A SAMPLE OF SEYFERT-2 GALAXIES WITH ULTRALUMINOUS GALAXY-WIDE NARROW-LINE REGIONS: QUASAR LIGHT ECHOES?^*

Deep ugriz CFHT/Megaprime data (Table 1) were obtained by us through Opticon proposal 2008BO01 at 2008 September 20 in excellent conditions to study a supercluster of galaxies at z = 0.45. Further details about these data can be found in Schirmer et al. (2011). J2240 at z = 0.326 is a serendipitous discovery in these images, extending over 7'' × 4''. With a pixel scale of 0186 and 07 image seeing the galaxy is well resolved. It has irregular morphology and peculiar colors similar to that of a GP (see Figure 1), with the exception of very low u-band flux (u − r = 4.06 mag). K-corrections and luminosities were calculated using kcorrect v4.2 (Blanton & Roweis 2007). Most noteworthy are a bright non-stellar nucleus in a disk-like body reminiscent of a spiral galaxy and a half-detached cloud about 12 kpc southwest of the nucleus. The cloud extends over 8 × 18 kpc.

Zoom In Zoom Out Reset image size

Table 1. Characteristics of the CFHT/Megaprime Data and J2240 Photometry

Band	texp (s)	Seeing	Mlim	MAB	L
u	4250	099	25.0	21.80	−19.4
g	3000	076	25.4	19.02	−21.3
r	5000	066	25.4	17.74	−22.2
i	6000	057	25.2	18.71	−23.0
z	5000	073	23.3	17.89	−23.2

Notes. The limiting magnitudes represent the 5σ completeness limit for non-stellar sources. Luminosities are dereddened.

Download table as:  ASCIITypeset image

We hereafter refer to these two components simply as the "Disk" and the "Cloud," without implying any properties about their physical shape. Spectroscopic observations reveal a much more complex picture, as can be seen in the right panel of Figure 1. We refer to the central ±4 kpc encompassing the nucleus as the "Center."

Figure 1 shows a fainter reddish galaxy with a photometric redshift of z = 0.37 ± 0.06, projected 20 east of the nucleus of J2240. A second galaxy with a prominent spiral arm is located 45 to the southeast and has the same spectroscopic redshift as J2240 (z = 0.326).

2.2.1. VLT/FORS2

2.2.2. VLT/XSHOOTER

2.2.3. Gemini/GMOS

2.2.4. Correction for Galactic Extinction

The theoretical value of the [O iii] λλ4960, 5008 intensity ratio has long been underestimated compared to observations (e.g., 2.89 by Galavis et al. 1997). Leisy & Dennefeld (1996) found 3.00 ± 0.08, consistent with high-S/N measurements of other nebulae, and encouraged improved theoretical calculations to be carried out.

Storey & Zeippen (2000) calculate precise line ratios for [O iii] λλ4960, 5008 and homologous cases (such as [N ii] λλ6550, 86, [Ne iii] λλ3869, 3968, or [Ne v] λλ3347, 3427) using relativistic corrections. The intensity ratios of these lines are independent of temperatures and densities in typical astrophysical nebulae, as the two lines come from a common upper level. The intensity ratio will change only if the lines become optically thick. This is unlikely, however, given their low absorption coefficients. If the density was high enough to cause significant optical depth, collisional de-excitation would suppress these lines in the spectrum (P. J. Storey 2012, private communication). The ratio of the transition probabilities between [O iii] λ5008 and [O iii] λ4960 is found to be 3.013, which translates into a flux ratio of 2.984 by multiplying with the wavelength ratio 4960/5008. This value has been confirmed (2.993 ± 0.014) by Dimitrijević et al. (2007) based on 34 SDSS AGN spectra. Likewise, for [N ii] λ6586 and [N ii] λ6550, the flux ratio evaluates to 3.05.

We use these theoretical values to deblend the Hα–[N ii] line complex and to verify our spectral analysis method with the [O iii] doublet.

As a first step we subtract the continuum, which we linearly interpolate across an emission line based on the adjacent ∼100 pixels on both sides. The resulting continuum is smoothed in dispersion direction with a 100 pixel wide median kernel and subtracted from the original spectrum. We do not correct line fluxes for potential stellar absorption, as the equivalent widths are high (several 100 Å to more than 1000 Å).

In the second step we extract images centered on each line of interest, encompassing their full spatial extent and 2.3 Å in width. These images are then stretched by a factor of λ_line/λ_Hβ to correct for the wavelength dependence of the velocity broadening. We choose Hβ as the reference frame, as it is in the middle between the blue [O ii] and red [S ii] lines. The re-projection onto Hβ also includes a two-fold binning along the dispersion direction to increase the S/N. We retain an effective spectral resolution of R ∼ 3800.

Third, we have to register the line images such that they overlap precisely. Distortion correction by the XSHOOTER pipeline left no measurable spatial offsets of the continuum recorded in the UVB and VIS arms. Registering the lines in dispersion direction is more difficult, as their individual appearances are different. Fortunately, almost all lines exhibit traces of the ENLR with low line widths, which we use as a reference mark. Remaining lines are registered using other common features. In this way registration in dispersion direction is accurate to ∼1 pixel, well within the oversampled spectral resolution.

Lastly, we smooth with a 2 pixel wide Gaussian kernel, cut off at a radius of 2 pixel. Prior to smoothing, the highest and lowest pixels in the aperture are rejected. In this way we suppress spurious noise features, which can get strong when calculating ratios with small denominators.

Error maps from the XSHOOTER pipeline are treated analogously and fully propagated.

Velocity broadening in J2240 leads to some overlap between Hα and [N ii] λλ6550, 86 and within the [S ii] λλ6718, 33 doublet. While the bright cores of these lines are well separated, their fainter wings overlap. We deblend them as follows.

For the Hα–[N ii] complex (Figure 2, panel 1) we have to perform several steps. First, we smooth Hβ with a 2 pixel wide Gaussian and subtract it from Hα after multiplication with 3.0. The latter is near the average Case B hydrogen ratio (3.08) in Seyfert-2 AGNs (Gaskell & Ferland 1984). In this way we remove about 90%–95% of the Hα contamination in the [N ii] lines (Figure 2, panel 2), thus preparing a first guess of the [N ii] doublet. Some residual Hα is still present, occurring due to variable extinction in J2240 (Section 4.1).

Zoom In Zoom Out Reset image size

Next we extract [N ii] λ6586 within a closed polygon line and then smooth and subtract it from the original Hα-[N ii] complex. We also divide it by 3.05 (Section 3.1), shift it to the position of [N ii] λ6550, and subtract it once more. The [N ii] doublet has thus been removed, and we have prepared a clean Hα line (Figure 2, panel 3). Lastly, we smooth the Hα image and subtract it from the original line complex, leaving the final [N ii] pair.

For [S ii] λλ6718, 33 we use a different approach, as this doublet is reasonably separated with little overlap. We define non-overlapping closed polygonal lines of identical shape around both lines. Pixels within each polygon line are smoothed with a small Gaussian kernel and subtracted, leaving the other line in the doublet isolated.

Note that our 2D analysis technique used in the following is confined to regions with high S/N. Areas where the [S ii] doublet or Hα and [N ii] overlap are already faint and mostly excluded.

As shown in Section 3.1, the [O iii] λλ4960, 5008 line ratio is independent over a large range of temperatures and densities. Both lines are strong and uncontaminated and thus serve as a good test case to validate our 2D analysis method. While a constant offset indicates a problem with flux calibration, systematic variations along spatial or spectral directions hint at bad registration of the two lines, or problems in the data processing or extinction correction.

Figure 3 shows that in our data the [O iii] ratio is constant within errors, thus validating our 2D analysis method. Utilizing the area where j_λ5008/j_λ4960 uncertainty is less than 0.03, we initially estimated a line ratio of 3.169 ± 0.012 (including full extinction correction; see Section 4.1). This is inconsistent with the theoretical value and amongst the highest values measured by Dimitrijević et al. (2007). Retracing our processing steps, we identified an unfortunate coincidence causing this discrepancy. The [O iii] λ4960 line is redshifted to 6577 Å, just redward of Hα. Being a white dwarf, the spectrophotometric standard Feige 110 shows a ∼20 Å wide Hα absorption line in the XSHOOTER spectrum. The reference flux values tabulated by Oke (1990), however, are based on a low resolution spectrum (∼5 Å pixel⁻¹), resulting in a measured line width of ∼40 Å. As the [O iii] line is narrow (1.5 Å) compared to the absorption line, we can easily determine a correction factor of 1.075 ± 0.015. The [O iii] intensity ratio then becomes 2.950 ± 0.015. The same effect is present in our FORS2 data (for which a different white dwarf served as flux calibrator), yielding a corrected ratio of 3.01 ± 0.04. No other emission lines are affected.

Zoom In Zoom Out Reset image size

4.1.1. Method

The extinction of a line with intrinsic intensity I_λ0 is given as I_λ = I_λ0 e^−C f(λ), where C is a constant and f(λ) parameterizes the extinction curve of the ISM and depends on the chemical and physical properties of the dust grains. Using the observed and intrinsic (before attenuation by dust) intensity ratio of two emission lines, the color excess is

$$\begin{equation} E(B-V) = -2.5\,{\rm log} \left(\frac{I_\lambda /I_\nu }{I_{\lambda 0}/I_{\nu 0}}\right) \frac{f(B)-f(V)}{f(\lambda)-f(\nu)}\;. \end{equation} \tag{ 1 }$$

We use the Hα/Hβ ratio and the standard R = 3.1 model (Cardelli et al. 1989) to calculate the reddening in J2240. This requires knowledge of the intrinsic value of Hα/Hβ, which equals ∼2.87 for Case B conditions in a typical H ii region. In the harder radiation field of an AGN collisional excitation of Hα becomes important, increasing the ratio to ∼3.08. However, it depends strongly on L_X/L_opt and the slope of the X-ray spectrum, as well as on metallicity (Gaskell & Ferland 1984). The intrinsic line ratio can get as low as 2.6 for high metallicities, or as high as 3.4 for low metallicities.

We take the metallicity into account in an explicit manner. Starting with a fixed intrinsic Balmer line ratio of 3.0, we compute an initial metallicity estimate following Storchi-Bergmann et al. (1998). Once a first guess for the metallicity is available, we update the intrinsic Balmer line ratio using an interpolation function to the values given in Gaskell & Ferland (1984). Pixel values now deviate individually from the initial guess of 3.0. With the improved extinction map we calculate a new abundance map. Convergence is achieved rapidly (less than 1% relative change for most pixels) after the first iteration; thus, we stop after a second iteration. Different starting point values of 2.7 and 3.2 for the intrinsic Balmer ratio converge to the same solution.

4.1.2. The Reddening Map

The resulting reddening map is shown in the right panel of Figure 4. It is based on the Seyfert 1 model X-ray spectrum of Gaskell & Ferland (1984). Adopting their ν⁻¹ power-law continuum (dashed line in their Figure 1), E(B − V) is systematically reduced by 0.16 mag, but the structures seen remain unchanged. We work with the higher extinction model for the rest of this paper, commenting on the effects of a different X-ray spectrum where applicable.

Zoom In Zoom Out Reset image size

The dust in J2240 is distributed unevenly. The area between the nucleus and the Cloud is highly reddened with E(B − V) = 0.5–0.7 mag. It is independent of radial velocity and likely caused by a large cloud of foreground dust in J2240. On the opposite side of the nucleus (above the dashed line in Figure 4), we find that reddening grows as a function of radial velocity (blueshifted components are less reddened than redshifted ones). In this area, the dust appears to be evenly embedded within the NLR (where it can survive; see Laor & Draine 1993), causing E(B − V) to increase from 0.05 to ∼0.7 mag as we look deeper into J2240. Other Seyfert-2 galaxies show similar (global) extinction values of E(B − V) = 0.1–0.6 for their NLRs (Ho et al. 1997; Rhee & Larkin 2005).

4.1.3. Error Analysis

4.1.4. Testing Different Dust Models

The electron temperature, T_e, for the medium-ionization zone can be obtained from [O iii], utilizing

$$\begin{equation} \frac{j_{\lambda 4960}+j_{\lambda 5008}}{j_{\lambda 4364}} = \frac{7.9 \, {\rm exp} (3.29\times 10^4 / T_{\rm e})}{1+4.5 \times 10^{-4}\,n_{\rm e} / T_{\rm e}^{0.5}} \end{equation} \tag{ 2 }$$

(Osterbrock & Ferland 2006). Similar relations exist for [Ne iii] ((j_λ3869 + j_λ3968)/j_λ3343) and [N ii] ((j_λ6548 + j_λ6583)/j_λ5755). For electron densities n_e ≲ 10⁵ cm⁻³, these temperature probes become independent of n_e. As the redshifted (and weak) [N ii] λ5755 line is lost in the atmospheric O₂ absorption feature at 7630 Å, we cannot use [N ii] to probe the lower ionization regions. Instead, we use [O ii] j_λ3727/j_{λ7321, 32} and Equation (21) from Peimbert (1967), together with an average electron density of 200 cm⁻³ for the Disk.

Due to the weak auroral lines, we cannot reconstruct 2D temperature maps, yet we can still measure global values. Using [O iii], we find 17,100 K and 14,700 K for the Disk and the Cloud, respectively (Table 3). [Ne iii] results in 18,100 K for the Disk, whereas the Cloud is undetected in [Ne iii] λ3343 and only a lower limit of 11,000 K can thus be obtained. Errors account for the significant blending of [Ne iii] λ3343 with the much brighter [Ne v] λ3347. For the lower ionization zone we find 13,000 K from [O ii].

The [S ii] j_λ6718/j_λ6733 ratio depends on n_e and T_e. We use T_e = 13, 000 K as obtained from [O ii] for the low-ionization zone. Results are shown in the lower right panel of Figure 5. The Disk has significantly higher density than the rest, starting from 150–200 cm⁻³ and peaking in a dynamically cool spot 1 kpc northeast of the nucleus, where n_e reaches 650 cm⁻³. Decreasing (increasing) T_e by 5000 K raises (lowers) the peak density by 130 (80) cm⁻³, which is significantly less than the density variations measured. These features are thus real and not a consequence of our simple assumption of constant temperature.

Zoom In Zoom Out Reset image size

The Cloud is in the low-density limit of the [S ii] probe; thus, only an upper limit of n_e < 50 cm⁻³ can be inferred. The area northeast of the nucleus, where J2240 is brightest in [S ii], has lower density than the Disk (80–150 cm⁻³).

Abundances for H ii regions are often obtained with the strong-line method (Kewley & Dopita 2002). A more accurate approach is the direct method (e.g., implemented in the nebular/ionic task in IRAF based on the five-level atomic model by De Robertis et al. 1987). Abundance determinations for NLRs are notoriously more difficult. Storchi-Bergmann et al. (1998) have developed an empirical strong-line method based on NLRs with embedded H ii regions. The underlying assumption is that both the H ii regions and the NLR share the same metallicity; thus, a calibration for NLRs without discernable H ii regions can be inferred. Storchi-Bergmann et al. (1998) offer two methods, one based on [N ii]/Hα and [O iii]/Hβ, and the other on log$({\rm [{\rm O}\,{\scriptsize{II}}]/[{\rm O}\,{\scriptsize{II}}]})$ and log$({\rm [{\rm N}\,{\scriptsize{II}}]/H\alpha })$. The latter depends on extinction correction as [O ii] and [O iii] are not neighboring lines. In this work we also included their correction term for electron density.

Both methods yield similar results of Z = (0.45–0.55)Z_☉, the second one with slightly larger values of Δlog(O/H) = 0.03 (Figure 5). Abundances are enhanced by about 0.1 toward the nucleus and in the Cloud and appear to increase systematically with growing radial velocity. Such sub-solar metallicities are uncommon in NLRs, as has been shown by Groves et al. (2006). Their study of 23,000 Seyfert-2 galaxies in SDSS yielded only ∼40 galaxies of evidently low metallicity. Using their prescription, we obtain average metallicities of 0.9 Z_☉ (from [N ii]/[O ii] and [O iii]/[O ii]) and ∼2.0 Z_☉ (from [N ii]/Hα and [O iii]/Hβ).

A large uncertainty in the abundance determination stems from the assumptions made about the ionizing spectrum. Ludwig et al. (2012) report that changes to the slope and shape of the spectrum can change metallicities from sub-solar level to several times the solar value. As we do not know the intrinsic X-ray spectrum of the AGN in J2240, the true absolute value of the abundance remains unknown. The variations measured in the abundance map may be real but can be altered by local variations of the X-ray spectrum filtered by the inhomogeneous ISM. We therefore do not consider abundances further in this work, leaving this task to future photoionization modeling once X-ray data become available. Also, note that increasing the metallicity from 0.5 Z_☉ to several Z_☉ changes the intrinsic Hα/Hβ Balmer ratio from 2.75 to about 2.60–2.65, increasing E(B − V) only marginally.

Once the metallicity is known, we can derive the ionization parameter U (Figure 5, lower left) from [O iii]/[O ii] (Kewley & Dopita 2002). Applying the best-fit relation from Penston et al. (1990), we find similar structures and values (not shown). Using [O iii]/[O ii] requires n_e < 10³ cm⁻³ (Komossa & Schulz 1997), since mixed-in higher densities will lead to an overestimation of U. This condition is well met (Figure 5, lower right).

We find two strong peaks in the ionization map located 0.0–2.0 kpc northeast of the nucleus (above the dashed line in Figure 5) and spread over a radial velocity range of 210 km s⁻¹. The spatial offset is coincident with that measured for the density peak and can either indicate the location of a second, deeply buried AGN or simply be the result of a shock or interaction with a jet. Two weaker peaks are found 4.5 and 8 kpc southwest of the nucleus (below the dashed line), significant on the 3σ and 2σ level, respectively. The first is located between the nucleus and the Cloud in the area with highest dust extinction, whereas the second is centered in the Cloud and redshifted by 110 km s⁻¹ with respect to the highest ionization peak.

The error map for U takes into account measurement errors, but not the uncertainty in metallicity. To test the latter, we lower the metallicity by ΔZ = 0.1 Z_☉, which corresponds to going from high- to low-metallicity areas. Log U then decreases by maximally 0.12, much less than the absolute variations observed (Δ(log U) = 0.85), and less than the direct measurement errors. Note that while the mentioned variations in U are real, its absolute value should not be taken at face value, as the absolute metallicity is unknown.

J2240 exhibits a relatively flat continuum with the brightest parts at 4500–7000 Å rest-frame wavelengths (Appendix A, Figure 12). With an extent of about 8 kpc in the 2D spectra, continuum radiation emerges from a more compact region than the line emission. A significant fraction of the continuum is stellar, as we clearly see the Ca ii H absorption line at 3934 Å. Ca ii K at 3969 Å is superimposed by [Ne iii] and H emission. The 4000 Å break is hardly visible in the data.

No continuum can be seen in the 2D spectra at the position of the Cloud. However, we can still test for the presence of continuum flux by integrating over all wavelengths excluding emission lines. The S/N obtained is sufficient to reconstruct broadband continuum spectral energy distributions (SEDs) for the Disk and the Cloud (Figure 6). While their colors are similar above 5500 Å, the Cloud is bluer than the Disk at shorter wavelengths. Both a nebular continuum and a different stellar population can cause this.

Zoom In Zoom Out Reset image size

We find a symmetric continuum profile within 5 kpc of either side of the nucleus (Figure 7). In the northeast the profile is nearly exponential with a Sersic index of 0.85 ± 0.04 and a disk scale length of 3.0 ± 0.1 kpc. In the southwest, we see a long tail through the Cloud (the bump at −10 kpc in Figure 7), traceable over 18 kpc.

Zoom In Zoom Out Reset image size

The velocity FWHMs (parameter-free, directly measured in the line images) as a function of slit position for selected lines are displayed in Figure 7, together with their intensity profiles. While the latter for low- and medium-ionization lines are well distinguished (bright and dark gray areas) within 8 kpc of the nucleus, the velocity profiles are similar. The notable exception is [O ii], whose FWHM is consistently larger by 222  ±  15 km s⁻¹ but otherwise has the same shape. This is a result of the absence of compact bright cores in [O ii], which are found in, e.g., Hα, [O iii], and [S ii] (see Figure 11).

Defining the maximum velocity as the point where the line profile drops to 20% of the maximum emission, we find v_max = (0.98 ± 0.05) × FWHM averaged over Hα, [O iii], and [S ii]. We find little variation (±0.14) along the slit. Highest velocities are blueshifted 1200 km s⁻¹.

The [O ii] velocity profile is remarkably flat between −14 kpc and +8 kpc, with line widths of 534 ± 69 km s⁻¹. Hα, [O iii], and [S ii] vary between 200 and 550 km s⁻¹ and drop to less than 100 km s⁻¹ for the largest nuclear distances. J2240 is embedded in an ENLR that corotates with the disk (albeit slower; the northeast part is redshifted by 195 km s⁻¹ with respect to southwest). The ENLR's low-luminosity, low line width, high-excitation state (${\rm log([{\rm O}\,{\scriptsize{III}}]/H}\beta)=0.86\hbox{--}0.98$) and rotation are typical (Unger et al. 1987; Durret 1989).

Deep GMOS imaging reveals a halo stretching 60 kpc from the nucleus (Figure 8, top panel). To distinguish between a stellar tidal stream and ionized gas, we use a 3 hr GMOS long-slit spectrum through a 15 slit, aligned along the major halo axis (at a position angle of 71°). Unresolved [O iii] λ5008 is detected out to 42 kpc (Figure 8, bottom panel); thus, an additional underlying stellar population can currently not be ruled out. The ENLR has a minimum diameter of 70 kpc.

Zoom In Zoom Out Reset image size

So far we have anticipated that the NLR is powered by an AGN. Observational evidence is presented in the following.

Hypothetically assuming that all line emission is powered by stars, we can estimate the necessary star formation rate using the extrapolated total line fluxes from Table 4 and the scaling relations from Kennicutt (1998). The result equals 290  ±  60 M_☉ yr⁻¹, needed to explain the Hα flux, 20 times higher than that in an average star-forming GP galaxy (Cardamone et al. 2009).

Table 4. Fluxes (Corrected for Galactic Extinction) and Luminosities of J2240 for Selected Emission Lines, and Their Relative Strength with respect to Hβ

Line	Islit	Lslit	Ltot	I(λ)/I(Hβ)
	(× 10−16 erg s−1 cm−2)	(× 1040 erg s−1)	(× 1040 erg s−1)
Hα	338.6 ± 2.8	1200.2 ± 9.9	2280 ± 360	4.25
Hβ	79.7 ± 1.2	282.4 ± 4.1	536 ± 85	1.00
Hγ	31.6 ± 0.9	111.9 ± 3.3	212 ± 34	0.40
Hδ	12.8 ± 0.3	45.6 ± 1.1	86 ± 13	0.16
He i λ5877	7.6 ± 0.1	26.7 ± 0.6	50 ± 8	0.09
He ii λ4687	10.6 ± 0.4	37.5 ± 1.3	70 ± 11	0.13
He i λ10833	23.7 ± 0.8	84.0 ± 2.9	159 ± 25	0.30
[O i] λ6302	44.0 ± 0.6	156.0 ± 2.1	296 ± 46	0.55
${\rm [{\rm O}\,{\scriptsize{II}}]}\,\lambda$3727	296.6 ± 4.1	1051.3 ± 14.7	1997 ± 316	3.72
[O iii] λ4364	10.3 ± 0.5	36.4 ± 1.9	69 ± 11	0.13
${\rm [{\rm O}\,{\scriptsize{ III}}]}\,\lambda$5008	841.5 ± 14.0	2982.7 ± 49.7	5667 ± 899	10.56
${\rm [{\rm N}\,{\scriptsize{ II}}]}\,\lambda$6586	187.7 ± 1.6	665.3 ± 5.5	1264 ± 199	2.36
${\rm [S\,{\scriptsize{II}}]}\,\lambda$7618,33	168.1 ± 1.6	595.6 ± 5.4	1131 ± 178	2.11
${\rm [S\,{\scriptsize{ III}}]}\,\lambda$9071	29.6 ± 0.9	104.9 ± 3.1	199 ± 32	0.37
${\rm [S\,{\scriptsize{III}}]}\,\lambda$9533	107.2 ± 2.6	379.8 ± 9.2	721 ± 115	1.34
${\rm [{\rm Ne}\,{\scriptsize{ III}}]}\,\lambda$3869	56.3 ± 0.9	199.6 ± 3.0	379 ± 60	0.71
${\rm [Ne\,\scriptsize{V}]}\,\lambda$3427	17.4 ± 0.4	61.8 ± 1.6	117 ± 18	0.22
${\rm [{\rm Mg}\,{\scriptsize{ II}}]}\,\lambda$2799	19.5 ± 0.5	69.1 ± 1.9	131 ± 21	0.24
${\rm [{\rm Ar}\,{\scriptsize{ III}}]}\,\lambda$7138	8.9 ± 0.1	31.4 ± 0.6	59 ± 9	0.11

Notes. Total values are corrected for slit losses by seeing and for the fact that the slit is smaller than the object. The correction factor used is 1.9 ± 0.3, its uncertainty conservatively estimated.

Download table as:  ASCIITypeset image

An AGN can be distinguished from star formation due to its harder radiation field changing the line ratios. Baldwin et al. (1981) present various tools for this purpose. Their BPT diagrams have been further developed by Veilleux & Osterbrock (1987), Kewley et al. (2001), Kauffmann et al. (2003), and Kewley et al. (2006). For J2240, the S/N is good enough to run this analysis on the basis of individual pixels. While the precise locations of the dividing lines between SF, AGNs, LINERs, and composite objects in the BPT plots are still under some debate (Cid Fernandes et al. 2010), the classification of J2240 as AGN is beyond doubt (Figure 10). [O i] λλ6302, 6366 emission across J2240 shows that shock/ionization fronts are present on a global level, emphasizing a powerful AGN.

When fitting the continuum across emission lines, we have ignored stellar absorption as the emission lines are much brighter than the continuum. If stellar absorption was present, Hα and Hβ fluxes are underestimated because of our negligence, driving the data points in the BPT diagrams away from star formation. Assuming a worst-case scenario with (uncorrected) 100% stellar absorption, BPT values are still far outside the star-forming area (see arrows in Figure 10).

While star formation appears to be negligible in J2240, Bennert et al. (2006b) warn that H ii regions can dominate the [O iii] flux at larger nuclear distances. Levenson (2007) and Kauffmann et al. (2003) also show that intense star formation is common in active galaxies. Indeed, a significant starburst can be buried by the [O iii] emission in J2240. However, while we do observe a decrease in log([O iii]/Hβ) with increasing radius (middle left panel in Figure 10), the AGN characteristics remain well preserved. In addition, if the NLR is due to a massive outflow of hot gas, the latter may have disrupted star formation (for an example, see Cano-Díaz et al. 2012).

J2240 has similar colors as typical GP galaxies. Applying the GP filter of Cardamone et al. (2009) to SDSS-DR8, we recover J2240 once we drop their maximum Petrosian radius of 20. Instead, we require a minimum radius of 20 and modified color cuts to select objects with particularly strong [O iii] fluxes (g − r > 1.0). Ninety-five percent of the objects found are spurious or have corrupted photometry (the latter in particular near R.A. = 355 ± 5 and decl. = 60 ± 10; see examples in Appendix A, Figure 13). If in doubt, we downloaded the original SDSS FITS images and created the color images ourselves. In 100% of these cases, the galaxies turn out to have normal colors or be artifacts.

Our SQL query for objects with 0.12 < z < 0.35 (where [O iii] falls into the r band) is shown in Appendix B. Shifting each bandpass by one filter toward redder wavelengths (e.g., replacing g − r with r − i, and leaving out the constraints requiring filters redder than the z band), we also select possible candidates at 0.39 < z < 0.69. Contamination by lower redshift star-forming galaxies with strong Hα emission is expected in this higher redshift bin.

Final manual selection results in a sample of 29 candidates (Table 5 and Appendix A, Figure 13). Redshifts and [O iii]/Hβ ratios have been measured with GMOS. All of our low-redshift candidates observed show high [O iii]/Hβ ratios, confirming their AGN character. Broad-line components are absent but not ruled out due to shallow data. Seven galaxies have a Petrosian radius between 24 and 26 and are good examples of GBs with particularly large NLRs. J2240 (listed as Object 016) is larger than any of the other galaxies (Petrosian radius of 35). Object 004 is noteworthy, as it shows a large extension, which indicates ejected matter or tidal interaction. Several GBs exhibit close companions of unknown redshift, indicating that encounters may play a role in the formation of GBs.

Table 5. GB Sample

ID	Name	SDSS objID	r	Rr	z	${\rm log}({{\rm [{\rm O}\,{\scriptsize{ III}}]}}/{{\rm H}\beta })$	FR	F24 μm	Comments
			(mag)	('')			(mJy)	(mJy)
001	SDSS J002016.4−053126	1237679077517557845	18.3	2.1	0.334	1.305 ± 0.005	9.38	11.7
002	SDSS J002434.9+325842	1237676441460474246	18.2	2.6	0.293	1.246 ± 0.003	...	25.4
003	SDSS J011133.3+225359	1237666091128914338	19.1	2.1	0.318	1.243 ± 0.016	...	23.1
004	SDSS J011341.1+010608	1237666340800364769	18.5	2.6	0.281	1.043 ± 0.006	1.18	39.7	SDSS redshift
005	SDSS J015930.8+270302	1237680284389015833	18.9	2.6	0.278	1.194 ± 0.004	...	18.1
006	SDSS J115544.5−014739	1237650371555229774	17.9	2.4	0.306	1.163 ± 0.006	undetected	16.9
007	SDSS J134709.1+545311	1237661386529374363	18.7	2.0	...	...	3.67	5.3
008	SDSS J135155.4+081608	1237662236402647262	19.0	2.1	0.306	1.161 ± 0.002	undetected	25.7
009	SDSS J144110.9+251700	1237665442062663827	18.5	2.2	0.192	1.086 ± 0.001	2.28	19.6
010	SDSS J145533.6+044643	1237655742407835791	18.5	2.0	0.334	1.151 ± 0.003	undetected	20.4
011	SDSS J150420.6+343958	1237662306730639531	18.7	2.2	...	...	undetected	7.6
012	SDSS J150517.6+194444	1237667968032637115	17.9	2.4	0.341	1.131 ± 0.001	4.89	49.9
013	SDSS J205058.0+055012	1237669699436675933	18.6	2.6	0.301	1.163 ± 0.003	...	49.5
014	SDSS J213542.8−031408	1237680191506678389	19.2	2.0	0.246	1.108 ± 0.004	...	3.2
015	SDSS J220216.7+230903	1237680306395415794	18.9	2.5	0.258	1.154 ± 0.006	...	24.8
016	SDSS J224024.1−092748	1237656538051248311	18.3	3.4	0.326	0.960 ± 0.003	2.85	37.4	Prototype; J2240
017	SDSS J230829.4+330310	1237680503434445439	19.1	2.0	0.284	1.258 ± 0.010	...	13.9
ID	R.A. (J2000)	Decl. (J2000)	i	Ri	z	${\rm log}({{\rm [{\rm O}\,{\scriptsize{ III}}]}}/{{\rm H}\beta })$	FR	F24 μm	Comments
			(mag)	('')			(mJy)	(mJy)
018	SDSS J015629.0+174940	1237679459756015771	19.2	2.1	...	...	...	5.7
019	SDSS J080001.8−095841	1237676676082173676	18.9	2.5	0.402	1.232 ± 0.010	...	14.7
020	SDSS J090015.7+604704	1237663530800382081	16.9	2.6	0.082	...	178.71	25.2	H ii region
021	SDSS J091353.4+601748	1237651274038837416	18.0	3.1	0.477	1.141 ± 0.0027	1.36	31.7
022	SDSS J110140.5+400423	1237661966349893812	18.2	2.6	0.456	1.296 ± 0.014	17.38	18.8	N. Zakamska
023	SDSS J113818.9+060620	1237654606942699779	18.8	2.2	0.499	1.132 ± 0.009	undetected	undetected
024	SDSS J122140.0+190442	1237667914346791043	18.6	2.9	0.542	1.163 ± 0.010	undetected	30.3
025	SDSS J161836.4−040942	1237668567712924379	17.6	2.2	0.211	0.155 ± 0.008	undetected	268.4	H ii region
026	SDSS J172145.7+632233	1237671938731213392	18.9	2.2	...	...	undetected	5.9
027	SDSS J191354.1+621147	1237671938736848903	18.8	2.2	...	...	undetected	32.7
028	SDSS J230901.0+035742	1237679004484370740	18.5	2.3	0.485	1.188 ± 0.007	undetected	12.8
029	SDSS J234130.3+031727	1237678598088425669	17.3	2.4	0.145	0.612 ± 0.014	851.95	121.8	H ii region; LINER
									strong [O i], weak [O iii]

Notes. R_{r, i} are the SDSS Petrosian radii. All but two redshifts originate from our GMOS pilot survey. High values of ${\rm log}({\rm [{\rm O}\,{\scriptsize{ III}}]}/{\rm H\beta })\sim 1$ confirm the presence of an AGN. The last two data columns contain VLA FIRST radio and WISE 24 μm fluxes. No number indicates that the source is either outside the FIRST survey area or no spectra have been taken. Object 016 is the main target analyzed in this paper. Spectroscopic data for Object 022 were taken from archival GMOS-North observations, executed by N. Zakamska (Gemini program GN-2010B-C-10).

Download table as:  ASCIITypeset image

In the higher redshift sample we find 12 objects, of which 7 have spectra taken. Five galaxies have genuine AGN spectra, whereas two are star-forming galaxies at lower redshifts for which Hα was mistaken as [O iii] by our SQL filter. All objects covered by the Very Large Array (VLA) Faint Images of the Radio Sky at Twenty-Centimeters (FIRST) catalog (White et al. 1997) are either non-detected or radio-quiet.

Apart from Object 015, all spectroscopically confirmed GBs have [O iii] lines extending over 15–20 kpc or more (Appendix A, Figure 14). Given integration times of only 300 s, the real extent of the NLRs is likely to be larger. In addition, the limited availability of guide stars is restricting the slit position angle, which consequently cannot be well aligned with the target's major axis in a majority of cases.

Type-2 AGNs are dusty objects with significant optical extinction. The WISE (Wright et al. 2010) mid-infrared fluxes can be used as a proxy for AGN activity, as mid-IR emission correlates with X-ray brightness over a wide range of luminosities (Asmus et al. 2011; Mason et al. 2012). In particular, the W4 filter at 24 μm is not affected by dust absorption. As a comparison sample we choose the 887 type-2 quasars from Reyes et al. (2008), 104 of which have similar redshifts as our GBs (0.25 ⩽ z ⩽ 0.35). The de-redshifted mid-IR spectra of our GBs are very red, following a power law with index 〈a_λ〉 = 1.99 ± 0.35, showing that the emission is likely of nuclear origin as compared to star formation. The 104 type-2 comparison quasars also have red spectra, albeit with a lower slope of 〈a_λ〉 = 1.59 ± 0.43. The null hypothesis that both samples have the same parent SED distribution is rejected on the 5% level based on the Kolmogorov–Smirnov test. The 24 μm luminosities of the GBs and the comparison sample are similar, though; thus, GBs might simply be particularly dusty objects.

In Figure 9, we plot $L_{\rm [{\rm O}\,{\scriptsize{III}}]}$ versus $L_{\rm 24\,\mu {\rm m}}$ for the obscured quasars of Reyes et al. (2008) and Greene et al. (2011). As our spectroscopic survey has not been flux calibrated, we have no direct measurements of $L_{\rm [{\rm O}\,{\scriptsize{ III}}]}$ for the GBs. However, we do know that the [O iii] equivalent widths are comparable to that of J2240, and that the spectra are generally similar. Assuming that [O iii] λ5008 contributes the same fraction to the total r-band flux as for J2240 (37%), we overplot the GBs in Figure 9. The 24 μm luminosities for GBs and quasars in the same redshift range (black dots) are indistinguishable. However, [O iii] luminosities of the GBs are 5–50 times higher than expected from their mid-IR emission. Since the latter mainly originates from the compact dusty torus, this indicates that the current AGN activity is too low to explain the NLR luminosity. The NLR may therefore reflect an earlier, more active state that subsided significantly in much less than a light-crossing time. We return to this light echo hypothesis below.

Zoom In Zoom Out Reset image size

6.1.1. AGN Character and Morphology

The galaxy-wide and ultraluminous NLR in J2240 is powered by an AGN, heating the medium- and low-ionization zones to 15, 000–18, 000 K and 13,000 K, respectively. The Disk has higher temperature than the Cloud. The flux of the auroral lines is sufficiently high to facilitate more reliable temperature maps with deeper data. Studying the profile of the continuum emission within ±5 kpc of the nucleus, we find a disk-like Sersic index of 0.85 and a disk scale length of 3 kpc, hence ruling out an elliptical host galaxy.

We find large amounts of dust as expected for type-2 AGNs. Dust is distributed within the NLR and also in a large foreground patch (Figure 4). Reddening variations are high, and small systematics due to the unknown X-ray spectrum may still be present. BPT line diagnostics (Figure 10) do not reveal significant star formation.

Zoom In Zoom Out Reset image size

J2240 is dynamically complex, showing seven distinct [O iii] peaks scattered over 30 kpc and 700 km s⁻¹ (Figure 11). A rotational component is well detected in the Disk. A typical ENLR stretches over 26 kpc northeast and 42 kpc southwest from the nucleus (Figure 8). It can be seen in all strong lines such as [O ii], [O iii], [Ne iii], [N ii], [S ii], and Hα, showing that J2240 is embedded in a rotating and quiescent, yet highly excited, bubble of gas.

Zoom In Zoom Out Reset image size

6.1.2. Shock Fronts and Jets

6.1.3. Globally Disturbed Gas Kinematics

6.1.4. GB Pilot Survey

6.2.1. Space Density and [O iii] Luminosities

6.2.2. Radio-loud Quasars and ULIRGs

6.2.3. Current Light Echo Samples

Can the extraordinary properties of J2240 be explained by mergers? We observe globally disturbed gas kinematics, and the highly ionized compact [O iii] sources can indicate multiplicity (Comerford et al. 2012). A tidally distorted neighboring galaxy (Figure 1) makes a multiple merger scenario plausible in which another, possibly gas-rich, galaxy is currently consumed by J2240. As Liu et al. (2012) demonstrate for wide separation binary AGNs, the SMBH accretion rate is increased in such double systems as the merger process funnels more material toward the centers. Log([O iii]) increases by 0.7 ± 0.1 when decreasing the separation from 100 to 5 kpc in these systems. Even closer pairs likely have correspondingly higher [O iii] luminosities.

Are we witnessing some violent process during the final stages of an AGN merger? SMBHs are common in the centers of massive galaxies; thus, galaxy mergers must also result in the coalescence of SMBHs. Accordingly, SMBH or AGN pairs should be common. However, with decreasing separation they are increasingly hard to identify. For example, numerous binary AGNs with separations of tens of kpc are known, yet with 3.6% their fraction among optically selected AGNs is already small (Liu et al. 2011). Shen et al. (2011) and Comerford et al. (2012) find kpc binary AGNs in galaxies with double-peaked [O iii] emission; however, that feature is more commonly caused by gas kinematics and cannot be used as a reliable indicator for AGN binarity (Shen et al. 2011; Fu et al. 2012). X-ray, infrared, or radio observations are needed to confirm such systems.

Only few binary AGNs with even smaller separations are known (Rodriguez et al. 2006; Fabbiano et al. 2011), including a system with subparsec scale and an orbital period of ∼100 years (Boroson & Lauer 2009). Eracleous et al. (2012) report several subparsec candidates, however, emphasizing that long-term monitoring is required to link observed line variability with orbital motions. From the X-ray perspective statistics are equally weak. For example, Teng et al. (2012) find that only 0%–8% of massive mergers actually harbor binary AGNs.

While the observational database is poor, simulations of the pre-coalescence state of SMBH mergers have been carried out (Hopkins & Quataert 2010; Khan et al. 2012; Van Wassenhove et al. 2012). During the orbital decay from kiloparsec to parsec scales no processes are found that explain an AGN flaring up for 10⁴–10⁵ years by 3–4 orders of magnitude. Neither can current accretion models explain a shutdown on similarly short timescales (Schawinski et al. 2010).

It is also unlikely that we are observing some effect or aftermath of the actual coalescence of two SMBHs. Tanaka et al. (2010) calculate the electromagnetic footprint of SMBH mergers. They find an increase of bolometric luminosity of about 10% per year over timescales of years or decades, together with an increase of X-ray hardness. This is, however, much less than the light-crossing time of a galaxy (10⁴–10⁵ years) needed to explain the size and luminosity of our NLRs.

The most prominent characteristics of the NLRs in our GBs are their large angular extent and high [O iii] luminosities. Certainly powerful AGNs must be responsible for this, but they are not evident from the SDSS imaging. Either they are deeply buried, or their activity has steeply declined over timescales much less than the light-crossing time of the NLR, in which case we are observing strong light echoes.

To this end we have to show that the current AGN activity is much lower than expected from the overall [O iii] luminosity, and that we are not just observing very obscured nuclei. The best way of doing this is to determine the current X-ray luminosities of GBs and compare them to their [O iii] luminosity. In case of a light echo, the X-ray output will not match the [O iii] luminosity. Using the best-fit relation between L_X and $L_{\rm [{\rm O}\,{\scriptsize{ III}}]}$ for type-2 quasars and Seyferts, we expect L_X ∼ 1 × 10⁴⁴ erg s⁻¹ in the 2–10 keV range (Jia et al. 2012; LaMassa et al. 2009). Note that, since ROSAT is only sensitive to soft energies, even the lowest plausible X-ray absorption is sufficient to account for the non-detection of J2240 by ROSAT. Therefore, we cannot constrain a possible quasar shutdown with the X-ray data currently available.

To overcome the lack of X-ray data, we use the mid-IR emission as a proxy for AGN activity, as it is unaffected by dust absorption and emanates from the immediate, parsec-scale AGN environment. Mid-IR emission is tightly correlated with the X-ray luminosity in AGNs (Asmus et al. 2011; Mason et al. 2012). We compare WISE 24 μm luminosities with [O iii] λ5008 luminosities for a large sample of type-2 quasars and GBs. Any change in AGN activity in GBs will need about a galaxy's light-crossing time before it is fully reflected in the NLRs' properties. We find the [O iii] luminosities of the GBs to be 5–50 times higher than expected from the mid-IR emission, strengthening the light echo scenario. If J2240 and the other GBs are indeed light echoes, they are spectacular examples of powerful QSOs currently shutting down.

Note, however, that the [O iii] luminosities of our GBs surpass the highest [O iii] luminosities of the reference type-2 sample at the same redshift (Figure 9). If the light echo interpretation is correct, then one expects quasars at similar redshifts with suitable mid-IR fluxes that have not shut down yet. Only quasars at z ≳ 0.4 match such high fluxes. We think that this is a selection effect, with GBs dropping out of the main SDSS spectroscopic target selection algorithm. The fact that our simple size and color selection of GBs yield about 95% spurious sources shows that the combination of broadband photometry and angular size is quite unusual. The SDSS algorithm designed to select sources for spectroscopic follow-up is optimized for a high success rate, allowing for only 5% unusual sources to be included (see, e.g., Reyes et al. 2008, and references therein). It is therefore not surprising that only one of our GBs has an SDSS-DR8 spectrum.

Considering the intrinsic luminosity that must be responsible for the observed optical NLR, the hard X-ray continuum of AGNs must be directly detectable, provided that the column density is not too great (N_H ≲ 10²³ cm⁻²). Even if the central engine is deeply obscured or truly hidden by Compton-thick absorption, the characteristic Fe Kα fluorescence line should be detectable. In this way a binary AGN with a few kpc separation can be directly confirmed through X-ray imaging or the shutdown timescale of the obscured quasars in GBs constrained. X-ray observations will also determine the slope of the ionizing spectrum, improving photoionization models of these NLRs.

Additional work going beyond our simple long-slit spectroscopic analysis has to be done. A realistic model of the NLR requires the use of photoionization codes and knowledge about the X-ray properties. The optical spectral analysis must also be extended to the full body of J2240. Such investigation may reveal ionizing sources missed by the long-slit observation and provide better extinction, temperature, and density maps. In addition, we can better constrain the dynamics, determine the gas mass, and investigate whether GBs have significantly higher gas masses than other obscured quasars showing massive outflows or large NLRs. To this end, GMOS IFU observations in good seeing conditions have been conducted of J2240 in 2012B at the Gemini Observatory. We have also applied for follow-up observations of several more GBs.