A systematic search for AGN obscuration variability in the Chandra archive

TL;DR

The paper addresses how often obscuration in AGN varies by systematically mining archival Chandra observations of 79 local ($z<0.1$) AGN and applying a hardness-ratio based method to flag $N_{ m H,los}$ variability between observation pairs. It finds that about 54% of the sample shows variability at 90% CL, with roughly 61% of Seyfert 1 and 47% of Seyfert 2 sources affected, acknowledging these as lower limits due to incomplete temporal sampling. Longer time baselines yield higher variability fractions, consistent with variability arising from more distant obscurers; a KS test confirms a significant time-dependence. A subset of 43 variable sources is highlighted as a valuable target for future detailed spectral modeling (e.g., with XMM-Newton/NuSTAR) to quantify actual $N_{ m H,los}$ changes and inform clumpy torus models such as UXCLUMPY and XCLUMPY.

Abstract

The nature of the obscuring material in active galactic nuclei (AGN) can be studied by measuring changes in the line-of-sight column density, $N_{\rm H,los}$, over time. This can be accomplished by monitoring AGN over long periods of time and at all timescales. However, this can only be done for a few selected objects as it is resource intensive. Therefore, the best option currently is to focus on population statistics based on the available archival data. In this work, we study 79 Seyfert 1 and Seyfert 2 galaxies from the Milliquas catalog to estimate a lower limit on the fraction of sources in the local $(z<0.1)$ universe that display spectral variability among observations. We find that 43 sources $(54\pm11\%)$, show indications of $N_{\rm H,los}$ variability at 90% confidence level. Interestingly, we also find that the variable fraction is similar for both Seyfert 1 $(f_{\rm Sy1}\sim61^{+13}_{-15}\%)$ and Seyfert 2 $(f_{\rm Sy2}\sim47\pm15\%)$ galaxies. The slightly higher $f_{\rm Sy1}$ fraction could be due to either a physical difference in the obscurers or the higher data quality in the Sy1 population. We also search for potential dependencies on the timescale between variable and non-variable observation pairs within a given source. In agreement with previous studies, we find evidence that more variability occurs on longer timescales than on shorter timescales. We present the 43 variable sources as a promising sample for future $N_{\rm H}$ variability studies.

Figures (7)

• Figure 1: Left: Redshift distribution for the final sample of 79 sources. Seyfert 1 galaxies are shown in orange while Seyfert 2 galaxies are shown in purple. Right: The average counts of the two observations in each pair considered. The observation pairs for PGC 94626 are not shown in this plot.
• Figure 2: All 79 sources with the number of ACIS-S (lighter shades) and ACIS-I (darker shades) observations. The sources flagged as variable (at 90 % confidence) have green bars while the sources with no detected variability (at 90 % confidence) have gray bars. Note that the horizontal axis has been broken since PGC 94626 has 90 total observations (28 ACIS-S and 62 ACIS-I). Inset: The fraction of sources that are flagged as variable as a function of the minimum number of observations available for the source. The colors represent different flagging thresholds with black being the most sensitive and light green being the least sensitive. The dashed lines represent the "corrected" fraction assuming the TPR and FPR values obtained in Cox2023.
• Figure 3: Stacked distributions of the number of observations of only the sources flagged as variable (in orange and purple) and those not flagged (in gray). Most of the variable sources have more than 2 observations and most of the unflagged sources only have 2. The vertical dashed lines show the median number of observations for the variable sources (black, 4.5) and the full sample (gray, 3.0). We double count each of the 6 overlap sources between ACIS-S and ACIS-I since the variability classification is made independently on each with a different number of observations.
• Figure 4: Left: Hardness ratio values for simulated UXCLUMPY spectra with photon indices ranging from $\Gamma=1.7$ to $\Gamma=2.1$. The spectra were simulated with $N_{\rm H,los}=10\times10^{22}$ cm$^{-2}$. The mean number of 2--10 keV counts for the spectra are 150. The transparent errorbars show the HRs of each individual observation (50 total for each photon index) while the solid error bars show the average HRs for each photon index. Right: Same as left panel, but the normalization is increased so that the mean number of 2--10 keV counts for the spectra is 1300.
• Figure 5: Left: The $\chi^2$ value of each observation pair for HR1 and HR2. The points are colored according to the average number of counts between the observations. The dashed lines are at the $\chi^2_c=2.706$ threshold. Right: The average count distributions for the flagged (blue) and unflagged (yellow) observation pairs. The vertical dashed lines enclose the central 95 % probability mass for each distribution.