NGC 3147: a prototypical low-luminosity active galactic nucleus with double-peaked optical and ultraviolet lines

We start our analysis by comparing the new G750L spectrum with the one taken in Cycle 25 and published in Bianchi et al. (2019). As evident from Fig. 1, the H$\alpha$ profile changed dramatically between the two observations. Most of the variation can be accounted for a lower flux of the relativistic component by a factor of $3.2\pm 0.2$ in the new data, with all the other parameters being the same. This allows for the detection of a further broad Gaussian component of the H$\alpha$ emission line, with a full width at half-maximum (FWHM) $\sim 5000$ km s^-1. This component is consistent with being present also in the old data, with a flux $1.8\pm 0.3$ times lower, but was not required by the fit because of the dominance of the relativistic line. All the narrow emission lines from the narrow-line region (NLR) are consistent with being constant, together with the continuum level. If a cross-normalization factor is included between the two data sets, an $\sim 5\%$ excess is found in the new data, well within any calibration uncertainties.

This best=fitting model gives a very good representation of both data sets, with the most significant residuals being bluewards the [S ii] doublet. We will investigate this issue with more detail below, with the high-resolution G750M spectrum. All the other parameters of the kerrdisk model (see Bianchi et al., 2019, for details) do not change significantly if allowed to be different between the two cycles, with the significant exception of the inner radius of the disc, which changed from $77\pm 15$ to $195\pm 15$ r_g (see Table 1). The underlying continuum is well modelled by a power law in the 5450-8000 Å band²²2The remaining 8000-9999 Å band is not used for the fit, since the wavelength-dependent corrections used for the non-standard extraction slit height (see Sect. 2) are significantly different with respect to lower wavelengths., with an index (in $F_{\nu}$) of $-1.40\pm 0.04$, and a flux density of $F_{\lambda}=1.44\pm 0.01\times 10^{-16}$ erg cm^-2 s^-1 Å^-1 at 7000 ų³3As reported in Sect. 2 and Appendix A, the red part of the G750L spectrum is likely contaminated by significant starlight, so that the intrinsic nuclear flux in this band is lower and the slope flatter than what observed.. Galactic reddening is taken into account with an extinction of $E(B-V)=0.022$, as derived from NED based on the prescription of Schlafly & Finkbeiner (2011) and variation with wavelength as specified by Cardelli et al. (1989).

The overall model used for G750L is qualitatively a good representation for the higher resolution G750M spectrum, when a cross-normalization is taken into account to correct for the extended nature of the source. A very good fit is found just by allowing the parameters to readjust (Fig. 2). While the parameters for the relativistic line remain unchanged, the FWHM of the further broad Gaussian component for the H$\alpha$ is smaller and much better constrained ($3760\pm 80$ km s¹). Note that its velocity shift is comparable to that of the narrow emission lines. All the best-fitting parameters relative to the narrow emission lines are reported in Table 1.

In particular, for the doublets we calculated the ratio of the two components. While we recover the expected ratio ($R=0.34\pm 0.01$) for the [N ii] doublet, the ratio for the [O i] is fixed to the expected value of 3. On the other hand, the observed [S ii] doublet ratio ($R=0.92\pm 0.03$) allows us to estimate the density of the gas to $n_{e}=680\pm 80$ cm^-3 (Sanders et al., 2016).

As already noted for the G750L, positive residuals are present bluewards the [S ii] doublet, approximately at the location of the peak of the red wing of the relativistic H$\alpha$ line. Since these residuals occur close to the disc line peak velocity, they may indicate that the thin flat Keplerian disc model, with a sharp inner boundary, may be an oversimplification of the inner disc boundary.

The HST STIS G140L spectrum is shown in Fig. 3. It is dominated by prominent broad Ly$\alpha$ and C iv emission lines, with complex profiles. Several foreground absorption lines due to our own interstellar medium (ISM) are apparent. They are modelled as Gaussian lines when appropriate in the following fits. Once corrected for the Galactic reddening, the underlying continuum is well modelled by a power law in the whole 1150-1680 Å band, with an index (in $F_{\nu}$) of $-2.0^{+0.3}_{-0.4}$, and a flux density of $F_{\lambda}=1.90\pm 0.05\times 10^{-16}$ erg cm^-2 s^-1 Å^-1 at 1300 Å.

The Ly$\alpha$ profile can be modelled with three components, as for the H$\alpha$ line: a disc line (Kerrdisk model), a broad and a narrow Gaussian line (see Fig. 4 and Table 2). The disc line profile is characterized by inner and outer radii r${}_{in}=300\pm 5$ and r${}_{out}=1020\pm 50$ r_g, and an inclination $i=27\pm 1\degr$. The inclination angle is in very good agreement with that found for H$\alpha$, and the locations of the blue and red peaks are again just where predicted by a simple disc emission. Both the inner and the outer radii are larger than that required by the H$\alpha$ profile.

The overall Ly$\alpha$ emission line is completed by a broad (FWHM$\sim 4500$ km s^-1) and a narrower (FWHM$\sim 1400$ km s^-1) Gaussian component. The fit also includes a broad N v doublet (F($\lambda 1238.82$)/F($\lambda 1242.80$) fixed to 2). Indeed, the emission feature at around 1270Å (rest frame), rather than being a strongly redshifted Si ii blend, lies exactly at the expected red wing of a further disc line component from the N v doublet, whose parameters are very similar to that of Ly$\alpha$, in particular the inclination angle and the inner radius (see Table 2).

The C iv emission line profile is also well modelled by a Kerrdisk disc line (see Fig. 5). Again, it is important to note how the inclination of $i=27\pm 1\degr$ is also recovered in this case. The inner and outer radii of the disc are r${}_{in}=270\pm 15$ and r${}_{out}=550\pm 70$ r_g, as reported in Table 2. A further broad Gaussian line with an FWHM=$5450\pm 250$ km s^-1 is also present, while a narrow component is not required by the fit. All the components are fitted with a doublet, fixing $F(\lambda 1548.19)/F(\lambda 1550.77)=2$. Although the overall profile is well reproduced by this model, residuals in excess are still present at the red wing of the line profile. These can be modelled by adding a further Gaussian line at $\sim 1586$ Å (rest frame) with an FWHM$\sim 2500$ km s^-1 and a flux of $\sim 5.9\times 10^{-15}$ erg cm^-2 s^-1. If interpreted as C iv emission, the corresponding velocity shift would be $7100\pm 200$ km s^-1.

Finally, a disc line profile is a good modellization of the Si iv doublet at 1393.76 and 1402.77 Å (see Fig. 6). Also in this case, the overall profile also includes a broad Gaussian component. A possible contribution from the O iv] blend cannot be excluded, although not strictly required. The N iv] line at 1486.5 Å is also very broad, but the lower statistics and the vicinity to the huge C iv line do not allow for a more complex profile decomposition.

The disc line best-fitting profiles for H$\alpha$ and Ly$\alpha$ are directly compared in velocity space in Fig. 7, renormalized with respect to their maximum. From this plot, it is evident that the blueward extension of all the lines is in good agreement, as expected, being it mostly set by the inclination angle of the disc. Moreover, the red wing is also very similar for Ly$\alpha$ and H$\alpha$ taken at the same epoch, while note the much more extended red wing in the H$\alpha$ profile as observed in the previous epoch, due to the smaller inner radius.

It is also interesting to look at the broad Gaussian components that are needed in all the modellizations of these emission lines. In Fig 7 (right-hand panel), the residuals to the data after taking into account the disc line profiles and the narrow Gaussian components for the NLR are shown. The broad Gaussian component is apparent in all three emission lines. While this component appears quite consistent between H$\alpha$ and Ly$\alpha$, the residuals for C iv are broader. However, it should be noted that the latter is a doublet, and this would also explain the apparent asymmetry of these residuals. Both H$\alpha$ and C iv show a further component at $\sim 7000$ km s^-1, as already noted in the previous fits. This component is not present in Ly$\alpha$.

Thanks to the very small slit adopted for the HST observation, we can build the first nuclear SED of NGC 3147. The continua observed in the HST G750L and G140L are well described by power laws with indices (in $F_{\nu}$) of $-1.40\pm 0.04$ and $-2.0^{+0.3}_{-0.4}$, respectively, as derived in Sect. 3.1 and 3.2 (see the left-hand panel of Fig. 8). On the other hand, the simultaneous Swift X-ray observation is well described by a power law with the same photon index measured in the 2015 NuSTAR observation ($\Gamma=1.69\pm 0.05$, Bianchi et al., 2017). The 2-10 keV flux is $2.5\pm 0.2\times 10^{-12}$ erg cm^-2 s^-1, i.e. at the same level of the NuSTAR observation and just above the historical average flux of the source (Bianchi et al., 2017).

The optical-to-X-ray SED is shown in the right-hand panel of Fig. 8. It is evident that it is very hard, with a peak (in $\nu L_{\nu}$) in the hard X-rays. By combining the Swift 2 keV observed density flux with the 2500Å density flux as extrapolated from the G140L spectrum, we obtain $\alpha_{ox}=-0.82\pm 0.16$ (see the right-hand panel of Fig. 8). This value is significantly flatter than the one derived by Bianchi et al. (2008) from the XMM-Newton Optical Monitor, which is dominated by the emission from the host galaxy, and is therefore not representative of the nuclear emission. For the same reason, the new Swift Ultra-violet Optical Telescope (UVOT) data are not displayed here, and cannot be used to build the nuclear SED.

Finally, we note that in the literature there are subarcsec near-infrared (NIR) and mid-infrared (MIR) fluxes of NGC 3147 taken with Gemini/Michelle in 2010 at 11.6 and 12.5 $\mu$m (Asmus et al., 2014; Mason et al., 2015), which would show a rise of the SED with respect to the optical emission. The resulting IR-to-optical power law would have an index (in $F_{\nu}$) of -1.66. However, the IR emission in NGC 3147 appears to be host-dominated also at subarcsec resolution, hence we decided not to include these data in the SED.

The presence of a thin accretion disc in the low-accreting AGN NGC 3147 was inferred by Bianchi et al. (2019) with the discovery of a broad H$\alpha$ emission line with a disc line profile. An H$\alpha$ line with the same profile is confirmed in the HST spectra presented in this work, taken 2 yr after, suggesting that the disc is persistent. Moreover, broad disc line profiles are observed for other prominent emission lines in UV, namely Ly$\alpha$, C iv, N v and Si iv. The thin disc interpretation is strongly reinforced by the fact that all these profiles present blue and red peak locations well matched with simple disc line predictions. Moreover, they all agree with a disc inclination between 23°and 27°.

The H$\alpha$ disc line component was much stronger in Cycle 25, although this does not appear to be accompanied by a significant flux variation of the source. The disc line variability may be due to a changing structure in the BLR, expected in a dynamical time-scale that could be of the order of $\sim 1$ month in such a compact BLR. A sort of ‘accretion event’ may have occurred in Cycle 25, with gas swirling in at $\sim 100$ r_g, and now we only see residual material further out. This scenario is reinforced by the larger inner radius measured in the new observation, indeed suggesting that the disc receded from $\sim 80$ to $\sim 200$ r_g. The disc line profiles for the UV lines are in good agreement with H$\alpha$ (notably for the inclination angle), and are all well modelled by inner radii between 200 and 300 r_g.

All the broad-line profiles require a Gaussian component in addition to the disc line, with a very similar FWHM around $4000-5000$ km s^-1. The gas that produces these lines should be at around $4000-5000$ r_g (neglecting any virial factor), i.e. $\sim 0.06-0.07$ pc for $\log\mathrm{M_{BH}/M_{\odot}}=8.49$. Since the sublimation radius for NGC 3147 should be located not farther than $\sim 0.03$ pc (Barvainis, 1987, but considering the whole bolometric luminosity here), this means that dust is expected to be present, and this broad component cannot be associated with a classical BLR. However, it may be possible that the very hard SED that characterizes this source is more efficient at destroying dust at farther distances.

The stellar velocity dispersion in NGC 3147 is $\sigma_{*}=233$ km s^-1 (Van Den Bosch et al., 2015), so we expect the FWHM of lines produced outside the sphere of influence of the central BH to be of the order of $2.35\sigma_{*}\sim 550$ km s^-1. This is indeed what is measured for [N ii], H$\alpha$ and [S ii]. On the other hand, [O i] is significantly broader. This is qualitatively in agreement with a correlation with the critical density, as already shown in Filippenko & Halpern (1984), since [O i] has the largest critical density and is therefore expected to be produced in a denser gas, closer to the BH. Indeed, a natural consequence of a radiation pressure compressed (RPC) NLR is that the lines with the highest critical density should be produced inside the BH sphere of influence, when the luminosity is low enough with respect to the BH mass (Stern et al., 2014). Using $\log\mathrm{M_{BH}/M_{\odot}}=8.49$, $L_{\mathrm{ion}}=L_{\mathrm{bol}}=2.9\times 10^{42}$ erg s^-1 (see Sect. 4.3), and a critical density $1.1\times 10^{6}$ cm^-3 (Martín et al., 2015), we can estimate FWHM${}_{[\text{O\,{i}}]}=1400$ km s^-1 (eq. 28 in Stern et al., 2014). Note that this is the maximum expected FWHM if all [O i] emission is produced at the critical density. In our spectrum, it appears that the bulk of the line is produced at farther distances, where the emissivity is still very high (see Fig. 6 in Stern et al., 2014). However, [O i] can potentially have a broader base.

It is interesting to check if the narrow lines’ ratios derived from our small-aperture HST spectrum are consistent with a Seyfert-like nucleus. From Table 1 we get $\log$ [N ii]/H$\alpha=0.528\pm 0.012$, which is in rough agreement with what found in larger aperture, ground-based spectra (0.3-0.43: Ho et al., 1997; Bianchi et al., 2012), and well inside the AGN region in standard BPT diagrams (e.g. Kewley et al., 2006). On the other hand, we have $\log$ [S ii]/H$\alpha=0.20\pm 0.02$ and $\log$ [O i]/H$\alpha=-0.10\pm 0.02$, which are larger than 0.06 (but it is 0.2 in Bianchi et al., 2012) and -0.82, respectively, as found in Ho et al. (1997). These values are more suggestive of a LINER classification with respect to a Seyfert, unless $\log$[O iii]/H$\beta$ is significantly larger than 0.79-0.85, as measured in Ho et al. (1997) and Bianchi et al. (2012). Unfortunately, our HST spectra do not cover the [O iii] and H$\beta$ band, so we cannot give the final word on the matter. However, we note that the electron density derived from the [S ii] doublet, i.e. $n_{e}=680\pm 80$ cm^-3, is significantly larger than what is found on average in LINERs, and more in agreement with the typical densities in Seyferts (e.g. Ho, 2008).

We have reconstructed the nuclear SED of NGC 3147 in Sect. 3.3. Notably, the optical, UV and X-ray data are taken simultaneously, and the HST data have subarcsec resolution, minimizing the host contamination⁴⁴4Although the Swift and NuSTAR have much larger spatial resolution, Chandra data and rapid variability conclusively showed that the X-ray emission in NGC 3147 is dominated by the nucleus (Bianchi et al., 2017). The resulting SED, as parametrized with power laws in different bands, is shown again in Fig. 9: a spectral index of -1.40 for the optical band, a steepening to -2.0 in the UV and a rise to -0.69 up to the hard X-rays. An integral of this SED from 1 $\mu$m to 200 keV gives a bolometric luminosity of $2.9\times 10^{42}$ erg s^-1, where we have extrapolated the X-ray power law introducing an exponential cutoff at 150 keV, as found on average in local Seyfert galaxies (e.g. Tortosa et al., 2018; Middei et al., 2019). With this revised bolometric luminosity, the Eddington ratio of NGC 3147 is $L_{\mathrm{bol}}/L_{\mathrm{Edd}}\sim 7.5\times 10^{-5}$, at the lower end of previous estimated ranges (e.g. Bianchi et al., 2019). Considering the 2-10 keV luminosity observed in the Swift observation ($5.6\times 10^{41}$ erg s^-1), this gives an X-ray bolometric correction of $\sim 5$, which is much lower than standard values, even at the low-luminosity end of the AGN population (see e.g. Duras et al., 2017). Most of the bolometric luminosity in NGC 3147 is released in the X-rays. Therefore, although the presence of double-peaked optical and UV lines shows that there is a thin disc in this source, this appears to be a rather ‘passive’ one, and most of the radiation is instead produced in some hot medium.

Indeed, the most striking characteristic of this SED is the lack of a UV bump. This is evident when compared to the standard SED of AGN (as in Richards et al., 2006), shown in green in the same figure, after renormalizing it at 2 keV⁵⁵5Above $\log\nu=15.5$ this SED is extrapolated in order to have an $\alpha_{ox}=-1.5$, a $-1.5$ index to model the ‘soft excess’ between 0.3 and 2 keV, then an index of $-0.8$ is adopted above 2 keV.. Indeed, in NGC 3147 the minimum energy is released in the UV, contrary to what found for a normal AGN. This reflects in a much flatter $\alpha_{ox}$ (-0.82) with respect to standard AGN (roughly between -1.2 and -1.6: e.g. Grupe et al., 2010; Vagnetti et al., 2010).

Such flat $\alpha_{ox}$ are instead more common in LLAGN (e.g. Ho et al., 1996; Ho et al., 2000; Sabra et al., 2003; Maoz, 2007; Ho, 2008; Fernández-Ontiveros et al., 2022). Indeed, the $\alpha_{ox}$ of luminous AGN is known to follow a tight inverse relation with luminosity (see e.g. Just et al., 2007). Extrapolating the relation found in Just et al. (2007) to the low UV luminosity of NGC 3147 (L${}_{2500\text{\AA}}\sim 7.6\times 10^{25}$ erg s^-1 Hz^-1 as extrapolated from our HST UV spectrum), we get $\alpha_{ox}=-0.92$, which is very close to what we actually observe in this source. It is quite impressive that the $\alpha_{ox}$ luminosity relation appears to extend down to these bolometric luminosities. This suggests that there is no clear transition from luminous AGN to LLAGN, but just a progressive systematic trend in the relative contribution of the X-rays to the UV bump, with the latter becoming weaker and colder at decreasing luminosity.

We can derive the frequency where a standard Shakura & Sunyaev (1973) disc should peak in the case of NGC 3147. From the bolometric luminosity derived above, we can estimate the mass accretion rate assuming efficiencies $\eta=0.057$ and $0.321$, i.e. as expected if the inner radius goes down to the last stable orbit, for $a=0$ and $a=0.998$, respectively (Novikov & Thorne, 1973). We get $\dot{\mathrm{M}}=1.1\times 10^{-3}$ and $2.0\times 10^{-4}$ M_⊙/yr⁶⁶6We drop the factor 2, valid only in the Newtonian approximation (e.g. Davis & Laor, 2011), and assume an inclination $\cos{i}=0.8$.. These values can be used to derive the maximum disc temperature, as in Laor & Davis (2011), and the relative frequency peak of the emission (in $\nu L_{\nu}$): $4.1\times 10^{14}$ and $6.4\times 10^{14}$ Hz, again for a non-rotating and a maximally rotating BH. This frequency range ($0.47-0.74\mu$m) corresponds approximately to the band covered by our HST optical spectra. Indeed, the flat (-1) slope we observe in the optical may be the accretion disc emission close to the peak. It is then possible that the observed steepening at higher frequencies to the UV is indeed the SED on the blue side of the accretion disc peak. In normal AGN this blue side is shortwards of $\sim 1000$Å, and is also a steep power law (e.g. Laor & Davis, 2014).

Therefore, the steep -2 UV power law we observe in NGC 3147 may be Comptonization of the cold disc by a warm corona, in analogy to one of the standard interpretations of the so-called ‘soft excess’ observed in the X-ray emission of more radiatively efficient AGN (e.g. Petrucci et al., 2018; Petrucci et al., 2020). In NGC 3147, this component is expected to shift to lower frequencies, together with the accretion disc peak, and indeed there is no clear evidence for an X-ray soft excess in NGC 3147 (Bianchi et al., 2012, 2017), although there are some hints of a different variability between the soft and the hard X-ray bands (Matt et al., 2012).

We searched in the literature for other LLAGN with double-peaked optical/UV emission lines and subarcsec HST UV observations allowing for a direct estimate of the intrinsic continuum slope in this band. We found two sources satisfying these criteria: M81 and NGC4579. Their SEDs were reconstructed using the optical/UV slopes derived from HST in Ho et al. (1996) (M81), Barth et al. (1996) and Molina et al. (2018) (NGC4579), the X-ray slopes observed by XMM-Newton for M81 (Cappi et al., 2006) and NGC4579 (Bianchi et al., 2009), and the $a_{ox}$ reported by Ho et al. (1996) (M81) and Barth et al. (1996) (NGC4579). The resulting SEDs are shown in Fig. 9. The overall shape of the SEDs of LLAGN is very similar to that of NGC 3147: a dominant X-ray emission and a weak UV emission characterized by a steep power law.

The inclusion of IR data to the SEDs of LLAGN generally leads to a rise of the emission up to around 10 $\mu$m (see e.g. Ho, 2008; Fernández-Ontiveros et al., 2022). While we are not confident that the existing subarcsec MIR and NIR photometry for NGC 3147 is dominated by the AGN, such a rise is observed also in NGC 3147, as briefly mentioned in Sect. 3.3. The IR emission in luminous AGN is dominated by dust reprocessing, generally associated to the torus. Indeed, NGC 3147 has a strong iron K$\alpha$ line in its X-ray spectrum (Bianchi et al., 2012, 2017; Matt et al., 2012), which suggests a significant contribution from Compton-thick neutral circumnuclear matter, as routinely found for more luminous AGN (e.g. Bianchi et al., 2009). However, hot dust is mostly heated by the optical-UV emission, so it is unlikely that the IR luminosity peak is much stronger than the UV peak. On the other hand, the IR emission in NGC 3147 can be significantly contributed by the accretion disc emission below the peak. This could be tested with future observations with NIR adaptive optics, as done in other very low luminosity AGN (Dumont et al., 2020).