[matt-pharr]

Release note

• Audience: users
• Numerical impact: none (harness @ ddad504)
• Migration: none

solve(EulerLagrangeProblem, Forward()) now runs kinetic stability problems with kinetic_source = "fixed", which previously only worked from a gpec.toml deck. The "calculated" source still needs the [KineticForces] profiles and remains deck-only. Kinetic runs with Riccati() or Galerkin() now fail immediately with a clear message.

Regression report

Regression Report: diiid_n1
Ref 1: develop  @ 10459f3 (2026-10-26)
Ref 2: feature/solve-fixed-kinetic  @ ddad504 (2026-10-26)
       env: julia 1.12.6, x86_64-linux-gnu, manifest aa94b997 (pinned), 8 threads/4 BLAS, jit build
Quantity                                      develop          feature/solve-fixed-kinetic  Diff       Status
total energy Re(et[1])                        8.006123e-01     8.006123e-01                 0.0e+00    OK
total energy Im(et[1])                        7.924925e-05     7.924925e-05                 0.0e+00    OK
plasma energy Re(ep[1])                       -1.372261e+00    -1.372261e+00                0.0e+00    OK
vacuum energy Re(ev[1])                       2.172874e+00     2.172874e+00                 0.0e+00    OK
# singular surfaces                           5                5                            0.0e+00    OK
PE plasma energy                              3.320677e+00     3.320677e+00                 0.0e+00    OK
PE toroidal torque                            1.119454e-04     1.119454e-04                 0.0e+00    OK
NTV torque FGAR [N·m]                         5.483495e-01     5.483495e-01                 0.0e+00    OK
...
Summary: 53 unchanged

Regression Report: solovev_kinetic_calculated
Quantity                  develop        feature/solve-fixed-kinetic  Diff     Status
vacuum energy Re(ev[1])   1.054785e+01   1.054785e+01                 0.0e+00  OK
||xi_clebsch_alpha||      7.717981e-03   7.717981e-03                 0.0e+00  OK
total energy Re(et[1])    1.827879e+00   1.827879e+00                 0.0e+00  OK
total energy Im(et[1])    -1.326825e+00  -1.326825e+00                0.0e+00  OK
plasma energy Re(ep[1])   -8.719975e+00  -8.719975e+00                0.0e+00  OK
# singular surfaces       2              2                            0.0e+00  OK
...
Summary: 17 unchanged

Notes for reviewers

• The "fixed" source builds its X-shaped test matrices in build_kinetic_matrix_splines without the KineticForces callback. The axis-validity boundary is 0.0 when no profiles are loaded and is only used in the "calculated" branch. So the placeholder KineticForcesControl() and nothing profiles that solve passes are never read on this path. The solve guard now rejects only kinetic_source = "calculated".
• solve checks Riccati()/Galerkin() + kinetic_factor > 0 before the metric and matrices are built. The deeper guards stay for the deck path, and the Riccati messages now match the Galerkin one.
• The ForceFreeStatesControl docstring now describes the "calculated" source instead of calling it unimplemented.
• New parity test in runtests_solve_api.jl: solve(...; kinetic_factor=1e-3) against main on the coarse Solovev fixture. It checks et[1] and psilim to rtol=1e-12, plus nzero, the mode range and the kinetic singular surfaces. The rejection tests match on the error message.
• Test - TEST - Add deck/API parity tests with the known drifts marked broken #546 (deck/API parity tests) marks this case @test_broken. Whichever of Test - TEST - Add deck/API parity tests with the known drifts marked broken #546 and ForceFreeStates - FEATURE - Accept the fixed kinetic source in solve(EulerLagrangeProblem) #550 merges second turns it into a pass and drops the duplicate.

🤖 Generated with Claude Code

[logan-nc]

Read in full. Ready to approve from my read; nothing to add. The "fixed" source path genuinely never reads the placeholder control or the nothing profiles (the axis-validity boundary is only consulted on the "calculated" branch), and the three rejection tests match on message text so a reworded guard cannot pass by accident.

[logan-nc]

Approving at 0d7b8fe on the comment review above. Nothing outstanding from my read.