SkyMapper colours of Seyfert galaxies and Changing-Look AGN

When SkyMapper was planned, the Sloan Digital Sky Survey had already revolutionised galaxy surveys by observing much of the Northern sky. As SkyMapper filters were designed, their scientific niche was seen to be improved studies of Milky Way stars.

The biggest difference between SkyMapper and SDSS filters is on the blue side of 400 nm (see Table 1 and Fig. 1): where SDSS has a $u$ band with $(\lambda_{\rm cen}/{\rm FWHM})=(358{\rm nm}/55{\rm nm})$ SkyMapper has a violet $v$ band (384/28) and a more ultraviolet $u$ band (349/42). This choice provides photometric sensitivity to the surface gravity of stars as the strength of the Hydrogen Balmer break affects the $u-v$ colour, as well as to metallicity as metal lines affect the $v-g$ colour. By exploiting this information, SkyMapper has facilitated the discovery of a large sample of extremely metal-poor stars (Da Costa et al., 2019) including the most chemically pristine star currently known (Keller et al., 2014). In this paper we demonstrate the use of this filter pair in near-field extragalactic work.

Filters that are useful for characterising stars should also help with characterising stellar populations in galaxies at redshifts close to zero. SEDs of more distant galaxies are of course redshifted out of sync with the passbands. We first note that the SkyMapper colour index $u-v$ takes on its reddest value for main-sequence A-type stars. We then expect the $u-v$ colour of a stellar population to depend on the fraction of A-type stars in the mix. After a quenching event, the young O- and B-type stars die away and leave the blue side of the galaxy SED dominated by the bright $(u-v)$-red A-stars. Conversely, in the event of a starburst a rapid increase in the fraction of O- and B-stars above typical levels makes the $u-v$ index bluer than usual. A continuously star-forming and steadily evolving population, however, has a constant O-to-A star ratio and will thus show a constant $u-v$ index, irrespective of the detailed history of an older underlying population.

In the following, we explore in detail how the $u-v$ colour responds to temporal changes in the star-formation rate of a galaxy and to AGN contributions. Colour indices made from $griz$ bands are expected to behave similar to the well-studied SDSS colours.

The aim of this step is to see what part of optical colour space may be expected to be occupied by galaxies considering a broad range of evolutionary options. It is not to interpret star-formation histories in the face of degeneracies in optical photometry. We target galaxies in the local ($z\la 0.1$) Universe, which may appear young or old, but often include a substantial amount of old or intermediate-age stars in either case. For stellar populations undergoing only secular evolution we assume no particular formation redshift and consider a large range of possibilities in the star-formation history. Mature galaxies might still undergo quenching or bursting events long after most of the stars in a galaxy have formed. To capture this, we model stellar populations with exponential star-formation histories proceeding for 10 Gyr, before a quenching or burst event happens. The choice of 10 Gyr is degenerate with other parameters of the evolution and hence not critical for predicting galaxy SEDs in a small number of bands. Immediately following the event, the SEDs of the populations change rapidly so we need to calculate their SED evolution with fine time resolution.

For population synthesis we use the Stochastically Lighting Up Galaxies (SLUG) package by Krumholz et al. (2015). In the long term, we intend to study the influence of stochasticity when considering small parts of well-resolved galaxies, possibly even at low surface brightness. In this first paper on galaxy colours in SkyMapper, however, we consider massive populations with little influence from stochasticity. Hence, we did not exploit SLUG’s capability to follow each single star in a large population and modified our approach to speed it up: we first run SLUG for 5 Myr with a constant star formation rate to create an elementary building block of a star-formation history with a total stellar mass of $10^{5}{\rm M}_{\odot}$. We sample the SED evolution of this building block at a high time resolution of 10 Myr for a duration of 14 Gyr. In a second step, we combine the building blocks of a more complex star formation history into a full SED. The main disadvantage of our method relative to the original SLUG mode is that we do not track metallicity evolution and instead assume a solar metallicity throughout. This choice affects mostly the age-tagging of old populations given the well-known age-metallicity degeneracy in broad-band SEDs (Worthey, 1994), but it does not shrink the space occupied by our models.

We consider a range of SFHs with a parametrised grid, where the principal stellar population is always described as an exponential model, $SFR(t)=Ce^{-t/\tau}$, with $\tau$ ranging from 3 to 100 Gyr. At $t=10$ Gyr, we place an event that is either a starburst or a quenching event. Quenching is parametrised by a second tau model with $\tau_{q}\in(0.03,0.1,0.3)$ Gyr. We also use $\tau_{q}=0$ for sudden complete quenching. Bursts are modelled with additional star formation that remains constant for 100 Myr and ranges from 1% to 100% of the instantaneous SFR at $t=0$. Figure 2 shows a sketch of this SFH parametrisation.

Figure 3 shows the SkyMapper colour locus at redshift 0, for the grid of galaxy SEDs described above, over a time range from 1 to 14 Gyr after formation, and assuming no dust reddening. The three panels on the right show that galaxy colours in the passbands $griz$ span effectively only a 1-parameter family of colours, where time since formation, the $\tau$ of secular evolution, and the strength of and phase within a quenching or bursting event are all degenerate. The dust vector in the figure shows that even dust reddens the SEDs along the same 1-parameter relation. In the SDSS system, a similar tight relation prevails, albeit shifted slightly due to differences between the two filter sets.

The left panels of Fig. 3 show information added by filters on the blue side of 400 nm. The $v-g$ index straddles the 4000Å-break and is thus close to a traditional $D_{4000}$ measurement, which is mostly driven by age of the stellar population or specific star-formation rate. It shows the effects of a second parameter in the galaxy SED family, driven by quenching and bursting events. As is well known from interpretations of colour-magnitude relations, the colour is also sensitive to metallicity in old stellar populations (e.g. Bower, Lucey & Ellis, 1992; Poggianti & Barbaro, 1997), which is not explored further in this work.

The left-most panel shows the $u-v$ index, which hardly depends on $\tau$, or on the time $t$ since formation, or on the mean age; $u-v$ is expected to be a constant for all galaxies undergoing secular evolution, irrespective of age! Instead $u-v$ responds distinctly to quenching and bursting events and is an indicator of recent change in SFR irrespective of dust extinction. This also implies that a $u-g$ index will have no more information on the age of a galaxy than $v-g$ and instead only more noise due to the fainter $u$ fluxes of galaxies. SDSS, in contrast, has a single, broader $u$ band, so that the single diagram of $g-i$ vs. $u-g$ contains all information on the SFH of a galaxy that is discernible from SDSS colours alone.

Pending an investigation of dust effects, it is conceivable that the SkyMapper filter set can disentangle the effects of mean age of the population and recent changes in the star formation history, while SDSS passbands offer too few constraints to disentangle these two parameters from $ugriz$ alone. While this is a promising outlook for further exploration in a separate paper, what matters for this paper is the space covered by galaxies relative to AGN.

Figure 3 also shows the colour locus for QSOs and nuclei of bright Seyfert 1 galaxies at redshift 0. We take the QSO spectral template from Vanden Berk et al. (2001), separate the emission-line contour from the continuum and reassemble them into a grid of QSO properties. We span a 2D-space of QSO SEDs using a range of slopes multiplied onto the mean continuum (from $\Delta\alpha=+1$ to $-1$), and independently vary the intensity of the emission-line spectrum by factors from $0.33\times$ to $2.4\times$ relative to the mean.

There is little distinction between the QSO and galaxy loci in $griz$. This may come as a surprise given the usually quite blue power law slopes of QSO continua. However, the QSO template used here was built from empirical spectra in the wavelength range of the SDSS spectrograph. The optical section of the template is thus dominated by contributions from low-redshift QSOs, which are largely of lower luminosity and thus show contributions from their host galaxies. While the host galaxy contribution explains the break in the power law of the template continuum near 400 nm, as shown convincingly by the improved QSO template of Selsing et al. (2016), the SDSS-based mixed template is appropriate for the low-redshift AGN in this work.

On the blue side of 400 nm, however, QSO and galaxy SEDs deviate from one another, and the $u-v$ index may provide the clearest distinction of any single measure. Naturally, we assume that the nuclear SED of an AGN can be approximated by a blend of light from a stellar population and an active nucleus. Seyfert-2 galaxies mostly add narrow emission lines and optionally a weak featureless continuum, which have never been successfully differentiated in broad-band optical photometry from the SEDs of star-forming galaxies. Seyfert-1 galaxies of increasing nuclear luminosity will form a mixing sequence from their nuclear stellar population to their relevant QSO SED. A similar mixing sequence arises when using SEDs from apertures of different sizes that include different amounts of host galaxy light.

We note that Pol & Wadadekar (2017) provide a spectral template specifically for Seyfert-1 galaxies, whose spectral index is redder at wavelengths of $\lambda>400$ nm by up to $\sim 0.5$ at the lowest redshift. However, this difference results from an increased contribution from the host galaxy as evidenced also by an increased strength of stellar absorption lines in the template. We continue to start from the somewhat purer QSO SED of Vanden Berk et al. (2001) and explore mixing of QSO+galaxy SEDs in Sect. 4.

Finally, we investigate briefly the utility of adding the GALEX near-ultraviolet (NUV) band to the SEDs. We find the $NUV-u$ colour index to range from $\sim-0.5$ in starbursts to $\sim+5.5$ in old populations, while the QSO grid at redshift 0 ranges from $\sim-0.2$ to $\sim+0.7$. Because of this overlap in colour between AGN and starbursts, the GALEX NUV band does not add any discriminating power to the SkyMapper $u-v$ index.

Given the intriguing possibility that SkyMapper $u-v$ could be used as a burst-and-quench or AGN indicator, we investigate how it changes with redshift. As galaxy SEDs are redshifted, their A star features will get increasingly out of sync with the two passbands $u$ and $v$, so the $u-v$ index will lose its diagnostic power as a burst-and-quench indicator eventually.

We recalculate the colour locus in steps of $\Delta z=0.01$ and find an increasing dispersion of event-free colours. This is, because the two filters no longer straddle the 365 nm Balmer break; at $z=0.1$ both the $u$ and the $v$ filter probe entirely the rest-frame near-UV region below 365 nm, and hence the $u-v$ index shows simply a UV slope that depends on star formation and mean age of the stellar population. We also find a trend whereby the mean $u-v$ colour of normal galaxies gets bluer with redshift, reducing the difference between normal galaxies and AGN. These effects tend to diminish the power of the $u-v$ index as an AGN indicator at redshifts of $z>0.05$. While redshift is known a-priori for many galaxies and its effect can be calibrated, the decline in discriminating power can not be overcome. We revisit this point with empirical data in Section 3.

6dFGS contains $\sim 114,601$ sources with recession velocity $cz>300$ km s^-1 and redshift quality flags of 3 or 4, which means that over 95% of the redshifts are considered reliable. Over 98% of them have a counterpart in SkyMapper within a matching radius of $2\arcsec$. We further require them to have in each of the six bands at least one Main Survey image with a photometric measurement of good quality (FLAGS$<4$ and NIMAFLAGS$<5$). The shallower 2MRS contains 25,398 sources with $v>300$ km s^-1 and a counterpart in DR3 within $2\arcsec$, most of which are duplicate with 6dFGS. The additional objects include very bright galaxies and galaxies at low Galactic latitude outside the 6dFGS footprint. We complement the selected 6dFGS dataset with missing sources from 2MRS, and given that those have larger footprints in the images, we include measurements with NIMAFLAGS$<20$, whose central photometry is still expected to be reliable.

DR3 contains tables of the galaxy catalogues that are already cross-matched to the SkyMapper master table, which can then be further joined with the photometry table using the unique object_id. The photometry table contains one row for every single detection of an object in a unique image and can be joined with the images table using the image_id. The entry exp_time in the image table reveals the type of image, where Main Survey images all have 100 sec, while Shallow Survey images range from 5 to 40 sec depending on the filter.

With a suitable SQL query, we selected objects with Main Survey photometry in all six bands and averaged their magnitudes among the repeat visits within any filter. This results in a sample of 25,958 galaxies in 6dFGS, complemented with 1,527 galaxies from 2MRS, with deep SkyMapper $uvgriz$ photometry.

The photometry table contains a sequence of magnitudes in apertures ranging from $2\arcsec$ to $30\arcsec$ diameter, as well as Petrosian and PSF magnitudes. The aperture magnitudes are corrected for aperture losses assuming a PSF shape (while the aperture flux columns are left uncorrected to preserve the original measurements). Seeing-induced aperture losses are thus compensated and aperture magnitudes will vary less with seeing changes than is usually the case for extended sources. They are, however, not perfectly stable against seeing variations since the varying amount of light scattered into the aperture from outside is not compensated (for more detail see Onken et al., 2019) .

We first use the 6dFGS sample to investigate the usefulness of the DR3 catalogue photometry in describing nuclear photometry of nearby galaxies. The fraction of total galaxy light captured by the apertures, and hence the contrast of the nucleus, depends obviously on the distance of a galaxy and on its physical extent.

To this end, we inspect the colour scatter in the red sequence of predominantly passive galaxies. We use galaxies in the redshift range of $4,000<v<8,000$ km s^-1 and exclude faint galaxies with a $5\arcsec$-aperture magnitude of $g_{\rm APC05}>16$.

We de-redden the photometry in $uvg$ bands with dust maps from Schlegel, Finkbeiner & Davis (1998) and filter coefficients derived in Wolf et al. (2018) using a Fitzpatrick (1999) extinction law. From the Petrosian $g$ band magnitude g_petro we determine a luminosity $M_{g}$ (without K correction given the low redshift of 0.02). After selecting red galaxies we obtain linear fits for the colour-magnitude relation, using $5\arcsec$ aperture colours, and find:

The absence of a slope in the $u-v$ colour-magnitude relation confirms our expectation that the $u$ band adds no sensitivity to age or metallicity and the $v-g$ index is a sufficient proxy for $D_{4000}$.

With an rms of 0.075 mag in $(v-g)_{0}$ and 0.05 mag in $(u-v)_{0}$, this aperture has the smallest colour scatter in the red sequence. Towards larger apertures the scatter increases, most likely due to colour gradients in early spirals mixed into the red sequence. In the smaller $3\arcsec$ aperture, the rms scatter also increases, probably due to noise and seeing variations, to 0.115 mag and 0.068 mag respectively. We thus continue to work with $5\arcsec$ aperture colours.

We test the dust correction empirically by looking for correlations in nuclear galaxy colour with interstellar foreground reddening, which ranges from $E(B-V)\approx 0.01$ to $0.27$ in this sample. We restrict the comparison to red-sequence galaxies, which have been successfully used as standard crayons by Peek & Graves (2010) given their tight and homogeneous colour distribution.

Wolf et al. (2018) derived synthetic reddening coefficients from the SkyMapper filter curves and the Fitzpatrick (1999) reddening law, finding $R_{v}-R_{g}=4.026-2.986=1.040$ and $R_{u}-R_{v}=4.294-4.026=0.268$; the latter value is close to zero, because $u$/$v$ is a close pair of relatively narrow filters²²2According to survey documentation at http://skymapper.anu.edu.au/filter-transformations, these $R$ values are to be used with $A=R\times E(B-V)_{\rm SFD}$ and not with $A=R\times 0.86E(B-V)_{\rm SFD}$, a misunderstanding that has affected literature on SkyMapper photometry.. From the red-sequence galaxy sample selected in the previous section, we find that the $v-g$ correction appears appropriate. A fit to the empirical colours yields a reddening coefficient of $R^{\rm obs}_{v-g}=1.114\pm 0.065$, which is within $\sim 1\sigma$ of $R^{\rm W18}_{v-g}=1.040$. However, we find with $4\sigma$ confidence that the $u-v$ colour is over-corrected, and indeed no evidence that a dust correction in $u-v$ is required at all. The empirical fit yields $R^{\rm obs}_{u-v}=-0.020\pm 0.038$, which is within $0.5\sigma$ of no reddening and inconsistent with $R^{\rm W18}_{u-v}=0.268$.

We note that $R^{\rm W18}_{u-v}=0.268$ is expected to reduce to $R^{\rm W18}_{u-v}=0.22$ for observations made at an increased airmass of 2, where the atmospheric extinction effectively reddens the $u$ band filter curve, and a Cardelli, Clayton & Mathis (1989) law reduces it further to $R^{\rm C89}_{u-v}=0.16$. These are subtle and interesting facts to consider for high-quality calibrations in the $u$ band, maybe also in light of a future refinement of the filter efficiency curve; we note, that we have no knowledge of the CCD sensitivity at $\lambda<350$ nm, i.e. for half of the filter transmission region. There is also the possibility that the photometric calibration in DR3 has a reddening-dependent bias as it is tied to Gaia DR2 in a way that involves a reddening-dependent extrapolation (see details in Onken et al., 2019).

Because of these subtleties and the fact that dereddening or not makes less than a $\pm 0.02$ mag difference for $>95$% of the sample, we choose to ignore reddening in $u-v$ and set $R_{u}-R_{v}\equiv 0$ from now on. Since $u-v$ is a burst-and-quench indicator and most galaxies are expected not to show much signal, we also remove the red-sequence cut now and find that galaxies overall have $\langle u-v\rangle=0.410$ with an rms scatter of $0.058$ mag.

In this work, we study the colours of galaxies with AGN and specifically investigate whether we can differentiate Seyfert-1 galaxies from other galaxies by colour. The requirement for spectra in this work is then to reveal the broad emission lines of true type-1 AGN. Spectra also allow some differentiation of type-2 AGN from non-AGN galaxies, depending on spatial resolution, but this will be a non-critical by-product. While the 6dF catalogue specifies redshifts, it contains no information on the galaxy type. We first consider known AGN from three, partially overlapping, literature sources:

1. 1.

   The Hamburg-ESO QSO Survey (HESQSO; Wisotzki et al., 2000) covers less than 20% of the Southern hemisphere, but is arguably the most complete survey for broad-line AGN as it is based on objective-prism spectra for AGN selection. At low redshift, this approach requires that objects display a broad H$\beta$ line.

2. 2.

   The RASS-6dFGS catalogue (Mahony et al., 2010) covers the whole 6dFGS footprint with a sample of X-ray-selected AGN in the ROSAT Bright Source Catalogue (Voges et al., 1999). Its completeness ought to be very high and unbiased between narrow-line and broad-line types for objects of sufficient X-ray luminosity. Especially, obscured type-2 AGN would be included, while dormant AGN with low accretion rate are expected to be missed. However, as we will see below, quite a few Seyfert galaxies of interest for this work are not contained in this catalogue. The RASS-6dFGS table contains a classification into narrow-line and broad-line objects based on the spectral atlas of the 6dFGS.

3. 3.

   The Siding Spring Southern Seyfert Spectroscopic Snapshot Survey (S7; Dopita et al., 2015; Thomas et al., 2017) has recently re-observed 131 previously known AGN with the integral-field unit WiFeS at the ANU 2.3m telescope (Dopita et al., 2007). The S7 lists a finer classification into subtypes based on their spectra and is limited in redshift to $z\la 0.02$.

The combined sample is not a volume-limited complete AGN sample, but it contains AGN of all subtypes and a complete subset of optically cross-identified, flux-limited X-ray sources.

We find that the colours of this AGN sample are generally as expected: AGN with broad emission lines are the bluest objects in $u-v$, with colours ranging from $-0.2$ to $+0.15$; while those with only narrow lines can be as red as the general galaxy population, up to $u-v\approx+0.5$. Searching the complete 6dF sample for AGN by eyeballing $\sim 25,000$ spectra is beyond the scope of this paper. An alternative approach to finding AGN would be to use automated line-fitting and a classification based on line ratios. Even though 6dFGS has been around for over a decade, such a catalogue has never been published, and we are not endeavouring to do so as part of this work either. However, at $z<0.1$ we notice over 300 objects with $u-v$ colours in the range of broad-line AGN and away from the normal galaxy distribution (see Sect. 4 for details), which we consider candidates for further broad-line AGN.

In order to extend our AGN sample with further broad-line AGN, we visually inspect the 6dFGS spectra of our $(u-v)$-blue candidate AGN and identify 51 further type-1 to 1.8 AGN not currently documented elsewhere. For the significant fraction of 6dFGS galaxies that were not observed by the 6dFGS itself, we adopt the classification from Simbad if it specified a Seyfert type. We also inspect the 6dFGS spectra of the literature AGN sample to verify subtypes consistently. We classify the spectra by eye, using the Seyfert types of 1, 1.5, 1.8, 1.9 and 2 as defined by Osterbrock (1981): Seyfert-1 galaxies feature several broad Balmer emission lines; type 1.5 denotes objects where the broad H$\beta$ line is clearly present but comparable in flux to the narrow [Oiii] lines, 1.8 denotes a barely noticeable H$\beta$ line, 1.9 an absent H$\beta$ line, and 2 a spectrum without any broad emission lines.

For differentiating Seyfert-2 and star-forming galaxies we considered line flux ratios seen by eye in [Nii]/H$\alpha$ and [Oiii]/H$\beta$ (BPT diagram; Baldwin, Phillips & Terlevich, 1981; Kewley et al., 2001; Kauffmann et al., 2003). Here, we find several composite spectra not clearly attributed to one type. While the line ratios could be refined with automated fitting, this is beyond the scope of this paper, and the quality of composite classification has no consequence for our results. Given the spatial mixing sequence from nuclear AGN spectra to star-forming host galaxy spectra (Davies et al., 2014), the AGN/composite divide depends most sensitively on spatial resolution as highlighted by the discussion of inclusion and dilution effects in Tremou et al. (2015). The spatial resolution in 6dFGS is low due to large fibres with $6\aas@@fstack{\prime\prime}7$ diameter and without better resolved spectroscopy the AGN/composite classification in this sample will remain sub-optimal. Line-fitting and classification of the full 6dFGS spectral atlas as well as resolution issues will be tackled in a separate work to produce a quantitative community resource.

Among the 6dFGS spectra we also notice several AGN, where a barely visible broad H$\beta$ line suggests type-1.8, while the H$\alpha$/H$\beta$ ratio appears regular (although uncertain flux calibration in 6dFGS makes this ratio unreliable); we suspect these are regular type-1 spectra diluted by a large host contribution in the 6dF fibres. This issue also relates to the discussion of low-luminosity AGN by Ho, Filippenko & Sargent (1995) and highlights the problem that even a well-defined Seyfert type depends on the observing conditions.

In total, we consolidated spectral types for 688 galaxies at $z<0.1$. This includes 6 BL Lac objects, 269 AGN of type Seyfert-1 to 1.5, 23 Seyfert-1.8, 57 Seyfert-1.9 and 98 Seyfert-2 galaxies, as well as 33 spectra we interpret as Sy-2/star-forming composite spectra, 132 star-forming galaxies and 61 galaxies with absorption-line spectra, and eight known Changing-Look AGN with variable type. In the appendix we list the newly identified Seyfert galaxies that were not reported previously to our knowledge, together with their Simbad classification from May 2020.

The diagnostic power of the $u-v$ index to reveal AGN or starbursts (and quenching events) relies on the alignment of the SkyMapper filters with crucial features in galaxy spectra that is only given at low redshift. The continuum colour of AGN and starbursts is nearly independent of redshift in virtually all galaxies that SkyMapper can spatially resolve (at $z<1$), but the spectra of common normally star-forming galaxies will reveal their different Hydrogen Balmer breaks in $u-v$ only at redshifts close to 0.

Thus, we look at how the $u-v$ index depends on redshift in the bulk of the galaxy population, and compare its trend with the colours of known Seyfert galaxies. Restricting the sample to $M_{g,\rm petro}<-19$ retains more than 90% of the galaxies at $z<0.1$, while losing only one known Seyfert-1 and four Seyfert-2 galaxies. Figure 6 shows the trend of $u-v$ with redshift for normal galaxies (light grey background dots), while objects with known spectroscopic types are rendered as larger symbols. Most type-1 to 1.5 AGN show bluer $u-v$ colours than the bulk of the galaxy population. We also see AGN on the normal galaxy locus, which are mostly narrow-line (type-2) AGN, whose weak featureless AGN continuum affects the $u-v$ colour very little relative to the colour of a pure stellar population.

Normal galaxies trend towards bluer $u-v$ with redshift as the Balmer break moves away from the point where it is straddled by the $u$ and $v$ band. This trend reduces the contrast between normal galaxies and type-1 AGN, whose colours are nearly independent of redshift (until the Mgii emission line enters the $u$-band at $z>0.2$). A second-order fit to this sample yields $\langle u-v\rangle=0.408+0.838z+34.9z^{2}$. At $z\la 0.05$ type-1 AGN appear well separated from normal galaxies, while the populations blend together towards $z\approx 0.1$. At $z>0.05$ we see a concentration of Seyfert-1 AGN with $u-v\approx 0\pm 0.1$, just as expected from the grid of QSO SED colours; these are clearly AGN with higher nuclear luminosity and subdominant host galaxy light. Starburst galaxies lie between the QSO grid and the bulk of the galaxy population. While there is significant overlap between starburst galaxies and type-1 AGN, the colour range of $u-v<0.1$ is clearly dominated by AGN. The synthetic colours of population synthesis suggested that a young starburst that has lasted for only 10 Myr can be as blue as $u-v\approx 0.05$ at redshift 0 and $u-v\approx-0.05$ at redshift 0.05. As the starburst continues and builds up a population with an increasing age, it gets redder in $u-v$, as the number of $(u-v)$-blue O stars remains stable while the number of $(u-v)$-red A stars grows continuously. In practice, the nuclear photometry of a galaxy will see a mix of a pre-existing older stellar population with the recent starburst event. E.g., a new starburst with a star formation rate of 1 M_⊙ yr^-1 only reaches $u-v\approx 0.1$ at $z=0$ for the first few million years when it is superimposed onto a 10 Gyr-old population with a stellar mass of $10^{9}$ M_⊙.

Obviously, our ability to identify AGN from the $u-v$ index depends on the type and luminosity of the active nucleus as well as the central stellar population it is superimposed on, and it weakens with increasing redshift. Details can only be quantified once all the spectra in the 6dFGS atlas are classified, which might reveal further AGN at lower luminosity than have been found in this work.

The SEDs of high-luminosity QSOs are dominated by AGN light, so the $u-v$ colour index clearly separates them from galaxies, but here we focus on Seyfert galaxies at lower luminosity. Their SEDs are composites of light from a stellar population and an active nucleus that form a galaxy-AGN mixing sequence. Fig. 7 shows the $u-v$ vs. $v-g$ colours of galaxies and known AGN. Close to $z=0$, inactive galaxies are concentrated around $u-v\approx 0.4$, as long as they are neither bursting nor undergoing quenching. Crucially, the $u-v$ colour is independent of the mean age and the absolute level of the current star-formation rate, which are well-known to strongly affect the strength of the 4000Å-break and the $v-g$ colour, but not the 365 nm Balmer break and the $u-v$ colour. In the $u-v$ vs. $v-g$ diagram, red-sequence galaxies are centred on $(u-v,v-g)_{0}\approx(0.4,1.8)$, while star-forming galaxies extend horizontally out to values of $(v-g)_{0}\approx 0.2$. Known AGN tend to be bluer than normal galaxies in $u-v$ irrespective of their $v-g$ colour. AGN of type 1.9 to 2 are closest to the colours of normal galaxies, but already extend bluewards from the non-AGN sequence in $u-v$, suggesting the presence of either some featureless AGN continuum or of starbursts in the central $5\arcsec$ apertures of the galaxies, a feature commonly seen in type-2 AGN (e.g. Cid Fernandes et al., 2001). AGN of type 1.5 to 1 are furthest offset to bluer $u-v$ colours and reach all the way to pure QSO SEDs (shown as a grid of dots).

With increasing redshift, the distribution of normal galaxies extends to bluer colours in $u-v$, because the two filters no longer straddle the 365 nm Balmer break. The $u-v$ index then probes a UV continuum slope and depends on mean age of the stellar population. Thus, it becomes highly correlated with the $v-g$ index and leads to overlaps between young galaxies and broad-line AGN.

Some AGN seem to defy the expectation that the continuum colour of a type 1.9-2 spectrum should differ little from that of normal galaxies. However, AGN might have changed their type between the early spectral epoch and the recent SkyMapper epoch. Eight such Changing-Look AGN (CLAGN) are known in the sample and marked with circles. They include e.g. NGC 2617 at $z=0.014$, which shows a type-1.9 spectrum in 6dFGS but has QSO colours in SkyMapper. CLAGN are further discussed in Sect. 5.

The lines in the diagram represent four mixing sequences, which start at representative points in the galaxy SED cloud and mix in increasing amounts of AGN light from different corners of the QSO SED grid. The colours of the known AGN relative to the mixing sequences emphasise that central stellar populations in AGN cover a broad range in colour, possibly with a preference for intermediate colours, which could be studied in the future.

Fig. 7 suggests that we may be able to separate type-1 and type-2 AGN based on the $u-v$ colour, provided the type-1 nucleus is luminous enough to move the composite SED away from normal galaxies. The contrast between AGN and stellar light depends obviously not only on the luminosity of the AGN but also on the central stellar population and the angular resolution of the aperture, hence on distance. We thus investigate the mixing sequences in terms of $u-v$ colour vs. surface luminosity in the central aperture. Surface brightness is independent of distance, except for cosmological surface brightness dimming, which we ignore here due to the small range of local redshifts; in this approximation, aperture luminosity density is equal to the extinction-corrected apparent aperture magnitude $u_{0}(5\arcsec)$ apart from an offset given by the distance modulus where a 5$\arcsec$-aperture covers a 1 kpc² area.

Our sample mostly ranges from $u_{0}\approx 14$ to $19$ mag, which corresponds to a $u$ band ABmag surface luminosity density of approx. $-19$ to $-14$ mag kpc^-2. If all light came from a nuclear point-source with a QSO spectrum, this range would correspond to nuclear luminosities of $M_{B}\approx-19$ to $-14$ at $z=0.01$. At our far end of $z\approx 0.08$, the $u_{0}=[14,19]$ range corresponds to $M_{B}=[-23.5,-18.5]$.

The AGN contribution makes the galaxy centre not only bluer but also brighter as illustrated in Fig. 8. The dashed lines represent the same mixing sequences as in Fig. 7 but also assume a brightness of the stellar population as an indicative starting point. In the lowest redshift bin, the colour of the stellar populations is so constrained in $u-v$ that colour alone may be sufficient to separate type-1 AGN from non-AGN galaxies, apart from increased scatter towards the faint end due to noise. The galaxy distribution extends to high aperture brightness or surface luminosity density, because the increase in spatial resolution towards $z\rightarrow 0$ focuses the aperture on the bright galaxy core. With increasing redshift the distribution loses its bright end as the $5\arcsec$ aperture averages the galaxy light profile over an increasing physical aperture. At this point, normal galaxies with young stellar populations show bluer $u-v$ colours that overlap with those of type-1 AGN, and the strongest indicator of a Seyfert-1 galaxy is the brightness of the nuclear region.

Using both axes, we define a selection cut for likely Seyfert-1 galaxies, which removes normal galaxies and type-2 AGN:

Whether this rule selects Sy-1 galaxies or not depends on the luminosity of their active nuclei. Starting from the an average non-starburst galaxy with $u-v=0.4$, we calculate how much AGN light is needed to move the $5\arcsec$ photometry across the selection threshold. At $z=0.01$, a median galaxy with $u\approx 17.5$ reaches the selection line after adding a nucleus of equal $u$ band brightness. A nucleus of $u=17.5$ corresponds to a luminosity of $M_{B}=-15.5$, meaning that our Seyfert-1 selection reaches a factor of 1000 below the luminosity of the classic QSO-Seyfert divide.

There are Seyfert-1 nuclei of lower luminosity known in more nearby galaxies that are seen with better spatial resolution and stronger contrast against the host stellar population (Ho, Filippenko & Sargent, 1995). An extreme example is the galaxy NGC 4395, which is $16\times$ nearer than $z=0.01$ and hosts a Sy-1 nucleus with $M_{B}\approx-10$ (Filippenko & Sargent, 1989). At the distance of NGC 4395 (2.6 kpc) our selection threshold would be $M_{B}<-9.5$, consistent with the expectation that the nucleus could be discerned in ground-based and seeing-limited observations. At the other end of the range considered here, $z=0.08$, we need to add a nucleus of $u\approx 18$ on typical host galaxies to reach the selection threshold, corresponding to $M_{B}\approx-19.5$.

We now consider the mix of galaxy types selected by Eq. 3. At $z<0.08$ and ignoring the known CLAGN, Eq. 3 selects $\sim 70$% of the known Sy-1 to 1.5 AGN and BL Lac objects. It also selects $\sim 40$% of the Sy-1.8, $\sim 30$% of the Sy-1.9 and $\sim 20$% of the Sy-2 AGN. Of the known composite objects and star-forming galaxies $\sim 7$% are selected. In terms of purity, the cut selects 246 (non-CLAGN) objects with known type, among which $\sim 75$% are Sy-1 to 1.5 AGN or BL Lac objects, $\sim 20$% are Sy-1.8 to 2 AGN, and $\sim 7$% are composites or starburst galaxies.

In Fig. 9 we show histograms of $u-v$ colour for different AGN types, restricted to $u_{0}<17.5$. A gradual trend is evident where the Balmer breaks get redder from type 1-1.2, over type 1.5, towards type 1.8-2, although even type-2 galaxies are usually not as red in $u-v$ as normally star-forming or quiescent galaxies. There are several reasons for overlap between the types:

• •

  Intrinsic SED mixing: redder active nuclei can appear bluer when accompanied by nuclear starbursts, and bluer active nuclei will appear redder when they are only of moderate luminosity and blended with redder stellar populations.

• •

  Intrinsic variability: AGN are variable sources, but the colours used in Fig. 9 are non-instantaneous and only accurate for non-variable SEDs. Even if the variability affected only brightness but not colour, the fact that different filters can be observed at different times may introduce spurious changes in colours.

• •

  Measurement errors: the formal mean 1$\sigma$-error in the $u-v$ index is $0\aas@@fstack{m}03$ and broadens the per-type distributions.

• •

  Evolution: an extreme form of variability is type evolution as seen in Changing-Look AGN.

The eight known Changing-Look AGN are found on both sides of the cut. It is also possible that some of the other objects in the sample have changed their AGN type between the time their spectra were taken and the more recent photometric observations with SkyMapper. How much the overlap of AGN types in the selection diagram of Fig. 8 is intrinsic and how much it is a result of objects evolving between the photometric and the spectroscopic observation cannot be answered without spectroscopy taken at the time of the photometric observations.

Whether the physical difference between type-1 and type-2 is dust obscuration or accretion rate, the observable spectra are similar. While the types 1 vs. 2 are defined by the presence of broad spectral lines, their presence is linked to the visibility of a blue featureless continuum from an accretion disk. Hence, we expect that spectral type could be inferred with some certainty from photometric SEDs that probe the central continuum SED of galaxies. Objects selected by Eq. 3 are very likely to be type-1 AGN, while objects outside of this cut are more likely to be type-2 AGN or inactive galaxies.

The conclusion is that objects with type-1 spectra in 6dFGS that are not selected by Eq. 3 are good candidates for present-day type-1.8 to 2 AGN and thus may be turn-off Changing-Look AGN. Conversely, we expect that those 46 objects selected by Eq. 3 that have type-1.8 to 2 spectra in 6dFGS are good candidates for present-day type-1 AGN and may be turn-on Changing-Look AGN. However, these objects might well have been type-1 AGN all along as a more nucleus-focused spectrum might reveal, but in order to test for any change, modern-epoch spectra need to be extracted with $6\aas@@fstack{\prime\prime}7$-apertures for a fair comparison.

A further 16 star-forming galaxies appear with bright nuclei and a blue Balmer break. We consider them contaminants of the broad-line AGN candidate list, while their origin may be either due to real blue transients such as tidal disruption events (TDE, see Sect. 6.1), extreme starbursts, or due to data artefacts as discussed in the following section.

All colour indices used in this work so far are averaged over all the available data and thus not instantaneous. It is not even guaranteed that the recorded mean colours represent the true mean colour of an individual object: seeing variations affect aperture photometry of extended objects and can thus introduce mild but spurious variability into the light curve of any single passband. When colours are made from two bands, they are only expected to be precise when the individual visits in the two bands occurred under a similar mix of conditions, which is not always the case. This issue introduces additional scatter into the distilled mean colours.

However, in the SkyMapper Southern Survey, the $uv$ bands are usually observed within minutes of each other, which allows near-instantaneous colour measurements. Variability in the light curves of extended objects that results from seeing variations will cancel out in the colours defined by bands with similar seeing levels as is the case for the two bands $u$ and $v$ that are very close in wavelength.

Variability that is coincident between two bands is often used as a strong indication of its true significance, see e.g. work on stellar flares in SkyMapper by Chang, Wolf & Onken (2020). For the particular case of variability in low-redshift AGN, the mixing sequences in Fig. 8 also illustrate how a true change in nuclear brightness is expected to be strongly related to a change in $u-v$ colour, providing further scrutiny on the nature of any detected variability signal. This point will be illustrated in the following section, where we investigate the light curves of known CL-AGN alongside instantaneous $u-v$ colours.

The SkyMapper colour index values discussed above are clipped weighted averages of measurements spread over the years 2014 to 2019, which may have averaged over low and high states and lowered the contrast for discovering changes. We thus discuss light curves of these objects with an emphasis on changes in the $u-v$ index. Now, we average photometry from multiple exposures only if they are taken in the same night, which usually happens only for Main Survey $uv$ exposures. The derived $u-v$ indices are thus from single nights³³3We allow for one exception: a $u-v$ index for Mrk 609 in late 2016 is formed from data separated by 63 days.. We include photometry from Shallow Survey images⁴⁴4Exposures in the Shallow Survey are on average 1.0 and 1.75 mag less deep than Main Survey images in $u$ and $v$ band, respectively. as well as long as their magnitude errors are $<0.05$.

Fig. 10 shows the path of the known CLAGN in the plane of colour vs. nuclear brightness, and Fig. 11 shows the light curves of the objects with the strongest changes. We note in particular, how the CL-AGN with larger variability nicely track the expected orientation of mixing sequences shown in Fig. 8.

In the following, we compile the scattered notes from the literature on the changes observed in individual objects and discuss how they compare with the observations included in DR3 of the SkyMapper Survey:

If any type-1 episodes are as short-lived as the case in 1998, SkyMapper is likely to miss them, as the overall number of repeat visits is small; the Main Survey schedule plans visits in the $uv$ filter pair only on two nights for any field. We still see modest variability in SkyMapper DR3, notably a decline of nearly 0.2 mag in $u$ and $v$ bands from 28 June 2018 to 18 Sep 2018 and a $\sim 0.1$ mag rise within the 24 hours to 19 Sep 2018. However, with $u-v\approx 0.4$ throughout the variations, we would expect the AGN to have been a type-1.9 or 2 at all observed epochs.

The fairly faint object ($u_{\rm APC05}\approx 18$) was observed by the SkyMapper Main Survey in $u$ and $v$ only on 29 October 2014. The $riz$ bands all appear to show a brightening by 0.14 mag on 10 August 2014 relative to 8 and 11 August 2014, however, these observations occurred in untypical seeing and are probably insignificant. Krumpe et al. (2017) report that Mrk 1018 reached a minimum around October 2016 and rebrightened in 2017. DR3 has further visits in $i$ and $z$ band in 2018 and 2019, showing that Mrk 1018 brightened by $\Delta i\approx 0.17$ from 31 July 2015 to 12 August 2018 and stayed at that higher level also on 23 Sep 2019.

The SkyMapper DR3 photometry shows modest variability in brightness from August 2014 to July 2015. By Aug/Sep 2018 the galaxy had brightened in all bands, from $\Delta z=-0.13$ to $\Delta u=-0.50$ and all of its colours became bluer with $u-v$ changing from $+0.26$ in Oct 2014 to $+0.19$ in 2018. Compared to other Seyfert galaxies in our sample, this makes us suspect that during 2014/5 a type between 1.8 and 2 was most likely, while by 2018 the object had come close to the Sy-1 selection in Eq. 3.

The SkyMapper DR3 photometry shows that Mrk 530 rose by $\sim 0.4$ mag in $u$ band from 2-7 July 2014 to 1 July 2015 (see Fig. 11); at the same time its $u-v$ colour changed from $0.0$ to $-0.10$. The $riz$ light curve shows that a high state occurred on 27 June only ten days before the lower state in July. We see another dip in September 2017 and another rise in July 2018. The colours suggest that the galaxy has been a type-1 from June 2014 to 2018.

In SkyMapper DR3 we see significant variability (see Fig. 11), with two low states in October 2014 and October 2016 showing $u-v=0.28$ in each case, and two high states, a brief one during late August 2015 and a possibly enduring one from August 2017 to September and December 2018, with $u-v\approx 0.14$. This makes the galaxy a candidate for changing between type-2 in 2014 to maybe type-1 to 1.5 in 2015, back to type-2 in 2016 and back again to type-1 to 1.5 in 2017 and 2018. However, a spectrum by Senarath et al. (2020) from 14 September 2018 shows it as a type-1.9.

SkyMapper DR3 has only one epoch for the $u-v$ colour, which was $+0.34$ on 29 November 2016, clearly consistent with a type-2 AGN. The photometry in the other bands shows little variability, with one exception: from January to August 2015 the galaxy brightened by $(\Delta i,\Delta z)=(-0.17,-0.11)$ mag, and by December 2016 it had fallen back to the level of January, where it was still seen in September 2019. This may suggest an epoch of activity in late 2015 that could have been accompanied by broad lines.

The SkyMapper data (see Fig. 11) show significant variability from 2014 to 2019. As the source faded by $\Delta u\approx 0.4$ from 11 March 2018 to 29 December 2018, its $u-v$ colour reddened from $-0.03$ to $+0.08$. However, by 10 April 2019 it had brightened by $\Delta u\approx-0.65$ and resumed its bluer level with $u-v-0.03$ again. The $griz$ band light curves have their lowest minimum in December 2018, at which stage we would still expect it to have been a type-1-1.5 based on the $u-v$ colour. The highest state within the coverage of DR3 was reached in January 2015, when the source was $\Delta v=-1.21$ mag higher than during the minimum. There might be a mild tension between the blue $u-v$ colour on 11 March 2018 and the ’weak H$\beta$ line’ seen by Oknyansky et al. (2018) in a spectrum from ’April/May 2018’.

The SkyMapper data (see Fig. 11) show the source in a low in 2014 with a colour of $u-v=0.42$ indicative of a Sy-1.9/2 AGN. By 16 February 2018 it had brightened by $\sim 1$ mag in $u$ band and reached $u-v=0.19$ getting close to our Sy-1 selection cut. By 16 March 2018 it had brightened by a further $\sim 0.2$ mag in $u$ band and reached $u-v=0.11$ making it a strong Sy-1 to 1.5 candidate. The next SkyMapper $uv$ observations were from 28 August 2018, in the declining phase after the peak observed by Oknyansky et al. (2019): the SkyMapper $u$ band was still brighter by an additional $\sim 0.5$ mag relative to March 2018 and the $u-v$ colour had reached $+0.06$, entirely consistent with a Sy-1.2 AGN. Two months later it had declined in $v$ band to the level in Feb/Mar 2018.

All the above variability from SkyMapper is consistent with nuclear-only changes in brightness, because on an absolute scale the Petrosian flux changes roughly in line with the nuclear aperture flux. The correlated changes in colour and brightness as well as the correlations between $u-v$ index and spectral types emphasise that the SkyMapper $u-v$ index is a useful proxy for Seyfert types. Monitoring AGN with the SkyMapper $u$ and $v$ bands is thus useful for detecting variability in AGN that goes beyond regular variability into the regime of type changes. Deciding whether CLAGN mark simply an extreme tail in a single mode of AGN variability, or constitute a separate phenomenon appearing in a specific subset of AGN, will require larger statistics of AGN variability.

TDEs produce blue featureless continua as much as AGN do, although their line emission may differentiate them from AGN, involving not only Balmer lines but He II emission and possibly Ne III emission resulting from Bowen flourescence, and by showing evolution as the TDE progresses (see e.g. Arcavi et al., 2014; Blagorodnova et al., 2019; Leloudas et al., 2019). As such, we expect that we will have no means to photometrically distinguish TDEs and CL-AGN based on only a single epoch of photometry. With multiple epochs, however, the two cases could be distinguished as TDEs usually fade back to nearly host level after less than 200 days in the SkyMapper $u$ band at $z\la 0.1$ (see example light curves in Holoien et al., 2018; Leloudas et al., 2019; Holoien et al., 2020).

Auchettl, Ramirez-Ruiz & Guillochon (2018) estimate the rate of TDEs to be 70 to 470 per million years per galaxy. If these were detectable for (optimistically) half a year in the SkyMapper $u-v$ colour, we might have between 1 and 6 TDEs in the sample of this paper. They could thus be a modest but noticeable portion among the 46 plus 16 candidates selected by Eq. 3, which have archival spectra showing type-1.8 to 2 AGN and normal galaxies, respectively. Correlating known TDEs with the SkyMapper data set will be reserved for a separate paper.

For supernovae, we estimate colours from template spectra of different supernova types (e.g. Filippenko, 1997; Foley et al., 2016) at the redshifts of $z\la 0.1$ considered here and find that their SkyMapper $u-v$ colours are redder than AGN continua due to metal absorption bands. This appears to be the case for supernovae of all common types including Ia, Ic and II, and irrespective of the fact that the SkyMapper $v-g$ and $g-i$ colours can successfully differentiate thermonuclear SNe Ia from core-collapse SNe Ic and II at these redshifts (Scalzo et al., 2017).

We find one particular entry in the Open Supernova Catalog (OSC, Guillochon et al., 2017), which is one of our turn-on CL-AGN candidates and also has $u-v$ data around its maximum: ASASSN-14ko is listed as a type IIn supernova discovered on 14 Nov 2014 by Holoien et al. (2014) about $0\aas@@fstack{\prime\prime}85$ from the centre of the galaxy ESO 253-G003 at $z=0.0425$ that has been typed as a Seyfert-2 galaxy from an archival spectrum. In the following, we attempt to establish whether this object is indeed a contaminating supernova or a type-1 AGN as suggested by our $u-v$ colour.

SkyMapper $uv$ photometry covers the object on 9+10 Nov 2014, 11 Mar 2018, 15 Aug 2019 and 8 Oct 2019. The object has a bright state with $(u,v)=(15.86\pm 0.01,15.82\pm 0.01)$ that corresponds to the possible SN peak in Nov 2014 and again with $(u,v)=(15.87\pm 0.01,15.80\pm 0.01)$ in Oct 2019. In between it is seen in a fainter state with $(u,v)=(16.60\pm 0.04,16.46\pm 0.07)$ in Mar 2018 and $(u,v)=(16.84\pm 0.05,16.75\pm 0.04)$ in Aug 2019. Based on the $u-v$ colour, ranging from $\sim 0.05$ in the bright state to $\sim 0.12$ in the faint state, we considered this object a candidate Seyfert-1 galaxy, and based on the earlier Seyfert-2 classification it is one of our 46 new candidate turn-on Changing-Look AGN.

Usually, AGN are not expected to occur $0\aas@@fstack{\prime\prime}85$ ($\sim 700$ pc in projection) away from the nucleus of a galaxy. However, the SkyMapper $i$ band images reveal that this object is a post-merger galaxy surrounded by a tidal arm; we see two nuclei with $\sim 1\aas@@fstack{\prime\prime}5$ separation, and the $u$ band image of Nov 2014 shows ASASSN-14ko as a bright point source precisely at the location of the North-Eastern nucleus (see Fig. 12). ASASSN-14ko is thus a nuclear transient and may just be a flare of an AGN. Recently published VLT/MUSE observations by López-Cobá et al. (2020) reveal even three nuclei, as the Southwestern nucleus is resolved into two cores (see Fig. 12).

Holoien et al. (2014) mention in their discovery telegram that strong AGN variability, while unusual for Type II AGN, cannot be ruled out. Dimitriadis et al. (2014) label it consistent with a SN IIn from a spectrum taken on 24 Nov 2014 at the ESO/NTT. We reproduce the spectrum⁵⁵5from https://wiserep.weizmann.ac.il/object/5901 in Fig. 12 but do not recognise any unique SN features; to our eyes, it is indistinguishable from the SDSS-based Seyfert-1 template by Pol & Wadadekar (2017). The MUSE data by López-Cobá et al. (2020) were taken in December 2015 and show a complex source structure. Importantly, broad Balmer lines are seen at the location of the North-Eastern nucleus, which appear similar to the earlier NTT data, more than a year after the initial ASASSN transient. Since the blue $u-v$ colour stretches at least five years, we claim that the transient ASASSN-14ko is not a supernova but the flaring tip of the iceberg of a highly variable, newly discovered Changing-Look AGN. Finally, Payne et al. (2020) argue in a new paper, that this object is even a periodically flaring AGN.