Nuclear Cross-Section Data Pipeline
How Kronos ingests, processes, and version-pins evaluated nuclear data so every neutronics run is reproducible and auditable.
Data is an input, not a detail
A neutronics result is only as trustworthy as the nuclear data behind it. Cross-sections, the probabilities that drive every collision in the transport equation, come from evaluated nuclear data libraries and must be processed into the point-wise, temperature-dependent form the Monte Carlo code consumes. Kronos treats this pipeline as a first-class, version-controlled L0 workload.
From evaluation to usable data
Raw evaluated files are processed: resonance parameters are reconstructed into point-wise cross-sections, Doppler broadening is applied at the operating temperatures of the breeder blanket and burner structure, and the data is formatted for the transport code. This processing is itself a compute job, and its output is a dataset that many downstream runs share.
- Resonance reconstruction to point-wise cross-sections
- Doppler broadening to material temperatures
- Thermal scattering treatment for moderating materials
- Format conversion and integrity hashing for the transport code
Version pinning and reproducibility
Every processed library is hashed and versioned. A neutronics run records exactly which library it used, so a breeder TBR value or a burner shielding flux can be re-derived years later against the identical data. Because cross-section evaluations are updated over time, pinning is the only way to keep old and new results comparable.
Data choice is also a source of uncertainty. Different evaluations disagree, especially for the lithium and beryllium reactions central to breeding. Kronos runs key studies against more than one library to bound how much the TBR estimate depends on the data itself, feeding that spread into the uncertainty quantification.
The pipeline serves both machines from one curated store. The breeder needs accurate 14 MeV and lithium data; the burner needs accurate D-D, D-T secondary, and structural-activation data. Curating them together, with shared processing and shared version control, keeps both machines' neutronics on a common, auditable footing.