Add mainline magnetic reconnection example

[rogeriojorge]

Problem

Main has no magnetic-reconnection example on the rFFT solver.

Change

One file, Examples/2D_magnetic_reconnection.py: periodic double Harris sheet with the Camporeale et al. (2006, sec. IV.C) parameters (m_i/m_e=16, omega_pe/Omega_ce=3, T_i=T_e, rho_i/a=2, guide field 5). Harris pressure balance gives v_th,e = sqrt(2)/6 c and hence a = sqrt(2) d_e; thermal speeds a/6, a/24 and drifts ±1/(6a) follow. Evolves in chunks, fits the half-sheet magnetic-mode growth, and writes a growth/flux plot, a J_z movie with A_z contours and an npz file. --perturbation 1e-4 is the linear diagnostic, 0.1 a finite-seed nonlinear island.

Effect (CPU, shared 36-core node)

• default 33x9, H=4, t=10: finite, 12 s
• 65x17, H=4, t=500, perturbation 0.1: closed island, 258 s; last-Hermite-mode fractions 4.3e-8, 2.6e-5, 3.5e-5
• Linear tearing (perturbation 1e-4, ny 5, nu 1, t = 1000; ky = 1 mode at the X point, fit over 500 <= t <= 1000) against Camporeale's 0.256 Omega_ci = 5.33e-3 omega_pe:

nx	H	gamma	vs paper	runtime
65	10	2.39e-3	-55 %	3.7 h
129	6	4.76e-3	-11 %	53 min
129	8	4.84e-3	-9 %	3.3 h
257	6	5.07e-3	-4.9 %	3.9 h

The rate converges in nx (the sheet needs nx >= 257 in this box); extrapolating nx (second order) and H gives about 5.25e-3, 1.5 % below the paper. The docstring records the table.

Tests

Example only, no solver change; full suite 7 passed (also with JAX_ENABLE_X64=1).

Update 2026-10-03 (head 87c4cf7)

• Growth observable and fit now match: the docstring names the observable (first ky mode of (Bx, By) over the left half-box), and gamma is a least-squares fit of its log over --fit-window (default: second half of the run). The output also has the fit rms log residual and the rates of the two halves of the window, as a check that the window is in a single exponential phase. The earlier fit through local maxima for t >= t_max/4 is gone.
• The movie shows the kinetic current J_z/Omega_ce from the Hermite moments (plasma_current). It is not labelled curl B, which also contains dE_z/dt.
• Hermite tails: hermite_tail_fraction per population (from Fix odd-grid RFFT energy weights; add a Hermite boundary diagnostic #9) replaces the unweighted all-species sum without Parseval weights. Depends on Fix odd-grid RFFT energy weights; add a Hermite boundary diagnostic #9. With main's two-population dCk bookkeeping, populations 3 and 4 keep their k=0 background, so the tail fractions are exact only after Fix multispecies perturbation bookkeeping #12.
• The convergence table is kept, but labelled as measured with the old peak fit and not re-measured. The "second-order extrapolation to 5.25e-3" is removed because no spatial order was measured.
• Run here: default smoke (33x9, H=4, t=10) and t=40 with --fit-window 10 40, both finite, on main + Fix odd-grid RFFT energy weights; add a Hermite boundary diagnostic #9 (4.6 s and 10.5 s on a laptop CPU). These short runs are transient, not tearing growth: the rms residual is 0.46 and 0.98, so the new output flags them.
• Equilibrium normalization: checked independently (see Correct Camporeale tearing sheet width and box size #33). Harris drift V = v_th^2/(Omega_ce a) = 1/(6a) and J_z/Omega_ce = (curl B)_z = sech^2/a with a = sqrt(2) d_e. This example uses the exact 1/3, 1/48 constants.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
see 2 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.