X-Ray and UV Monitoring of the Seyfert 1.5 Galaxy Markarian 817

Active galactic nuclei (AGNs) are powered by accretion onto a supermassive black hole. At high fractions of the Eddington limit, this process is mediated by an accretion disk that peaks in the UV. A "corona" is also inferred through X-ray emission in many AGNs. Though the corona and accretion disk may be physically distinct, the X-ray and UV emission from these regions may interact with each other. On long timescales (weeks to months), variations in the mass accretion rate through the disk should cause changes in the rate of viscous dissipation and thereby the UV flux; Compton upscattering in the corona may then drive variations in the X-ray flux. On shorter timescales (less than one to two days), rapid variability in the corona—perhaps due to magnetic activity—can drive X-ray flux changes that are reprocessed into UV emission in the disk. On both long and short timescales, then, a very simple picture predicts clear lags and correlations.

Recent research into X-ray and UV variability has produced mixed results. Several teams looking at different sources have found cases where the X-ray emission leads the UV emission (e.g., McHardy et al. 2014; Edelson et al. 2015, 2017; Troyer et al. 2016; McHardy et al. 2018; Pal & Naik 2018), consistent with X-ray reprocessing. These lags are longer than predicted, though, and they are attributed to larger accretion disks or extra reprocessing mechanisms. In the case of NGC 5548, the X-ray emission has been found to lead the optical emission on timescales of  1 yr (Uttley et al. 2003), but it has been noted recently that the X-ray and UV/optical emissions correlate poorly on shorter timescales, which again require extra reprocessing mechanisms to explain such results (Edelson et al. 2015; Gardner & Done 2017; Starkey et al. 2017). Others, however, have found no correlation between the X-ray and UV emission, attributing this to light bending of a centrally compact X-ray source near the black hole, X-ray flux in our line of sight not being correlated with the X-ray emission that is reprocessed in the disk, and/or other mechanisms driving the observed UV variability (Robertson et al. 2015; Buisson et al. 2018).

It has also been previously noted that the appearance of significant lags can change depending on the interval or timescale being studied. For example, Gallo et al. (2018) examined X-ray and UV light curves of Mrk 335 and found no correlation between the two, over an 11 year span. However, when they restricted their analysis to only the eighth year of monitoring, they found a highly significant positive correlation of the X-ray leading the UV. They attributed this correlation to a giant X-ray flare that occurred during their monitoring. This and the examples above reveal the complexity of X-ray/UV radiation and the need for a larger number of multiwavelength studies to better understand disks and disk–corona connections in AGNs.

In this paper, we examine the relationship between X-ray and UV variability of the nearby (z = 0.03145) Seyfert 1.5 galaxy Mrk 817 ($\mathrm{log}({M}_{\mathrm{BH}}/{M}_{\odot })=7.586;$ Bentz & Katz 2015). The X-ray emission of Mrk 817 has changed drastically since its discovery decades ago. Previously, the modest X-ray flux of Mrk 817 made it largely inaccessible. However, by 2009, its X-ray flux had increased by a factor of ∼40 since 1990 (Winter et al. 2011), making Mrk 817 one of the brightest Seyferts in the X-ray regime. In contrast, the UV emission only varied by a factor of ∼2.3 over a 30 yr period (Winter et al. 2011). The fact that the X-ray light curve has previously shown marked variability, while the UV light curve does not, makes Mrk 817 a particularly intriguing AGN. Yet, in comparison to other X-ray-bright Seyfert galaxies, Mrk 817 has been overlooked.

The most recent X-ray spectral study of Mrk 817, by Winter et al. (2011), used the six X-ray observations from Swift and a single XMM-Newton exposure that were available at the time. They found that the optical/UV fluxes were similar, and the optical–X-ray SED changes between observations were due to variations in the X-ray luminosity/spectral shape, concluding that the UV and X-ray are not correlated. Winter et al. (2011) also reported a strong positive correlation between the X-ray spectral slope and X-ray luminosity. We re-examine these results more thoroughly using a cross-correlation analysis on 94 new X-ray and UV observations from Swift . In Section 2, we describe the observations and data reduction. Section 3 details and provides results of a cross-correlation analysis of the X-ray and UV light curves. We end with a discussion of our results in Section 4.

Mrk 817 was monitored simultaneously using the Swift XRT (Burrows et al. 2005) and the Swift UVOT (Roming et al. 2005) from 2017 January 2 to 2018 April 20. Over this 473 day period, 94 observations were considered, separated in time by an average of ∼5.1 days.

The XRT exposures were taken in "photon counting" mode with a 23.6 × 23.6 arcmin field of view. Source events were extracted from a circle with a radius of 20 pixels, centered on coordinates provided in NED.³ Background events were extracted using a circular region of radius 20 pixels, off-center from the source coordinates and outside the source region. Ancillary response files were created for each observation using xrtmkarf, each of which included the provided exposure map to account for bad CCD pixels and columns. Using grppha, the spectra were grouped such that there are at least 10 counts in each bin.

The extracted spectra were then fitted with a power-law model over the 0.3–10.0 keV energy range (mean count rate of 0.5466 ± 0.0025 cts s⁻¹) using xspec v.12.9.1p. An absorbed power-law model is not considered, as the galactic H i column density is too small (N_H = 1.50 × 10²⁰ cm⁻²; Dickey & Lockman 1990) to produce any discernible effects within the short exposures. For each spectrum, the photon index and normalization, and the corresponding 1σ errors on these parameters, were recorded. The flux in the fitting band was obtained using the "flux" command within XSPEC; flux errors were calculated by assuming that the flux has the same fractional error as the power-law normalization. We find that this gives a more conservative estimation of error than the one provided by xspec. The X-ray light curve can be seen in the panel labeled "X-ray" of Figure 1.

Zoom In Zoom Out Reset image size

We separately modeled all X-ray spectra using a broken power-law model, in an attempt to allow for a physically distinct "soft excess." The break energy was fixed at 2 keV in each observation, and the fluxes for the soft (0.3–2.0 keV) and hard (2.0–10.0 keV) X-rays were recorded. The flux errors were calculated in the same manner as before. The soft and hard X-ray light curves are shown in the panels labeled "Soft" and "Hard" of Figure 1, respectively.

The UV images were taken with the UVOT/UVM2 filter, which has the smallest red leak of the Swift UVOT UV filters. The UVM2 filter has a central wavelength of 2246 Å, an effective wavelength of 2231 Å, and an FWHM of 498 Å (Poole et al. 2008). Source events were extracted from the images using a circular region of radius 5 arcsec centered on the source. Background emission was accounted for using an annulus (33 arcsec inner radius, 43 arcsec outer radius) centered on the source. We calculated the flux density of the UVM2 images using the ftools function uvotsource. It should be noted that there can be coincidence losses due to the brightness of the source, and uvotsource provides a reconstructed "true" flux density; this is the flux density we recorded for each observation. The error on each flux density value was calculated as $\sqrt{\mathrm{stat} \mbox{-} {\mathrm{err}}^{2}+\mathrm{sys} \mbox{-} {\mathrm{err}}^{2}},$ where stat-err and sys-err are the statistical error and systematic error on the flux density, respectively. The light curve for UVM2 is given in the panel labeled "UVM2" of Figure 1. The mean corrected count rate for the entire monitoring period is 46.66 ± 0.11 cts s⁻¹ with a mean corrected count rate factor of 1.627 ± 0.004. It is well known that UV "dropouts" can occur due to suspect regions of the detector (e.g., Edelson et al. 2015, 2017). We do not account for "dropouts" here, as the UVM2 light curve has very few obvious points that demonstrate such features.

Light curve data and exposure times for each of the 94 observations are provided in Table 1.

3.1. Photon Index/X-Ray Relationship

A plot of the power-law photon index and corresponding X-ray (0.3–10.0 keV) flux is provided in Figure 2. The mean power-law photon index is 2.0047 ± 0.0117. We calculated a Spearman correlation coefficient of r = 0.34 with a two-sided p-value of p = 9.6 × 10⁻⁴. In contrast, Winter et al. (2011) found a positive correlation between the photon index and X-ray luminosity, with a Spearman correlation coefficient of r = 0.77 and a two-sided p-value of p = 0.05. Our analysis includes a much larger set of observations and may therefore more accurately reflect the degree to which the power-law properties are correlated. Combining our points with the data presented in Winter et al. (2011) slightly increases the correlation we originally calculated: r = 0.38 with p = 1.1 × 10⁻⁴.

Zoom In Zoom Out Reset image size

We also provide plots for the soft X-ray photon index versus the soft X-ray flux and the hard photon index versus the hard X-ray flux in Figure 3. The soft photon index and hard photon index correspond to the 0.3–2.0 keV and 2.0–10.0 keV energy ranges of the broken power-law model, respectively. The mean soft photon index and mean hard photon index are 2.142 ± 0.015 and 1.601 ± 0.032, respectively. The soft photon index and soft X-ray flux give a Spearman correlation coefficient of r = 0.12 with a two-sided p-value of p = 0.24. The hard photon index and hard X-ray flux produce a Spearman correlation coefficient of r = 0.25 with a two-sided p-value of p = 0.013.

Zoom In Zoom Out Reset image size

3.2. X-Ray/UV Correlation Analysis

A plot (Figure 4) of the X-ray (0.3–10.0 keV) flux versus the UVM2 flux shows minimal correlation between the two bands. Indeed, an analysis of the two produces a Pearson correlation coefficient of r = 0.11 with a two-sided p-value of p = 0.29.

Zoom In Zoom Out Reset image size

A cross-correlation analysis was performed in order to test for correlations at different lags. We use the discrete correlation function (DCF; Edelson & Krolik 1988; Robertson et al. 2015) because of the uneven sampling of the X-ray and UV observations. A positive value on the lag axis corresponds to the X-ray flux leading the UV flux. The DCF requires the lag time axis to be separated into bins of equal width, δτ, which is a free variable. By the definition of the DCF, these bins must contain at least one data point, and—if errors are to be calculated—each bin must contain at least two data points. Choosing a value for δτ is a tradeoff between statistical accuracy and resolution; the former requires a larger δτ and the latter requires a smaller δτ. We set δτ = 50 days, as this allowed for at least 20 points in each bin, providing reasonable statistical accuracy and a DCF with discernible details.

We calculated 95% and 99% confidence curves to evaluate the significance of various potential lags. We first simulated 1000 X-ray light curves using the method of Timmer & König (1995). This procedure requires knowledge of the power spectral density, which we estimate to be a power law with a slope of −1.1 and a normalization value of 0.011; this estimation follows the method of Zoghbi et al. (2013). Values are then selected from each simulated light curve such that they correspond to the times at which the real X-ray observations were recorded. The DCF between these simulated values and the real UV light curve is calculated. This produces a distribution of 1000 correlation values at each lag, which we then used to calculate 95% and 99% confidence values.

We report no correlation above a 95% confidence value at any given lag, as can be seen in the first panel of Figure 5. However, via visual inspection, there appears to be an anticorrelation between the X-ray and the UV light curves from 2018 January 12 to April 20 (MJD 58130 to MJD 58228), a time period of approximately 98 days in which the observation cadence increases. The DCF plot shows an anticorrelation at a lag of ∼18.3 days (see Figure 6), significant at the 95% level of confidence, corresponding to the X-ray flux leading the UV flux. Owing to this modest significance, the lack of a clear physical interpretation for an anticorrelation, and the robust nature of the DCF,⁴ this putative signature must be regarded skeptically. The light curves were cross-correlated for a period of less frequent observation (2017 January 2–December 122, MJD 57755 to MJD 58098), and no significant correlation was found.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The soft and hard X-ray bands were each cross-correlated with UVM2 as well; these DCFs are plotted in the second and third panels of Figure 5, respectively. The soft X-ray band produces a DCF similar to the full X-ray band (0.3–10.0 keV), showing no correlation. The DCF of the hard X-ray band with the UVM2, however, produces a 99% confident anticorrelation at a lag of ∼284 days, meaning the X-ray leads the UV by ∼284 days. The fact that the lag timescale is comparable to the period over which the monitoring was conducted indicates that this signal should be regarded cautiously. Horne et al. (2004) suggested that the length of the monitoring period should be at least three times longer than the lags of interest, whereas Welsh (1999) recommended a monitoring period of at least four times longer, but preferably ∼10 times longer, than the lags of interest. In this paper, we follow the suggestion of Horne et al. (2004).

A cross-correlation of the hard and soft X-ray light curves reveals a 99% significant anticorrelation at a lag of ∼235 days, corresponding to the hard X-ray flux leading the soft X-ray flux (see the fourth panel of Figure 5). There also exists 95% confident positive correlations at lags of ∼144 and ∼349 days. These positive correlations may also be spurious: the lags are not highly significant, and the robust nature of the DCF can produce such results. Here again, the longest lags are also close to the full period over which the monitoring was conducted. Visually inspecting the soft and hard X-ray light curves, it appears that the two light curves are tracking each other. We find that the hard and soft X-ray light curves are indeed correlated at lag τ = 0 by calculating a Pearson correlation coefficient of r = 0.464 with a two-sided p-value of p = 2.49 × 10⁻⁶, corresponding to a >99% significance. The disagreement at τ = 0 between the DCF and the Pearson coefficient is due to the binning requirement of the DCF.

We have undertaken an extensive X-ray and UV monitoring campaign to better understand Mrk 817, a bright, nearby, Seyfert 1.5 AGN. Although the source is now as bright as the best-known Seyferts, it was much fainter in past decades, and as a result it has not been studied extensively. We cross-correlated the X-ray and UV light curves using the standard DCF; separate cross-correlations were undertaken for the soft and hard X-ray flux, in recognition of potentially distinct components within the spectrum. We do not find any lag signals at a high level of statistical significance. In this section, we offer plausible physical reasons why strong correlations might be absent.

Many similar analyses have found significant short lags, with the X-rays leading the UV in a manner broadly consistent with reprocessing (McHardy et al. 2014; Edelson et al. 2015, 2017; Troyer et al. 2016; McHardy et al. 2018; Pal & Naik 2018). In some cases, the observed lags are slightly longer than anticipated by standard thin disk models; this may suggest that AGN disks do not follow such prescriptions or the lags are contaminated by the continuum emission from the broad-line region (Korista & Goad 2001; Cackett et al. 2018). In the case of Mrk 817, however, the DCFs that we calculated find no robust correlations above the 95% confidence between the X-ray (0.3–10.0 keV) and UV light curves. Putative lag signals on long timescales—noted above—are too close to the duration of the monitoring to be credible. Though uncommon, non-correlations have been reported recently (Robertson et al. 2015; Buisson et al. 2018). This lack of a significant X-ray/UV correlation is inconsistent with the optical variability of Mrk 817, as it has been reverberation-mapped, i.e., lags have been calculated between the Hβ and continuum light curves (Peterson et al. 1998; Denney et al. 2010). This suggests that, in this source, the X-ray variability is not the primary driver of photoionization and line variability in the broad-line region. This is at odds with the standard lamppost model whereby a small, centrally located X-ray source drives variability in the UV/optical continuum and emission lines.

Geometric beaming—or perhaps "funneling"—of the X-ray emission represents one potential means by which reprocessing of X-ray emission into UV flux might be prevented. For an AGN with a high-Eddington fraction, for instance, the inner accretion disk may fail to cool efficiently, giving it an increased scale height above the local surface of the disk. This geometry may serve as an inner funnel that could block the X-ray emission from larger disk radii. This explanation requires that we have a privileged viewing angle into the funnel. Moreover, Mrk 817 has a low-Eddington fraction: L_bol/L_Edd = 0.1950 ($\mathrm{log}\,{L}_{\mathrm{bol}}=44.99$, Woo & Urry 2002). Recent work on NGC 5548 and NGC 4151, in particular, has suggested that additional geometries may be needed to account for X-ray and UV correlations (e.g., Edelson et al. 2017; Gardner & Done 2017). Moreover, X-ray observations of NGC 4151 find evidence of a warp or bulge in the inner accretion disk (Miller et al. 2018; A. Zoghbi et al. 2018, in preparation) though its Eddington fraction is even lower than that of Mrk 817.

A qualitatively distinct but physically similar scenario is that the X-ray flux is relativistically beamed away from the disk, owing to bulk motion normal to the disk. If the X-ray corona is the base of a jet (Markoff et al. 2005; Miller et al. 2006), where acceleration away from the disk may occur, then the velocities required to escape the regions closest to the black hole may serve to naturally beam the X-ray emission away from the disk. Like most other Seyferts, however, Mrk 817 is not radio-loud, so if this process is at work, the corona might be the base of a "failed" jet.

It is also possible that a dense outflow from the inner disk could inhibit the reprocessing of X-rays into UV flux in the disk. A disk wind that covers the inner accretion disk might scatter a fraction of the X-ray emission before this flux reaches the optically thick accretion disk. Winter et al. (2011) found no evidence of X-ray absorption consistent with an outflow; however, disk winds—especially in the inner disk—may be primarily equatorial. If we observe Mrk 817 at a low inclination (closer to the pole), a wind may operate but not be detected. Recent observations of NGC 5548 have also revealed evidence of dense, transient outflows from the inner disk (Kaastra et al. 2014). Additional X-ray and UV spectroscopy of Mrk 817 may reveal if the source was in a similar state during the period of our monitoring and/or when Winter et al. (2011) studied the source with XMM-Newton. Obscuration by transient clouds from a warm absorber and a neutral absorber can mask the intrinsic variability of the X-ray emission, which could lead to a lack of correlation between the X-ray and UV emission.