VARAHA: A Fast Non-Markovian sampler for estimating Gravitational-Wave posteriors

TL;DR

VARAHA is a novel, fast non-Markovian sampler that significantly accelerates gravitational-wave posterior estimation, enabling rapid analysis for multi-messenger astronomy and other applications.

Contribution

It introduces a new sampling algorithm that discards low-likelihood regions to efficiently estimate gravitational-wave posteriors, reducing computational time.

Findings

VaraHA localizes GW170817 30 times faster than LALInference.

It can estimate sky locations and source classifications within minutes.

The method reduces analysis time from days to minutes.

Abstract

This article introduces VARAHA, an open-source, fast, non-Markovian sampler for estimating gravitational-wave posteriors. VARAHA differs from existing Nested sampling algorithms by gradually discarding regions of low likelihood, rather than gradually sampling regions of high likelihood. This alternative mindset enables VARAHA to freely draw samples from anywhere within the high-likelihood region of the parameter space, allowing for analyses to complete in significantly fewer cycles. This means that VARAHA can significantly reduce both the wall and CPU time of all analyses. VARAHA offers many benefits, particularly for gravitational-wave astronomy where Bayesian inference can take many days, if not weeks, to complete. For instance, VARAHA can be used to estimate accurate sky locations, astrophysical probabilities and source classifications within minutes, which is particularly useful for multi-messenger follow-up of binary neutron star observations; VARAHA localises GW170817 $\sim 30$ times faster than LALInference. Although only aligned-spin, dominant multipole waveform models can be used for gravitational-wave analyses, it is trivial to extend this algorithm to include additional physics without hindering performance. We envision VARAHA being used for gravitational-wave studies, particularly estimating parameters using expensive waveform models, analysing subthreshold gravitational-wave candidates, generating simulated data for population studies, and rapid posterior estimation for binary neutron star mergers.