Toward mapping turbulence in the intracluster medium IV. Using NewAthena/X-IFU and simulation based inference to constrain turbulence

TL;DR

This study advances turbulence inference in the intracluster medium by applying simulation-based inference (SBI) with SNPE to end-to-end NewAthena/X-IFU mock observations. By training neural density estimators on a fast surrogate model of turbulent velocity fields and their X-ray observables, it accounts for the stochastic nature of turbulence and measurement variance, providing robust posteriors for the 3D velocity power spectrum. The results show that the normalization of the power spectrum (σ) is the most tightly constrained parameter, while the spectral slope (α) remains weakly constrained due to sample variance, and the injection scale (L_inj) is only moderately constrained; this underscores the necessity of accounting for variance with SBI. The work highlights the importance of advanced modeling and multi-target observing strategies to unlock reliable turbulence measurements with X-IFU/NewAthena in the presence of substantial cosmic variance and instrumental effects.

Abstract

Context. The NewAthena mission planned for launch in the late 2030s will carry X-IFU, an integral field unit spectrometer that will obtain unique insight into the X-ray hot universe through its combination of spectral and spatial capabilities. Its high spectral resolution will allow a mapping of turbulent velocities of the hot gas in galaxy clusters, providing an unrivaled way to study the complex dynamics within galaxy clusters. Aims. This is the fourth in a series of papers aimed at forecasting the ability to investigate turbulence in the intracluster medium through the observation of the centroid shift caused by turbulent motions of the gas. In this paper we improve on previous methods by investigating the ability of simulation-based inference (SBI) to constrain the underlying nature of velocity fluctuations through the use of standard observational diagnostics, such as the structure function. Methods. We rely on a complex architecture of neural networks in order to model the likelihood and posterior distributions relevant to our case. We investigate its capability to retrieve the turbulence parameters on mock observations, and explore its capability to use alternative summary statistics. Results. Our trained models are able to infer the parameters of the intracluster gas velocity power-spectrum in independently simulated X-IFU observations of a galaxy cluster. We evaluated the precision of the recovery for different models. We show the necessity to use methods such as SBI to avoid an under-estimation of the sources of variance by comparing the results to our previous paper. We confirm that sample variance severely impacts the precision of recovered turbulent features. Our results demonstrate the need for advanced modeling methods to tackle the complexity of the physical information nested within future observations expected from X-IFU/NewAthena.

Figures (15)

• Figure 1: Centroid shift (top) and line broadening (bottom) maps extracted from a mock observation of 19 X-IFU pointings simulated with SIXTE, centered on a galaxy cluster with a turbulent ICM. The $R_{500}/2$ size of the cluster is shown as a circle on the maps.
• Figure 2: Similar figure as Fig. \ref{['fig:output_maps']}, showing expected values (inputs maps, see text) of centroid shift and line broadening free of any instrumental noise.
• Figure 3: Centroid shift (top) and line broadening (bottom) structure functions extracted from an observation of 19 X-IFU pointings simulated with SIXTE, with the predicted structure function distributions as percentile envelopes.
• Figure 4: Scatter plots of the binned differences between the output and input maps of centroid shift and broadening, as a function of the distance from the center of the cluster. The different radial bins used for modeling this difference are plotted as vertical lines. The distribution in each radial bin is shown as a violin plot, with the mean shown as a black line and $1\sigma$ values shown as blue lines.
• Figure 5: Retrieved average value and standard deviation of the marginalized posterior distribution of each parameter, obtained for 100 observations of 100 velocity field realizations generated with the same input parameters. The density estimator used SNPE with default parameters and was trained using the structure function of the centroid shift and the broadening.