Needle in a Haystack: Finding Supermassive Black Hole-related Flares in the Zwicky Transient Facility Public Survey

Transient accretion events onto supermassive black holes (SMBHs), such as tidal disruption events (TDEs), Bowen Fluorescence Flares (BFFs), and active galactic nuclei (AGNs), which are accompanied by sudden increases of activity, offer a new window onto the SMBH population, accretion physics, and stellar dynamics in galaxy centers. However, such transients are rare and finding them in wide-field transient surveys is challenging. Here we present the results of a systematic real-time search for SMBH-related transients in Zwicky Transient Facility (ZTF) public alerts, using various search queries. We examined 345 rising events coincident with a galaxy nucleus, with no history of previous activity, of which 223 were spectroscopically classified. Of those, five (2.2%) were TDEs, one (0.5%) was a BFF, and two (0.9%) were AGN flares. Limiting the search to blue events, the fraction of TDEs nearly doubles to 4.1%, and no TDEs are missed. Limiting the search further to candidate post-starburst galaxies increases the relative number of TDEs to 16.7%, but the absolute numbers in such a search are small. The main contamination source is supernovae (95.1% of classified events), of which the majority (82.2% of supernovae) are of Type Ia. In a comparison set of 39 events with limited photometric history, the AGN contamination increases to ∼30%. Host galaxy offset is not a significant discriminant of TDEs in current ZTF data, but might be useful in higher-resolution data. Our results can be used to quantify the efficiency of various SMBH-related transient search strategies in optical surveys such as ZTF and the Legacy Survey of Space and Time.


INTRODUCTION
Wide-field optical surveys have recently found new types of transients occurring exclusively in galaxy centers. These transients are thought to be associated with enhanced accretion events onto supermassive black holes (SMBHs). As such, they have the potential to reveal the presence and properties of otherwise inactive SMBHs, as well as constrain physics of accretion and related radiative processes. Notable examples of such transients are optical-ultraviolet tidal disruption events (TDEs) and Bowen Fluorescence Flares (BFFs). Both types of events are characterized by a sudden increase of flux by several orders of magnitude and are thus much more dramatic than the few tens of percent level variability seen in most active galactic nuclei (AGNs), which host SMBHs with steadily accreting material.
A TDE is the result of the disruption of a star by an SMBH (Hills 1975). In such an event, half of the stellar material is expected to accrete onto the SMBH (Rees 1988). For disruptions occurring outside the event horizon (expected for solar-type stars disrupted by SMBHs of masses ≲ 10 8 M ⊙ , for example), the accretion event will be accompanied by an observable flare. Several such flares have been detected in X-rays, as expected from directly observed accretion emission (see Saxton et al. 2020 for a recent review). However, somewhat surprisingly, a class of optical-ultraviolet TDEs has also been discovered (Gezari et al. 2012;Arcavi et al. 2014). These events are mostly blue, with blackbody temperatures of a few 10 4 K lasting for several months to years, and show broad emission lines of H and/or He in their spectra. The emission mechanism leading to these observed properties is a topic of active debate (see van Velzen et al. 2020 and Gezari 2021 for recent reviews).
In addition to their emission mechanism puzzle, optical-ultraviolet TDEs show a peculiar and strong host galaxy preference for post-starburst galaxies (Arcavi et al. 2014;French et al. 2016. This preference is not yet fully understood, but might be related to the spatial distribution and dynamics of various stellar populations in the centers of such galaxies (French et al. 2020). Studying optical-ultraviolet TDEs can thus also help shed light on the stellar dynamics in galaxy nuclei, which are responsible for driving up TDE rates in post-starburst environments (e.g. Madigan et al. 2018).
Like TDEs, BFFs are also blue and show H and He lines in their spectra, leading Tadhunter et al. (2017) to classify the first such observed event as a TDE. However, a second (Gromadzki et al. 2019) and then third (Trakhtenbrot et al. 2019) event showed that their typical spectral line widths are much narrower than those of TDEs, and that their light curves decline much more slowly than those of TDEs. This led Trakhtenbrot et al. (2019) to classify BFFs as a separate observational class. There are now also hints that BFFs occur in previously existing AGN (Makrygianni et al. 2023), meaning that they could be the result of accretion instabilities in an AGN disk, or of a TDE occurring in an AGN and interacting with its existing accretion disk (e.g. Chan et al. 2019).
BFFs are named as such because they exhibit certain emission lines (such as He II 4686Å and N III 4640Å, among others) that are associated with the Bowen fluorescence mechanism (Bowen 1928). In this mechanism, extreme-ultraviolet and X-ray photons excite certain He II transitions, which in turn launch a cascade of transitions observed in the optical and ultraviolet regimes. This process requires the presence of extreme-ultraviolet photons hitting high-density and high-optical-depth material, and is was indeed predicted decades ago to occur in AGNs (Netzer et al. 1985). Since the identification of this mechanism in BFFs, it has also been suggested to occur in some TDEs (Blagorodnova et al. 2019;Leloudas et al. 2019), hinting at a possible connection between the conditions of matter and radiation in these two types of events related to SMBH accretion.
It is clear that studying more TDEs and BFFs is necessary in order to better constrain their nature, emission mechanisms, and the physics they can teach us in rela-tion to SMBHs and their associated accretion processes. However, these events are intrinsically rare. The exact TDE rate remains uncertain, but is likely to be in the range of 10 −5 -10 −4 events per galaxy per year (e.g. Wang & Merritt 2004;Stone & Metzger 2016). The BFF rate is not yet estimated at all, but observationally, they are less common than TDEs (this could be due in large part to selection effects, as discussed below). In addition to their intrinsic rarity, finding TDEs and BFFs is also observationally challenging. As events that occur in galaxy centers, their detection is contaminated by image-subtraction artifacts, "regular" AGN activity, unresolved non-SMBH-related transients, and even variable stars that cannot be easily distinguished from distant or compact galaxies. For these reasons, only a few dozen TDEs and a few BFFs have been identified so far.
Attempts have been made to devise selection criteria to weed out such transients from the large alert streams produced by wide-field transient surveys. Such criteria typically include selecting candidates by the significance of the flare (since TDEs and BFFs are luminous), color (since TDEs and BFFs are blue), and host properties (since TDE hosts are mostly quiescent). Hung et al. (2018) searched a set of 493 nuclear transients (0. ′′ 8 from their host galaxy center) from the intermediate Palomar Transient Factory, for events with g − R < 0 mag residing in galaxies with u − g > 1 mag and g − r > 0.5 mag. These cuts reduced the set of candidates to just 26, of which two are TDEs. Still, the contamination fraction is large. A substantial amount of telescope time is required to vet 13 candidates (through spectroscopy or ultraviolet colors) for each bona fide TDE.
One way to further reduce the amount of transient contamination in TDE searches is to focus the search on galaxies most similar to the post-starburst hosts that TDEs seem to prefer. French & Zabludoff (2018) used galaxies from the Sloan Digital Sky Survey (SDSS; York et al. 2000) Data Release 12 main galaxy survey (Strauss et al. 2002;Alam et al. 2015), with similar spectral properties as those of actual TDE hosts, to train a machinelearning algorithm to identify such galaxies from photometry alone. They then used this algorithm to identify several tens of thousands of TDE host galaxy candidates in archival survey data. Arcavi et al. (2022c) found indeed that using the French & Zabludoff (2018) catalog of galaxies reduces contamination by roughly a factor of 3-50 (depending on the subset of galaxies used from the catalog) compared to filtering just on quiescent galaxies, and that the only contaminant transients in such galaxies are Type Ia supernovae (SNe). That study, however, was based on archival data alone.
Here we perform a systematic real-time search for TDEs, BFFs, and other possible SMBH flares in the Zwicky Transient Facility (ZTF; Bellm 2014; Graham et al. 2019) alert stream, as parsed by the Lasair broker (Smith et al. 2019). We use various search criteria that rely on candidate brightness and color and their host galaxy properties, and compare their effectiveness in selecting TDEs and BFFs against actual spectroscopic classifications obtained by us and by the rest of the community. We focus on rising events (i.e. events discovered before their peak), selected using visual inspection, for two main reasons. First, such events are more scientifically valuable, as they include the peak time and brightness, as well as the early pre-peak emission, both of which contain important information for constraining models of SNe and TDEs. In addition, rising events present a way of decreasing the number of events to a more manageable subsample for spectroscopic classification, while avoiding biasing the sample toward a particular class (the selection of rising events is done before their classifications are known). 1 Our goal is to quantify the contamination fraction for the various search criteria and to check whether any of them miss TDEs and BFFs. Here, we do not constrain the intrinsic rates of SNe, TDEs, or BFFs in nature, but rather the observed fractions of events, to help guide searches in ongoing and future transient surveys and to help prioritize limited spectroscopic classification resources. We detail our search criteria in Section 2, present and analyze our results in Section 3, and discuss them and conclude in Section 4.

METHODS
We searched the ZTF real-time alert stream for transients in galaxy centers every day between 2020 November 3 and 2022 March 6 (UT dates), with the exception of a ∼2-month break due to a ZTF technical outage between 2021 December 5 and 2022 February 17. In total, our search includes alerts from 414 days. We used the custom query builder on version 1.0 of the Lasair broker 2 to filter the alerts. Lasair uses a contextual classifier called Sherlock 3 . Sherlock is a boosted decision tree algorithm that provides an initial classification of every nonmoving object by performing a spatial crossmatch against data from historical and ongoing astronomical surveys, including catalogs of nearby galaxies, variable stars, and AGNs (see Section 4.2 of Smith et al. 2020 for more details).
Our queries, which are based on the TDE queries by M. Nicholl on version 1.0 of Lasair 4 , filter ZTF alerts according to the following criteria (for each, we state the corresponding Lasair query condition): 1. The candidate is within a certain threshold distance of the nearest Sherlock catalog source. For 80% of our sample, we choose a threshold of 0. ′′ 5 5 . For the rest, we increased the separation threshold to 1 ′′ to check if this has a strong effect on the results: sherlock classifications.separationArcsec < 0.5 or sherlock classifications.separationArcsec < 1 We found that the value of the threshold has no significant effect on the results (see Appendix B), and therefore analyzed the joint sample of both separation thresholds together to increase our sample size. This condition, regardless of the separation threshold, filters out "hostless" events, i.e. those with no host in the Sherlock catalog.
2. The nearest catalog source is likely a galaxy rather than a star 6 : objects.sgscore1 < 0.5 3. The Sherlock classification of the candidate is either "SN" (Supernova) or "NT" (Nuclear 8. At least one of those detections was no more than 14 days ago (in order to avoid old objects that might already be fading): objects.ncandgp 14 > 1 9. The candidate is more than 10 • away from the Galactic plane (in order to filter stellar flares or variability): objects.glatmean > 10 OR objects.glatmean < -10 Conditions 4 and 5 could introduce a bias against finding BFFs (which might be associated with preexisting 8 A ZTF alert is reported to the brokers with a 30 day history, which may contain prediscovery detections. Lasair marks the time of the first detection in this 30 day history as jdmin. 9 Sporadic false detections in galaxy centers may occur, sometimes years before a real event occurs at the same position, and based on false detections that are later filtered out by the brokers. In such a case, the real event would have an old name from when the false detection occurred years before. Removing such events might thus undesirably filter out interesting candidates. In order to avoid losing many candidates, but still not being inundated with variable sources, we decided to allow events with ZTF18 names (several bad subtractions in 2018 caused false events then; E. Bellm, private communication), while removing those with ZTF17 and ZTF19 names. 10 ncand is the total number of detections from ZTF, which can be either positive or negative subtraction residuals (i.e. a brightening or fading with respect to the reference image). ncandgp counts only 'good and positive' detections, i.e. a positive flux with respect to the reference and having a ZTF machine-learning real-bogus score >0.75. This criterion requires that most detections are good and positive, but allows for one or two light-curve points with poor real-bogus scores if, for example, the transient was detected when it was very young and the subtraction residuals at the earliest epochs have a low real-bogus score due to a relatively low signal-to-noise ratio.
AGNs; Makrygianni et al. 2023) and TDEs occurring in AGNs. However, these conditions are necessary in order to remove "normal" AGN activity, which can otherwise be a major contaminant (see below). In addition to conditions 1-9, we create two variations of the query, each with a different magnitude limit: 10a. The latest g-or r-band magnitude of the candidate is brighter than 19: objects.rmag < 19 OR objects.gmag < 19 10b. The latest g-or r-band magnitude of the candidate is brighter than 19.5: objects.rmag < 19.5 OR objects.gmag < 19.5 The motivation for these variations is due to several spectrographs on dynamically scheduled telescopes (which are ideal for the rapid classification of transients) -such as the Floyds spectrographs (Brown et al. 2013 et al. 2014;French et al. 2016), as implemented in the "E+A Galaxies" watchlist 12 on Lasair. In total we have six queries, which we hereby number as follows: I. Conditions 1-9, with a limiting magnitude <19 (Condition 10a), blue (Condition 11), and in a PS galaxy.
Obviously, these are not independent, with some queries being subsets of others, and all being subsets of Query VI. These queries produced roughly 30 new candidates per day in total, which we inspected manually. Only those showing a coherently rising light curve were marked as candidates of interest. Candidates not obviously rising at discovery were monitored for an extra epoch of ZTF photometry and checked again. Candidates for which it was still not clear whether they were rising or not were monitored for another week. This step removed events that had a flat, varying, or incoherent light curve, which could be due to "normal" AGN variability, the stellar variability of Galactic objects, or artifacts of the ZTF image-subtraction pipeline. In addition, this removed true transients that were already after their peak luminosity and that are not part of our sample as defined here. Of all our filtering steps, this is the most subjective, as it requires visual inspection, rather than some strict criterion for what constitutes a "coherently rising" light curve. However, by checking each candidate during multiple epochs, we aim to make this step as inclusive as possible. In addition, since this step is performed before the classification of the candidate is known, it should not bias the search against a particular type of transient (except extremely rapidly rising events, with rise times ≲3 days). 12 https://lasair.roe.ac.uk/watchlist/321/ After these cuts, we were left with a total of 345 candidates of interest (from our entire 414 day search, i.e. ∼0.83 candidates of interest per day, on average), which we attempted to classify spectroscopically within a few days of discovery.
Version 3.0 of Lasair (also known as "Iris") was released in 2021 March. To improve performance, not all of the information that was available in version 1.0 (such as the full detection histories of all candidate events) was carried over to version 3.0. To check for any differences in query results, we add four more queries that we ran on Lasair 3.0 (Iris) between 2022 April 6 and 2022 August 2 (for a total of 118 days): VII. Conditions 1-9, with a limiting magnitude <19 (Condition 10a).
X. Conditions 1-9, with a limiting magnitude <19.5 (Condition 10b) and in a PS galaxy.
Color information was not available as a query parameter in Iris, therefore here we cannot filter by condition 11 here. We perform the same manual cuts as above and are left with 39 events, which is an average of 0.33 candidates per day. Version 4.0 of Lasair was released in 2022 May, but we do not test it here.
We obtain a total of 345 candidates of interest from Lasair 1.0 (310 of which from using a separation threshold of 0. ′′ 5 in Criterion 1, with the rest from using a separation threshold of 1 ′′ ) and 39 candidates of interest from Lasair 3.0 (all of which from using a separation threshold of 1 ′′ ).
For all those brighter than 19th magnitude, we requested spectra through the Las Cumbres Observatory Floyds spectrographs mounted on the 2 m FTN and FTS telescopes at Haleakala (United States) and Siding Spring (Australia) observatories, respectively. Weather, technical issues, and oversubscription of the telescopes mean that not all the requested spectra were obtained or that some were obtained first by the community and reported to the Transient Name Server (TNS) 13 . We were able to obtain spectra of 83 candidates of interest, taken through a 2 ′′ slit placed on the candidate along the parallactic angle (Filippenko 1982). One-dimensional spectra were extracted and the flux and wavelength were . Fainter targets, accessible from La Silla Observatory, were sent for consideration to the ePESSTO+ collaboration, for classification with the NTT. All of our classification spectra, as well as those obtained by the ePESSTO+ collaboration, were publicly reported to the TNS. Many of our candidates of interest were classified by other members of the community and also reported to the TNS. In total, 246 of our 384 candidates of interest (64.1%) were classified on the TNS. We take these classifications as reported to the TNS and analyze their distribution in the next section.

RESULTS AND ANALYSIS
The full list of our candidates of interest can be found in Table 8 in Appendix C. The redshift distribution of all classified transients with a determined redshift on the TNS (244 events) is presented in Figure 1. 15 While our queries can in principle find TDEs out to a redshift of z ∼0.16 (using our magnitude limit of 19.5 and a TDE typical peak absolute magnitude of -20; van Velzen et al. 2020), the median redshift of our classified candidates of interest is z = 0.069, and all but one of the TDEs are at redshifts z < 0.04 (the most distant TDE, AT 2022csn at a redshift of z = 0.148, is also more luminous than typical TDEs; Y. Dgany et al. in preparation). The reason that most classified events are much closer than our redshift limit is likely because nearby events are typically prioritized for spectroscopic classification over more dis-14 https://github.com/LCOGT/floyds pipeline 15 One classified transient, AT 2022amc, has no redshift determination, since its spectrum consists of a blue continuum with no clearly identifiable lines tant events. At the median redshift, our angular cut of 0. ′′ 5 from the galaxy nucleus corresponds to a physical cut of ∼0.66 kpc (assuming the cosmology of Hinshaw et al. 2013). The distribution of the classifications of our candidates of interest, per query, are listed in Table 1 and presented in Figures 2 and 3. In the interest of simplicity, we consolidate the various SN classifications into one category, which we name "SN". These include SNe of undetermined type, SNe I of undetermined subtype, SNe Ia and their various subtypes, as well as SNe Ib, Ic, Ic-BL, II, IIn, IIb, and superluminous SNe (SLSNe) of Types I and II. A breakdown of the number of events per SN type is available in Table 3 in Appendix A.
For all of our events of interest in Lasair 1.0 (Query VI), we find that the vast majority (95.09%) of classified events are SNe, 2.23% (five events) are TDEs, 0.45% are BFFs, and 0.89% are flaring AGN. 16 The remaining 1.34% consist of one variable star, one event classified as "Galaxy" (which means that it was either an artifact or it faded before the spectrum was obtained), and one classified as "Other". The "Other" event is AT 2022amc, which displays a featureless blue continuum. This could have been a young core-collapse SN or some other hot flare, including an SMBH-related one, such as a TDE or BFF. Unfortunately, no follow-up spectra were posted to TNS or, to our knowledge, published elsewhere, so its nature remains undetermined.
The   Table 1. The numbers in parentheses next to each subplot title denote the total number of events in that query.
The BFF is AT 2021seu (ZTF21abjciua, also named ATLAS21bbfi; Arcavi et al. 2021), also classified by this effort using Floyds (Arcavi 2021). The classification is based on a possible N III / He II emission complex on top of a blue continuum, not seen in an archival SDSS spectrum at that position (Arcavi et al. 2021), and resembling the spectra of BFFs in Trakhtenbrot et al. (2019).
All five TDEs pass the "blue criterion" (Criterion 11) and were found by Query V, but the number of SNe passing this criterion is much smaller, nearly doubling the percentage of TDEs among classified blue transients. Limiting the search to candidates that are both blue and brighter than magnitude 19 at discovery (Query II) keeps all TDEs and further increases their percentage to 5.15%. Looking at events in PS galaxies (Query IV), 17 only one of the five TDEs remains, but it is one 17 Here, we study all events that were both in a PS host and blue.
There were two more transients in PS hosts that were not blue: ZTF20acselme (a Type Ia SN) and ZTF22aabsemf (an unclassified event).
of six (16.67%) classified transients there. This is consistent (to 1.2σ) with the finding of Arcavi et al. (2022c) that 10.0% ± 5.5% of classified transients in PS galaxies should be TDEs. For all of our events of interest in Lasair 3.0 (Query VIII), SNe are still the majority of classified events (72.73%), with the rest all flaring AGNs. The fraction of flaring AGNs in Lasair 3.0 is thus 30 times larger than in Lasair 1.0. Some of this difference is likely explained by the fact that, at least initially, Lasair 3.0 did not provide the full multiyear light-curve history of each candidate. This precluded filtering most AGNs by their historical activity.
In order to quantify the significance of the difference in fractions between queries, we calculate their confidence bounds using the Clopper-Pearson method (Clopper & Pearson 1934). Gehrels (1986) discusses how this method, which uses binomial statistics to estimate lower and upper confidence bounds for ratios, is especially useful for ratios of different event types, when the numbers of observed events are small. The 1σ confidence bounds  calculated with this method (and used hereafter) are shown in Table 2.
The fraction of TDEs in our global Lasair 1.0 query (Query VI) is 2.23%±0.98% and that of BFFs is 0.45%± 0.44. Requiring candidates be blue (Query V), increases the TDE fraction by a factor of 1.84 ± 1.11. Adding the requirement for a PS host (Query IV) increases the TDE fraction by a factor of 7.47±7.53 compared to the global query (Query VI). Without the full light-curve history of Lasair 3.0, the fraction of AGNs there increases by a factor of 30.55±23.86 in Query VIII compared to Query VI.
We wish to check whether the offset of a source from its host galaxy center can be used as a way to better select for TDEs and BFFs. To do that, we retrieved the distnr parameter of each detection of each SN, TDE, and the BFF in our sample using the Automatic Learning for the Rapid Classification of Events (ALeRCE) broker 18 (Förster et al. 2021). The distnr parameter is provided with each detection of a source in the ZTF alert packets 19 . It denotes the distance of that detection to the nearest source in the reference-image PSF catalog (within 30 ′′ ), in units of pixels (which are equal to units of ′′ , since the ZTF pixel scale is 1 pixel per ′′ ). We plot the distribution of these values in Figure 4. TDEs show a slightly lower average distnr value than SNe, but the difference is much smaller than the spread of values of each type of event and hence not significant. Our one BFF actually shows a larger average distnr value than the SNe. However, this is based on detections of a single event, and could be driven by the centroid measurement of its particular host galaxy in ZTF. We conclude that there is no significant difference in the distnr values between SNe, TDEs, and BFFs in ZTF, and therefore that this parameter is not a good discriminant.   Figure 4. Distribution of the distnr parameter, which quantifies the offset of a source from its host galaxy center (shown in a stacked histogram) for all detections of SNe, TDEs, and the BFF. The vertical lines at the top denote the average value for each class. The number of total detections per class of events is shown in parentheses in the legend. While TDEs show a lower average offset compared to SNe, the difference is not large enough to be used a discriminant.
There are 24 additional events that are classified as TDEs on the TNS, from the time period of our search, which do not appear here. Of those, 16 have robust TDE classifications (i.e. at least one public spectrum showing clear broad He II and/or Hα emission and blue colors). The rest have either no public spectrum available, very noisy spectra, or show no clear spectral features. Of the 16 robust TDEs, most (11) have not been identified here because they were deemed not to have a coherent rising light curve at discovery. Three events were missed due to a bug in the queries (which was later fixed), and two did not pass Criterion 7, regarding the number of unreliable detections (one required increasing the number of unreliable detections from <3 to <5, and one required it be increased to <21). However, relaxing Criterion 7 would have likely also led to an increase in contaminant candidates.

DISCUSSION AND CONCLUSIONS
Our results quantify the "needle in a haystack" problem of finding TDEs and BFFs in wide-field transient surveys. We find that the photometric history of candidates is crucial for removing most AGN contamination. Even so, roughly one in 35-45 events is a TDE, and one in 170-220 is a BFF. This sets a significant challenge for identifying these events in current transient surveys, and for identifying even a small subset of the thousands of TDEs expected to be discovered by the Legacy  The fraction of TDEs increases by almost a factor of 2 to roughly one in 20-25 events when selecting only blue (g − r < 0.05) transients. This cut does not remove any TDEs. The fraction further increases by another factor of ∼3 to roughly one in 5-6 events when selecting probable TDE host galaxies. However, such galaxies are rare, making the total number of TDEs discoverable in this way small. Given the huge increase in expected TDE discovery fractions, though, it would be beneficial to update the French & Zabludoff (2018) galaxy catalog and to expand its coverage using new spectroscopic and photometric surveys such as SDSS-V (Kollmeier et al. 2017).
An additional ∼50% of TDEs are found when relaxing Criterion 7 to allow events with a smaller number of reliable detections, and roughly three times as many TDEs are found when relaxing the condition that the event be rising in brightness at discovery. However, the number of contaminants that relaxing such criteria adds is significant. For example, changing the threshold of Criterion 7 from <3 to <5 (which would have added one TDE to the sample) increased the number of daily candidates by a factor of 2-3, and changing it to <21 (which would have added a second TDE to the sample) increased the number of daily candidates by an order of magnitude.
We find no significant difference between the offsets of TDEs vs. nuclear SNe from their host galaxy centers. Therefore, this parameter cannot be used as a discriminant for selecting more likely TDEs, at least not in ZTF, as quantified by the distnr parameter. LSST, with its higher spatial resolution, might be able to make host nucleus offsets a more viable distinguishing parameter.
An additional possible discriminant for selecting TDEs, but which was not tested here, is their ultraviolet to optical colors (e.g. Hung et al. 2018). Obtaining ultraviolet photometry rapidly and for many targets is currently possible almost exclusively with the the Neil Gehrels Swift Observatory (Gehrels et al. 2004) Ultraviolet/Optical Telescope (Roming et al. 2005). Indeed, Hung et al. (2018) have shown that selecting transients by a combination of their ultraviolet to optical colors using Swift and the optical colors of their host galaxies increased the fraction of TDEs to 1 in 4.5. However, Swift is limited in the number of transients it can vet. The upcoming Ultraviolet Transient Astronomy Satellite (ULTRASAT; Sagiv et al. 2014), with its wide-field ultraviolet imager, is expected to obtain ultraviolet photometry for thousands of TDEs. This will be an excellent way to discriminate TDEs from other transients without the need for substantial classification resources.
Another approach is to train machine-learning algorithms to classify transients from photometry alone. This has been done for distinguishing between some SN types (e. Of course, any such filtering based on the photometric properties of the transient or the characteristics of its host galaxy can also bias population studies of TDEs and BFFs. A specific population that almost all current searches (including ours) are biased against is that of slowly evolving transients with years-long evolution.
Such events are already suppressed by the alert mechanisms of most transient surveys, even before reaching the brokers. ZTF alerts, for example, are generated by comparing a new image to a reference image taken up to a few months to a few years earlier. Therefore, events that rise on a time scale of several years will not be much brighter in the new image compared to the reference, and thus an alert might never be issued. Since the set of images used as references is updated from time to time, such transients could remain hidden during the lifetimes of the surveys. Indeed, when Lawrence et al.
(2016) compared images from PS1 to images obtained a decade earlier by SDSS, they found a population of slowly rising nuclear transients. This population is not seen in current transient surveys, which are optimized to find transients that change on shorter time scales.
It will thus continue to be challenging to find TDEs and BFFs in optical transient surveys in an unbiased way, even for events evolving on time scales of days to months. One way forward is to use some combination of well-defined photometric and host galaxy filters, such as those used here. However, making searches as complete as possible will still require ample spectroscopic resources for vetting large numbers of SMBH-related transient candidates.
In order to assess if there is a statistically significant difference between the fractions resulting from the different thresholds, we calculate the 1σ Clopper-Pearson confidence bounds per each separation threshold subsample in Table  6 (for the 0. ′′ 5 threshold) and Table 7 (for the 1 ′′ threshold).
There are no statistically significant differences between the results of the two separation thresholds. It does appear that there is a much higher fraction of TDEs in the subsample of the 1 ′′ threshold (of order 10%) compared to the  Figure 5. Distribution of the separationArcsec parameter from Criterion 1 (shown in a stacked histogram) for all detections of our five TDEs. There is no significant difference in the separations between the TDEs in the 0. ′′ 5 threshold subsample (AT 2020vwl and AT 2021ehb) and the TDEs in the 1 ′′ threshold subsample (AT 2022bdw, AT 2022csn, and AT 2022dbl), meaning that the change of separation threshold does not have a strong effect on the number of TDEs discovered. 0. ′′ 5 threshold (of order 1%). However, this is not statistically significant and is a result of small number statistics. To demonstrate this, we plot all of the separationArcsec values of all the detections of our five TDEs in Figure 5.

C. LIST OF EVENTS
We list in Table 8 the full set of events considered transients of interest in this work, their publicly available classification and redshift, and the query number(s) in which they were found. 3.70% ± 3.61% 11.11% ± 6.01% 0 0 0 3.70% ± 3.61%  Table 8 continued  Table 8 continued     Table 8 continued   Table 8 continued        Table 8 continued b Lower than the expected signal-to-noise ratio in the spectrum attempt.
c Faded before the spectrum was attempted (could have been rapidly evolving or an artifact). Note-Since some queries are subsets of other queries, here we list only the most stringent query that produced each event.