Beyond Gauss? A more accurate model for LISA astrophysical noise
Riccardo Buscicchio (University of Milano-Bicocca) gave a KICC special seminar at the Institute of Astronomy, University of Cambridge on Tuesday 15th October. The talk was titled “Beyond Gauss? A more accurate model for LISA astrophysical noise” and was based on his two new papers published on arXiv: “A test for LISA foreground Gaussianity and stationarity. I. Galactic white-dwarf binaries” and “A test for LISA foreground Gaussianity and stationarity. II. Extreme mass-ratio inspirals”.
The Laser Interferometer Space Antenna (LISA) is a planned space-based gravitational-wave (GW) detector that is expected to be launched in the mid-2030s. One of the key challenges for LISA data analysis is the presence of astrophysical noise, also known as confusion noise. This noise arises from the superposition of GW signals from millions of unresolved sources, such as double white dwarfs (DWDs) in our Galaxy and extreme mass-ratio inspirals (EMRIs) in other galaxies.
Current LISA data analysis pipelines often assume that the astrophysical noise is Gaussian and stationary, meaning that its statistical properties do not change over time [2019JCAP…11..017C,2020CQGra..37u5017K,2020JCAP…07..021P,2021JCAP…01..059F,2023JCAP…04..066B,2024PhRvD.109d2001M,2021PhRvD.103j3529B,2023PhRvD.107l3531H]. However, Buscicchio’s work shows that these assumptions are not always valid.
Buscicchio’s team simulated the LISA data stream, including the signals from millions of DWDs and EMRIs. They found that the astrophysical noise exhibits significant non-Gaussianity and non-stationarity, particularly at low frequencies. The non-stationarity is induced by the motion of LISA as it orbits the Sun, which couples to the inhomogeneous distribution of sources in the sky.
To illustrate the non-Gaussianity and non-stationarity of the astrophysical noise, Buscicchio presented a toy model consisting of two periodic signals at 1 mHz and 5 mHz, a bunch of stochastic signals, and a region around 10 mHz where the number of sources drops below 1 [2410.08263].
They then applied a statistical test to the toy model data and found that it revealed both non-Gaussianity and non-stationarity. They also showed that the non-Gaussianity is washed away when the astrophysical noise is combined with the instrumental noise of LISA.
Buscicchio’s findings have important implications for LISA data analysis. First, they suggest that we need to develop more accurate noise models that account for the non-Gaussianity and non-stationarity of the astrophysical noise. Second, they highlight the importance of using robust statistical methods that are not sensitive to these deviations from the Gaussian and stationary assumptions.
Several groups are currently working on developing more sophisticated LISA data analysis pipelines that address these challenges [2301.03673,2405.04690,2403.15318v3]. These pipelines are based on blocked Gibbs sampling, a statistical method that allows us to break down the high-dimensional parameter space of the LISA data into smaller, more manageable blocks.
Buscicchio concluded his talk by discussing two specific reasons why it is important to account for the non-Gaussianity and non-stationarity of the astrophysical noise. First, it can help us to discover hidden Milky Way satellites. Second, it can prevent source misidentification, which could lead to biases in our estimates of cosmological parameters.
Overall, Buscicchio’s seminar provided a valuable overview of the challenges and opportunities associated with modelling astrophysical noise in LISA data. His work highlights the need for continued research in this area in order to ensure that we can extract the full scientific potential of this groundbreaking mission.