Searches after Gravitational Waves Using ARizona Observatories (SAGUARO): Observations and Analysis from Advanced LIGO/Virgo's Third Observing Run

Deep reference images were constructed based on nearly three years of CSS data, with an average limiting magnitude of G ≈ 23.0 mag. Data are run through a custom pipeline in real time that performs image subtraction on each 4 × 30 s median-combined image using ZOGY ²⁵ (an image subtraction algorithm developed for optimal transient detection and flux measurement in the limit of background-dominated noise; Zackay et al. 2016). Candidates with an S/N > 5 are ingested into our database. PSF photometry is performed directly on the subtracted image, following the equations in Zackay et al. (2016), and calibrated to the CSS zero point calculated from Gaia DR2. A graphical summary of the entire candidate vetting process is shown in Figure 2. “Automatic vetting” and simple “human vetting” were performed for triggered SAGUARO fields in real time, with the results reported via GCN to the community. In this paper, we present a more detailed vetting of candidates in the context of the updates discussed in Section 3.

Zoom In Zoom Out Reset image size

We make use of various catalogs and databases, including the Minor Planet Center Orbit Database ²⁶ (MPCORB) to identify known moving objects, and the Transient Name Server ²⁷ (TNS) and the Zwicky Transient Facility (ZTF; Graham et al. 2019) to identify known transients. We also cross-match candidates with the Galaxy List for the Advanced Detector Era (GLADE; Dálya et al. 2018) for possible associations to nearby cataloged galaxies, and the Gaia DR2 catalog (Gaia Collaboration et al. 2018) to reduce contamination by stellar sources (see Section 3). Candidates are then assigned a machine-learning (ML) score, based on the likelihood of being a real astrophysical source. A cut of ML score >0.7 is made before continuing (see Section 3). This process falls under our “automatic vetting” and is done in real-time without human intervention to prepare candidates for “human vetting.”

We then move onto our human vetting through our web interface. We manually remove any remaining artifacts caused by satellite trails, bright stars, or optical reflections. Slow-moving objects that are not removed during the median stacking of the four images or found in the MPCORB were then rejected using our moving object calculator (see Section 3.3). Next, candidates coincident with point-like sources that we infer to be stellar with PanSTARRS1 multiband images (Chambers et al. 2016) that were not removed by the Gaia cross-matching, were rejected as potential kilonova candidates. If candidates are clearly spatially coincident with a galaxy, the redshift of the host galaxy (if known) is compared to the distance inferred for the GW event reported by LIGO/Virgo. Candidates with detections prior to the GW event in either CSS or ZTF were ruled out as being related to the GW event. Candidates with light curves inconsistent with the rise or fall times of kilonovae (see Section 3.5) built from a combination of CSS and ZTF data were rejected. Additionally, candidates already ruled out by the community through spectroscopic or photometric means (reported either through GCNs or papers) were discounted. The remaining candidates are then analyzed in more detail to assess their association with the GW event and provide value-added information. First, individual images are checked to assess shifts in the source's shape and position, as well as the presence of the candidate in each image, to confirm the candidate as a real source. Next, a literature search for spatially coincident galaxies using a radius, r < 4″ (equating to spatial scales of ∼2–13 kpc for distances of 100–500 Mpc) is performed. If no galaxy is found within the literature, we search the DESI Legacy Survey DR8 (Dey et al. 2019) data for a detection or limits (based on nearby detections) on a faint coincident host. For candidates embedded within a galaxy, the position of the candidate and offset, δ r, is reported relative to the position of the galaxy. For galaxies without measured redshifts, we check Pan-STARRS DR2 (Flewelling 2018) for photometric redshifts (z_photo). The final candidates for all events presented here are summarized in Table 2, with the details discussed in depth in the relevant individual event sections (Sections 4.6, 4.8, 4.9, 4.12, 4.16).

In addition to vetting our own candidates, we make use of normal CSS operations to rule out candidates reported by the community that were not followed up and are covered by CSS at a later date. For candidates within the CSS footprint, the positions of these candidates are put through the same automatic and human-vetting procedures as described above to assess their viability as genuine kilonova.

Starting alongside LIGO/Virgo's O3, SAGUARO had overlapping operations for 9.5 months with a duty cycle of 53.6% for the 24 hr window following each event and 78.6% for the 120 hr window after each event. ²⁸ Including all observing constraints, CSS was available to trigger on 27 nonretracted events with δ t < 24 hr (where δ t is defined as the time after the event) and on 38 events within δ t < 120 hr. Our trigger criteria focused on events that contained potential NS components (where the sum of the classifications containing a possible NS, ∑_{(BNS,NSBH,MassGap)} > 20% at the time of trigger) for which CSS could cover part of the 90% probability region when considering airmass and moon constraints described in Lundquist et al. (2019a); but also included unusual events such as S200114f, the “burst” event. Two additional components of importance “HasNS” and “HasRemnant” accompany each event. These describe the probability that the event has an NS and the probability that there is matter outside the final compact object (calculated based on the masses and spins inferred), respectively. With these classifications, we would expect events with large HasRemnant values to produce EM emission. While 38 events were available to trigger on, taking our trigger criteria and the GW sky localization under consideration, we triggered on 8 events. After taking data quality under consideration, we present seven triggered events during O3, including an initial trigger to test our system.

Candidates are given an ML score between 0 (bogus) and 1 (real) based on an ML model to reduce the number of bogus candidates. Following the example of Wright et al. (2015), this model was created using a random forest classifier that was trained on 10 × 10 pixel substamps of a sample that included 3678 detections of known transients and 56,062 bogus detections, including artifacts from bright stars and oversubtractions. To optimize our random forest model, we performed a grid search over the parameter space for the number of trees, the number of samples required to split an internal node, and the number of samples required to be a leaf node. Final values of 1000, 2, and 2 for these parameters were chosen respectively. To reduce overfitting in our ML model and improve generalization, we add Gaussian noise into our training sample data at the level of 1% of the peak brightness. To improve rotation invariance, we add versions of each image in our training sample that have been rotated by 90, 180, and 270 deg, effectively increasing our training set fourfold.

We evaluated the ML scores on a test sample that included 1285 detections of known transients. By selecting a lower limit on the ML score of 0.70, such that we only assess candidates with ML > 0.70, we recovered 94.3% of these known transients with an average score of 0.87. With every classifier, there is a trade-off between the number of false positives and the number of missed detections. Given this recovery rate of known transients, we use this score to maximize the number of real candidates and limit the number of false positives during human vetting.

To reduce the number of known variable stellar sources that produce residuals in subtracted images, all candidates are cross-matched against high proper motion objects in the Gaia DR2 catalog (Gaia Collaboration et al. 2018) upon ingestion. If a candidate matches a high proper motion source (μ > 3σ, where σ is the error on proper motion measurement) in the Gaia catalog, it is classified in the candidate database as a stellar object, as these sources must be stars within the Milky Way, and thus removed from our candidate vetting process. This reduces the number of candidates that need to be human vetted by ∼7.2%.

Candidates are cross-matched against known moving objects from the MPCORB to remove these before human vetting. However, unknown moving objects, especially extremely slow-moving objects that move on the order of a pixel or less between each image and thus are not removed in the median image, can still persist as potential candidates. To help eliminate these slow-moving objects whose small shifts can be difficult to determine by eye, we implemented a moving object calculator. This calculator uses the position of the source in each of the four individual images and determines the shift between the first and each of the subsequent images along with the standard deviation of these shifts. With this method, we are able to identify moving objects that move as little as ∼1 pixel between each of the individual images i.e., ∼16/8 minutes.

While GW 170817 peaked in the r band at δ t ≈ 0.65 days and faded at a rate of ∼1 mag per day (e.g., Cowperthwaite et al. 2017; Kasen et al. 2017; Nicholl et al. 2017; Valenti et al. 2017; Villar et al. 2017), kilonova models describing various progenitor and remnant scenarios predict a diversity in peak times and fading behaviors (e.g., Metzger & Fernández 2014; Kasen et al. 2015; Barbieri et al. 2020; see Section 3.5 and Figure 4). As CSS maps ∼1000 deg² of the sky each night in search of NEOs and PHAs, this provides the opportunity to analyze additional data outside our triggered area and time frame in search of optical counterparts. Thus motivated, we ingest incoming CSS survey data to search for fields that serendipitously fall within the 90% contour of the localization of GW events within δ t < 5 days (i.e., <120 hr). We perform image subtraction between these fields and our deep reference images using ZOGY, and perform the same cross-matching and ingestion as described above. When only considering the triggered fields, SAGUARO covers between 15–180 deg² for each triggered event. The inclusion of these serendipitous fields results in increased coverage of up to 2115 deg² for purely serendipitous events. For events with triggered fields, we see up to ∼9 times in additional area coverage (see Figure 3). In total, we have serendipitous coverage of up to ∼43% of the LVC total probability (P_total) for individual events. Thus, we present an additional 10 events with serendipitous fields, defined as those with coverage in the range of δ t < 120 hr.

Zoom In Zoom Out Reset image size

To place constraints on photometric evolution, for events classified as BNS, NSBH, or MassGap, we compare each candidate to a number of kilonova models that represent a large range of progenitor and remnant scenarios. To describe the data of GW 170817/AT 2017gfo, we choose the two-component model described in Kasen et al. (2017), which we smooth with a Savitsky–Golay filter. For NSBH events, we compare to a model by Kawaguchi et al. 2020 parameterized by an ejecta mass M_ej = 0.06M_⊙. ²⁹ We correct for the ∼0.2 mag artificial excess brightness (determined through private communication with the authors) due to the use of a restricted line list in calculating the ejecta opacity and smooth using a Savitsky–Golay filter. We also compare to three BNS models with resulting hypermassive NS (HMNS) remnant lifetimes of 0 ms (prompt formation of a BH), 100 ms, and infinite (corresponding to an indefinitely stable remnant; Kasen et al. 2015). The overall effect of a long-lived NS remnant, as opposed to prompt collapse to a BH, is that the remnant radiates neutrinos into the postmerger disk, resulting in a bright and blue slow-rising kilonova (Metzger & Fernández 2014; Perego et al. 2014; Metzger 2017). Finally, to describe a massive merger event like GW 190425 (with a chirp mass of Abbott et al. 2020a), we choose two models. The first model's ejecta masses are ≳0.1M_⊙ with the DD2 equation of state, representing an optimistic transient brighter than AT 2017gfo (Barbieri et al. 2020). The second assumes ejecta mass M_ej = 0.04M_⊙, velocity v_ej = 0.15c, and lanthanide fraction X_lan = 10⁻², and predicts a kilonova less luminous than AT 2017gfo across all bands (Foley et al. 2020). We show these models in rest-frame time units (δ t_rest) in Figure 4, along with the absolute magnitude of our 5σ transient depth for individual fields. Even for the slowest rising model, given by an indefinitely stable remnant, the peak of emission occurs at δ t ∼ 2.5 days. We can thus use the rise times to rule out candidates that continue to rise for δ t ≫ 3 days. Given the time evolution of all models, we can also use the fading time to rule out candidates that have detections on longer timescales, i.e., for candidates detected months after the GW event. In general, SAGUARO is able to search to depths that would allow the detection of kilonova emission out to ∼150–400 Mpc, assuming detection at peak brightness and depending on the model assumed (see Figure 4).

Zoom In Zoom Out Reset image size

A candidate GW signal was identified on 2019-04-08 at 18:18:02.288 UTC using data from L1, H1, and V1 (LIGO Scientific Collaboration & Virgo Collaboration 2019f). The final classification for this event is >99% BBH at a distance, d = 1473 ± 358 Mpc. This event, as the first of the O3 alerts, served as a test for our system, and we triggered three fields. The initial follow-up for these triggered fields is discussed in Lundquist et al. (2019a). No serendipitous fields were covered by CSS within our search interval of δ t < 120 hr. The localization of the event, along with the CSS fields that fall within the 90% probability region, is shown in Figure 5.

A total of 11 candidates were initially reported by the community via GCN (MASTER-Network, Pan-STARRS, Gaia; Lipunov et al. 2019a; Smith et al. 2019a; Kostrzewa-Rutkowska et al. 2019a; Lipunov et al. 2019b; Kostrzewa-Rutkowska et al. 2019b). Initial follow-up, either through photometry or spectroscopy, ruled out four of these as viable candidates (Valeev et al. 2019b; Jonker et al. 2019b; Kostrzewa-Rutkowska et al. 2019c; Onori et al. 2019; Chen et al. 2019). SN 2019daj was found to have detections prior to the GW event (Smartt et al. 2019a) and later classified as a Type Ia SN (Burke et al. 2019). Later follow-up classified another candidate, SN 2019dma, as a Type II SN (Prentice et al. 2019). Another candidate, AT 2019dpg, was detected by routine ZTF survey operations out to ∼35 days after the GW event (Nordin et al. 2019). Having faded by <1.5 mag over a month, we consider it to be unrelated to the GW event based on photometric evolution. Searching public ZTF data, we find that another candidate, AT 2019deu, shows variability for over a year after the GW event (Masci et al. 2019). We thus consider it to be unrelated to the GW event. Of the remaining four candidates, CSS observed the positions of AT 2019doo, AT 2019dmc, and AT 2019dev. We found no detections down to a limit of 20.8, 21.0, and 21.2 mag at δ t = 74.7, 65.7, and 65.8 days, respectively. With these limits, we are not able to rule out these as potential candidates. No other viable candidates were found within our data.

A candidate GW signal with classification >99% BNS was identified on 2019-04-25 at 08:18:05.017 UTC using data from L1 and V1 (LIGO Scientific Collaboration & Virgo Collaboration 2019g). This event was later confirmed to be a real BNS merger with a total mass of that occurred at (Abbott et al. 2020a). We triggered 12 fields in the northern part of the localization with δ t < 24 hr. The initial follow-up of these triggered fields are discussed in more detail in Lundquist et al. (2019a). In total, we obtained additional serendipitous coverage of 49 fields (245 deg²) at δ t < 96 hr. Including the triggered fields, the total probability coverage for this event was P_total = 6.55% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.2.

A total of 54 candidate optical counterparts were reported by the community from ZTF, ATLAS, Pan-STARRS, Swift/UVOT, and Gaia (McBrien et al. 2019a; Smith et al. 2019b, 2019c; Kostrzewa-Rutkowska et al. 2019d, 2019e, 2019f, 2019g; Kasliwal et al. 2019; Breeveld et al. 2019; Anand et al. 2019). Hosseinzadeh et al. (2019) compiled a list of all public follow-up searches that were reported for this event at the time, including these 54 candidates and an additional 15 pre-event transients reported by Kasliwal et al. (2019). Of these 54 candidates, 6 were excluded after spectroscopic classification (Short et al. 2019a; Jonker et al. 2019a; Perley et al. 2019b; Buckley et al. 2019b; Pavana et al. 2019; Nicholl et al. 2019), 8 had redshifts inconsistent with the inferred LIGO/Virgo distance (Short et al. 2019b; Anand et al. 2019), while an additional 9 were detected prior to the GW event (Coughlin et al. 2019a; Andreoni & Bellm 2019a; Andreoni et al. 2019c). Five of these 54 candidates had marginal detections (McBrien et al. 2019a), with follow-up imaging failing to confirm these as real candidates (Perley & Copperwheat 2019; Steeghs et al. 2019). Later follow-up by Coughlin et al. (2019b) ruled out an additional four candidates due to photometric evolution that was inconsistent with a kilonova. Routine survey operations by Pan-STARRS resulted in photometric coverage of an additional candidate, AT 2019eby, ∼1 month after the GW event (Chambers et al. 2016). Having faded by <0.4 mag over a month, we consider it to be unrelated to the GW event based on photometric evolution. Of the remaining 21 candidates, the positions of 20 were covered by CSS survey operations over the subsequent months. Only one candidate, AT 2019ech, was identified, just below our automatic vetting S/N threshold (with S/N ≈ 4.2) and 5σ transient depth of 21.0 mag, ∼17 days later. Having faded by <0.4 mag, we are able to rule it out as a viable candidate. None of the other candidates were detected in our data. We find no new candidates within our serendipitous data.

GW 190426_152155 was identified as a candidate GW event on 2019-04-26 15:21:55.337 UTC in data from H1, L1, and V1 (Ligo Scientific Collaboration & VIRGO Collaboration 2019h). This event is classified as a MassGap and is either an NSBH or BBH with masses of and at Mpc (Ligo Scientific Collaboration & VIRGO Collaboration 2019a; Abbott et al. 2020b). We triggered 12 fields for this event in the interval δ t = 24–48 hr (Table 1). The initial follow-up of these triggered fields are discussed in more detail in Lundquist et al. (2019a). No additional fields within the 90% localization were covered by CSS during normal operations over the subsequent 120 hr after the detection. The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.3.

A total of 30 optical candidate counterparts were reported by the community from LCOGT, DECam/GROWTH, ZTF, GRAWITA, and Gaia (Arcavi et al. 2019a; Izzo et al. 2019a; Coughlin et al. 2019c; Perley et al. 2019c; Andreoni et al. 2019e; Kostrzewa-Rutkowska et al. 2019h, 2019i). Hosseinzadeh et al. (2019) compiled a list of all public follow-up searches that were reported for this event at the time. Of these candidates, five were spectroscopically classified and excluded as kilonova candidates (Valeev et al. 2019c; Hu et al. 2019d), three were detected prior to the GW event (Andreoni et al. 2019a; Arcavi et al. 2019b; Andreoni & Bellm 2019b), and one had a redshift inconsistent with the inferred LIGO/Virgo distance (Perley et al. 2019c). Later follow-up by Kasliwal et al. (2020) ruled out 10 candidates based on spectroscopic classification, photometric evolution, or association with a stellar source or artifact. CSS coverage of one of these sources, ZTF 19aaslszp/AT 2019snj, reveals a rising light curve ∼1 month after the GW event, consistent with the finding of Kasliwal et al. 2020 that it is unrelated to the GW event. Seven DECam/GROWTH candidates discussed in Goldstein et al. (2019) were initially ruled out as they were outside the updated LALInference localization; however, GWTC-2 included a new localization that had these candidates within the 90% region. Four of these candidates, DG19ftnb, DG19kqxe, DG19nmaf, and DG19zyaf, were covered by CSS ∼1 week after the event. None were detected to a limit of ∼21.0 mag. DG 19zdwb was covered by CSS ∼40 days after the event and was not detected down to a limit of ∼20.9. The other two candidates, DG 19ouub and DG 19vkgf, were only covered by CSS operations ∼278 days after the GW event, with nondetections to a limit of 21.6 and 22.1 mag, respectively. With these limits, we are not able to rule out any of these as potential candidates. None of the remaining four candidates were covered by CSS, and no new candidates were found within our data.

GW 190521 was identified as a candidate GW signal using data from H1, L1, and V1 on 2019-05-21 03:02:29.447 UTC (Ligo Scientific Collaboration & VIRGO Collaboration 2019i) and later confirmed to be a BBH merger of two unusually high mass components of and (Abbott et al. 2020b, 2020e), at Gpc (Ligo Scientific Collaboration & VIRGO Collaboration 2019j). There are no triggered fields for this event, and our total serendipitous coverage is seven fields (35 deg²) in the interval δ t = 96–120 hr, with P_total = 2.27% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.4.

Graham et al. (2020) reported the detection of a potential counterpart, ZTF 19abanrhr, which they found to be consistent with a kicked BBH merger inside the accretion disk of an active galactic nucleus (AGN). Ashton et al. (2020), however, found insufficient data to confidently associate this optical counterpart with the GW event. They determined the odds (defined as the ratio of the probabilities between the scenario where the AGN is caused by the BBH and a random coincidence) for this event to be 1–12, depending on the waveform model used, compared to the 10⁶ found for GW 170817 and AT 2017gfo. The location of ZTF 19abanrhr and host, J124942.3+344929, is within the CSS footprint; however, we have no data for the period over the duration of the flare (which started at δ t ∼ 50 days and had a a duration of ∼50 days). Our closest coverage was at δ t ∼ 47 days, 3 days before the approximate start of the flare, which can be used to help constrain the start of the flare and preflare activity. At this time, we find no detection of a transient or significant change in magnitude for the galaxy relative to the deep reference to a limit of 20.7 mag. Over the course of 53 epochs prior to the event, the source maintained a constant flux within 1σ of 19.00 ± 0.19 mag. Additionally, Podlesnyi & Dzhatdoev (2020) reported limits from a search for high-energy γ-rays in publicly available Fermi-LAT data in the 100–300 MeV range for this candidate. No other candidates were reported by the community, and a search of our serendipitous data found no candidates within the covered fields.

A candidate GW signal was identified using data from L1 and V1 on 2019-06-30 at 18:52:05.180 UTC (Ligo Scientific Collaboration & VIRGO Collaboration 2019k). The final classification for this event, GW 190630_185205, is a BBH merger with masses of and at Mpc (Abbott et al. 2020b). There are no triggered fields for this event. In total, we covered 26 fields serendipitously with δ t < 120 hr, totaling 130 deg², with P_total = 25.29% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.5. Only a single Gaia candidate was reported within the localization (Kostrzewa-Rutkowska et al. 2019j); however, no additional follow-up was conducted. This candidate was not covered by CSS. No new candidates were found within our serendipitous data.

A candidate GW signal was identified using data from L1 and V1 on 2019-09-01 at 23:31:01.838 UTC (Ligo Scientific Collaboration & VIRGO Collaboration 2019l). The final classification for this event, S190901ap, is 86% BNS at d = 241 ± 79 Mpc, with a 14% probability of being terrestrial in nature. We triggered 12 fields (60 deg²) at δ t < 24 hr. Additionally, we covered 97 fields serendipitously (485 deg²) in the interval δ t = 24–96 hr. In total, we observed 109 fields with δ t < 96 hr, totaling 545 deg² with P_total = 5.79% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.6.

Initially, 19 candidates were reported by the community via GCN (ZTF, Stein et al. 2019b, 2019c; Kool et al. 2019; MASTER-Network, Lipunov et al. 2019c; GOTO, Ackley et al. 2019; Gaia Kostrzewa-Rutkowska et al. 2019k). Of these, 17 were subsequently ruled out as possible kilonova candidates by the community due to association with a galaxy at a distance outside the inferred LIGO/Virgo distance range, spectroscopic classification, or photometric evolution (Burdge et al. 2019a, 2019b; Izzo et al. 2019b; Salmaso et al. 2019; Rosell et al. 2019; Kumar et al. 2019; Wei et al. 2019; Nascimbeni et al. 2019; Kankare et al. 2019; Kasliwal et al. 2020). We find a detection ∼1 month later for ZTF 19abvjnsm, initially ruled out by an inconsistent photo z (Kool et al. 2019), confirming it to be unrelated to the GW event due to inconsistent photometric evolution. Likewise, we also find a detection ∼1 month later for ZTF 19abwsmmd/AT 2019pnc, consistent with the findings of Kasliwal et al. 2020, which ruled it out as a candidate due to a slow decline rate. Of the two candidates not followed up by the community, we find a detection ∼1 month later for Gaia 19dzi/AT 2019piw, showing inconsistent photometric evolution for a kilonova and thus ruling it out as a candidate.

During our search, we found a single new candidate, SAGUARO19k/AT 2019aaid (Paterson et al. 2020) at R.A. = 04^h06^m262 and decl. = −12^d01^m13^s (see Table 2). Performing PSF photometry on the subtracted image, we obtain m_can = 20.95 ± 0.27 mag at δ t_rest = 0.48 days, making it consistent with an AT 2017gfo–like kilonova when transforming to absolute magnitude space using the inferred distance from LIGO/Virgo. However, the candidate only has a single detection, with no pre-event data and no additional coverage of the field by CSS until three months after the event, by which time the source is no longer detected to a limit of 20.6 mag. Searching public ZTF data, we find nondetections to the limits of r = 20.42 and 20.27 mag at δ t ∼ 4.5 and 7.5 days, respectively, and g = 20.36 mag at δ t ∼ 7.5 days. AT 2019aaid is isolated with no cataloged galaxy at the position of the source and no coincident galaxy detected in our images. Searching the Legacy Survey data, we find no source at the position of this candidate with limits based on nearby sources of g > 25.56, r > 24.62, and z > 24.50 mag. With no constraints on the photometric evolution, we are not able to concretely conclude that this candidate is related or unrelated to the GW event.

S190923y was identified as a candidate GW signal using data from H1 and L1 on 2019-09-23 12:55:59.646 UTC (Ligo Scientific Collaboration & VIRGO Collaboration 2019m). The final classification for this event is 68% NSBH merger at d = 438 ± 133 Mpc and a 32% probability of being terrestrial. There are no triggered fields for this event. In total, we covered nine fields serendipitously in the interval δ t = 96–120 hr, totaling 45 deg² with P_total = 4.05% (Table 2). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.7. A single candidate within the localization was reported by the MASTER-Network (Lipunov et al. 2019d), but spectroscopic follow-up ruled it out due to its Galactic origin (Buckley et al. 2019a). Searching our serendipitous fields, we find no new candidates.

GW 190930_133541 was identified as a candidate GW event on 2019-09-30 13:35:41.247 UTC using data from L1 and H1 (Ligo Scientific Collaboration & VIRGO Collaboration 2019n). The final classification for this event is a BBH with masses of 7.8 and 12.3 M_⊙ at d = 709 ± 79 Mpc (Abbott et al. 2020b). We triggered 12 fields (60 deg²) at δ t < 24 hr. From CSS operations, we have an additional 55 fields (275 deg²) of serendipitous coverage with δ t < 72 hr. In total, we covered 67 fields with δ t < 72 hr, totaling 335 deg² with P_total = 13.13% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.8.

Only two candidates were initially reported, both by the MASTER-Network (Lipunov et al. 2019e, 2019f). Both were reported as likely Galactic CVs, but only one was followed up (ATLAS, Smartt et al. 2019b). Searching through our data, we find a single new candidate, SAGUARO19n/AT 2019aaig (Paterson et al. 2020) at RA = 21^h53^m140 and Dec = −06^d43^m38^s (see Table 2). Performing PSF photometry on the subtracted image, we obtain m_can = 20.60 ± 0.23 mag at δ t_rest = 2.17 days. As a BBH event, however, we do not expect EM emission. Nevertheless, comparing AT 2019aaig to the kilonova models discussed here, we find that AT 2019aaig has a luminosity of 8.75 × 10⁴² erg s⁻¹ at the median inferred distance of the GW event, making it ∼3 mag brighter, or 20 times more luminous, than the brightest model at δ t_rest = 2.17 days. Searching our data, we find a nondetection to the limit of 21.2 mag ∼12 days prior to the event and a nondetection to the limit of 20.9 mag seven months after the event. Searching the literature, we find AT 2019aaig is coincident with the galaxy SDSS J215314.00–064338.3 (Alam et al. 2015). SDSS J215314.00–064338.3 has a z_photo = 0.208 ± 0.030, within 3σ of the inferred LIGO/Virgo distance for this event. Searching Pan-STARRS DR2, we find z_photo = 0.168 ± 0.041, in agreement with the SDSS z_photo, and in better agreement with the inferred LIGO/Virgo distance. We find δ r ∼ 05 between the position of AT 2019aaig and the position of the galaxy from SDSS DR12.

S190930t was identified as a candidate GW event on 2019-09-30 14:34:07.685 UTC, nearly one hour after GW 190930_133541 was detected Ligo Scientific Collaboration & VIRGO Collaboration (2019o). S190930t was only identified in data from L1 and thus has a very large localization. The final classification for the event is 74% NSBH at d = 108 ± 38 Mpc. As a result of the large sky localization for this event, no specific fields were triggered. In total, our serendipitous data covered 425 fields with δ t < 120 hr after the event, totaling 2125 deg² with P_total = 9.13% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.9.

Eleven candidates were reported by the community (MASTER-Network; Lipunov et al. 2019e, 2019f, ZTF; Stein et al. 2019d, ATLAS; Smartt et al. 2019b, Swift/UVOT Oates et al. 2019; Tohuvavohu et al. 2019), including the two candidates from the GW 190930_133541 localization (see Section 4.8). Six candidates were followed up in detail and ruled out as possible candidates due to spectroscopic classification (Dahiwale et al. 2019; Karambelkar et al. 2019; Dahiwale & Fremling 2020), inconsistent photometric properties (Smartt et al. 2019b), or associated distances outside the inferred LIGO/Virgo range (Tohuvavohu et al. 2019). One of the remaining candidates, AT 2019rpt, was observed several times by CSS after the GW event. Having faded by ∼2.5 mag over ∼136 days, we consider it to be unrelated to the GW event based on photometric evolution. None of the remaining six candidates were detected by CSS.

Searching through our serendipitous data, we find three new candidates (see Table 2). The first candidate, SAGUARO19l/AT 2019aaie (Paterson et al. 2020), is at R.A. = 20^h40^m052 and decl. = −01^h39^m09^s. Performing PSF photometry on the subtracted image, we obtain m_can = 20.90 ± 0.26 mag at δ t_rest = 1.49 days, making it consistent with the Kawaguchi et al. (2020) NSBH model discussed here when transforming to absolute magnitude space using the inferred LIGO/Virgo distance. Searching our data, we find a nondetection to the limit of 20.82 mag one month prior to the event and a nondetection to the limit of 20.96 mag eight months after the event. Searching the literature, we find AT 2019aaie is coincident with a galaxy (SrcID: 472567686234) in the VISTA Hemisphere Survey DR4.1 (VHS; McMahon et al. 2013). Searching the Legacy Survey, we find the coincident galaxy (Brick: 3101m017, ObjID: 7531) with g = 22.49, r = 21.49, and z = 20.88 mag. We find δ r ∼ 03 between the position of AT 2019aaie and the position of the galaxy from the Legacy Survey. The second candidate, SAGUARO19m/AT 2019aaif (Paterson et al. 2020), is at R.A. = 20^h15^m458 and decl. = −07^d55^m44^s. Performing PSF photometry on the subtracted image, we obtain m_can = 18.63 ± 0.22 mag at δ t_rest = 2.44 days. Comparing the absolute magnitude of AT 2019aaif to the kilonova models discussed here, we find that AT 2019aaif overlaps with the Kasen et al. (2015) HMNS model when considering the error on the median inferred LIGO/Virgo distance. Searching our data, we find a nondetection to the limit of 20.97 mag three months prior to the event and a nondetection to the limit of 21.05 mag eight months after the event. AT 2019aaif is isolated with no cataloged galaxy at the position of the source. No coincident galaxy is detected in our images, and it is not covered within the Legacy Survey footprint. The third candidate, SAGUARO19o/AT 2019aaih (Paterson et al. 2020), is at R.A. = 23^h22^m034 and decl. = −23^d15^m11^s. Performing PSF photometry on the subtracted image, we obtain m_can = 20.28 ± 0.31 mag at δ t_rest = 4.49 days. Comparing the absolute magnitude of AT 2019aaih to the kilonova models discussed here, we find that AT 2019aaih overlaps with the Kasen et al. (2015) HMNS model when considering the error on the median inferred LIGO/Virgo distance. Searching our data, we find only one other observation a year after the event, with a nondetection to the limit of 20.55 mag. AT 2019aaih lies in the outskirts (with δ r = 79) of the GLADE galaxy PCG 802942, at ∼330 Mpc (z_photo = 0.07 ± 0.03; Bilicki et al. 2014). This z_photo lies within 3σ of the inferred LIGO/Virgo distance. Searching Pan-STARRS DR2, however, we find z_photo = 0.113 ± 0.008, >8σ from the inferred LIGO/Virgo distance. Given the disagreement between the z_photo for the host and the general uncertainties associated with the z_photo calculations, we do not rule AT 2019aaih out as a potential candidate.

The GW candidate S191205ah was discovered using data from H1, L1, and V1 at 21:52:08.569 UTC by the gstlal pipeline (Ligo Scientific Collaboration & VIRGO Collaboration 2019p). The final classification is 93% NSBH at d = 385 ± 164 Mpc. We triggered 12 fields (60 deg²) at δ t < 24 hr and have an additional 45 serendipitous fields (225 deg²) with δ t < 96 hr. In total, we covered 57 fields with δ t < 96 hr, totaling 285 deg² with P_total = 10.91% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.10.

During the real-time analysis of our data, we reported a single SAGUARO candidate, SAGUARO19j (Paterson et al. 2019b). It was later found to have a previous detection in Pan-STARRS (McBrien et al. 2019b) and was subsequently classified as an SN (Yan 2020). It was thus ruled out as being related to this GW event. An additional 13 candidates were reported by the community via GCN (ZTF; Andreoni et al. 2019f, MASTER-Network Lipunov et al. 2019g, 2019h, Gaia Kostrzewa-Rutkowska et al. 2019l). Of these, nine were ruled out as possible kilonova candidates by the community due to spectroscopic classification or photometric evolution (Valeev et al. 2019a; Hu et al. 2019a, 2019b, 2019c; Valeev et al. 2019d; Kasliwal et al. 2020). One of the remaining four candidates, AT 2019wjr, was detected by ATLAS during routine survey operations two days after the initial detection. Still on the rise ∼7 days after the GW event, we consider it to be unrelated to the GW event based on photometric evolution. The remaining three candidates lie outside the CSS footprint and were thus not covered by us. Searching our data, we find no new additional candidates.

A candidate GW signal was identified using data from H1, L1, and V1 on 2019-12-13 at 04:34:08.142 UTC (Ligo Scientific Collaboration & VIRGO Collaboration 2019q). The final classification for this event, S191213g, is 76.8% BNS at d = 201 ± 81 Mpc and a 23.2% probability of being terrestrial. We did not trigger any fields for this event but covered 28 fields serendipitously with δ t < 72 hr, totaling 170 deg² with a P_total = 2.10% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.11.

In total, 21 candidates were reported in GCNs by ZTF, MASTER-Network, and Pan-STARRS (Stein et al. 2019a; McBrien et al. 2019c; Andreoni et al. 2019g; Denisenko 2019). All candidates were subsequently ruled out and determined to be either SNe (Perley et al. 2019a; Brennan et al. 2019; Castro-Tirado et al. 2019; Vogl et al. 2019), due to AGN activity (Brennan et al. 2019), stellar in nature (Denisenko 2019; Castro-Tirado et al. 2019), or unrelated to S191213g (Andreoni et al. 2019b; Coughlin et al. 2020). Searching through our serendipitous fields, we find no new candidates.

A subthreshold candidate GW signal, with a high terrestrial probability (97.3%), was identified in data from L1 and V1 on 2020 January 5 16:24:26.057 UTC (LIGO Scientific Collaboration & Virgo Collaboration 2020b). The event was not retracted, however, as the data suggested the probability of it being astrophysical in nature was greater than calculated by the real-time processing due to its chirp structure (LIGO Scientific Collaboration & Virgo Collaboration 2020b, 2020c). The offline parameter estimation classifies the event as an NSBH at d = 283 ± 74 Mpc (LIGO Scientific Collaboration & Virgo Collaboration 2020d) with a 90% localization of 7373 deg². There are no triggered fields for this event. CSS serendipitously covered 53 fields with δ t < 48 hr, totaling 265 deg² with P_total = 2.84% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.12.

In total, 24 candidates were reported by ZTF (Stein et al. 2020; Ahumada 2020) and Gaia (Kostrzewa-Rutkowska & Gaia Alerts Team 2020). All candidates were ruled out through spectroscopic classification (Anand et al. 2020c; Hu et al. 2020; Valeev & Font 2020; Castro-Tirado & Font 2020), slow photometric evolution, or identification as a stellar source or moving object Anand et al. 2020c. We detect five of the ZTF candidates in our data, all showing photometric evolution inconsistent with a kilonova, in agreement with Anand et al. (2020c). Searching our serendipitous fields, we find a single new candidate, SAGUARO20h/AT 2020abgt (Paterson et al. 2020), at R.A. = 07^h41^m240 and decl. = 09^d27^m59^s (see Table 2). Performing PSF photometry on the subtracted image, we obtain m_can = 18.86 ± 0.24 mag at δ t_rest = 0.49 days. Comparing the absolute magnitude of AT 2020abgt to the kilonova models discussed here, we find that it is ∼3 mag brighter (a factor of 20 larger in luminosity) than the brightest model at δ t_rest = 0.49 days, using the median inferred LIGO/Virgo distance. Given the uncertainty associated with NSBH models and the unknown parameters of the system, we do not rule out this candidate based purely on the above. Searching our data, we find a nondetection to the limit of 21.31 mag just over one month prior to the event and a nondetection to the limit of 21.07 mag one month after the event. AT 2020abgt is isolated with no cataloged galaxy at the position of the source. No coincident galaxy is detected in our images, and it is not covered within the Legacy Survey footprint.

On 2020 January 14 02:08:18.23 UTC, the coherent Wave Burst (cWB) pipeline (Klimenko et al. 2016) triggered on a candidate unmodeled burst signal, S200114f, using data from the L1, H1, and V1 (LIGO Scientific Collaboration & Virgo Collaboration 2020e). As a burst event, S200114f was detected without a template or prior knowledge of the waveform, with no constraint on the distance. The FAR was 1 per 25.838 yr, and the 90% localization covered 403 deg². We triggered 36 fields (180 deg²) within δ t < 24 hr, covering P_total = 86.30% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.13.

In total, 33 optical candidates were reported by ZTF (Andreoni 2020a, 2020b) and SAGUARO (Lundquist et al. 2020b). Ten of these were classified as SN and excluded (Valeev & Castro-Rodriguez 2020; Cartier et al. 2020; McCully et al. 2020), two (including one SAGUARO detection) were excluded for having previous detections (Lundquist et al. 2020a; Andreoni 2020b), and one was classified as a QSO and excluded (McCully et al. 2020). Subsequent reanalysis of the three remaining original SAGUARO candidates suggests that two (SAGUARO20b and SAGUARO20d) have underlying point sources and are likely the result of stellar variability. Of the remaining 17 candidates, we find a detection of AT 2020aco within our data prior to the event on 2019 November 24, thus ruling it out as being related to the GW event. Searching through our data, we find no new candidates for this event. A more in-depth discussion on the progenitor of this burst event, along with possible models and the efficiency of our search with regard to these “burst” events, will be presented in M. J. Lundquist et al. (2021, in preparation).

S200115j was identified as a candidate GW signal using data from H1, L1, and V1 on 2020 January 15 04:23:09.742 UTC (LIGO Scientific Collaboration & Virgo Collaboration 2020f). The final classification for this event is >99% MassGap at d = 332 ± 78 Mpc. With a HasNS probability of > 99%, this would suggest that the event was a merger between one object of 3 < M/M_⊙ < 5 and an NS with <3M_⊙. The calculated HasRemnant for this event is >99% (LIGO Scientific Collaboration & Virgo Collaboration 2020g), suggesting a high probability of EM emission. There are no triggered fields for this event, and our total serendipitous coverage is 11 fields (55 deg²) in the interval δ t = 72–120 hr, with P_total = 2.64% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.14. Three optical candidates were reported by the community, all from ZTF (Anand et al. 2020a). Two of these candidates had inconsistent redshifts with the inferred LIGO/Virgo distance, while multiple nondetections in optical imaging follow-up of the third failed to confirm it as a source (Ahumada et al. 2020a, 2020b; de et al. 2020; Mazaeva & IKI FuN 2020). Searching through our serendipitous fields, we find no new candidates.

A candidate GW signal, S200128d, was identified using data from H1 and L1 on 2020 January 28 02:20:11.903 UTC (LIGO Scientific Collaboration & Virgo Collaboration 2020h). The final classification for this event is 97% BBH at d = 3702 ± 1265 Mpc. There are no triggered fields for this event, and our total serendipitous coverage is 64 fields (320 deg²) with δ t < 96 hr and P_total = 11.80% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.15.

No counterparts were reported by the community. Searching our serendipitous fields, we find a single new candidate, SAGUARO20i/AT 2020abgu (Paterson et al. 2020), at R.A. = 14^h16^m090 and decl. = −06^d22^m43^s (see Table 2). Performing PSF photometry on the subtracted image, we obtain m_can = 20.47 ± 0.26 mag at δ t_rest = 0.28 days. As a BBH event, however, we do not expect EM emission. Nevertheless, comparing the absolute magnitude of AT 2020abgu to the kilonova models discussed here, we find that AT 2020abgu is ∼6 mag brighter (a factor of 240 larger in luminosity) than the brightest model at δ t_rest = 0.28 days, using the median inferred LIGO/Virgo distance. Searching our data, we find a nondetection to the limit of 20.72 mag just over one month prior to the event and a nondetection to the limit of 20.93 mag roughly three months after the event. Searching the Legacy Survey, we find AT 2020abgu is coincident with a galaxy (Brick: 2140m065, Objid: 6389) with g = 19.51, r = 19.07, and z = 18.73 mag. Searching Pan-STARRS DR2, we find a z_photo = 0.065 ± 0.039. This distance is vastly different from the inferred LIGO/Virgo distance (by >3000 Mpc), but, due to the large errors in the inferred LIGO/Virgo distance, still falls within 3σ of the median. From our images, AT 2020abgu appears to be nuclear in nature with δ r ∼ 17 between the position of AT 2020abgu and the position of the galaxy from the Legacy Survey. The nature of both the merger (BBH and very far) and the candidate (extremely bright if assuming the median LIGO/Virgo distance, the host z_photo, and apparent nuclear position) implies that AT 2020abgu is most likely due to AGN activity, or possibly a tidal disruption event, and not related to the GW event.

S200224ca was identified as a candidate GW signal using data from H1, L1, and V1 on 2020-02-24 22:22:34.406 UTC (LIGO Scientific Collaboration & Virgo Collaboration 2020i). The final classification for this event is >97% BBH at d = 1575 ± 322 Mpc (LIGO Scientific Collaboration & Virgo Collaboration 2020j). There are no triggered fields for this event, and our total serendipitous coverage is five fields (25 deg²) in the interval δ t = 24–72 hr, with P_total = 11.42% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.16.

In total, 27 optical candidates were initially reported by the community (DESGW; Morgan et al. 2020a, 2020c, GRAWITA Grado et al. 2020). Initial follow-up ruled out 14 optical candidates due to photometric evolution consistent with an SN (Morgan et al. 2020a) and thus too slow for a kilonova. While the DESGW candidates are generally too faint for the limiting magnitude of CSS, we are able to identify AT 2020dlt and AT 2020dlu below our automatic vetting S/N threshold (with S/N ≈ 2.2 and 2.9, and 5σ transient depth of 21.6 mag and 21.1 mag, respectively) in images taken 6 days prior to the event. Thus, we are able to rule them out as being related to the GW event. Searching through our data, we find nondetections for the remaining 13 candidates, both prior to and after the event. No new candidates were found within our serendipitous data for this event.

A candidate GW signal, S200316bj, was identified using data from H1, L1, and V1 on 2020-03-16 21:57:56.157 UTC (LIGO Scientific Collaboration & Virgo Collaboration 2020k). The final classification for this event is >99% MassGap at d = 1178 ± 283 Mpc. With a HasNS probability of <1%, this would suggest the event was a merger between one object of 3 < M/M_⊙ < 5 and a BH with > 5M_⊙ (LIGO Scientific Collaboration & Virgo Collaboration 2020l). There are no triggered fields for this event, and our total serendipitous coverage is 20 fields (100 deg²) at δ t < 48 hr and P_total = 73.24% (Table 1). The localization of the event, along with the CSS fields that fall within the 90% region, is shown in Figure 5, subfigure 5.17. No candidates were reported by the community, and no candidates were found in our serendipitous data for this event.