openWFN 0.13.0

openWFN 0.13.0

Release date: October 12, 2026.

Version 0.13.0 is a stabilization release for existing scientific workflows,
terminal interfaces, exports and documentation.

What changes

This release focuses on existing workflows rather than adding scientific methods.
Guided file selection, settings, navigation and saves have regression coverage.
CLI results use readable tables and short explanations; JSON keeps the full
result, units, warnings, scientific status and input provenance.

The terminal assistant handles supported short file questions directly and
improves clarification, streamed responses and cancellation. General scientific
conversation still needs a configured model. Model explanations are not
independently verified, and model response speed depends on the runner and hardware.

Density integration and density, orbital and NCI cube exports process bounded
chunks instead of retaining all coordinate arrays. Grid ceilings and numerical
checks remain in place. Batch scheduling reduces requested workers when its
memory estimate requires it; this estimate is not an operating-system memory quota.

Switching offline workbench views clears the previous fields and measurements.
Reader edge cases, reports, exports and interface consistency have additional
regression tests. Documentation uses simpler instructions, corrected examples
and links, and improved narrow-screen layouts.

What the checks establish

The stabilization work was checked through the full Linux software suite,
cross-platform CI, independent reference jobs, installed-wheel workflows and
offline browser tests. Larger molecular and source-property examples supplement
the packaged everyday-QC corpus. The publication workflow must repeat its
installed-wheel and public-PyPI checks for this version before release is complete.

These checks cover the named fixtures and reference scope. They do not prove that
every method works for every molecule or input file. A reader cannot recover
records that the source file does not contain. Run capabilities first and inspect
warnings, numerical controls and convergence before using a result in research.

Scientific limits

Existing method-specific validation boundaries remain unchanged. In particular,
QTAIM, ELF/LOL, NCI, spectroscopy and excited-state analyses retain their documented
limits; no basin integration or universal validation claim is added.

Missing data remain unavailable, and conservation failures remain partial.
Density integration still needs grid convergence checks. An orbital-energy gap
is not an optical excitation energy.

Normal installation includes the supported readers, Python API, local MCP and
terminal assistant. Model weights and Gaussian's external formchk utility are
not bundled. Existing standalone HTML exports must be regenerated to receive
the workbench fixes.

openWFN 0.12.0

openWFN 0.12.0

Release date: October 7, 2026.

Using the toolkit

The normal installation includes supported format readers, output-property
extraction, Python analysis, local MCP and terminal chat. You do not need to
select extras for these interfaces. Existing extra names still work.

openwfn FILE opens guided mode in a terminal. openwfn FILE analyze gives a
bounded overview without population or grid calculations. Outside a terminal,
a file-only command now returns that overview; scripts relying on the old
summary schema should use explicit summary.

Guided mode shows eligible workflows, units and indexing, allows settings to be
changed, shows reproducible commands, and confirms save paths. Failed exports
leave earlier output files intact. Input aliases cannot be replaced by rendered
results, even with --overwrite.

openwfn chat opens with or without a file or model connection. /connect offers
local-runner download links and model selection; /models switches models and
/disconnect clears the connection without stopping the runner. Connections and
history are session-only. No model is downloaded automatically.

openwfn FILE chat --model MODEL can use a configured model to select tools.
Supported short file questions also work without a model. openWFN supplies the scientific facts. Capabilities,
input hashes, status, warnings, units and provenance stay attached to the result.
Remote endpoints require explicit permission; no model weights, paid calls or
public servers are activated by installation.

General scientific conversation uses the selected model, is clearly labelled,
and is not independently verified. No automated citation verification is
included. Model response time varies; a connected server is not proof of fast
inference. Ctrl+C cancels the current request and returns to the prompt.

The interactive CLI, chat, README, website and offline HTML exports use the
shared pixel wordmark. Machine-readable output stays banner-free; plain,
no-colour and narrow-terminal modes have compact fallbacks.

The read-only MCP interface shares these limits and checks. It adds density
integration after approval of settings and explicit-point ELF/LOL/NCI and
derivative requests. External chat hosts control their own explanations and
need separately configured connections; no public plugin listing is claimed.

Scientific scope

The numerical engines and validation tolerances are unchanged. Scientific
status is method- and fixture-specific, not a consequence of the package version.
The validation matrix is the authority.

QTAIM critical-point search, ELF/LOL, NCI, spectroscopy and output extraction
retain their documented Experimental boundaries. QTAIM searches are bounded
and non-exhaustive; basin integration is not included. Missing wavefunction data
are not reconstructed from structures, grids or incomplete logs.

MOL/SDF charge parsing now follows the property-record precedence described in
the CTfile specification.
Structure-only electronic multiplicity remains unknown unless a supported source
record provides it. Binary .chk conversion still requires external Gaussian
formchk; that utility is not installed by pip.

Verification

Local source, clean-wheel CLI/API/MCP, real-file and configured local-model
checks are recorded before publication. The local source suite passed 1,303
tests with 60 skipped; skipped optional scientific references are not passes.
The local macOS sandbox cannot enumerate processes for the resource matrix;
the Linux installed-wheel matrix is a required CI/publication gate instead.
Exact-commit CI, platform checks,
distribution checks and a public-PyPI reinstall are required by the release
workflow. Skipped optional scientific-reference tests are not reported as passes.

openWFN 0.11.0

openWFN 0.11.0

Release date: October 4, 2026.

openWFN 0.11.0 is a stable software release that integrates three major feature lines into the public package: native Hirshfeld population analysis, vibrational spectroscopy, and method-general excited-state/UV–Vis post-processing. Package stability does not promote every scientific method to the same validation status; each analysis retains its documented evidence boundary.

Highlights

Native Hirshfeld populations and charges

• Adds ordinary neutral-pro-atom Hirshfeld populations and charges through the shared analysis engine.
• Uses packaged, reproducible H/C/N/O pro-atom reference data with provenance and integrity checks.
• Reports electron/population/charge closure diagnostics and does not renormalize final charges to hide integration residuals.
• Rejects unsupported elements, ECP/pseudopotential ambiguity, and ghost-center ambiguity instead of guessing a fallback.
• Independently compared against HORTON-PART for the named ten-case H/C/N/O all-electron validation set.
• Scientific status: Validated only for the documented H/C/N/O all-electron scope and fixed tolerances; unsupported chemistry remains explicit.

Vibrational spectroscopy

• Adds typed Gaussian harmonic vibrational modes, signed imaginary frequencies, source IR intensities, source Raman activities, and source normal-mode vectors.
• Adds deterministic Gaussian broadening, IR/Raman CSV and figure export, CLI/Python/MCP parity, report integration, and a dedicated offline Workbench workspace.
• Raman activities are preserved as source activities and are not silently converted to experimental Raman intensities.
• Scientific status: Experimental. Parser/regression/interface evidence is not yet an independent external spectroscopy validation campaign.

Excited states and UV–Vis

• Adds a universal typed excited-state/job-block model with deterministic multi-job isolation.
• Adds source adapters for Gaussian, ORCA, and Q-Chem.
• Preserves source method detail while classifying broad method families and amplitude conventions conservatively.
• Adds excited-state tables, one-state inspection, transition dipoles, source oscillator-strength sticks, and deterministic UV–Vis broadening.
• Uses a documented default Gaussian FWHM of 0.20 eV in energy space and applies the proper energy-to-wavelength Jacobian for wavelength-domain profiles.
• Preserves dark states (f = 0), distinguishes missing oscillator strengths from zero, and excludes invalid/negative optical values from simulated curves with explicit reasons rather than clamping or inventing data.
• Adds JSON/CSV/PNG/SVG export, HTML report integration, and an Excited States workspace in the offline Workbench.
• Scientific status: Experimental. Recognition of a source state is not a claim that every underlying TDDFT/EOM/ADC/multireference implementation is independently validated.

Interfaces

The new capabilities use the same scientific engine across public surfaces:

• command line;
• typed Python API and named analysis registry;
• local read-only MCP analysis routing;
• CSV/JSON/PNG/SVG exports;
• HTML research reports;
• offline Workbench workspaces.

Representative commands:

openwfn calculation.fchk population hirshfeld
openwfn frequency.log vibrations
openwfn frequency.log spectra ir --fwhm 20 --points 2001 --export ir.csv
openwfn excited.log excited states
openwfn excited.log excited state 1
openwfn excited.log excited dipoles
openwfn excited.log spectra uvvis --fwhm-ev 0.20 --export uvvis.svg

Reproducibility and release verification

The stable publication workflow remains fail-closed and version-driven. Before the 0.11.0 tag or PyPI upload is created, the exact release commit must pass the repository test, documentation, security, and offline Chromium workflows. The workflow then:

1. checks synchronized release metadata;
2. builds and validates wheel and source distributions;
3. installs the built wheel in a fresh environment with interoperability/resource extras;
4. installs the packaged eleven-molecule corpus;
5. runs the 110-command bounded resource/workflow matrix against the installed package;
6. creates the annotated v0.11.0 tag;
7. publishes to PyPI through OIDC trusted publishing;
8. creates the GitHub stable release for the same exact commit;
9. installs openwfn==0.11.0 back from public PyPI in a second clean environment and reruns the same 110-command matrix.

Command/resource completion and stable packaging are evaluated separately from scientific validation status.

Compatibility and support

• Python 3.10–3.13.
• Existing 0.10 public analysis/result schemas remain supported unless a capability explicitly exposes new additive fields.
• Optional IOData interoperability remains pinned to the tested 1.0.1 contract.
• Optional cclib source-output extraction remains Experimental and is not a substitute for the typed excited-state adapters.
• Gaussian formatted-checkpoint files remain the preferred native full-wavefunction input path.

Install

python -m pip install --upgrade openwfn

For an exact research environment:

python -m pip install "openwfn[interop,resources]==0.11.0"

Review the validation matrix and method-specific limitations before using any quantity for research reporting. Stable software packaging does not erase Experimental or Unsupported scientific boundaries.

openWFN 0.10.1

openWFN 0.10.1

Release date: October 2, 2026. This is a release-quality and reproducibility patch for the 0.10 stable line. It does not change scientific numerical kernels, formulas, validation tolerances, model schemas or result schemas.

What changed

• Replace development-era validation prose with a current-state evidence page.
• Extend the canonical validation/manifest.json so current capability status, historical captures and resource-benchmark meaning are machine-readable in one place.
• Mark the earlier three-molecule regression snapshot and grid-ESP convergence capture as historical evidence rather than current status authorities.
• Package the complete eleven-molecule everyday-QC Molden corpus with the wheel and install it under everyday-qc/ through the existing openwfn examples install command.
• Make the resource benchmark consume an explicit example directory and hash the actually imported openWFN package, allowing built-wheel and public-PyPI verification to test the installed artifact rather than repository source files.
• Resolve explicit relative example-directory paths to absolute paths before per-workflow temporary working-directory changes, so installed-corpus verification uses the intended 11 inputs.
• Require the complete eleven-input corpus and all 99 expected workflow records; an empty or incomplete corpus cannot pass.
• Derive release version, tag and release-note path from package metadata rather than embedding a previous version in the publication workflow.
• After publication, install openwfn[interop,resources] from public PyPI and rerun the same installed-corpus real-workflow benchmark used before publication.

Resource evidence wording

The release gate records that 99/99 prescribed CLI workflows completed successfully within resource limits; scientific result status is evaluated separately.

Those 99 runs are eleven real molecular inputs multiplied by nine prescribed command workflows. They test command completion and bounded resource behavior. They do not convert a partial or Experimental scientific result into a Validated one. Capability-specific reference tests and per-result diagnostics remain authoritative for scientific status.

Installed reproduction corpus

python -m pip install "openwfn[interop,resources]==0.10.1"
openwfn examples install installed-examples
python scripts/benchmark_resources.py \
  --examples-dir installed-examples/everyday-qc \
  --output resource-report.json

The packaged Molden files are byte-identical copies of the versioned examples/everyday-qc/ inputs used by repository validation.

Scientific scope

The scientific scope remains the documented 0.10 everyday-QC scope. Successful orbital-composition, Mayer, DOS, PDOS and native point-ESP results retain their documented validation status and reference boundaries. Conditional failures in normalization, conservation, conditioning or convergence remain partial or Experimental instead of being promoted by the package-level stable label.

Native Hirshfeld, spectroscopy/excited-state workflows, NTOs, analytic density derivatives and advanced topology/basin methods remain outside this release.

openWFN 0.10.0

openWFN 0.10.0

Stable release dated October 2, 2026. This release supersedes 0.10.0rc1 and
expands everyday QC interpretation and stabilizes the tested
automation and resource workflows. It does not claim completion of every
scientific phase.

Install

python -m pip install "openwfn[interop,resources]==0.10.0"
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 issues with the exact version, input checksum, command,
structured result and independent expected value.

Stable release scope

This promotes the tested candidate implementation without scientific engine,
model/result schema or validation-label changes. Stable describes the package's
supported workflows. Experimental source readers, partial/convergence results
and the remaining scientific gates above keep their explicit status.
The candidate notes and previous release histories are retained.

openWFN 0.10.0rc1

openWFN 0.10.0rc1

Release candidate dated October 2, 2026. The stable release remains 0.9.2.
This candidate expands everyday QC interpretation and stabilizes the tested
automation and resource workflows. It does not claim completion of every
scientific phase.

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.

openWFN 0.9.2

openWFN 0.9.2

Release date: October 1, 2026. This patch addresses summary and CLI usability
issues. It adds no analysis methods or broader scientific validation claims.

Install

python -m pip install --upgrade openwfn==0.9.2
python -m openwfn.cli --version

For optional interoperability, QC output properties, and local MCP:

python -m pip install --upgrade "openwfn[interop,outputs,mcp]==0.9.2"

Corrections

• Structure-only summaries use reported element identities for formula and atom
  count even when effective nuclear charges are missing. Explicit ghost centers
  are excluded. Unknown ghost/ECP classification and electronic properties stay
  unknown; the result remains partial.
• Human-readable CLI output separates analysis validation, result status, and
  source-job termination. FCHK output does not establish source convergence.
• Point ESP explains nuclear, atomic-charge, and density-grid singularities.
  Zero-charge ghost centers do not create false nuclear singularities.
• Property arrays longer than eight entries are summarized in human-readable
  output. Diagnostics and warnings remain visible; --verbose, JSON, CSV, and
  Python API results retain full values. JSON field names are unchanged.

Verification and boundaries

The patch adds 12 regression cases. Local verification: 608 passed, 2 skipped,
plus lint, distribution tests, and a strict documentation build. A real XYZ
summary was checked for CLI/API/stdio MCP parity; a real ORCA output was checked
for compact presentation with complete API coordinates.

MCP, output-property extraction, the workbench, and grid-based electronic ESP
remain Experimental. These checks do not establish support for every molecule,
program version, or job type. See the 0.9.1 notes for the previous
resource and export safety corrections.

openWFN 0.9.1

openWFN 0.9.1

Release date: October 1, 2026. This is a corrective patch release; it does not
add new analysis methods or broaden scientific validation claims.

Install

python -m pip install --upgrade openwfn==0.9.1
python -m openwfn.cli --version

For optional formats, output properties and local MCP:

python -m pip install "openwfn[interop,outputs,mcp]==0.9.1"

Corrections

• Molecular grids above 2,000,000 points are rejected before allocation. AO
  evaluation chunks shrink for larger bases. Increase spacing or reduce padding
  when a request is rejected; settings are never silently changed. Check density
  conservation and convergence after changing them. This is not a guarantee
  that every calculation fits on every machine.
• Molden ingestion recognizes Psi4-style .molden.input filenames and the
  [Molden Format] header on otherwise unrecognized inputs. Batch discovery
  recognizes the compound extension.
• Atomic-charge ESP retains population conservation warnings and partial status.
• HTML/Markdown reports display warnings and partial status. Workbench payloads
  retain scientific status and warnings; the viewer labels ghost centers and
  displays field warnings. Files without a LUMO retain a usable HOMO field.
• Structure exports warn when charge or spin multiplicity will be lost. They
  remain structure exports, not electronic-state round-trips.
• Python distance, angle and dihedral methods accept isolated structure data
  without a complete wavefunction. Periodic minimum-image geometry is not provided.

Verification and boundaries

The patch adds 23 regression cases, including a safe pre-allocation check,
CLI JSON/API resource errors and generated viewer behavior. Local verification:
596 passed, 2 skipped, plus lint, packaging tests and a strict documentation build.

MCP, source-reported output extraction, the workbench and grid-based electronic
ESP remain Experimental. Ingestion support does not imply availability of every
analysis. Validated results remain scoped to their named fixtures and tolerances.

See the 0.9.0 notes for interoperability and API migration details.

openWFN 0.9.0

openWFN 0.9.0

Release date: October 1, 2026. This release adds capability-aware format
ingestion, mixed-format workflow hardening, and optional Experimental interfaces.

Install

python -m pip install "openwfn[interop,outputs,mcp]==0.9.0"
openwfn --version

The optional extras are separate. The base installation does not require
IOData, cclib, or the MCP SDK. Install only the extras required by your workflow.

Support boundaries

Workflow	Status and required data
Native FCHK CLI/API analysis	Existing interfaces retained; each analysis requires the corresponding source records. Numerical validation remains fixture-scoped.
Additional format ingestion	Stable fixture contract for the pinned 25-format inventory, using IOData 1.0.1 where required. Not a guarantee for every file variant.
Structure-only inputs	Partial summaries and supported structural operations; no invented orbitals, populations, or density.
Mixed-format batches	Capability-aware results and explicit unsupported/failed inputs. Format hints/maps may be needed for ambiguous files.
QC output properties / read_output()	Experimental source-reported extraction through cclib 1.8.1; missing properties remain null or empty. Not a full wavefunction or registered batch analysis.
Local MCP	Experimental stdio adapter using SDK 2.2.0. Inspection, registry analyses, and output properties only. Binary .chk inputs are rejected.
Periodic and integral-only inputs	Preserve supported components; general molecular electronic analyses are not promised for these inputs.
REST or remote MCP service	Not provided. No public upload or authentication service is included.

See the format matrix,
output properties, and MCP.

Python migration

load(path).data is now an OpenWFNData container. Scripts that accessed
low-level calculation fields directly should use .data.calculation and check
that it is not None. Prefer the high-level methods and file-specific
capability inspection for automation. Result schema 1.0 remains in use.

The new properties command reads QC logs independently of load(). A log may
report frontier energies or atomic charges without containing the coefficients,
basis, and density matrices needed to recompute those quantities.

Verification

The release preparation recorded 573 passing local tests and two skips. CI on the
merged source passed Python 3.10–3.13 tests, macOS/Windows smoke checks,
interoperability contracts, scientific validation, optional-interface tests,
and wheel installation. Final metadata and distribution checks are required before publication.

Five water representations come from one originating calculation and test
conversion consistency. Selected real ORCA, Q-Chem, and Gaussian outputs were
compared with source values. Neither establishes universal molecule, method,
or file-version coverage. Quantitative density work still requires grid
convergence and inspection of conservation diagnostics.

Citation and provenance

The software citation identifies 0.9.0 and its release date. No paper, DOI,
affiliation, or unverified authorship details are added. Preserve the exact
software version, input checksum, source program, and numerical settings
with each research result.

openWFN 0.8.2

What's Changed

• Fix post-release scientific edge cases by @sha786muhammed in #30
• Release openWFN 0.8.2 by @sha786muhammed in #32

Full Changelog: v0.8.1...v0.8.2