First-time assessment of glitch-induced bias and uncertainty in inference of extreme mass ratio inspirals

TL;DR

The paper addresses how streams of non-Gaussian glitches bias EMRI parameter estimation for LISA. It adopts a Fisher-matrix approach, validated against MCMC, to quantify glitch-induced biases and uncertainty inflation using shapelet glitches drawn from LISA Pathfinder data and Kerr EMRI waveforms. Key findings show that moderate glitch mitigation ($\rho_{\rm opt}\lesssim 90$) yields biases well below $1\sigma$ and only modest covariance inflation, while higher-SNR glitches ($\rho_{\rm opt}\lesssim 400$) can approach $1\sigma$ biases; overall, EMRI inference is more robust to glitches than short-duration sources. The study informs glitch-mitigation requirements for LISA EMRI science and highlights the utility and limits of the Fisher-mmatrix framework for rapid bias assessment in realistic noise environments.

Abstract

This work investigates the impact of streams of transient, non-Gaussian noise artifacts or "glitches" on the parameter estimation of extreme mass ratio inspirals (EMRI) in the Laser Interferometer Space Antenna (LISA). Glitches cause biased and less precise inference for short-duration signals such as massive black hole binaries, but their effect on long-lived sources such as EMRIs has not been quantified. Using simulated LISA observations containing injected EMRIs and streams of shapelet-based glitches drawn from the LISA Pathfinder catalog, we estimate the glitch-induced parameter biases and uncertainties through a Fisher-matrix-based analysis whose accuracy we verify with Markov-Chain Monte Carlo. We find that moderately mitigated glitch streams i.e. ones containing only glitches of up to moderate SNRs ($ρ\lesssim 90$) induce negligible to minor biases $[\sim0.04σ,\sim0.6σ]$ in the inferred EMRI parameters. In contrast, weakly mitigated glitch streams containing higher-SNR events ($ρ\lesssim 400$) can produce biases nearing $1σ$. These results demonstrate that, when compared to inference of other sources such as massive black hole binaries, EMRI inference is notably more robust to glitches. We stress that at least some amount of glitch modeling and mitigation remains essential for unbiased EMRI analyses in the LISA era.

Figures (6)

• Figure 1: Distribution of network optimal SNRs for individual LISA-injected glitches from the LPF catalog, computed using the second-generation SciRDv1 noise PSD. Vertical dashed lines indicate our adopted glitch mitigation thresholds at $\rho_{\mathrm{A,E}} = \{8,90,400\}$, corresponding approximately to the 50th, 75th, and 90th percentiles of the glitch $\rho_{\mathrm{A,E}}$ distribution.
• Figure 2: Each EMRI's count of runs in which a given parameter had the largest absolute normalized statistical fluctuation i.e. $\arg \max |\Delta\theta_{\text{glitches}}^i/\Delta\theta_{\text{stat}}^i|$. The dotted colors represent intrinsic parameters. By proportion, each bar is the same as $\mathbb{P}\left(\arg\max_i \frac{|\Delta\theta_{\text{glitches}}^i|}{\Delta\theta_{\text{stat}}^i}=i\middle|g\right)$. We found little to no variation in results across glitch mitigation levels, so the counts are combined across all mitigation levels. For the prograde and strong-field EMRIs, extrinsic parameters dominate, consistent with their weaker Fisher constraints and susceptibility to phase perturbations.
• Figure 3: Schematic diagram taken from PhysRevD.107.083019 of the LISA constellation showing the three spacecraft (S/C 1, S/C 2, S/C 3) and their respective moving sub-optical assemblies (MOSA). Each MOSA is labeled with indices $(i,j)$ where $i$ denotes the host spacecraft and $j$ indicates the spacecraft from which it receives laser light. Red arrows indicate the inter-spacecraft links with associated light-travel distances $D_{ij}$.
• Figure 4: Marginal posteriors of the prograde EMRI as obtained by MCMC alongside the FM-derived posterior. The data contained only the EMRI signal in zero noise. The close agreement validates our Fisher matrix calculations.
• Figure 5: Marginal posteriors of the strong-field EMRI as obtained by MCMC alongside the FM-derived posterior. The data contained only the EMRI signal in zero noise.