Assessing signal cross talk between extreme-mass-ratio inspirals and Galactic binaries in LISA data

TL;DR

This paper quantifies cross talk between a single EMRI signal and the population of Galactic binaries in LISA data using the Sangria dataset and FEW-based EMRI waveforms. It shows that resolvable Galactic binaries can mimic EMRIs, causing substantial parameter biases and false detections, while the unresolved Galactic foreground behaves as Gaussian noise once resolvable binaries are removed. The authors employ Bayesian inference with parallel-tempering MCMC to compare multiple data configurations, demonstrating that removing resolvable GBs is essential for reliable EMRI inference, whereas the stochastic GB foreground poses a manageable background. These results inform practical global-fit strategies for LISA, reinforcing the need to subtract resolvable Galactic binaries before EMRI searches and enabling robust EMRI science in the presence of GB confusion.

Abstract

The future space-based gravitational wave observatory, the Laser Interferometer Space Antenna, is expected to observe between 1-1000s extreme mass-ratio inspirals (EMRIs) per year. Due to the simultaneous presence of other gravitational wave signals in the data, it can be challenging to detect EMRIs and accurately estimate their parameters. In this work, we investigate the interaction between a gravitational signal from an EMRI and millions of signals from inspiralling Galactic white dwarf binaries. We demonstrate that bright Galactic binaries can contaminate the detection and characterization of EMRIs. We perform Bayesian inference of EMRI parameters after removing resolvable Galactic binaries and confirm an accuracy comparable to that expected in Gaussian noise.

Figures (10)

• Figure 1: EMRI and total Sangria signal (excluding MBHBs) vs frequency (log scale).
• Figure 2: Influence of the full population of Galactic binaries on EMRI parameter estimation. Corner plot shows Bayesian inference of the four most important binary parameters. The true point is marked in black. Contour lines mark the 50% and 90% credible regions. The posteriors for the noiseless run are given in blue for comparison.
• Figure 3: $\rho=(h^\text{EMRI}(\boldsymbol{\theta})| h_\text{resolvable}) / \text{SNR}_\text{EMRI}$ vs $\Delta M$, where $\boldsymbol{\theta} = (M_\text{true} + \Delta M, \mu_\text{true}, \dots)$. The matched filtering SNR does not show a clear dependence on the MBH mass and resembles a popcorn-like noise. Similar behavior is observed for other parameters.
• Figure 4: Top: $(h^\text{EMRI}|h_\text{GB})$; Bottom: Overlap$(h^\text{EMRI},h_\text{GB})$ for individual resolvable GBs plotted against each GB's frequency. The EMRI waveform is computed at $\boldsymbol{\theta}_\text{Sangria}$.
• Figure 5: Top: Cumulative matched filtering SNR, $\rho(t)$, of EMRI signal with parameters $\boldsymbol{\theta}_\text{Sangria}$ and all resolvable GBs. Bottom: Time-frequency plot for the $\boldsymbol{\theta}_\text{Sangria}$ point. The spectrogram demonstrates only TDI A and is displayed on a log scale. The blue and red points mark detected sharp jumps in the cumulative inner product of EMRI with individual resolvable GBs.