Validation and Numerical Status¶
Validation is divided into three distinct layers:
- Initial-condition integrity — the complete 12-halo, 48-file MUSIC2 suite is tested before simulation; see MUSIC2 IC-suite validation.
- End-to-end code integrity — a dust-enabled zoom is evolved from the initial conditions to \(z=0\), with whole-box conservation and particle fields audited.
- Physical convergence — scientific observables are compared across resolution levels. A successful code or IC test is not a convergence claim.
End-to-end validation run¶
A dust-enabled Halo 295 / 512 zoom completed from \(z=98.1046\) to \(z=0\). The run exercises star formation, stellar enrichment, explicit PartType6 creation and evolution, domain decomposition, FOF/SUBFIND, and particle exchange.
Whole-box conservation benchmark¶
| Quantity | Result |
|---|---|
| Initial baryonic mass | 2.611556729 × 10¹² M☉ |
| Final baryonic mass | 2.611556749 × 10¹² M☉ |
| Fractional baryonic-mass drift | 7.747311 × 10⁻⁹ |
| Fractional dark-matter mass drift | 7.375289 × 10⁻¹⁶ |
| Dust C + silicate / total dust | 1.000000000000 |
| Non-finite masses | 0 |
| Negative masses | 0 |
| Invalid dust carbon fractions | 0 |
At \(z=0\), the box contains 136,764 gas particles, 1,044 star particles, and 8,365 surviving dust particles. The initial gas count was 137,808, so final gas + star particle counts close exactly at 137,808.
The current mass and composition-integrity checks therefore PASS.
Metal bookkeeping¶
Gas metallicity, stellar metallicity, dust mass, and the tracked gas-element reservoirs evolve smoothly from zero initial enrichment. C, N, O, Ne, Mg, Si, and Fe are audited separately.
GasZ + StarZ + Dust is treated as a bookkeeping trend rather than an automatically conserved scalar, because formal conservation depends on the precise yield, stellar-remnant, and metallicity conventions used by the enrichment implementation.
Code validation versus physical convergence¶
The successful 512 run demonstrates end-to-end numerical integrity; it does not establish convergence of galaxy-scale dust predictions.
For Halo 295 at \(z=0\), the central 30 pkpc contains only:
| Component | Particle count |
|---|---|
| Gas | 63 |
| Stars | 152 |
| Dust | 9 |
The surviving dust mass inside 30 pkpc is \(1.743\times10^3\,M_\odot\), about 0.283% of the surviving dust mass inside \(R_{200}\). At this resolution, the central ISM, stellar population, enrichment history, and dust distribution are too sparsely sampled for this result alone to be interpreted as physical dust evacuation.
The convergence program therefore repeats the same diagnostics at 1024³, 2048³, and 4096³-equivalent zoom resolution.
Recent regression fixes¶
The current validation baseline includes several fixes that materially changed what should be considered a trustworthy production run:
- the legacy instantaneous enrichment contribution was removed from the active enrichment path in favor of the delayed stellar-feedback treatment;
- stellar metallicity bookkeeping was corrected to use the intended hydrogen mass fraction convention;
- dust state and cadence bookkeeping across restarts was repaired and is now explicitly serialized/tested; and
- snapshot continuation includes dust-ID and particle-state safeguards before the initial domain decomposition.
Because enrichment changes can strongly alter the dust budget even when gas and stellar masses change modestly, results produced before these fixes should not be mixed quantitatively with the current production baseline without a targeted comparison.
See Resolution and gas smoothing length for the separate
MinGasHsml convergence tests.
Current caveats¶
- 512³-equivalent zooms are validation runs. Internal galaxy structure, D/G, D/Z, dust morphology, and dust transport require resolution testing.
- LRN dust production is not yet finally calibrated. The source channel is implemented and produces tagged PartType6 particles, but its absolute yield normalization remains under investigation.
- Composition is presently represented by an evolving carbon fraction. More detailed hybrid/Astrodust-inspired composition models are a future direction.
- Transport histories require evolving halo centers. Raw displacement between
BirthPosand present-day position contains halo bulk motion and must not be interpreted directly as a dust-outflow distance.
Regression benchmark¶
Future changes to particle exchange, feedback, enrichment, or dust mass-transfer routines should rerun the whole-box conservation audit. A material degradation relative to this Halo 295 / 512 benchmark should be investigated before accepting the change.