Bahamas: BAyesian inference with HAmiltonian Montecarlo for Astrophysical Stochastic background
Federico Pozzoli, Riccardo Buscicchio, Antoine Klein, Daniele Chirico
arXiv:2506.22542v1 Announce Type: new
Abstract: The LISA datastream will be populated by large instrumental and astrophysical noises, both potentially exhibiting long-term non-stationarities. Modelling and inferring on them is a challenging task, central for accurate signal reconstruction. In this paper, we introduce $texttt{bahamas}$, a codebase designed to characterize noises and stochastic gravitational wave backgrounds (SGWBs) in LISA. $texttt{bahamas}$ adopts a time-frequency data representation, based on the Short Time Fourier Transform, to accurately describe the signal temporal evolution and accommodate for the presence of data gaps. In addition, $texttt{bahamas}$ supports a variety of SGWB spectral models proposed in literature, enabling joint inference on them. Posterior sampling leverages No-U-Turn sampling an efficient variant of Hamiltonian Monte Carlo, inheriting the cross-hardware capabilities provided by NumPyro (CPU/GPU/TPU). We benchmark $texttt{bahamas}$ performances on a simple test case, and present ongoing developments to appear in future releases.arXiv:2506.22542v1 Announce Type: new
Abstract: The LISA datastream will be populated by large instrumental and astrophysical noises, both potentially exhibiting long-term non-stationarities. Modelling and inferring on them is a challenging task, central for accurate signal reconstruction. In this paper, we introduce $texttt{bahamas}$, a codebase designed to characterize noises and stochastic gravitational wave backgrounds (SGWBs) in LISA. $texttt{bahamas}$ adopts a time-frequency data representation, based on the Short Time Fourier Transform, to accurately describe the signal temporal evolution and accommodate for the presence of data gaps. In addition, $texttt{bahamas}$ supports a variety of SGWB spectral models proposed in literature, enabling joint inference on them. Posterior sampling leverages No-U-Turn sampling an efficient variant of Hamiltonian Monte Carlo, inheriting the cross-hardware capabilities provided by NumPyro (CPU/GPU/TPU). We benchmark $texttt{bahamas}$ performances on a simple test case, and present ongoing developments to appear in future releases.