AT 2021loi: A Bowen Fluorescence Flare with a Rebrightening Episode Occurring in a Previously Known AGN

3.1.1. Optical, MIR, and UV

We retrieved point-spread function (PSF)–fit photometry of reference-subtracted images from the ZTF forced photometry service ¹⁶ from MJD = 58,194 to MJD = 59,975 and the Asteroid Terrestrial-impact Last Alert System (ATLAS; Tonry et al. 2018) forced photometry service ¹⁷ from MJD = 57,232 to MJD = 59,975 at the position of AT 2021loi. We find no signs of strong variability in the years preceding the flare (Figure 1).

Zoom In Zoom Out Reset image size

We also obtained optical photometry in the gri bands starting on 2021 July 1 with the Las Cumbres Observatory network of 1 m telescopes (Brown et al. 2013). As there are no available pretransient Las Cumbres images of AT 2021loi, and it has yet to fade, we use Sloan Digital Sky Survey (SDSS; York et al. 2000) reference images obtained on 2006 October 6 for all three bands. We used lcogtsnpipe ¹⁸ (Valenti et al. 2016) in order to process the Las Cumbres data. The pipeline is used to generate the PSF for the Las Cumbres images and the SDSS reference images. After that, the pipeline uses an implementation of the High Order Transform of PSF ANd Template Subtraction (hotpants; Becker 2015) to perform image subtraction. We performed aperture photometry at the source position because we find that, for the Las Cumbres data, PSF fitting produces varying residuals in the image after subtracting the PSF model. There is a noticeable difference between the ZTF and Las Cumbres r-band measurements, where the latter appears to be fainter (Figure 2). This difference is likely due to the difference in the transmission profiles of the two r filters. In Figure 3, we demonstrate this difference using an example optical spectrum of AT 2021loi along with the two r-band filter transmission functions. ^{19 , 20} While the ZTF filter encompasses the observed broad Hα emission line of AT 2021loi, the Las Cumbres filter does not, thus explaining why more flux is observed in ZTF r versus Las Cumbres r.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The location of AT 2021loi was repeatedly observed in the mid-infrared (MIR) regime by the Wide-Field Infrared Survey Explorer (WISE; Wright et al. 2010) and its extensions, NEOWISE (Mainzer et al. 2011) and the NEOWISE reactivation mission survey (NEOWISE-R; Mainzer et al. 2014). The measurements were conducted in both the W1 and W2 bands (3.4 and 4.6 μm, respectively). We queried the NASA/IPAC Infrared Science Archive ²¹ for MIR detections within 5″ from the ZTF-determined position of AT 2021loi. This position was visited roughly twice a year by NEOWISE-R (changes in the scanning pattern may result in more observations like in this case, where there are two measurements within a few days in 2016 and 2021). We rebinned the measurements available for each such visit (using a weighted mean) into a single representative measurement per year per band. The resulting MIR light curve ²² is shown in Figure 1 (top panel). Prior to the optical outburst, the MIR flux shows limited variability of up to a few tenths of a magnitude. Fitting a linear magnitude-versus-time model to the preflare data (MJD ≤ 59,210) yields a rate of change of 0.067 ± 0.013 mag yr⁻¹. The first MIR visit to show a dramatic flux increase (>1.5 mag in both MIR bands) is indeed the first measurement taken after the optical outburst on 2021 July 22, i.e., ∼70 days after the initial optical detection of the transient and ∼30 days after the peak optical emission.

The earliest available WISE-based measurements associated with the source hosting AT 2021loi (obtained through the AllWISE catalog; Cutri et al. 2021), using coadded images taken during 2010 December 11 to 2011 January 17, yield a MIR color of W1 − W2 = 0.08 mag. The average MIR color during the period 2014 January 18 to 2020 December 31 probed by NEOWISE-R, i.e., the ∼7 yr prior to the optical outburst (part of it, ∼4.5 PSF, is shown in Figure 1), is W1 − W2 = 0.02 mag. These MIR colors are consistent with the emission being related to an AGN, typically defined at W1 − W2 ≳ 0.06 mag in the AB system (e.g., Stern et al. 2012; Assef et al. 2018).

We initiated near-UV (NUV) follow-up observations (PI: L. Makrygianni) with the Neil Gehrels Swift Observatory (Gehrels et al. 2004) Optical/UV Telescope (UVOT; Roming et al. 2005). We obtained 23 Swift epochs from 2021 June 21 to 2022 October 11 in all six UVOT filters (uvw2, uvm2, uvw1, U, B, and V). The UVOT photometry was extracted using the HEASARC pipeline and the standard analysis task uvotsource. We used a circular aperture with a radius of 75 for all UV photometry, partially motivated by our desire to compare the postflare measurements to archival Galaxy Evolution Explorer (GALEX) photometry (see below). For the longer-wavelength bands, we use smaller apertures with radii of 5″ (for the U and B bands) and 4″ (for the V band). ²³ In all cases, a nearby circular region with a radius of 75 was used for background extraction. We used the archival NUV flux measurement from GALEX observations taken on 2006 November 9, obtained with a 75 radius aperture, yielding m_NUV = 20.05 ± 0.14 mag (Seibert et al. 2012).

Preflare photometry of WISEA J010039.62+394230.3 is presented in Table 1. For AT 2021loi, all photometry was corrected for Milky Way extinction using E(B − V) = 0.045 mag (Schlafly & Finkbeiner 2011), ²⁴ the Cardelli et al. (1989) extinction law, and R_V = 3.1. Our photometry is presented in Figure 1, the bottom panel of Figure 2, and Table 2.

Table 1. Preflare Photometry of WISEA J010039.62+394230.3

Filter	Magnitude	Error	Source
FUV	19.86	0.21	GALEX
NUV	20.05	0.14	GALEX
u	18.55	0.02	SDSS
g	17.91	0.01	SDSS
r	17.43	0.01	SDSS
i	17.00	0.01	SDSS
z	16.85	0.01	SDSS
W1	16.34	0.03	WISE
W2	16.26	0.03	WISE

Note. Photometry is in the AB system. No extinction correction applied.

Download table as:  ASCIITypeset image

Table 2. Photometry of AT 2021loi

MJD	Filter	Magnitude	Error	Source
59,341.49	r	18.99	0.15	ZTF
59,341.49	r	18.73	0.14	ZTF
59,342.49	r	18.63	0.08	ZTF
59,342.49	r	18.59	0.09	ZTF
59,344.49	r	18.45	0.08	ZTF
59,344.49	r	18.63	0.11	ZTF
59,345.48	r	18.45	0.08	ZTF
59,345.49	r	18.36	0.08	ZTF
59,345.49	r	18.38	0.08	ZTF
...	...	...	...	...
59,385.39	r	16.89	0.02	ZTF
59,387.44	r	16.82	0.01	ZTF
59,391.41	r	16.83	0.02	ZTF
59,393.45	r	16.80	0.01	ZTF
59,399.39	r	16.83	0.02	ZTF
59,401.44	r	16.86	0.01	ZTF
59,403.43	r	16.77	0.01	ZTF
59,405.43	r	16.82	0.01	ZTF
...	...	...	...	...
59,428.20	g	17.64	0.01	Las Cumbres
59,444.40	g	17.33	0.03	Las Cumbres
59,450.10	g	17.24	0.02	Las Cumbres
59,461.20	g	17.33	0.02	Las Cumbres
59,469.40	g	17.69	0.01	Las Cumbres

Note. Photometry is in the AB system. This table is published in its entirety in machine-readable format. A portion is shown here for guidance regarding its form and content.

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

3.1.2. X-Ray and Radio

We retrieved an archival spectrum of WISEA J010039.62+394230.3, the host galaxy of AT 2021loi, from the fifth data release of the LAMOST survey (LAMOST/DR5; Zhao et al. 2012) obtained on 2013 December 22. ²⁷ The LAMOST spectrum covers the range 3500–9000 Å with a resolution of R = 1800. The archival spectrum shows AGN features, in particular prominent broad Balmer emission lines and strong narrow [O iii] λ5007.

The classification spectrum of AT 2021loi (Graham et al. 2021) was obtained with the LRIS on Keck I, which covers the wavelength range between 3400 and 10300 Å with a resolution of R = 600–1000, on 2021 June 7. This spectrum is shown in Figure 4, together with the archival LAMOST spectrum. In addition, Figure 14 shows the two spectra normalized in a way that would yield a common integrated [O iii] λ5007 line flux level to further emphasize the changes in the continuum and broad-line emission (see Appendix A).

Zoom In Zoom Out Reset image size

We initiated a spectroscopic monitoring campaign for AT 2021loi using the FLOYDS spectrograph (Sand et al. 2011) mounted on the robotic 2 m telescope at Haleakalā Hawaii (which is also part of the Las Cumbres Observatory network). During the 16 months of spectroscopic monitoring, we obtained 20 spectra. FLOYDS covers the 3500–10000 Å range in a single exposure by capturing two spectral orders simultaneously, yielding a spectral resolution of R ∼ 400. The exposure time of the spectra varies between 1500 and 1800 s, and we have used a slit width of 2″.

The Las Cumbres Observatory spectra were reduced with a custom data reduction pipeline ²⁸ built with the iraf-free python-based aspired tool kit ²⁹ (Lam & Smith 2022; Lam et al. 2023). The spectral images were each cropped into a red and a blue image. Standard data reduction procedures were applied to trace and then optimally extract the spectral information using the Horne (1986) algorithm. Wavelength calibration was performed using the built-in calibrator powered by rascal ³⁰ (Veitch-Michaelis & Lam 2020, 2021). Standard stars from the same night were used for flux calibration when available; otherwise, the ones observed closest in time to the science observations were used. Finally, atmospheric extinction and telluric absorption were removed. We present all spectra in Figure 5.

Zoom In Zoom Out Reset image size

Basic details regarding all of the optical spectra of AT 2021loi used in this work are provided in Table 4 (in Appendix B), along with key spectral measurements (see Section 4.2).

The optical brightness of AT 2021loi increased by a factor of ≈4 between 2021 May 7 (first detection by ZTF) and 2021 June 22, from the initial M_r = −19.04 ± 0.15 mag to peak magnitudes of M_r = −21.08 ± 0.03 and M_g = −20.98 ± 0.04 mag (all reference-subtracted). The latter corresponds to ν L_ν (g) ≃ 7.8 × 10⁴³ erg s⁻¹. Following the peak, the optical light curve shows a slow, steady decline until 2022 mid-June (Figure 2). We fit the decline phase between the main peak and the rebrightening phase (59,386 < MJD < 59,750) with a power law of the form . We find a power-law slope of α ≃ −0.48 if we fix the value of t₀ to the discovery time of the optical flare, and α ≃ −0.78 if we let t₀ (which yields t₀ = 59253 ± 40) be a free parameter, with formal uncertainties of ≈0.01 and 0.1, respectively. The reduced χ² in both cases is ≈25, which shows that a power-law model does not provide a good fit for our data. In any case, the declining phase and the best-fit power laws are much shallower than the α = −5/3 associated with TDEs (e.g., Rees 1988), and, as we found, a power law does not provide a good fit to the data (especially for the r band; Figure 2). After 2022 mid-June, AT 2021loi experiences a rebrightening and shows a second peak in the ZTF r band at M_r = −20.59 ± 0.05 mag on 2022 July 15, about 390 days after the main peak. This second peak is also seen in the ZTF g and ATLAS o and c bands. An observational gap did not allow the Las Cumbres gri bands to cover the rebrightening. The decline after the second peak, although faster (α ≃ −0.9, r band), is still very different from that of a TDE.

We compare the uvm2 UVOT observations to the archival GALEX (NUV; λ_eff = 2304 Å) data, as uvm2 (λ_eff = 2245 Å) is the most relevant UVOT filter for such a comparison. In the first UVOT observation, which also resulted in the highest NUV flux we obtained during our monitoring campaign, we measure a uvm2 absolute magnitude of M_uvm2 = −21.06 ± 0.07 mag, which corresponds to ν L ν(NUV) = 2.3 × 10⁴⁴ erg s⁻¹. This is a factor of ∼20 increase from the archival GALEX detection, where M_NUV = −17.84 ± 0.14 mag translates to ν L_ν (UV) = 1.1 × 10⁴³ erg s⁻¹. Following this measurement, the NUV flux shows a gradual decline before the rebrightening mentioned above. The monochromatic NUV luminosity we measure ∼14 months after the initial transient detection (and 1 yr after the initial Swift detection) is ν L_ν ≃ 5.7 × 10⁴³ erg s⁻¹, i.e., a drop of ∼75% but still higher than the archival GALEX measurement by a factor of ∼5.

As already mentioned in Section 3.1.1, the MIR light curves (Figure 1, top panel) show limited variability prior to the optical flare and a noticeable flux increase in the first MIR measurement that follows the optical flare. This enhanced MIR emission is separated by at most ∼64 days from the detection of the optical outburst, or ∼30 days compared with the peak optical emission (both in the rest frame). These timescales are mere upper limits on the real delay, driven by the (low) cadence of the NEOWISE-R monitoring. To put these timescales in context, we note that the inner edge of the dusty torus, i.e., where the temperature reaches the sublimation threshold, is expected to be ∼100–300 lt-days away from the central engine based on our near-peak L_bol estimate for AT 2021loi. This range is driven by the range in possible dust grains (see Equations (2) and (3) in Mor et al. 2009) and is supported by NIR reverberation mapping campaigns (e.g., Suganuma et al. 2006; Koshida et al. 2014).

We can thus infer that, given the data in hand, the increase in MIR emission is broadly consistent with the reprocessing of the (sharp) rise in UV/optical radiation by circumnuclear dust in a way that is not markedly different from what is seen in normal (persistent) AGNs. From the NEOWISE-R light curves, there is also some variability in the W1 − W2 color, which changes from around 0.02 mag prior to the outburst to ∼−0.24 mag during the transient state. While the former, preoutburst MIR color is close to the cut used to identify AGNs (i.e., ≳0.06 mag in the AB system), the latter postflare color is, in fact, less consistent with what is found for persistent AGNs. This may not be surprising given the possibility that the emission from either the exceptionally UV-bright flare and/or the hottest, pure graphite part of the torus is enhancing the shorter-wavelength W1 band (i.e., relative to normal AGNs).

There is no ROSAT counterpart to WISEA J010039.62+394230.3 in the all-sky survey conducted during 1990–1991. From this, we infer an archival 3σ X-ray upper limit of F(0.1–2.4 keV) ≲ 10⁻¹³ erg cm⁻² s⁻¹ (Boller et al. 2016), which translates to L(0.1–2.4 keV) ≲ 2 × 10⁴² erg s⁻¹ but also L(2–10 keV) ≲ 2.5 × 10⁴² erg s⁻¹. In order to compare this archival ROSAT upper limit with the postflare XRT upper limits described in Section 3.1.2, we utilize the observed relation between X-ray and NUV emission in persistent AGNs, which is commonly quantified through the anticorrelation between the optical–to–X-ray spectral slope, , and the monochromatic NUV luminosity (see, e.g., Just et al. 2007; Lusso & Risaliti 2016; Nanni et al. 2017, and references therein).

Specifically, we rely on the relation derived in Equation (2) of Nanni et al. (2017), of the form . From the (extinction-corrected) archival GALEX NUV magnitude of AT 2021loi, m_NUV = 19.65 mag, we derive an archival NUV luminosity of L_ν (2500 Å) = 8.5 × 10²⁷ erg s⁻¹ Hz⁻¹. Plugging this into the Nanni et al. (2017) relation, the expected spectral slope is α_ox = −1.12, which in turn yields an intrinsic monochromatic X-ray luminosity at 2 keV of ν L_ν (2 keV) = 5.6 × 10⁴² erg s⁻¹. Extrapolating the 2 keV fluxes to the 2–10 keV range assuming a photon index of Γ = 1.8, we derive an expected X-ray luminosity of L(2–10 keV) ≃ 6 × 10⁴² erg s⁻¹, which is a factor of ∼3 higher than the upper limit from ROSAT. We stress that this estimate for the expected X-ray emission of AT 2021loi carries significant uncertainties. In particular, a 1σ uncertainty of ≈±0.2 on α_ox at any given L_ν (2500 Å) (consistent with what is reported in Nanni et al. 2017) would translate to an uncertainty of almost 0.6 dex on the expected X-ray luminosity we derive through this approach (i.e., a factor of ∼4). ³¹ Following the same approach for the near-peak Swift NUV measurement of m_NUV = 16.82 mag yields α_ox ≃ −1.33 and L(2 keV) = 3.1 × 10⁴³ erg s⁻¹. Extrapolating the 2 keV flux to the 2–10 keV range using Γ = 1.8, we derive an expected X-ray luminosity of L(2–10 keV) = 5.0 × 10⁴³ erg s⁻¹. The upper limits from Swift/XRT correspond to L(2–10 keV) < (2.4–7.8) × 10⁴² erg s⁻¹. At face value, we find that the source would have been detected in X-rays based on the near-peak NUV emission and if the broadband spectral energy distribution (SED) of AT 2021loi would have followed that of typical, persistent broad-line (unobscured) AGNs. However, the large uncertainties on this prediction limit our ability to draw definite conclusions regarding the SED of AT 2021loi based on its nondetection in X-rays (see Section 3.1.2). We discuss this further in Section 5.1.

Figure 6 shows the color evolution of the UV-bright AT 2021loi compared to that of the BFF AT 2017bgt (Trakhtenbrot et al. 2019b) and the BF TDE AT 2018dyb (Leloudas et al. 2019). The color of AT 2021loi shows limited evolution similar to that of AT 2017bgt. Also, AT 2021loi is redder compared to AT 2018dyb.

Zoom In Zoom Out Reset image size

We fit a blackbody to the ZTF (g and r), ATLAS (c and o), LCO (g and i), and all Swift photometry for each epoch in which we have Swift data (linearly interpolating neighboring optical epochs) using the blackbody fitting tool superbol ³² (Nicholl 2018). We then calculate the bolometric luminosity by integrating each observed SED, approximating the missing flux outside the observed bands, and using the blackbody fits. Our results are presented in Figure 7. The peak blackbody temperature (i.e., at UV peak but also the peak temperature) is (21.1 ± 1.5) × 10³ K. After ∼43 days (in the rest frame), it declines to (15.8 ± 1.2) × 10³ K, eventually reaching (14.4 ± 1.8) × 10³ K. At these temperatures, we are fully sampling the blackbody continuum, especially with the UV data (Arcavi 2022).

Zoom In Zoom Out Reset image size

Compared to a large compilation of TDEs (van Velzen et al. 2020), AT 2021loi is on the low-temperature end. Moreover, AT 2021loi has a steady slow decline, unlike several TDEs that show a decline followed by a rise in temperature. In terms of inferred radius, AT 2021loi has larger blackbody radii (1.8–2.6 × 10¹⁵ cm) compared to the vast majority of the van Velzen et al. (2020) TDEs (typically ≲6 × 10¹⁴ cm) and shows very little evolution. In contrast, TDE radii decline a few weeks after peak. The peak bolometric (blackbody) luminosity of AT 2021loi is , which is higher than the majority of the TDEs presented in van Velzen et al. (2020). In comparison, the estimated temperature at the peak of the OGLE 17aaj BFF is 4 × 10⁴ K, which is much higher than that of AT 2021loi. (Gromadzki et al. 2019).

We use the pyqsofit fitting procedure ³³ (Guo et al. 2018) to model the archival LAMOST spectrum, the first flare Keck spectrum, and the FLOYDS spectra. pyqsofit fits and decomposes the continuum emission (with AGN and host galaxy components) and fits the broad and narrow emission lines. The continuum is fitted as a simple power law, and we add a polynomial component when fitting fails without it.

All narrow emission lines are modeled with a single Gaussian, whereas the broad lines are modeled with either one or two Gaussians. We initially tried to model all broad emission lines with a single (broad) Gaussian, which works for simpler and/or weaker broad line profiles (e.g., Hδ). However, we found that some line profiles (e.g., Hβ) are more complex, and two broad components are required to achieve a satisfactory fit. These multiple broad Gaussians are not constrained in terms of offset velocity, peak intensity, or any other parameter. We stress that the choice to use either one or two broad Gaussian profiles was applied consistently to all spectra (i.e., all epochs). This fitting approach is in line with common practice in fitting the (sometimes complex) broad-line emission profiles in normal, persistent AGNs (see, e.g., Greene & Ho 2005; Shen et al. 2011; Trakhtenbrot & Netzer 2012; Mejía-Restrepo et al. 2018; Rakshit et al. 2020). We fit the 4500–5100 Å region, which covers the entire N iii λ4640 and He ii λ4686 region, as well as Hβ and the [O iii] λ λ4959, 5007 lines. In addition, we fit the Hα region, including the [N ii] λλ6548, 6584 and [S ii] λλ6718, 6732 doublets.

We first fit the archival spectrum from LAMOST with all of the available components within pyqsofit (i.e., the host component, power-law continuum, Balmer continuum, and iron emission). The fit results are presented in Figure 8. This spectrum shows broad Balmer emission lines with FWHM(Hα) and , thus clearly capturing a broad-line, unobscured AGN. In addition, it also shows strong narrow [O iii] λ5007 emission. The key diagnostic narrow line ratios are and , which would correspond to the composite region in the BPT (Baldwin et al. 1981); i.e., both star formation and AGN activity are responsible for producing the narrow lines in this galaxy (Kewley et al. 2006). Table 4 (in Appendix B) lists some key spectral measurements resulting from our analysis.

Zoom In Zoom Out Reset image size

Based on relations between luminosity and BLR size derived from reverberation mapping campaigns, the LAMOST spectrum yields a BLR size of R_BLR ≃ 2.9 and 5.4 lt-days using the prescriptions of Trakhtenbrot & Netzer (2012) and Bentz et al. (2013), respectively. We use the L(Hα) luminosity and the respective FWHM(Hα) to calculate the SMBH mass following Equation (2) in Mejía-Restrepo et al. (2022) and find M_BH = 1.0 × 10⁷ M_⊙. The uncertainties of such mass estimates are of the order of 0.3–0.5 dex, including uncertainties on both the R_BLR–L₅₁₀₀ relation and the so-called virial factor (see, e.g., Shen 2013; Mejía-Restrepo et al. 2022, and references therein). We estimate the bolometric luminosity (L_bol) from the host galaxy–subtracted monochromatic luminosity at rest frame 5100 Å (3.2 × 10⁴² erg s⁻¹) and a bolometric correction of f_bol(5100 Å) ≡ L_bol/L₅₁₀₀ = 9.26 (e.g., Shen et al. 2008; MacLeod et al. 2010). This estimate has an uncertainty of order ∼0.3 dex due to the uncertainty on the bolometric correction (e.g., Runnoe et al. 2012; Duras et al. 2020, and references therein). From L_bol and M_BH, we derive an estimate for the preflare Eddington ratio of L/L_Edd ≈ 0.02 (with large systematic uncertainties stemming from both L_bol and M_BH). Despite the large uncertainties associated with this Eddington ratio estimate, we conclude that the source in its preflare state was well within the sub-Eddington accretion regime.

The earliest spectrum after the flare, which is also the one nearest the optical peak time, is the Keck/LRIS spectrum obtained 31 days after discovery. Since the Keck spectrum has a lower resolution compared to LAMOST, we use the host component as measured for the LAMOST spectrum by pyqsofit to apply host subtraction to the Keck spectrum. After that, we process the Keck spectrum with pyqsofit (Figure 9) without applying host decomposition but only continuum, iron, and emission line fitting.

Zoom In Zoom Out Reset image size

For the Keck spectrum, we find L₅₁₀₀ = 3.0 × 10⁴³ erg s⁻¹. From this, along with the M_BH estimated from the LAMOST spectrum and the same bolometric correction as mentioned above, we derive an Eddington ratio at the epoch of the Keck spectrum (i.e., near peak) of L/L_Edd ≈ 0.2. This is an increase by a factor of ∼10 compared to the preflare phase, but it is still well within the sub-Eddington regime. The UV peak luminosity found in Section 4.1, combined with a UV bolometric correction of f_bol(UV) = 3.5 (see Kaspi et al. 2000; Netzer et al. 2016), yields a bolometric luminosity of 7.7 × 10⁴⁴ erg s⁻¹, which in turn gives L/L_Edd ≃ 0.5. The higher value of the UV-based L_bol estimate is a result of the UV-bright nature of the transient. Although more than twice as high as what is found from L₅₁₀₀, the UV-based L/L_Edd estimate is still within the sub-Eddington regime, although we note that the large systematic uncertainty on this estimate means it is formally consistent with L/L_Edd ≈ 1. The bolometric luminosity generated from the blackbody fit in the previous section is 4.8 × 10⁴⁴ erg s⁻¹, which is lower than the one estimated from the UV flux, resulting in an even lower Eddington ratio. In any case, notwithstanding the range of L/L_Edd estimates and the significant systematic uncertainties associated with them, we can still conclude that the data in hand do not support a scenario in which the AT 2021loi flare approached the Eddington limit of the accreting SMBH.

In Figure 10, we plot the LAMOST and Keck spectra divided by the power-law continuum and the iron component as fitted by pyqsofit in order to see the difference in the strength of the emission lines with respect to the continuum and iron complex. We see that the relative strength of the feature around 4680 Å to Hβ has increased. In order to quantify the changes in the He ii λ4686+N iii λ4640 region between the LAMOST preflare and Keck (near optical peak) spectra, we perform the following procedure on both spectra. We start by fitting both the He ii λ4686 and N iii λ4640 complex with two Gaussian components for each transition, forcing the widths to match that of the Hβ broad component. Each of these broad components is allowed to have an FWHM in the range of 2000–10,000 km s⁻¹. In the case of LAMOST, although it is possible to fit the emission feature around 4680 Å (FWHM > 6000 km s⁻¹) as either one or two components (i.e., only He ii λ4686 or He ii λ4686 blended with N iii λ4640), it is not clear if the feature indeed consists of two peaks. In the case of the Keck spectrum, we see two peaks more clearly, similar to previously identified BFFs. We list the flux (measured with pyqsofit) ratios between the emission lines of interest in Table 3.

Zoom In Zoom Out Reset image size

We look for differences in the integrated flux density of the total He ii λ4686 and N iii λ4640 region relative to the integrated flux density of Hβ. We do this by integrating the flux density after host, continuum, and iron subtraction between 4580 and 4750 Å (the region that covers He ii λ4686 and N iii λ4640) and between 4800 and 4900 Å (the region that covers Hβ). We use two approaches to quantify this ratio. In the first case, we integrate the observed flux in the region with no emission line fitting. We find that the ratio between the integrated fluxes in the two regions is 0.53 for the LAMOST spectrum and 1.12 for the Keck spectrum, meaning that the ratio increased by a factor of ≈2. Using the flux as extracted from the Gaussians fit to the emission lines, we find a similar line ratio increase from F(He II λ4686 + N iii λ4640)/F(Hβ) ∼ 0.58 in the LAMOST spectrum to F(He II λ4686 + N iii λ4640)/F(Hβ) ∼ 1.09 in the Keck spectrum.

We also find that the ratio between the integrated flux without fitting the Hδ spectral region (4050–4150 Å) to that of Hβ increased from 0.22 ± 0.01 in the LAMOST spectrum to 0.35 ± 0.02 in the Keck spectrum. The flux ratio of F(Hδ)/F(Hβ) from Gaussian fitting increased from 0.24 ± 0.02 in the LAMOST spectrum to 0.36 ± 0.03 in the Keck one. On the other hand, the flux ratio of Hγ to Hβ remained consistent with F(Hγ)/F(Hβ) = 0.51 ± 0.02 for LAMOST and 0.48 ± 0.03 for Keck (without fitting) and at F(Hγ)/F(Hβ) = 0.48 ± 0.05 for LAMOST and 0.41 ± 0.03 for Keck (with fitting). We find, though, that the F(Hγ)/F(Hβ) ratio remained constant, whereas F(Hδ)/F(Hβ) significantly increased by ∼1.5, providing more support for the BF mechanism, as the Hδ feature is likely blended with N iii emission at λ λ4097, 4104 (Netzer et al. 1985).

Although the spectrum appears to have a strong iron bump around 4500–4600 Å after the pyqsofit fitting, we find that He ii at 4687 Å was present in the archival LAMOST spectrum. Typically, in AGNs, F(He II λ4686)/F(Hβ) ≤ 0.05 (Vanden Berk et al. 2001). For the LAMOST spectrum, the intensity ratio F(He II λ4686 + N iii λ4640)/F(Hβ) is ∼0.5, which is already 10 times higher than typical AGNs. For the Keck spectrum, where the N iii λ4640 and He ii λ4686 lines can be clearly separated, we find that F(He II λ4686)/F(Hβ) = 0.71 ± 0.09, F(N iii λ4640)/F(Hβ) = 0.47 ± 0.07, and F(N iii λ4640)/F(He II λ4686) = 0.66 ± 0.11. Given the F(He II λ4686)/F(Hβ) and F(N iii λ4640)/F(Hβ) ratios seen in AT 2021loi and the predicted relative intensities for AGNs as suggested by Netzer et al. (1985), the gas producing the BF must have a very high density (n_H > 10^9.5 cm⁻³) and higher N and O abundances relative to the cosmic ones. We find that the [He i] λ5875 flux relative to [O iii] λ5007 increased from 1.17 in the LAMOST spectrum to 2.53 in the Keck spectrum; however, the flux ratio to Hβ has decreased from 0.27 in LAMOST to 0.13 in Keck.

To trace the evolution of the key spectral features of AT 2021loi, we apply the pyqsofit fitting to the lower-resolution Las Cumbres/FLOYDS spectra using the same approach as we did for the Keck spectrum (i.e., subtracting the host component as extracted from the archival LAMOST spectrum). We use the same two approaches as before, i.e., first integrating over the observed flux in the region with no emission line fitting and then finding the integrated fluxes after Gaussian fitting for He ii λ4686 and N iii λ4640. In the top panel of Figure 2, we plot the ratio of the total He ii λ4686 and N iii λ4640 integrated flux to the respective integrated flux of the broad Hβ as a function of rest-frame time with respect to MJD = 59,381. The flux is host-, continuum-, and iron-subtracted in both cases (i.e., with and without fitting).

The top panel of Figure 2 shows that there is enhanced emission in the region around 4680 Å (relative to Hβ) consistent with BF brightening. This enhanced emission lasts for at least 27 rest-frame days after the Keck spectrum and 58 rest-frame days after the initial transient detection. The ratio of the flux of the BF features to Hβ appears to become weaker than the respective ratio in the LAMOST spectrum after 35 rest-frame days from the Keck spectrum, but it is evident that the emission around 4680 Å still exists. The lower resolution of the FLOYDS spectra compared to LAMOST and Keck may underpredict our measured ratios due to blending with Fe features.

We now turn to investigating the high-ionization iron lines. These lines have been seen in the past in AGNs, and for Fe x λ6375, it has been found that the maximum F([Fe X] λ6375)/F([O iii] λ5007) ratio is ∼0.24 (Nagao et al. 2000). The LAMOST spectrum of AT 2021loi shows an F([Fe X] λ6375/F([O iii] λ5007) ratio of ∼0.6. In the Keck spectrum, we find this value to be 2.06.

In the LAMOST spectrum, the FWHM of Fe x λ6375 is ∼1900 km s⁻¹ lower than that of Hβ, Hα, He i (∼3000 km s⁻¹), and the BF lines (He ii λ4686+N iii λ4640 ∼ 7000 km s⁻¹). For the Keck spectrum, the Fe x λ6375 emission line’s FWHM is ∼2000 km s⁻¹. We note that this is surprising for two reasons: (1) the line seems almost unchanged between the two spectra, and (2) in both cases, the line is consistent with a BLR, whereas in Wang et al. (2012), coronal emitters are found to have Fe x λ6375 with an FWHM between 200 and 1000 km s⁻¹, i.e., consistent with a narrow-line region.

Here we refer to the basic properties of AT 2021loi that suggest it is part of the BFF class and compare it to previously reported BFFs. We focus on the three events used by Trakhtenbrot et al. (2019b) to identify this class, namely, AT 2017bgt (first reported in Trakhtenbrot et al. 2019b), F01007–2237 (Tadhunter et al. 2017), and OGLE 17aaj (Gromadzki et al. 2019).

After the initial increase in brightness, AT 2021loi shows a very slow decline until rebrightening ∼13 months after the initial detection (see Figures 1 and 2 and Section 4.1). The slow decline is comparable to that of the BFFs presented in Trakhtenbrot et al. (2019b); i.e., after more than a year, the optical transient emission still persists. In Figure 11, we present the r-band light curve of AT 2021loi along with those of the Trakhtenbrot et al. (2019b) BFFs. The rebrightening and second peak are very clear for AT 2021loi. There is evidence for a rebrightening in F01004–2237 too, which we point out explicitly here for the first time. However, in this case, the photometric cadence is rather low, and the host subtraction is more challenging. The most up-to-date ZTF and ATLAS photometry for AT 2017bgt is also indicative of another bump, roughly 14 months after the initial detection (and 13 months after the first peak); however, this possible rebrightening is even less robust than the one in F01004–2237. ³⁴ A late-time rebrightening of the (optical) emission may thus emerge as another intriguing common property of BFFs, although the evidence in hand is still limited.

Zoom In Zoom Out Reset image size

The optical spectroscopy of AT 2021loi shows a prominent double-peaked emission feature around 4680 Å, consistent with blended He ii λ4686 and N iii λ4640 emission. The spectra also show other BF features, such as O iii lines at 3133 and 3444 Å. A strong feature around 4680 Å is also seen in the preflare spectrum of AT 2021loi. To our knowledge, only one of the three BFFs studied by Trakhtenbrot et al. (2019b) has preflare optical spectroscopy: the HST/STIS spectrum of F01004–2237 (Tadhunter et al. 2017). That spectrum (Figure 12) also shows preexisting N iii λ4640 and He ii λ4686 features that became stronger following the optical flare. Tadhunter et al. (2017) attributed the preexisting N iii λ4640 and He ii λ4686 BF emission to WN Wolf–Rayet stars in the host galaxy, interpreting the flare and He ii λ4686 enhancement as a result of a TDE in F01004–2237.

The preflare F(He ii λ4686 + N iii λ4640)/F(Hβ) ratio of AT 2021loi (0.58) is already higher than what is expected for AGNs (∼0.05; see, e.g., Vanden Berk et al. 2001) and similar to that measured for F01004–2237 (0.49 preflare). For AT 2021loi, this ratio increases to ∼1.1 near the optical peak, which is slightly lower than what was found in the spectrum of F01004–2237 during its flare (∼1.8) and slightly higher than what was found for OGLE 17aaj (∼0.9; Gromadzki et al. 2019).

Zoom In Zoom Out Reset image size

In Figure 13, we compare the near-peak spectrum of AT 2021loi with those of the BFFs discussed in Trakhtenbrot et al. (2019b). In addition to the ∼4680 Å feature, AT 2021loi also shows a broad and possibly double-peaked feature near Hδ, similar to what is seen for OGLE 17aaj (Gromadzki et al. 2019). As previously mentioned, the Hδ emission line coincides with the N iii BF features expected at 4097 and 4104 Å. These two lines are expected to have relative intensities to N iii λ4640 of 0.117 and 0.082, respectively (Netzer et al. 1985). Given the limited resolution of the AT 2021loi spectra in hand, it is not possible to properly decompose these lines from Hδ, but we suggest that since the F(Hδ)/F(Hβ) ratio increased more from pre- to during-flare spectra than the F(Hγ)/F(Hβ) ratio, the Hδ spectral feature is affected by intensified N iii emission.

Zoom In Zoom Out Reset image size

Object AT 2021loi also shows signs of enhanced emission (between pre- and during-flare spectra) in some higher-ionization coronal lines, specifically Fe x λ6375. This line was also found in the BF nuclear transient AT 2019avd (Malyali et al. 2021). In that case, the presence of coronal emission lines was argued to require and be associated with the intense soft X-ray radiation seen in that event. The initial discovery of the X-ray flare by e-ROSITA measured a 0.2–2 keV luminosity of 2.7 × 10⁴² erg s⁻¹, which is 90 times brighter than the preflare upper limits. The X-ray emission of AT 2019avd remained rather steady for 4 months, after which it further increased (L(0.2–2 keV) = 1.9 × 10⁴³ erg s⁻¹). For AT 2021loi, there is no X-ray emission detected by our Swift/XRT monitoring, with 3σ upper limits in the range (1.5–5.8) × 10⁴² erg s⁻¹ at 2–10 keV. While there is no X-ray emission detected, we note that the expected X-ray luminosity for AT 2021loi at 2keV based on the NUV flux (ν L_ν = ∼ 2.2 × 10⁴³ erg s⁻¹; see Section 4.1) is higher than what was detected in AT 2017bgt (L(2–10 keV) = 1.2 × 10⁴³ erg s⁻¹) and AT 2019avd (2.7 × 10⁴² erg s⁻¹).

Thus, the X-ray emission from AT 2021loi and the AGN in its preflare state is intrinsically weaker than what is seen in normal, persistently accreting AGNs (taking into account the ≈0.6 dex uncertainty on the expected X-ray emission; see above); considerably weaker than what is seen in other BFFs; and/or delayed compared with what is seen in other BFFs. It is highly unlikely that intrinsically luminous X-ray radiation is obscured by (dusty) gas along our line of sight, given the prominent blue continuum and broad-line emission. Finally, we cannot rule out a more complex scenario in which the X-ray-emitting corona was disrupted by the UV flare, as suggested in at least one other AGN-related UV-luminous transient (Ricci et al. 2020). At any rate, the X-ray nondetection following the UV-luminous flare in AT 2021loi does not follow what is seen in normal AGNs.

Compared to the archival GALEX measurement of its host, AT 2021loi is associated with increased UV emission (by a factor of ∼20). The preflare optical spectrum clearly shows the presence of an actively accreting SMBH in the nucleus of the host galaxy. The rise of the UV emission, peaking at ν L_ν ≃ 2.3 × 10⁴⁴ erg s⁻¹, combined with the simultaneous and significant brightening of the Bowen features, strongly suggests that the AT 2021loi BF features are driven by the accretion flow onto the SMBH.

We can use the observed and modeled SEDs of AT 2021loi to assess whether there is indeed sufficient EUV emission to account for the BF features. First, by integrating the best-fit (17.2 × 10³ K) blackbody SED of the flare near peak, we find that only ∼10⁻¹² of the total luminosity is emitted below 228 Å (the range required to ionize He ii), corresponding to a luminosity of the order of 10³² erg s⁻¹. The observed Bowen lines near peak, however, have respective (He ii λ4686 and N iii λ4640) luminosities of order 10⁴¹ erg s⁻¹. Thus, another ionizing source outside the measured blackbody continuum might be required to account for the BF emission. This is similar to what was found for the BF-emitting TDE AT 2018dyb (Leloudas et al. 2019).

Second, we can use a (simplified) AGN SED to assess the radiative (EUV) energetics, scaling to either the pre- or during-flare measured luminosities of AT 2021loi. We use the Marconi et al. (2004) SED, scaled to either or 44.46 for the pre- or during-flare states, respectively. The integrated EUV luminosities, i.e., over λ < 228 Å, are 7.4 × 10⁴² and 5.3 × 10⁴³ erg s⁻¹ for the pre- and during-flare near-peak states, respectively. At face value, it appears that the EUV emission from the AGN is sufficient to account for the energetics of the BF features (≳10⁴¹ erg s⁻¹), even before the optical/UV flare. However, given that the (optical) luminosity of the preexisting AGN in this system is not exceptionally high, we recall that BF lines are not ubiquitous in persistent AGNs (e.g., Netzer et al. 1985; Vanden Berk et al. 2001), suggesting that other considerations apart from pure (radiative, EUV) energetics are at play.

We next compare AT 2021loi to the growing sample of TDEs detected in the optical/UV regime (see van Velzen et al. 2020 and Gezari 2021 for recent reviews). The rise time of AT 2021loi, about a month, is similar to what is observed in optical/UV TDEs. However, the decline of the optical emission in AT 2021loi, which can be described by a power law with an index shallower than α ≃ −0.8 (second decline α ≃ −0.9), is much slower than what is seen in optical/UV TDEs, which are typically compared to α ≃ −5/3. This is also apparent in Figure 11, where we compare the optical/UV light curve of AT 2021loi with that of the TDE PS1-10jh (Gezari et al. 2012, 2015). The TDE light curve fades within a few months, whereas AT 2021loi lasts for more than 1 yr and also presents a secondary brightening. If we fit the postpeak decline of AT 2021loi with a t^−5/3 power law, the required disruption time is MJD = 59,031 ± 11 (i.e., 2020 June–July), which is approximately 1 yr before the initial ZTF detection of the flare. While the tidal disruptions of giant stars are indeed expected to have longer timescales (e.g., MacLeod et al. 2012b), this would also imply longer rise times (i.e., months instead of weeks), which is not the case with AT 2021loi. We conclude that the optical/UV light curve of AT 2021loi is markedly different than what is seen in (optical/UV) TDEs. We conclude that the overall shape of AT 2021loi is very different from what we know so far for TDEs.

In terms of spectral properties, in many cases, optical/UV TDEs also show broad emission features around 4680 Å (associated with He ii λ4686 and N iii λ4640; see van Velzen et al. 2020 and references therein). However, we first note that the emission lines (such as He ii and Balmer) in TDEs (FWHM ∼ 10⁴ km s⁻¹) are much broader than those observed in AT 2021loi (FWHM ∼ 3000 km s⁻¹). In terms of broadband SEDs, the color of AT 2021loi is redder than what is found for TDEs in the early stages and shows very limited evolution (Figure 6). When modeling the transient emission in AT 2021loi with a blackbody, we find that the derived temperature is much lower than what is seen in TDEs, whereas its (peak) bolometric luminosity and inferred blackbody radius are higher than in TDEs (see Section 4.1). These features lend further support to our interpretation that the transient emission in AT 2021loi is not driven by a TDE—at least not of the common optical/UV kind—as emphasized by Trakhtenbrot et al. (2019a) for the BFF class as a whole.

Also, AT 2021loi is not compatible with scenarios of partial TDEs (e.g., Zhong et al. 2022), double TDEs (e.g., Mandel & Levin 2015), or extreme mass-ratio inspirals (EMRIs; e.g., Sari & Fragione 2019). The second peak in AT 2021loi occurred ∼400 days after the first peak, which is longer than predicted in the partial TDE scenarios (i.e., ≲150 days for main-sequence stars). On the other hand, in the case of EMRIs, repeated flares are expected to happen on longer timescales (i.e., ≳several years). However, more detailed simulations of such events are needed to investigate their possible emission signatures.