HST unveils a compact mildly relativistic broad-line region in the candidate true type 2 NGC 3147

ABSTRACT

NGC 3147 has been considered the best case of a true type 2 AGN: an unobscured AGN, based on the unabsorbed compact X-ray continuum, which lacks a broad-line region (BLR). However, the very low luminosity of NGC 3147 implies a compact BLR, which produces very broad lines, hard to detect against the dominant background host galaxy. Narrow (0.1 arcsec × 0.1 arcsec) slit HST spectroscopy allowed us to exclude most of the host galaxy light, and revealed an H α line with an extremely broad base (FWZI|${\sim }27\, 000$| km s⁻¹). The line profile shows a steep cut-off blue wing and an extended red wing, which match the signature of a mildly relativistic thin accretion disc line profile. It is indeed well fit with a nearly face on thin disc, at i ∼ 23°, with an inner radius at 77 ± 15 r_g, which matches the prediction of |$62^{+18}_{-14}$|r_g from the R_BLR–L^1/2 relation. This result questions the very existence of true type 2 AGN. Moreover, the detection of a thin disc, which extends below 100 r_g in an L/L_Edd ∼ 10⁻⁴ system, contradicts the current view of the accretion flow configuration at extremely low accretion rates.

galaxies: active, galaxies: individual: NGC3147, galaxies: Seyfert

1 INTRODUCTION

The polarized flux (scattered light) spectra of many type 2 Seyfert galaxies show a big blue bump and broad emission lines, showing the presence of a hidden Seyfert 1-like nucleus. Because the per cent polarization is very high (⁠|${\gtrsim }15{{\ \rm per\ cent}}$|⁠), and approximately perpendicular to the radio structure axis,¹ it was inferred that photons can only escape from the nuclear regions by travelling along the axis and then scattering into the line of sight; escape along the equatorial latitudes is blocked by a ‘dusty torus’ or similar structure. This leads to the Unified Model (UM; Antonucci 1993, 2012), the simplest version of which states that the spectral types differ solely due to inclination of the axis with respect to the line of sight. The nucleus is seen directly when the inclination is small.

However, a significant fraction of Seyfert 2 galaxies do not show a hidden broad-line region (BLR), even in high-quality spectro-polarimetric data (Tran 2001, 2003, but see the smaller fraction in Ramos Almeida et al. 2016). A possible explanation is that any mirror reflecting the broad lines has low scattering efficiency or is obscured (Miller & Goodrich 1990; Heisler, Lumsden & Bailey 1997). The lack of polarized broad lines appears also to be associated with a stronger dilution from the host galaxy or from a circumnuclear starburst, making their detection more challenging (Alexander 2001; Gu, Maiolino & Dultzin-Hacyan 2001). Nevertheless, the X-ray emission often provides clear evidence for a buried type 1 AGN, as indicated by the detection of a highly absorbed, or reflection-dominated, X-ray continuum (e.g. Marinucci et al. 2012).

Some low-luminosity Seyfert 2s show instead an unobscured X-ray continuum, strongly suggesting a direct view of the innermost regions close to the massive black hole (BH) (as much as 30 per cent of the entire type 2 population, according to Panessa & Bassani 2002; Netzer 2015, but see the much lower fraction suggested by Malizia et al. 2012). Furthermore, in some of these objects the optical and/or the X-ray continuum varies rapidly, clearly excluding that we are instead observing large-scale reflection of the inner nucleus (e.g. Hawkins 2004; Bianchi et al. 2012, 2017). Are these true type 2 AGN, intrinsically lacking a BLR? Does the AGN unification break down in low-luminosity AGN?

A plausible model for the origin of the BLR gas is a disc wind, driven by radiation pressure on UV resonance lines (Shlosman, Vitello & Shaviv 1985; Murray et al. 1995; Proga & Kallman 2004), on dust grains (Czerny & Hryniewicz 2011; Baskin & Laor 2018), or magnetically driven (Emmering, Blandford & Shlosman 1992; Lovelace, Romanova & Biermann 1998). This wind is expected to disappear for low luminosities, or accretion rates (Nicastro 2002; Laor 2003; Elitzur & Shlosman 2006; Elitzur & Ho 2009). Indeed, the true type 2 best candidates have low Eddington ratios: NGC 3147 (log L_bol/L_Edd ≃ −4: Bianchi et al. 2008), Q2131-427 (log L_bol/L_Edd ≃ −2.6: Panessa et al. 2009), NGC 3660 (log L_bol/L_Edd ≃ −2: Bianchi et al. 2012). These are unambiguous candidates since the lack of the optical broad lines was found together with simultaneous X-ray observations which certify an absorption-free line of sight to the nucleus.

Have the above theoretical predictions, that the BLR disappears at very low luminosities/accretion rates, been vindicated by these objects? One needs to be careful as the optical spectra of low L_bol/L_Edd AGN are heavily dominated by the host galaxy emission. It is extremely difficult to set tight upper limits on the flux of any broad lines and their equivalent widths relative to any nuclear continuum in the heavily starlight-dominated ground-based spectra (see Antonucci 2012, for a comprehensive criticism on true type 2s). Furthermore, if the BLR does exist, it must be very compact, since the size of the BLR scales as |$L_\mathrm{bol}^{1/2}$| for L_bol in the range of 10⁴⁰–10⁴⁷ erg s⁻¹ (e.g. Kaspi et al. 2007; Bentz et al. 2013). This relation is naturally explained by assuming that the outer boundary of the BLR is determined by the dust sublimation radius, which is set by the continuum luminosity (Laor & Draine 1993; Netzer & Laor 1993). The compact size leads to a high Keplerian velocity in the BLR, which can make the lines extremely broad, and thus even harder to detect. Indeed, narrow slit HST observations revealed very broad H α lines in a number of low-luminosity AGN (Ho et al. 2000; Shields et al. 2000; Barth et al. 2001; Balmaverde & Capetti 2014).

NGC 3147 (z = 0.009346: Epinat et al. 2008) is the best of the three true type 2 Seyfert candidates with simultaneous optical and X-ray observations (Bianchi et al. 2012), being the one with the most extensive X-ray and optical coverage. Its X-ray spectrum shows very tight constraints on the column density of any obscuring gas along its line of sight, and a modest fraction of reprocessing components (iron K α line and Compton hump) even at energies larger than 10 keV, thus excluding with high confidence that we are observing the reflected component of a Compton-thick source (Panessa & Bassani 2002; Bianchi et al. 2008; Matt et al. 2012; Bianchi et al. 2017). Significant X-ray variability on time-scales as short as weeks (Bianchi et al. 2017), adds further support to a direct view of a compact emitting region. Bianchi et al. (2008) analysed the (galaxy-subtracted) optical spectrum taken at the Observatorio de Sierra Nevada simultaneously with the XMM–Newton observation, reporting an upper limit to a broad (full width at half-maximum, FWHM, fixed to 2000 km s⁻¹) component of the H α line corresponding to a luminosity of 2 × 10³⁸ erg s⁻¹. This is significantly lower than the value of ≃10⁴⁰ erg s⁻¹, expected from the X-ray luminosity, based on the relation followed by type 1 AGN (Stern & Laor 2012a). This is also true for the |$\mathrm{L_{H\beta }^{broad}/\nu L_\nu ^{1.4\, GHz}}$| and |$\mathrm{L_{H\alpha }^{broad}/L_\nu ^{10\, \mu m}}$| ratios (Shi et al. 2010).

Is this enough to classify it as a true type 2 AGN? The above-mentioned limits on any broad component are valid only if the width and profile of the line are not extreme. The only way to definitely exclude the predicted BLR emission is narrow slit HST spectroscopy, which allows for a clear detection of the low-luminosity AGN emission. Here, we present an HST observation of NGC 3147, designed to exclude the bulk of the host emission, thanks to its narrow slit (0.1 arcsec).

2 OBSERVATION AND DATA REDUCTION

The target was observed with the Space Telescope Imaging Spectrograph (STIS) on board the Hubble Space Telescope on 2018 May 17. The observations were performed using the CCD with the 52 arcsec × 0.1 arcsec slit and the G750L low-resolution grating. The pointing aimed at the host galaxy photocenter, and automatically re-centred to the location of the brightest visible source in the acquisition image. The image shows that the AGN is clearly visible as a point source at the centre of the field of view. Four more spectra in contiguous regions were also taken, in case the automatic centering procedure did not accurately fit the AGN in the 0.1 arcsec slit. Each spectrum was taken with a total exposure time of 410s, split into two CR-SPLIT exposures of equal duration to remove cosmic ray events. The automatic re-centring procedure was able to accurately place the target at the centre of the slit.

We retrieved the raw data and the calibration files from the Mikulski Archive for Space Telescopes (MAST). We processed the data to correct for charge transfer inefficiency (CTI) using the python script provided by the STScI STIS team, based on the model by Anderson & Bedin (2010). In addition to providing a pixel-based CTI correction for both the dark frame and each of the raw images, the script also performs the basic calibration steps (bias and dark subtraction and flat field correction) and combines the two CR-SPLIT images into a single 2D spectral image. The nuclear spectrum was rectified and extracted from a 2 pixel aperture (0.1 arcsec) using the pyiraf task splot. We used a synthetic PSF obtained with tiny tim to derive the appropriate aperture correction from 2 pixel to the standard 7 pixel aperture.

The spectrum was analysed with xspec 12.10.1 (Arnaud 1996), and the fitting model convolved with a Gaussian smoothing with σ = 3.14Å, to match the instrumental resolution.² In the following, (statistical only) errors and upper limits correspond to the 90 per cent confidence level for one interesting parameter, apart from the plots, where 1σ error bars are shown. The adopted cosmological parameters are H₀ = 70 km s⁻¹ Mpc⁻¹, |$\Omega _\Lambda = 0.73$|⁠, and Ω_m = 0.27.

3 SPECTRAL ANALYSIS

The HST STIS G750L spectrum of NGC 3147 is shown in Fig. 1. Along with the lines from the narrow line region (NLR), it is evident the presence of a very broad emission line centred at the H α wavelength, extending from 6450 to 7050 Å. This corresponds to a full width at zero intensity (FWZI) of 27 000 km s⁻¹. We therefore focus our analysis in the 6000–7200 Å range, and started by modelling the narrow emission lines from the NLR. In our fits, we assumed that the wavelengths of the components of the [O i], [N ii], and [S ii] doublets have the expected ratios λ₂/λ₁ = 6363.81/6300.32, 6583.39/6548.06, and λ₂/λ₁ = 6730.78/6716.42, respectively, and share the same width. We also assumed that |${\mathrm{F}([{\rm N\, {\small {II}}]}} \lambda 6583.39)/\mathrm{F}([{\rm N\, {\small {II}}}] \lambda 6548.06) = 3$| and |$\mathrm{F}([{\rm O\, {\small I}}] \lambda 6300.32)/\mathrm{F}([{\rm O\, {\small I}}] \lambda 6363.81) = 3$|⁠, as required by the ratio of the respective Einstein coefficients, and the width of the H α is the same as that of the [N ii] doublet. The properties of the narrow emission lines modelled in the 6000–7200 Å range are listed in Table 1.

Figure 1.

A remarkably prominent and very broad H α line profile (in red) is revealed in the HST STIS G750L spectrum of NGC 3147, carrying the signature of a relativistic thin Keplerian disc, with an inner emitting radius of only 77 r_g. Narrow lines from the NLR are in green, the continuum in gray. The fit residuals are shown in the bottom panel.

Table 1.

Emission line properties in the 6000–7200 Å range. Laboratory wavelengths (Å) in air are from Bowen (1960). FWHMs are in km s⁻¹, fluxes in 10⁻¹⁵ er g cm⁻² s⁻¹.

Line	λl	FWHM	Flux
Narrow Emission Lines
|$\mathrm{[{O\, \small {I}}]}$|	6300.32	|$1030^{+200}_{-160}$|	2.7 ± 0.4
|$\mathrm{[{O\, \small {I}}]}$|	6363.81	|$1030^{+200}_{-160}$|	|$0.9^{+0.3}_{-0.6}$|
|$\mathrm{[{N\, \small {II}}]}$|	6548.06	680 ± 30	|$3.90^{+0.13}_{-0.09}$|
H α	6562.79	680 ± 30	4.1 ± 0.3
|$\mathrm{[{N\, \small {II}}]}$|	6583.39	680 ± 30	|$11.7^{+0.4}_{-0.3}$|
|$\mathrm{[{S\, \small {II}}]}$|	6716.42	|$520^{+150}_{-110}$|	1.5 ± 0.3
|$\mathrm{[{S\, \small {II}}]}$|	6730.78	|$520^{+150}_{-110}$|	2.2 ± 0.3
Broad H α disc line profile
i	rin	rout	α	Flux
|$23^{+2 \circ }_{-1}$|	77 ± 15 rg	570 ± 60 rg	1.9 ± 0.3	67 ± 2

The residuals left after the inclusion of the narrow emission lines from the NLR show a very broad and asymmetric profile. In particular, it is characterized by a steep cut-off blue wing and an extended red wing, which are the signature of a mildly relativistic thin accretion disc line profile (e.g. Laor 1991). We therefore adopted the Kerrdisk model (Brenneman & Reynolds 2006), which reproduces the line profile from accretion disc systems around BHs. This model accurately reproduces the overall profile (Fig. 1), tightly constraining the emission parameters: the inclination angle is |$i=23^{+2 \circ}_{-1} $|⁠, the inner and outer radii are r_in = 77 ± 15 and r_out = 570 ± 60 r_g, respectively, where r_g = GM/c² is the gravitational radius. The line emissivity index (see Brenneman & Reynolds 2006) is constrained at α = 1.9 ± 0.3, while the BH spin, as expected since r_in is much larger than the innermost stable orbit, is completely unconstrained. The rest-frame line wavelength of the disc line component appears to be blueshifted with respect to the narrow H α line³ by |$850^{+100}_{-200}$| km s⁻¹. It is interesting to note that all the derived parameters are linked to clear properties of the observed profile. The inclination angle is mostly set by the cut-off at lower wavelengths, while the inner radius is determined by the extension of the red wing tail. The distance between the two peaks, instead, is constrained by the outer radius. The flux of the broad component of the H α emission line is 6.7 ± 0.2 × 10⁻¹⁴ erg cm ⁻² s⁻¹, i.e. 16 times larger than the narrow H α line. This corresponds to a line luminosity of 1.30 ± 0.04 × 10⁴⁰ erg s⁻¹, and an equivalent width of 449 ± 13 Å, close to the mean value of 570 Å observed in low-z type 1 AGN (e.g. Stern & Laor 2012a).

An important question is whether the broad component of the H α line was hidden by the host contamination in previous, ground-based spectra or is a new feature just arisen at the time of the HST observation. Fig. 2 shows on the same linear scale with the new HST spectrum three ground-based spectra analysed in literature: the 1986 Palomar spectrum (Ho, Filippenko & Sargent 1995), the 2003 Keck spectrum (Tran, Lyke & Mader 2011) and the 2006 Observatorio de Sierra Nevada spectrum (Bianchi et al. 2008). The very small aperture of our HST observation allowed us to exclude the starlight continuum and part of the NLR, which dominate the other spectra. Even the less contaminated (i.e. the highest S/N and smallest aperture) Keck spectrum is compatible with the presence of the broad H α line observed by HST. We have therefore no evidence in favour of a variability of this component, which is likely hidden by the host galaxy contamination in earlier observations. This scenario is consistent with the X-ray flux state of NGC 3147 in a TOO Swift observation we triggered shortly after the HST observation (July 27th): the 2–10 keV flux is (1.5 ± 0.5) × 10⁻¹² er g cm⁻² s⁻¹, i.e. the average flux of the source (Bianchi et al. 2017), thus excluding this is a ‘changing-look’ AGN.

Figure 2.

Comparison of the available ground-based optical spectra of NGC 3147 with the HST STIS G750L spectrum analysed here. The starlight in the 0.1 arcsec HST spectrum is almost completely eliminated, revealing the AGN continuum, which is a factor of 12 smaller than the host light in the 1 arcsec Keck spectrum, and 40–50 smaller than the 2 arcsec Palomar and OSN spectra.

4 DISCUSSION

Based on the analysis presented in this paper, it is unavoidable to conclude that NGC 3147 is not a true type 2 AGN, but a low accretion rate type 1 AGN with a large BH mass. The BLR is clearly present in this source, and the previous lack of its detection was caused by the combined effect of a very broadened emission profile and an intrinsic weakness with respect to the host galaxy, as expected for a low accreting high-mass object. Since NGC 3147 likely represented the best candidate for this class of sources, this result questions the very existence of true type 2 AGN.

Most models for the origin of the BLR predict its disappearance at low accretion rates, although for different reasons: the disc wind cannot form if the accretion rate is lower than the minimum value for which a standard Shakura & Sunyaev (1973) disc is stable (Nicastro 2002); the photoionized gas in the inflated disc remains dusty, making the BLR emission inefficient (Baskin & Laor 2018). Inserting the stellar velocity dispersion σ_* = 233 ± 8 km s⁻¹ measured by van den Bosch et al. (2015) in the M–σ_* relation for ellipticals (Gültekin et al. 2009), we get a BH mass log M_BH/M_⊙ = 8.49 ± 0.11 for NGC 3147 (the reported dispersion of the relation is 0.31 dex). Adopting the most recent X-ray bolometric corrections (e.g. Duras et al., in prep), a bolometric luminosity of 4.0 ± 1.2 × 10⁴² erg s⁻¹ can be derived from the average 2–10 keV X-ray luminosity (3.3 ± 1.0 × 10⁴¹ erg s⁻¹; Bianchi et al. 2008; Matt et al. 2012; Bianchi et al. 2017). Therefore, the Eddington ratio of NGC 3147 is L_bol/L_Edd = 1.0 ± 0.4 × 10⁻⁴. This accretion rate is much lower than the minimum value required by Nicastro (2002) to launch the BLR wind (1–4 × 10⁻³, slightly depending on the BH mass and the accretion properties), and of the same order of the critical accretion rate below which dust effectively suppresses BLR emission, but only if low metallicity is allowed (Baskin & Laor 2018).

The luminosity of the broad H α line in the HST spectrum (⁠|$\log L_\mathrm{b\mathrm{H}\alpha}/\mathrm{erg\,s^{-1} }=40.11$|⁠) is consistent with the predicted values of 40.5 ± 0.5 and 40.6 ± 0.5 from the L_X versus |$L_{\rm b\mathrm{H}\alpha }$| and |$L_{[{\rm O}\, {\small {III}]}}$| versus |$L_{\rm b\mathrm{H}\alpha }$| relations in type 1 AGN (Stern & Laor 2012a,b), given the above reported X-ray luminosity and the (extinction corrected) [O iii] luminosity of 1.6 × 10⁴⁰ erg s⁻¹(Bianchi et al. 2008). NGC 3147 appears like a standard type 1 AGN at the very low end of luminosity ranges.

From the measured broad H α luminosity, we can also predict the BLR radius (as mapped by H α) in this source. By combining the |$L_\mathrm{b\mathrm{H}\alpha }$| versus L₅₁₀₀ and the R_BLR versus L₅₁₀₀ relations reported in Greene & Ho (2005), we get R_BLR = 2.8 ± 0.2 × 10¹⁵ cm (the uncertainty being dominated by the uncertainties in the parameters of the relations). For the BH mass reported above, this translates into |$R_\mathrm{BLR}=62^{+18}_{-14}$|r_g (the uncertainty being dominated by the uncertainty on the BH mass). This is in very good agreement with the r_in = 77 ± 15 r_g found with the disc line profile fit.

The disc line profile fit yields the emitting radii in units of r_g, independently of the luminosity and the BH mass. We can therefore do the opposite reasoning, and estimate the BH mass in NGC 3147 on the basis of the inner radius and the luminosity of the H α line. By re-arranging and re-normalizing the above-mentioned formulae, we get

\begin{eqnarray*} \log \frac{M_{\rm BH}}{\mathrm{M}_\odot } =\left(10.22\pm 0.05\right) + \left(0.553\pm 0.017\right) \log L_\mathrm{b\mathrm{H}\alpha }^{40} - \log \frac{{r}_{\rm in}}{{r}_{\mathrm g}} \end{eqnarray*}

(1)

where |$L_\mathrm{b\mathrm{H}\alpha }^{40}$| is the Hα luminosity in units of 10⁴⁰ erg s⁻¹. Inserting the values measured in our HST spectrum, we have log M_BH/M_⊙ = 8.40 ± 0.05, in agreement with the 8.49 ± 0.11 obtained via the stellar velocity dispersion.

H α disc line profiles have been previously observed in a handful of objects (e.g. Eracleous & Halpern 1994, 2003; Strateva et al. 2003; Tran 2010; Storchi-Bergmann et al. 2017), but with significantly larger inner radii at least in the range |$10^2-10^3\, {r}_\mathrm{g}$|⁠, the smallest being 98 r_g in SDSS J0942 + 0900 (Wang et al. 2005). The larger inner radii are indeed expected, since these objects have generally larger accretion rates than NGC 3147 and the results are commonly interpreted as a torus like structure, possibly associated with a geometrically thick low accretion rate configuration. Here, we find that the BLR forms a thin disc, extending down well below 100 r_g. Standard BLR photoionization models require column densities N_H > 10²³ cm⁻². The disc gas mass within R_BLR is then |$\pi R_\mathrm{BLR}^2 m_{\rm p} N_\mathrm{H}\gt 2\times 10^{-3}\mathrm{M}_\odot$|⁠, which is >3 times the accretion required per year to sustain the observed L_bol. Therefore, this compact thin disc is likely the inner part of the reservoir of gas which feeds the AGN, i.e. the accretion disc. The presence of a thin accretion disc at L/L_Edd ∼ 10⁻⁴ is likely to change our current view of the accretion flow, being in contradiction with the standard paradigm, that at low accretion rates, the accretion configuration becomes optically thin and quasi-spherical (Blandford & Begelman 1999).

Even broader H α profiles are expected to be observed when an object like NGC 3147 is seen at a higher inclination (see Fig. 3). One should therefore look carefully for weak quasi continuum features which can extend over a region >1000 Å wide. Despite their weakness, these features should be clearly detected once the host contamination is fully excluded, and the net AGN emission is observed at a reasonable S/N ratio. Optical detections of relativistic line profiles may provide a new tool to explore the innermost disc structure: by exploring L/L_Edd ∼ 10⁻⁵ and M_BH ∼ 3 × 10⁹ M_⊙ (typical in LINERs and FR Is) we may be able to go down by an additional factor of 10 (in terms of r_g) in the inner radius, and see the emission from the innermost disc at unprecedented resolution and S/N, as the optical photons detection rate is 10⁵ higher than in the X-rays.

Figure 3.

Predicted disc line profiles with the same total flux and disc parameters as NGC 3147, but with different inclination angles. Even broader profiles are expected to be observed in higher inclination low-luminosity AGN. The expected profiles for different inner radii are shown in the inset.

ACKNOWLEDGEMENTS

We thank the anonymous referee and H. Tran for the 2003 Keck spectrum of NGC 3147. SB acknowledges financial support from the Italian Space Agency under grant ASI-INAF 2017-14-H.O. FJC acknowledges financial support through grant AYA2015-64346-C2-1P (MINECO/FEDER). The data described here may be obtained from the MAST archive at http://dx.doi.org/10.17909/t9-7vzv-2j06.

Footnotes

1

That excludes interstellar dichroic absorption, and applies to the per cent polarization of the scattered light alone (Antonucci 2002).

2

http://www.stsci.edu/hst/stis/performance/spectral_resolution

3

Keeping the two wavelengths linked yields a statistically worse fit, but the other disc line parameters are only slightly affected, showing that the overall modelling of the profile is robust.