OPTICAL SPECTRAL PROPERTIES OF SWIFT BURST ALERT TELESCOPE HARD X-RAY-SELECTED ACTIVE GALACTIC NUCLEI SOURCES

In order to fit the emission lines as correctly as possible, great care must be taken in modeling the continuum. For an AGN source, we expect the continuum to be a combination of non-thermal emission from the AGN and stellar light from the host galaxy. To model the contribution from stellar light, we used the population synthesis models from the GALAXEV package⁹ (Bruzual & Charlot 2003) in the 3200–9300 Å range. The spectral resolution of these models (≈3 Å) is directly comparable to that in the SDSS and KPNO samples. We assume that the galaxy light is the sum of bursts of formation at different ages, using stellar populations at three different ages (25, 2500, and 10,000 Myr) to determine whether the host is consistent with a young, intermediate, or old population (or any combination of these three). While a three component stellar model (young, intermediate, old) does not fully describe the spectra of all galaxies, we have found (see the Appendix) that adding more components results in degenerate solutions with different sums of the 10 spectral models in the Bruzual & Charlot instantaneous burst models. We thus use the three component models adopted and recognize that this may not be a fully accurate description of the stellar components of the host galaxies. Additionally, we used three metallicity levels: 0.05 Z (2.5 Z_☉), 0.02 Z (Z_☉), and 0.004 Z ($\frac{1}{5}\, Z_{\odot }$). We use the same code described in Tremonti et al. (2004), which was used to measure the continuum in a sample of 53,000 SDSS galaxies. As described in Tremonti et al. (2004), the best fit is obtained using a nonnegative least-squares fit using the same metallicity for all three of the "age" groups, attenuated by dust (which is modeled as a free parameter). The χ² values, using different metallicity populations, are compared to find the best-fit metallicity range. We note, however, that these models depend upon necessary assumptions, such as stellar populations created in an instantaneous burst of star formation (see Conroy et al. 2009 for a discussion of many of the associated uncertainties in single stellar population models), which are not physical.

To test the effectiveness of the galaxy continuum fits, we first applied the models to our set of template galaxies obtained at KPNO. The final input to the Tremonti et al. (2004) code is the galaxy's velocity dispersion, a quantity that is unknown for many of our AGN host galaxies. Therefore, we fit each of the templates with a range of dispersion values to obtain the best fit. These values were then compared to the known galaxy parameters, listed in LEDA¹⁰ (Paturel et al. 2003). The galaxy type and velocity dispersion, as well as the fitted values, are listed in Table 5. On average, we find that the fitted dispersion velocities (for a Gaussian, FWHM =2.35 × σ) are in agreement with the central velocity dispersions listed in LEDA (〈σ_M〉 = 132 km s⁻¹, while the LEDA values give 〈v_disp〉 = 159 km s⁻¹). From a comparison of the galaxy type to the light fraction (at 5500 Å) from the young, intermediate, and old stellar populations, there are no obvious contradictions. Our sample includes late spirals through ellipticals and we find that the models suggest the light is dominated by intermediate to old stellar populations in most of the galaxies (this is consistent with the color analysis of the images found by M. Koss et al. 2010, in preparation). In Figure 1, we plot examples of the results of the stellar continuum fitting. We find that the models are particularly accurate at fitting the blue end (below 5000 Å) of the spectra. While the addition of more stellar populations (at different ages) would provide better fits to the spectra, Tremonti et al. (2004) point out that the fits are often degenerate (they use 10 different population ages). Therefore, in an effort to get a broad understanding of the stellar properties of the AGN host galaxy properties, we confine our fits to the young, intermediate, and old populations indicated above.

Zoom In Zoom Out Reset image size

Additionally, we created a grid of test spectra using different combinations of the three stellar populations indicated. Random noise was added to the test spectra, which were then broadened with FWHM =300 km s⁻¹ and an instrumental resolution of 75 km s⁻¹, and reddened using the Charlot & Fall (2000) law. The results of continuum fits to these test spectra are presented in Appendix A. As shown, we find that the velocity dispersion is well determined for our test spectra while the metallicities are not. We can clearly distinguish young stellar populations from the intermediate/old populations, however, there is a degeneracy between the intermediate and old populations when they are combined with the young populations. These degeneracies are taken into account in the following discussions. We also created a grid of test spectra including a power-law contribution similar to that of our sample along with the stellar populations, from which we found no degeneracy between the power law and stellar components (see Appendix A).

In order to subtract a continuum from the KPNO and the SDSS spectra, we modified the galaxy modeling code to include a non-thermal power-law contribution from the AGN ($p_0 \times \lambda ^{p_1}$, where p₀ is constrained to range from 0 to 1 and p₁ > 0). In our model, separate reddening values were fitted for both the power-law component and the stellar component. Additionally, in our fits we masked out regions near prominent emission line positions (i.e., Hβ, [O iii] λ5007) at a standard width of 500 km s⁻¹ and used a larger width of 7000 km s⁻¹ around Hα. For the broad-line sources (identified as such by visual inspection of the optical spectra), we masked a larger region with a width of 10,000 km s⁻¹ around prominent hydrogen and helium emission features (Hγ, Hδ, Hβ, Hα, He i, and He ii). The results of these fits are presented in Table 6. Average values of p₁ for our sources were 0.67, very similar to the power-law slopes found for luminous quasars by Richards et al. (2006), with a range of fitted values from 0 to 2.89. The average value for p₀ is 0.47, with values ranging from 0 to 1. As listed in the Table 6, p₀ was calculated for the specific flux at 1 Å and has units of 10⁻¹⁷ erg s⁻¹ cm⁻² Å⁻¹.

As we show in Appendix A, we found no statistical degeneracies between the power-law component and stellar continua based upon our simulations. However, the issue of separating stellar and non-thermal AGN continua is complex. In order to assess the degree of degeneracy in our models, we carefully analyzed the results of our model fits. From our models, 37% of the narrow-line sources (sources in this category tend to be classified as Sy 1.8/1.9 sources by other authors) and 38% of the broad-line sources have contributions of 50% or greater from a power law. We examined the spectra of these sources in the region from 3800 to 4200 Å, which includes the important stellar diagnostic lines of Ca H and K as well as the Hδ absorption. For broad-line sources with high power-law contributions, we find that absorption lines tend to be weak, while [Ne iii] (at 3869 and 3968 Å) and occasionally weaker hydrogen Balmer (Hζ, H, Hδ) emission lines are comparatively strong. For the narrow-line sources, sources with strong power-law contributions tend to have weak to no clearly evident absorption features. Nearly half of these narrow-line spectra have either poor fits to the data (χ² > >1) or no spectral coverage at the blue wavelengths which include important stellar lines like Ca H and K (making the fits less reliable). Therefore, the effects of any degeneracies between power-law and stellar population models are likely small for our purposes (i.e., rough estimates of the continuum).

In Figures 2 and 3, we show examples of the continuum results. Both the original and continuum subtracted spectra are plotted in black with the continuum plotted in blue. For the majority of sources, we find acceptable fits with the stellar + power-law continuum models. Particularly, good fits are obtained for the narrow-line sources. For the broad-line sources, the presence of broad Balmer lines makes it particularly hard to obtain a good fit to the spectrum below ≈4500 Å (see, for example, the spectrum of MCG +04-22-042).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To show how the spectra and continuum models for spectra taken at KPNO compare to the SDSS spectra, we plot the KPNO and SDSS spectra + continuum fits for the four sources with spectra from both in Figure 4. We chose to show the region from 3700 to 6200 Å, a region which includes both prominent emission lines (i.e., Hβ and [O iii]) and intrinsic absorption features (Ca H and K, the G Band, Mg ib, and Na iD). Both the SDSS and KPNO spectra of Ark 347 are well fitted with a continuum dominated by an old stellar population at solar metallicity. The KPNO spectrum of Mkn 417 is found to be dominated by a power law, while the SDSS continuum is dominated by a solar metallicity old stellar population. For the broad-line source MCG +04-22-042, neither the KPNO or SDSS spectra are fitted well at the blue end of the spectrum (due to the hydrogen Balmer lines), making it unsurprising that the models do not match.

Zoom In Zoom Out Reset image size

Finally, for Mkn 18, different metallicities (low in the SDSS spectrum and high in the KPNO spectrum) and galaxy contributions are found. However, as our test models showed, the metallicities are not well determined with the continuum models. The young stellar population contributions are similar for both the KPNO and SDSS spectra, leaving the discrepancy in the intermediate and old contributions as a likely effect of the degeneracy we found in our test models between the intermediate and old populations. The difference in the continuum flux between the KPNO and SDSS spectra of Mkn 18 is an extreme case, likely due to the fact that Mkn 18 is highly elliptical and inclined along the E–W direction of the slit in the KPNO observation (15''), while the circular fiber of the SDSS (3'') misses out on this flux.

To measure the properties of the emission lines in the KPNO and SDSS spectra (including the FWHM and flux of each line), we adopted two separate methods for the narrow-line and broad-line spectra. For the narrow-line spectra, we first measured the prominent lines in two distinct regions, the regions surrounding Hβ and Hα. At the blue end of the spectrum, we fixed the positions of the Hγ, Hβ, and [O iii] lines (λ4959 and λ5007), requiring that the velocity offset and FWHM of the lines remain the same for all of the lines measured, and fit for the flux and equivalent width. For spectra whose wavelength range includes [O ii] λ3727, an important diagnostic for distinguishing low-ionization narrow emission line regions (LINERs; Heckman 1980), we include this line in the fits to the blue end of the spectrum. Additionally, we followed the same procedure to fit the prominent emission lines surrounding and including Hα, [O i] λ6300, [N ii] λ6548, [N ii] λ6584, [S ii] λ6716, and [S ii] λ6731. The intensities of the [N ii] lines are fixed such that the λ6548 line is at a 1:2.98 ratio with the λ6584 line, as dictated by atomic physics. For all of the narrow-line fits, the FWHM was corrected for the instrumental resolution (200 km s⁻¹ at 5007 Å for the KPNO spectra and 75 km s⁻¹ for the SDSS spectra) and we placed the restriction that the FWHM values have a lower limit of 50 km s⁻¹and an upper limit of 1000 km s⁻¹. The results are recorded in Table 7. In Table 8, we include the intensity ratios for additional weaker lines (i.e., Hδ, [N i], He i) measured in the spectra.

For the broad-line sources, two complications arise which prevent us from performing the same analysis as for the narrow-line sources. First, greater uncertainties exist in the continuum measurements. Second, the lines cannot be fitted by simple Gaussians with the same widths. While the hydrogen Balmer lines of many of the broad-line sources show asymmetries, we chose to fit both Hα and Hβ with a combination of narrow and broad Gaussians. To ensure the uniform measurements of the lines in our spectra, we used an automated process which focused on fitting lines in a narrow region surrounding both the Hβ and Hα lines, separately.

In the Hβ region, defined as the region from 4600 to 5200 Å, we fit a combination of three narrow Gaussians to [O iii] 5007 Å. The use of three Gaussians allowed us to reproduce the shape more robustly, since this line often shows extended wings. The narrow-line shapes, particularly the widths of these lines, were applied to the narrow He ii 4686 Å, Hβ 4861 Å, and [O iii] 4959 Å lines. Both the flux and velocity offsets of each line were allowed to vary. The continuum was fitted with a linear function in a region unaffected by the prominent lines. Finally, these fitted narrow components were combined with a broad Hβ line, which was modeled with a single broad Gaussian component, and re-fitted. The use of essentially a narrow and broad Gaussian allows us to estimate the flux and width of each component, important in estimating the black hole mass (based on the FWHM in the broad component) and emission-line ratios (which depend on the ratio of the narrow lines). Results of these fits are included in Table 9, including the measured continuum flux at 5100 Å. The recorded values of FWHM for the narrow component apply to the strongest narrow-line component of the three Gaussians used to fit the [O iii] 5007 Å line.

In the Hα region, defined as the region from 6200 to 6900 Å, we used the narrow [O i] 6300 Å line to define the initial guess for the velocity offset of the measured lines and the set FWHM of a single Gaussian component. The offset velocities and fluxes of the remaining narrow Hα line, [N ii] lines, and [S ii] lines were allowed to vary. However, the intensities of the [N ii] lines are fixed such that the λ6548 line is at a 1:2.98 ratio with the λ6584 line. A linear continuum was fitted in a region unaffected by the emission lines. The narrow lines were added to a single broad Gaussian for broad Hα and re-fitted. Results from these fits are recorded in Table 10. Examples of fits to both the Hβ and Hα regions are shown in Figure 5. The largest uncertainties involved in these fits are associated with the measurements of Hα and the two [N ii] lines, which are blended in our broad-line spectra, particularly for a source such as MCG +04-22-042.

Zoom In Zoom Out Reset image size

Additionally, weaker lines that are also present in the spectra were measured by manually selecting a continuum region surrounding the selected emission feature. The fluxes of each of these measured lines are included in Table 11. Where broad lines were present and clearly separable from a narrow component, the indicated flux is for the narrow component.

As an alternate method of determining ages of the host galaxies from the stellar continuum fits, we measure the strength of stellar absorption features directly from the non-galaxy continuum subtracted (both with and without subtraction of the AGN non-thermal component) spectra of our sources. This method is analogous to the work measuring Lick-indices by Worthey & Ottaviani (1997). However, instead of broadening our spectra to the velocity dispersion of the Lick/IDS spectral library (9 Å), we follow the procedure outlined in Kauffmann et al. (2003b) for SDSS spectra, which instead compares the measured indices to the Bruzual & Charlot (2003) stellar models. For further details on the SDSS analysis, along with a comparison of the measured indices with additional high resolution stellar libraries, see the discussion in Kauffmann et al. (2003b).

Two particularly important indicators of the age of a stellar population were used extensively in galaxy studies using SDSS spectra (Kauffmann et al. 2003a, 2003b; Gallazzi et al. 2005; Kewley et al. 2006). These are the 4000 Å break (measured with D_n(4000)) and the equivalent width of Hδ absorption (measured with Hδ_A). Among these, the 4000 Å break, or Ca ii break, is observed as a discontinuity in the optical spectrum, caused mainly by the presence of absorption features from metals below 4000 Å. Since the opacity of metals in young, hot stars is low, this feature is weak in young stellar populations and strong in old populations. As a measurement of the Ca ii break, we use the definition of Balogh et al. (1999) to compute

$$\begin{equation} D_n(4000) = \frac{\int ^{4000}_{4100} f_{\lambda } d\lambda }{\int ^{3850}_{3950} f_{\lambda } d\lambda }. \end{equation} \tag{ 1 }$$

While strong Ca ii breaks indicate old populations, strong equivalent widths of Hδ absorption indicate a recent burst of star formation within 0.1–1 Gyr (Worthey & Ottaviani 1997). Therefore, we measure

$$\begin{equation} {\rm H}\delta _A = (4083.50 - 4122.25)(1 - (F_I/F_C)), \end{equation} \tag{ 2 }$$

where F_I is the flux of the line within the bandpass of the feature (4122.25–4083.50) and F_C is the flux in a pseudo-continuum. The pseudo-continuum is defined as the line drawn through the average of the flux in the continuum immediately blueward (λλ 4041.60–4079.75) and redward (λλ 4128.50–4161.00) of the Hδ absorption feature.

Half of the spectra show an Hδ emission line (10 narrow-line sources and 16 broad-line sources), while emission from [Ne iii] 3869 Å is often present in the pseudo-continuum from which D_n(4000) is measured. For the narrow-line sources in our sample, we subtracted the measured narrow lines before calculating these age indicators. Such a calculation is not straight forward for the broad-line sources, where broad emission features are often present in the region containing Hδ_A (with Hδ emission) and D_n(4000) (including [Ne iii] + H7 λ3968 Å and Hδ). In Figure 6, we plot examples of spectra for both narrow and broad-line sources, where stellar absorption features are seen.

Zoom In Zoom Out Reset image size

In Figure 7, we plot Hδ_A versus D_n(4000) for our sources, excluding broad-line sources with prominent Hδ emission. We plot the values measured both after subtracting the power-law continuum (Table 6; top plot) and from the original dereddened spectrum (bottom plot). From each of these measurements, the D_n(4000) break does not change appreciably whether or not the power-law component is subtracted, with a median value of 1.26 for narrow-line sources and 0.91 for broad-line sources when the power law is subtracted and 1.41 (narrow) and 0.92 (broad) without the subtraction. The Hδ_A values are affected, however, for the narrow-line sources with median values of 0.81 (narrow) and −2.15 (broad) with the power law subtracted and 1.73 (narrow) and −2.15 (broad) without the subtraction. To test whether any aperture effects influenced our measurements, we plotted each of these diagnostic measurements against redshift. With no correlation in either Hδ_A or D_n(4000) with z, we conclude that there are no obvious aperture effects to be accounted for in our measurements.

Zoom In Zoom Out Reset image size

The majority of the narrow-line sources occupy the area expected from our stellar population model tests, discussed in Appendix A and plotted in Figure 19. The broad-line sources, however, occupy a region with considerably lower values of Hδ_A. This is true even for sources where an Hδ emission line is not seen in the spectrum (as for the sources plotted). From visual inspection of the Hδ region of our sources, we find that unlike the narrow-line sources, we cannot clearly identify an Hδ absorption feature in any of the broad-line sources. In most cases, we see emission features that are often broad. The low values of Hδ_A measurements for broad-line sources are therefore a likely effect of emission in this region.

In addition to these stellar age diagnostics, we measured additional absorption indices for common stellar absorption features. These values were measured using the same method as used for the Hδ_A index, first subtracting the emission-line spectra for the narrow-line sources and subtracting the power-law component for all of the sources. Bandpasses and continuum ranges are defined in Worthey et al. (1994) and Worthey & Ottaviani (1997). In Table 12, we present the stellar age indicators (D_n(4000) and Hδ_A) along with six metallicity indicators, chosen to sample indices sensitive to several different elements (i.e., C, N, Ca, Mg, Fe). Two of these indices are combinations of other indices, defined in González (1993):

$$\begin{equation} [{\rm MgFe}] = \sqrt{\rm Mgb \langle Fe\rangle }{ \rm and } \langle {\rm Fe}\rangle = \frac{1}{2} ({\rm Fe 5270 + Fe 5335}). \end{equation} \tag{ 3 }$$

We use the modified form of [MgFe]', defined by Thomas et al. (2003) as

$$\begin{equation} [{\rm MgFe}]^{\prime } = \sqrt{\rm Mgb (0.72 Fe5270 + 0.28 Fe5335)}. \end{equation} \tag{ 4 }$$

Additionally, to better understand our results, we also measured these stellar absorption indices for a sample of test spectra created from the stellar population models used for the continuum fits. We discuss these results, where we used different combinations of stellar ages and metallicities, in Appendix A. Of the additional stellar absorption indices recorded in Table 12, Hδ emission could affect the value measured for CN₁. Additionally, He ii 4686 Å is within the range of C₂ 4668 and [N i] 5199 Å is within the range of Mgb. Since [N i] 5199 Å is weak in our broad-line sources, we expect little error in our Mgb measurements. In Figure 8, we plot various metallicity indicators and the age indicator D_n(4000) versus the metallicity indicator Mgb for our target sources. Comparing with our results from the test spectra, it appears that C₂ 4668 is the most affected by "contaminating" emission features. The narrow-line sources should be unaffected, however, since we have subtracted the emission components from their spectra.

Zoom In Zoom Out Reset image size

Based on a comparison of the plots in Figure 8 with the test spectra values, we find that the [MgFe]' versus Mgb and 〈Fe〉 versus Mgb plots are the best indicators of the metallicity of the stellar populations. However, the only clear result is that we do not find old, high-metallicity (2.5 Z_☉) populations within our sample (all of the old population test spectra have Mgb ≲2, as determined from the D_n(4000) versus Mgb plot). Since there is little difference in the parameter space occupied by solar and low-metallicity populations, we cannot discern anything more from our measured stellar absorption indices.