Testing the performance and accuracy of the RELXILL model for the relativistic X-ray reflection from accretion disks

TL;DR

This study systematically evaluates the RELXILL model's ability to accurately recover black hole spin and ionization parameters from simulated NuSTAR X-ray spectra, highlighting the importance of fitting procedures and data handling.

Contribution

It provides a comprehensive analysis of RELXILL's parameter constraints, emphasizing the impact of fitting strategies, data binning, and statistical methods on model accuracy.

Findings

High confidence in parameter recovery at high reflection fraction and spin

Fitting results depend strongly on initial parameter estimates

Proper data binning and statistical choices are crucial for accurate analysis

Abstract

The reflection spectroscopic model RELXILL is commonly implemented in studying relativistic X-ray reflection from accretion disks around black holes. We present a systematic study of the model's capability to constrain the dimensionless spin and ionization parameters from $\sim$6,000 NuSTAR simulations of a bright X-ray source employing the lamppost geometry. We employ high count spectra to show the limitations in the model without being confused with limitations in signal-to-noise. We find that both parameters are well-recovered at 90% confidence with improving constraints at higher reflection fraction, high spin, and low source height. We test spectra across a broad range - first at 10$^6-$10$^7$ and then $\sim$10$^5$ total source counts across the effective 3-79 keV band of NuSTAR, and discover a strong dependence of the results on how fits are performed around the starting parameters, owing to the complexity of the model itself. A blind fit chosen over an approach that carries some estimates of the actual parameter values can lead to significantly worse recovery of model parameters. We further stress on the importance to span the space of nonlinear-behaving parameters like $log~\xi$ carefully and thoroughly for the model to avoid misleading results. In light of selecting fitting procedures, we recall the necessity to pay attention to the choice of data binning and fit statistics used to test the goodness of fit by demonstrating the effect on the photon index $\Gamma$. We re-emphasize and implore the need to account for the detector resolution while binning X-ray data and using Poisson fit statistics instead while analyzing Poissonian data.