Skip to content

openWFN 0.10.0rc1

Historical release-candidate notes from October 2, 2026. At that time, the stable release was 0.9.2. This candidate added everyday-QC analyses and improved automation and resource controls; it did not complete every planned scientific method.

Install

python -m pip install --pre "openwfn[interop,resources]==0.10.0rc1"
openwfn --version

The source tag includes eleven real RHF/UHF molecular examples and reference records. The wheel retains the installed water example:

openwfn examples install installed-examples
openwfn installed-examples/water.fchk orbitals cube --mo homo --output homo.cube
openwfn installed-examples/water.fchk orbitals composition --mo homo
openwfn installed-examples/water.fchk bondorder mayer
openwfn installed-examples/water.fchk orbitals dos --export dos.csv
openwfn installed-examples/water.fchk report build report.html
openwfn installed-examples/water.fchk workbench workbench.html

Included capabilities

  • Signed MO cubes with one-based/HOMO/LUMO selection and alpha/beta routing.
  • Named Mulliken/Löwdin orbital compositions and spin-corrected Mayer orders.
  • Orbital-energy Gaussian DOS and atom/element/angular PDOS, with CSV/JSON/plots.
  • Native Gaussian-integral point ESP with convergence and work limits.
  • Shared CLI/Python/batch/report/MCP numerical results and source provenance.
  • Tested guided terminal tables and offline Chromium workbench controls/geometry.
  • Lower grid temporary memory, primitive-aware AO chunks, cached normalization, atomic streamed cubes, and optional trusted-workflow timeout/RSS/output checks.

MO/composition/Mayer/DOS/PDOS results are Validated for the documented eleven-molecule scope. Local MCP and the offline workbench interface are Stable within their tested scope. Failed source, conditioning, normalization or convergence diagnostics remain failed or partial/Experimental; data are never invented to obtain a successful result.

Evidence and resource boundaries

The scientific/resource implementation checkpoint passed four CI workflows, including Python 3.10–3.13, Linux/macOS/Windows, independent references, CLI/API/batch/report/MCP parity, wheels, Chromium, documentation and security. Its Linux resource benchmark passed 99/99 real workflows with peak observed tree RSS about 122 MiB, longest command about 1.74 s and maximum per-command output about 4.34 MiB. These are machine-specific measurements. The release publication additionally gates on all checks for the exact release commit and tests the newly installed wheel's real workflows.

Sampled RAM/output monitoring is not an OS hard quota or adversarial sandbox. Coarse density previews and explicit grid ESP retain convergence warnings. Read-only MCP filesystem and size/basis restrictions remain enforced.

Remaining gates

Native Hirshfeld, vibrational/excited-state spectra, genuine transition-data NTOs, analytic density derivatives and advanced topology/basin methods are not included. Source-output readers, Cartesian Löwdin cross-normalization comparison and broader browser/device/program coverage retain documented limitations. No schema version changes or blanket scientific promotions.

Report candidate issues with the exact version, input checksum, command, structured result and independent expected value.