Statistically-informed deep learning for gravitational wave parameter estimation

Abstract We introduce deep learning models to estimate the masses of binary components black hole mergers, (m1,m2) , and three astrophysical properties post-merger compact remnant, namely, final spin, $a_\mathrm f$?> overflow="scroll">af frequency damping time ringdown oscillations fundamental $\ell = m 2$?> overflow="scroll">?=m=2 bar mode, $(\omega_\mathrm R, \omega_\mathrm I)$?> stretchy="false">(?R,?I(m1,m2,af,?<mml:mi five holes: GW150914 GW170104 GW170814 GW190521 GW190630 use PyCBC Inference directly compare traditional Bayesian methodologies for parameter estimation based distributions. results show network predict distributions encode physical correlations, data-driven median 90% confidence intervals are similar those produced gravitational wave analyses. This methodology requires single V100 NVIDIA GPU values within two milliseconds each event. network, tutorial its use, available at Data Learning Hub Science