Skip to content

openWFN 0.8.1

Historical release notes. Version 0.8.1 added checks against silent numerical errors and improved source tracking. It retained the stable 0.8 CLI and Python interfaces.

Install

python -m pip install --upgrade openwfn

For reproducible work, pin the exact release:

python -m pip install openwfn==0.8.1

These notes describe 0.8.1. The README PyPI badge follows the latest package version and is not an identifier for this historical release.

Scientific correctness safeguards

  • ECP and ghost centers now use source effective nuclear charges from Gaussian FCHK Nuclear charges when available.
  • Density expectations prefer source electron-count records; deterministic fallbacks warn instead of silently guessing.
  • Population and density conservation failures retain usable values but return status="partial", Experimental validation status, and a warning.
  • Zero-target spin-density validation uses an absolute-error criterion.
  • Density cubes are marked Validated only when the generated grid passes electron-conservation validation.
  • Post-HF analyses explicitly identify when the parsed SCF density is being used rather than implying a correlated density.

Frontier orbitals and spin handling

Unrestricted calculations can now request a spin-complete frontier result:

openwfn FILE orbitals frontier --spin all

The same selection is available in batch workflows and the Python API. The overall HOMO is chosen from both alpha and beta channels, ROHF is distinguished from closed-shell restricted references, and frontier selection is occupation aware rather than assuming LUMO = HOMO + 1.

ECP, ghost, and structural behavior

Population charges, nuclear ESP, and cube atom charges now use effective nuclear charges. Ghost centers are excluded from physical formula, center-of-mass, inferred bond, fragment, and nuclear-ESP summaries. Bond and fragment counts remain covalent-radius heuristics rather than authoritative connectivity.

Numerical robustness

Löwdin analysis reports overlap-matrix conditioning diagnostics. Density-grid AO evaluation is chunked to reduce peak memory while preserving Cartesian point order and numerical values. The default 0.15 bohr density spacing remains an accuracy/performance starting point; quantitative research should converge spacing and padding.

Automation and reproducibility

Batch workflows preserve scientifically usable partial analyses instead of collapsing them into errors. Frontier spin selection is included in reproducibility fingerprints, and analysis warnings are merged with source/provenance warnings rather than overwritten.

Validation scope

The release is covered by regression fixtures for ECP effective charges, ghost centers, post-HF SCF-density provenance, UHF cases where the beta HOMO is higher, ROHF classification, zero-spin density integration, cube validation, warning propagation, and chunked density equivalence.

These fixtures are implementation evidence, not universal independent validation of every ECP family, basis set, post-HF method, or chemical system. Existing Validated, Experimental, and Unsupported capability labels remain scoped to their documented evidence.

The repository release gates include Python 3.10–3.13, Linux/macOS/Windows smoke coverage, strict documentation builds, wheel and source builds, twine check, scientific validation, external repository benchmarks, and installed-example workflows.