Do AGN triggering mechanisms vary with radio power? II. The importance of mergers as a function of radio power and optical luminosity

The radio-intermediate HERGs were selected from the catalogue of 18, 286 local radio galaxies produced by Best & Heckman (2012). These were selected as radio-intermediate based on their 1.4 GHz radio powers³³3In the range 22.5 $<$ log(L_1.4GHz) $<25.0$ W Hz^-1. Studies of local radio luminosity functions for AGN and star-forming galaxies show that the latter decline steeply above L${}_{\rm 1.4GHz}\sim 10^{23}$ W Hz^-1 (e.g. Sadler et al., 2002; Best & Heckman, 2012), but since all radio sources in the Best & Heckman (2012) catalogue were additionally identified as AGN using their optical properties, a lower luminosity density limit was considered for the selection., but were separated into two samples covering different ranges for the analysis in this paper. The optical spectroscopic data were extracted from the value-added catalogue maintained by the Max Planck Institute for Astrophysics and the Johns Hopkins University (the MPA-JHU value-added catalogue), which provides raw measurements and derived properties (e.g. stellar masses and star-formation rates) based on analysis of the SDSS DR7 spectra⁴⁴4Available at: https://wwwmpa.mpa-garching.mpg.de/SDSS/DR7/..

Our sample of high-radio-power AGN was selected from the Spinrad et al. (1985) catalogue of 3CR radio galaxies. The full sample comprises all 84 objects with redshifts in the range 0.05 $<z<$ 0.3. The lower redshift limit restricted the sample to galaxies with high radio powers (L${}_{\rm 1.4GHz}\gtrsim 10^{25}$ W Hz^-1), while the upper redshift limit ensured that the imaging observations had sufficient sensitivity and resolution for the detection of faint interaction signatures. However, the morphologies of 11 of the selected sources have been studied in detail in our previous work (Ramos Almeida et al., 2011; Ramos Almeida et al., 2012; Ramos Almeida et al., 2013), and were thus excluded from the current sample. One further object could not be used for the classification analysis, since a saturated bright star ruined the image appearance at its location (3C 452). The sample studied here comprises the remaining 72 3CR sources.

The spectroscopic properties of the vast majority of the objects in the sample have been determined by Buttiglione et al. (2009, 2010, 2011). The measured [OIII]$\lambda$5007 emission-line luminosities or upper limits⁵⁵5These are 3$\sigma$ upper limits determined from measurements of the noise level in the spectral regions near to the expected positions of the lines, assuming the instrumental resolution as their width (Buttiglione et al., 2009). from these studies are listed in Table 3 (corrected for Galactic reddening), and were only unavailable for three of the sources: the line was not detected in the spectra of 3C 52 and 3C 130; and 3C 405 (Cygnus A) is not included in the Buttiglione et al. samples. For 3C 52 and 3C 130, the upper limits provided for the H$\alpha$ emission-line fluxes were used to estimate [OIII]$\lambda$5007 luminosity upper limits, under the assumption that the noise levels were equivalent in the two spectral regions.

Buttiglione et al. (2009, 2010, 2011) also used the spectroscopic data to classify the galaxies as HERGs and LERGs, mostly using the excitation index (EI) scheme from Buttiglione et al. (2010). In the following cases, however, EI indices could not be calculated due to the lack of detections for key spectral lines: 3C 35; 3C 52; 3C 89; 3C 130; 3C 258; 3C 319; 3C 346; and 3C 438. For these objects, we use the [OIII]$\lambda$5007 equivalent width (EW) method adopted by Best & Heckman (2012) to obtain classifications, with HERGs having EW${}_{\rm[OIII]}>5$Å and LERGs having EW${}_{\rm[OIII]}<5$Å. Measurements of the continuum level were estimated from visual inspection of the spectra in Buttiglione et al. (2009) and Buttiglione et al. (2011). For the objects that lacked clear [OIII]$\lambda$5007 detections, the measured H$\alpha$ fluxes or upper limits were instead used as upper limits on the [OIII]$\lambda$5007 fluxes, for use with the equivalent width criterion (decisive for 3C 52 and 3C 89). When these values were unavailable or were not decisive for the classification, the spectral region near the expected position of the [OIII]$\lambda$5007 line was visually inspected, in order to place an upper limit on its line flux (necessary for 3C 130, 3C 319 and 3C 438).

For 3C 405, the [OIII]$\lambda$5007 emission-line luminosity was calculated from the flux measured by Osterbrock & Miller (1975) – corrected for Galactic reddening using the average $E(B-V)$ value of 0.48 determined by Taylor et al. (2003) from optical spectra of the outer regions of the galaxy – from which it is identified as a HERG using the EW criterion outlined above. The [OIII]$\lambda$5007 luminosity provided by Buttiglione et al. (2010) for 3C 321 is inconsistent with other measurements in the literature, and we instead adopted the value presented in Dicken et al. (2010).

Basic information on the properties of the 3CR sample is presented in Table 3. The 1.4 GHz radio powers were derived from the 178 MHz flux densities in Spinrad et al. (1985), assuming a standard spectral index of $\alpha=-0.7$. Stellar mass estimates were derived following the procedure outlined in §3.1. Note that we make no distinction between Type 1 and Type 2 AGN in the 3CR sample, since we find that the presence of a broad-line nucleus in the 11 Type 1 objects does not affect the ability to detect large-scale tidal features. The distributions of redshift, stellar mass, 1.4 GHz radio power, and [OIII]$\lambda$5007 emission-line luminosity are shown separately for the 3CR HERGs and LERGs in Figure 2.

Our Type 2 quasar sample consists of all 25 objects in the SDSS-selected sample of Reyes et al. (2008) with [OIII]$\lambda$5007 emission-line luminosities above $\rm L_{[OIII]}\geq 10^{8.5}$ L_⊙⁶⁶6Consistent with the Type 2 quasar definition used by Zakamska et al. (2003). ($\rm L_{[OIII]}\gtrsim 10^{35}$ W), redshifts $z<0.14$ and right ascension coordinates in the range 00h 30m $<\alpha<$ 12h 30m. These comprise part of a larger sample of 48 Type 2 quasars (the QSOFEED sample), which is described in detail in Ramos Almeida et al. (2021; in preparation).

As for the radio-intermediate HERG samples, the 1.4 GHz radio powers were derived from their NVSS flux densities whenever possible to avoid sensitivity losses due to resolution effects. However, FIRST flux densities were instead used for J0858+31 and J1218+08, where the FIRST images suggested that the NVSS flux densities would be strongly contaminated by other radio sources. FIRST flux densities were also used for J0818+36 and J1015+00, which were not detected by NVSS. J1036+01 and J1223+08 were not detected by either radio survey. For these objects, upper limits on the 1.4 GHz radio powers were estimated using the 1 mJy beam^-1 flux density selection limit for the FIRST catalogue (White et al., 1997). The sample is found to cover a broad range in radio power and includes 21 radio-intermediate sources ($10^{22.5}<$ L_1.4GHz $<10^{25}$ W Hz^-1) and two traditionally radio-loud sources (L_1.4GHz $>10^{25}$ W Hz^-1): J1137+61 and 3C 223⁷⁷73C 223 was observed and classified twice as part of both the Type 2 quasar sample and the 3CR sample. The object is considered as part of both samples throughout the analysis, although the classifications obtained from the Type 2 quasar image are used when they are combined, due to the higher image quality. The 1.4 GHz radio power and [OIII]$\lambda$5007 luminosity from Table 3 are also used for this object in Table 4, for consistency..

Calculation of the stellar mass estimates for the hosts of the Type 2 quasars was as performed for the 3CR sources (see §3.1). Basic information for the Type 2 quasar sample is presented in Table 4. The redshift, stellar mass, 1.4 GHz radio power and [OIII]$\lambda$5007 emission-line luminosity distributions for the targets in the sample are presented in Figure 3.

All deep optical imaging data were obtained using the Wide-Field Camera (WFC) attached to the 2.5m Isaac Newton Telescope (INT) at the Observatorio del Roque de los Muchachos, La Palma. The WFC consists of four thinned, anti-reflection-coated 2048 $\times$ 4100 pixel CCDs separated by gaps of $660-1098$ µm, pixel sizes of 0.333 arcsec pixel^-1, and a resulting large total field-of-view of 34 $\times$ 34 arcmin². The vast majority of the images were taken using the WFC Sloan r-band filter ($\rm{\lambda_{eff}}=6240$ Å, $\Delta\lambda=1347$ Å), chosen for consistency with the radio-intermediate HERG observations in Paper I and the observations of radio-loud galaxies in the 2 Jy sample, performed by Ramos Almeida et al. (2011). However, images were taken using the WFC Sloan i-band filter ($\rm{\lambda_{eff}}=7743$ Å, $\Delta\lambda=1519$ Å) for six of the 3CR sources (3C 52, 3C 61.1, 3C 63, 3C 130, 3C 405, 3C 410), in order to reduce the influence of the bright moonlight in the period in which these observations were performed. 3C 198 was observed using the WFC Harris R-band filter ($\rm{\lambda_{eff}}=6380$ Å, $\Delta\lambda=1520$ Å), which has a different response function to the WFC Sloan r-band filter but covers a broadly similar wavelength range.

The observations for the 3CR, RI-HERG high and Type 2 quasar samples were conducted in several separate runs between December 2012 and January 2020, on the dates listed in Tables 2, 3 and 4. Observations of the RI-HERG low sample were taken in March 2017, and are described in Paper I. Individual seeing full width at half maximum (FWHM) values for the observations were obtained from averaging the measured FWHM values of foreground stars in the final coadded images (following the reduction outlined in §2.5). These values are presented in Tables 2, 3 and 4. The seeing FWHM measurements range from 0.93 to 2.61 arcsec across the four samples, with a median of 1.39 arcsec. The RI-HERG low and RI-HERG high samples were observed with the best typical seeing, with medians of 1.31 and 1.36 arcsec and standard deviations of 0.15 and 0.26 arcsec, respectively. The 3CR and Type 2 quasar samples were typically observed in poorer seeing conditions, with medians of 1.51 and 1.40 arcsec and standard deviations of 0.39 and 0.30 arcsec, respectively. A summary of the seeing measurements for each sample is presented in Table 5.

The vast majority of the targets were observed using 4 $\times$ 700s exposures, yielding total exposure times of 2800s per target. These integration times were necessary for the detection of low-surface-brightness tidal features, but were divided into separate exposures to avoid target galaxy saturation in individual images. A square dithering pattern was employed for the observations (four pointing offsets of 30 arcsec) in order to cover the gaps in the images introduced by the spacings between the CCDs. Flat-field and image defect corrections were also improved by this process.

Several targets were, however, observed with different total integration times. 3C 405 and 3C 410 lie in crowded fields that contain a large number of stars due to their close proximity to the Galactic plane. The observations of these targets were divided into additional separate exposures that would yield similar total exposure times and reduce the number of saturated stars in individual images: 5 $\times$ 500s for 3C 405 and 7 $\times$ 400s for 3C 410. Only three exposures were obtained for 3C 63 (3 $\times$ 700s), J0818+36 (3 $\times$ 700s) and 3C 458 (3 $\times$ 600s), while five were used for 3C 236, 3C 326 and J0945+17 (each 5 $\times$ 700s), six for 3C 349 (6 $\times$ 700s) and seven for J1015+00 (7 $\times$ 700s). Four shorter exposures were taken for J1223+08 (4 $\times$ 400s), due to time constraints on the night on which this target was observed. In addition, image quality issues meant that only 3 $\times$ 700s individual frames could be used for some targets: J0836+05 and 3C 236 (satellite trails); J0858+31, J0945+17 and J1200+31 (bad CCD striping). Targets with three observations were still dithered sufficiently to overcome the issues related to the WFC CCD gaps.

All processing of the target images, from initial reduction to construction of the final mosaic images, was carried out using THELI (Schirmer, 2013; latest version available at https://github.com/schirmermischa/THELI). Bias and flat-field corrections for each observation were performed via the subtraction of master bias frames and division by the master flat-field frames, respectively. For the i-band observations, fringing effects were removed using a two-pass image background model. Brighter background features were removed on the first pass, where no detection and masking of objects was performed. Fainter background features were modelled in the second pass, performed after the detection and masking of all objects in the field above a certain threshold (signal-to-noise per pixel of 1.5 for objects with 5 or more connected pixels).

Astrometric solutions for the 3CR, RI-HERG high and Type 2 quasar calibrated images were calculated in THELI through cross-matching object catalogues produced for each image by SExtractor (Bertin & Arnouts, 1996) with the GAIA DR2 catalogue (Gaia Collaboration et al., 2018). The all-sky USNO-B1 catalogue (Monet et al., 2003) was used for the RI-HERG low sample, since these data were reduced with an earlier version of THELI (see Paper I). A sky model was produced and subtracted from all calibrated images by THELI to correct for remaining variations in the sky background level before final coaddition.

Photometric zero points were calculated in THELI through comparison of the derived instrumental magnitudes for stars in the final coadded image fields with their catalogued magnitudes in the Pan-STARRS1 (Chambers et al., 2016) First Data Release⁸⁸8Note that the analysis in Paper I considered catalogued SDSS r-band magnitudes when determining the zero points for the RI-HERG low sample. However, the Pan-STARRS1 magnitudes are used when comparing between the different samples in the current work. Both surveys use the AB magnitude system and the difference between the r-band magnitudes is minor, characterised by the following equation: $r_{\rm SDSS}=r_{\rm P1}-0.001+0.011(g-r)_{\rm P1}$ (Tonry et al., 2012).. This method has the key advantage of automatically correcting for photometric variability throughout the nights, since the calibration stars are observed at the same time and at the same position on the sky as the galaxy targets. The reliance on average zero points derived from standard star observations is also removed.

Measurements of the surface brightness depth of the observations were obtained using the same procedure as outlined in Paper I, which closely follows that used by Atkinson et al. (2013). These measurements were carried out after all reduction had been performed, including the flattening achieved through the subtraction of the sky background model. The total counts detected within 40 unique circular apertures of one arcsecond in radius were measured, placed in regions of the sky background with minimal or no influence from the haloes of bright objects in the field.

The standard deviations in the sky background level within the apertures ($\sigma_{\rm sky}$) were converted to apparent magnitudes, using the photometric zero points derived as outlined above, to provide final measurements of the surface brightness depth. The median $3\sigma_{\rm sky}$ surface brightness depths over all r-band observations of the targets is 27.0 mag arcsec^-2, with a standard deviation of 0.3 mag arcsec^-2. The value for the i-band observations of the 3CR targets is 25.7 mag arcsec^-2 ($3\sigma_{\rm sky}$), with a standard deviation of 0.2 mag arcsec^-2. The differences in individual and total exposure times for some target observations (see §2.4) did not significantly affect the surface brightness depths achieved. Limiting surface brightness measurements for individual targets in the RI-HERG low sample were presented in Paper I. Values for the individual target observations are not presented for the RI-HERG high, 3CR and Type 2 quasar samples. However, a summary of the average values for each sample is presented in Table 5.

Prior to identifying suitable matched control galaxies, estimates of the stellar masses were required for all of the targets to be matched. All objects in the RI-HERG low and RI-HERG high samples had existing stellar mass estimates in the MPA-JHU value added catalogue from which the control galaxies were also extracted⁹⁹9These were obtained from fitting of the stellar population synthesis models of Bruzual & Charlot (2003) to the SDSS broad-band ugriz photometry for the galaxies. The method is similar to that described by Kauffmann et al. (2003), who instead used spectroscopic features to characterise the fits. A comparative discussion of these methods is available at: https://wwwmpa.mpa-garching.mpg.de/SDSS/DR7/mass_comp.html., and so no additional calculations were required for these sources. However, MPA-JHU measurements were not available for all of the objects in the 3CR and Type 2 quasar samples. Initial stellar mass estimates for the galaxies in these samples were instead derived from their Two Micron All Sky Survey (2MASS; Skrutskie et al., 2006) $K_{s}$-band luminosities, using the colour-dependent mass-to-light ratio prescription of Bell et al. (2003) for this waveband. This method was used to derive stellar mass estimates for all of the objects in both the 3CR and Type 2 quasar samples (i.e. MPA-JHU values were not used for any of these galaxies), for consistency both within and between the two samples.

The Bell et al. (2003) mass-to-light ratio prescription for the 2MASS $K_{s}$-band has the following form (in solar units):

Where possible, the “total" $K_{s}$-band magnitudes¹⁰¹⁰10Derived by extrapolating the radial surface brightness profile measured within a 20 mag arcsec^-2 elliptical isophote to a radius of around four scale lengths (from Sérsic-like exponential fitting), in an attempt to account for the flux lost below the background noise. listed in the 2MASS Extended Source Catalog (Jarrett et al., 2000) were used to derive the luminosities for the calculations, in order to ensure that as much as possible of the galaxy flux was encapsulated (XSC magnitudes, hereafter). These values were available for 44 of the 73 3CR sources (60 per cent) and 19 of the 25 Type 2 quasars (76 per cent). For the remaining sources in the two samples, the magnitudes listed in the 2MASS Point Source Catalog (Skrutskie et al., 2006), derived from PSF profile fitting, were considered (hereafter PSC magnitudes). These were available for all 6 of the remaining Type 2 quasars and 26 of the 29 remaining 3CR galaxies (90 per cent).

Three of the 3CR sources – 3C 61.1, 3C 258, and 3C 458 – did not have either 2MASS XSC or PSC magnitudes available. Stellar mass estimates were not calculated in these cases, and, consequently, the control matching was not performed for these sources. However, these objects were still considered for the morphological classifications and are included in comparisons of the results obtained for the 3CR sample and their matched controls (§4). This was done under the assumption that the stellar masses for these sources, which represent a small minority of the full 3CR sample (4 per cent), are similar to those of the other 3CR objects.

The main disadvantage of using the PSC measurements was loss of sensitivity to emission from the extended regions of the target galaxies. In an attempt to account for this effect, a correction was derived from the average difference between the XSC and PSC magnitudes for the large number of galaxies in the MPA-JHU catalogue that had both measurements available. Considering all MPA-JHU galaxies with redshifts in the range covered by both the 3CR and Type 2 quasar samples ($0.05<z<0.3$), a median magnitude difference of $K_{s}^{\rm PSC}-K_{s}^{\rm XSC}=0.811$ was determined, with a standard deviation of 0.311. No significant evidence for a relationship between these differences and the measured PSC magnitude or redshift was found, as confirmed by Pearson correlation tests ($r=-0.121$ and $r=0.100$, respectively, with negligible p-values). As a result, a fixed value of 0.811 was subtracted from the PSC magnitudes to convert them to estimated XSC magnitudes in all cases, with the latter then being used in the subsequent calculations.

Following this, an extragalactic extinction correction was subtracted from the XSC or corrected PSC magnitudes, using the K-band values of Schlafly & Finkbeiner (2011)¹¹¹¹11Downloaded from the IRSA Galactic Dust Reddening and Extinction online service, available at: https://irsa.ipac.caltech.edu/applications/DUST/.. A cosmological k-correction was also applied, using the $k(z)\approx(2.1\pm 0.3)z$ formulation given in Bell et al. (2003), which is independent of galaxy spectral type. The corrected magnitudes were then converted to solar luminosities by assuming a solar 2MASS K_s-band absolute magnitude of 3.27 (Willmer, 2018), after which Equation 1 was applied to obtain the stellar mass estimates. An AGN contribution to the near-infrared light was not subtracted before calculating the stellar masses. However, we do not expect this to have a strong influence on our main results, since: (i) high-resolution near-IR imaging of nearby 3CR radio galaxies showed that the contribution from Type 2 AGN was typically small near $K_{s}$-band wavelengths ($<$20 per cent of the light in 0.9” diameter nuclear apertures; Ramírez et al., 2014a, b); (ii) Type 1 AGN objects only comprise 15 per cent of the 3CR sample (11/72) and 7 per cent of the full AGN sample (11/154), and Type 1 quasars only 7 per cent and 3 per cent, respectively.

As a final step, the same technique was used to determine $K_{s}$-band stellar mass estimates for all galaxies in the MPA-JHU catalogue with 2MASS XSC magnitudes available. This was done with the goal of obtaining a correction to align the $K_{s}$-band stellar mass estimates with the MPA-JHU values. Since individual SDSS $g-i$ colours were available for the MPA-JHU galaxies, a second prescription from Bell et al. (2003) that utilized these measurements was employed in this instance: $\mathrm{log}(M/L)_{K_{s}}=-0.211+0.137\times(g-i)$. The procedure was otherwise identical to that described above; this again included the subtraction of an additional 0.15 dex to convert to a Kroupa IMF. All XSC-matched galaxies with MPA-JHU stellar mass estimates in the range 10.7 $\leq$ $\rm log(M_{*}/M_{\odot}$) $\leq$ 12.0 were considered for the comparison, in approximate agreement with the range of $K_{s}$-band stellar mass estimates derived for the 3CR and Type 2 quasar samples – while some of the 3CR sources had $K_{s}$-band estimates larger than $\rm log(M_{*}/M_{\odot}$) = 12.0, this upper limit was chosen because of the large uncertainties determined for MPA-JHU estimates above this value.

The $K_{s}$-band and MPA-JHU stellar mass estimates for the $\sim$240,000 galaxies considered for the comparison are presented in Figure 4. The line of best fit displayed in the figure has the form $\rm log(M_{*}/M_{\odot})_{K_{s}}$ = 0.84 $\rm log(M_{*}/M_{\odot})_{\rm MPA-JHU}$ + 1.77, as derived from linear regression. The typical scatter around the relation is 0.01. No significant evidence that the difference between the stellar mass estimates varies with $g-i$ colour, redshift, 2MASS XSC $K_{s}$-band magnitude, or the values of either of the stellar mass estimates was found, based on Pearson correlation tests. Consequently, this relation was employed to convert $K_{s}$-band stellar mass estimates to corresponding MPA-JHU stellar mass estimates in all cases. The final MPA-JHU-equivalent stellar mass estimates for the 3CR and Type 2 quasars are presented in Tables 3 and 4, respectively.

The MPA-JHU value-added catalogue contains a large amount of raw and derived data for 927,552 galaxy spectra from SDSS DR7, including the spectroscopic redshifts and stellar mass estimates that were crucial for the control matching. Given the significant crossover between the fields covered by the INT/WFC images and the SDSS DR7 survey footprint, the catalogue therefore provided a suitable “pool" from which to select matched control galaxies for the active galaxies with matching observation properties (technique, image depth, observing conditions). Prior to performing the control matching, however, the following steps were taken to limit the control pool to galaxies with suitable properties.

1. 1.

   Coordinate restriction. Galaxies were required to lie within the regions of sky covered by the INT/WFC images of the active galaxies. This was the most restrictive constraint, with only 2,744 of the objects in the catalogue meeting this criterion (0.3 per cent).

2. 2.

   Removal of likely AGN. The MPA-JHU catalogue was constructed solely of objects that had been spectroscopically classified as galaxies in SDSS DR7, with the exception of some objects that had originally been targeted as galaxies but were later classified as quasars. However, the spectral types of both galaxies and quasars were also subclassified based on the properties of their [OIII]$\lambda$5007, H$\alpha$, H$\beta$, and [NII]$\lambda$6583 emission lines (if strongly detected). All sources that had been subclassified as either AGN or non-star-forming broad-line objects¹²¹²12Objects with lines detected at the 10 $\sigma$ level, with velocity dispersion $>200$ km s^-1 at the 5 $\sigma$ level. were removed from consideration. Star-forming or starburst galaxies were not removed. This left 2,615 galaxies in the control pool.

3. 3.

   Removal of remaining targets. The three galaxies from our target samples that had not been identified as AGN based on their SDSS spectra in step (ii) were removed, leaving 2,612 objects.

4. 4.

   Removal of duplicate objects. The MPA-JHU catalogue contains duplicate identifications for a large number of galaxies due to repeat SDSS spectral observations. Any remaining duplicates were taken out of the control pool at this point, which left 2,413 sources available for the matching.

All matched controls used for the analysis were selected from this final restricted pool of 2,413 galaxies. As mentioned, this matching was not performed for the three galaxies in the 3CR sample for which stellar mass estimates could not be obtained from 2MASS magnitudes (§3.1): 3C 61.1; 3C 258; and 3C 458. The Type 2 quasars were also not considered at this stage, although matching was performed for these objects after the classifications had been obtained (see below). A total of 127 active galaxies (58 radio-intermediate HERGs and 69 3CR sources) were therefore considered for control matching. Repeat selections of controls that matched multiple active galaxies were permitted throughout the matching process, in order to maximise the number of possible matches available for each target.

Redshift matches were determined using separate criteria for the radio-intermediate HERG and 3CR samples:

1. 1.

   $z_{\rm target}-0.01<z_{\rm control}<z_{\rm target}+0.01$, for matching to the RI-HERG low and RI-HERG high samples;

2. 2.

   $z_{\rm target}-0.02<z_{\rm control}<z_{\rm target}+0.02$, for matching to the 3CR sample.

1. 1.

   (log(M_∗/M_⊙) + $\sigma$)_control $>$ (log(M_∗/M_⊙) $-$ $\sigma$)_target ;

2. 2.

   (log(M_∗/M_⊙) $-$ $\sigma$)_control $<$ (log(M_∗/M_⊙) $+$ $\sigma$)_target ;

In addition to the three 3CR sources not considered for the matching (listed above), no matches were found for a further 10 of the 3CR sources using these criteria: 3C 130; 3C 234; 3C 323.1; 3C 346; 3C 371; 3C 382; 3C 405; 3C 410; 3C 430; and 3C 433. These sources had stellar mass estimates in the range $\rm 11.3\leq log(M_{*}/M_{\odot})\leq 12.7$, with a median of 11.8, and the lack of success in the matching was found to be driven by the large stellar masses of these objects. Overall, the procedure was thus successful in finding controls for 117 out of the 127 targets considered for the matching (92 per cent).

In order to reduce the total number of galaxies requiring classifications, only the five controls with the smallest differences between the target and control stellar mass estimates were considered for each active galaxy. This requirement had the additional advantage of counteracting the potential selection of controls with large uncertainties on their stellar mass estimates. In cases where the target had fewer than five matches, all of the available controls were considered: J1036+38 (RI-HERG low), with 4 matches; 3C 52, with 3 matches; 3C 236, with 2 matches; and 3C 132, 3C 388, 3C 438, J1630+12 (RI-HERG low), J0752+45, J1147+35, and J1436+05 (RI-HERG high), with one match each. A total of 551 control selections (388 unique galaxies, 163 repeats) were made for the 118 matched targets remaining at this point, an average of 4.7 per active galaxy.

19 of these control galaxies were found to be unsuitable for the classifications due to issues with the images (e.g. bad image regions, defects, crowding/source confusion). In these cases, images for the next closest matches in stellar mass were inspected until a suitable replacement control was found. All 19 controls were successfully replaced, but with 10 additional repeat selections (551 total selections with 173 repeats). This left a final sample of 378 unique control galaxies with images to be used for the classification analysis.

Due to their late inclusion in the project, control matching for the Type 2 quasar sample was performed after the interface classifications had been obtained. The controls for these targets were hence selected from the 378 control galaxies selected as matches to objects in the other active galaxy samples. The criteria used for the matching were identical to those used for the RI-HERG low and RI-HERG high samples: (i) $z_{\rm target}-0.01<z_{\rm control}<z_{\rm target}+0.01$, for the redshift matching; (ii) (log(M_∗/M_⊙) + $\sigma$)_control $>$ (log(M_∗/M_⊙) $-$ $\sigma$)_target and (log(M_∗/M_⊙) $-$ $\sigma$)_control $<$ (log(M_∗/M_⊙) $+$ $\sigma$)_target, for the stellar mass matching.

Using these criteria, it was found that 202 of the 378 controls were also matches to at least one of the galaxies in the Type 2 quasar sample. This included matches for each of the Type 2 quasar objects, with the number of matches for each individual target ranging from 4 to 47. As before, the five controls with the smallest differences between active galaxy and control galaxy stellar mass were selected for the control sample, and repeat selections were allowed. Only four matched controls were available for J1100+08, which were hence all included in the sample. A total of 124 control selections were therefore made for the 25 Type 2 quasar objects (91 unique control galaxies, 33 repeats).

Figure 5 shows the stellar mass and redshift distributions for each of the active galaxy samples and their respective matched control galaxy samples, which demonstrate the success of the matching. Two-sample Kolmogorov-Smirnov tests performed on these distributions provided no significant evidence for rejecting the null hypothesis that the targets and their matched controls are drawn from the same underlying distribution, in any of the cases. The results of these tests are presented in Table 6.

The online interface used to obtain morphological classifications for the project was made using the Zooniverse Project Builder platform at Zooniverse.org¹³¹³13Available at: https://www.zooniverse.org/lab., a citizen science web portal that stemmed from the initial Galaxy Zoo project by Lintott et al. (2008, 2011). Through this interface, eight researchers (all authors except PB, PDG) were blindly shown images of the active galaxies and control galaxies in a randomised order, and were asked to answer multiple choice questions concerning their optical morphologies.

No additional information on the nature of the galaxy to be classified was provided, in order to avoid introducing any biases related to the individual galaxy properties (e.g. active/non-active, target name, stellar mass, redshift, radio power, optical luminosity). However, two scale bars of 10 kpc in size were included on each image to assist with determination of multiple nuclei classifications (see below). The images were centred on the targets and fixed to be of 200 kpc $\times$ 200 kpc in size at the redshift of the galaxy in question, in all cases (as in Gordon et al., 2019). The images used for the classifications for all active galaxies (except those in the RI-HERG low sample) are available in the supplementary information.

One disadvantage of carrying out the classifications in this way was that the ability to fully manipulate the image contrast levels (as for those in Ramos Almeida et al., 2011; Ramos Almeida et al., 2012; Bessiere et al., 2012; Pierce et al., 2019; Ellison et al., 2019) was lost. The interface was set up with several features that partially accounted for this issue. Firstly, two postage stamp images with different contrast levels were displayed alongside each other for each galaxy: one of high contrast, for clearer identification of high-surface-brightness tidal features and the overall morphological types (spiral/disk, elliptical, etc.); and one of low contrast, for clearer identification of faint morphological structures. The two image contrast levels were chosen manually on a case-by-case basis, with consideration given to the appearance of both the target galaxy structures and the objects and/or image defects within the 200 kpc square surrounding region. In addition, classifiers could zoom in or out and pan around the image to look at specific regions in more detail. Rotation or inversion of the image was also possible.

Using the interface, classifiers were required to answer up to three multiple choice questions related to the morphological appearance of the subject galaxies. The first of these required the classifiers to answer the question “Does this galaxy show at least one clear interaction signature?" using one of the following options: (i) “Yes"; (ii) “No"; or (iii) “Not classifiable (e.g. image defect, bad region/spike from saturated star)". The last option was included for cases in which it was not possible to determine whether or not the galaxy was disturbed due to issues with the displayed image (e.g. because of presence of major image defects). For consistency with the previous study of the RI-HERG low sample (Paper I), dust lanes were included as one of the interaction signature classifications at this stage. However, dust lanes were not considered as a clear signature of a galaxy merger or interaction, and these classifications were not used for the analysis presented in §4.1 and §4.2.

Should the classifier have answered “Yes" to this first question, they were then asked to identify the type(s) of interaction signature that they had seen. To do this, they had to answer the question “What types of interaction signature are visible?" using the following options:

1. 1.

   Tail (T) – a narrow curvilinear feature with roughly radial orientation;

2. 2.

   Fan (F) – a structure similar to a tail, but that is shorter and broader;

3. 3.

   Shell (S) – a curving filamentary structure with a roughly tangential orientation relative to a radial vector from the main body of the galaxy;

4. 4.

   Bridge (B) – a feature that connects a radio galaxy with a companion;

5. 5.

   Amorphous halo (A) – the galaxy halo is misshapen in an unusual way in the image;

6. 6.

   Irregular (I) – the galaxy is clearly disturbed, but not in a way that fits any of the alternative classifications;

7. 7.

   Multiple nuclei (2N, 3N…) – two or more brightness peaks within a distance of 10 kpc;

8. 8.

   Dust lane (D) – a clear linear dark structure within the galaxy;

9. 9.

   Tidally interacting companion (TIC) – a companion galaxy shows clear morphological disturbance that is suggestive of a tidal interaction with the main target (e.g. with direction aligned towards/away from the central target).

Finally, the classifiers were required to answer the question “On first impression, what is the morphological type of the galaxy?" using one of the following responses: (i) “Spiral/disk"; (ii) “Elliptical"; (iii) “Lenticular"; (iv) “Merger (too disturbed to classify as above)"; or (v) “Unclassifiable (due to image defects, not merger)". Again, these options were chosen to be consistent with the host type classifications obtained from visual inspection of the RI-HERG low sources in Paper I. As for the first question, the last option was included for cases where the classifier thought that issues with the image quality meant that they could not provide an accurate classification.

The main goal of the project was to determine how the importance of galaxy mergers and interactions for triggering AGN varies with their radio powers and/or optical emission-line luminosities. This topic was addressed by the first question asked to the classifiers in the online interface: “Does this galaxy show at least one clear interaction signature?", with the possible answers of “Yes", “No" and “Not classifiable (e.g. image defect, bad region/spike from saturated star)" (see §3.3).

This question was answered by all eight classifiers for every galaxy in the sample. Only one of the listed responses could be selected when answering this question. A threshold of 5 out of 8 votes was chosen as the lower limit for accepting a certain classification in all cases (i.e. a simple majority). In this instance, the goal was simply to test whether or not the classifier believed that the galaxy had been disturbed by a merger or interaction. Here, cases in which 5 or more votes were recorded for “Yes" were taken to confirm that the galaxy was disturbed, and those where 5 or more votes were recorded for “No" were taken to indicate that the galaxy was not disturbed. Any other distribution of votes (including any number for “Not classifiable") was considered as an uncertain case.

The second question in the online interface asked the researchers to identify the specific types of interaction signature that they had seen in the galaxy images, with the goal of better characterising the types and stages of the mergers and interactions identified. Classifiers were asked to answer the question “What types of interaction signature are visible?" using one or more of the categories outlined in §3.3.

This question was only answered in cases where classifiers had already indicated that clear interaction signatures were visible in the galaxy images by responding “Yes" to the first question. Multiple responses could be selected, since several different types of interaction signature could be present at the same time. As a result of these two factors, the total number of votes recorded for each galaxy across the different interaction signature categories varied. An interaction signature classification was therefore required to meet two criteria in order to be accepted. Firstly, the threshold of 5 out of 8 votes must have been met for the first question, ensuring that only the galaxies accepted as being disturbed were included in the analysis. Secondly, the majority of the classifiers that had answered “Yes" to the first question must have voted for that particular interaction signature category, i.e. 3 out of 5, 4 out of 6, 4 out of 7, or 5 out of 8.

The main aim of this question was to determine whether the galaxy appeared to be in the early stages or late stages of a merger or interaction, i.e. prior to or following the coalescence of the galaxy nuclei (“pre-coalescence" or “coalescence/post-coalescence", respectively). For this purpose, the “Bridge (B)", “Tidally interacting companion (TIC)" and “Multiple nuclei (MN)" classifications were considered to indicate early-stage or pre-coalescence interactions, while the remainder were classified as late-stage or coalescence/post-coalescence interactions (late-stage or post-coalescence hereafter, for brevity). This approach is mostly consistent with that of Ramos Almeida et al. (2011); Ramos Almeida et al. (2012) and Bessiere et al. (2012), with the exception of the new “Tidally interacting companion (TIC)" category that was added for this work. On the occasions when both early-stage and late-stage interaction signature classifications were accepted, the interaction was considered to be of early-stage/pre-coalescence, for consistency with these previous studies.

Figure 10 shows the proportions of disturbed radio galaxies, Type 2 quasar objects, and control galaxies with secure interaction signature classifications that are suggestive of early-stage/pre-coalescence or late-stage/post-coalescence events based on the categorisation outlined above. The measured proportions for each sample are provided in Table 8.

Late-stage interactions appear to be slightly favoured for the disturbed galaxies in each radio galaxy sample, although the late-stage and early-stage proportions are consistent with being equal for the RI-HERG low sample. This preference is also found when considering the measured proportions for all radio galaxy samples combined: 31 ${}^{+7}_{-6}$ per cent and 69 ${}^{+6}_{-7}$ per cent for early- and late-stage interaction signatures, respectively. These results agree well with the those found for the powerful radio galaxies in the 2Jy sample, for which 35 $\pm$ 11 per cent and 65 $\pm$ 11 per cent of the disturbed objects with $z<0.3$ show a preference for pre- and post-coalescence interactions, respectively (Ramos Almeida et al., 2011).

On the other hand, the Type 2 quasar hosts show a preference for early-stage (69 ${}^{+8}_{-13}$ per cent) relative to late-stage (31 ${}^{+13}_{-8}$ per cent) interactions. These proportions are, however, consistent with those found for the disturbed galaxies in the control sample, for which 59 ${}^{+6}_{-7}$ per cent and 41 ${}^{+7}_{-6}$ per cent were found to have early-stage and late-stage interaction signatures, respectively. Two-proportion Z-tests suggest that the null hypothesis that the proportions of early-stage (or late-stage) interactions in the control sample and Type 2 quasar sample are the same can only be rejected at a confidence level of 0.7 $\sigma$. Similar fractions of pre-coalescence and post-coalescence interactions were also found for the disturbed hosts of Type 2 quasars with moderate redshifts ($0.3<z<0.41$) studied by Bessiere et al. (2012): 47 $\pm$ 13 per cent and 53 $\pm$ 13 per cent, respectively.

The measurements for the Type 2 quasar sample and control sample contrast significantly with those found for the radio galaxies, with the proportions of early-stage (or late-stage) interactions differing with those in the combined radio galaxy sample at confidence levels of $\sim$3 $\sigma$, in both cases. Note that all of the above results are preserved if the radio galaxies with [OIII]$\lambda$5007 luminosities above the quasar-like threshold ($\rm L_{[OIII]}\geq 10^{35}$ W) are not considered for the analysis – the early-stage and late-stage proportions become 30 ${}^{+8}_{-6}$ per cent and 68 ${}^{+6}_{-8}$ per cent, respectively.

Overall, the interaction signature classifications from the interface therefore tentatively suggest that radio AGN are preferentially triggered in the late stages of galaxy mergers and interactions, while Type 2 quasars are preferentially triggered in their early stages. Regardless of this, AGN host galaxies with both pre- and post-coalescence interaction signatures are found in each of the active galaxy samples. Therefore, if the galaxy mergers and interactions are responsible for triggering each of the types of AGN considered, this can occur at several different phases during these events.

The final question in the online interface asked the classifiers to indicate the overall morphological type of the galaxy in the image. Classifiers were required to answer the question “On first impression, what is the morphological type of the galaxy?" by selecting spiral/disk, elliptical, lenticular, merger (too disturbed to categorise), or indicating that it was unclassifiable due to image issues (§3.3). This was done to test if there was any variation in the longstanding relationship between radio AGN and early-type host galaxies with radio power, as suggested by the mixed population of late- and early-type galaxies found for the RI-HERG low sample in Paper I. All classifiers were required to answered this question, and only one of the available responses could be selected for each galaxy. As a result, a threshold of 5 out of 8 votes was again used for accepting a classification. If this was not met for any of the options (excluding the “Unclassifiable (due to image defects, not merger)" category), the galaxy was classed as having an uncertain host type.

The measured morphological type proportions for all samples are provided in Table 9. With the exception of the RI-HERG low sample (27 ${}^{+9}_{-6}$ per cent early type, 47 $\pm$ 9 per cent late type), it is found that the majority of the galaxies in each radio galaxy sample were classified as having early-type morphologies (elliptical and lenticular; dominated by the former), and small proportions were deemed to have late-type morphologies (spiral or disk-like). Their respective matched controls show relative excesses in late-type morphologies in all cases, with the same exception. The majority of the Type 2 quasar hosts are also classed as ellipticals, and a significant excess in the merger category relative to their matched controls is also found (3.1 $\sigma$), consistent with the high overall rate of disturbance determined for the sample (64 ${}^{+8}_{-10}$ per cent; §4.1).

In order to investigate the relationship between AGN host type and radio power, the proportions of active galaxies with morphologies classified as early-type and late-type were compared across the full range of radio powers covered. Here, all galaxies classified as either elliptical or lenticular were considered to be of early-type, and those classed as spirals or disks were considered to be of late-type, consistent with the analysis presented in Paper I.

Figure 11 shows the proportions of early-type and late-type galaxies in bins of 1.4 GHz radio power and [OIII]$\lambda$5007 emission-line luminosity for the full active galaxy sample and, separately, for only the HERGs and Type 2 quasars. A strong positive correlation is observed between the early-type proportions and medians of the 1.4 GHz radio power bins for both the full active galaxy sample ($r_{\rm full}=0.994$, $p_{\rm full}=0.006$) and the HERG and Type 2 quasar subset ($r_{\rm HERG}=0.998$, $p_{\rm HERG}=0.002$), according to Pearson correlation tests. This is coupled with a strong negative correlation for the proportion of late-type galaxies in both cases ($r_{\rm full}=-0.993$, $p_{\rm full}=0.007$ and $r_{\rm HERG}=-0.997$, $p_{\rm HERG}=0.003$). Pearson correlation tests also suggest strong correlations with [OIII]$\lambda$5007 luminosity, although these are found to be of lower significance than the relationships with 1.4 GHz radio power: $r_{\rm full}=0.868$, $p_{\rm full}=0.132$ and $r_{\rm HERG}=0.972$, $p_{\rm HERG}=0.028$, and $r_{\rm full}=-0.978$, $p_{\rm full}=0.022$ and $r_{\rm HERG}=-0.985$, $p_{\rm HERG}=0.003$ for the early- and late-type proportions, respectively. In contrast with the rates of disturbance, this suggests that the host types are more strongly linked with the radio power of the AGN than the optical emission-line luminosity.

While the control galaxies were matched to the targets in terms of their stellar masses and redshifts, matching of their morphological types was not performed. It is therefore important to check that the difference between the rates of disturbance for the active galaxies and control galaxies were not caused by the general preference for late-type morphologies exhibited by the latter objects (Table 9). Across the full sample of galaxies studied in the project, including all active galaxies and control galaxies, it is found that 30 ${}^{+3}_{-2}$ per cent of the early types and 26 $\pm$ 4 per cent of the late types were classified as disturbed. A two-proportion Z-test indicates that the null hypothesis that these two proportions are the same can be rejected at the 0.9 $\sigma$ level. There is thus little evidence for a significant difference between the disturbance rates of the early-type and late-type galaxies studied, and the general excess of late-type galaxies in the control samples relative to the active galaxy samples should therefore not affect the comparison of their disturbance fractions. Matching of the morphological types would hence have had little effect on the results presented in §4.1 (as found by Gordon et al., 2019).

Taken together, the results described in this section are consistent with the longstanding idea that the most powerful radio AGN are hosted by massive early-type galaxies. In addition, the fraction of early-type hosts is found to decrease strongly with decreasing radio power, while the proportion of late-type hosts shows the opposite trend. Since the sample is dominated by HERGs and Type 2 quasar objects, this supports the picture of a transition in the dominant host types for radiatively-efficient AGN from massive early-type galaxies at high radio powers to late-type hosts at low radio powers (i.e. like Seyfert galaxies), supporting the results from Paper I.

The online interface provided a means for obtaining morphological classifications of a large sample of active galaxies and matched control galaxies in a time-efficient manner. The standardised image format, set classification questions/categories and the randomisation of the galaxy image presentation also allowed the levels of individual classifier bias to be greatly reduced. Prior to comparing with results from the literature, however, we here consider how classifications obtained for the RI-HERG low sample using the online interface compare with those obtained from the more detailed visual inspection (i.e. with the ability to manipulate the image contrast and scale as required) in Paper I.

When considering the online interface classifications, it is found that 11 out of 30 (37 ${}^{+9}_{-8}$ per cent) of the galaxies in the RI-HERG low sample are classed as disturbed, compared with the 16 out of 30 (53 $\pm$ 9 per cent) found from the more detailed analysis. The results from the two methods therefore show good overall agreement, with the same classifications (either both disturbed or both not disturbed) being determined for 77 per cent of the RI-HERG low objects (23 out of 30). Most importantly, this includes the eight galaxies that in Paper I were identified as “highly disturbed" based on cursory visual inspection, all of which were classified as disturbed by either 7 (J1351+46, J1358+17) or all 8 (J0757+39, J0836+44, J0902+52, J1243+37, J1257+51, J1412+24) of the researchers in the online interface.

Of the remainder, 3 galaxies (10 per cent of the sample) were classified as having an uncertain level of disturbance (4 votes disturbed, 4 votes not disturbed) through the online interface but as disturbed in the more detailed analysis (J0827+12, J1601+43, J1609+13). A further 3 galaxies (10 per cent) had secure classifications as not disturbed from the interface (i.e. meeting the 5-vote threshold) that disagreed with the disturbed classifications from the more detailed analysis (J0725+43, J0911+45, J1236+40). As can be seen from the images presented in Paper I, all of these galaxies exhibit subtle morphological signatures of disturbance, and, because of the limited image manipulation afforded by the interface method, the lack of agreement between the two methods is therefore unsurprising. The final galaxy, J1324+27, was the only object classified as disturbed when using the online interface but as not disturbed in the detailed analysis. This galaxy was a borderline case, however, with 5 votes recorded for disturbed and 3 for not disturbed, and it is seen to exhibit an unusual spiral structure that could be interpreted either as a sign of disturbance or as that of an undisturbed late-type galaxy.

When dividing the sample into two halves by radio power, the two methods give the same proportion of disturbed galaxies for the half with the highest radio powers (10 out of 15; 67 ${}^{+10}_{-13}$ per cent), but a much reduced proportion is found for the lower-radio-power half from the interface classifications relative to the detailed inspection – 1 out of 15 (7 ${}^{+13}_{-2}$ per cent) and 6 out of 15 (40 ${}^{+11}_{-13}$ per cent), respectively. This suggests that the galaxies in the higher-radio-power half of the sample exhibit higher levels of disturbance than those in the lower-radio-power half, in support of the conclusions drawn based on the “highly disturbed" galaxies in the sample in Paper I. From the interface classifications, the two-proportion Z-test now indicates that the null hypothesis that the two proportions are equal can be rejected at a confidence level of 3.4 $\sigma$, compared to the value of 1.5 $\sigma$ obtained previously. Repeating this analysis in terms of [OIII]$\lambda$5007 luminosity, proportions of 7 out of 15 (47 $\pm$ 17 per cent) and 4 out of 15 (27 $\pm$ 11 per cent) are measured for the high-luminosity and low-luminosity halves of the sample, respectively, a difference only at the 1.1 $\sigma$ level.

These results hence appear to support the idea that the importance of mergers and interactions for triggering radio AGN is strongly dependent on radio power but more weakly dependent on optical emission-line luminosity, as suggested in Paper I. However, they are based on the measured proportions of disturbed galaxies, which, as shown in §4.1, increase strongly with stellar mass for both active and non-active galaxies. Within the RI-HERG low sample, Pearson correlation tests suggest that there is a moderate but significant correlation between stellar mass and 1.4 GHz radio power ($r=0.569$, $p=0.001$), but no significant correlation between stellar mass and [OIII]$\lambda$5007 luminosity ($r=0.095$, $p=0.617$). While the disturbed proportions for radio-intermediate active galaxies are also positively correlated with radio power in the current project, this is not observed when the matched control galaxy proportions are taken into consideration (i.e. the enhancement ratios in Figure 8). These factors therefore suggest that the apparent relationship with radio power from the Paper I results is in fact a consequence of an underlying trend with stellar mass in the general galaxy population, showing the importance of performing the control matching.

Considering the results for the host types, it is found that the classifications from the two methods agree for 18 of the 30 galaxies in the sample (60 per cent). However, 8 of the remainder were classed as uncertain (27 per cent), and so secure classifications (with $\geq$5 votes) from the interface only disagreed for 4 of the galaxies (13 per cent). Of this latter group, 3 out of 4 galaxies were classed as lenticular by one of the two methods and as late-type (spirals/disks) by the other, a difference that could be caused by the reduced ability to identify finer structures when using the interface. Interestingly, in the more detailed analysis, the host types for 5 out of the 8 uncertain cases from the interface classifications were deemed too disturbed to classify (the “Merger" class), which could explain this categorisation. Therefore, although the rate of complete consistency is lower than for the classifications of morphological disturbance, the general agreement still appears to be good between the two methods, given these factors.

Overall, this comparison suggests that the interface classifications provide good sensitivity to major levels of disturbance but not to minor levels, and the derived rates of disturbance should therefore be treated as lower limits. Furthermore, the lower sensitivity to more subtle morphological details could lead to the preferential classification of early-type hosts relative to late-types. However, any limitations introduced by the interface method should affect the active galaxies and their matched control samples equally, and so conclusions based on relative comparisons between the two should be secure in all cases. This again highlights the importance of the control matching process carried out for the current work.

The importance of galaxy mergers and interactions for triggering AGN has been the subject of much debate. The most widely accepted model suggests that radiatively-efficient AGN (e.g. HERGs/SLRGs/quasars) and radiatively-inefficient AGN (e.g. LERGs/WLRGs) differ in their dominant triggering and fuelling mechanisms (e.g. Heckman & Best, 2014; Yuan & Narayan, 2014). In this picture, radiatively-efficient AGN are fuelled by a high Eddington rate cold gas flow from a standard accretion disk (c.f. Shakura & Sunyaev, 1973), and hence a sufficient supply of such gas must be available to the central SMBH in order to initiate and sustain this type of nuclear activity. The strong inflows of gas caused by the tidal forces associated with galaxy mergers and interactions (e.g. Barnes & Hernquist, 1996; Gabor et al., 2016) therefore provide an attractive mechanism for triggering and fuelling the AGN in these objects.

Radiatively-inefficient AGN are then thought to be fuelled by an optically thin, geometrically thick accretion flow of hotter gas at low Eddington rates (c.f. Narayan & Yi, 1994, 1995; Narayan, 2005). At high radio powers, the favoured fuelling mechanisms in this case are often linked with the prevalent hot gas supply in the host galaxy haloes and the dense larger-scale environments in which these active galaxies typically lie (e.g. Baum et al., 1992; Best et al., 2005; Hardcastle et al., 2007; Gaspari et al., 2013), with mergers and interactions thus being relatively less important. The results obtained from the interface classification analysis are here discussed in this context.

Evidence from many past observational studies suggests that powerful radio AGN ($\rm L_{1.4GHz}\gtrsim 10^{25}$ W Hz^-1) are predominantly associated with massive elliptical galaxies (e.g. Matthews et al., 1964; Dunlop et al., 2003; Best et al., 2005). A minority of the objects in the high-flux-density selected 3CR and 2Jy samples do however exhibit disk-like morphologies upon cursory visual inspection, with these typically being found towards the lower end of the radio power range covered (Tadhunter, 2016). Some of these objects, in fact, have intermediate radio powers, which supports previous suggestions that late-type hosts could be more common in this range (e.g. Sadler et al., 2014). Furthermore, the results presented in Paper I showed that HERGs in the intermediate radio power range have a mixture of early-type and late-type morphologies. In combination, these results could suggest that the hosts of radiatively-efficient AGN move towards the predominantly late-type morphologies of Seyfert galaxies (e.g. Adams, 1977) at lower radio powers.

The general trends observed for the morphological type proportions in the current analysis provide strong support for this picture. Pearson correlation tests revealed evidence for strong positive correlations with radio power for the proportions of active galaxies classified as early-type in both the full active galaxy sample and the HERG and Type 2 quasar subset, which are coupled with strong decreases in the proportions classified as late-type with increasing radio power – each of these trends are clearly demonstrated in Figure 11. While similar trends are also seen with [OIII]$\lambda$5007 emission-line luminosity, these are found to be of lower significance, thus indicating that the host types are more strongly linked with AGN radio power. These results therefore suggest that there is a gradual transition in the dominant host types of radio AGN from early-type galaxies at high radio powers to late-type galaxies at lower radio powers, at least for radiatively-efficient objects. They are also consistent with the idea that secular triggering mechanisms related to galaxy disks (e.g. Hopkins & Quataert, 2010; Heckman & Best, 2014) become increasingly important towards lower radio powers, as suggested in Paper I, or lower total AGN powers (as in e.g. Treister et al., 2012).

Looking at the results for the current Type 2 quasar sample, it is found that the majority of host galaxies are classed as early-type (52 ${}^{+9}_{-10}$ per cent elliptical, 4 ${}^{+8}_{-1}$ per cent lenticular) and only 2 out of 25 (8 ${}^{+9}_{-3}$ per cent) and 3 out of 25 (12 ${}^{+10}_{-4}$ per cent) are classed as late-type or “merger", respectively; the remaining 24 ${}^{+10}_{-6}$ per cent have uncertain host types. A preference for early-type hosts has also been found in other imaging studies of low to intermediate redshift Type 2 quasars, from both visual inspection (Urbano-Mayorgas et al. 2019; but see Zhao et al. 2019) and detailed light profile fitting (Greene et al., 2009; Wylezalek et al., 2016). The preference for early-type hosts is enhanced when the radio-loud quasar-like AGN in the 3CR sample (with $\rm L_{[OIII]}\geq 10^{35}$ W) are considered together with the Type 2 quasars (73 ${}^{+6}_{-8}$ per cent), in agreement with previous results that suggest both radio-loud and radio-quiet quasars are predominantly hosted by elliptical galaxies (Dunlop et al., 2003). The host type classifications determined using the interface therefore appear to show good general agreement with those from these previous studies, and lend favour to the idea that powerful quasar-like activity in the local universe is largely associated with massive, early-type galaxies.