Data products

data/profile_datasets.yaml is the machine-readable contract for profile generation. It declares the profile-data version, density model, electronic-structure settings, radial grid, QA grid, density cutoffs, selected basis families, element coverage, charge-class selection, and state-role selection.

The generated data files live under data/profiles/, data/radii/, and data/qa/. Profile, radii, and QA tables are generated data products: do not hand-edit them. Change the source/configuration layer instead, regenerate the tables, and run the validation checks.

Profile datasets

dataset ID	basis	coverage	selected rows	selected states
pbe0_sfx2c_x2cqzvpall_h-rn_spherical_v2	x2c-QZVPall	H-Rn	430	all curated states in range
pbe0_sfx2c_dyallv4z_h-lr_spherical_v2	dyall-v4z	H-Lr	501	all curated states in range
pbe0_sfx2c_x2cqzvpalls_h-rn_spherical_v2	x2c-QZVPall-s	H-Rn	192	neutrals and anions; cations excluded
pbe0_sfx2c_dyallav4z_h-ba_hf-ra_spherical_v2	dyall-av4z	H-Ba/Hf-Ra neutrals plus selected anions in the same intervals	166	neutrals and anions; cations excluded

The primary datasets are deliberately not split into separate neutral, cation, and anion products. The supplemented/augmented branches follow the same rule for the states they support: each branch groups its neutral and anion rows under one basis identity, while cations are excluded because they are compact and less relevant to tail-sensitivity comparisons. x2c-QZVPall-s is an NMR-shielding-oriented supplemented branch rather than a generic diffuse basis. dyall-av4z has discontinuous coverage; its generated branch contains H-Ba/Hf-Ra neutrals plus selected anions in the same available intervals, including Fr and Ra monoanions.

The dataset ID is part of the public data identity. It encodes the method family, basis family, element coverage, spherical density convention, and major data version in a compact name. The full machine-readable record remains in data/profile_datasets.yaml and in each generated metadata.json file.

Radial profile tables

Each generated profile dataset contains:

• data/profiles/<dataset_id>/profiles.csv
• data/profiles/<dataset_id>/metadata.json

profiles.csv is a wide table. It has one shared radial column, r_bohr, and one density column for each selected atomic state. Density columns use deterministic curated state IDs:

rho_e_bohr3__<state_id>

The density unit is electrons/bohr³. The tabulated density is the spin-summed spherical proatomic density described in the scientific model. The stored release grid has 1200 logarithmic radial points from 1e-6 to 60 bohr.

metadata.json records the dataset identity, profile-data version, basis ID and basis checksum, density model, method settings, radial grid, QA grid, density cutoffs, state list, column map, related data paths, and generator provenance. The local SCF paths recorded in metadata are regeneration provenance; those files are intentionally ignored by Git.

The directory-level details are documented in data/profiles/README.md.

Density-cutoff radii

Each generated radii dataset contains:

• data/radii/<dataset_id>/radii.csv
• data/radii/<dataset_id>/metadata.json

radii.csv stores one row per selected state. For each declared density cutoff, radii are reported in bohr and ångström:

• r_iso_0.003_e_bohr3_bohr and r_iso_0.003_e_bohr3_angstrom
• r_iso_0.001_e_bohr3_bohr and r_iso_0.001_e_bohr3_angstrom
• r_iso_0.0001_e_bohr3_bohr and r_iso_0.0001_e_bohr3_angstrom

The 0.003 and 0.001 electrons/bohr³ radii are the primary practical size descriptors. The 0.0001 electrons/bohr³ radius is retained mainly as a tail and interpolation diagnostic.

The directory-level details are documented in data/radii/README.md.

QA data

Each generated QA dataset contains:

• data/qa/<dataset_id>/qa.csv
• data/qa/<dataset_id>/metadata.json

The aggregate QA layer contains:

• data/qa/qa_summary.csv
• data/qa/qa_report.md
• data/qa/metadata.json

The current generated QA summary contains four datasets, 1289 dataset-state rows, and zero validation failures. The QA layer records SCF completion, independent electron-count integration, finite-density checks, tail coverage, cutoff-radius consistency, angular sphericity, and linear-dependency diagnostics. Linear-dependency warnings are reported as diagnostics; they are not validation failures when the generated density passes the numerical QA gate.

The directory-level details are documented in data/qa/README.md.

Basis-sensitivity QA

Supplemented/augmented basis-sensitivity QA is stored below:

data/qa/basis_sensitivity/
  basis_sensitivity.csv
  basis_sensitivity_summary.csv
  basis_sensitivity_outliers.csv
  basis_sensitivity_metric_distributions.csv
  metadata.json

  dyall-v4z/
    basis_sensitivity.csv
    basis_sensitivity_summary.csv
    basis_sensitivity_outliers.csv
    basis_sensitivity_metric_distributions.csv

  x2c-QZVPall/
    basis_sensitivity.csv
    basis_sensitivity_summary.csv
    basis_sensitivity_outliers.csv
    basis_sensitivity_metric_distributions.csv

The dyall-v4z/ subdirectory compares dyall-v4z with dyall-av4z for matched neutral and anion states. The x2c-QZVPall/ subdirectory compares x2c-QZVPall with x2c-QZVPall-s for matched H-Rn neutral and anion states. The root-level CSV files are aggregate compatibility outputs. check_basis_sensitivity.py emits every configured supplemented/augmented comparison by default when the corresponding generated profile datasets are present.

Current generated basis-sensitivity counts are:

comparison	rows	low	moderate	high-sensitivity outliers	validation failures
dyall-v4z vs dyall-av4z	166	132	20	14	0
x2c-QZVPall vs x2c-QZVPall-s	192	192	0	0	0
aggregate	358	324	20	14	0

The sensitivity metrics classify how much the radial density distribution changes when the supporting basis branch is used. Large sensitivity can be scientifically expected for some formal or highly charged anions and is not, by itself, a validation blocker. The narrative interpretation and recommended next analyses are summarized in the Results.

Primary basis-family comparison QA

The primary basis-family comparison is stored below:

data/qa/basis_comparisons/
  metadata.json
  x2c-QZVPall__dyall-v4z/
    basis_comparison.csv
    basis_comparison_summary.csv
    basis_comparison_outliers.csv
    basis_comparison_metric_distributions.csv

This comparison is not a diffuse-basis sensitivity test. It compares the primary x2c-QZVPall and dyall-v4z branches over their H-Rn overlap, matching exact state_id values and state-record digests. The current data product contains 430 matched rows, zero integrity failures, and one high-difference formal multianion outlier. Signed deltas are dyall-v4z minus x2c-QZVPall.

Regeneration policy

Regenerate profiles, radii, QA, basis-sensitivity QA, primary-basis-comparison QA, and documentation-derived outputs with:

python scripts/extract_profiles.py --force --check
python scripts/check_basis_sensitivity.py --force
python scripts/check_basis_comparisons.py --force
python scripts/check_profile_artifacts.py --require-generated
python scripts/prepare_docs.py --write

The expensive local SCF material is stored under:

• local-data/scf/<dataset_id>/<state_id>/scf.chk
• local-data/scf/<dataset_id>/<state_id>/scf.npz
• local-data/scf/<dataset_id>/<state_id>/scf.json
• local-data/scf/<dataset_id>/<state_id>/scf.log

local-data/ is ignored by Git. It is required for regeneration but is not part of the public release tables.

Multiwfn interoperability files

The configured Multiwfn interoperability root is data/multiwfn_artifacts/. It contains generated density-only .rad files and neutral-atom PROAIM .wfn files derived from the same local SCF checkpoint, NPZ, and metadata artifacts used to produce the release profile layer. These files are committed derived products for Multiwfn-facing workflows; they do not replace the project-native profile, radii, QA, or comparison tables.

Current generated contents are:

format	count	basis branches	intended role
.rad	931	x2c-QZVPall H--Rn and dyall-v4z H--Lr primary branches	density-only atomic radial references for stockholder/Hirshfeld-like use
.wfn	86	neutral x2c-QZVPall H--Rn atoms	atomwfn-style wavefunction containers for workflows that need GTF and spin-orbital information

The .rad files are evaluated on the fixed Multiwfn atmrad grid from SCF density matrices; they are not generated by interpolating the committed profiles.csv tables. The .wfn files use the validated atomref spin-orbital convention: alpha orbitals first, beta orbitals second, occupations at or below one, and explicit $MOSPIN labels 1 and 2.

These files are useful when the atomref-proatoms reference gauge should be used inside Multiwfn workflows involving promolecular and deformation-density maps, Hirshfeld/Hirshfeld-I-like charges and populations, VDD and ADCH-style charge analyses, fuzzy atomic-space integrations, orbital composition with Hirshfeld partitioning, real-space density and spin-density maps, topology/basin/domain analyses, molecular-surface analyses, weak-interaction visualizations such as RDG/NCI, IGM/IGMH/mIGM/aIGM, IRI and DORI, and density-difference or charge-transfer inspection.

The main scientific advantage over relying on ad hoc Multiwfn atom generation is that the atomref densities are spherical at the SCF-model level. They are not ordinary anisotropic open-shell atom calculations that were sphericalized only after SCF convergence. The product also provides a broader, versioned coverage contract: .rad files for all configured primary-branch states over H--Rn and H--Lr, and neutral H--Rn .wfn files where the chosen WFN basis-format boundary is safe.

The manifest uses repository-relative paths. The optional file field is kept only as a compatibility alias for older diagnostics and is required to match the canonical path field. One-letter neutral WFN filenames intentionally retain Multiwfn's atomwfn spacing convention, for example H .wfn and O .wfn.

The folder-level contract and regeneration commands are documented in data/multiwfn_artifacts/README.md.