AT 2019avd: A novel addition to the diverse population of nuclear transients

We report on SRG/eROSITA, ZTF, ASAS-SN, Las Cumbres, NEOWISE-R, and Swift XRT/UVOT observations of the unique ongoing event AT 2019avd, located in the nucleus of a previously inactive galaxy at $z=0.029$. eROSITA first observed AT 2019avd on 2020-04-28 during its first all sky survey, when it was detected as an ultra-soft X-ray source ($kT\sim 85$ eV) that was $\gtrsim 90$ times brighter in the $0.2-2$ keV band than a previous 3$\sigma$ upper flux detection limit (with no archival X-ray detection at this position). The ZTF optical light curve in the $\sim 450$ days preceding the eROSITA detection is double peaked, and the eROSITA detection coincides with the rise of the second peak. Follow-up optical spectroscopy shows the emergence of a Bowen fluorescence feature and high-ionisation coronal lines ([\ion{Fe}{X}] 6375 {\AA}, [\ion{Fe}{XIV}] 5303 {\AA}), along with persistent broad Balmer emission lines (FWHM$\sim 1400$ km s$^{-1}$). Whilst the X-ray properties make AT 2019avd a promising tidal disruption event (TDE) candidate, the optical properties are atypical for optically selected TDEs. We discuss potential alternative origins that could explain the observed properties of AT 2019avd, such as a stellar binary TDE candidate, or a TDE involving a super massive black hole binary.

Whilst the sample of ignition events in galactic nuclei was previously limited to only a few objects, the advance of widefield, high-cadence surveys over the last decade has facilitated the discovery of an increasing number of extreme state changes. This has resulted in tighter constraints on the timescales of flaring events for these systems. For example, Trakhtenbrot et al. (2019b) recently reported a new class of SMBH accretion event that sees a large amplitude rise in the optical/UV luminosity over timescales of months.
In addition to triggering drastic changes in the accretion rate in AGNs, TDEs can also cause quiescent black holes to transition into short-lived active phases. In a TDE, a star that passes too close to a BH is torn apart by strong tidal forces, with a fraction of the bound stellar debris then being accreted onto the BH (Hills 1975;Young et al. 1977;Gurzadian & Ozernoi 1981;Lacy et al. 1982;Rees 1988;Phinney 1989). Early TDE candidates were first identified through detection of large-amplitude (at least a factor of 20), ultra-soft X-ray flares (black-body temperatures between 40 and 100 eV) from quiescent galaxies during the ROSAT survey (Bade et al. 1996;Komossa & Bade 1999;Komossa & Greiner 1999;Grupe & Leighly 1999;Greiner et al. 2000). Since then, the vast majority of TDE candidates have been optically selected, such as through the Sloan Digital Sky Survey (SDSS; e.g. van Velzen et al. 2011;Merloni et al. 2015), the Panoramic Survey Telescope and Rapid Response System (Pan-STARRS; e.g. Gezari et al. 2012;Holoien et al. 2019a), the Palomar Transient Factory (PTF; e.g. Arcavi et al. 2014), the Intermediate Palomar Transient Factory (iPTF; e.g. Blagorodnova et al. 2017;Hung et al. 2017), the All Sky Automated Survey for SuperNovae (ASAS-SN; e.g. Holoien et al. 2014Holoien et al. , 2016Wevers et al. 2019;Holoien et al. 2019b), and the Zwicky Transient Facility (ZTF; e.g. van Velzen et al. 2019van Velzen et al. , 2020. Optically selected TDEs are characterised as blue nuclear transients with light curves showing longer/ shorter rise and decay timescales relative to supernovae (SNe)/ AGN 1 , and a relatively smooth power-law decline. Optical spectroscopic follow-up of these events post-peak reveals blue continua (blackbody temperatures ∼ 10 4 K) with various broad emission lines (full width at half maximum, FWHM 10 4 km s −1 ); a recent characterisation of the different TDE spectroscopic classes was presented by van . Although a number of TDE candidates have also been found through UV selection (GALEX, Gezari et al. 2008, and X-ray selection (XMM-Newton Slew, Esquej et al. 2007Esquej et al. , 2008Saxton et al. 2012Saxton et al. , 2017, most of our understanding of TDEs is currently biased towards this set of observed properties of optically-selected TDEs. Whilst most previous TDE searches focused on identifying TDEs in quiescent galaxies, an increasing number of candidates for TDEs in AGNs are being proposed in the literature (Merloni et al. 2015;Blanchard et al. 2017;Trakhtenbrot et al. 2019a;Liu et al. 2020;Ricci et al. 2020). In certain cases, the distinction between TDE and non-TDE-induced SMBH accretion state changes is becoming increasingly blurred (see also Neustadt et al. 2020). Variants of TDEs have also been proposed to explain more exotic phenomena, such as the recently observed quasi-periodic eruptions (QPEs) in a few galactic nuclei (Miniutti et al. 2019;Giustini et al. 2020;King 2020), and periodic flaring seen in an AGN (Payne et al. 2020). Other origins for extreme nuclear transients involve SNe in the AGN accretion disc (Rozyczka et al. 1995), or interaction of SMBH binaries (SMBHB) with an accretion disc (Kim et al. 2018). It is clear that such different physical origins may result in a diverse range of observed variability behaviours.
In this paper, we report on the ongoing extreme event AT 2019avd, which is a novel addition to the already diverse population of nuclear transients. AT 2019avd is associated to the previously inactive galaxy 2MASX J08233674+0423027 at z = 0.029 (see Fig. 1), and was first reported as ZTF19aaiqmgl at the Transient Name Server (TNS 2 ) following its discovery by ZTF on 2019-02-09 UT 3 . The transient was independently detected more than a year later on 2020-04-28 as a new ultra-soft nuclear X-ray source (Malyali et al. 2020) during the first all-sky survey of the eROSITA instrument (Predehl et al., in press) on-board the Russian/German Spectrum-Roentgen-Gamma (SRG) mission.
This work presents X-ray (SRG/eROSITA, Swift/XRT), optical/UV/mid-infrared (MIR) photometric (ZTF, ASAS-SN, NEOWISE-R, Swift/UVOT), and optical spectroscopic (NOT/ALFOSC, Las Cumbres Floyds, ANU/WiFeS) observations of AT 2019avd. In Section 2, we report our X-ray observations and analysis of AT 2019avd, whilst the photometric evolution and host galaxy properties are presented in Section 3. We then present details of our optical spectroscopic follow-up campaign in Section 4, before discussing possible origins for AT 2019avd in Section 5, and conclude in Section 6. We adopt a flat ΛCDM cosmology throughout this paper, with H 0 = 67.7 km s −1 Mpc −1 , Ω m = 0.309 (Planck Collaboration XIII 2016); z = 0.029 thus corresponds to a luminosity distance of 130 Mpc. All magnitudes will be reported in the AB system, unless otherwise stated. Brunner et al. in prep.) was systematically examined for new soft X-ray sources associated with the nuclei of galaxies that showed no prior indication of being an AGN. The eROSITA data for AT 2019avd are composed of four consecutive scans with gaps of 4 hr each and a midtime of 2020-04-28. The total on-source exposure amounts to 140 s (see Table 1). The source was localised to (RA J2000 , Dec J2000 )=(08h23m37s, 04 • 23 03 ), with a 1σ positional uncertainty of 2 , which is consistent with the nucleus of the galaxy 2MASX J08233674+0423027.

X-ray observations
Photons were extracted using the eSASS task SRCTOOL (version 945) choosing a circular aperture of radius 36 centred on the above position (84 counts were detected within this region). Background counts were selected from a circular annulus of inner and outer radii 72 and 144 , respectively. Using the best-fit spectral model (see Section 2.3), we derived a 0.2−2 keV flux of (1.4±0.2)×10 −12 erg cm −2 s −1 (1σ).

Swift follow-up
Triggered by the eROSITA detection, a series of follow-up observations were performed with the Neil Gehrels Swift Obser-  Table 2. Swift UV photometry (corrected for Galactic extinction using the UVOT correction factors in Table 5 of Kataoka et al. 2008 Malyali & B. Trakhtenbrot). Observations were obtained roughly every 7 days, until the source was no longer visible due to Sun angle constraints; a further Swift observation was then obtained ∼ 3 months later. A log of the observations can be found in Table 1. The XRT observations were performed in photon counting mode. The data were reduced using the xrtpipeline task included in version 6.25 of the heasoft package. Spectra for each of the five epochs were extracted using the xrtproducts task. Source counts were extracted from a circular aperture of radius 47 and background counts extracted from a circular annulus of inner and outer radii 70 and 250 , respectively 7 .
Observations with the Ultraviolet and Optical Telescope (UVOT; Roming et al. 2005) were obtained simultaneously with the XRT observations. Imaging was performed at three epochs (00013495001, ..004, ..005) using the UVW1 filter with exposures of 1.36, 1.95, and 1.93 ks, respectively. The remaining three observations utilised all six UVOT filters (UVW2, UVM2, UVW1, U, B, V) with accordingly shorter exposure times.
The UVOT flux was extracted with the uvotsource task using a 9 radius aperture centred on the optical position of AT 2019avd, whilst a nearby circular region with 15 radius was used for background subtraction. The photometry was extracted from each unique Swift observation ID, and is presented in Table 2 (we note that this photometry includes both AGN and host galaxy emission in order to be consistent with the SED fitting in Section 3.3). Relative to UV photometry obtained prior to the initial optical outburst (see Section 3.3 and Fig. 7), AT 2019avd has brightened by ∼ 1 mag in the UVW1, UVM2, and UVW2 bands, and brightens only by ∼ 0.1 − 0.2 mag over Swift observations between 2020-05-13 and 2020-09-16.  . BXA fit to the eROSITA eRASS1 spectrum. Black markers are the binned observed data, whilst the red represents the fitted convolved model for tbabs*blackbody (darker and light red bands enclose the 68 % and 95 % posterior uncertainty on the model at each energy). Both the black-body and power-law fits to the (low count) eRASS1 spectrum suggest that the source is ultra-soft (see Table 4).

X-ray spectral fitting
X-ray spectra were analysed using the Bayesian X-ray Analysis software (BXA, Buchner et al. 2014), which connects the nested sampling algorithm MultiNest (Feroz & Hobson 2008) with the fitting environment CIAO/Sherpa (Freeman et al. 2001) and XSPEC (Arnaud 1996). The spectra were fitted unbinned using the C-statistic (Cash 1976), and the eROSITA and XRT backgrounds were both modelled using the principal component analysis (PCA) technique described in Simmonds et al. (2018). For each set of eROSITA and XRT spectra, a joint fit on both the source and background spectra was run. Two different models for the source spectra were used: (i) an absorbed black body (tbabs*blackbody), and (ii) an absorbed power law (tbabs*powerlaw). The equivalent Galactic neutral hydrogen column density, N H , was allowed to vary by 20% from its tabulated value in the HI4PI survey of 2.42 × 10 20 cm −2 (HI4PI Collaboration et al. 2016) during fitting. The complete set of priors adopted under each model is listed in Table 3, whilst an example of the BXA fit to the eROSITA spectrum is shown in Fig. 3, and spectral fit results are presented in Table 4.
Over the course of the six weeks following the initial eROSITA detection, there was no major variability in the 0.2 − 2 keV flux between the eROSITA and XRT observations (Table 4 and Fig. 4). However, the 0.2 − 2 keV flux in the last Swift epoch increased by a factor of about six relative to the previous observation.
AT 2019avd remained in an ultra-soft state during the Swift monitoring campaign, although there is variability in the inferred black-body temperatures (kT ranges between minimum and maximum values of 72±8 eV and 132±10 eV, respectively). The inferred black-body temperatures are similar to those measured in the X-ray emission of previously observed thermal TDEs (45 kT 130 eV, e.g. van Velzen et al. 2020), and are also consistent with the temperatures of the soft excess shown in AGN (e.g. Table A1 in Gliozzi & Williams 2020).

Optical evolution
The region around the position of AT 2019avd has been monitored by ZTF Graham et al. 2019)  For MJD<58855 (2020-01-07), we obtained a forced photometry ZTF light curve for AT 2019avd (Masci et al. 2019). For MJD>58855, we downloaded the ZTF light curve of AT 2019avd using the Lasair alert broker (Smith et al. 2019), which processes and reports to the community on transients detected within the large ZTF data streams. Both of these light curves are constructed from PSF-fit photometry measurements run on ZTF difference images. We also obtained additional photometric observations with the Spectral Energy Distribution Machine (SEDM; Blagorodnova et al. 2018) on the Palomar 60-inch telescope. The SEDM photometry was host-subtracted using SDSS reference images, as described in Fremling et al. (2016). These two light curves, and the host-subtracted SEDM photometry, were then combined for subsequent analysis, and are shown in Fig. 5.
After the initial detection on 2019-02-09, AT 2019avd continued to brighten until reaching its maximum observed brightness of r ∼ 16.8 mag on 2019-02-20. Between 2019-02-24 and 2020-01-01, the g-band magnitude of the host nucleus decayed nearly monotonically from 17.13±0.09 mag to 20.08±0.20 mag, followed by a re-brightening to 18.58±0.13 mag on 2020-05-03. The late time SEDM photometry around 2020-09-19 revealed a further brightening to r and g-band magnitudes of ∼ 17.6 mag and ∼ 18.4 mag respectively. The first eROSITA observation occurred during the rise of the second major peak of the ZTF light curve (Fig. 5).
The location of AT 2019avd has also been monitored in the V-band by ASAS-SN (Shappee et al. 2014;Kochanek et al. 2017) from February 2012 to November 2018, and in the g-band from October 2017 to September 2020 (the time of writing). No major optical outbursts were seen in the ASAS-SN light curve prior to the ZTF detection (Fig. B.1); given the joint ASAS-SN and ZTF light curves, it is likely that the system 'ignited' around MJD = 58510 (2019-01-27). Table 3. Summary of priors adopted in the BXA analysis of the eROSITA and XRT spectra. For each fit, a log-uniform prior on N H between (0.8N H , 1.2N H ) was defined, where N H = 2.42 × 10 20 cm −2 (see Section 2.3). Γ denotes the slope of a power law, kT the black-body temperature, A the normalisation. The prior over A is in units 1.05×10 −6 erg cm −2 s −1 .

Model
Priors Table 4. X-ray spectral fit results from applying BXA to the extracted eROSITA and XRT spectra, with uncertainties enclosing 68% of the posterior for each parameter. F 0.2−2keV is the inferred observed (unabsorbed) flux under each model.  (Table 5). No significant variability before the initial 2019 outburst is observed in the host nucleus of AT 2019avd with archival NEOWISE-R and ASAS-SN observations (Fig. B.1). The NEOWISE-R observations pre-outburst are observed with mean W1, W2 marked out in the top panel by the cream and orange dashed lines respectively. For plotting clarity, we omit the high-cadence ZTF Partnership observations obtained between MJD 58820 and 58860, and we rebin the ∼ 3 SEDM observations in each filter into a single data point.

Rise and decay timescales in the light curve
In the following, we fit the light-curve model presented in equation 1 of van Velzen et al. (2019), which models the rise with a half-Gaussian function, and an exponential function for the decay, to the first and second peaks of the ZTF light curve, using UltraNest 9 (Buchner 2016, 2019) as our sampler. Whilst such a model is not physically motivated, it enables a comparison of the timescales involved in the light curve of AT 2019avd with those of the population of ZTF nuclear transients presented in van Velzen et al. (2019). While fitting the first peak, we first filter out observations outside of the MJD period between 58450 and 58650, and we then run a joint fit of the g and r band observations in flux space. Our model has seven free parameters, defined following the notation of van Velzen et al. (2019): σ r and σ g , the rise timescale of the light curve in the r and g bands respectively; τ r and τ g , the decay timescale of the light curve in r and g bands; F peak,r and F peak,g , the peak flux in r and g bands; t peak , the time of the peak of the light curve (to enable a comparison with van Velzen et al. 2019, we assume that the light-curve model peaks at the same time in both of these bands). For the second peak, we filter out observations outside of the MJD period 58840 and 59115 (the late-time SEDM datapoints are used in the fitting), and because we do not sample the decay of this peak, we only model the rise here. The model for the second peak has five free parameters, with τ r and τ g now being omitted. We list our priors in Table A.1, and present the fits in Fig. 6.
From the posterior means, we infer σ r = 7.9 ± 0.3, σ g = 7.2±0.2, τ r = 58.2±0.5 and τ g = 39.8±0.4 days for the first optical peak (68% credible intervals). Whilst the rise timescales in each filter are consistent with each other to within 2σ, the decay timescales in each filter significantly differ. With τ r > τ g , the first peak shows a potential cooling signature during its decay phase, although we are unable to constrain the temperature evolution during this because of a lack of contemporaneous observations in other wavelength bands. Relative to the population of nuclear transients in van Velzen et al. (2019), one sees that these are short rise and decay timescales relative to those of AGN flares, and are thus more similar to those in the van Velzen et al. (2019) sample of TDEs and SNe (Fig. 6). As expected from Fig. 5, the inferred rise times for the second peak are longer and more AGN-like, with τ r ∼ 88 days and τ g ∼ 93 days.

Mid-infrared variability
The location of AT 2019avd was observed in the W1 (3.4 µm) and W2 (4.6 µm) bands by the Wide-Field Infrared Survey Explorer mission (WISE, Wright et al. 2010)  The MIR light curve was observed to be flat prior to the initial ZTF outburst, but showed significant brightening in the 9 https://github.com/JohannesBuchner/UltraNest 10 https://irsa.ipac.caltech.edu/frontpage/ , with red and green stars computed from the fitted model components for each respective filter. The red and green vertical lines mark the e-folding rise time of the second optical peak in the r and g bands, respectively. We also plot the rise and decay e-fold timescales inferred from the ASAS-SN V-band light curve of the nuclear transient AT 2017bgt (Trakhtenbrot et al. 2019b; see also Section 5.1) with a black marker. Not only is the double-peaked light curve of AT 2019avd clearly distinct from the other light curves of sources in the AT 2017bgt nuclear transient class, but the first peak of AT 2019avd decays much faster than the AT 2017bgt flare, whilst the second peak rises much slower than the AT 2017bgt flare.
first NEOWISE-R epoch obtained thereafter. Observations obtained ∼ 6 months later found the source to still be in the bright state despite having faded by ∼ 3 mag in the optical. The MIR brightening was also accompanied by a significant reddening, evolving from W1 − W2 ∼ 0.08 mag in AllWISE, to a more AGN-like W1 − W2 ∼ 0.6 mag during flaring. The W1 − W2 colour before the outburst is much lower than the suggested cuts (W1 − W2 0.7 mag) for identifying AGNs in previous MIR classification schemes (Stern et al. 2012;Assef et al. 2013Assef et al. , 2018, further supporting the hypothesis that there was no strong recent AGN activity in AT 2019avd at that time (although the use of WISE colours for selecting AGNs is less effective at lower AGN luminosities; see discussion in Padovani et al. 2017).

Host-galaxy properties
The spectral energy distribution (SED) of the host galaxy of AT 2019avd was compiled from archival 11 UV to MIR photometry from GALEX (FUV, NUV), SDSS DR12 (g, r, i, z), UKIDSS (y, J, H, K), and AllWISE (W1, W2). The SED was modelled using CIGALE (Burgarella et al. 2005;Boquien et al. 2019), which allows the estimation of the physical parameters of a galaxy by fitting composite stellar populations combined with recipes describing the star formation history and attenuation. The best-fitting model (see Fig. 7) is that of a galaxy with a stellar mass of (1.6 ± 0.8) × 10 10 M , a star formation rate (SFR) of 0.17 ± 0.05 M yr −1 , and little attenuation, E(B − V) = 0.03 ± 0.02 mag, which experienced a burst of star formation 3.7 ± 0.2 Gyr ago. The inferred stellar mass and SFR place the host galaxy of AT 2019avd in the 'green valley' between the star-forming main sequence and quenched elliptical galaxies (adopting the green valley definition presented in Law- Smith et al. 2017).
The SED fit suggests that the host galaxy did not show strong signs of nuclear activity prior to the detection of AT 2019avd. This is further supported by the absence of a radio counterpart in the FIRST catalogue (Becker et al. 1995) within 30 of AT 2019avd, with a catalogue upper detection limit at this position of 0.96 mJy/beam 12 .

Spectroscopic observations
On 2019-03-15, ∼33 days after the first observed peak in the ZTF light curve, an optical spectrum of AT 2019avd was obtained by Gezari et al. (2020) with the Alhambra Faint Object Spectrograph and Camera (ALFOSC) 13 on the 2.56 m Nordic Optical Telescope (NOT). The spectrum was obtained with a 1 . 0 wide slit, grism #4 (covering the wavelength region from 3650-9200 Å), and the slit was positioned along the parallactic angle at the beginning of the 1800s exposure. Reductions were performed in a standard way using mainly iraf based software, including bias corrections, flat fielding, wavelength calibration using HeNe arc lamps imaged immediately after the target and flux calibrations using observations of a spectrophotometric standard star.
No further spectra were taken until after eROSITA had detected the large-amplitude soft-X-ray flare from AT 2019avd in late April 2020, which triggered a further five epochs of spectroscopy (dates listed in Table 5) using the FLOYDS spectrographs (Brown et al. 2013) mounted on the Las Cumbres Observatory 2m telescopes at Haleakala, Hawaii, and Siding Spring, Australia. Each spectrum was taken with a 3.6ks exposure, using the 'red/blu' grism and a slit width of 2 . The spectra were reduced using PyRAF tasks as described in Valenti et al. (2014). FLOYDS covers the entire 3500-10000 Å range in a single exposure by capturing two spectral orders (one red and one blue) simultaneously, yielding R ∼ 400. The different orders are usually merged into a single spectrum using the region between 4900 and 5700 Å, which is present in both the red and blue orders. However, in this case, in order to avoid erroneous wavelength shifts at the blue edge of the red order (where there are fewer arclines), all FLOYDS spectra were merged using a reduced stitching region of 5400 to 5500 Å 14 . This stitching was done manually in Python, by replacing fluxes in that wavelength range with an average of the linear interpolations of the two orders.
In addition, a higher resolution spectrum (R∼3000) was obtained on 2020-05-29 with the Wide Field Spectrograph (WiFeS; Dopita et al. 2007Dopita et al. , 2010) mounted on the 2.3m ANU telescope at Siding Spring Observatory. We employed the R3000 and B3000 gratings, and obtained an arc lamp exposure after each target exposure. The total spectral range from the two gratings is 3500 to 7000 Å. The data were reduced using the PyWiFeS reduction pipeline (Childress et al. 2014), which produces threedimensional data (data cubes). These spectra are bias subtracted, flat-fielded, wavelength and flux calibrated, and corrected for telluric absorption. We then extracted the spectra from the slitlets that captured AT 2019avd using the IRAF (Tody 1986) task apall which allowed for background subtraction.
A comparison of the NOT and WiFeS spectra is presented in Fig. 8, and the spectral evolution in the FLOYDS spectra is shown in Fig. 9. A log of the spectroscopic observations of 12 http://sundog.stsci.edu/cgi-bin/searchfirst. 13 http://www.not.iac.es/instruments/alfosc 14 The most extreme arcline used to calibrate each order is at ∼5460 Å. AT 2019avd is presented in Table 5. We note that we have not found any archival optical spectra of the host galaxy that were obtained prior to the initial 2019 outburst discovered by ZTF.

Summary of the main observed features of the optical spectra
The NOT spectrum from 2019-03-15 appears similar to broad line AGN spectra, showing a relatively flat continuum (in terms of F λ ) and broad Balmer emission lines (Hα, Hβ, Hγ, Hδ; Fig. 8). However, the strong Fe ii complex that is frequently seen in some AGNs is not present.  Pelat et al. 1987), which is not detected. The FLOYDS spectra (Fig. 9) show no major evolution in the Balmer emission line profiles, and show the broad emission feature around 4680 Å from 2020-05-10 (for epochs with sufficiently high S/N ratios in the blue wavelength range), which was reported to the TNS (and first identified) in Trakhtenbrot et al. (2020).

Optical spectrum modelling
For the two higher resolution spectra (NOT and WiFeS), the region around the main observed emission lines is fitted separately (Hγ, 4240Å < λ < 4440Å; He ii, 4500Å < λ < 4800Å; Hβ, 4700Å < λ < 5000Å; Hα, 6364Å < λ < 6764Å; [S ii] doublet, 6650Å < λ < 6800Å; and ±100Å of the line centre for [O iii] 5007 Å, [Fe x] 6375 Å). Each emission line complex is modelled with multiple Gaussians (an overview of these is presented in Table A.2), and each complex is fitted independently of the others. For all spectral fits, we assume a flat continuum component during the fitting process, and run our model fitting using the region slice sampler option within UltraNest. Spectral fits for the NOT and WiFeS spectra are shown in Figs. 10 and 11, whilst the spectral fit results are listed in Tables 6, 7, and 8.   3.0 ± 0.5

Balmer emission
From the best-fitting spectral models, we infer a broad Balmer decrement, F(Hα b )/F(Hβ b ), of 3.4 in the WiFeS spectrum (we use superscripts 'b' and 'n' to refer to the broad and narrow components of a given emission line when such are clearly detected). Such a decrement is consistent with what is observed in AGNs (e.g. Dong et al. 2005Dong et al. , 2007Baron et al. 2016), and is slightly higher than the predicted value of around 2.74-2.86 15 for case B recombination (Baker & Menzel 1938) and thus a photoionisation origin. Whilst it was originally thought that the observed 15 The predicted value is dependent on the assumed gas density and temperature.  . The WiFeS spectrum is of much higher resolution relative to the NOT spectrum, and therefore is able to better resolve narrow emission lines, such as the [S ii] doublet at 6716 and 6731 Å. Neither are shown corrected for Galactic extinction. The NOT spectrum was normalised by its continuum flux in the 5100-5200 Å range (rest frame), whilst the blue and red arms of the WiFeS spectra were normalised in the 5100-5200 Å and 6400-6450 Å ranges respectively (rest frame).
broad and narrow Balmer emission lines respectively (using the Calzetti et al. 2000 extinction law) 16 . We note that the E(B-V) inferred from the Balmer decrement is larger than that inferred from SED fitting, which was performed on photometry that included light emitted from a larger region in the host galaxy than that probed by the Balmer decrement analysis. strongest Fe ii transitions in the 4500-4600 Å or ∼5150-5350 Å ranges (e.g. Kovačević et al. 2010). When comparing the WiFeS AT 2019avd spectrum to the composite SDSS quasar spectrum presented in Fig. 2 Netzer et al. (1985) predicted the relative Bowen line intensities in AGNs under a range of different metal gas densities and abundances, where they found that to produce the high F(He ii)/F(H β b ) ratios seen in AT 2019avd as well as the high observed F(N iii 4640)/F(H β b ) ratio, the gas producing the Bowen fluorescence must have very high density (n H > 10 9.5 cm −3 ) and high N and O abundances relative to cosmic abundances.

Coronal lines
From the line fitting seen on the WiFeS spectrum in Fig. 11, we infer the luminosities of the [Fe x] 6375 Å and [Fe xiv] 5303 Å emission lines to be ∼ 2 × 10 39 and ∼ 3 × 10 39 erg s −1 . We also infer relative intensities of  Wang et al. (2012).
The Fe coronal lines are narrower relative to the He ii and N iii 4640 Å emission lines (Table 8), with FWHM for the [Fe xiv] 5303 Å and [Fe x] 6375 Å of 1560 ± 140 and 770 ± 40 km s −1 respectively. Under the assumption that the line widths are set by the virial motion of the gas, this suggests that the coronal lines are produced further away from the BH than the Bowen lines, and also with the higher ionisation coronal lines being produced closer to the BH than the lower ionisation lines. The width of [Fe xiv] 5303 Å is comparable to the observed Balmer emission. We also note the differing line profiles of the [Fe xiv] 5303 Å and [Fe x] 6375 Å emission, with the latter showing a stronger blue asymmetry (Fig. 11).
As discussed in Wang et al. (2012), the weakness of [Fe vii] emission relative to [Fe x] and [Fe xiv] may be explained through the coronal line gas either being overionised under a high X-ray flux, or due to collisional de-excitation of [Fe vii], because it has a lower critical density (∼ 10 7 cm −3 ) compared with the higher ionisation lines (∼ 10 10 cm −3 , Korista & Ferland 1989).

Black hole mass estimate
We assume that the gas that produces the broad Hβ emission is virialised around the SMBH at the centre of the galaxy, and use the 'single epoch' mass-estimation technique (e.g. Vestergaard & Peterson 2006) to infer the black hole mass using the following scaling relation from Assef et al. (2011):  (Assef et al. 2011). We note that using this technique requires the correlations between continuum luminosity and radius of the broad line region (BLR; e.g. Kaspi et al. 2005) obtained in previous AGN reverberation mapping experiments to also hold for the BLR around the SMBH in AT 2019avd.  (Baldwin et al. 1981), such line ratios suggest that a blend of star formation and AGN activity is responsible for producing the narrow line emission in the host galaxy of AT 2019avd (Kauffmann et al. 2003;Kewley et al. 2006). Without an archival spectrum though, it is unclear whether the [O iii] 5007 Å and [N ii] 6583 Å lines have increased in intensity since the initial ZTF outburst, or an AGN-like ionising source has always been present.   Table A.2). The lower plots in each panel show the residuals in the spectral fitting, where δF λ is the difference between the observed F λ and the model F λ , normalised by the model F λ . We note that the double peaked appearance of the He ii emission line in the WiFeS spectrum is most likely non-physical and due to the noisy optical spectrum, as no other broad lines show such similar line profiles.

Mapping out the BLR
Assuming that each observed emission line is broadened due to its virial motion around the central BH, we can use the measured FWHMs to obtain rough estimates of the distances from the central ionising source at which each line is produced (Fig. 12). Similar to previous work (e.g. Korista et al. 1995;Kollatschny 2003;Bentz et al. 2010), we also find evidence for a stratified BLR, whereby the higher ionisation lines are produced in regions closer to the BH.

Discussion
Based purely on its X-ray luminosity evolution, AT 2019avd most likely involves an accreting SMBH at the centre of a galaxy. Whilst the large amplitude X-ray flaring (factor of 600), soft X-ray spectrum, lack of previous strong (and sustained) AGN activity, and the implied unabsorbed X-ray peak luminosity in the 0.2 − 2 keV energy range of 2 × 10 43 erg s −1 (using spectroscopic z = 0.029, see section 4.1) initially made the source a strong TDE candidate, this is clearly discordant with the doublepeaked optical variability seen in the ZTF observations (it does not look like a prototypical, single-event TDE as observed elsewhere). In the following section, we discuss potential origins of the rich phenomenology seen in AT 2019avd.

AT 2019avd as non-TDE-induced AGN variability
If AT 2019avd is related to AGN activity that was not induced by a TDE (herein referred to simply as AGN 'activity' or 'vari-ability' 20 ), then the combination of its X-ray and optical light curves make it one of the most extreme cases of AGN variability observed to date.
It is clear that the X-ray spectrum of AT 2019avd (section 2.3) is far softer than what is commonly seen in Seyfert 1s; for example, the power-law slope for Swift OBSID 00013495001 was 5.3 +0.4 −0.4 , whilst Nandra & Pounds (1994) model the observed power-law slope distribution with a Gaussian distribution of mean 1.95 and standard deviation 0.15. However, based on the measured FWHMs of the broad Balmer emission lines in the optical spectrum, it would be classified as a NLSy1, and softer spectral indices have also been observed in the NLSy1 population; a systematic ROSAT study of this by Boller et al. (1996) found power-law slopes of up to ∼ 5. NLSy1s are also known to exhibit rapid, large-amplitude X-ray variability (e.g. Boller et al. 1996). As the X-ray variability of NLSy1s over longer timescales has not been extensively monitored before, how common AT 2019avd-like X-ray flares are within this population is currently unclear. For this reason, the X-ray properties alone cannot be used to state that the observed variability in AT 2019avd was induced by a TDE.
However, AT 2019avd shows a number of features in its optical spectrum that are infrequently seen in NLSy1s. First, NLSy1s commonly show strong Fe ii emission (e.g. Rakshit et al. 2017), whereas this is not seen in the WiFeS spectrum, and only a weak Fe ii complex is seen in the NOT spectrum in AT 2019avd. Instead, the most prominent Fe emission we observe are the transient, ECLE-like higher ionisation coronal lines of [Fe xiv] 5303 Å and [Fe x] 6375 Å in the WiFeS spectrum. During our spectroscopic follow-up campaign, we also observe the appearance of He ii 4686 Å and N iii 4640 Å emission lines (attributed to Bowen fluorescence). The optical spectrum at late times appears similar to the recently identified new class of flaring transients by Trakhtenbrot et al. (2019b), and we present a comparison of AT 2019avd with this class in Fig. 13. Whilst AT 2019avd shares the broad emission feature around 4680Å with the AT 2017bgt flare class, the optical spectrum of AT 2019avd is distinguishable from the other members based on its much weaker [O iii] 5007Å emission line. A likely reason for this is that the host galaxies of the other flares had persistent, higher luminosity AGNs in them prior to the optical outburst, relative to AT 2019avd. In addition, AT 2019avd's large amplitude, ultra-soft X-ray flare, and its optical light-curve evolution make it unique amongst the AT 2017bgt flare class.
Finally, we stress that the double-peaked optical variability shown by AT 2019avd is unprecedented for a NLSy1, which when combined with its X-ray properties, make AT 2019avd clearly unique relative to all previous examples of AGN variability. Further examples of NLSy1 variability seen during the ZTF survey will be presented in a separate publication ).  The pericentre and circularisation radii are computed assuming a Sun-like star incident on this BH with its closest approach at the tidal radius. Similarly to Kollatschny et al. (2014), we see evidence for a stratified BLR. The coloured lines represent length scales that were obtained based on observations of AT 2019avd, whilst the grey dashed lines are based on various scaling relations in the literature (BLR radius based on Kaspi et al. 2005, whilst the inner torus radius was computed using equation 1 of Nenkova et al. 2008, assuming a dust sublimation temperature of 1500K).

Canonical tidal disruption event
As AT 2019avd shows a very-large-amplitude, soft-X-ray flare from the nucleus of a galaxy that shows no strong signs of prior AGN activity, it appears similar to the predicted observational signatures for TDEs (e.g. Rees 1988) and most of the previous X-ray-selected thermal TDE candidates (Bade et al. 1996;Komossa & Bade 1999;Komossa & Greiner 1999;Grupe & Leighly 1999;Greiner et al. 2000;Saxton et al. 2019). On the other hand, its optical spectrum shows a far weaker blue continuum component relative to that seen in optically selected TDEs, as well as narrower Balmer emission lines (for TDEs where these are detected); based on these two pieces of evidence, it would be straightforward to declare that AT 2019avd is not a TDE candidate, according to criteria for optical TDE selection in van .
The observed broad Balmer emission lines in AT 2019avd instead appear more like those commonly seen in the broad emission lines of Seyfert 1s. With such similarity, a mechanism analogous to the broad line emission in AGNs is likely operating in AT 2019avd, whereby the line widths of hydrogen recombination lines are set by the gas kinematics (whereas some TDEs may have line widths set by repeated non-coherent electron scattering; e.g. Roth & Kasen 2018), and the high densities in the BLR result in the line intensity responding effectively instantaneously to changes in the continuum flux. In the limit of a weak TDE-like reprocessing layer 21 , the optical spectrum of a TDE may appear similar to that of an AGN, as has been previously suggested (e.g. Gaskell & Rojas Lobos 2014). The timescales for the evolution of the spectral features in such systems may be different from those observed in optically selected TDEs, as they originate from a region further away from the BH than the reprocessing layer. The optical emission mechanism in TDEs is currently not well understood, although it is thought to arise either from shocks produced from stellar debris stream self-intersections Piran et al. 2015), or from debris reprocessing the emission from an accretion disc (e.g. Loeb & Ulmer 1997;Ulmer et al. 1998;Roth et al. 2016;Roth & Kasen 2018). However, it is unclear how luminous the shocks are from stream self-intersections, whilst for the reprocessing scenario we still do not understand where the reprocessor is situated, where it forms, how large its covering angle would be from the BH, how efficiently it converts disc emission into the optical wavebands, or how all of these aspects are affected by the properties of the BH and those of the disrupted star. There is currently not a large 21 And likely a lack of optically-selected observed TDE features. enough sample of TDEs selected through both X-ray and optical surveys to test these various models of optical emission, and to properly assess the various complex underlying selection effects likely present in the existing TDE candidate population. A key example of these effects is the fact that only a small fraction of optically selected TDEs show transient X-ray emission (∼ 25% of optically selected TDEs in van Velzen et al. 2020 were X-ray bright); Dai et al. (2018) suggested that the observed properties of a TDE may be dependent upon the viewing angle to the newly formed disc.
Given the above, and that there are also no X-ray selected, non-relativistic TDEs in the literature that have high-cadence optical photometric light curves available 22 , we cannot rule out a TDE-related origin for AT 2019avd simply on the basis of a lack of optically selected TDE features in the optical spectrum. However, we do disfavour the canonical TDE interpretation (seen in optically selected TDEs) for this flare on the basis of the doublepeaked optical light curve, which has not been observed in any of the TDEs identified by ZTF so far. Secondary maxima have previously been seen in the light curves of some TDE candidates (a compilation is presented in Fig. 8 of Wevers et al. 2019), though not at optical wavelengths and of far smaller amplitude increase compared with AT 2019avd (with the exception of the TDE in an AGN candidate in Merloni et al. 2015).

A more exotic variant of a tidal disruption event?
A large fraction of stars may exist in binary systems (e.g. Lada 2006). Mandel & Levin (2015) studied the various outcomes of a binary star passing close to a SMBH from a nearly radial orbit. In ∼ 20% of such approaches, a double tidal disruption event (DTDE) is produced, whereby both stars in the binary are disrupted in succession. These latter authors estimated that ∼10% of all stellar tidal disruptions could be associated with DTDEs, with such events expected to produce double-peaked light curves.
We can use the inferred rise-to-peak timescales from the ZTF light curves to test the feasibility of whether AT 2019avd may have been triggered by a DTDE, specifically for the case where each peak is associated with the rise to peak mass fallback of each successive disruption. Guillochon & Ramirez-Ruiz (2013) present the time taken for a single TDE to reach peak mass fallback rate (in their equation A2): where B γ is a function of β, the ratio of the tidal radius of the BH to the pericentre of the orbit of the star, γ is the polytropic index of the star 23 , M BH is the black hole mass, and M and R are the mass and radius of the star being disrupted. Similarly to Merloni et al. (2015), we then generate a grid of M and β, log-uniformly between (0.1M , 100M ) and (0.5, 4), respectively, and compute R for each M using the mass-radius relationship for zero-age main sequence stars presented in Tout et al. (1996). For each possible combination of M and β, and for a black hole with log[M BH /M ] ∼ 6.3, we check whether it can produce t peak (using equation 2) within 20% of the observed peak timescales in the ZTF light curves (∼ 24 days and ∼ 260 days for the first and second peak respectively). We also enforce the  constraint that its tidal radius lies outside of the Schwarzschild radius for the system, so that it can produce a TDE with the star being swallowed whole by the black hole. We plot the permitted regions of the M , β parameter space in red in Fig. 14, where we see that no main sequence binary star configuration can reproduce the observed rise times for both the first and second peaks. It would also be possible to obtain further constraints on the feasibility of this scenario based on the observed peak luminosities (similar to Merloni et al. 2015) and their ratio, as well as from the inferred properties of the binary itself, such as from the time between the two observed peaks (which could be used to constrain the semi-major axis) and the inferred mass ratio. However, the constraints provided from t peak are perhaps the simplest to implement and are sufficient to highlight the caveats of a simple DTDE interpretation. Bonnerot & Rossi (2019) recently suggested that following the disruption of a stellar binary, the two separate debris streams may collide prior to their fallback onto the black hole. These collisions then shock-heat the gas, and were predicted to produce an optical flare prior to the main flare of the disruption event. Such a model for a binary TDE could potentially explain the observed double-peak light curve, and the observed emergence of the Bowen feature after the second peak (the soft X-rays can only be emitted once the accretion disc has formed). However, a caveat to this interpretation is that both a strong ionising flux and high gas densities are required for Bowen fluorescence to be produced, and we cannot confidently state here that the reason for not observing Bowen lines in the NOT spectrum is the absence of an X-ray-emitting accretion disc during that observation, because the absence of Bowen lines may also be due to insufficiently high gas densities (not all TDEs that are X-ray bright have displayed Bowen emission lines). We do not rule out this more complex DTDE scenario for AT 2019avd here, but do not perform a detailed comparison between the simulations in Bonnerot & Rossi (2019) and AT 2019avd in the present paper. Another alternative could be that AT 2019avd involved some type of TDE about a SMBH binary (e.g. Liu et al. 2009;Coughlin et al. 2017), where in such systems, the presence of the secondary BH can perturb the accretion flow onto the primary, leading to intermittent light curves.

Could AT 2019avd be supernova-related?
The spectra of Type IIn SNe can appear similar to those of AGNs (e.g. Filippenko 1989), as they can show broad and narrow emission lines, an absence of P-Cygni profiles, and higher luminosities and slower decay timescales relative to normal Type II SNe (Nyholm et al. 2020). Type IIn SNe typically also show the highest X-ray luminosities amongst all SNe. However, AT 2019avd has a L 0.2−2keV that is about an order of magnitude higher than what is seen in most X-ray-luminous Type IIn SNe, when considering the sample of IIn shown in Fig. 3 of Dwarkadas & Gruszko (2012). Furthermore, the X-ray emission from Type IIn SNe is predicted to be hard (e.g. Ofek et al. 2013), whilst that of AT 2019avd is ultra-soft. Based on the X-ray emission alone, we disfavour the idea that both optical peaks in AT 2019avd are related to a single Type IIn supernova.
Given the observed peak and decay timescales (Fig. 6), the peak absolute magnitude of the optical light curve (∼ −18.5), the small amount of reddening seen in the ZTF light curve during the decay phase, and the NOT spectrum, the first optical peak may have been associated with a Type IIn SN. The second optical peak would then be associated with a 'turn on' event in the SMBH that sees a vast increase in the accretion rate and the luminosity of the BH. This scenario would then explain why the He ii, Bowen, and coronal lines are not seen in the NOT spectrum, and only in the spectra taken after the second peak. However, the probability of observing both a Type IIn SN and an AGN 'turn on' event within just over a year of each other is extremely small given the apparent rarity of extreme 'turn-on' events in AGNs (especially those showing an AT 2019avd-like X-ray outburst) and the expected detection rates for Type IIn SNe (e.g. Feindt et al. 2019), and we therefore disfavour a scenario where AT 2019avd is the chance coincidence of a Type IIn SN and extreme AGN ignition event within roughly one year of each other.

Conclusions
This paper presents an overview of a set of multi-wavelength observations of an exceptional nuclear transient, AT 2019avd, whose main observed features are as follows: 1. eROSITA detected an ultra-soft (kT ∼ 85 eV) X-ray brightening ( 90 times brighter than a previous 3σ upper flux limit) from a previously X-ray-inactive galaxy (Section 2).

AT 2019avd was initially observed on a weekly basis with
Swift XRT/UVOT for 6 weeks following the eROSITA detection. The host had brightened in all UVOT bands by ∼ 1 mag relative to archival GALEX observations, and was observed with 0.2 − 2 keV X-ray flux consistent with the eROSITA detection (Section 2). A further Swift observation ∼5 months after the initial eROSITA detection revealed a brightening by a factor of approximately six in the 0.2 − 2 keV band relative to the eROSITA detection. AT 2019avd therefore shows a net brightening in the 0.2 − 2 keV band by a factor of at least 600 relative to the 3σ upper detection limit derived from an XMM-Newton pointing in 2015. 3. In the 450 days prior to the eROSITA detection, ZTF observed a double-peaked light curve (Section 3). The first optical peak shows rise and decay timescales akin to TDEs and SNe, whilst the rise time of the second peak is more similar to those seen in AGNs. No optical outbursts were detected during ASAS-SN observations over the seven years preceding the initial outburst seen by ZTF. 4. Optical spectroscopic follow-up finds transient He ii emission, Bowen fluorescence lines, and high-ionisation coronal lines ([Fe x] 6375 Å, [Fe xiv] 5303 Å) in the spectra taken after the second optical peak, but not in the spectrum taken 30 days after the first peak. The presence of such a set of lines requires an intense source of soft X-ray emission and extremely high densities. Broad Balmer emission lines were detected in spectra 30 days after the first peak in the ZTF light curve, as well as in all spectra taken in the weeks after the eROSITA detection with FWHM ∼ 1400km s −1 (Section 4).
AT 2019avd thus shows a set of observed features which have never been observed together in the same nuclear transient before, and further complicates the non-trivial task of distinguishing the physical origin of large-amplitude variability seen in galactic nuclei. Whilst a discussion on the potential origins of this transient is presented in Section 5, it is still unclear what has triggered such exotic behaviour. Detailed simulations would be welcome to distinguish between the various possible scenarios. These will be well complimented with future planned observations (Swift, NICER, XMM-Newton) monitoring the late-time evolution of AT 2019avd. Finally, we note that during its eight successive all-sky surveys in the following years, eROSITA will systematically monitor the X-ray variability of AGNs and map out the population of nuclear transients. With this information, we will be able to better understand the extent of the X-ray variability shown by AT 2019avd, and make a more informed judgement on the origin of this transient.