betaFeedback

WorksPhysics and Astronomy

StochAXtic: a JAX-Accelerated, End-to-End Differentiable Framework for Gravitational-Wave Backgrounds from Compact Binary Populations

Abstract

We present StochAXtic, a package for computing the dimensionless energy density spectrum ΩGW(f)Ω_{\mathrm{GW}}(f) of the stochastic gravitational-wave background generated by compact binary coalescences. Rather than sampling binary populations via Monte Carlo, StochAXtic exploits the scale invariance of the waveforms: at fixed mass ratio, spins and tidal deformabilities, the waveform depends on the total mass only through the dimensionless combination MfMf. This allows all computationally expensive waveform evaluations using LALSuite to be performed once and stored in a population-independent lookup table. The subsequent integrals over mass and redshift can then be recast as convolutions and evaluated efficiently using fast Fourier transforms. We further employ truncated singular value decomposition to compress the lookup table, reducing both computational cost and memory requirements. Once the table has been computed and cached, StochAXtic evaluates ΩGW(f)Ω_{\mathrm{GW}}(f) in a few milliseconds, achieving a relative numerical error below 10−310^{-3} at frequencies up to 11 kHz. Its results agree with those of popstock within the latter's statistical uncertainties. Implemented entirely in JAX, StochAXtic also provides gradients of ΩGW(f)Ω_{\mathrm{GW}}(f) with respect to population hyperparameters, enabling its direct integration with gradient-based inference methods including Hamiltonian Monte Carlo.

Paper details

vlnc:W_d4pfxuta
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
arXiv nonexclusive license
  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…