Speeding Up the Spinning, Precessing Effective One-Body--Numerical Relativity (SEOBNRv3) Code by ~10,000x

West Virginia University Research Corporation, Morgantown WV

Investigators

Abstract

When LIGO detects a gravitational wave, a group of data analysts within the LIGO Scientific Collaboration (LSC) addresses the question "What exactly produced this wave?", by employing a suite of tools designed to estimate source parameters of detected signals. Their findings extend our understanding of the universe, and inspire the next generation of scientists. The process of parameter estimation requires that the observed wave be compared to many millions of gravitational waves generated by theoretical models. However, this poses a major challenge: the software currently used for this task is far too slow, requiring up to 1,000 years to estimate parameters for a single detected gravitational wave. This grant will support an effort to improve this software platform performance by a factor of 10,000, which will greatly increase the usefulness of this state-of-the-art and widely-used model for LSC parameter estimation. The grant will also support the implementation of an automated software validation system for the LSC, to prevent certain types of software "bugs" from influencing parameter estimation. For cases where the sources of interest are spinning black-hole pairs, the LSC primarily makes use of two highly-reliable theoretical models based on supercomputer-generated solutions to Einstein's equations of General Relativity. One of these models is called SEOBNRv3 (short for Spinning Effective One Body-Numerical Relativity model, version 3). The SEOBNR series of gravitational waveform models are among the best available for parameter estimation (PE) of spinning black-hole binaries, filling gaps in perturbative waveform approximants with results from numerical relativity simulations. Despite their great efficiency and reliability when compared to numerical relativity waveforms, SEOBNR codes are still far too slow to be directly useful for standard Markov-Chain Monte Carlo (MCMC)-based PE. This award supports work to drastically improve the performance of the official SEOBNRv3 (v3) software. In addition to this effort, the award will support the education of graduate students in STEM areas of research.