minke 2.2.1

2.2.1

This is a bug-fix release. Injections made by earlier releases had incorrect SNRs and noise levels, and WaveformDict.project applied the inclination factors twice for lalsimulation models, so results produced with 2.2.0 and earlier should be regenerated. It also fixes the GWF frame-writing failure which the 2.2.0 workaround did not resolve, and removes the dependency on puddin.

Bug fixes

Optimal SNR
calculate_network_snr_for_distance and make_injection truncated the FFT to N // 2 points and used df = 1 / sample_rate, so the PSD was evaluated at 0-16 Hz and SNRs came out roughly 100x too low (the 2.2.0 "fix" for this did not resolve it). A new minke.filters.optimal_snr_squared computes the standard :math:4 \int |h(f)|^2 / S(f) \, df over the in-band frequencies, matching bilby and simple-pe to 0.1%.

Noise normalisation
LALSimulationPSD.time_series produced noise with power :math:4/T times the requested PSD (1/16 for a 64 s segment), and evaluated the PSD 20 Hz too high because the LAL frequency series started at lower_frequency while the frequency axis started at 0. The generated noise now reproduces the requested PSD independent of segment length. The Nyquist bin is only zeroed for even-length series.

Projection
WaveformDict.project applied the :math:(1 + \cos^2\iota) and :math:\cos\iota factors a second time for models which already evaluate :math:h_+ and :math:h_\times at an inclination (those with an inclination parameter, i.e. lalsimulation models). Only the antenna response and the phase mixing are now applied to such waveforms, and theta_jn is no longer required when inclination is given.

GWF frame writing
The 2.2.0 epoch workaround had no effect with gwpy 4, and roughly half of all epochs still failed with "Series start time is earlier than frame start time". _write_gwf_epoch_safe now creates the frame directly with LALFrame, using the epoch of the LAL series as the frame start time, so the two cannot disagree.

Python and packaging
minke.__version__ is now read with importlib.metadata rather than pkg_resources, which is no longer shipped with recent setuptools. minke.asimov now imports importlib.resources explicitly, which fixes an AttributeError on Python 3.10.

Changes

puddin is no longer required
minke.bagpuss.read_injection_parameters now calls lalsimulation.SimInspiralTransformPrecessingNewInitialConditions directly for the spin frame transformation instead of puddin.lalsim.spins_to_lalsim. The bagpuss tutorial no longer asks for puddin to be installed.

Tests
Added tests for the optimal SNR (against an independent LAL PSD reference), noise PSD recovery, projection, and GWF round-tripping. Updated the make_injection and cache-file tests, which predated make_injection returning (injections, frame_files, network_snr) and the <ifo>_<framefile>.cache naming.

Known issues

Lalsimulation models only accept inclination; an iota key (as returned by minke.bagpuss.read_injection_parameters) is silently dropped, giving face-on waveforms. minke.bagpuss also passes m1_source/m2_source without applying (1 + z).