Varaha: A promising sampler for obtaining gravitational wave posteriors

Vaibhav Tiwari

2024-05-26

Abstract:Nested sampling has been the workhorse for estimating parameters of compact binaries from the gravitational wave signals. It operates using a set of live points and accepts/rejects points drawn from prior such that the live point with the lowest likelihood is replaced by a point with a higher likelihood value. Each such iteration shrinks the volume enclosed by the live point of lowest likelihood by a known factor. The estimated sampling density along with the likelihood values of the new and old live points quantifies the probability mass enclosed by them. Although robust, nested sampling often discards a majority of the sampled points~($\sim 99.9\%$) at which likelihood was calculated. However, for small dimensional problems, which happens to be the case for compact binaries, the sampling density of all the sampled points can be explicitly calculated thereby removing the need to discard samples. The points' sampling density and likelihood values constitute the posterior distribution. Avoiding the wastage implies significant improvement in the sampling efficiency. In a previous method, we presented a recipe to reconstruct the live volume; the volume that encloses most of the posterior probability mass. We build on this method and present a methodology that explicitly records the sampling density of all the points. For small dimensions, the presented sampler is significantly more efficient than the canonical nested samplers and is embarrassingly parallel.

High Energy Astrophysical Phenomena