- The aim of this project is to determine the computational cost of applying the Bayesian methods of source parameter estimation to gravitational-wave searches for LIGO. By implementing parallel-tempered Markov-chain Monte Carlo (PTMCMC) with the Message Passing Interface (MPI) protocol, which distributes computational tasks across a cluster of parallel CPUs, one might in theory be able to complete data analysis in near real-time. The ability to complete parameter estimation analyses in near real-time would have important implications for searches, because parallel tempering also directly computes the evidence term and thus the Bayes factor, which allows us to select between models describing the data as either noise of noise plus signal. The Bayes factor is proposed as an alternative detection statistic that may be more robust than the signal-to-noise ratio (SNR), as using Bayesian evidences may result in fewer false dismissals of real signals. The SNR is cheap to compute, while the computational cost of calculating the Bayes factor is currently unclear. This project quantifies the cost of a search algorithm that uses the Bayesian methods of parameter estimation, with implications for the ways in which LIGO conducts data analysis.
DCC Version 3.3.0, contact
Document Database Administrators