Mid-infrared Outbursts in Nearby Galaxies (MIRONG). II. Optical Spectroscopic Follow-up

Aiming at detecting potential characteristic emission-line variation, we began our analysis with subtraction of continuum. These spectra are first corrected for Galactic extinction using the extinction map of Schlegel et al. (1998) and the extinction curve of Fitzpatrick (1999).

Because the SDSS spectra are dominated by starlight, we fit them using the procedure PPXF (Cappellari & Emsellem 2004; Cappellari 2017) and MILES spectral library (Vazdekis et al. 2010). Nevertheless, the AGN continuum is likely not negligible for eight broad-line objects (Section 3.2), while the PPXF is a pure starlight component fitting code. So for these eight sources, we used the spectral decomposition results of Paper I instead. The fitting of J1647+3843 is shown as an example in Figure 1.

Zoom In Zoom Out Reset image size

For the DBSP spectra observed after the MIR flare, an additional continuum component associated with the MIR outburst is possible. We thus also placed a reddened power-law continuum assuming an extinction curve of the Galactic dust (Fitzpatrick 1999), as normally appears in broad-line AGNs, in the fitting. On the other hand, the shape of the starlight component remains exactly the same as that of the SDSS spectrum. Accordingly, we fit the continuum of DBSP spectra using the following formula:

$\begin{eqnarray}&&A* S(\lambda )+B* {\lambda }^{\alpha }* {10}^{-0.4* E(B-V)* k(\lambda )}.\end{eqnarray} \tag{ 1 }$

S(λ) is the starlight component of the same source obtained by applying the PPXF procedure to the SDSS spectrum, and the power-law index α was set to be in the range between −2 and −1. k(λ) is the extinction curve and E(B − V) varies from 0 to 2.5.

The fitting was done using the Python code MPFIT, ⁹ and the errors of the parameters were estimated from Monte Carlo simulations. We show an example in the bottom panel of Figure 1.

After subtracting the continuum, the emission-line spectrum is modeled with a combination of Gaussian functions in several segments to measure the line fluxes of various broad and narrow lines. The broad lines include Hα, Hβ, He ii, and Bowen line N iii. The narrow lines contain [O iii], [N ii], [S ii] and [Ne ııı]. Each narrow line is modeled with one or two Gaussians, while broad Hα and Hβ are fitted simultaneously with Gaussians of the same width and centroid in velocity. The number of Gaussians is determined according to the 95% threshold of improvement with the F-test by adding one more Gaussian to the fit. We show an example of Hα and Hβ fitting results in Figure 2.

However, reliable detection of broad Hα requires not only at least one broad component as suggested by the F-test but a 5σ criterion was also demanded. Accordingly, eight sources in our sample have reliable broad Hα detection in the SDSS spectra, which are taken before the MIR flare (see Appendix A). For DBSP spectra, usually with multiple epochs, we have fitted and tried to detect broad Hα in each of them. If there is no broad component detected in Hα, we take the flux at the 90% confidence level as the upper limit, which is derived from Δχ² = 2.7 (Avni 1976). We will adopt broad Hα, which is usually the strongest, if it exists, to represent the evolution of broad components.

We consider seven common coronal lines, namely [Fe vii]λ3759, [Fe v]λ4071 [Fe xiv]λ5304, [Fe vii]λ5722, [Fe vii]λ6087, [Fe x]λ6376, and [Fe xi]λ7894. Each is fitted with one Gaussian of FWHM < 1500 km s⁻¹ while a linear function has also been added to represent the local residual continuum (see an example in Figure 2). An F-test is employed to assess whether the Gaussian component is necessary or not. A line is detected if its flux is above the 3σ level. The presence of a coronal line is defined as the detection of three coronal lines or two lines with one above the 5σ level in the same epoch, or a coronal line in multiple epochs. According to the above rule, no coronal lines were present in all our SDSS spectra but they show up in the post-outburst spectra of nine objects.

Next we begin to check the emission-line variability between DBSP and SDSS spectra. Obviously, if broad Hα is reliably detected in the SDSS spectrum, one has to measure the line flux to assess whether it is variable or not. This requires a reliable flux calibration. Although our spectra were already flux calibrated using a standard star, due to changes in weather conditions over the night, significant uncertainty could be introduced. With a reliable broad Hα indicating significant nuclei activity, recalibration was done based on the normalization of the [O iii]λ5007 flux, which is assumed to be constant for the Seyfert galaxy over the monitoring timescale (Peterson et al. 2013). The variability of broad Hα then was acquired by checking whether its flux at any epoch during the follow-up observations has changed by 3σ relative to the SDSS one or not. However, recalibration could not be implemented for J2215–0107, one of the eight sources with reliable broad Hα detection in SDSS spectra, due to the too-low spectra quality, and we just removed it without further discussion (hence, 53 left). Significant variation of broad Hα was found in the left seven. Counting those with newly emerged broad Hα, 22 in our samples display variation in broad Hα and the appearance of coronal lines was found in 9 of them. According to whether or not the variability of Hα,_B or coronal lines is detected, our sample can be classified into three subclass (see Table 1). It is interesting to note that there is no object in our sample that shows only variation in coronal lines.

Additionally, as the [O iii]λ5007 flux of the seven with reliable broad Hα in SDSS spectra and other Seyfert 2 galaxies was expected to be constant in our multiple DBSP follow-up spectra and the same as that of SDSS, the scatter of the [O iii]λ5007 flux normalized by the SDSS ones could thus be used to estimate the accuracy of the original flux calibration. The median value of the scatter is about 0.17 dex. When we focused on the long-term evolution (see Section 3.3.2), such uncertainty would be taken into consideration for MIRONG starting from normal galaxies without a reliable broad Hα detection in SDSS, and only when the variation amplitude exceeds 0.17 dex do we think there is an upward or downward trend.

3.3.1. The Variability in Different Galaxy Types

First, the percentage of objects that display emission-line variability differs greatly in different spectral types (see Table 1). It is remarkable yet not surprising to note that Seyfert galaxies possess the highest fraction given that AGNs display ubiquitous variability in both continuum and emission lines. In particular, all four Seyfert 1 galaxies show obvious Hα,_B variations, with coronal lines being detected in two of them. Though obvious variability is not expected for normal Seyfert 2 galaxies (e.g., Yip et al. 2009), the line variability found for the Seyfert 2 galaxies in our sample is understandable, because there is a high possibility for them to be changing-look AGNs considering the selection criteria. In contrast to Seyfert galaxies, the star-forming (SF) and composite galaxies present a significantly lower portion of variability (33.3% and 35%, respectively), with the former dominated by Hα,_B while the latter is dominated by Hα,_B-CL variations. Low-ionization nuclear emission-line regions (LINERs), which are usually considered to be photoionized by a weak AGN (Ferland & Netzer 1983; Halpern & Steiner 1983), have the same ratio as star-forming galaxies though.

Zoom In Zoom Out Reset image size

3.3.2. Long-term Evolution of Broad Hα and Coronal Lines

All 22 objects with detection of emission-line variability with respect to SDSS spectra show an apparent increase in broad Hα emission while only a small fraction of them show coronal-line variations. Except for J1442+5558 and J1003+0202, the strength of Hα,_B declines again in the subsequent observations although most of them have not returned back to the flux level of SDSS as of the latest epoch. It indicates that the Hα,_B emission fades slowly and a minority of them may sustain for a long time. Despite their weakness, coronal lines clearly fade out after the first follow-up observation except for J1442+5558, whose coronal lines have just emerged in recent DBSP spectra while Hα,_B has been detected in the first DBSP. We show the spectral evolution of J1442+5558 in Figure 3 and all 22 sources with emission-line variation in Appendix B.

Zoom In Zoom Out Reset image size

3.3.3. N iii and He ii λ4686

Among the 53 useful sources, 22 show obvious emission variability in at least broad Hα, while the other 31 appear unchanged in optical spectra according to our observations. It is necessary to explore whether the nondetection is intrinsic or simply caused by selection effects. First, we have examined the S/N of the spectroscopic data of the variable and nonvariable subsamples as the higher-quality data are undoubtedly more helpful for the detection of weak emission lines embedded in the continuum. However, no obvious difference is found.

As we have shown in Section 3.3.2, the Hα,_B and iron coronal lines of most objects show apparent decreasing trends after brightening. Some of them have even disappeared rapidly, demanding quick spectral follow-ups to capture the variations. For this reason, if the spectroscopic observations were conducted too late, we might miss the fleeting emissions. We have calculated the time separation between the spectroscopic observational date and the first MIR brightening epoch (3σ brightening in either the W1 or W2 band; see details in Paper I). There is no significant difference between the spectral variable and nonvariable subclass according to a Kolmogorov–Smirnov test (K-S test). It seems that the nondetection of spectral variability should not be a consequence of too-late observations if the two subclasses share comparable decay timescales.

Dust is ubiquitous in galaxies from galactic disks to galactic centers. It is especially considerable for our MIRONG sample because they were selected by IR echoes radiated by the circumnuclear dust around SMBHs. The dust along the light of sight can obscure the optical photons, making them undetectable in both continuum and emission lines. Figure 4 shows the W1-band variability amplitude ΔW1, W1 peak luminosity L _W1 , and dust temperature T_dust (data drawn from Paper I, also a significant difference in ΔW2 and L_W2). All three parameters show a significant difference between the variable and nonvariable subclass based on a K-S test with the p-value being 3.1 × 10⁻⁵, 1.1 × 10⁻⁵, and 3.8 × 10⁻⁶, respectively. In detail, the nonvariable sources have lower ΔW1, L _W1 , and T_dust, which can be explained by the self-obscuration of dust emission. We have checked the narrow-line Balmer decrements of the two populations and found no difference, indicating a similarity in galactic-scale dust. The dust discrepancy between them might be in the galactic nucleus.

Zoom In Zoom Out Reset image size

We have also looked into the distributions of other parameters, including M_BH and the rising and decay timescales of MIR light curves. All show negligible differences.

In fact, the SN scenario has been largely excluded (Paper I) based on MIRONG's high MIR luminosity (see also Wang et al. 2018b) and proximity to the galaxy center. Spectroscopic evolution further supports this conclusion.

First, the varying components in the optical spectra overwhelmingly feature emission but without notable absorption lines, which is atypical for conventional SNe (Filippenko 1997). Further, the broad Hα lasts typically 2–4 yr or more from their first detection in our sample, which is also in sharp contrast with the fast decline for a typical Type II SN (e.g., Uomoto & Kirshner 1986; Sahu et al. 2006). Only a subclass of SNe (Type IIn), which shows signatures of the interaction between the SN ejecta and the circumstellar medium and exhibits obvious Balmer emissions, most notably Hα, with a weak or absent broad absorption component (Schlegel 1990; Filippenko 1997), bears some similarities with our sample. And some objects, such as SN2010jl (Fransson et al. 2014; Moriya et al. 2020), SN2015da (Tartaglia et al. 2020), SN2005ip (Stritzinger et al. 2012; Fox et al. 2020), and SN2006jd (Stritzinger et al. 2012), displayed a long-lasting broad Hα with similar line luminosity to our sample and could be accompanied by coronal lines. However, their late-time spectra usually show prominent low-ionization emission lines such as near-infrared Ca ii triplets and O i λ8446, which are completely absent in our spectra. Hence, our spectroscopic results argue against an SN origin.

The final sample of the 22 spectroscopically variable sources consists of 7 Seyfert galaxies according to their SDSS BPT classification. In addition, we note that there are three more galaxies in other classes, that is J1315+0727, J1332+2036 (composite), and J1133+6701 (LINER), that show evident broad Hα in their SDSS spectrum. Therefore, we grouped the 10 sources (Seyferts or with a broad line) into a subsample of unambiguous AGNs before the MIR outburst and the remaining 12 into a subsample without or with very weak AGNs. We begin our discussion with the latter subsample.

4.2.1. One Bona Fide Turn-on AGN Candidate

4.2.2. Most are Good TDE Candidates

4.2.3. The Possibility of Sporadic Gas Accretion

The real physics behind MIRONG in active galaxies could be complicated and diverse, with turn-on AGNs, unusual AGN variability, and TDEs all possible. Distinguishing among them is quite challenging to our current knowledge. We will try to give our initial impression of them here, leaving a detailed analysis to our subsequent works.

The amplitude of stochastic variability, which is a defining characteristic of AGNs, is generally not higher than tenths of a magnitude within months to years although it increases toward longer timescales (e.g.,Vanden Berk et al. 2004; MacLeod et al. 2010). Only the most extreme tail of AGN variability distribution may show an amplitude comparable with TDEs over a short period of time (e.g., MacLeod et al. 2012; Rumbaugh et al. 2018). Moreover, conventional AGN variability does not show regular and smooth patterns, such as the power-law decay seen in TDEs. Nevertheless, there indeed exists a population of AGNs displaying major flares at timescales of years, which resembles TDEs more, in their light curves, (e.g., Graham et al. 2017; Kankare et al. 2017; Trakhtenbrot et al. 2019). The nature of these AGN flares remains very elusive even with spectroscopic observations as we still lack adequate information to distinguish each scheme from impostors at the benchmark level (Zabludoff et al. 2021), as we emphasized in the last section.

Exploration of our MIRONG sample becomes even difficult without direct variability information of their original emission. Luckily, emission-line variability, which can be considered as an alternative tracer, has been detected in most objects (see Table 1). The detection of Hα,_B variation itself is not a surprise for AGNs, yet their rapid evolution may shed light on the central ionizing sources. Characteristic emissions other than Hα,_B, namely iron coronal lines and He ii λ4686, might give further clues to the outburst. Only three objects (J1105+5941, J1402+3922, and J1657+2345) show these emissions, and their Hα,_B has either been restored or is steadily declining. We thus take the three as tentative TDEs in AGNs for the time being as both iron coronal and He ii λ4686 are representative features of TDE spectra, although not decisive.

For the remaining sources, only Hα,_B variability has been found and their physics are even harder to constrain. To be consistent with prior definitions, we categorize J1003+0202, which was a Seyfert 2 and presented stable Hα,_B after the brightening, as a turn-on subclass while the rest as AGN flares. We note that J1315+0727 only has two observations in 2017 and its Hα,_B was still rising. More recent spectra will confirm whether it has faded away now. It should be emphasized again that the flares here may be caused by distinct processes in different sources, such as TDE, sporadic gas accretion, or special accretion disk instability.

The nature of the other 31 objects without clear spectral variations is still mysterious, yet might be even more fascinating. One possibility is that they belong to a new population of fast transients, which evolve much faster than variable sources. The high-cadence surveys over the past decade have indeed revealed a population of transients with timescales obviously shorter than those of SNe and TDEs, simply called "Rapidly Evolving Transients" or "Fast-Blue Optical Transients" (FBOTs) in the literature (e.g., Drout et al. 2014; Pursiainen et al. 2018) as their physical origin remains a puzzle. For instance, the nearest and most well-studied case, AT 2018cow, shows a rise-to-peak timescale of only three days and faded away within a couple of months (Margutti et al. 2019; Perley et al. 2019). The reprocessed MIR light curves cannot reflect the intrinsic outburst timescales unambiguously due to their coupling with dust distributions. Moreover, the optical surveys that overlapped with the outbursting period of our MIRONG sample have either poor cadence (e.g., PanSTARRS, PTF) or depth (e.g., ASASSN), thus they cannot put effective constraints on their optical emission. For likely the same reason, most of them have been missed by optical surveys. Future selection of more recent MIRONG such as the outbursts after 2018 with both excellent WISE and ZTF light curves, may help test the fast transient hypothesis as the ZTF survey has a much higher cadence and depth compared with previous surveys.

The discrepancy in the dust properties (see Section 3.4) suggests that dust obscuration may play a vital role and those spectral nonvariable sources correspond to more seriously obscured ones. In this scenario, they are good candidates for obscured TDEs, turn-on AGNs, or any other types of SMBH transients that are undetectable in optical bands. The discrepancy in IR emission can be also partly accounted for by the possibility that the outburst energy of the nonvariable sources is systematically lower than that of variable sources if their dust contents (covering factor) are identical. Further investigations with the aid of multiwavelength observations (e.g., radio; Mattila et al. 2018) are needed to make a conclusion about whether they are real dust-obscured events or an intrinsically distinct population.