AGN X-ray Spectroscopy with Neural Networks

TL;DR

This paper demonstrates that neural networks can efficiently estimate physical parameters from AGN X-ray spectra, achieving accuracy comparable to traditional spectral fitting but with significantly faster computation, especially when combined with PCA.

Contribution

The study introduces a neural network approach for AGN X-ray spectral analysis that is faster and avoids false minima, enhanced by PCA for improved accuracy and simplicity.

Findings

Neural networks achieve comparable accuracy to spectral fitting.

PCA improves parameter estimation accuracy.

Speed is increased by approximately three orders of magnitude.

Abstract

We explore the possibility of using machine learning to estimate physical parameters directly from AGN X-ray spectra without needing computationally expensive spectral fitting. Specifically, we consider survey quality data, rather than long pointed observations, to ensure that this approach works in the regime where it is most likely to be applied. We simulate Athena WFI spectra of AGN with warm absorbers, and train simple neural networks to estimate the ionisation and column density of the absorbers. We find that this approach can give comparable accuracy to spectral fitting, without the risk of outliers caused by the fit sticking in a false minimum, and with an improvement of around three orders of magnitude in speed. We also demonstrate that using principal component analysis to reduce the dimensionality of the data prior to inputting it into the neural net can significantly increase the accuracy of the parameter estimation for negligible computational cost, while also allowing a simpler network architecture to be used.