Untangling the Complex Nature of AGN Variability with Fairall 9

TL;DR

This study interrogates AGN variability in Fairall 9 by combining time-resolved SED modelling with Fourier-domain timing, revealing that large-scale disc changes drive the SED while the X-ray power stays nearly constant. Using Gaussian Process reconstructions, the authors measure PSDs, lags, and coherence across a three-year campaign and develop an analytic timing framework that attributes UV–X-ray coupling to seed-photon modulation, disc reverberation, and a wind/BLR reprocessing component. The timing signals show a low-frequency UV-leading lag during the rise phase and a later coherence drop, which the analytic framework can accommodate through interference between independently varying disc and corona components. Overall, the results imply an evolving inner accretion structure on timescales of hundreds of days and challenge simple reverberation or propagating-fluctuation scenaria, with implications for future monitoring campaigns and timing analyses of AGN.

Abstract

The accretion flow in AGN is not well understood, motivating intensive monitoring campaigns of multiwavelength variability to probe its structure. One of the best of these is the 3 year optical/UV/X-ray approximately daily monitoring campaign on Fairall\,9, a fairly typical moderate accretion rate AGN. The UV lightcurve shows a clear increase over $\sim 50$ days between years 1 and 2, strongly coherent with the X-ray lightcurve rise. This changes the average spectral energy distribution such that the disc component is stronger while the X-ray spectrum steepens, so that the total X-ray power remains roughly constant. Outside of this global change, we apply a Fourier resolved analysis to test stochastic models where intrinsic fluctuations in the UV disc propagate down into the hard X-ray emission region via both changing the seed photon flux for Compton scattering (short light travel timescale) and changing the electron density (longer propagation timescale). Unlike these models, the hard X-rays are not particularly well correlated with the UV, and also have the wrong sign in that the hard X-rays marginally lead the UV fluctuations. We show that this is instead consistent with uncorrelated stochastic fluctuations in both the UV (slow) and X-ray (fast), which are linked together only weakly via light travel time. These variability properties, as well as the changes in the SED, has implications for our understanding of AGN structure and physics, as well as future monitoring campaigns.

Figures (15)

• Figure 1: Top rows: Light-curves for Fairall 9 for Swift UVW2 (top, magenta) and HX (1.5-10.0 keV, bottom, blue) taken from Edelson24. These are shown as fractional flux to allow for a more direct comparison, given that the native units of each light-curve differ (cts/s for HX and erg s$^{-1}$ cm$^{-2}$ Å$^{-1}$ for UVW2). The dashed lines indicate the segments used to create time-averaged SEDs. The shaded grey region indicates a sharp rise in the light-curves, which we have chosen to exclude from the SED analysis due to the possibility of a change in the inner geometry. Bottom row: SEDs averaged over each of the segments indicated in the light-curves. The orange data-points show the unfolded, de-absorbed, data, while the grey points indicate the absorbed data. We note that the x-errorbars in the UVOT data represent the full band-pass. During the fitting procedure the filter response is also taken into account. The solid black line shows the total SED model, with the dashed lines indicating the model components to the total SED. These are: a warm Comptonising disc (green), a hot Comptonising corona (blue), and neutral reflection of optically thick material (magenta).
• Figure 2: A comparison of the predicted intrinsic SEDs for each light-curve segment highlighted in Fig. \ref{['fig:LC_and_SEDs']}. Here the solid green line shows segment 1a, the dashed orange line shows segment 2, and the dashed-dotted purple line shows segment 3. The shaded regions highlight the Swift UVW2 and HX band-passes. It is clear that the majority of the variability between segments occurs in the disc-like region. The time-averaged, integrated, hot Coronal component displays little variability, suggesting that the long term variability in the HX band-pass is mainly due to the change in photon-index.
• Figure 3: Posterior distributions for the free agnsed parameters from our MCMC fit, normalised such that they integrate to 1. Here green corresponds to the segment 1a SED, orange to the segment 2 SED, and purple to the segment 3 SED. There is clearly a significant increase in mass-accretion rate between segment 1a and segments 2 and 3, corresponding to before and after the strong rise in the UVW2 light-curve. There is also a significant softening of the hot corona photon index, suggesting an increase in seed photon flux entering the corona leading to increased Compton cooling. Additionally we see a softening in the warm corona photon index, which suggests an increase in power dissipated within the mid-plain. The reduction in truncation radius, $r_{\rm{hot}}$, is a consequence of the energy balance compensating for the increased mass-accretion rate (seen in the optical/UV) while the integrated X-ray power remains remarkably constant. The segment 3 posteriors are generally broader than in segments 1a & 2 as during this time the cadence of the campaign was reduced to $\sim 4$ days, resulting in fewer observations in our stack and thus lower S/N.
• Figure 4: Swift UVW2 (top) and HX (bottom) light-curves, including our Gaussian Process (GP) model. The mean GP prediction is shown as the solid green line in both light-curves, while the $1\sigma$ dispersion is shown by the green shaded regions. As in Fig.\ref{['fig:LC_and_SEDs']} the dashed lines indicate the segments used for our SED and now Fourier analysis. Here we also include the segment 1b for the Fourier analysis, in order to evaluate the impact of the sharp rise to the variability.
• Figure 5: Power-spectra of Fairall 9 for UVW2 (magenta open circles) and HX (turquoise stars) for each segment in the light-curves, as predicted by our Gaussian Process model. In the leftmost panel we also show the power-spectrum from Segment 1a+1b as the open red triangles (UVW2) and filled red diamonds (HX). These have all been calculated by averaging over 5000 realisations of the GP model. The x-error indicates the width of the frequency bin, while the y-error is the combined error calculated from the standard error on the underlying noise process and the 1$\sigma$ dispersion from the 5000 GP realisations. At low frequencies the y-error is dominated by the noise intrinsic to the periodogram, while at high frequencies the dispersion in the GP realisations dominates. The dashed black line shows the expected noise level for the HX power-spectra, while the dotted black line shows the noise for UVW2, both calculated following Uttley14. We also include the predicted power-spectra from the propagating fluctuations model of Hagen24a as the solid lines for both UVW (magenta) and HX (turquoise). The red hatched region displays the frequency range beyond $f_{NY}$ for each segment, calculated from the average sampling rate in each segment (given in the top left corner). Due to the significant drop in cadence in Segment 3, the two highest frequency bins extend beyond $f_{NY}$, and are thus based on the properties learned by the GP model in previous (higher cadence) segments. As such they should not be trusted.