Reconnection-rate scaling with FLASH¶
A fresh Simjecture 0.7.0 investigation falsified the proposed slope band −0.55 ≤ p ≤ −0.45. The independent reviewer accepted the counterexample from fresh production trajectories. The explanatory repair remained unvalidated.
The upper rows show density with magnetic streamlines and out-of-plane current from three saved states of the fresh production trajectory. The lower panel compares exploratory rates with the qualified production contrast. The dashed orange curve is the original \(p=-1/2\) reference, normalized to the lowest-\(S_\eta\) exploratory rate; it is not fitted to the production results. Download figure PDF.
The question and the counterexample¶
The operator supplied a commissioned FLASH resistive-MHD island-coalescence application and the bounded hypothesis. The worker fixed the flux-window rate measurement, admissibility checks and numerical uncertainty rule, then examined how reconnection changes with the inverse-resistivity control \(S_\eta=1/\eta\).
The decisive contrast uses \(S_\eta=1000\) and \(2000\):
Quantity |
Fresh reviewed result |
|---|---|
Rate at \(S_\eta=1000\) |
0.0496886 |
Rate at \(S_\eta=2000\) |
0.0382421 |
Effective log-slope |
−0.3778 |
Numerical sensitivity interval |
[−0.4314, −0.3247] |
Conjectured band |
[−0.55, −0.45] |
The entire numerical interval lies above −0.45. This rejects the stated band, not merely an exact −1/2 exponent. The interval propagates measured numerical sensitivities; it is not a statistical confidence interval.
A current sheet only a few cells wide made a stable integrated rate insufficient. Independent methods review required matched-flux profile checks and a magnetic-null inventory as well as spatial, timestep, cadence and flux-reconstruction controls. When the \(256^2\to384^2\) rate shift grew, the worker retained that failed check and refined to \(512^2\). Two fresh \(384^2\) production trajectories followed methods approval, with fine-grid uncertainty transported into the decisive comparison.
The run completed 17 MHD trajectories, six diagnostic analyses and three failed trajectory receipts. GPT-6.1 Sol at medium effort served as worker; the director and independent reviewers used high effort in separate contexts. The operator extended the allowance and restored recorded interruption time. The investigation then closed with an accepted report. Root falsification is established; the subsequent repair attempts did not pass their fresh validation, so no replacement law is claimed.
Inspect and reproduce¶
The fresh compact audit
contains the original inputs, source snapshots, all 26 receipts, methods and claim
reviews, accepted report and omission ledger. The numerical runtime was
0.7.0, commit cfca558cb30868d7bc12af7a9b636303f7bc500b.
From the repository root:
uv run python scripts/verify_research_audit.py demos/resistive_mhd_island_coalescence/fresh_v070_record
uv run --extra flash-demo python scripts/render_fresh_plasma_examples.py
These commands check retained bytes and redraw figures without simulations or model calls. Bulk FLASH states are explicitly omitted from the public audit; complete native-data replay requires the original artifacts. A fresh investigation requires an operator-supplied, qualified FLASH executable and a configured agent backend. The retained commissioning inputs are starting material, not evidence for a new run.
The earlier investigations and coordinate-diagnostic audit remain available with their original findings and figures. Their wider slope band and unresolved outcomes belong to those historical studies.