GRANDMA observations of advanced LIGO’s and advanced Virgo’s third observational campaign

ABSTRACT

1 INTRODUCTION

Gravitational-wave (GW) signals detected by Advanced LIGO (Aasi et al 2015) and Advanced Virgo (Acernese et al 2015) since 2015 have revealed the existence of compact binary mergers which included neutrons stars (NSs, Abbott et al. 2017a, 2020a) and black holes (BHs, Abbott et al. 2019e), providing unprecedented new insights into these objects.

During the first and second Advanced LIGO and Advanced Virgo observing runs, O1 and O2 (2015–2017, Advanced Virgo joined at the end of O2), GWs from 10 binary BH (BBH) systems and a binary NS system (BNS) were detected (Abbott et al. 2017a, 2019e). Additional events produced by BBHs were also claimed by other groups analysing the LIGO/Virgo data (Zackay et al. 2019). The GW observations of these compact binary sources have allowed gravity to be probed in the ultrastrong field regime (Abbott et al. 2016), and have yielded new ideas on the formation channels of the systems (Abbott et al. 2019b).

The first GW observation from a coalescing BNS, GW170817 (Abbott et al. 2017a) and the joint observations of GRB 170817A by the Fermi/GBM and INTEGRAL/SPI ACS gamma-ray detectors (Goldstein et al. 2017; Savchenko et al. 2017) firmly established the association between BNS mergers and short-duration gamma-ray bursts (GRBs). The estimation of the source location of GW170817 and its subsequent identification at a distance of 40 Mpc (Coulter et al. 2017) initiated a ground-breaking electromagnetic (EM) follow-up campaign from the X-ray to radio bands. These EM observations provided evidence that a successful jet was launched and that the initial gamma-ray emission were from a structured jet viewed approximately 20° off-axis (Mooley et al. 2018; Ghirlanda et al. 2019).

Furthermore, optical, ultraviolet, and infrared observations of the optical counterpart AT 2017gfo (Coulter et al. 2017) monitored the glow of a ‘kilonova,’ produced by the radioactive decay of r-process elements in the neutron-rich ejecta (Lattimer & Schramm 1974; Li & Paczynski 1998; Kasen et al. 2017) and showed that the source evolved from blue to red over a week (Abbott et al. 2017b; Arcavi 2018). The impact of this discovery motivated a large number of small aperture telescopes to systematically scan the credible regions provided by LIGO/Virgo GW candidate event triggers during subsequent LIGO/Virgo observational runs. The most recent run, O3, which began in 2019 April, was split into two parts to allow for a commissioning break: O3a (2019 April–2019 September) and O3b (2019 November–2020 March 27; somewhat more than a month earlier than planned because of the COVID-19 pandemic).

Since the discovery of the EM counterpart of GW170817 no other counterpart associated with a GW candidate event has been observed despite significant coordinated efforts by numerous observing groups, for example Coughlin et al. (2019c), Hosseinzadeh et al. (2019), Goldstein et al. (2019), Andreoni et al. (2020a), Ackley et al. (2020), and Brunn et al. (2019).

The potential BNS sources¹ reported so far by the LIGO/Virgo network in O3 (a sample of five with a BNS source probability of |$\gt 50{\rm {per \, cent}}$|⁠) have median estimated distances in the range ∼157–241 Mpc with only one source, GW190425 (Abbott et al. 2020), within 200 Mpc.

Additionally, the |$90{\rm {per \, cent}}$| confidence intervals of the sky localizations for this sample cover the range 1131–24226 deg². The combination of large credible regions coupled with the dimmer optical counterparts expected at such large distances have made searches for EM counterparts a highly challenging (and ultimately unsuccessful) endeavour during O3.

When conducting searches of large GW sky localizations with a single instrument, a telescope with a large field of view (FoV) is optimal. However, there is a delicate balance; large credible regions are often the result of more distant sources observed at relatively lower signal-to-noise ratio (S/N) – but sources at greater distances are more accessible to larger diameter telescopes. It is both technically and financially challenging to have a large aperture telescope capable of covering a large FoV. This has motivated the use of multiple telescopes working in unison within global networks for the purpose of identification and characterization of counterparts; see, for example Coughlin et al. (2020b). In addition to GRANDMA (Global Rapid Advanced Network Devoted to the Multi-messenger Addicts), the heterogeneous telescope network described in this paper, other networks such as the Global Relay of Observatories Watching Transients Happen (GROWTH, Coughlin et al. 2019c), MASTER-Net (Lipunov et al. 2010), and KMTNet² have also been conducting counterpart searches during O3, supported by a host of individual facilities such as Pan-STARRS1 (Flewelling et al. 2016), ATLAS (Tonry 2011), DECam (Herner et al. 2017), MeerLICHT (Bloemen et al. 2016), DDOTI (Watson et al. 2020), and GOTO³ (O’Brien 2018). The O3 campaign has also seen the emergence of a variety of new tools and multimessenger platforms to coordinate and optimize the follow-up: galaxy ranking can be optimized through the online catalogue MANGROVE (Ducoin et al. 2020a); the gwemopt open software can be used to schedule GW follow-ups (Coughlin et al. 2018a); and the GW TreasureMap (Wyatt et al. 2020) is a system to report and visualize the progress of searches for EM counterparts to GW events.

In a previous paper (Antier et al. 2020), we published the GRANDMA results obtained during O3a. This current paper continues in a similar vein through the study of candidate events from O3b. The paper is organized as follows: Section 2 outlines the GW candidate alerts received during O3, and Section 3 describes the new telescopes that joined the GRANDMA collaboration for O3b. Section 4 provides details on the adopted observational strategies. Section 5 presents details on the GRANDMA observations, focuses on some specific events, and provides new constraints on the GW progenitors. We finish by presenting our conclusions in Section 6.

2 OVERVIEW ON THE THIRD OBSERVATIONAL CAMPAIGN O3

The third observational campaign organized by LIGO and Virgo, O3, started 2019 April 1 at 15:00 UTC (LIGO Scientific Collaboration & Virgo Collaboration 2019a). After a month-long commissioning break in October, the second period of the campaign finished on 2020 March 27 at 17:00 UTC, for a total of 330 d, short of the goal of a full 365 d, cut short by the COVID-19 pandemic. For comparison, the second observational campaign lasted 269 d and the first observational campaign lasted 129 d (Veitch et al. 2015).

2.1 O3 alerts

During O3, the LIGO Scientific Collaboration and the VIRGO Collaboration (LVC) sent 80 alerts to the astronomical community (and about 100 in total for the O1, O2, and O3 observational campaigns combined). Among them, 24 were false positives resulting in retractions due to data quality issues (see for instance LIGO Scientific Collaboration & Virgo Collaboration 2019p, 2020b, f) and 56 are still considered as potential signals. 52 were classified as GW signals with a probability higher than 50 per cent⁴ via the GCN platform (Gamma-ray burst Coordinates Network⁵) thanks to a new dedicated alert system developed by the LVC.⁶ The alert system was still being debugged for the first two months (April and May), producing a large delay of dozens of minutes between the alert delivery and the GW trigger time (the event UTC time). The delay was typically less than 10 min for the remainder of the run, although some exceptional delays did occur for non-technical reasons, that is S200219ac (LIGO Scientific Collaboration & Virgo Collaboration 2020j), S200105ae (LIGO Scientific Collaboration & Virgo Collaboration 2020a), and S191105ae (LIGO Scientific Collaboration & Virgo Collaboration 2019n) during O3b.

The GW alerts were distributed with three different types of GCN/LVC notices with different distribution delays: preliminary (within 5 min), initial (within 30 min, produced after human vetting), and update notices (within 2–4 d, produced after a refined parameter estimation analysis). The notices contain information such as the trigger time, the online pipeline that generated the trigger, the event false alarm rate, a link to the sky localization probability map (Bayestar, Singer & Price 2016 or LALInference, Veitch et al. 2015), and the estimate of the distance to the source. The alert also contains different source classification information if available for compact binary searches: a BNS [p(BNS)], a BHNS [p(BHNS)], or terrestrial noise [p(terrestrial)] (Kapadia et al. 2020), and an indicator to estimate the probability of producing an EM signature considering the candidate is of astrophysical origin [p(HasRemnant)] (Chatterjee et al. 2019). The GW trigger can be also classified as ‘MassGap,’ completing the possible classifications. In the case of a Mass Gap event, the initial estimate is that at least one of the compact objects has a mass of 3–5 M_⊙, implying it is unclear whether it is an NS or a BH. The other object can be a lighter NS or a more massive BH. Up to the time when this paper was written, only the ‘superevent’ S190425z has been confirmed as a real astrophysically produced GW event, GW190425 (Abbott et al. 2020).

Out of the 52 GW candidates detected in O3, 36 have been classified as likely being emitted from a BBH merger, five from mass gap mergers, five from BHNS mergers, five from BNS mergers, and one was a GW ‘burst’ type trigger (unmodelled search) of unidentified origin (LIGO Scientific Collaboration & Virgo Collaboration 2020d; Klimenko et al. 2016). Except for the case of the burst GW signal, the rest have information on their distance derived from the GW binary coalescence signal modelling. The final event classification can differ from the low latency analysis due to the further study subsequently conducted. This is especially the case for low-mass binaries with one object close to the lower or upper bound of the 3–5 M_⊙ gap that can be reclassified ‘NSBH’, ‘BBH’, or ‘Mass Gap’: S190814bv was given a final classification of 99 per cent ‘NSBH’ (LIGO Scientific Collaboration & Virgo Collaboration 2019j, k), and S190728q as 52 per cent Mass Gap (LIGO Scientific Collaboration & Virgo Collaboration 2019h, i). In addition, the indication regarding probability to be of astrophysical origin has also evolved for some of the GW candidates: S190426c and S190510g now have a preferential origin of being ‘Terrestrial’ (LIGO Scientific Collaboration & Virgo Collaboration 2019b, c, l); on the contrary, S190727h is now considered as an astrophysical event with 92 per cent ‘BBH’ (LIGO Scientific Collaboration & Virgo Collaboration 2019g); S200105ae is still classified as 97 per cent ‘Terrestrial’ but the LVC reported that the low false alarm rate for this trigger was due to only one detector (LIGO Livingston) responding to the event and offline analysis is likely to increase its significance (LIGO Scientific Collaboration & Virgo Collaboration 2020a, c).

The localization area for GW alerts also differs from the initial to the updated analysis for some events (e.g. due to the inclusion of Virgo data in the analysis for S200225q, LIGO Scientific Collaboration & Virgo Collaboration 2020k). These issues will be discussed in Section 2.2. Note that the reports for the distance estimate can also vary (e.g. S200219ac, LIGO Scientific Collaboration & Virgo Collaboration 2020j), but we will not discuss this distance analysis in this paper.

Fig. 1 displays the most recently issued 90 per cent credible localization region produced for a GW candidate versus its most recently updated luminosity distance (posterior mean distance and posterior standard deviation of distance): all detailed numbers can be found in Antier et al. (2020) for O3a and Section 5 for O3b. We split the diagram into four sections, one for each of the O1, O2, O3a, and O3b observational campaigns. We indicate the names of detected events and candidates if their error boxes are smaller than 30 deg² containing one confirmed BNS event (GW170817, Abbott et al. 2017a, 2019e), an NS–BH candidate (S190814bv, LIGO Scientific Collaboration & Virgo Collaboration 2019k), and two BBH candidate events (S200225q, LIGO Scientific Collaboration & Virgo Collaboration 2020k, and S200208q, LIGO Scientific Collaboration & Virgo Collaboration 2020g).

Figure 1.

The most recently updated 90 per cent credible region area versus the most recently updated luminosity distance (posterior mean distance and posterior standard deviation of distance) for all LIGO/Virgo GW events/candidates of runs O1, O2, O3a, and O3b. Vertical red dotted lines are the expected limiting distances to detect BNS mergers by the LIGO/Virgo detectors. BBH, MG, and NSBH are represented in different colours. Four candidates for run O3 are indicated by open symbols corresponding to a predominantly terrestrial classification (above 50 per cent) and the classification indicated in these plots is the second most likely.

Open in new tabDownload slide

For each kind of event, a cumulative distribution function on the distance can be constructed by approximating each measured distance and error bar with a Gaussian function. The combined distribution is shown in Fig. 2, along with the predictions provided by the LVC (Abbott et al. 2018a). There is a clear separation between the BNS and BBH distributions. This is due to the much higher rate of BNS mergers compared to BBH mergers⁷ (Abbott et al. 2019e), and to the much smaller distance at which BNS mergers can be detected by GW detectors compared to BBH mergers. Hence, BBH mergers are detected often at large distances where the large volume compensates for the low rate. In particular, there is no BBH candidate or event detected below 200 Mpc.

Figure 2.

The cumulative distribution function of all LIGO/Virgo GW events and candidates in O3 with a Terrestrial indicator below 50 per cent versus the distance is shown in solid lines, compared to the predictions for these distribution from Abbott et al. (2018a) shown in dashed lines.

Open in new tabDownload slide

The observed distances are larger than the prediction by approximately a factor 2 for BNS and BBH event types. This could be due to the conservative detection threshold on network S/N of 12 used in the prediction described in Abbott et al. (2018a) and by masses for the binary system being larger than those used in the simulations. The study of this discrepancy is beyond the scope of this paper. In particular, the median distance of BNS candidates in the run O3 is 200 Mpc, 5 times further than GW170817. As a consequence, the apparent magnitudes of kilonovae associated with the BNS events of the run O3 are expected to be ∼3.5 mag fainter than for GW170817, assuming the same intrinsic light curve.

The 90 per cent credible region areas can be combined to obtain a cumulative distribution function for the localization area. This is shown in Fig. 3 and compared with the predictions provided by the LIGO/Virgo collaboration (Abbott et al. 2018a). For BBH mergers, the localization area is on average a factor 2 larger than predicted. Similar to the distance distribution, the difference could be due to larger masses and smaller network S/N of the signals detected as opposed to those from the simulations, as both properties are expected to reduce localization performance. For BNS mergers the discrepancy with the prediction is larger by a factor of 10. This is more difficult to explain, but we note that three out of eight BNS candidates have a significant probability (above 50 per cent) of being Terrestrial. GW190425 is the only confident detection from O3 announced by the LVC so far (Abbott et al. 2020), but it was observed by only two GW detectors and with a very low S/N in one of them, which explains its poor sky localization. In comparison with O2, the sky localization areas of the eight BNS candidates produced during the run O3 are spread over a median value of 4500 deg² (i.e. 10 per cent of the sky), 300 times larger than for GW170817, which was observed by three GW detectors.

Figure 3.

The cumulative distribution function of all LIGO/Virgo GW events and candidates in O3 with a Terrestrial indicator below 50 per cent versus the 90 per cent credible localization area is shown in solid lines, compared to the predictions for these distributions from Abbott et al. (2018a) shown in dashed lines.

Open in new tabDownload slide

The large GW alert rate in O3 provides motivation for a first estimate of the rate at which each classification of event (BNS, NS–BH, MassGap, and BBHs); see Abbott et al. (2018a, 2019e) for a discussion giving a more sophisticated approach. Our method is as follows: for each type of event we calculate the mean distance for which they are detected. Next, from this distance we can determine the enclosed volume that it represents, and estimate an event rate. To do this, we divide the number of GW candidates during the observational period by its duration, and divide that result by the volume determined for the type of event as presented in Table 1. A Monte Carlo study shows that this method yields, on average, the correct rate estimate if all events of the same type can be detected up to the same limiting distance (i.e. there is no large dispersion in the chirp masses). This is approximately true for BNS events, but for BBH systems there is a wide distribution in masses which will lead to a large systematic error (Abbott et al. 2019e). We cannot quantify this bias as the chirp mass of the systems are not yet publicly distributed.

2.2 Localization and observational constraints

In order to give insight into the O3 campaign, it is useful to quantify the possibility of observing during the nights following the detection, as well as the benefit of observing after the initial sky localization is released but before the updated sky localization is available. Due to factors such as high airmass and the Sun’s altitude, the possibility of observing from a certain location may be different than others, and so it is important to take this into consideration when attempting to follow-up candidate events. Both initial and updated sky localization areas are available for 41 candidates. Fig. 4 displays the cumulative distribution function for the observable probability within the localization from three representative locations (Palomar, The Cerro Tololo Inter-American Observatory, and Hawaii) for all candidate events during O3. As can be seen in the figure, around 20 per cent of events were not observable at all from at least one of the three locations in the week following detection. We also find that a further |${\sim }16{\rm {per \, cent}}$| of candidates had |$\lessapprox 4{\rm {per \, cent}}$| of the probability observable from all three locations; all of these transients have 90 per cent credible regions spanning less than 300 deg² (except for S190602aq, LIGO Scientific Collaboration & Virgo Collaboration 2019d, for which it spans ∼750 deg²) and were classified to very likely be BBH mergers (aside from S190924h, LIGO Scientific Collaboration & Virgo Collaboration 2019m, which was classified as a mass gap candidate at |$\gt 99{\rm {per \, cent}}$|⁠). Additionally, |$20{\rm {per \, cent}}$| of candidates had more than |$40{\rm {per \, cent}}$| observability from only one of the three sites, demonstrating the utility of coordinating network-level telescope observations to obtain large coverage of the localizations. The importance of obtaining such observations at significant depth and coverage (regardless of whether an EM counterpart is ultimately discovered) cannot be overstated, as it has a direct impact on the ability to constrain the ejecta mass and binary parameters for the event if it is truly astrophysical in origin (Coughlin et al. 2020b).

Figure 4.

Probability contained within the observable region of the sky localization for all GW candidates during O3; this is shown from three different locations (Palomar (33.3563°, −116.8648°), CTIO (−30.1691°, −70.8062°), and Hawaii (19.8968°, 155.5828°); the LALInference sky localization was used as long as it was available for the candidate/event in question.

Open in new tabDownload slide

Fig. 5 presents the distribution of the percent overlap between the 90 per cent credible regions of the initial Bayestar (Singer & Price 2016) and the updated LALInference (Veitch et al. 2015) sky localization areas for 41 candidate events from O3. To find the percent overlap, we compute the intersection in pixels between the two maps and then sum the probability contained in that region using the LALInference sky area. From the results, we find that the events fall into one of two possible categories: a low percentage overlap (∼0–30 per cent) or a high percentage overlap (close to 100 per cent). Around 27 per cent of the events belong to the first category, showing that a significant number of candidates experience a large shift in their 2D probability distribution. This shift is likely due to a variety of factors. For one, some multi-instrument detections did not factor in data from one of the GW detectors when generating the initial Bayestar skymap (e.g. S190720a, LIGO Scientific Collaboration & Virgo Collaboration 2019f), and for other events, the sky area decreased significantly between the two sky localizations (e.g. S200225q, LIGO Scientific Collaboration & Virgo Collaboration 2020k, with the inclusion of Virgo, and S190630ag, LIGO Scientific Collaboration & Virgo Collaboration 2019e). In addition, disagreement between the time delays of the detected signals for the event and the results of the parameter estimation, as well as unanticipated issues (e.g. the occurrence of scattered light glitches in some of the detectors in the case of S191213g, LIGO Scientific Collaboration & Virgo Collaboration 2019o, which possibly affected the sky position) may have led to the low percentage overlap for some events.

Figure 5.

Overlap ratio between the initial and the most up-to-date 90 per cent credible regions of all GW candidate events during O3 (given that both sky localizations were available). Note that the overlapping regions can be discontinuous.

Open in new tabDownload slide

3 UPDATE OF THE GRANDMA COLLABORATION

GRANDMA, a worldwide telescope network, was presented in Antier et al. (2020) in the context of O3a. Recently, new groups have joined the collaboration, as will be described below; also presented are updates on the observation strategies and results of certain subgroups. To date, GRANDMA is composed of 25 telescopes with both photometric and spectroscopic facilities, with a large amount of time allocated for observing transient alerts as a telescope network (see Fig. 6, and Tables 2 and 3). The GRANDMA collaboration includes 20 observatories, 29 institutions, and groups from 12 countries.

Figure 6.

Locations of the 20 observatories involved in the GRANDMA network. The colour encodes the observation strategy followed by telescopes at a given observatory: red for tiling, blue for targeting galaxies, and green for following-up candidates.

Open in new tabDownload slide

For each of the observatories we provide a 5σ limiting magnitude as an estimate of the faintness of sources that can be reliably discovered by our system. The 5σ limit is the usual choice of most astronomical surveys (SDSS, York et al. 2000; LSST, LSST Science Collaboration 2009; PanSTARRS, Chambers et al. 2016, etc.) as well as other GW follow-up efforts (e.g. GROWTH, Coughlin et al. 2019c, and GOTO, O’Brien 2018), and it implies that only one out of 1.7 million sources above this threshold would be a false detection. However, since in our case we are just looking for point-like sources in images with well-known point spread functions (PSFs), our system is capable of filtering out many of these sources (which do not have stellar PSFs) and significantly reducing the number of false positives. Our current neural network system (see Section 4) has shown to be able to reduce the number of false positives by a factor of several thousand. This makes the 5σ limit provided here a conservative one.

3.1 Canada–France–Hawaii Telescope

Twelve hours of observing time have been awarded (PI: A. Coleiro) on the Canada–France–Hawaii 3.6-m world-class telescope (CFHT) for period 2020A in order to trigger target-of-opportunity photometric observations. The telescope is located on top of the Mauna Kea summit in Hawaii and is equipped with two imaging instruments: a 1° × 1° FoV optical camera, MegaCam, complemented by a 20 arcmin × 20 arcmin FoV near-infrared detector, WIRCam. By providing high-sensitivity and wide FoV optical (i^′, r^′, and |$z$|^′) and near-infrared (J, H, and K_s) data, CFHT is an essential element of the GRANDMA network, allowing for a fast identification of the EM counterpart and/or its physical characterization. To strengthen their scientific impact, we asked all data to be made freely available to the whole community when delivered to the GRANDMA team. However, due to the premature end of the scientific operations of both CFHT and the GW observatories at the end of 2020 March, no observations were triggered on CFHT during period 2020A.

3.2 FRAM network

FRAM (F/(Ph)otometric Robotic Atmospheric Monitor) is a series of small robotic telescopes primarily designed for a continuous atmospheric monitoring of astronomical sites in order to measure atmospheric transparency with high spatial or temporal resolution. The original FRAM (Prouza et al. 2010) has been operated at the Pierre Auger Observatory in Argentina for more than a decade, while three more FRAMs (Janeček et al. 2019) are prepared to be used for a real-time atmospheric monitoring at Cherenkov Telescope Array (CTA) locations. Two of the FRAM sites are equipped, in addition to primary wide-field telephoto lenses, with larger diameter traditional telescopes, with larger diameter traditional telescopes intended for a follow-up and monitoring observations of a wide range of astrophysical transient objects (GRBs, variable stars, comets, etc). These two telescopes are used for GRANDMA network observations. The first is a 30 cm f/6.8 telescope located at Pierre Auger observatory, Malargue, Argentina, and equipped with B, V, R, and I filters, with an FoV of 60 arcmin × 60 arcmin and pixel scale of 0.92 arcsec pixel⁻¹. The second is a 25 cm f/6.3 telescope located at Observatorio del Roque de los Muchachos, La Palma, Canary Islands, Spain, and is equipped with B, V, R, and |$z$|^′ filters, with a 26 arcmin × 26 arcmin FoV and 1.52 arcsec pixel⁻¹ pixel scale.

3.3 HETH group at IAA

The HETH (High-Energy Transients and their Hosts) group at the Instituto de Astrofísica de Andalucía (IAA) in Granada, Spain, continues to obtain competitive time at multiple telescopes. The programmes at the Observatorio de Sierra Nevada (OSN, PI: Blažek) and the Centro Astronómico Hispano en Andalucía (CAHA, PI: Kann) continue as detailed in Antier et al. (2020).

Competitive time was obtained at the Observatorio de Javalambre (OAJ) for further observations with the T80 telescope to obtain tiling observations (proposal number 1900160, PI: Kann, see Antier et al. 2020 for further details). Once again, the proposal was to observe 1.4° × 1.4° fields with an exposure time of 180 s each, reaching a typical limiting magnitude of r^′ > 21.0 mag. In this proposal, based on the lessons learned from O3a, the requested time was increased to 12.46 h (up to 100 tiles), including late-time re-observations for image-subtraction purposes. The second part of the proposal, multicolour KN follow-up, was dropped as GRANDMA has access to facilities that are better suited for such deep, narrow-field, targeted observations.

The Gran Telescopio Canarias⁸ (GTC) is the world’s largest single-aperture optical telescope, a 10.4-m Ritchey–Chrétien telescope with a segmented mirror located at 2267 m altitude at Roque de los Muchachos observatory, La Palma, Canary Islands, Spain, at 28°45^′24^″N, 17°53^′31^″W. It is equipped with a large suite of instruments. The following three are part of the proposal.

The Optical System for Imaging and low-Intermediate-Resolution Integrated Spectroscopy (OSIRIS) is a multipurpose imager and spectrograph mounted on the Nasmyth-B focus. It is equipped with SDSS u^′g^′r^′i^′|$z$|^′ filters and multiple grisms and volume-phased holographic (VPH) gratings allowing for different spectral resolutions. We generally employ the R1000B (wavelength range 3630–7500 Å, resolution 1018 at 5455 Å) grism and the R2500I (wavelength range 7330–10000 Å, resolution 2503 at 8650 Å) VPH grating, as well as g^′r^′i^′|$z$|^′ imaging. To obtain spectra with sufficient S/N, exposures of different integration times are employed: 4 × 1200 s if the target has r^′ > 22 mag, 3 × 900 s for r^′ ∼ 20–22 mag, and 2 × 900 s if r^′ < 20 mag.

The Especrografo Multiobjeto Infra-Rojo (Infrared Multi-Object Spectrograph, EMIR), installed at the Nasmyth-A focus, is a multichannel IR imager and medium-resolution spectrograph. For spectra, it can achieve an S/N of 7 for a 21 mag source with 4800 s of exposure in the YJ grism. In imaging, limiting AB magnitudes at 5σ of Y > 24.2, J > 23.6, H > 23.1, and K_S > 22.2 mag can be achieved with exposure times of 180, 240, 420, and 600 s, respectively.

Finally, for the purpose of observing the host galaxy of a GW event in 3D, an application was made for time to use the Multi-Espectrógrafo en GTC de Alta Resolución para Astronomía (High-Resolution Multi-Spectrograph for Astronomy at GTC, MEGARA) in integral field unit (IFU) mode, covering a field of 12.5 arcsec × 11.3 arcsec with a spaxel size of 0.62 arcsec. Two grisms would be used to cover the spectral lines Hα, [N ii], and [S ii] as well as Hβ and [O iii].

An award was made (proposal 118-MULTIPLE-3/20A, PI: Kann) 8 h of time with OSIRIS, 5 h with EMIR, and 2 h with MEGARA, ranked ‘A’. The proposal was geared toward obtaining detailed observations of a single confirmed EM counterpart of a GW event, and not meant for classifying candidates. The observing time was valid for the last two months of O3, in 2020 March and April; however the worldwide COVID-19 pandemic caused GTC to be closed down already in mid-March, shortly thereafter followed by the end of O3b; no observations were ever triggered.

3.4 Thai National Telescope

The Thai National Telescope (TNT) located at Doi Inthanon National Park is the main facility for the National Astronomical Research Institute of Thailand (NARIT). It is s 2.4 m diameter, Ritchy–Chrétien telescope with an f/10 ratio, providing a plate scale of 8.6 arcsec mm⁻¹ at the two Nasmyth foci. The telescope system is an alt-az mount that can slew at a speed of 4° s⁻¹; therefore, it can track or acquire optical transient sources rapidly with a pointing accuracy at less than 3 arcsec, and can track targets without autoguiding for 10 min or more with a pointing error of less than 0.5 arcsec. There are many scientific instruments that can be used upon request for specific observations such as ULTRASPEC (Dhillon et al. 2014) with SDSS filters ranging from 330 to 1000 nm, spectrographs varying from low to high resolution, and different types of scientific cameras.

The telescope was chosen to be built at the summit of the highest mountain in Thailand (Lat. 18.57372E°, Long. 98.48219E°, Alt. 2457 m). Its observational area and seeing conditions ensure dark skies for B = 21.5 and V = 21.9 mag arcsec⁻², and the median seeing is stable throughout the night at ≈0.9 arcsec.

3.5 VIRT

The Virgin Islands Robotic Telescope (VIRT) is the primary research telescope at the University of the Virgin Islands and is located at the Etelman Observatory on the island of St. Thomas in the US Virgin Islands. The facility is at an elevation of 381 meters, latitude of 18.3N and longitude of 64.9W. The VIRT is a 0.5-m Schmidt–Cassegrain telescope with a Finger Lakes Instrumentation (FLI) Proline 4240 back illuminated mid-band camera with an FLI Johnson/Bessel UBVRI filter set. Typical seeing at the site is 1.5−2.0 arcsec.

3.6 The kilonova-catcher citizen science program

Since the beginning of O3, GRANDMA has developed a citizen science program called kilonova-catcher (Antier et al. 2020). It allows non-professional astronomers to add their observational capabilities to the GRANDMA network in order to perform optical follow-up of GW candidates on a best-effort basis. During the O3b run, 33 kilonova-catcher users around the globe registered to the GRANDMA GW alert stream. In Fig. 7, we show the locations of the telescopes they could operate during the O3b run.

Figure 7.

Locations of the 34 telescopes involved in the GRANDMA citizen science program: kilonova-catcher. Some citizen astronomers can operate several telescopes represented in this Earth map and also external remote facilities like the iTelescope Network (https://www.itelescope.net/), not represented here.

Open in new tabDownload slide

During the O3b campaign, the kilonova-cather program was active during the follow-up of two BNS merger candidates S191213g and S200213t. Because of the narrow FoV of the amateur telescopes, we coordinated their observations with a strategy based on targeted searches for optical transients located in promising host galaxies. From these two citizen observational campaigns, we obtained 57 follow-up images in total. A detailed description of the amateur observation strategy and results can be found in Section 5. GRANDMA is unique with this global initiative of citizen observers. O3 illustrated the huge potential for GW astronomy to vitalize sky observations and EM counterpart candidates, while also highlighting the need of training and coordinated joint analysis.

4 UPDATE FOR THE ALERT INFRASTRUCTURE, OBSERVATION STRATEGY AND WEB INTERFACE, AND DATA ANALYSIS USED IN GRANDMA

We use a similar infrastructure – ICARE (Interface and Communication for Addicts of the Rapid follow-up in multi-messenger Era) – for conducting coordinated observations of GW alerts to that discussed in Antier et al. (2020). ICARE relies on the automatic reception of a GW alert, creation of an automatic observation plan that is then sent to the telescopes of the network, a central database, and cloud applications to monitor the full network. Similar tools and infrastructure have been developed by other teams for a similar purpose, that is the GROWTH Marshal (Kasliwal et al. 2019). In this section, we will outline some improvements since the end of the O3a run, and in particular, during the break between O3a and O3b. We are going to focus on work done for the daily use version of ICARE that ran during O3b.⁹ These modifications introduced some minor negative impacts on the productivity of the collaboration at the level of the end-user communication (i.e. mostly with regard to the preparation of GCNs), which had no effect on observations, nor results reported in this manuscript.

The python-based alert distribution system used by GRANDMA during O3a remained mainly the same. Upon receiving a GW candidate alerts through a GCN, the coverage of the sky localization region was generated for the GRANDMA collaboration telescopes (Table 2). For BBH-type events, only large FoV instruments were included, while all instruments were involved for NSBH or BNS type events if the posterior mean distance was below 300 Mpc. The resulting observation plans were then sent via VOEvent as soon as the computation of a single plan was done using Comet.¹⁰ We improved this system for O3b by introducing a forwarding of the retraction notice to avoid unnecessary observations.

Our observation plan algorithm benefitted from the continuous improvements of the gwemopt¹¹ telescope scheduling software (Coughlin et al. 2018a). Its latest version includes slicing sky localizations in right ascension to separately schedule different lobes in multilobed maps (Coughlin et al. 2019a), and automatically determining fields capable of being observed multiple times by several telescopes (Almualla et al. 2020). In doing so, the coverage of all our instruments, and especially the wide FoV telescopes, was improved. Our galaxy targeting strategy, that is when small FoV telescopes were utilized to survey known galaxies) was also revised using the dedicated MANGROVE galaxy catalogue presented in Ducoin et al. (2020a). We adopted the selection of preferential galaxies based on dependence on galaxy stellar mass. The most problematic issue in Galaxy Targeting Mode from O3a, however, that has since been addressed was the re-imaging of tiles when observations were executed. In short, this issue arose from an avoidance of considering whether or not a given galaxy requested for observation resided in the FoV of a preceding observation. As such, it resulted in an inefficient management of telescope observing time, and reduced the area covered by our collaboration. This improvement additionally benefited our strategy, as it led to extra time available for observing the most promising events.

The start of the O3 run demonstrated the need for a standard data reduction pipeline in order to homogenize the photometry within the network. First, we still use individual online image processing developed by the various groups such as TAROT and FRAM (Noysena et al. 2019). Secondly, we continue the development of a common detection pipeline Gmadet¹² adaptable for each telescope: it makes use of the popular AstrOmatic¹³ software (scamp, swarp, psfex, and SExtractor) to perform the astrometric calibration, align and stack images, estimate the PSF and extract sources. The detected sources are crossmatched to different catalogues (Pan-STARRS, Gaia, USNO-B1) using Xmatch, the CDS crossmatch service, to discard already known objects. A crossmatch with known Solar system objects is also performed using SkyBoT (Berthier et al. 2006). Particular attention is paid to the photometric calibration so that magnitudes are expressed in the AB system. Image substraction with a reference image is performed using HOTPANTS (Becker 2015), where Pan-STARRS stacked images can be used as reference. A Convolutional Neural Network algorithm is trained to identify stellar sources and discard most of the cosmics, bad pixels or artefacts coming from the subtraction process. In practice, during O3b Gmadet was used to detect transient candidates in parallel to individual standard online image processing made by the groups. The results presented in this work correspond for Gmadet only for the OAJ observations of S200213t. The other results use the individual image processing described in Antier et al. (2020) to allow for a direct comparison of our results against those for O3a. In case of GW alert observations engaged by GRANDMA, each individual image was processed with a search for point sources detected at the 5-or greater level. All the false positives which were deemed to be artieacts or cosmic rays, were eliminated by human inspection.

5 GRANDMA ELECTROMAGNETIC FOLLOW-UP CAMPAIGN OF O3A AND O3B

5.1 Observational summary of GRANDMA for O3a and O3b

Over the 11 months of observations of the O3 campaign, GRANDMA followed-up 49/56 candidate events: 27/33 for O3a and 22/23 for O3b. The observation efficiency for GRANDMA is therefore ∼82 per cent for O3a and 96 per cent for O3b. This high cadence is possible due to the organization of the GRANDMA collaboration and the automated infrastructure discussed above. It is significantly above the reported observational rate of other ground-follow-up teams, aside from MASTER-Net (Lipunov et al. 2010) that reports GCNs for every alert based on their automated observations. Based on the GCN traffic, GRANDMA can be compared to GROWTH (about 30 per cent of alerts, Coughlin et al. 2019c), GOTO (about 15 per cent of alerts, O’Brien 2018), ENGRAVE (about 5 per cent of alerts, Ackley et al. 2020), GRAWITA (about 25 per cent of alerts, Salmaso et al. 2019), J-GEM (about 7 per cent of alerts, Kaneko & J-GEM Collaboration 2020), NOWT (about 4 per cent of alerts, Zhu et al. 2019), SAGUARO (about 11 per cent of alerts, Lundquist et al. 2019), and SVOM (about 30 per cent of alerts, Xin, SVOM Multi Messenger Astronomy Team & GWAC Team 2020).

As shown in Fig. 8, GRANDMA performed its first observation at most 1.5 h after the GW trigger time for more than half of the alerts, and at most 30 min for 15 per cent of the alerts during O3. The minimal delay between the GW trigger time and the first GRANDMA observation is 15 min, which includes the ∼5 min delay for sending the alert by LIGO/Virgo, ∼5 min for computing the observation plan, and ∼5 min for scheduling by the telescope. This has been achieved for both the TAROT and FRAM networks. Moreover, we see some improvement in terms of the rapid response by GRANDMA between O3a and O3b: this is due to improvements not only from the LIGO/Virgo infrastructure system (Abbott et al. 2019a) but also the GRANDMA infrastructure improvements discussed above. This demonstrates once more the need for a heterogenous network of telescopes sharing joint tools to face the challenge of follow-up of GW triggers and their requirement of rapid EM observational responses. Indeed, Metzger (2019) and references therein, predict different light-curve evolutions for the first hours after the merger of two compact objects (Arcavi 2018) depending on the properties of the initial binary system. Early time observations after the merger (less than two hours and in different passbands if possible) might have provided important constraints on properties of the fast ejecta components, the composition of the ejected material, and the heating rate for the unbound material. Other contribution channels or kilonova-precursors powered by free neutron decay can also be ruled out by early observations (Metzger et al. 2015).

Figure 8.

Cumulative distribution of the delay between the GW candidate/event trigger time and the first observation done by GRANDMA. The minimal delay is about 15 min obtained with both the TAROT and FRAM networks. About 50 per cent of the sky localization areas of GW candidate/event alerts have been observed with a minimal delay of 1.5 h.

Open in new tabDownload slide

GRANDMA covered 213 ± 173 deg² on average for the GW alerts during O3. As shown in Fig. 9, more than 100 deg² of the sky has been observed by the GRANDMA network for more than half of the alerts. We do not see major differences in terms of sky coverage between O3a and O3b; however, improvements in the network-level scheduling as discussed in Coughlin et al. (2019a) in 2019 June led to a significant increase of the maximum coverage reachable by GRANDMA (typically by a factor of 2). The total maximal coverage for a single alert was for S191204r, a BBH candidate event, at 550 deg², representing 90 per cent of the sky localization area. The lowest coverage was only 10 per cent of the candidate S200225q’s most updated sky localization area; although the event was well localized, with a 90 per cent credible region of 22 deg² as seen in Appendix A and LIGO Scientific Collaboration & Virgo Collaboration (2020k), the vast majority of this region was behind the sun at the time of the event. In Fig. 10, we can see the total sky coverage of GRANDMA over the full O3 campaign. There is a clear distribution of Northern and Southern hemispheres among the different telescopes due to the global coordination (see Section 4). In total, we covered 9218 deg² with six wide FoV telescopes, with multi-epoch observations for most of the fields. Our near real-time analysis did not find any interesting kilonova candidates (see Section 5 and Antier et al. 2020).

Figure 9.

Cumulative distribution of the total sky coverage of the GRANDMA network (in deg²) engaged for each GW candidate/event alert. The distribution depends on the sky localization area size of the alerts and observability constraints, as discussed in Section 2.

Open in new tabDownload slide

Figure 10.

Total sky coverage of the wide FoV telescopes of GRANDMA during O3. Note the distribution of observations due to the GRANDMA coordinated program. Also note that the heterogeneity of the maps is imprinted from observing constraints and the distribution of the alerts over the sky.

Open in new tabDownload slide

According to Fig. 2, the median distance of O3 BNS merger candidates is ∼ 200 Mpc, corresponding to a distance modulus, |$\rm {DM} = 36.5$|⁠. The median distance of BBHs is ∼1.3 Gpc (⁠|$\rm {DM} = 40.6$|⁠). Considering an apparent limiting magnitude of 17, this corresponds to approximate absolute magnitude non-detection limits of −19.5 for BNSs and −23.6 for BBHs. Fig. 11 displays the absolute magnitude limit based on the distance of each GW event versus the observed area, and in Fig. 12, versus the time of the first optical observations. It shows that only 12 GW events were observed by GRANDMA covering more than 80 per cent of the error box. The names of these events are indicated in the Fig. 11. Essentially only BBH events have full coverage of the credible region; this is due to a combination of the statistical rate of observability (see Section 2.2), GRANDMA coverage capacity (∼200 deg²), and the poor localization of NSBH or BNS events. S190814bv, with a current classification of NSBH, remains the exception. In conclusion, the disparity of alerts during O3 in terms of localization and their observational accessibility, the evolution of the localization over time, as well as the possible nature of candidates, shows a clear need for monitoring of routine observations of the alerts and calls for rapid reactions at all times. In this way, by observing all alerts during O3, GRANDMA has gained training and experience and is ready to observe a large number of alerts during the subsequent GW observation campaigns, starting with O4.

Figure 11.

All LIGO/Virgo GW candidates/events observed by GRANDMA telescopes are plotted as limiting absolute magnitude versus the percentage of the 90 per cent credible region that was observed. We added the TAROT/TRE limit of GW 170814 published in Noysena et al. (2019). candidates/events having a percentage of the 90 per cent credible region greater than 150 per cent are plotted at 150 per cent to keep clarity of the plot.

Open in new tabDownload slide

Figure 12.

All LIGO/Virgo GW candidates/events observed by GRANDMA telescopes are plotted as limiting absolute magnitude versus the delay between the GW trigger and the first GRANDMA observations. Filled symbols are used for candidates/events for which more than 80 per cent of the 90 per cent credible region have been observed. We added the TAROT/TRE limit of GW170814 published in Noysena et al. (2019). The dashed horizontal bars indicate the duration of the first visit for the best candidates/events (selected from the Fig. 11). The black lines are the typical limits of LGRB light curves, blue lines for SGRBs, and red lines for the GW170817 kilonova.

Open in new tabDownload slide

5.2 Report summary of GRANDMA observations during O3b

Specifically for O3b, we report our observations for the BBH candidates in Table 4, and for the systems potentially containing at least one NS in Table 5. The unmodelled GW burst candidate S200114f is listed in Table 6. All the sky localization coverage are shown in Appendix A.