A Deep Learning Approach to Galaxy Cluster X-Ray Masses

Abstract

We present a machine-learning approach for estimating galaxy cluster masses from Chandra mock images. We utilize a Convolutional Neural Network (CNN), a deep machine learning tool commonly used in image recognition tasks. The CNN is trained and tested on our sample of 7,896 Chandra X-ray mock observations, which are based on 329 massive clusters from the IllustrisTNG simulation. Our CNN learns from a low resolution spatial distribution of photon counts and does not use spectral information. Despite our simplifying assumption to neglect spectral information, the resulting mass values estimated by the CNN exhibit small bias in comparison to the true masses of the simulated clusters (-0.02 dex) and reproduce the cluster masses with low intrinsic scatter, 8% in our best fold and 12% averaging over all. In contrast, a more standard core-excised luminosity method achieves 15-18% scatter. We interpret the results with an approach inspired by Google DeepDream and find that the CNN ignores the central regions of clusters, which are known to have high scatter with mass.derstanding of the scatter in the mass-observable relationship.A number of mass proxies can be derived from Xray observations of galaxy clusters. X-ray cluster observations probe a portion of the baryonic component of clusters -the hot intracluster medium -which emits X-ray radiation primarily through bremsstrahlung. Hydrodynamical simulations (Nelson et al. 2014b;Hahn & Angulo 2016;Le Brun et al. 2017;Barnes et al. 2018;McCarthy et al. 2018) can be used to model observablemass relations in clusters.Cluster luminosity (L X ) is correlated with mass, and excising the inner ≈ 0.15 R 500c reduces scatter further arXiv:1810.07703v2 [astro-ph.CO]

Galaxy cluster mass estimation with deep learning and hydrodynamical simulations

“…9, we see that the F xs image has substructure at the top-right, which is also seen in the gradient image. The central regions appear to be relatively uninformative, in agreement with the conclusions of Ntampaka et al (2019). In addition, the lack of sensitivity of the central region with the DD images generally mimics the shape of the cluster itself (it is clear, for instance, in the lowest two rows of Fig.…”

“…We ran one iteration of DD for each data tracer, selecting images from a range of true cluster masses. (We also ran two iterations, as did Ntampaka et al (2019), and found similar results.) The gradient images for these examples are shown in the middle columns of Figs 8-11.…”

“…Our CNN is implemented using the keras package with a Tensorflow back end written in PYTHON. Our network architecture is similar to that used by Ntampaka et al (2019), which is a simplified version of that used by Simonyan & Zisserman (2014):(1) 3 × 3 convolution with 16 filters (2) 2 × 2, stride-2 average pooling (3) 3 × 3 convolution with 32 filters (4) 2 × 2, stride-2 average pooling (5) 3 × 3 convolution with 64 filters (6) 2 × 2, stride-2 average pooling (7) Flatten (8) Fully connected with 200 neurons (9) 10 per cent dropout (10) Fully connected with 100 neurons (11) 10 per cent dropout (12) Fully connected with 100 neurons (13) 10 per cent dropout (14) Fully connected with 20 neurons (15) Output neuron For our multichannel network, each data channel is convolved, pooled, and flattened separately, using the same architecture as the single-channel network. As shown in Fig.…”

“…Cohn & Battaglia (2020) and Armitage, Kay & Barnes (2019) use multiple machine-learning algorithms to estimate cluster mass from a set of observable quantities. Green et al (2019) use X-ray observational parameters, Ntampaka et al (2019) use mock X-ray images, and Gupta & Reichardt (2020) use simulated microwave sky to train neural networks to predict cluster mass. Here, we extend their work and utilize CNNs to predict cluster masses from stellar mass data, X-ray data, Compton y data, and from combinations of them.…”

Using X-Ray Morphological Parameters to Strengthen Galaxy Cluster Mass Estimates via Machine Learning

“…ML-based methods of estimating galaxy cluster masses from X-ray observations, including the method presented here as well as others in the literature (e.g., Ntampaka et al 2019b), offer a promising step towards extracting the maximum information content present in imminent datasets such as eROSITA. Modern ML methods will enable the completion of an unprecedentedly accurate cosmic census and position the halo mass function to be used to place ever stronger cosmological constraints.…”