Extracting the X-ray reverberation response functions from the AGN light curves using an autoencoder
Sanhanat Deesamutara, Poemwai Chainakun, Tirawut Worrakitpoonpon, Kamonwan Khanthasombat, Wasutep Luangtip, Jiachen Jiang, Francisco Pozo Nuñez, Andrew J. Young · 2025
Read Extracting the X-ray reverberation response functions from the AGN light curves using an autoencoder by Sanhanat Deesamutara, Poemwai Chainakun, Tirawut Worrakitpoonpon, Kamonwan Khanthasombat, Wasutep Luangtip, Jiachen Jiang, Francisco Pozo Nuñez, Andrew J. Young (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-14618.txt.
Introduction
Sanhanat Deesamutara and colleagues use an autoencoder to extract X-ray reverberation signals from active galaxies. Their 2025 arXiv paper studies how X-rays echo off accretion disks. The echoes carry facts about the disk and the hot corona above it. Most past work analyzed these signals in the frequency domain. The team instead works directly from light curves.
A variational autoencoder was trained on simulated light curves. The simulations varied the height of the lamp-post corona. They included matching reflection signals from the disk. Later tests added disk fluctuations and random noise for realism. The model learned to rebuild the response profiles well. Results suggest machine learning can recover reverberation shapes from complex real data.
Catalog record
| CREATOR | Sanhanat Deesamutara, Poemwai Chainakun, Tirawut Worrakitpoonpon, Kamonwan Khanthasombat, Wasutep Luangtip, Jiachen Jiang, Francisco Pozo Nuñez, Andrew J. Young |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 51 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-14618
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Extracting the X-ray reverberation response functions from the AGN light curves using an autoencoder Sanhanat Deesamutara, Poemwai Chainakun, Tirawut Worrakitpoonpon, Kamonwan Khanthasombat, Wasutep Luangtip, Jiachen Jiang, Francisco Pozo Nuñez, Andrew J. Young Abstract We study the X-ray reverberation in active galactic nuclei (AGN) using the variational autoencoder (VAE), which is a machine-learning algorithm widely used for signal processing and feature reconstruction. While the X-ray reverberation signatures that contain the information of the accretion disk and the X-ray emitting corona are commonly analyzed in the Fourier domain, this work aims to extract the reverberation response functions directly from the AGN light curves. The VAE is trained using the simulated light curves that contain the primary X-rays from the lamp-post corona varying its height and the corresponding reflection X-rays from the disk. We use progressively more realistic light-curve models, such as those that include the effects of disk-propagating fluctuations and random noises, to assess the ability of the VAE to reconstruct the response profiles. Interestingly, the VAE can recognize the reverberation patterns on the light curves, hence the coronal height can be predicted. We then deploy the VAE model to the XMM-Newton data of IRAS 13224–3809 and directly estimate, for the first time, the response functions of this source in various observations. The result reveals the corona changing its height between 3 r g 3~r_{\rm g} and 20 r g 20~r_{\rm g} , which is correlated with the source luminosity and in line with previous literature. Finally, we discuss the advantages and limitations of this method. Unified Astronomy Thesaurus concepts: Reverberation mapping (2019); X-ray astronomy (1810); Active galactic nuclei (16); Black hole physics (159) I Introduction Active galactic nuclei (AGN) are the most luminous objects in the Universe, which are powered by an accretion of gas onto the central supermassive black hole. The accreting gas can form a disk-like structure referred to as an accretion disk that converts the gravitational energy into thermal optical/UV radiation. The disk photons gain more energy by Compton up-scattering with hot electrons inside the corona, raising the photon energy to X-rays [ 59 ] . The X-rays from the corona can travel back to the disk, be re-processed, and reflect off the disk in the form of a reflection spectrum, carrying out the characteristic features of the emission lines to an observer [ 29 , 62 , 61 , 28 , 27 ] . How much the Doppler and relativistic effects distort the lines determines how close the inner edge of the disk is to the black hole, hence allowing us to probe the innermost region of the AGN. The X-ray variations in AGN occur on a wide range of timescales [ 66 , e.g.] . The variabilities on long timescales relate to the fluctuations of the mass accretion rate that propagate inwards along the disk, modulating the outer (spectrally softer) regions first before the inner (spectrally harder) regions, hence producing the hard lags [ 42 , 3 ] . On shorter timescales, the fluctuations in the direct continuum cause the reflection to reverberate, so the changes in the reflection-dominated energy bands are expected to lag behind the changes in the continuum-dominated band. The amplitude of the reverberation lags relate to the light travel time between the corona and the reflecting region, providing us insights about the disk-corona geometry [see 65 , 7 , for a review] . Theoretical works in calculating the X-ray reverberation lags have been done early on [ 60 , 72 ] before the first hints of the reverberation were seen in Ark 564 [ 48 ] , followed by the first robust detection in 1H0707–495 [ 24 ] where the fluctuations in the soft (reflection dominated) energy band lagged behind those in the hard (continuum dominated) band by \sim 30 s. Since then, the number of AGN that exhibit the X-ray reverberation lags have increased [ 18 , 40 , e.g.] . While the nature of the corona is still uncertain, the dynamics of the corona were evident, e.g., in Mrk 335 [ 70 ] , IRAS 13224–3809 [ 2 , 6 , 11 ] , 1H 0707–495 [ 32 , 47 ] , and NGC 4051 [ 45 ] . By studying the time delays between UV and X-ray bands, a change in the geometry of the system (e.g. the height of the corona) was also suggested [ 44 ] . Meanwhile, the full-relativistic X-ray reverberation models were continuously developed based on the ray-tracing simulations that trace photon trajectories along the Kerr geodesics between the corona, the disk and the observer [ 69 , 8 , 21 , 14 , 23 , e.g.,] . In this way, we can calculate the flux of the reflection photons from different parts of the disk as seen by the observer and as a function of time after the direct continuum. This is known as the response function that can be used to generate realistic light curves based on the convolution theorem. The lags between the light curves in different energy bands then are estimated in the Fourier-frequency space [ 65 ] . Comparing the lags between the observations and the models is traditionally performed in the Fourier domain, either by fitting the lag-frequency spectra [ 2 , 5 , 6 , e.g.] or the power spectral density (PSD) obtained from the modulus square of the Fourier transform [ 22 , 11 , e.g.] . This work aims to develop machine learning (ML) models to extract the response functions from the AGN light curves so that they can be directly fitted with the theoretical response profiles. We employ the variational autoencoder (VAE) which is based on the multilayer perceptron (MLP) for signal reconstruction and reverberation feature extraction. Several applications of autoencoding architectures have been applied to astronomical problems such as SED feature extraction [ 26 ] , strong lensing identification [ 16 ] , and X-ray binary pattern recognition [ 52 ] . However, there was no previous work that has been applied to predict the X-ray reverberation response function. Thus, we are motivated by these previous studies to adopt the VAE to our problem and examine it in terms of performance and sensitivity. Simulated AGN light curves in several scenarios are produced to test the effectiveness of the VAE model in reconstructing the response functions. This is very challenging since the AGN light curves are known to be stochastic (i.e. different realizations of the same underlying physical process may look entirely different). This work can provide an independent tool to discover the underlying process straightforwardly in the time domain and to constrain the key parameters of the system. In addition, we also apply the model to IRAS 13224–3809 which is the well-known AGN that exhibits strong and variable signatures of the X-ray reverberation due to its dynamic corona, where the coronal height seems to increase with the source luminosity [ 2 , 6 , 11 , e.g.] . In brief, IRAS 13224–3809 has been identified as a Narrow-line Seyfert 1 galaxy [ 67 ] , located at redshift z of 0.06580; 1 1 1 Fiducial redshift obtained from http://ned.ipac.caltech.edu this is corresponding to the luminosity distance D_{\rm L} of 288 Mpc –