Speaker
Description
The subtraction of compact-binary foregrounds is a central challenge for stochastic gravitational-wave background searches with third-generation detector networks. In particular, it remains unclear whether imperfect removal of individually resolved binary black hole signals could produce a residual foreground that limits sensitivity to cosmological or astrophysical stochastic backgrounds. Previous studies have reached differing conclusions, in part because population-scale analyses have often relied on Fisher-level approximations, while full Bayesian parameter estimation for $\mathcal{O}(10^5)$ events is computationally prohibitive. In this work, we develop a scalable full-posterior foreground-subtraction framework based on neural posterior estimation. Using a normalizing-flow parameter-estimation pipeline as a tractable surrogate for full Bayesian inference, we propagate event-level posterior uncertainty through the subtraction procedure at realistic population scale. We find that the residual from resolved binary black hole subtraction is reduced to a level well below the sensitivity of a third-generation detector network, indicating that such residuals do not constitute a sensitivity floor for stochastic-background searches in the scenarios considered. Our results demonstrate that neural posterior estimation can retain the essential posterior information required for realistic foreground subtraction while remaining computationally viable for next-generation observing campaigns. This framework provides a general methodological foundation for binary black hole foreground mitigation in future stochastic gravitational-wave background analyses.
| Research Area | Gravitational waves: black holes |
|---|