Measuring the Vertical Structure of Active Galactic Nuclei Disks with Transformer Models and the Vera C. Rubin Observatory
Amy Secunda, Sebastian Wagner-Carena, Helen Qu, Shirley Ho · 2026
Read Measuring the Vertical Structure of Active Galactic Nuclei Disks with Transformer Models and the Vera C. Rubin Observatory by Amy Secunda, Sebastian Wagner-Carena, Helen Qu, Shirley Ho (2026) 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-2601-08926.txt.
Introduction
Amy Secunda and three co-authors use AI to weigh the height of disks round huge black holes. Light from such a disk rings back with a lag. The lag by hue tells us the span of the disk. The team wants the height as well, which is far more hard to read from lags. A new huge sky survey will soon give rich lag data for many disks.
They train a transformer model, a net that weighs which past points count, on sim lag data. The model learns to read height signs in the lag shape. A reader in space work gets a tool set for the new survey. It may turn lag lists into full disk shapes for a vast set of black holes.
Catalog record
| CREATOR | Amy Secunda, Sebastian Wagner-Carena, Helen Qu, Shirley Ho |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-08926
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Measuring the Vertical Structure of Active Galactic Nuclei Disks with Transformer Models and the Vera C. Rubin Observatory Amy Secunda, Sebastian Wagner-Carena, Helen Qu, Shirley Ho Abstract Reverberation mapping is one of the main techniques used to study active galactic nuclei (AGN) accretion disks. Traditional continuum reverberation mapping uses short lags between variability in different wavelength AGN light curves on the light crossing timescale of the disk to measure the radial structure of the disk. The harder-to-detect long negative lag measures lags on the longer inflow timescale, opening up a new window to mapping out the vertical structure of AGN disks. The Vera Rubin Observatory, with its 6 wavebands, long baseline, and high cadence, will revolutionize our ability to detect short and long lags. However, many challenges remain to detect these long lags, such as seasonal gaps in Rubin light curves, the weak signal strength of the long lag relative to the short lag, and the enormous influx of data for millions of AGN from Rubin. Machine learning techniques have the potential to solve many of these issues, but have yet to be applied to the long negative lag problem. We develop and train a transformer-based machine learning model to detect long and short lags in mock Rubin AGN light curves. Our model identifies whether a light curve in our test set has a long negative lag with 96% recall and 0.04 % 0.04\% contamination, and is 98% accurate at predicting the true long lag. This accuracy is an enormous improvement over two baseline methods we test on the same mock light curves, the interpolated cross correlation function and javelin , which are only 54% and 21% accurate, respectively. I Introduction Active Galactic Nuclei (AGN) accretion disks fuel radiation and outflows that can impact galaxy evolution on large scales ( 21 ; 43 ) . AGN disks also provide an important laboratory for studying accretion physics. Unfortunately, due to their great distance, it is not possible to spatially resolve the vast majority of AGN accretion disks. Instead, most of what we observe about AGN disk structure comes from a technique known as disk continuum reverberation mapping, which was first adapted by 14 from a similar idea for mapping the AGN broad line region used by 7 . Continuum reverberation mapping is often interpreted in the context of the lamp-post model. In the lamp-post model, time-varying X-ray radiation is emitted by the corona and moves radially outward on the light crossing timescale to the UV- and optical-emitting regions of the disk where it gets absorbed, reprocessed, and re-emitted at different wavelengths based on the local black body temperature of the disk. The reprocessing of X-ray variability causes variability in longer wavelength light curves to lag variability in shorter wavelength light curves. Measuring these lags, which we will call short lags here, allows us to measure the radial temperature profile and radial extent of an AGN disk ( 64 ; 13 ; 11 ; 17 ; 20 ; 18 ; 19 ; 37 ; 23 ; 68 ; 31 , e.g.,) . The lamp-post model assumes that reprocessed X-ray variability is the main driver of variability in UV and optical light curves. However, there is growing evidence of intrinsic variability not driven by X-ray variability in UV-optical AGN light curves ( 2 ; 52 ; 29 ; 4 , e.g.,) and a number of examples of AGN with only weak to moderate correlations between X-ray and UV-optical light curves ( 60 ; 8 ; 19 ; 12 ; 39 , e.g.,) . Recent radiation magnetohydrodynamic simulations suggest that fluctuations in the AGN disk driven by the magnetorotational instability ( 3 , MRI,) and convection can produce intrinsic variability in AGN light curves ( 35 ; 62 ; 63 ; 40 ) . Identifying this intrinsic variability in AGN light curves could allow us to probe the internal physics of AGN disks. Also tantalizing is the prospect of being able to measure long negative lags between intrinsic variability in different waveband light curves coming from the inward propagation of fluctuations in AGN disks on the viscous or inflow timescale of the disk. In the standard 65 thin disk model, the inflow timescale can be parameterized as \tau_{\rm{inflow}}=260\frac{M}{10^{8}M_{\odot}}\frac{0.1}{\alpha}\left(\frac{h}{0.01r}\right)^{2}\left(\frac{r}{30r_{\rm{g}}}\right)^{3/2}\rm{~years}, (1) where M is the supermassive black hole mass, \alpha is the 65 stress parameter, h/r is the aspect ratio, where h is the scale height of the disk and r is the disk radius, and r_{g} is gravitational radii. In the thin disk model, h/r=0.01 leading to an inflow timescale on the order of 100 years, or far too long to detect in current AGN light curves. However, there is growing evidence from observations ( 51 ; 37 ; 28 , e.g.,) and simulations ( 24 ; 36 ; 34 ; 35 ; 32 , e.g.,) that the standard thin disk model does not fully describe an AGN disk ( 42 ; 1 , see also,) . Because the inflow timescale depends on the height of the disk, long negative lags can be used to measure the height of AGN disks and serve as a fundamental test of the standard thin disk model. There is already some evidence for long negative lags in AGN light curves. 52 and others measured propagating temperature fluctuations in AGN photometry and spectroscopy ( 53 ; 70 , see also,) . 73 directly measured a long negative lag ranging from around -10 to -70 days between several different waveband light curves for the AGN Fairall 9. This long negative lag suggests an aspect ratio of h/r\approx 0.2 ( 61 ) , significantly thicker than is typically assumed in the standard thin disk model. Observing more long negative lags is crucial to develop better models for these lags and understand how they can be used to study AGN disk structure. The Vera Rubin Observatory Legacy Survey of Space and Time, which has 6 wavebands, a planned 10-year baseline, and will contain over 100 million AGN ( 33 ) , is the ideal instrument for detecting long negative lags. Unfortunately, 61 tested several traditional reverberation mapping methods on mock Rubin light curves, and found that the roughly consistent \sim 6 month seasonal gaps in these light curves will lead to problems with aliasing and false signal detections on timescales comparable to the long negative lag. In addition, 61 found that the confounding signal from the short lag, which should in most cases be a stronger signal than the long lag signal in order to be consistent with observations, prevents traditional methods from accurately detecting the long negative lag consistently. In this paper, we develop a machine learning method with a transformer-based architecture ( 72 ) to predict the posterior distribution over the short and long lag. We train our transformer on a test set of mock Rubin AGN light curves in preparation for applying our model to real Rubin AGN light curves as they become available in the coming years. While training, our model implicitly learns to marginalize over the uncertainty introduced by missing data (gaps) and the confounding signal (short lag). Although transformers have not yet been widely applied to the analysis of AGN light curves, previous studies have used other