Multivariate Regression Analysis of Gravitational Waves from Rotating Core Collapse
We present a new multivariate regression model for analysis and parameter estimation of gravitational waves observed from well but not perfectly modeled sources such as core-collapse supernovae. Our approach is based on a principal component decomposition of simulated waveform catalogs. Instead of reconstructing waveforms by direct linear combination of physically meaningless principal components, we solve via least squares for the relationship that encodes the connection between chosen physical parameters and the principal component basis. Although our approach is linear, the waveforms' parameter dependence may be non-linear. For the case of gravitational waves from rotating core collapse, we show, using statistical hypothesis testing, that our method is capable of identifying the most important physical parameters that govern waveform morphology in the presence of simulated detector noise. We also demonstrate our method's ability to predict waveforms from a principal component basis given a set of physical progenitor parameters.
Additional Information© 2014 American Physical Society. Received 15 June 2014; published 8 December 2014. We acknowledge helpful discussions with and help from members of the LIGO Scientific Collaboration and Virgo Collaboration Supernova Working Group, in particular Sarah Gossan, I. Siong Heng, and Nelson Christensen. BE and RF are supported in part by NSF grant PHY-1205952. CDO is partially supported by NSF CAREER grant PHY-1151197, NSF gravitational physics grant PHY-0904015, The Sherman Fairchild Foundation, and the Alfred P. Sloan Foundation. Some of the computation performed towards the results presented here used NSF XSEDE computing resources under award TGPHY100033.
Published - PhysRevD.90.pdf
Submitted - 1406.1164v1.pdf