Input data

The v1 datasets are generated from two compact tracked input layers: atomic-state records and frozen Gaussian basis bundles. The active selection and numerical settings are combined in data/profile_datasets.yaml.

Atomic-state layer

The curated state table is:

• data/states/curated/atom_states_v1.json

The selection file used to build it is:

• data/states/selection/required_states_v1.csv

Each curated record contains the fields needed to construct a one-center PySCF atom and assign spherical fractional occupations: element, charge, electron count, configuration, multiplicity, spin-resolved angular-momentum counts, state role, curation status, and occupation policy.

The active v1 profile datasets select neutral recommended states from this table. A charged record may exist in the curated state layer without being part of any v1 generated profile dataset.

Source convention

Neutral and positive-ion configuration labels are prepared from the NIST Atomic Spectra Database Ground States and Ionization Energies interface, NIST Standard Reference Database 78. This repository keeps compact configuration labels needed for generator reproducibility. It does not redistribute raw NIST ASD pages, ionization-energy tables, uncertainty records, or bibliography rows.

Small formal-anion configuration records are stored separately under data/states/source/. They are part of the state curation layer, not active v1 profile outputs unless selected by the active profile-dataset configuration.

The state-layer details are documented in data/states/README.md, and the build entry point is scripts/build_atom_states.py.

Basis-set layer

The frozen basis bundles are stored under data/basis_sets/. Each bundle contains one NWChem-format spherical basis file, a manifest, SHA256 checksum, local reference notes, and a bundle README.

Active v1 branches:

basis ID	role	active coverage
x2c-QZVPall	primary H-Rn scalar-relativistic branch	H-Rn
dyall-v4z	primary H-Lr heavy-element branch	H-Lr

Auxiliary frozen bundles may be present for sensitivity checks. They are not active profile datasets unless explicitly selected in data/profile_datasets.yaml.

The basis text checksum is the basis-data identity. Ordinary validation is offline and checks required files, checksums, NWChem spherical headers, manifest consistency, and declared element coverage. If PySCF is installed, representative basis-parse smoke checks are also run.

The basis-layer details are documented in data/basis_sets/README.md, and the validation entry point is scripts/check_basis_bundles.py.

Dataset specification

data/profile_datasets.yaml connects the input layers to the generated data products. It declares:

• the profile-data version;
• PySCF version expected for release generation;
• PBE0, UKS, sf-X2C-1e, and pure/spherical basis settings;
• SCF convergence and DFT grid settings;
• the stored radial profile grid;
• the independent QA grid;
• density cutoffs used for radii;
• active dataset IDs, basis IDs, element coverage, and neutral-only selection.

This YAML file should be treated as the central machine-readable contract for the v1 release configuration.