REVERBERATION MAPPING MEASUREMENTS OF BLACK HOLE MASSES IN SIX LOCAL SEYFERT GALAXIES

Spectra of the nuclear region of our complete²⁵ sample (see Table 1) were obtained daily (weather permitting) over 89 consecutive nights in 2007 Spring with the 1.3 m McGraw-Hill telescope at MDM Observatory, and supplemental spectroscopic observations of most targets were obtained with the 2.6 m Shajn telescope of the Crimean Astrophysical Observatory (CrAO) and/or the Plaskett 1.8 m telescope at Dominion Astrophysical Observatory (DAO) to extend the total campaign duration to ∼120 nights. We used the Boller and Chivens CCD spectrograph at MDM with the 350 grooves mm⁻¹ grating (i.e., a dispersion of 1.33 Å pixel⁻¹) to target the Hβ λ4861 and [O iii] λλ4959, 5007 emission-line region of the optical spectrum. The position angle was set to 0°, with a slit width of 50 projected on the sky, resulting in a spectral resolution of 7.6 Å across this spectral region. We acquired the CrAO spectra with the Nasmith spectrograph and SPEC-10 1340 × 100 pixel CCD. For these observations a 30 slit was utilized, with a 90° position angle. Spectral wavelength coverage for this data set was from ∼3800 to 6000 Å, with a dispersion of 1.8 Å pixel⁻¹ and a spectral resolution of 7.5 Å near 5100 Å. The actual wavelength coverage is slightly greater than this, but the red and blue edges of the CCD frame are unusable due to vignetting. The DAO observations of the Hβ region were obtained with the Cassegrain spectrograph and SITe-5 CCD, where the 400 grooves mm⁻¹ grating results in a dispersion of 1.1 Å pixel⁻¹. The slit width was set to 30 with a fixed 90° position angle. This setup resulted in a resolution of 7.9 Å around the Hβ spectral region. Figure 1 shows the mean and rms spectra of our sample based on the MDM observations. Table 2 gives more detailed statistics of the spectroscopic observations obtained for each target, including number of observations, time span of observations, spectral resolution, and spectral extraction window.

Zoom In Zoom Out Reset image size

A relative flux calibration of each set of spectra was performed using the χ² goodness of fit estimator algorithm of van Groningen & Wanders (1992) to scale relative fluxes to the [O iii] λ5007 constant narrow-line flux. This algorithm not only makes a multiplicative scaling to account for the night-to-night differences in flux in this line caused primarily by aperture effects, but it also makes slight wavelength shifts to correct for zero-point differences in the wavelength calibration and small resolution corrections to account for small variations in the line width caused by variable seeing. The best-fit calibration is found by minimizing residuals in the difference spectrum formed between each individual spectrum and the reference spectrum, which was taken to be the average of the best spectra of each object (i.e., those obtained under photometric or near-photometric conditions). Because of this multiple-component calibration method, the final, scaled [O iii] λ5007 line flux in each spectrum is not exactly the same as the reference spectrum. Instead, there is a small standard deviation in the mean line flux due to differences in data quality that averages ∼1.2% across our sample.

In addition to spectral observations, we obtained supplemental V-band photometry from the 2.0 m Multicolor Active Galactic NUclei Monitoring (MAGNUM) telescope at the Haleakala Observatories in Hawaii, the 70 cm telescope of the CrAO, and the 0.4 m telescope of the University of Nebraska. The number of observations obtained from each telescope and the time span over which observations were made of each target are given in Table 3.

The MAGNUM observations were made with the multicolor imaging photometer (MIP) as described by Kobayashi et al. (1998a, 1998b), Yoshii (2002), and Kobayashi et al. (2004). Photometric fluxes were measured within an aperture with radius 83. Reduction of these observations was similar to that described for other sources by Minezaki et al. (2004) and Suganuma et al. (2006), except the host-galaxy contribution to the flux within the aperture was not subtracted and the filter color term was not corrected because these photometric data were later scaled to the MDM continuum light curves (as described below). Also, minor corrections (of order 0.01 mag or less) due to the seeing dependence of the host-galaxy flux were ignored.

The CrAO photometric observations were collected with the AP7p CCD mounted at the prime focus of the 70 cm telescope (f = 282 cm). In this setup, the 512 × 512 pixels of the CCD field project to a 15' × 15' field of view. Photometric fluxes were measured within an aperture diameter of 150. For further details of the CrAO V-band observations and reduction, see the similar analysis described by Sergeev et al. (2005).

The University of Nebraska observations were conducted by taking and separately measuring a large number of one-minute images (∼20). Details of the observing and reduction procedure are as described by Klimek et al. (2004). Comparison star magnitudes were calibrated following Doroshenko et al. (2005a, 2005b) and Chonis & Gaskell (2008). To minimize the effects of variations in the image quality, fluxes were measured through an aperture of radius 80. The errors given for each night are the errors in the means.

Except where noted below for individual objects, continuum and Hβ light curves were created as followed. Continuum light curves for each object were made with the V-band photometric observations and the average continuum flux density measured from spectroscopic observations over the spectral ranges listed in Table 2 (i.e., rest frame ∼5100 Å). Continuum light curves from each source were scaled to the same flux scale following the procedure described by Denney et al. (2009b). Figure 2 (top panels) shows these merged light curves, where measurements from each different observatory are shown by the different symbols described in the figure caption.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Light curves of the Hβ flux were made by integrating the line flux above a linearly interpolated continuum, locally defined by regions just blueward and redward of the Hβ emission line. The Hβ emission line was defined between the observed frame wavelength ranges given for each object in Table 2. The Hβ light curves formed from each separate spectroscopic data set (i.e., MDM, CrAO, and DAO) were placed on the same flux scale (i.e., that of the MDM observations) by again following the scaling procedures described by Denney et al. (2009b). An additional flux calibration step was used for NGC 3516, however, because it has a particularly extended [O iii] narrow-line emission region. In an attempt to decrease the uncertainties in our relative flux calibration from slit losses of this extended emission, we made an additional correction to each MDM Hβ flux measurement to account for possible differences in the observed [O iii] λ5007 flux due to seeing effects. To measure the expected differences in [O iii] λ5007 flux entering the slit as a result of changes in the nightly seeing, we followed the procedure of Wanders et al. (1992), using their artificially seeing-degraded narrow-band image of the [O iii] λ5007 emission from the nuclear region of NGC 3516 (details regarding the narrow-band data are described by Wanders et al.). Using the differences in measured flux, we scaled our MDM flux measurements accordingly. We could only do this for the MDM measurements, since we do not have accurate seeing estimates for the CrAO and DAO data sets. Because of our deliberately large aperture (see Table 2, Column 8), the effect was not appreciable for most observations, and there is no indication that our inability to complete the same analysis for the CrAO and DAO data had any measurable effect on the subsequent time-series analysis. The lower panels of Figure 2 show the Hβ light curves for each object after merging the separate data sets into a single Hβ light curve.

Before completing the time-series analysis, the light curves shown in Figure 2 were modified in the following ways.

• 1.  
  An absolute flux calibration was applied to both continuum and Hβ light curves by scaling to the absolute flux of the [O iii] λ5007 emission line given for each object in Column 3 of Table 4. For objects in which there was not a previously reported absolute flux, we calculated one from the average line flux measured from only those observations obtained at MDM under photometric conditions.
• 2.  
  The host-galaxy starlight contribution to the continuum flux was subtracted. This contribution, listed for each target in Column 5 of Table 4, was determined using the methods of Bentz et al. (2009b) for all objects except Mrk 290, which had not been targeted for reverberation mapping prior to our observing campaign.²⁶ For Mrk 290, we use an estimate made from the spectral decomposition (following decomposition method "B" described by Denney et al. 2009a) of an independent spectrum taken at MDM with nearly the same setup as our campaign observations but covering optical wavelengths from 3500 to 7150 Å with a 15 slit. This value is only a lower limit, however, since this slit width was smaller than that of our campaign observations (i.e., 50).
• 3.  
  We "detrended" any light curves in which we detected long-term secular variability over the duration of the campaign that is not associated with reverberation variations (Welsh 1999; see also Sergeev et al. 2007, who show that there is little correlation between long-term continuum variability and Hβ line properties, demonstrating the independence of this variability on reverberation processes). Detrending is important because if the time series contains long-term trends (i.e., compared to reverberation timescales), the flux measurements are not randomly distributed about the mean and are, thus, highly correlated on these long timescales. These long timescale correlations then dominate the results of the cross-correlation analysis that determines the time delay, biasing the desired correlation due to reverberation. Welsh (1999) strongly recommends removing these low-frequency trends with low-order polynomials (a linear fit at the very least) to improve the reliability of cross-correlation lag determinations. We took a conservative approach and only linearly detrended light curves in which there was evidence for secular variability and for which the cross-correlation analysis was improved upon detrending: both light curves from Mrk 290, the Hβ light curve from Mrk 817, and the continuum light curve from NGC 3227 (see Section 2.4 for further discussion). These fits are shown in Figure 2 for each of these respective light curves. It was unnecessary to detrend all light curves, as no improvement in the cross-correlation analysis would result from detrending light curves that already have a relatively flat mean flux. Also, it is not surprising for associated continuum and line light curves to exhibit different long-term secular trends, since the relationship between the measured continuum and the ionizing continuum responsible for producing the emission lines may not be a linear one (Peterson et al. 2002), and the exact response of the line depends on the detailed structure and dynamics of the BLR.
• 4.  
  We excluded the points from the Mrk 817 light curve with JD < 2,454,200 because (1) there is a large gap in the data between these points and the rest of the light curve and (2) there is little to no coherent variability pattern seen here (i.e., the continuum is relatively flat and noisy, and the Hβ fluxes are particularly noisy and are of otherwise little use, given there are no continuum points at earlier times).

Tabulated continuum and Hβ fluxes for all objects, except for NGC 4051 which were previously reported by Denney et al. (2009b), are given in Tables 5 and 6, respectively. Values listed represent the flux of each observation after completing all flux calibrations described above (i.e., absolute flux calibration based on the [O iii] λ5007 emission-line flux and host-galaxy starlight subtraction), but before detrending, since this results in an arbitrary flux scale normalized to 1.0. The final calibrated light curves used for the subsequent time-series analysis are shown for each object in the left panels of Figure 3. Statistical parameters describing these calibrated light curves (again, before detrending) are given in Table 7, where Column 1 lists each object. Columns 2 and 3 are mean and median sampling intervals, respectively, between data points in the continuum light curves. The mean continuum flux is shown in Column 4, while Column 5 gives the excess variance, calculated as

$$\begin{equation} F_{\rm var} = \frac{\sqrt{\sigma ^2 - \delta ^2}}{\langle f \rangle }, \end{equation} \tag{ 1 }$$

where σ² is the variance of the observed fluxes, δ² is their mean square uncertainty, and 〈f〉 is the mean of the observed fluxes. Column 6 is the ratio of the maximum to minimum flux in the continuum light curves. Columns 7–11 display the same quantities as Columns 2–6 but for the Hβ light curves.

We performed a cross-correlation analysis to evaluate the mean light-travel time delay, or lag, between the continuum and Hβ emission-line flux variations. We primarily employed an interpolation scheme (Gaskell & Sparke 1986; Gaskell & Peterson 1987, with the modifications of White & Peterson 1994). Using this method, we first interpolate (with an interval equal to roughly half the median data spacing, i.e., ∼0.5 day) between points in the emission-line light curve before cross-correlating it with the original continuum light curve, calculating cross-correlation coefficients, r, for many potential lag values (both positive and negative). We then average these cross-correlation coefficients with those measured by imposing the same set of possible lag values in the case where we cross-correlate an interpolated continuum light curve with the original emission-line light curve. This gives us a distribution of average cross-correlation coefficients as a function of possible lags, known as the cross-correlation function (CCF). We checked the results from this method with the discrete correlation method of Edelson & Krolik (1988), also employing the modifications of White & Peterson (1994), but we do not show these results here, since they are consistent with our primary cross-correlation method and provide no additional information.

The right panels of Figure 3 show the adopted cross-correlation results for each object (i.e., after detrending selected light curves; see below for a discussion of the effect of detrending on this analysis). Here, the autocorrelation function (ACF), computed by cross-correlating the continuum with itself, is shown in the top right panel for each object, and the CCF computed by cross-correlating the Hβ light curve with that of the continuum, is shown in the bottom right. Because the CCF is a convolution of the transfer function with the ACF, it is instructive to compare the two distributions, as the lag measured through this type of cross-correlation analysis will depend not only on the delay map, but also on characteristic timescales of the continuum variations (see, e.g., Netzer & Maoz 1990). We characterize the time delay between the continuum and emission-line variations by the parameter τ_cent, the centroid of the CCF based on all points with r ⩾ 0.8r_max, as well at the lag corresponding to the peak in the CCF at r = r_max, τ_peak. Time dilation-corrected values of τ_cent and τ_peak were determined for each object using the redshifts listed in Table 1, i.e., τ_rest = τ_obs/(1 + z), and are given in Table 8. Uncertainties in both lag determinations are computed via model-independent Monte Carlo simulations that employ the bootstrap method of Peterson et al. (1998), with the additional modifications of Peterson et al. (2004).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Visual inspection of the CCFs of selected objects before and after detrending was made to determine if detrending these light curves was warranted. Based on the combined properties of the light curves shown in Figure 2 (whether or not an overall slope appeared in the flux across the extent of our campaign) and the CCFs, shown in Figure 4 for Mrk 290, Mrk 817, and NGC 3227 before and after detrending, we ultimately decided to adopt the detrending for the following reasons listed for each object.

Zoom In Zoom Out Reset image size

Mrk 290. The top panels of Figure 4 show that before detrending (left), the peak of the CCF is broader than the detrended peak (right) and is blended with an aliased peak at ∼30 days. Since the reverberation lag is clearly seen in the Mrk 290 light curves in Figures 2 and 3 and the peak of highest significance is the same both before and after detrending, the presence of this alias only acts to decrease the precision of our lag measurements. While τ_cent is roughly one day smaller after detrending (a difference less than even the measured uncertainty) due to the reduced significance of the aliased peak at ∼30 days by a factor of almost 10, the detrended CCF is narrower and the measured lags more precise, so we adopt the detrended measurements.

Mrk 817. The middle panels of Figure 4 show the original (left) and detrended (right) CCFs from the analysis of Mrk 817. The choice to detrend was marginal in this case. The process resulted in a larger observed lag (τ_cent = 14.48 days versus τ_cent = 11.93) after detrending, contrary to the typical expectation that lags will be underestimated after detrending (since the process removes low-frequency variability). We adopt the detrended results because the resulting CCF is narrower, particularly with respect to lags ≲0.0 days, and the resulting lag measurement is more consistent with past results that we hold to be reliable (see Section 5.1).

NGC 3227. The bottom panels of Figure 4 show the original (left) and detrended (right) CCFs from the analysis of NGC 3227. Here it is obvious that not detrending the light curves results in a non-physical measurement of the lag at ∼−33 days with a broad peak (due to aliasing effects between the features with the highest flux in each of the original continuum and Hβ light curves). While the physical peak (i.e., with positive lag, as seen and measured from the detrended CCF) is present, every lag is of low significance, i.e., r ≲ 0.4. After detrending, the CCF peak at negative lags is still present; however, the "true" reverberation signal at a lag of ∼4 days is rightfully more significant.

Some of the objects in this campaign were targeted, at least in part, because they have previously appeared as outliers on AGN scaling relationships, in particular, the R_BLR–L relationship. As such, all objects except Mrk 290 have previous reverberation results, several of which were suspect for one reason or another and warranted re-observation. Based on the outcomes of the current analysis, we will group our results into three categories: (1) new measurements for an object never before targeted, i.e., Mrk 290, (2) replacement measurements for objects that had uncertain results (typically due to undersampling) and for which our results completely replace any previous measurements of the Hβ reverberation lag, i.e., NGC 3227, NGC 3516, and NGC 4051, and (3) additional measurements of objects for which we already trust the previous lag measurements, i.e., NGC 5548 and Mrk 817. In this context, we can compare our new results to previously published results.

5.1.1. New Measurements

5.1.2. Replacement Measurements

5.1.3. Additional Measurements

To investigate the outcome of our goal to improve the calibration of scaling relations by re-examining objects that had large measurement uncertainties and/or that appeared as outliers on these scaling relationships, we place our new measurements in context to the R_BLR–L relationship most recently calibrated by Bentz et al. (2009b). Luminosities were measured from the average, host-corrected continuum flux density measured within the 5100 Å rest-frame continuum windows listed for each object in Table 2. For most objects, we simply corrected for Galactic reddening along the LOS (Schlegel et al. 1998); however, NGC 3227 and NGC 3516 show evidence of internal reddening that must be taken into account in determining the luminosity. Gaskell et al. (2004) argue that the UV–optical continua of AGNs are all very similar, so that the reddening can be estimated by dividing the spectrum of a reddened AGN by the spectrum of an unreddened AGN. In the case of NGC 3227, we use the value of A_B determined by Crenshaw et al. (2001) by comparing the UV–optical spectrum of NGC 3227 to the unreddened spectrum of NGC 4151. For NGC 3516, we consider two methods for estimating the reddening, which result in consistent estimates of A_B: (1) we follow the Crenshaw et al. method, comparing the spectrum of NGC 3516 again to that of NGC 4151, which results in A_B = 1.72, and (2) we use the Balmer decrement measured from the broad components of the Hα and Hβ emission lines to estimate a reddening of A_B = 1.68. These two values are highly consistent, and we adopt the average between the two methods of A_B = 1.70. Our measured luminosities are given in Column 9 of Table 8, where the uncertainties in the luminosities are the standard deviation in the continuum flux over the course of the campaign, except for NGC 4051, where the uncertainty in the distance is added in quadrature to this (see Denney et al. 2009b).

The top panel of Figure 6 shows the Bentz et al. (2009b) R_BLR–L relationship, reproduced from the bottom panel of their Figure 5. Here, we have differentiated the objects targeted for our present campaign with solid squares, while all other objects presented by Bentz et al. are open squares. The bottom panel of Figure 6 shows our current results, where the objects for which our new measurements are either truly new (i.e., Mrk 290) or have become replacements for old values are shown by the solid stars, and we no longer plot the old values. Our additional measurements for NGC 5548 and Mrk 817 are shown with the open stars, and the previous weighted average lags and luminosities for these objects as reported by Bentz et al. are still present in this bottom panel. The reader should immediately notice the increased precision and accuracy of our new and replacement measurements, where it is important to note that we have not determined a new fit to the data.²⁷ Clearly, these better measurements emphasize the small intrinsic scatter in this relationship, reinforcing the apparently homologous nature of AGNs, even over many orders of magnitude in luminosity. The results from this campaign also support the conclusion of Peterson (2010) that improving this relationship further will not come from simply obtaining more BLR radii measurements to "beat down" the noise, but rather, from more reliable, higher-precision measurements.

Zoom In Zoom Out Reset image size

The cleanest cases of a velocity-resolved reverberation response are for NGC 3516, NGC 3227, and NGC 5548, where we see kinematic signatures indicating apparent infall, outflow, and non-radial, or "virialized," motions, respectively. Denney et al. (2009c) discuss the velocity-resolved results for these three objects and the implications of these different kinematic signatures in the context of our overall understanding of the BLR and the use of BLR radii measurements for determining BH masses. In addition, Denney et al. (2009b) present and discuss the marginally velocity-resolved lags shown here for NGC 4051, and so those results are not discussed further here.

The objects not discussed in previous publications are Mrk 290 and Mrk 817. Figure 5 shows that there is very little variation in the reverberation lag across the full width of the Mrk 290 line profile, indicating that any differences in the reverberation lag across the extent of the Hβ-emitting region in this object were unresolvable with the sampling rate of our campaign. An additional possibility for the uniform response we observed (i.e., small range in lags and no short lags observed) could be that the highest velocity gas seen in the wings of the mean spectrum is optically thin, and therefore does not respond to the continuum variations. This is supported by the narrowness of the Hβ profile in the rms spectrum compared to that observed in the mean spectrum. On the other hand, based on the relative emission-line strengths of the high-velocity wings in several AGNs, Snedden & Gaskell (2007) argue against this interpretation.

At first glance, Mrk 817 appears to show an outflow signature similar to that of NGC 3227, however, cross-correlation between the continuum light curve and those derived from the line flux in the first four velocity bins actually results in lag determinations that are, though negative, largely consistent with zero lag. Ignoring these first bins gives results similar to Mrk 290, where no velocity-dependent differences in the lags are resolved. Taken at face value, this result is curious. We present binned light curves of the Mrk 817 line profile in Figure 7, where to increase the clarity of the discrepancy between the red and blue sides of the line for this discussion, we have combined sets of two bins to make a total of four bins instead of eight, i.e., we plot the flux from bin 1 added to that of bin 2, bin 3 added to bin 4, etc. For completeness, we also recompute the CCFs (also shown in Figure 7) and velocity-resolved lag measurements for these four combined bins and find results consistent with simply taking the average of the lags of each set of two bins that we combined, though the uncertainties in the newly measured lags are generally smaller, particularly for the bluest and reddest bins. Upon inspection of the individual light curves for these bins, it becomes apparent that the cross-correlation analysis for these bins essentially failed, not finding a strong correlation between the continuum flux variability and that seen in the light curves of bin 1 and bin 2. The light curves show a lack of variability in the flux in these bins during the first half of the campaign, and then a fairly monotonic rise in flux during the second half, so the peak in the continuum flux seen near JD2,454,230 is not seen in the light curves of bins 1 and 2, and instead, the feature the cross-correlation analysis picks up is the trough near ∼JD2,454,282, apparently seen in the bins 1 and 2 light curves ∼8–10 days earlier. This combination causes the cross-correlation analysis to give unreliable results. Furthermore, no real indication of the expected positive lag can be seen by eye, as with the other bins (and other objects, for that matter). The observations could be explained by some gas having an unresolved velocity structure near the mean radius measured for this object and there also being an outflowing component in the BLR of this object, so that the blueshifted gas is primarily along the LOS and a resulting zero-day lag is measured. However, given that (1) the overall variability observed in this object was small during this campaign and (2) the Hβ profile is very broad, leading to a small variability signal spread over a large wavelength range, we cannot make any strong conclusions at this time. Future efforts will be made both to glean further information from the velocity–delay map reconstructed from our current data as well as to reanalyze the previous monitoring data on this object in an attempt to search for any other indications of velocity-resolved signatures.

Zoom In Zoom Out Reset image size

Despite the differences we see in the velocity-resolved kinematics across our sample of objects, we do not believe that there is cause for concern for the masses derived from the mean BLR radii measured from these reverberation lags. Obviously, observing unresolved, virial, or infalling gas motions certainly does not question the validity of our assumption that the BLR motions are gravitationally dominated, but indications of outflow may be more problematic. However, even given these signatures, the mean lag we measure is still consistent with lags derived from the majority of the emission-line gas. Besides, it is only gas outflowing at velocities larger than the escape velocity that would break the validity of our assumptions, and this does not seem to be the case. There are good observational and theoretical reasons to believe that there are multiple components within the BLR (e.g., disk and wind components), and the disk–wind model of Murray et al. (1995), for example, is still able to justify the constraint of the BH mass by the reverberation mapping radii measurements, even with the presence of a wind (see Chiang & Murray 1996).

From velocity-resolved studies, such as the one discussed here and in our previous publications on this data set (Denney et al. 2009b, 2009c), it is clear that high-cadence reverberation mapping studies are beginning to push the envelope with respect to the amount of information we are able to glean from data of high quality and homogeneity. The next goal is to attempt a reconstruction of the velocity-resolved transfer function through the production of velocity–delay maps, with priority placed on the objects shown here and discussed by Denney et al. (2009c) that exhibit statistically significant kinematic signatures of infall, outflow, and virialized motions (NGC 3516, NGC 3227, and NGC 5548, respectively). Preliminary results from this analysis show the potential to reveal the types of structured maps that will hopefully provide additional constraints on future models of the BLR and more clearly reveal distinct kinematic structures responsible for the velocity-resolved signatures we presented here.