SUPERMASSIVE BLACK HOLES WITH HIGH ACCRETION RATES IN ACTIVE GALACTIC NUCLEI. III. DETECTION OF Fe ii REVERBERATION IN NINE NARROW-LINE SEYFERT 1 GALAXIES

We observed 10 NLS1s identified as SEAMBH candidates, spectroscopically and photometrically, between 2012 October and 2013 June. Object names and coordinates are listed in Table 1 of Paper II. Hβ time lags for three of the sources (Mrk 335, Mrk 142, IRAS F12397+3333) are presented in Paper I, and for five additional sources (Mrk 1044, Mrk 382, MCG +06–26–012, Mrk 486, Mrk 493) in Paper II. Like many other NLS1s (e.g., Boroson & Green 1992; Sulentic et al. 2000; Boroson 2002; Zhou et al. 2006), all the sources in our sample show strong Fe ii emission lines and high L_bol/L_Edd. This sample is therefore different from most AGNs in the local universe, and none of the results presented below should be compared with earlier studies, like Hu et al. (2008),  that address the general population properties.

The spectra were obtained using the Yunnan Faint Object Spectrograph and Camera, which was mounted on the Lijiang 2.4 m telescope at the Yunnan Observatory of the Chinese Academy of Sciences. A long slit with a projected width of 2 5 was oriented to take the spectra of the object and a nearby non-varying comparison star simultaneously, following, e.g., Maoz et al. (1990) and Kaspi et al. (2000). The comparison star was then used as a standard for flux calibration. For the Lijiang 2.4 m telescope, the rotator is accurate and the tracking is stable. Thus, the object and the comparison star were kept within the slit during the typical 30 minute exposures. The distance between the object and the comparison star along the direction of the slit width is less than 1 pixel (0 283).

Grism 14 was used and yielded spectra covering the wavelength range of 3800–7200 Å , with a dispersion of 1.8 Å pixel⁻¹. The final spectral resolution, obtained by comparing the width of the [O iii] emission line with the one measured from the Sloan Digital Sky Survey (SDSS; York et al. 2000) spectrum of the same object, is roughly 500 km s⁻¹. All spectra are extracted in a uniform, large aperture of 8 5 to minimize light loss. The flux calibration is based on a comparison of the object flux to the comparison star flux in the 2 5 × 8 5 apertures. However, the host galaxy flux calibration is different because of the non-stellar image—the galaxy is extended and resolved. This can result in apparent flux variations due to variable seeing and mis-centering. This requires a different calibration procedure that has a large impact on the various light curves. Appendix A details this procedure.

The traditional method for measuring the flux of the continuum and the broad emission lines, in most RM studies, is simple integration (e.g., Kaspi et al. 2000). A straight line is set between two line-free windows to define the AGN continuum, and the flux of the emission line is measured using a simple integration above the line. This method works well for single, strong emission lines, e.g., Hβ, because there are no (or only weak) other emission lines in this range and the wavelength window is narrow enough for the continuum to be approximated by a straight line. In the case of Fe ii, neither condition is satisfied.

Fe ii emission consists of thousands of lines that form a pseudo-continuum. The most prominent features in the optical band are two bumps between 4500 and 5500 Å: one between Hβ and Hγ, and the other on the red side of [O iii] λ5007. Even for quasars, whose host galaxies are extremely faint relative to AGNs, it is hard to find "pure" continuum windows, and it is inappropriate to assume a straight line over such a wide wavelength range of more than 1000 Å. The strong host galaxy contribution in most of our objects, and the contamination by other emission lines (e.g., He ii and coronal lines), makes the situation much worse. Appendix B presents the Fe ii light curves that were measured by the traditional integration method. All light curves have a large scatter, and only less than half show rough structures. A simultaneous fitting that includes the Fe ii emission and all the other spectral components is necessary.

Template-fitting is a widely used method for measuring Fe ii emission in single-epoch spectra of AGNs (see, e.g., Hu et al. 2008 for a brief overview). Input–output simulations show that template-fitting is a reliable measurement of Fe ii emission with an equivalent width (EW) > 25 Å in quasars (Hu et al. 2008). It is also common to include a host galaxy component into the fitting of Type I AGNs with strong host contribution (e.g., Zhou et al. 2006; Ho & Kim 2009). This method has recently been adopted to measure the light curves in a few RM studies, proving to be successful. Bian et al. (2010) reanalyzed the spectra of PG 1700+518 using data from Kaspi et al. (2000). They detected a lag of $209_{-147}^{+100}$ days for Fe ii emission. Barth et al. (2013) performed a very careful spectral decomposition including a power law continuum, a host galaxy component, an Fe ii template, and other emission lines (Hβ, [O iii], He ii, and He i). Using this technique they were able to obtain statistically significant reverberation lags of the Fe ii lines in two Seyfert 1 galaxies, NGC 4593 and Mrk 1511.

The spectral fitting in this paper follows the algorithm of Hu et al. (2012), where several spectral components are fitted simultaneously by minimizing the χ² via the Levenberg–Marquardt method. All the light curves are obtained directly from the results of the fitting. Before fitting, we correct the calibrated spectra for Galactic extinction and redshift. We use the R_V-dependent Galactic extinction law given by Cardelli et al. (1989) and O'Donnell (1994), and R_V is assumed to be 3.1. The redshift z and V-band extinction are taken from the NASA/IPAC Extragalactic Database⁸ and Schlafly & Finkbeiner (2011), as listed in Papers I and II.

The left panel of Figure 1 shows the fitting to the spectrum of Mrk 382 taken at Julian date (JD) 2456298, after Galactic extinction and redshift correction. The fitting is performed in the rest frame wavelength range 4150–6280 Å, except two narrow windows around Hγ  and He i λ5876. The two lines are not blended with the major part of Fe ii emission, and their study is beyond the scope of this paper. We keep them out of the fitting to avoid introducing too many unnecessary parameters. The observed spectra in the fitting windows are plotted in green, while those left out of the fitting are in black. The fitting includes the following components: (1) a single power law, (2) Fe ii emission, and (3) the host galaxy. These three components, which when combined together form the pseudo-continuum,⁹ are plotted in blue. (4) The fitting also includes the Hβ  emission line plotted in magenta; (5) the broad He ii λ4686 emission line plotted in cyan; and (6) the narrow emission lines plotted in orange, including [O iii] λλ4959, 5007, He ii λ4686, He i λ4471, and several coronal lines. The summed model is plotted in red. The bottom panel shows the residual spectrum. Note that, although He i λ5876 is not in the fitting window, its profile is well recovered after removing the Na i λλ5890, 5896 (Na D) absorption lines from the host galaxy.

Zoom In Zoom Out Reset image size

Limited by the signal-to-noise ratio (S/N) for the individual-night spectra, there are degeneracies between several pairs of spectral components, including: (a) the AGN power-law continuum and the host galaxy, (b) Fe ii emission and the broad He ii line, and (c) Fe ii emission and the coronal lines. Thus, we first fit the high-S/N mean spectrum with all parameters set free. We then fit each individual-night spectrum with the values of some parameters fixed to those obtained from the fitting of the mean spectrum. The details of each spectral component, and the fitting parameters, are described in the following subsections.

3.1.1. AGN Power-law Continuum

3.1.2. Fe ii Emission

3.1.3. The Hβ  Line

3.1.4. Narrow Emission Lines

Besides the strong [O iii] λλ4959, 5007, there are many other narrow emission lines in the wavelength range of our fit spectrum (Vanden Berk et al. 2001). We identify narrow He ii λ4686, He i λ4471, and several high-ionization forbidden coronal lines. In some objects, these narrow lines are so strong that their contamination is non-negligible to the Fe ii measurement. Figure 2 shows an individual-night spectrum of Mrk 335 and our fit. The strong coronal lines included in this case are [Fe vii] λ5158, [Fe vi] λ5176, [N i] λ5199, [Ca v] λ5309, [Fe vii] λ5721, and [Fe vii] λ6086. Adding these lines significantly improves the fit.

Zoom In Zoom Out Reset image size

The intensities of the narrow lines differ from one object to the next. For each object, we first identify narrow lines in the mean spectrum by testing the goodness of the fit. Only lines that are identified in the mean spectrum are included in the fitting of individual-night spectra. Each narrow emission line is modeled by a Gaussian. These lines are coming from the narrow-line region and hence do not vary on the campaign timescale. Thus, we constrain them to have the same velocity width and shift as those of the [O iii] λ5007 line. The flux ratio relative to [O iii] λ5007 is also kept constant, as given by the best fit to the mean spectrum (the flux ratio of [O iii] λ4959 to λ5007 is fixed to the theoretical value of 1/3), in order to avoid the degeneracy between these lines and Fe ii. Only the flux, velocity width, and shift of [O iii] λ5007 are set free.

3.1.5. Broad He ii Emission Line

3.1.6. Host Galaxy

Table 1 lists the values of the parameters fixed in the fitting of individual-night spectra for all the objects in our sample. Columns (5)–(12) list the narrow emission lines included in the fits. As explained, their velocity widths and shifts are constrained to be the same as those of the [O iii] λ5007 line, while the relative intensity ratios are kept as measured from the mean spectrum. The last column gives the galaxy template used. In the table, a blank entry means that a spectral component is not included for the object.

Figure 3 shows examples of fittings to individual-night spectra for the eight objects not presented so far. The notations and colors are as the same as those in Figure 1 for Mrk 382. Notes on two objects with unusual treatments are as follows.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

IRAS 04416+1215. This object has the highest redshift (0.089) in our sample. From the images taken for slit centering, the point-spread function of the object is as broad as that of the comparison star, meaning that there is a negligible galaxy contribution. For this object, the comparison star is fainter than the AGN, especially in the red part. We found that the shape of the spectrum of this object is not as well calibrated by the comparison star as in the other objects. Thus, we fit in a relatively narrow wavelength window (4430–5550 Å), but covering the two bumps of Fe ii emission on either side of Hβ. We also let the spectral index, α, of the AGN power-law continuum free to vary in individual-night spectra to compensate for the difference in spectral shape caused by the calibration. As shown in Figure 3, the spectrum can be fitted well by including only a power law, Fe ii emission, Hβ, [O iii], and a few narrow lines.

IRAS F12397+3333. As discussed in Paper I, the spectrum of this AGN is probably affected by host galaxy extinction. The observed spectra (see Figure 4 of Paper I for an example) have a very red color, and are badly fitted by our spectral model. We perform an extinction correction after de-redshifting, assuming the Galactic extinction law and A_V = 1.44 mag estimated from the Balmer decrement (see details in Paper I). This dereddened spectrum is fit well with a typical spectral index, as shown in Figure 3.

For each individual-night spectrum, the fluxes, FWHMs, and velocity shifts (with respect to [O iii]) of Fe ii and Hβ  are calculated from the best-fit model. The means of these properties are considered to be the measurements of the line profile, except for the width of Hβ (FWHM_Hβ; see below). The standard deviations on those means are used as the uncertainties.

FWHM_Hβ  is underestimated in the fitting, as we do not include the narrow component of Hβ  in our model (see Section 3.1.3). Thus, we use a different method to estimate this width. We add a Gaussian to our model to represent the narrow Hβ  component. The velocity width and shift are constrained to be the same as those of [O iii]. We then fit the mean spectra twice, once assuming the flux of the narrow Hβ  line is 10% of the flux of [O iii] λ5007 and once assuming a flux ratio of 0.2. The width of the broad Hβ obtained with a flux ratio set to 0.1 is adopted as the FWHM_Hβ. The uncertainty is obtained from the fit with a flux ratio set to 0.2 and the original fit that assumed no narrow Hβ. This method essentially resembles the one used in Papers I and II, while the spectral model here is more sophisticated.

Table 2 lists the measurements of the 10 objects. The instrumental broadening (FWHM ∼500 km s⁻¹, Section 3.1.3) has been taken into account in the listed FWHMs. Note that the Hβ of IRAS 04416+1215 has a large velocity-shift with respect to [O iii] λ5007, and its profile features no narrow component with the same shift of [O iii]. Adding a narrow component makes the width of the broad Hβ  even narrower. Thus, the FWHM_Hβ of IRAS 04416+1215, as with other properties, was obtained from the measurements of individual-night spectra.

Our fitting successfully reduces the scatter in the light curves due to the influence of the host galaxy contamination; thus, F_AGN represents the real AGN continuum much better than the simply integrated 5100 Å flux (F₅₁₀₀), as shown in Appendix A. The light curves of the emission lines are also generated directly from the best-fit values of the corresponding parameters that were obtained from the fits of individual-night spectra. The errors of the fluxes given by the fitting are not large enough to account for the scatter in the fluxes of successive nights; an additional systematic error is estimated for each light curve (as in Paper I). This systematic error is added in quadrature to the fitting error for the calculations of variability amplitudes and time lags below. Note that our treatment is different from that of Barth et al. (2013), who measured the light curve of Hβ  by integrating the continuum-subtracted spectra.

We use the quantity F_var defined in Rodríguez-Pascual et al. (1997) to represent the variability amplitude. This quantity is an estimate of the intrinsic variability over the errors. The uncertainties are calculated following Edelson et al. (2002). The results for Fe ii and Hβ  are listed in Table 3. For the variability amplitude ratio of Fe ii to Hβ, the range is about 0.6 (Mrk 486) to 1.2 (Mrk 42), except for Mrk 382, which has the largest host galaxy contamination. On average (except for Mrk 382), the Fe ii value is about 10% smaller, consistent with previous results (e.g., Vestergaard & Peterson 2005; Barth et al. 2013).

The time lags between the AGN continuum variations (F_AGN) and the emission lines (F_Fe and F_Hβ) are measured from the cross-correlation functions (CCFs) of the relevant light curves. We use the interpolation cross-correlation function (Gaskell & Sparke 1986; Gaskell & Peterson 1987; White & Peterson 1994) method to calculate the CCF, and adopt the centroid of the CCF, above 80% of the peak value (r_max) as the time lag (Koratkar & Gaskell 1991; Peterson et al. 2004). The uncertainty in the time lag measurement is estimated from the cross-correlation centroid distribution (CCCD) given by random subset selection/flux randomization Monte Carlo realizations (Maoz & Netzer 1989; Peterson et al. 1998).

All of the objects in our sample, except for Mrk 42, have reliable time lag measurements for both Fe ii and Hβ. The right columns of Figure 4 show the results of the CCF analysis. For each object, the top right panel shows the autocorrelation function of the F_AGN light curve, which is shown in the top left panel. The two lower panels in the right column show the CCFs (in black) for Hβ (middle right) and Fe ii (bottom right), with respect to F_AGN  light curves. The blue histograms are the corresponding CCCDs. Most CCCDs have a rather symmetric profile, except those for IRAS 04416+1215.

Table 4 lists the time lags of Fe ii (τ_Fe) and Hβ (τ_Hβ), their uncertainties, and the corresponding r_max for the nine objects with reliable lag measurements. The listed time lags and uncertainties are in the rest frame after time-dilation correction. Both of the time lags for IRAS 04416+1215 have uncertainties of highly unequal upper and lower limits as the sequence of their asymmetric CCCDs. For other objects, the time lags are well determined by the single-peak CCFs with high r_max and symmetric, narrow CCCDs.

Table 4.  Time Lag Measurements

Object	Fe ii		Hβ
	rmax	τFe	rmax	τHβ
Mrk 335	0.48	$26.8_{-2.5}^{+2.9}$	0.70	$8.7_{-1.9}^{+1.6}$
Mrk 1044	0.48	$13.9_{-4.7}^{+3.4}$	0.62	$10.5_{-2.7}^{+3.3}$
IRAS 04416+1215	0.61	$12.6_{-6.7}^{+16.7}$	0.60	$13.3_{-1.4}^{+13.9}$
Mrk 382	0.40	$23.8_{-6.0}^{+6.0}$	0.47	$7.5_{-2.0}^{+2.9}$
Mrk 142	0.78	$7.6_{-2.2}^{+1.4}$	0.84	$7.9_{-1.1}^{+1.2}$
MCG +06–26–012	0.86	$22.4_{-6.3}^{+9.3}$	0.92	$24.0_{-4.8}^{+8.4}$
IRAS F12397+3333	0.63	$10.6_{-1.9}^{+7.0}$	0.71	$9.7_{-1.8}^{+5.5}$
Mrk 486	0.77	$17.3_{-3.7}^{+5.8}$	0.79	$23.7_{-2.7}^{+7.5}$
Mrk 493	0.63	$11.9_{-6.5}^{+3.6}$	0.81	$11.6_{-2.6}^{+1.2}$

Note. Time lags are in the rest frame, and in units of days.

Download table as:  ASCIITypeset image

The detection rate of the Fe ii time lag in our sample is extremely high (9 in 10) compared with previous RM experiments. This is mainly attributed to the common feature of strong Fe ii emission in the SEAMBH sample, the high-cadence observation, and sophisticated spectral fitting. The failure to obtain reliable τ_Fe and τ_Hβ in Mrk 42 is caused by the F_AGN light curve, which has large scatter and no clear structure (see Figure 5). However, the light curves of F_Fe and F_Hβ show hints of a similar structure, which, unfortunately, is not enough to establish a time lag.

The clear reverberation of the Fe ii emission in response to the continuum supports the hypothesis that the Fe ii emission originates from photoionized gas in the BLR. The high detection rate indicates that this is prevalent in NLS1, which are AGNs with high L_bol/L_Edd. It is possible that the origin of Fe ii emission in AGNs with low L_bol/L_Edd is different, but there is little evidence to support this claim. In fact, Barth et al. (2013) already detected the reverberation of Fe ii in two broad-line Seyfert 1 galaxies with Hβ widths of FWHM ≳ 4000 km s⁻¹ and much lower L_bol/L_Edd. Comparing the time lags of Fe ii and Hβ, along with their velocity widths, provides some hints about the location and geometries of the two emission regions. We will discuss these ideas in Section 5.2 below.

In Paper I and Paper II, the reported time lags of Hβ are obtained from the light curves that were measured by the more traditional integration method, without taking into account the contamination by the host galaxy and the narrow emission lines. Figure 6 shows a comparison between the previously reported τ_Hβ that is based on the integration method and the new results given by the fitting method reported here. The dashed diagonal line denotes the 1:1 ratio and the region between the two dotted lines gives the 0.1 dex deviation from the line. For seven out of nine objects, the differences between τ_Hβ given by the two methods are less than 0.1 dex. The two exceptions are IRAS 04416+1215 and Mrk 1044. In Paper II we show that using the integration method, we cannot obtain a significant τ_Hβ for IRAS 04416+1215. The lag for Mrk 1044 that is reported in that paper is about a factor of 2 shorter than obtained here, with a very large uncertainty ($4.8_{-3.7}^{+7.4}$). This value is consistent with the new one presented here within the uncertainties. In addition, the uncertainty of τ_Hβ in Mrk 493 given by the integration method is very large, with a negative lower limit ($12.2_{-16.7}^{+3.5}$). The fitting method yields a much smaller uncertainty with basically the same τ_Hβ. This is a direct result of the much smoother Hβ  light curve of Mrk 493 given by the fitting method (see Figure 4, and also Figure 2 of Paper II). Thus, the time lags of Hβ given by the two methods are consistent with each other and τ_Hβ  is better defined by the fitting procedure in some of the cases.

Zoom In Zoom Out Reset image size

A simple theoretical expectation based on photoionization is that ${{R}_{{\rm BLR}}}\propto {{L}^{\alpha }}$ with α ∼ 0.5. The R_BLR–L relation for Hβ has been compared with such predictions in several earlier publications (e.g., Kaspi et al. 2000; Bentz et al. 2009, 2013, and references therein). Regarding Fe ii lines, Chelouche et al. (2014) provide a tentative R_BLR–L relation for the first time, from a small inhomogenous sample of six AGNs. Among them, two objects are reported by Barth et al. (2013) using the standard spectral fitting method, and four others come from Rafter et al. (2013) and Chelouche et al. (2014), using the MCF scheme of Chelouche & Zucker (2013). Here, we revisit this issue by more than doubling the size of the sample by adding our nine newly measured sources. For a proper comparison with the Chelouche et al. (2014) results, we used our best observed fluxes and uncertainty, and a cosmology with ${{H}_{0}}=70\;{\rm km}\;{{{\rm s}}^{-1}}$ Mpc⁻¹, Ω_m = 0.3, and Ω_Λ = 0.7.

Figure 7 shows τ_Fe  versus the 5100 Å luminosity for the nine objects in our sample (black dots) alongside the earlier results (kindly provided by D. Chelouche). The green squares are those obtained by the MCF scheme (including three with insignificant results; see Figure 4 of Chelouche et al. 2014 for details). The blue triangles are the two objects from Barth et al. (2013) for which we used the centroid of the published CCFs for Fe ii and the V-band light curves. Using the FITEXY method (Press et al. 1992) to fit all the 18 objects, we find

$$\begin{eqnarray}\begin{array}{ccccccccccccccc} {\rm log} ({{R}_{{\rm Fe}\,{\rm II}\,{\rm BLR}}}) & = & (-22.0\pm 1.1) \\ {} & {} & +(0.54\pm 0.02){\rm log} (\lambda {{L}_{\lambda }}(5100\overset{\circ}{\rm A} )), \\ \end{array}\end{eqnarray} \tag{ 1 }$$

which is plotted as a solid line. This line deviates considerably from the relation presented in Chelouche et al. (2014; shown as a dotted line), but  is close to that of Bentz et al. (2009) for Hβ (dashed line). This result is expected since most of the objects in our sample show roughly equal τ_Fe  and τ_Hβ (see Table 4), except for three objects (Mrk 335, Mrk 382, and Mrk 486), while the previous studies have  τ_Fe greater than τ_Hβ. Our sample by itself shows no correlation at all between τ_Fe and the 5100 Å luminosity. This is not surprising given the relatively small range of luminosity we have considered here, and the fact that for this sample, the correlation between τ_Hβ and the continuum luminosity is also very weak. We suggest that this lack of correlation is also related to the large L_bol/L_Edd of the objects in our sample, a topic we discuss in great detail in Paper IV (Du et al. 2015).

Zoom In Zoom Out Reset image size

As shown in Table 4 τ_Fe is roughly equal to τ_Hβ (counting the uncertainties in both lags) in six of our objects, longer in two (Mrk 335 and Mrk 382), and somewhat shorter in one (Mrk 486). A common feature of the three objects with unequal lags is the relatively low Fe ii/Hβ  intensity ratio. Figure 8 shows the plot of τ_Fe/τ_Hβ versus F_Fe/F_Hβ (defined as R_Fe). The three objects with different time lags are labeled by their names and located in the region of R_Fe < 1. The six other objects show R_Fe > 1. We also mark the approximate positions of the two objects that were measured by Barth et al. (2013).¹⁰ Based on eye estimates of the published light curves shown in their paper, both objects have R_Fe ∼ 0.75. Their positions in Figure 8 are marked around 0.75 (and horizontally shifted for clarity). They both have different time lags than Hβ and R_Fe < 1, consistent with the results of our sample.

Zoom In Zoom Out Reset image size

We note, again, that our sample is highly biased toward AGNs with high L_bol/L_Edd and thus high R_Fe. In an unbiased AGN sample, the fraction of sources with R_Fe < 1 is much larger (see, e.g., Figure 7 of Sulentic et al. 2000 and Figure 1 of Shen & Ho 2014). Hu et al. (2008) analyzed the properties of a large number of low-redshift AGNs. They found that the velocity width of Fe ii is systematically narrower (∼3/4) than that of Hβ, which may suggest, given Keplerian velocities, that in those sources τ_Fe is longer than τ_Hβ. Our present sample contains only 10 sources, which is too few to systematically test any of these ideas.