFlattening]
widths = { "2/1" = 0.05, "3/2" = 0.03 }

A [Tearing] holding only this sub-table runs no layer solve. In a script: island_flattening(ffs.equil; widths=Dict("2/1" => 0.05)). Docs: docs/src/inner_layer.md.

Also

Utilities.record_inputs! keeps recorded sub-tables a result does not state, so [Tearing.IslandFlattening] is recorded beside [Tearing] rather than replacing it. The rule and src/Utilities/Output.jl are identical to #665 (MRE with saturated-island roots). A test merge with #665 conflicts only where both add to the same lists (includes, exports, deck sections, main's return, docs). Each conflict resolves by keeping both sides, with one exception: in tearing_from_deck each branch strips its own sub-table before deciding whether a layer solve runs, and the merge strips both. On the merged tree the island-flattening, modified-Rutherford, result-struct, SLAYER-runner, deck round-trip, rerun and solve-API suites all pass.

Review package

The rule drawn out, plus four checks on the DIII-D-like example: the profile, the leading-order limit, convergence of the Fortran grid rule to the closed form, and overlap counted once. https://claude.ai/artifact/SDY8PUQKf6fa4FdQLQLhHQ

Tests

• test/runtests_island_flattening.jl checks four things:
  • The Fortran trapezoid rule on a fine grid converges to the closed form.
  • Narrow islands: ΔW → 0 and ΔW ∝ w, with the leading-order coefficient (3/2)(−p′) w V(ψ_s).
  • Overlapping islands are flattened once, checked against an independent quadrature.
  • Edge and inner clipping, the bootstrap fraction (uniform and peaked profiles), bad input, and the HDF5 layout.
• Deck and API parity, including replay from the recorded deck, in test/runtests_solve_api.jl.
• New harness case diiid_n1_island_flattening, which puts 3/2, 2/1 and 3/1 islands on the DIII-D-like example.

Release note

• Audience: users
• Numerical impact: none on existing quantities. All 27 existing cases are unchanged. The new case diiid_n1_island_flattening has no develop baseline (develop rejects the section): W = 5.83e5 J, ΔW/W = 8.72e-2. (harness @ 5d0483f vs develop f11d75a; the later f6aa34b changes only the input record and two comments) The full suite passes except runtests_equil.jl:312 (TokaMaker i-file), which fails identically on develop.
• Migration: none. main returns one more field, island_flattening.

New [Tearing.IslandFlattening] section: static pressure flattening across islands of prescribed width, giving the stored energy and the pressure (pedestal-top included) the islands remove.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BgNxCBwLNHQmVq7iXMCTae