Deep Neural Emulation of the Supermassive Black Hole Binary Population

• DOI: 10.3847/1538-4357/adb4ef
• Publication Date: 2025-03-19T06:32:58Z
• Authors: Nima Laal; Stephen R. Taylor; Luke Zoltan Kelley; Joseph Simon; Kayhan Gültekin; David Wright; Bence Bécsy; J. Andrew Casey-C...; Siyuan Chen; Alexander Cingora...; Daniel J. D’Orazio; Emiko C. Gardiner; William G. Lamb; Cayenne Matt; Magdalena S. Siwek; Jeremy M. Wachter

Abstract While supermassive black hole (SMBH) binaries are not the only viable source for low-frequency gravitational wave background (GWB) signal evidenced by most recent pulsar timing array (PTA) data sets, they expected to be likely. Thus, connecting measured PTA GWB spectrum and underlying physics governing demographics dynamics of SMBH is extremely important. Previously, Gaussian processes (GPs) dense neural networks have been used make such a connection being built as conditional emulators; their input some selected evolution or environmental binary parameters output emulated mean standard deviation strain ensemble distribution over many Universes. In this paper, we use normalizing flow (NF) emulator that trained on entirety distribution, rather than deviation. As result, can predict distributions mirror simulations very closely while also capturing frequency covariances in well statistical complexities tails, non-Gaussianities, multimodalities otherwise learnable existing techniques. particular, feature various comparisons between NF-based GP approach extensively past efforts. Our analyses conclude outperforms GPs ease computational cost training but fidelity distributions.