AstroAI Workshop 2026
Sara Gholamhoseinian
Calibrated Selection Functions for Binary Black Holes via Normalizing Flows
Presenter: Sara Gholamhoseinian (University of Massachusetts Dartmouth)
Title: Calibrated Selection Functions for Binary Black Holes via Normalizing Flows
Date/Time: Monday, June 15, 4:00 PM - 5:30 PM
Abstract: Accurate selection functions are essential for inferring the astrophysical population of binary black holes from gravitational-wave catalogs. Semi-analytic approaches based on post-Newtonian (PN) amplitude scaling and signal-to-noise ratio thresholds are widely used because they are fast and interpretable, but they neglect spin precession, higher modes, and merger-ringdown physics. We present a conditional normalizing flow trained on a large set of IMRPhenomXPHM waveform simulations in the LIGO-Hanford, Livingston, Virgo three-detector network at design sensitivity, which learns the full distribution of the optimal network SNR conditioned on the binary parameters. An analytic non-central χ² noise model maps this to the observed SNR, yielding calibrated, threshold-agnostic detection probabilities. We benchmark the flow against an exact analytic baseline, two linear-regression semi-analytic models, and a formula-free data-driven baseline that uses no PN extrapolation. The flow matches the exact baseline on calibration, recovers data-driven sensitive volumes across the chirp-mass and effective-spin plane, and corrects a systematic high-mass volume bias inherent to PN-extrapolated semi-analytic methods. The resulting selection function is fast to evaluate, differentiable, and applicable across detector configurations without retraining.