Validation¶
Every headline number on this site comes from an acceptance gate that was actually run, not from design intent. Three headline gates anchor the project – the second of them itself a bundle of four internal gates. Each one compares the bridge against an independent baseline, and each one only counts when both sides run the same pinned Julia 1.12.7 – a comparison across Julia versions measures version drift, not the bridge.
Gate 1: CJ detonation speed¶
The reacting-flow capstone (reacting_flow) simulates a Chapman–Jouguet
detonation: a Fortran Euler driver with a vendored Clawpack kernel, coupled
to Catalyst chemistry through Strang splitting.
| Quantity | Value |
|---|---|
| Simulated CJ speed | 7.12391 |
| Analytic CJ speed | 7.12470 |
| Deviation | 0.011% |
The simulated wave speed landing within 0.011% of the analytic CJ value means the operator splitting, the bridge crossings inside it, and the Julia chemistry integrator all conserve the physics to within discretization error.
Gate 2: bridge vs native – four internal gates, zero deviation¶
The GRI ignition case (gri_ignition) is run twice: once through the bridge
and once as a native Julia baseline. Four internal acceptance gates compare
the outputs.
| Setting | Result |
|---|---|
| Julia version, both sides | 1.12.7 (pinned) |
| Gates compared | 4 |
| Deviations found | 0 |
| Residuals | < 1e-12 |
Residuals below 1e-12 are at the level where floating-point non-associativity ends and exact agreement begins: the bridge changes nothing about the arithmetic.
Why the pin is part of the metric
An earlier comparison run once failed for a reason that had nothing to do with the bridge: a version shim silently resolved to a different Julia minor version than the one the bridge embeds, and the baseline drifted. Since then, both sides of every comparison are pinned to the same 1.12.7 binary.
Gate 3: GRI ignition delay¶
The same machinery computes the ignition delay of a homogeneous GRI-Mech 3.0 mixture (53 species, 325 reactions):
- t_ign = 1.1075 ms, taken as
argmax |dT/dt|– the time of maximum heat-release rate, the standard sharp definition that does not depend on a threshold convention.
A physically meaningful ignition delay from a 325-reaction network is an
end-to-end check: mechanism parsing (ChemMechSim.jl, open source; this
validation used v1.2.0), network construction (Catalyst), stiff integration
(OrdinaryDiffEq), and the data path back into Fortran arrays.
Summary¶
| Metric | Value | Gate |
|---|---|---|
| CJ detonation speed | 7.12391 vs 7.12470, 0.011% off | reacting_flow, Clawpack1d + Strang splitting |
| Bridge vs native | 4 gates, 0 deviations, residuals < 1e-12 | gri_ignition, both sides on 1.12.7 |
| GRI ignition delay | t_ign = 1.1075 ms (argmax abs(dT/dt)) | gri_ignition, GRI-Mech 3.0 via ChemMechSim.jl v1.2.0 |
Reproduction steps, timing methodology, and the deployment-mode ladder (JIT 17.3 s → sysimage 1.90 s → AOT first call ~12 μs) are covered in the User Guide and the developer annexes.