Reproduction code — Conditional maximum entropy production in Earth-system branch selection: transport closure, quasi-potentials, and dynamical accessibility

This dataset is the self-contained replication package for the paper "Conditional maximum entropy production in Earth-system branch selection: transport closure, quasi-potentials, and dynamical accessibility" (S. Kaneko). The paper asks when an Earth-system maximum-entropy-production (MEP) ordering selects the realized branch among the coexisting attractors of a specified slow transport dynamics. Once the slow transport variable is given an effective dynamics, branch selection is governed by the full Freidlin-Wentzell quasi-potential - of which the resolved entropy production is only the time-antisymmetric part - rather than by the dissipation rate. The omitted non-gradient structure, of which the time-symmetric activity (frenesy) is one part, is set by the eddy and turbulent transport closure that empirical parameterizations fix by hand; it is identified here through the closure-induced non-gradient current rather than decomposed. The scripts make this concrete in minimal predator-prey closures and in the standard reduced model of equatorial superrotation, and they quantify when a closure reverses the dissipation ordering. The package regenerates every figure (Figs. 1-11) and every numerical diagnostic in the paper, including: the closure-to-activity (Lyapunov) circulation table; the weak-noise Fokker-Planck selection scan within the bistable window, with its saddle-node threshold and crossing; the directly computed quasi-potential gap by the geometric minimum-action method (gMAM), with the Omega=0 gradient-limit validation and path-resolution convergence; the self-consistent single-channel model (MEP-aligned: the higher-dissipation branch is selected, frenesy not overturning the ordering); a two-channel closure that yields a genuine but conditional reversal (the lower-dissipation regime selected in a minority of cases); a deterministic straight-line quasi-potential predictor that forecasts which closures reverse in about nine of ten cases; a structurally distinct second (Hill-toggle) closure confirming the reversal is not a mass-action artifact; the superrotation quasi-potential under both the Hadley-cell-feedback and wave-jet-resonance mechanisms; and the Eyring-Kramers prefactor / Maxwell-degeneracy analysis. Contents: 11 Python reproduction scripts; the 11 figures they produce, as PDF and PNG; a README.md mapping each script to its paper figure and section and tabulating the numerical audit trail (grids, domains, noise strengths, boundary conditions, convergence); requirements.txt (minimum versions) and requirements-lock.txt (the exact verified build); run_all.sh; and run_all_expected_output.txt (expected stdout for the main-text scripts). All computations are deterministic - the predictor scan draws its closures from fixed random seeds - so all reported results are reproducible. Verified with Python 3.12, numpy 2.4.4, scipy 1.17.1, and matplotlib 3.10.8.