Skip to content

openWFN 0.8.0a2

Historical alpha notes. Version 0.8.0a2 improved installation, batch runs and release checks. It added no scientific methods or input formats.

Installation channels

Install the current stable release:

python -m pip install --upgrade openwfn

Install the newest pre-release:

python -m pip install --pre --upgrade openwfn

After publication, pin this alpha for a reproducible environment:

python -m pip install openwfn==0.8.0a2

These setup notes describe the alpha at the time it was released. The online handbook now follows main; use Installation for current setup.

Installed first analysis

The wheel now contains the maintained water fixture and its provenance record. It can be copied without a source checkout:

openwfn examples install ./openwfn-examples
openwfn ./openwfn-examples/water.fchk summary

Existing files are not replaced unless --overwrite is supplied.

Parallel batch behavior

Multi-worker batches now use a bounded completion-driven queue. A finished input is persisted atomically without waiting for an earlier slow input, while the final JSON manifest and CSV index remain in deterministic input order. Unexpected worker failures stop the run instead of writing an apparently complete manifest.

Benchmark evidence

The repository benchmark runs real parsing, summary analysis, and result writing over deterministic structure inputs. A full Linux validation host run uses:

python scripts/benchmark_batch.py \
  --count 10000 \
  --workers "$(nproc)" \
  --workspace /tmp/openwfn-benchmark \
  --output openwfn-benchmark-10000.json

The output is JSON. CI checks correctness at a smaller count and does not use a timing threshold.

CLI startup

Plotting imports are now lazy. openwfn --version, inspection commands, and batch commands do not initialize Matplotlib unless a plot is requested.

Scientific boundary

The capability states are unchanged. Methods documented as Validated keep their named evidence scope; Experimental and Unsupported capabilities remain explicitly labeled. Gaussian formatted-checkpoint data remains the primary full-wavefunction input for this alpha.

See the 0.8.0a1 notes for the model, result-envelope, parser, and scientific changes introduced by the first 0.8 alpha.