JF Tools

Collection of small useful helper tools for Python by Johannes Feist.

Short Iterative Lanczos Backend Selection

The jftools.short_iterative_lanczos solver provides two backends:

• numba for array-based propagation.
• python for the reference implementation.

Select backend behavior with the backend function argument:

• auto (default): select numba only when both the Hamiltonian and the input state are compatible with the Numba implementation; otherwise fall back to python.
• python: force the original Python implementation.
• numba: force the Numba backend.

Current Numba compatibility is intentionally narrower than the Python backend:

• States must be numpy.ndarray (converted to complex128) or qutip.Qobj states.
• Hamiltonians may be dense NumPy arrays, SciPy CSR matrices/arrays, callable Hfun(t, phi, Hphi) operators working on array states, or the specialized sum-operator form (H0, (H1, f1), ...).

The Python backend uses the same NumPy/Qobj-style state handling as the reference implementation.

Example:

prop = jftools.short_iterative_lanczos.lanczos_timeprop(H, maxsteps=14, target_convg=1e-12, backend="auto")

Installation

jftools is a pure Python package again. There is no compiled extension build step.

Recommended (uv-only) workflow:

uv sync --group dev
uv run python -m pip install -e .

For a non-editable local install through uv:

uv run python -m pip install .

QuTiP Compatibility

Short iterative Lanczos targets modern QuTiP (5.x):

• QuTiP Hamiltonians are converted through H.data.as_scipy() when available.
• QuTiP state outputs preserve dims and state shape.

Publishing

Before publishing a new version:

1. Update __version__ in jftools/init.py.
2. Commit the version bump.
3. Build locally.

For a local packaging sanity check on your current machine only:

uv build