Assessing the data-analysis impact of LISA orbit approximations using a GPU-accelerated response model

TL;DR

This paper introduces 'fastlisaresponse', a GPU-accelerated tool for accurate LISA gravitational wave response modeling, demonstrating that current approximations can bias parameter estimation for loud signals, and emphasizing the need for more precise models.

Contribution

The paper presents a new GPU-accelerated code for the generic LISA TDI response, enabling efficient and accurate gravitational wave data analysis beyond common approximations.

Findings

Equal-armlength approximation introduces biases in parameter estimation.

The GPU-accelerated code significantly speeds up response computations.

Accurate modeling is crucial for high SNR source analysis.

Abstract

The analysis of gravitational wave (GW) datasets is based on the comparison of measured time series with theoretical templates of the detector's response to a variety of source parameters. For LISA, the main scientific observables will be the so-called time-delay interferometry (TDI) combinations, which suppress the otherwise overwhelming laser noise. Computing the TDI response to GW involves projecting the GW polarizations onto the LISA constellation arms, and then combining projections delayed by a multiple of the light propagation time along the arms. Both computations are difficult to perform efficiently for generic LISA orbits and GW signals. Various approximations are currently used in practice, e.g., assuming constant and equal armlengths, which yields analytical TDI expressions. In this article, we present 'fastlisaresponse', a new efficient GPU-accelerated code that implements the generic TDI response to GWs in the time domain. We use it to characterize the parameter-estimation bias incurred by analyzing loud Galactic-binary signals using the equal-armlength approximation. We conclude that equal-armlength parameter-estimation codes should be upgraded to the generic response if they are to achieve optimal accuracy for high (but reasonable) SNR sources within the actual LISA data.