v2.6.0

[ColmTalbot]

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 (#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 (BUG: remove incorrect window-based PSD scaling #967)

---

This discussion was created from the release v2.6.0.