bilby-dev/bilby: v2.6.0

Colm Talbot; Gregory Ashton; Moritz Hübner; Matt Pitkin; plasky; asb5468; Michael J. Williams; Rory Smith; Aditya Vijaykumar; SMorisaki; John Veitch; Nikhil Sarin; Duncan Macleod; Daniel Williams; MarcArene; JasperMartins; C P L Berry; Vivien Raymond; Ceciliogq; Ivan Markin; David Keitel; AlexandreGoettel; Lorenzo Pompili; Mick Wright; oliviawilk; noahewolfe; jacobgolomb; Shichao Wu; Rhiannon Udall; Michael Pürrer

doi:10.5281/zenodo.15918940

Published July 15, 2025 | Version v2.6.0

Software Open

bilby-dev/bilby: v2.6.0

1. University of Southampton
2. TNG Technology Consulting ANZ Pty Ltd
3. University of Portsmouth
4. Caltech
5. Canadian Institute for Theoretical Astrophysics
6. University of Glasgow
7. Cardiff University Gravity Exploration Institute (@cardiffgravity)
8. Potsdam University
9. Universitat de les Illes Balears
10. Max Planck Institute for Gravitational Physics
11. Utrecht University
12. LIGO Lab, Caltech
13. University of Rhode Island

We noticed a longstanding mis-definition in the GravitationalWaveTransient likelihood that caused the PSD to be incorrectly normalised when using time domain data with a Tukey window (https://github.com/bilby-dev/bilby/issues/869). By extension, the SNR and log-likelihood is overestimated. While the peak of the likelihood is in the correct place, the posterior is overly constrained and due to prior effects may not peak in the correct place.

The specific form of the previous version of the likelihood is

$$ \hat{p}(d | \theta) = \frac{p(d | theta)}{\beta} $$

where $\beta$ is the power loss factor in the Tukey window ($\beta = 1 - 5 \alpha / 8$).

The impact of this bug is most pronounced for short-duration signals, with a long turn on for the Tukey window. To reproduce the old behaviour, users can set the environment variable BILBY_INCORRECT_PSD_NORMALIZATION=TRUE.

Fixed

Fix PSD normalisation in GravitationalWaveTransient (https://github.com/bilby-dev/bilby/pull/967)

Files