The Accretion Disk Size Problem in AGN Disk Reverberation Mapping is an Obscuration Effect: A Uniform AGN Sample Study with Swift

TL;DR

Swift-based high-cadence UVOIR reverberation mapping of nine AGN shows the accretion disk size problem persists but is not universal; lags exceed thin-disk predictions primarily in obscured AGN, with a mean lag normalization excess of about $3.45$ across the sample and $5.21$ for obscured vs $1.00$ for unobscured. X-ray spectral modeling reveals two distinct populations by $N_H$ and hardness ratio, enabling robust separation of obscured and unobscured sources, and lag normalizations scale with $M$ and $\dot{m}$ as $\tau_0 \propto (M^2 \dot{m})^{1/3}$, with a fitted coefficient $\alpha \approx 3.4\times10^{-5}$ days $M_\odot^{-2/3}$. Multivariate regression shows column density explains >80% of the variance in lag excess, while fractional RMS variability provides no independent predictive power when $N_H$ is included. The results argue that line-of-sight obscuration drives the lag excess via additional reprocessed emission, rather than implying a universal failure of thin-disk theory, and emphasize accounting for obscuration in future lag-based disk size studies.

Abstract

In the past decade, Swift has performed several AGN high-cadence reverberation mapping campaigns, and generally found that the UV/optical interband lags are $\sim$3 times longer than predicted for a standard thin disk, thus coined "the accretion disk size problem". Here we present a systematic sample of Swift-monitored AGN. In this analysis, we confirm the accretion disk size problem, but find that the lag excess occurs only in the subset of obscured AGN, which show a significantly elevated mean normalization of $5.21 \pm 0.47$ ($p = 0.008$), whereas the unobscured AGN exhibit a mean excess consistent with standard disk predictions ($1.00 \pm 0.31$). Correlation and regression analyses similarly reveal X-ray column density as the strongest predictor of lag excess, explaining over 80% of its variance. We interpret these results as line-of-sight obscuration being linked to the too-long lags via additional reprocessed emission from the absorbing material itself. The consistency of lags in the unobscured subgroup with standard disk predictions suggests that the accretion disk size problem is not the result of shortcomings of standard accretion disk theory nor contamination by the broad-line region (BLR). X-ray to UV lag amplitudes and correlations show more complex and variable behavior in obscured AGN, suggesting that obscuration may disrupt or complicate the connection between high- and low-energy emission potentially through reprocessing, scattering, and/or ionization changes.

Figures (9)

• Figure 1: Swift X-ray spectra for the AGN in our sample. The sources fall into two broad categories based on their spectral shapes, which reflect varying levels of obscuration. Spectra shown in pink correspond to unobscured sources, as shown by smooth, power-law–like continua with relatively minimal spectral features. In contrast, the green spectra exhibit pronounced curvature, particularly between $\sim$1–4 keV, indicative of higher levels of obscuration from neutral and ionized gas along the line of sight. These visually apparent differences are later formalized using a Gaussian mixture model, whose classifications are consistent with this initial inspection. For detailed inspection, per-source spectra are shown in the Appendix (Figure \ref{['fig:spectra_grid']}).
• Figure 2: The two Gaussian components of the Gaussian mixture model, corresponding to the obscured and unobscured subgroups. Each distribution was fit by maximizing the likelihood of the observed column density and hardness ratio values, incorporating measurement uncertainties via a Monte Carlo approach. Covariance between the variables is accounted for, producing tilted ellipses, most notable in the obscured subgroup. The overlap between the two components is minimal, resulting in very high cluster assignment probabilities ($>99\%$) for all sources.
• Figure 3: Measured interband lags as a function of wavelength using the ICCF approach. The dashed black line shows the best-fit power-law relation, $\tau \propto \lambda^{4/3}$, with the normalization treated as a free parameter. The orange dotted line shows the same relation, but with the normalization instead predicted from standard accretion disk given the black hole mass and accretion rate of each source. As such, the differences between these measured and predicted curves reflect departures from theoretical expectations (the "normalization excess").
• Figure 4: Best-fit lag normalizations ($\tau_0$), corresponding to the dashed black curves in Figure \ref{['fig:lags_vs_wavelength']}, plotted against $M^2 \dot{m}$. The orange dashed line shows the predicted scaling $\tau_0 \propto (M^2 \dot{m})^{1/3}$, computed from a modified version of Equation 12 in Fausnaugh_2016, assuming standard disk parameters stated in the text. Most sources lie above the theoretical prediction, highlighting a systematic excess in lag normalizations relative to expectations. However, while the obscured sources are well described by the best-fit relation, the unobscured sources tend to lie closer to the predicted scaling, suggesting that the excess lag normalizations are most significantly driven by the obscured population.
• Figure 5: Normalization excesses--defined as the ratio of the measured lag normalization to the predicted value--for each AGN in the sample. Dotted lines indicate the sample means for each unobscured and obscured subgroup, with shaded bands denoting the standard error of the mean. The full sample has a mean excess of $3.45 \pm 0.27$, with the obscured subgroup showing a larger average excess ($5.21 \pm 0.47$) than the unobscured group ($1.00 \pm 0.31$). A similar trend is observed in the U-band lag excesses on their own. Hypothesis testing confirms that the excesses in the full and obscured samples are significantly greater than one ($p = 0.011$ and $p = 0.008$, respectively), while the unobscured group (as expected) shows no significant deviation from theoretical expectations ($p = 0.50$). The one super-Eddington source (Mrk 142) occupies a distinct region of the parameter space.