The Calm Before the (Next) Storm: No Third Outburst in 2019–2020, and Ongoing Monitoring of the Transient AGN IC 3599

We report on follow-up observations of the Seyfert 1.9 galaxy IC 3599 with the NASA Neil Gehrels Swift mission. The detection of a second X-ray outburst in 2010 by Swift after the first discovery of a bright X-ray outburst in 1990 by ROSAT led to the suggestion of two very different explanations. The first one assumed that IC 3599 exhibits outbursts due to repeated partial tidal stripping of a star, predicting another outburst of IC 3599 in 2019/2020. The second, alternative scenario assumed that the event observed in X-rays is due to an accretion-disk instability, which would suggest a much longer period between the large outbursts. Our continued monitoring campaign by Swift allowed us to test the first scenario that predicted a repetition of high-amplitude flaring activity in 2019/2020. We do not find any evidence of dramatic flaring activity with factors of 100 since the last X-ray outburst seen in 2010. These observations support the accretion-disk scenario. Further, while IC 3599 remains in low-emission states, the long-term X-ray light curve of IC 3599 reveals ongoing strong variability of a factor of a few. The most remarkable event is a miniflare of a factor of 10 in X-rays in 2022 December. After that flare, the otherwise supersoft X-ray spectrum shows an exceptional hardening, reminiscent of a temporary corona formation.


INTRODUCTION
Since the proposed scenario of the tidal disruption of a stars by a supermassive black holes (e.g., Rees 1988), there have been many observations confirming this scenario.In the 1990s the X-ray mission ROSAT (Truemper 1982) had been instrumental for the discovery of the first stellar tidal disruption events (TDEs), in the form of giant-amplitude X-ray outbursts from quiescent host galaxies (e.g., NGC 5905 (Bade et al. 1996;Komossa & Bade 1999), RX J624.9+7554 (Grupe et al. 1999), and RX J1242.6-1119(Komossa & Greiner 1999)).More recently, TDEs have been detected by all major X-ray observatories, including the discovery of jetted events with the Neil Gehrels Swift observatory, first in Swift J164449.3+573451(e.g., Burrows et al. 2011;Bloom et al. 2011;Zauderer et al. 2011).The discovery of TDEs from quiescent galaxies, however, is not limited to the X-ray sky.GALEX led to the first detection of events in the UV-optical band (Gezari et al. 2006), and SDSS enabled the first identification of events with transient optical emission lines such as HeII, broad Balmer lines, and coronal lines (Komossa et al. 2008).TDE candidates continue to be identified at all wavebands (see Komossa 2015;van Velzen et al. 2021;Webb et al. 2023;Komossa & Grupe 2023, for recent reviews).
Although the TDE model is required to explain sudden, luminous X-ray outbursts in non-active galaxies, this picture is far from being clear in AGN.In the case of an AGN, the power is coming from a long-lived accretion disk surrounding the central supermassive black hole.This power is not constant and varies, and in extreme cases a significant change in the disk properties can cause a dramatic increase in the AGN luminosity.An increasing number of optical changing-look AGN which do change their luminosity output along with their optical broad emission lines, have been identified in recent years (see Komossa & Grupe 2023, for a review) even though the first cases were known already in the 1970s-80s (e.g., Penston & Perez 1984;Alloin et al. 1985).Different theoretical scenarios related to accretion disk instabilities or other disk processes have been explored in recent years (e.g., Nicastro 2000;Grupe et al. 2015;Ross et al. 2018;Noda & Done 2018;Dexter & Begelman 2019;Sniegowska et al. 2020;Pan et al. 2021;Kaaz et al. 2023;Cao et al. 2023).There will be a major discovery space for both, TDEs and changing-look AGN once the Vera Rubin Telescope will go online 1 .
One of the earliest examples of an AGN that might have exhibited either a TDE or an accretion disk instability was the Seyfert 1.9 galaxy IC 3599 (Zw 159.034; 1RXS J123741.2+264227;α 2000 = 12 h 37 m 41. s 2, δ 2000 = +26 • 42 ′ 27 ′′ , z=0.0215).IC 3599 was discovered as a very X-ray bright AGN during the ROSAT All-Sky Survey (RASS, Voges et al. 1999) in 1990.Subsequent pointed ROSAT observations during the following years revealed that it had faded by a factor of initially 60 and even more than 100 in later observations.Both, Brandt et al. (1995) and Grupe et al. (1995) speculated at the time that this was caused by either a TDE or high-amplitude AGN variability.What made both scenarios plausible was the discovery of a strong evolution in the optical emission line spectrum.The spectrum taken about half a year after the RASS observation shown in Brandt et al. (1995) showed very strong permitted lines from Hydrogen and Helium along with transitions from other elements.These strong emission lines led Brandt et al. (1995) to classify IC 3599 as a Narrow-Line Seyfert 1 galaxy (NLS1), albeit without FeII emission.Grupe et al. (1995) showed spectra obtained years after the RASS which displayed the spectrum of a Seyfert 2 galaxy but with fading coronal iron lines.The fading coronal lines were confirmed by new optical spectroscopy of Komossa & Bade (1999) who still detected a broad component in Hα, making IC 3599 a Seyfert 1.9 galaxy.These authors re-emphasized that TDEs are best identified in quiescent host galaxies, because in AGN, accretion-disk-related activity is the most plausible explanation for variability Rees (see also 1990).Note, no FeII lines, the signature lines in Narrow Line Seyfert 1 galaxies, were present at any time in the optical data of IC 3599.Doubt remained given that this is an AGN, if a TDE really was an explanation of the X-ray outburst.Instead, the spectacular changes in its optical emission-line spectrum combined with the dramatic X-ray variability make IC 3599 one of the most extreme examples of a changing-look AGN (Komossa & Grupe 2023).
1 https://rubinobservatory.org/After ROSAT, IC 3599 was again observed in X-rays by Chandra in 2002 (Vaughan et al. 2004) and was still found in an X-ray low state.In two observations by the Neil Gehrels Swift mission (Gehrels et al. 2004, Swift hereafter) in 2010, it was seen again in an X-ray outburst state with its X-ray flux at a similar level as during the RASS (Grupe et al. 2015).Based on the RASS observation from 1991, the Chandra observation from 2002, and the Swift observation from 2010, Campana et al. (2015) speculated that the 2010 outburst was due to the tidal stripping of a star, repeating an initial 1991 tidal stripping event of a star orbiting the black hole, with a low black hole mass of 3 × 10 5 M ⊙ and a 9.5-year orbital period.This kind of repeated flaring has been suggested e.g. by Payne et al. (2021) to explain the quasi-periodic eruptions in ESO 253-G003/AT2014ko.Campana et al. (2015) predicted that another outburst in IC 3599 would be visible in 2019/2020.Although this is an intriguing model, Grupe et al. (2015) argued that the 1990 and the 2010 outbursts were due to accretion disk instabilities.Major arguments were the optical light curve from the Catalina observatory which showed a relatively gradual increase in the optical flux, as well as a low black hole mass not being supported by any observational parameter (instead, it was measured to be of the order of 2 − 20 × 10 6 M ⊙ ).
To test these different models, we continued to monitor IC 3599 with Swift typically once per month for a nine-month period each year.Swift is unable to observe IC 3599 for a three-month period between August and November due to the sun-constraint (a 45 • sun avoidance angle).Here we report on the entire Swift monitoring campaign from 2010 to June 2023.During the entire monitoring period no further outbursts have been observed in IC 3599, supporting the accretion disk instability scenario suggested by Grupe et al. (2015).Over the last few years there has been, however, a general trend of IC 3599 slowly becoming brighter in X-rays, and it has recently been observed in an intermediate state.
This paper is organized as follows: in Section 2 we will describe the Swift observations, present the results in Section 3, and then discuss them in Section 4. Throughout the paper spectral indices are denoted as F ν ∝ ν −α .Luminosities are calculated assuming a ΛCDM cosmology with Ω M =0.286, Ω Λ =0.714 and a Hubble constant of H 0 =70 km s −1 Mpc −1 , resulting in a luminosity distance of D=91.4 Mpc using the cosmology calculator by Wright (2006).All errors are 1σ unless stated otherwise.

Swift OBSERVATIONS AND DATA REDUCTION
Swift has observed IC 3599 since 2010.When it was discovered in 2013 that it had exhibited an X-ray outburst in 2010 (Grupe et al. 2015) we started observing it once a month.These observations were supported by another set of monthly observations (PI Campana) in 2019 which essentially made it a two-week cadence.All Swift observations are listed in Table 4.The full machinereadable table is available on Zenodo: 10.5281/zenodo.10899673.
The Swift X-ray telescope (XRT; Burrows et al. 2005) was operating in photon counting mode (pc mode, Hill et al. 2005) and the data were reduced by the task xrtpipeline which is part the HEASOFT package 6.30.1.Source counts were extracted in a circle with a radius of 47.1 ′′ (except the 2010 outburst observation for which we used 70.7 ′′ ).The background counts were extracted in a nearby circular region with a radius of 247.5 ′′ .For all spectra we used the most recent response file swxpc0to12s6 20130101v014.rmf.The X-ray spectra were analyzed using XSPEC version 12.12.1 (Arnaud 1996).If a sufficient number of counts (>30) was available, we used W-statistics (Cash 1979) for single observations.Although often the number of counts was too low to allow for a spectral analysis, we still determined a hardness ratio, defined as HR = hard−sof t hard+sof t with the soft and hard bands in the 0.3-1.0 and 1.0-10 keV range, respectively by applying the program by Park et al. (2006).The count rates in the XRT were determined by using the Swift XRT product page at the Swift Data Center at the University of Leicester2 (Evans et al. 2007), as well as the Living Swift XRT Point Source Catalogue3 (Evans et al. 2023).
To perform a spectral analysis of different X-ray flux states, we merged the data of various time periodd into single data sets within XSELECT.These periods and spectral analysis results are listed in the appendix in Table 2.We then constructed the auxiliary response files (arf) by using the FTOOLS command addarf, using the weighting factor of each individual arf by dividing the individual exposure time of each single observation by the total exposure time of the merged spectrum.
The UV-optical telescope (UVOT; Roming et al. 2005) data of each segment were co-added in each filter with the UVOT task uvotimsum.In general, all observations were performed in all 6 UVOT filters, except if the observation was interrupted by Gamma-Ray Burst detections or higher priority Target of Opportunity observations.Source counts in all 6 UVOT filters were selected in a circle with a radius of 5 ′′ and background counts in a nearby source free region with a radius of 20 ′′ .UVOT magnitudes and fluxes were measured with the task uvotsource based on the most recent UVOT calibration as described in Poole et al. (2008) and Breeveld et al. (2010).The UVOT data were corrected for Galactic reddening (E B−V = 0.015; Schlegel et al. 1998).The correction factor in each filter was calculated with equation (2) in Roming et al. (2009) who used the standard reddening correction curves by Cardelli et al. (1989).The fluxes in X-rays and in the UV/optical obtained by Swift are listed in Table 3.This table also contains the observations published already in Grupe et al. (2015) which were re-analyzed due to updates in the calibration files.The full machine-readable table is available on Zenodo: 10.5281/zenodo.10899673.Note that many 0.3-10 keV fluxes were estimated from the count rates derived from the Leicester Swift-XRT Product tool (see footnote 1) and then converted to fluxes based on the closest X-ray data that allowed a spectral analysis.These estimated fluxes are marked in Table 3.
Figure 1 displays the long-term X-ray light curve of IC 3599.The data for this light curve are available on Zenodo: 10.5281/zenodo.10899673.The light curves clearly show the two giant outbursts seen by ROSAT and Swift about 20 years apart.The long-term light curve also displays a mini flare which shows an increase in X-ray flux by a factor of about 10 in December 2022.
Figure 2 displays the full Swift light curve between 2010 and August 2023.Note that due to the low background in the Swift XRT even a low number of counts results in a 3σ detection.However, the number of counts is often too low to allow for a reliable determination of the hardness ratio.It displays the outburst in 2010 and the remarkable flat light curve since Swift resumed the observations with our monitoring program in 2013.The green vertical lines mark the year 2019 when Campana et al. (2015) predicted that the hypothetic star would return and undergo another partial disruption event while going through periastron and thus form a temporary accretion disk around the black hole again, causing another outburst.Clearly, this did not happen.There are no signs of unusual activity in X-rays or UV, yet.There is, however, a general trend since about 2021 that IC 3599 is slowly becoming brighter in X-rays (see below in Section 3.1.3).The hardness-ratio light curve suggests strong spectral variability in X-rays as well.
Figure 3 shows the Swift XRT and UVOT light curves from November 2022 to August 2023.It shows a very dense monitoring phase in January 2023 due to the unusually high count rate of 0.03 counts s −1 seen in December 2022.This is a factor of 10 higher than typically observed when it is in its very low state (see also Figure 2).The light curve also displays a remarkable hardening of the X-ray spectrum after the mini flare from being very soft at almost HR=-1 to becoming harder with HR>0.However, from about March 2023 to the end of the observing campaign in August 2023, IC 3599 has become softer again.

Variability Analysis
We have monitored IC 3599 regularly since 2013 which has led to a total number of more than 100 observations.This rich data set allows for a detailed temporal analysis.As the first step we can apply the Fractional Excess Variance as defined by Rodríguez-Pascual et al. (1997) with an uncertainty of <f > 2 following the definition by Edelson et al. (2002).Here, σ is the variance of the light curve, δ is the mean value of the uncertainties of the fluxes/count rates, and < f > is the mean value of the fluxes/count rates.We applied this to the Swift XRT and UVOT light curves including and excluding the 2010 outburst data.The results of the fractional excess variance are listed in Table 1.
While the results of the fractional excess variance including the 2010 outburst data seem to follow the decrease of F var with increasing wavelength as expected from accretion disc reprocessing models (e.g.Cackett et al. 2007), excluding these data shows a different picture: The same overall trend is still visible, however 2 bands (U and V) deviate from the systematic trend.When working with a variance and mean, the assumption is that the underlying distribution is Gaussian (Gauß 1821).We checked the flux distributions in X-rays and in each UVOT filter using R (e.g., Craweley 2009) by looking at the histogram of the flux distribution as well as at the quantile-quantile plot (QQ plot).We found that this assumption is well-justified for the UVOT observations.Except for the high state data from 2010, the data of all UVOT observations are well-described by a Gaussian distribution in all 6 filters.Figure 4 displays the distribution and the QQ plot of the W2 magnitudes of IC 3599 which clearly shows that this is a Gaussian distribution.The other UVOT filters show similar distributions which are all consistent with a Gaussian distribution.
However, the X-ray flux variations are not represented by a Gaussian distribution.This is clearly shown in Figure 5 which displays the histogram and the QQ-plot.
For non-Gaussian distributions, the variance and the mean are not good measures of a sample and in these cases, the median is a much better measure than the mean.In order to study the overall variability in all Swift XRT and UVOT light curves we designed a new measure: a variability parameter which is like a reduced χ 2 value.This parameter, P var , is a measure of how strongly a source deviates from the median value.Here   with small uncertainties.We estimate the uncertainties on the values of P var by ∆P var = δ med x med P var .The values for P var are listed in Table 2.These values also confirm the stronger variability in X-rays than in the UV and optical.The variability parameter P var also shows that these values are not dominated by the 2010 outburst measurements.It allows a direct comparison with the variability strength between Gaussian distributed data and non-Gaussian data, like the X-ray flux distribution of IC 3599, simply because it does not assume a Gaussian distribution and it is therefore a more general measure of the variability.
Just a Gaussian flux distribution by itself does not prove that these are just random fluctuations.In order to check on any variability we performed a periodogram within R using spec.pgram.Besides some red noise at lower frequencies, we only found white noise in the data.The V filter data are dominated by white noise.This variability pattern is typical for type 2 AGN (Wang et al. 2024) and may suggest that IC 3599 remains in its Seyfert 1.9 state at this time.Figure 6 displays the long-term Swift XRT light curve with the data averaged over each year, except for the 2013-2016 period because the number of observations is relatively small and the count rate was low, of the order of 0.003 counts s −1 .We binned the 2013-2015 and 2015-2017 observations into one bin each and after 2017, the observations of each year were combined.The binned data are displayed as red data points in Figure 6.The 0.3-10 keV fluxes and the hardness ratios for these periods are listed in Table 2.There is clearly an overall slow brightening in X-rays over the last decade.The hardness ratio light curve also suggests spectral variability with IC 3599 becoming softer from 2013 to 2020, but becoming harder again since.However, there is no correlation between the hardness ratio and the X-ray brightness of IC 3599.
Figure 7 displays the UVOT light curves between 2013 and 2023 with the merged data shown in red.Overall, there is no significant variability from year to year.The results for the merged data are listed in Table 4. Figure 7 and Table 4 show that there is no significant variability on long or short time scales, not even a trend like seen in X-rays.This is somewhat expected given the Gaussian distribution of the UVOT data.

X-ray Spectral Analysis
For the comparison of the high and low states we merged the 2010 February and May data into a high state spectrum and did the same for the low state by merging all data from 2013 to 2023.The high and low state data were then fitted with various spectral models.The results of these fits are listed in Table 3.All models were fitted with the z=0 absorption fixed to the Galactic value of 1.17 × 10 20 cm −2 .
The total exposure time of the high state spectrum was 3354s and the count rate 0.115 counts s −1 which means a total number of source counts of 260 counts which account for 99.3% of the counts in the spectrum.The relatively low number of counts in the high-state spectrum only allowed for an analysis using W statistics (Cash 1979).
The overall count rate of the low-state spectrum is 0.0046 counts s −1 and has a total exposure time of 302 ks.This results in a total number of about 1400 source counts, which accounts for 89% of the counts in the spectrum.The relatively large number of counts allows us to fit the low-state spectrum with χ 2 statistics.The binning of the low-state spectrum was 20 counts per bin.This spectrum is displayed in Figure 8.
The high and low flux state spectra were first fitted by a single power law model.While in the high state case the data can be fitted by this model acceptably, the low state spectrum clearly deviates from a simple power law model as shown from the fit parameters in Table 3 and in Figure 8.One thing to keep in mind is that the number of counts in the high state data above 1.5 keV is so low that we do not have any information on the spectrum above this energy.Next, we applied a broken power law model and a blackbody plus power law model.These are phenomenological models that can describe the spectra quite well in both cases.Finally, we used a neutral partial covering absorber model.For the high state data, the parameters of the partial covering absorber are unconstrained.There are no data at higher energies that are needed to fit this model.For the low state data however, the spectrum is well described by a partial covering absorber model with a column density of 6 × 10 22 cm −2 and a covering fraction of 80%.
Figure 9 displays an absorbed blackbody plus power law model fit to the merged high state data from 2010 and the 2013-2022 low state data.Note, that for display purposes the high state data have been binned, however the spectral fits were performed with unbinned data using W-statistics (Cash 1979).The blackbody component is displayed in red while the power law component is shown in green.As for the change between the high and low state black body and power law components, the black body component changes by a factor of 48±13 and the power law component by a factor of 23±6.This suggests that the blackbody component decreased stronger than the power law component.

Strong X-ray Spectral Change in January 2023
As shown in the hardness ratio light curve in Figure 3 there is a strong spectral change from being a very soft source to a significantly harder AGN around MJD 59954-59977 (2023-January-10 to 31).In order to study the change in the X-ray spectrum, we merged the data after IC 3599 emerged from its sun constraint on 2022-October-31 to 2023-January-10 when it was in the soft state and between 2023-February-10 and May-01 when it was in the hard state.Both spectra can be fitted by simple power law models.For the soft and hard phases, the photon indices were Γ = 4.03 +0.37 −0.33 and Γ = 1.70 +0.87  −0.54 , respectively.During the soft state the AGN is also brighter with a flux of F 0.3−10keV = (4.33 ± 0.40) × 10 −16 W m −2 , while during the hard state the flux was at (1.34 ± 0.58) × 10 −16 W m −2 .Since about March 2023 IC 3599 has transitioned again into a softer X-ray state.

UV Morphology
Figure 10 displays the W2 image which was merged from all low-state W2 data from 2013-2023.This image has a total exposure time of 99 ks.The image clearly shows the spiral structure of the host galaxy, tracing the young, blue, stars in the spiral arms.The projected angular diameter of the galaxy is about 75" which corresponds to a physical size of 33 kpc; larger than a typical dwarf galaxy, suggesting that the SMBH mass of IC 3599 is not particularly low.

UV Colors
Although the Gaussian distributions of the magnitudes in all 6 UVOT filters during the low state suggest random fluctuations, we still checked if there may be any dependence on the colors with fluxes in X-rays and the UV.For this purpose, we calculated U-B and W2-W1 colors and correlated them with various properties, like W2 magnitude, count rate and hardness ratio.As expected, we did not find any correlations among these parameters.

Spectral Energy Distribution
The    cause to determine the 2 keV data point we assumed a simple power law model, which may not be the correct model at 2 keV, but with the lack of available data it is the value derived from the X-ray spectral fit.
For the 2010 outburst data and applying the simple power law model, we derived a 0.2-2.0keV luminosity of L X = (9.9± 0.8) × 10 35 W . Applying the relation logL bol = 1.23 × log(L X ) − 7.36 as given in Grupe et al. (2010) we estimated the bolometric luminosity as 8 × 10 36 W. Assuming a black hole mass of IC 3599 of 2.5×10 6 M ⊙ the Eddington luminosity is 3.15 × 10 37 W which then results in an Eddington ratio of L/L Edd = 0.25.
As for the low state 2013-2023 data, assuming the broken power law model as shown in Table 3, the 0.2 -2.0 keV luminosity is 1.8 × 10 34 W which results in a bolometric luminosity of 6 × 10 34 W and an Eddington ratio of L/L Edd = 2 × 10 −3 .
In addition, the SED (Figure 11) also displays the SED of the data during the December 2022 flare period (open magenta circles) and the data after the mini flare from February to May 2023 (open green triangles).Note that for visualization purposes we have binned the data with a binning of 10.For the spectral fits, no binned data were used applying W statistics (Cash 1979).The Xray data after the mini flare indicate a significant flux increase in the hard X-ray band.The spectral analysis of these data results in a very flat X-ray spectrum with Γ = 1.70 +0.81  −0.51 , significantly flatter than what has been observed during the entire 2013-2022 low state period.Note that the low number of counts did not allow for a more sophisticated spectral analysis.

High-Amplitude Variability of AGN
In principle, any single one of a huge number of small or large flares in any blazar or radio-quiet AGN could be a TDE.However, TDEs are very rare events, and almost every single AGN or blazar is known to vary (see Ulrich et al. 1997;Gaskell 2008;Gallo 2018, for reviews).Therefore, exceptional positive evidence is required for any claim of a TDE in an AGN, as emphasized by, e.g., Rees (1988) and Komossa & Bade (1999).The situation is completely different, when a quiescent galaxy suddenly shows a huge X-ray outburst, because the only known model to explain a quasar-luminosity outburst in a galaxy without long-lived accretion disk is a TDE.IC 3599 is a bona fide AGN with a long-lived NLR (Komossa & Bade 1999; Grupe et al. 2015), and closely resembles the increasing number of changing-look AGN identified in recent years (e.g.Alloin et al. 1986; Runco  et al. 2016;MacLeod et al. 2019;Frederick et al. 2019;Ochmann et al. 2024).Therefore, so far, no definite positive evidence for a rare TDE in this highly variable AGN has been presented.However, we note that Mandel & Levin (2015) mentioned in passing the possibility of tidal disruption of one star and later tidal stripping of the second star of a binary star system in context of IC 3599.In that case, the near-identical peak luminosities of the two outbursts of IC 3599 would be unexpected.Further, w.r.t. the two outbursts of IC 3599, Komossa et al. (2014) pointed out that during certain phases of binary SMBH evolution, or in the presence of a recoiling SMBH, TDE rates can be temporarily strongly enhanced (Chen et al. 2009;Komossa & Merritt 2008;Stone & Loeb 2011), but two events within decades would still be rare, and IC 3599 does not show evidence for a recent major merger (e.g., Figure 10), and the near-identical peak luminosities of the two outbursts would have to be coincidence.Therefore, Komossa et al. (2014) re-emphasized that TDEs are most reliably identified in quiescent galaxies.Further, Grupe et al. (2015) commented on the TDE model of Liu et al. (2009) (see also Liu et al. 2014) in application to IC 3599.Liu et al. (2009) predicted TDE light curves for stellar disruption in a binary SMBH system, showing characteristic recurrent flare and dip events.However, the timing and amplitudes of IC 3599 did not fit that model. 5he one TDE model with explicit predictions which we can now test and rule out with the new monitoring data presented here, is the published repeat-tidalstripping scenario for IC 3599, because that model made testable predictions for the time frame covered in our new observations.We come back to this model in the following paragraphs.Beyond that, we continue to consider AGN scenarios, rather than TDE scenarios, for this highly variable AGN.Below, we also discuss the other aspects of the long-term variability of IC 3599, unrelated to the two big outbursts.

4.2.
No Outburst of IC 3599 in 2019/2020 Campana et al. (2015) predicted that there would be another outburst due to the partial tidal stripping of an orbiting star in 2019 and 2020.They predicted a 9.5 year period.The model also assumed a black hole mass of the order of a few 10 5 M ⊙ .While in general a TDE scenario cannot be excluded in principle, as pointed out by Grupe et al. (2015) and Komossa & Bade (1999), IC 3599 is an AGN and the dramatic flaring in IC 3599 does not have to be associated with a tidal disruption event and that given the black hole mass estimates of IC 3599 of the order of several millions of solar masses, an accretion disk instability scenario would be more likely.Campana et al. (2015)'s prediction of repeat flaring made it possible to test the tidal stripping scenario.The Swift monitoring programs from 2019 and 2020 initiated by us and Campana clearly show that there is no outburst seen during this period (see Figure 2).This result clearly favors the accretion disk scenario as suggested by Grupe et al. (2015).

Slow Increase in X-ray Flux
Figure 6 suggests that there has been a slow overall increase in X-ray flux since about 2018.In the accretion disk scenario that could be explained by a slow fill-up of the inner part of the accretion disk which emptied out during the last outburst in 2010.In Grupe et al. (2015) we estimated the fill-up time to be several decades, following the relation τ fill = 0.33α −8/10 M 6/5 6 given in units of month following the relation by Saxton et al. (2015).The other parameters are the viscosity α for which we assumed a standard value of 0.1, the black hole mass M 6 in units of 10 6 M ⊙ , the infill rate Ṁ edd in units of the Eddington limit.The truncation radius R trunc and the inner radius R 0 are given in units of the gravitational radius R g = GM c 2 .While the X-ray flux seems to develop some long-term trend to become slightly brighter, there is no sign of any significant variability in the optical and UV.IC 3599 seems to be fairly constant in all 6 UVOT filters and the fluctuations seem to be random, which is supported by the Gaussian distributions of the magnitudes seen in each filter.

X-ray Mini Flare and Spectral Hardening Event in 2022/2023
As shown in Figure 6, there is an X-ray mini flare at the beginning of December 2022.Right after this shortlasting flare we see a dramatic hardening of the X-ray spectrum, as shown in the hardness ratio light curve in Figure 6.This hardening of the spectrum can also be seen in the spectral energy distribution in Figure 11.The data from this hardening event as displayed as green open triangles.
The mini flare started on 2022 December 01 and peaked on December 10 with a peak flux of 1.29±0.45×10 −15 W m −2 , which is a factor of about 10 higher than the flux on December 01 (see also Table 3).While the spectrum was very soft during the flare, the onset of the spectral hardening started around 2023 January 03 and reached the hardest state with a hardness ratio of 0.73±0.25 on March 31.The spectrum became soft again by the end of May 2023 and has remained soft in all observations after that.
The question is raised, which physical process(es) cause this temporary spectral hardening.The hard Xray spectral component, typically observed in all AGN, is usually interpreted as the presence of an accretion-disk corona.It is therefore possible that a corona formed temporarily, following the mini flare, and disappeared again in the low-state.Alternatively, we might speculate that a temporary absorption event caused an apparent spectral hardening.For instance, it is possible that due to the flaring activity a temporary outflow was triggered because of the increase in the accretion luminosity (Takeuchi et al. 2013).When a clumpy outflow partially covers the intrinsic emission, a spectral hardening can result.However, even at the peak of the mini flare, L/L Edd is only about 0.02, whereas launching of radiation-pressure driven outflows requires values closer to 1.
The Swift snapshot observations do not contain enough photon statistics to distinguish between these different spectral models (like emission from a tempo-rary corona, versus partial covering absorption, or more complex spectral models involving the interplay of several components).Deeper, triggered, follow-up observations during the flare decline phase will be very useful, if new mini flares are detected in the future.

Black Hole Mass Estimates
While the repeating TDE model of Campana et al. (2015) requires a small black hole mass of a few 10 5 M ⊙ , all observational parameters, like the width of the [OIII]λ5007 line, suggest a black hole mass that is at least 10 times lager (see our discussion in Grupe et al. (2015), where we also used BLR line width).The large physical size of IC 3599 as shown in Figure 10 may support a larger black hole mass as well, but the previous SMBH mass estimates using the various scaling relations are the most restrictive.

Conclusions
Despite the speculation in 1995 by Brandt et al. (1995) and Grupe et al. (1995), and the repeated-TDE model by Campana et al. (2015) that the X-ray outburst seen during the RASS could have been caused by a TDE, our long-term Swift monitoring campaign suggests otherwise, since the third outburst predicted by the repeat-TDE model in 2019/2020 did not happen.The X-ray outbursts seen in 1990 by ROSAT and 2010 by Swift are most likely caused by an accretion disk instability as favored by Grupe et al. (2015), since IC 3599 is not a quiescent galaxy but a bona fide AGN.In fact, IC 3599 is an extreme case of an optical changing-look AGN (Komossa & Grupe 2023; Xu et al. 2024).
Independent of the nature of the big outbursts or their absence in 2019/2020 in particular, we have also studied the long-term X-ray flux and spectral variability properties of IC 3599 during its more quiescent levels.IC 3599 turns out to be highly variable in the X-ray regime.One event stood out: We found that IC 3599 exhibited an increase in its X-ray flux by a factor of about 10 in December 2022.This mini flare was followed by a remarkable hardening of the X-ray spectrum in the following months.One possible explanation for such a behavior could be the temporary formation of an accretion disk corona.
We will continue to monitor IC 3599 with Swift.It depends on the model, if and when a new outburst is expected.For instance, if the previous two outbursts were powered by a disk instability, it would depend on the evolving and variable AGN disk properties, if and when a new instability occurs (note that the X-ray emission of IC 3599 in low-state is not constant but still varies, as in the majority of AGN).If, instead, there was an underlying strictly periodic process behind the outbursts, like variants of binary SMBH models where for instance a secondary SMBH interacts with the primary's disk once during its orbit, or binary SMBH models where stream-feeding from a circum-binary disk occurs, then the next outburst is expected around the year 2030.In case IC 3599 will exhibit another X-ray outburst again, we have a pre-approved ToO program with XMM that would allow us to study the inner-most environment of the central black hole.
We would like to thank Swift PI Brad Cenko for approving our continued requests to observe IC 3599 and the Swift Science Operations team for executing the observations.We would also like to thank the anonymous referee for useful comments and suggestions.This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, Caltech, under contract with the National Aeronautics and Space Administration.This work made use of data supplied by the UK Swift Science Data Centre at the University of Leicester (Evans et al. 2007).This research has made use of the XRT Data Analysis Software (XRTDAS) developed under the responsibility of the ASI Science Data Center (ASDC), Italy.This research has made use of data obtained through the High Energy Astrophysics Science Archive Research Center Online Service, provided by the NASA/Goddard Space Flight Center.Software: HEASoft (https://heasarc.gsfc.nasa.gov/docs/software/heasoft/) with XSPEC (Arnaud 1996), ESO-MIDAS (https://www.eso.org/sci/software/esomidas/), the R programming language (https: //www.r-project.org/), and SuperMongo (https://www.astro.princeton.edu/∼ rhl/sm/).  The hardness ratio is defined as HR = hard−sof t hard+sof t with the soft and hard bands in the 0.3-1.0 and 1.0-10 keV bands, respectively, applying the BEHR program by citetpark2006 4 The data from the mini flare in December 2022. 5After flare data from February to May 2023 2 The hardness ratios are as defined in Table 2 3 The Fluxes in the UVOT filters are given in units of 10 −15 W m −2 . 40.3-10 keV fluxes estimated from the course rates based on the latest possible spectral analysis.

Figure 1 .
Figure 1.Long-term 0.2-2.0keV X-ray light curve of IC 3599.The data during the 1990s are from ROSAT, the 2002 observation from Chandra, and all others after 2010 from Swift.Data are available at 10.5281/zenodo.10899673 x med and δ med are the sample median values of the fluxes and the median of the uncertainties.The values x i and δ i are the individual values with their uncertainties.The term δ 2 med δ 2 i is a weighing factor that takes the individual uncertainties into account so that values with large uncertainties carry less weight in the calculations than those values

Figure 2 .
Figure 2. Observed Swift X-ray to optical light curve of IC 3599 from 2010 to August 2023.The fluxes in the X-ray and UVOT bands are given in units of 10 −16 W m −2 .The grey-shaded area displays the range of the 2023 Swift light curve displayed in Figure 3.

Figure 3 .Figure 4 .Figure 5 .
Figure 3. Observed Swift X-ray to optical light curve of IC 3599 from 2022 November to 2023 August.The fluxes in the X-ray and UVOT bands are given in units of 10 −15 W m −2 .

Figure 6 .
Figure 6.Swift XRT 0.3-10 keV flux (in units of 10 −16 W m −2 ) and hardness ratio long-term light curves.The black data points displayed in Figure 2 show detections.The red data points display merged data.
Spectral Energy Distribution (SED) of IC 3599 during the February 2010 outburst and the 2013-2023 low state is shown in Figure 11.It displays how dramatically different the SED of IC 3599 was during the 2010 outburst.The optical/UV to X-ray spectral slope α ox 4 during the outburst was α ox =1.57±0.07 and during the low state α ox =1.79±0.04.Note that the uncertainties on the α ox value during the outburst maybe larger, be-Å)), where l 2500 Å and l 2keV are the luminosity densities at 2500 Å and 2 keV.

Figure 7 .
Figure 7. Swift UVOT long-term light curves.The fluxes are given in units of 10 −15 W m −2 .The black data points display as in Figure 2 each observation with a detection.The red data points display merged data as listed in Table 4.The vertical green lines mark the median values in each of the filters.

Figure 8 .
Figure 8. Upper panel: Swift XRT spectrum of all low state 2013-2023 data merged fitted by an absorbed single power law model.Lower panel: The ratio between the data and the model.

Figure 11 .
Figure 11.The SED of IC 3599 between the outburst in February 2010 (red), the merged 2013-2023 low state data (blue), the mini flare data in December 2022 (magenta), and the merged data after the mini flare (green).Note that the UVOT data during this flare are not shown due to their similarity to the overall low-state data.

Table 1 .
Fractional excess variance and the mean fluxes < f > of the Swift XRT and UVOT light curves including all data and only the low state 2013-2023 data.

Table 2 .
Variability parameter Pvar and the median flux f med of the Swift XRT and UVOT light curves of the low state 2013-2023 data

Table 3 .
Spectral analysis of the high and low state XRT spectra.The high state data were fitted with W statistics, while for the low state data, χ 2 statistics was applied.The redshifted (z=0.021)partial covering absorber column density is given in units of 10 22 cm −2 .

Table 2 .
Merged XRT data as shown in Figure6

Table 3 .
Swift XRT and UVOT fluxes of IC 3599 The full machine-readable table is available on Zenodo: 10.5281/zenodo.10899673.The observed 0.3-10 keV fluxes are given in units of 10 −16 W m −2 .

Table 4 .
Merged UVOT data as shown in Figure7.