Search for the optical counterpart of the GW170814 gravitational wave event with the VLT Survey Telescope Free

ABSTRACT

1 INTRODUCTION

2017 August 14 is a milestone in the gravitational waves (GW) astronomy. The Virgo interferometer (Acernese et al. 2015) detected its first signal and the triangulation with the LIGO interferometers (Aasi et al. 2015) led to a strong improvement of the sky localization of the event GW170814, shrinking the area of the original error box obtained by using only the LIGO detectors, from 1160 to only 60 deg² after using all three detectors (Abbott et al. 2017a).¹ This was a fundamental step forward in the newly born era of multimessenger astrophysics which was fully exploited three days later with the detection of the binary neutron stars merger GW 170817 event and its electromagnetic (EM) counterpart (Abbott et al. 2017b; Pian et al. 2017).

The GW170814 event was detected at 10:30:43 UTC; it is attributed to the merger of two black holes (BH) with a false-alarm rate of 1/27000 yr. The estimated BH masses are |$30.5^{+5.7}_{-3.0}$| and |$25.3^{+2.8}_{-4.2} \, \mathrm{M}_{\odot }$|⁠, while the inferred luminosity distance is |$540^{+130}_{-210}$| Mpc² corresponding to a redshift |$z = 0.11^{+0.03}_{-0.04}$| (Abbott et al. 2017a). In the O1 and O2 LIGO and Virgo observing run a total of 10 BBHs (binary black holes) have been observed (The LIGO Scientific Collaboration et al. 2018b), allowing us to put constraints on a certain number of parameters associated to the population of BBHs. Among them it has been possible to infer a BBH merger rate density of ∼ 60 Gpc⁻³yr⁻¹. Of high valuable scientific interest is their mass distribution, it comes out that there is a significant reduction in the merger rate for BBHs having primary masses larger that ∼ 45 M_⊙, while the limited sensitivity does not allow to place reliable constraints on the minimum mass of BHs. The limited size sample and the local nature of the events so far observed have prevented to infer the BBH redshift distribution. A binary BH merging is not expected to produce an EM counterpart. However, there are alternative models (de Mink & King 2017; Perna, Lazzati & Giacomazzo 2016; Perna et al. 2018; Zhang 2016; Murase et al. 2016; Yamazaki, Asano & Ohira 2016; Stone, Metzger & Haiman 2017) which predict possible EM radiation in the optical/near-infrared spectral region. If a fraction of the progenitors mass lost during the prior evolution of the binary stellar system, is retained as circumbinary disc, an infrared emission to medium-energy X-rays might be observable within hours after the GW event (de Mink & King 2017). In particular Perna et al. (2019) find that after one day from the GW event an optical signal in the R band could be detected at a magnitude of ∼22.5 for a source at a distance of GW150914.

The detection of an EM counterpart of a GW emitted by a stellar BBH would reduce GW parameters degeneracy (see Pankow et al. 2017), it would allow the measurement of the source redshift (see Schutz 1986), and provide hints on the formation channel of the BBH binary by exploring the environment in which the merger occurred. Indeed dynamical formation (Rodriguez et al. 2016) is likely to take place in dense star clusters, while the isolated binary evolution is expected in typical field galaxies (Kalogera et al. 2007). The results of the search of optical counterpart to GW170814 event was recently published (Doctor et al. 2019). By using the Dark Energy Camera, 225 deg² were imaged corresponding to 90 per cent of the LALinference probable sky area. The observations have been carried in the i band reaching a depth of ∼ 23 mag. With a selection criteria optimized to search for fast declining (few days) transients two candidates were found. However, as stated by the authors, these candidates are most likely not associated with GW170814. This work is carried out within the GRAvitational Wave Inaf TeAm collaboration (Brocato et al. 2018) and is organized as follows: Section 2 describes the VST observational strategy, and Section 3, the processing of data and search for counterpart. In Section 4 is described the survey detection efficiency calculation. Finally, we draw our conclusions in Section 5.

2 VST OBSERVATIONAL STRATEGY

We carried out the search for the optical counterpart of the GW170814 event using the ESO VLT Survey Telescope (VST, Capaccioli, Mancini & Sedmak 2003) located at Cerro Paranal in Chile. VST is a 2.6-m telescope specifically designed for wide-field imaging. It is equipped with OmegaCAM, a CCD mosaic camera made of 32 (2048 × 4096) pixel devices for a total 16k × 16k pixels, providing a field of view of ∼ 1 deg² (Kuijken 2011). The standard filter set includes ugriz Sloan filters (Doi et al. 2010).

We started our observations 17.5 h (Greco et al. 2017) after the LIGO/Virgo alert triggered by four independent low-latency pipelines at 2017 August 14 10:30:43 UTC (The LIGO Scientific Collaboration & the Virgo Collaboration 2017a). The sky area tiling was obtained using a dedicated tool named GWsky.³GWsky is a python⁴ tool developed to generate accurate sequences of pointings (tiles) covering the sky localization of a GW. It is equipped with a Graphical User Interface (GUI) optimized for fast and interactive telescope-pointing operations and supplies information and descriptive statistics on telescope visibility, GW localization probability, availability of reference images, and coverage of catalogued galaxies for each Field of View (FoV) footprint. GWsky supports two observational strategies according to the preliminary information reported in the LIGO/Virgo GCN notices/circulars. They are (1) the tiling and (2) targeted searches. The tiling (or coverage) strategy mainly provides a list of pointings that (i) optimizes the telescope’s visibility during the pre-assigned epochs, (ii) maximizes the contained localization probability, and (iii) avoids areas with too bright objects and/or severe crowding. The second strategy generates a list of pointings covering galaxies from the Glade catalogue (Dálya et al. 2018) ranked from the highest to the lowest luminosity taking into account the initial airmass.

The posterior luminosity distance and standard deviation of GW170814⁵ were automatically taken from the fits header of the initial BAYESian TriAngulation Rapid localization (BAYESTAR) sky map (Singer & Price 2016) (DISTMEAN and DISTSTD) for rapidly setting an observational strategy. Given the nature of the event and its posterior mean distance, the number of galaxies contained in the sky error volume was found to be too high for a targeted search so that the coverage strategy was chosen. Starting from the maximum probability pixel of the BAYESTAR sky map, (RA[deg] = 41.06842, Dec.[deg]  = −45.48795) a sequence of nine consecutive tiles (each tile is made of 3 × 3 pointings⁶) were defined limiting the airmass to 2 and maximizing the enclosed probability in each tile. The survey was performed in the r band covering an area of |$81\, \mbox{deg}^2$|⁠, enclosing the initial sky localization probability of ∼ 76 per cent. The choice of the r band is a trade-off between telescope and camera efficiency and expected spectral distribution of the EM signal.

On 2017 August 16, an updated localization from LIGO and Virgo data around the time of the compact binary coalescence was provided (The LIGO Scientific Collaboration & the Virgo Collaboration 2017b). Source parameters estimation were performed using the LALInference algorithm (Veitch et al. 2015) and a new sky map was available for retrieval from the GraceDB⁷ event page (hereafter, preliminary LALInference). The maximum probability pixel of the preliminary LALInference is localized at the coordinates RA[deg] = 46.53409, Dec.[deg]  = −44.59799. The 90 per cent credible regions of the initial (BAYESTAR) and his updated localization have significant overlap, but the 50 per cent credible region is shifted east by its entire width. To compensate such shift, two additional tiles were added (shown in light green in Fig. 1), for a total observed area of 99 deg², starting from the third epoch enclosing a total probability of ∼ 50 per cent of the Preliminary LALInference. On 2017 September 27, a refined sky localization was issued taking into account calibration and waveform modelling uncertainties (The LIGO Scientific Collaboration & the Virgo Collaboration 2017c).

Figure 1.

Footprints of the VST r-band observations over the contours of both the initially distributed BAYESTAR and refined LALinference sky localization maps of GW 170814. Each small square represents a VST observation. The blue and red external contours are the initial localization and refined map at 90 per cent confidence level, respectively. The inner contour lines represent the enclosed probabilities of 50 per cent and 10 per cent confidence level, respectively. The tiles in light green are observed since the third epoch.

Open in new tabDownload slide

The sky coordinates of the VST pointings and the enclosed probability for each tile are reported in Table 1. The enclosed probability is calculated with respect to the BAYESTAR (The LIGO Scientific Collaboration & the Virgo Collaboration 2017a), LALInference preliminary (The LIGO Scientific Collaboration & the Virgo Collaboration 2017b) and the final LALInference (hereafter, refined LALInference) sky maps. The refined LALInference was recently published in the Gravitational-Wave Transient Catalogue – GWTC-1 (The LIGO Scientific Collaboration et al. 2018a). The initial BAYESTAR sky localization and the refined LALInference sky localization, as well as the VST tiles, are shown in Fig. 1. We covered a total area of 99 deg², taking into account uncovered gaps among ccds, haloes, and spikes due to bright stars, we encloses ∼ 77 per cent and ∼ 59 per cent of the initial and refined localization probability region, respectively.

The VST observations are organized in observing blocks (OB) consisting of a mosaic of nine pointings – |$1 \times 1\, \mbox{deg}^2$| – hence covering a total area of |$3 \times 3\, \mbox{deg}^2$|⁠. Each pointing is made of two exposures of 40 s each dithered by |$0.7 - 1.4\, \mbox{arcmin}$|⁠. The dithering allows us to fill the gaps in the OmegaCAM CCD mosaic. The VST observations of the patrolled area were repeated six times distributed over ∼ two months in order to have enough data points to sample the light curve of the optical transients. A 50 per cent point sources completeness at r ≃ 22.0 AB mag was reached in most epochs.

A summary of the VST observations performed for the GW170814 event is reported in Table 2. In some epoch, it was not possible to cover the whole area in one single night. In Table 2, we also report the average seeing (FWHM), the minimum and maximum values of the airmass, the magnitude corresponding to 50 per cent completeness and the minimum and maximum value for the different pointings.⁸ An overview of the temporal distribution of the observations along with the observing conditions is shown in Fig. 2.

Figure 2.

Overview of the survey monitoring progress. On ordinates are indicate months and days. On abscissa the VST pointings of 1 × 1 deg². The colour indicates the seeing achieved for each specific pointing. Please refer to Table 2 to see the way nights are grouped for each epoch.

Open in new tabDownload slide

3 DATA PROCESSING

3.1 Pre-reduction

Immediately after acquisition, the images were mirrored to the ESO Garching data archive, and then transferred by an automatic procedure from the ESO Headquarters to the VST Data Center in Naples (Italy). The first part of the image processing was carried out using VST-tube (Grado et al. 2012), which is a data reduction pipeline specifically developed for the VST-OmegaCAM mosaics. It performs pre-reduction, astrometric and photometric calibration, and mosaic production.

To remove the instrumental signatures, after overscan, bias and flat-field corrections, we performed gain equalization of the 32 CCDs and illumination correction. The last step was achieved by comparing the magnitude of stars with the Sloan Digital Sky Survey (SDSS) DR12 catalogue (Alam et al. 2015) in equatorial fields observed during the same nights. The magnitude difference is function of the position over the whole field of view is fitted using the ‘generalized additive method’ (Hastie & Tibshirani 1986). This approach has been shown to be more robust compared to a plain surface polynomial fit in particular for what concerns the behaviour at the edges of the image. Then, the position-dependent zero-point correction was applied to science images. The relative photometric calibration of the images was obtained by minimizing the quadratic sum of differences in magnitude between sources in overlapping observations. The tool used for both the astrometric and relative photometric calibration tasks is SCAMP (Bertin 2006). The used reference astrometric catalogue is Two Micron All-Sky Survey (2MASS) (Skrutskie et al. 2006). The absolute photometric calibration was obtained with the Photcal tool (Radovich et al. 2004) using as reference the equatorial photometric standard star fields from SDSS DR12 (Alam et al. 2015). Finally the two dithered images for each pointing were co-added. In order to simplify the subsequent image subtraction analysis, for each pointing the mosaics at the different epochs were registered and aligned to the same pixel grid. This assures that each pixel in the mosaic frame corresponds to the same sky coordinates for all epochs. For further details on the data reduction, see Capaccioli et al. (2015).

The detection of transients is affected by the presence of artefacts on the images. These artefacts can be due to the telescope, like spikes and haloes around bright stars, and/or to the camera, like electronic cross-talk. It is desirable to detect and correct them or at least to identify the pixels affected by the artefact. With this purpose we produce a flag image with specific code for different kind of contaminants that can be used during image analysis. The procedure to mask the spikes around bright stars is described in Appendix  A whereas in the following we describe the approach adopted for cross talk. In fact, two CCDs of the OMEGACAM camera (ccd65 and ccd66 in ESO names coding; cf. Mieske 2018) are affected by electronic cross-talk, i.e. the appearing of spurious sources or holes due to the presence of bright sources in the corresponding pixels of the other CCDs. This produces spurious sources that can be confused with transients. In principle with the knowledge of the cross-talk coefficients one can attempt to remove the ghost objects. In practice, the coefficients change with time then preventing a safe removal of the spurious sources. For that reason we decided to adopt a conservative approach masking the pixels affected by cross-talk ghosts with a specific flag value.

With the current available hardware, the time needed to produce a 1 deg² fully calibrated co-added image is ∼20 min, including the time for data transfer from Cerro Paranal to Naples and the production of the SExtractor catalogues.

3.2 Transient search

In order to search for variable and transient sources, we applied our analysis procedure first described in Brocato et al. (2018). The pipeline is based on two independent yet synergistic procedures. One, called ph-pipe, is based on the comparison of the photometric measurements of all the sources in the VST field obtained at different epochs. The second (diff-pipe) is based on the analysis of the difference of images following the approach of the supernova (SN) search program conduced with the VST for several years now (Cappellaro et al. 2015).

The ph-pipe is typically more rapid, providing an excellent quick-look facility able to single out candidates in single-epoch campaigns, while the diff-pipe is more effective for pinpointing sources projected over extended objects or in case of strong crowding. The two approaches are complementary and since they are independent the comparison of the results allow us to spot possible missed transients.

As discussed in Brocato et al. (2018), the list of candidates contains a large number of spurious objects that can be related to small misalignments of the images, improper flux scaling, incorrect PSF (point spread function) convolution or not well-masked CCD defects and cosmic rays. In most cases, spurious transients could be easily identified by visual inspection. However due to the large number of candidates it would be very time consuming. For this reason both pipelines apply several criteria to attribute a ‘score’ to select ‘bona fide’ candidates, based on their morphological parameters, magnitude variations, positions with respect to nearby galaxies, etc. (cf. Cappellaro et al. 2015). With this approach we can readily reject most spurious sources. The remaining candidates go through a further selection after being visually inspected.

In consideration of the largely unknown properties of the possible EM gravitational wave counterpart, we decided not to use model-based priors in the candidate selection. The main goal of our analysis is therefore to simply identify sources showing a ‘significant’ brightness variation (5σ between at least two epochs that in our case corresponds typically to ∼ 0.5 mag), either rising or declining in flux during the monitoring.

The implementation of the diff-pipe requires adequate reference images. Ideally, these images should be obtained before the event to allow for a prompt transient search. If they are not available, as it is in our case, they can be obtained at later epochs. Alternatively, we can use as references archival images obtained with other telescopes. Archival images of the survey area of GW170814 are available from the Dark Energy CAMera (DECAM) survey archive.⁹ This option is not immune to risk due to a possible mismatch of the instrument/telescope band-passes, objects with a peculiar spectrum give rise to spurious transients after image subtraction (notice however that such objects will give a constant residual for all the different VST epochs).

Conservatively, in the case of GW170814, we performed the transient search using as reference the VST late time images whereas DECAM reference images were used to produce the transient light curves (cf. later this section). This was also useful to verify the possible use of archive DECAM images for future prompt searches.

The difference images produced with hotpants for the 99 pointings were searched with SExtractor resulting in a cumulative list of 3 974 951 variable sources. The list is large because we adopted a low detection threshold (1.5σ above the background noise) to make sure to get the faintest transients. This also means that most of the detected sources are spurious due to noise fluctuation, image defects, not well-removed cosmic rays or imperfect subtraction of bright stars, etc. Following Cappellaro et al. (2015) to filter out most of the spurious candidates we used a scoring algorithm based on several metrics of the detected candidates following the scheme reported in Table 3.

The numbers in Table 3 includes multiple detections at different epochs. In fact, the final list with score >0 counts 9342 distinct candidates of which 1687 with high score (>=60). The first condition in Table 3 adds a penalty of 60 to objects that appear to be non stellar in the difference image. The second condition helps to remove cosmics that were not detected in the early stages of the image processing. The third condition gives a strong penalty to those sources caused by a local mismatch of the PSF that usually gives a ‘dipolar’ source with positive and negative pixels counts. Then the scoring is repeated with more stringent parameters but with less heavy penalties. Ratings are applied in order to penalize spurious detections with the exceptions of ‘near galaxy sources’, a condition that raises the rank of these objects to a positive score.

The scoring algorithm and reference value were chosen after extensive artificial star experiments. We found that with the above scoring scheme and selecting a score threshold >0 we can reject >99 per cent of spurious candidates at the cost of loosing 5 per cent of the real transients. After adopting a score threshold of 60 per cent the number of spurious events decreases by a factor ∼5, whereas the fraction of real transients that are lost may increase up to by 15 per cent.

In this work, we visually inspected all candidates with score >0 which led to a list of 246 validated transients.

We cross-checked the candidates with public data sets, e.g. SIMBAD (Wenger et al. 2000), the NASA/IPAC Extragalactic Database (NED),¹⁰ and Sky Body Tracker (SkyBoT) (Berthier et al. 2006), aiming at identifying known objects. This allowed us to remove from the list two RR Lyrae, and 21 asteroids. We also removed transients that are bona fide variable stars because they coincides with stellar sources present in all our images and also in the DECAM archive images.

We are left with a total number of 53 candidates (36 were identified by both pipelines and 17 only by diff-pipe), see Appendix  B.

Since all the selected candidates are also located in the survey area of the Dark Energy Survey (DES), we took DES images observed before the GW event as reference frames to investigate the sources variability. Details of the 53 candidates transients, namely their photometric catalogue and light curve analysis, are reported in Tables B1 and 4. We have differentiate the transients found by both the search pipelines (the prefix c in the Id) from the ones found only with the image subtraction pipeline (prefix d in the Id). We reported whether a source was found in the NED (within 5 arcsec) or SIMBAD (within 3 arcsec) (columns 4 and 5 of table 4, respectively) and the identification that comes from these databases. In Fig. 3 we show the light curves for all 53 candidates. In case of missing detection it is reported the limiting magnitude of the image at the specific epoch.

Figure 3.

The light curves of the transient candidates found with the VST in |$99\, {\rm deg^2}$| covering the GW170814 error area. Arrows indicate the upper limit magnitude for the undetected sources.

Open in new tabDownload slide

For transients with only one detection, we can exclude that they are normal SNe (of any kind) or AGN (active galactic nucleus, they are not associated with galaxy nuclei). We cannot exclude unknown asteroids, peculiar fast transients, and flare stars (even if they are much rarer). When multiple detections for a given transient were available we attempted to perform a transient classification using the approach described in section 4 of Cappellaro et al. (2015). In short, the transient light curve was compared with that of templates SNe of different types assuming as a free parameters the phase and redshift. In practice, for each template we compute a grid of simulated light curves in a range of redshift by applying the appropriate k-correction and time dilation. The best fit is chosen by chi-square minimization (due to the lack of colour information we cannot constrain the extinction that is neglected). In this approach, the shape of the light curve is the main constraint for the template selection while the apparent/absolute magnitude comparison set the redshift for the specific template. In fact, only in two cases the redshift of the host was known from NED. Yet, because of the very sparse light curve sampling, limited number of epochs and lack of colour information no firm classification could be secured. We could only conclude that two dozen transients have the light curve consistent with that of a specific SN type at a given redshift (and are then considered SN candidates). In addition, ten events are identified as AGN candidates because they are hosted in galaxy nucleus and shows erratic variability. Finally, for 19 transients, with only one detection no specific class could be assigned.

One of the candidate AGN is indeed classified as a QSO in the NED data base. Also, three of the SNe candidate are listed in the Transient Name Server (TNS)¹¹ as follows:

• the transient d3 of Table B1 has a light curve shown in Fig. 4 top panel and was obtained by diff-pipe. By assuming the known redshift z = 0.0815 (Miller et al. 2009), we identified it as a superluminous type II SNe similar to SN 2008es.

  Indeed we found that the candidate d3 is coincident with SN2017eni (Gaia17blw). This was announced by Gaia Alerts (Delgado et al. 2017) on 2017 June 6 with G = 17.7 mag and indicated as an SN candidate in the galaxy 6dFGS gJ030511.0−453304. It was spectroscopically classified as a Type IIn superluminous SN by Strader, Chomiuk & Tremou (2017). Photometry from the All-Sky Automated Survey for SuperNovae (ASAS-SN) survey finds a peak at V = 16.9 mag on 2017 Jun 6.44, implying an absolute magnitude of M_V  = −21.0 mag.

• Transient c32 is coincident with AT 2017gqz (Gaia17cgz) announced as a Gaia transient on 2017 September 08 17:47:02 with G = 18.79 mag. Our light-curve fitting shows that the transient is consistent with a type Ia SN at redshift 0.05 (see Fig. 4, mid panel).

• Transient d9 is coincident with AT 2017fat (Gaia17bqm) announced as a Gaia transient on 2017 June 05 00:41:45 with G = 18.18 mag. The light-curve fitting is consistent with that of a type Ia SN at a redshift 0.08 (see Fig. 4, bottom panel).

Figure 4.

Top panel: best fit for transient d3 (SN2017eni, blue points) with the template SLSN 2008es (green dots). Mid panel: best fit for transient c32 (AT 2017gqz, blue points) and the SN Ia 1994D (green dots). Bottom panel: best fit for transient d9 (AT 2017fat, blue points) and the SN Ia 1994D (green dots). In the upper right boxes are reported the template used, the redshift and the phase. The last values, always equal to zero, are the extinction.

Open in new tabDownload slide

We compared our search with the work carried out by Doctor et al. (2019). The Table 5 summarize the comparison of the two follow-up programs in terms of area covered, filter used, depth, number of epochs, time span of the search, and total uncertainty enclosed by the surveyed area.

In Doctor et al. (2019) two candidates with coordinates (1) RA= 42.35047 Dec.=−40.32632 and (2) RA=47.63365 and Dec. = −36.36045 were found. The second object is outside our searching region, while in the position of the first one we have detected a source at magnitude between 22.5 and 23.5 (as reported by Doctor et al. (2019)), which is too faint to be qualified as a transient in our catalogue.

4 DETECTION EFFICIENCY

To evaluate the survey efficiency, we need to know the expected optical light curve of the transient following a BBH merger. The latter is very uncertain although most authors agree that the optical emission should be expected much fainter and more rapid declining than the kilonova following a Binary Neutron Stars (BNS) merger. To explore the performance of our search, we use as reference the most recent models of Perna et al. (2019) that likely encompass the range of possible EM transient from BBH merger. The different prediction are characterized by: (1) jet energy in the range 10⁴⁶–10⁴⁹ erg; (2) jet opening angle in the range 10–40 deg; and (3) jet Lorentz factor Γ (10 or 100). All models are computed for an ambient number density 0.01 cm⁻³. We immediately verify that a large class of the Perna et al. (2019) models predict optical transients that, for the given distance of GW170814, are too faint for our instrumentation. In particular all models with energy ≤ 10⁴⁷ erg are too faint regardless of the Γ factor and jet opening angle.

We are just marginally sensitive to models with energy 10⁴⁸ erg only if the jet opening angle is small and our line of sight is along the jet. In the more favourable case of high Γ factor, the transient remain above mag 22 for about one day.

We have a measurable efficiency only for 10⁴⁹ erg events characterized by high Γ factors.

To estimate the chance of detection of a possible optical transient after BBH we used the simulation of Perna et al. (2019), as follows.

For each model:

1. we compute the predicted light curve in apparent magnitude at the distance of the transient (540 Mpc) for all the different viewing angles;

2. for each pointing and epoch we verify whether the predicted magnitude is brighter that the magnitude limit.

3. by considering the probability area covered by the specific transient and the solid angle fraction covered by the viewing solid angle, we compute the chance of detection for the specific pointing.

4. the chance of detection for the full survey is obtained by summing the probability for all pointings. In the best case of Γ  = 100 and jet energy 10⁴⁹ erg, with a view angle of 40 deg we obtain a survey efficiency of 5.1 per cent.

A more accurate estimation of the survey efficiency would require a Monte Carlo simulation where synthetic sources extracted from a population following the Perna et al. (2019) models are added to the images. Then the pipeline procedure to detect the transients is applied to recover the simulated sources. Considering the low efficiency, even for the best case, this procedure would not change significantly the results.

5 DISCUSSION AND CONCLUSIONS

We reported on the search for the optical counterpart for the GW event GW170814, by exploiting the capabilities of the VST. We covered a search area up to 99 deg², corresponding to 59 per cent of the credible region, repeated in six epochs distributed over ∼ 2 months. A 50 per cent point source completeness at r ≃ 22 AB mag was reached in most of the epochs. This threshold corresponds to a luminosity limit of L_optical ∼ |$1.4^{+0.7}_{-1}$| in unit of 10⁴² erg s⁻¹ after assuming a light flat spectrum and a GW170814 median distance. The errors on the luminosity mostly derive from the uncertainty on the GW source distance. To estimate the survey efficiency we have considered the recent model for EM counterpart from BBHs mergers described in Perna et al. (2019). In the most favourable case of a very energetic jet emission (10⁴⁹ erg) and high Γ factor the expected magnitude of the transient source, at the distance of GW170814, is estimated to remain brighter than mag 22 for about a day. Therefore, under these specific conditions, the corresponding survey efficiency is considerably low, about |$5 {{\ \rm per\ cent}}$|⁠. For optical transients search, we developed a pipeline based on two independent analysis algorithms, one based on source extraction and magnitudes comparison between different epochs and the other based on transient identification obtained through image subtraction techniques. We present the catalogue of these transients, including, when possible, matches with the SIMBAD within 2 arcsec or NED catalogues and/or possible identification based on light-curve fitting. We identified two dozen candidates SNe, nine AGN candidates, one QSO. Nineteen transients have not been previously catalogued and since they have only one detection cannot be classified. For the sake of completeness, we have reported in the paper all (53) candidates found. However, if we consider only the ones that fall into the refined localization map, the number decreases to 40. After removing d9, which was identified as a Gaia transient before the occurrence of the GW event, we are left with 39 ‘bona fide’ candidates.

Our first two runs were carried out within two days from the GW event then the observations became more sparse (this also because we prioritize observations of the kilonova AT 2017gfo). It is worth noting that the tiles added to improve the coverage of the refined map started from the third epoch and that they cover almost half of the total contained probability. This implies that our observations tend to constrain models that predict slow-declining light curves like in Bartos (2016) AND Stone et al. (2017) where a new BBH formation channel inside AGN discs is proposed. In these models, the presence of substantial gas densities around the merging stellar mass BHs can produce strong, slow-transient EM counterpart. In this respect, our work is complementary to the work of Doctor et al. (2019) that put constraints on fast fading transients. Particularly one of the two candidates that they found is outside our search area and the other one is too faint.

The VST plan for future EM follow-up of BBH mergers is to have two or more observing epochs within the first couple of days after the alert to search for fast-declining transients and then keep observing for few weeks to put constraints on those models that predict slower declining light curves.

ACKNOWLEDGEMENTS

The authors thank the anonymous reviewer for constructive comments on the manuscript. This paper is based on observations made with the European Southern Observatory/VST. We acknowledge the usage of the OmegaCam and VST Italian Guaranteed time. We also acknowledge Istituto Nazionale di Astrofisica (INAF) financial support of the project ‘Gravitational Wave Astronomy with the first detections of adLIGO and adVIRGO experiments’. This research has made use of public data of the LIGO Scientific Collaboration and the Virgo Collaboration. LIGO is funded by the U.S. National Science Foundation. Virgo is funded by the French Centre National de Recherche Scientifique (CNRS), the Italian Istituto Nazionale della Fisica Nucleare (INFN), and the Dutch Nikhef, with contributions by Polish and Hungarian institutes. SC acknowledges partial funding from Agenzia Spaziale Italiana-Istituto Nazionale di Astrofisica grant I/004/11/3. AR acknowledges support from Premiale LBT 2013. We acknowledge partial support by the PRIN-INAF 2016 with the project ‘Towards the SKA and CTA era: discovery, localization, and physics of transient sources’ (PI: M. Giroletti).

This project used public archival data from the Dark Energy Survey (DES) as distributed by the Science Data Archive at NOAO. Funding for the DES Projects has been provided by the DOE and NSF (USA), MISE (Spain), STFC (UK), HEFCE (UK), NCSA (UIUC), KICP (U. Chicago), CCAPP (Ohio State), MIFPA (Texas A|$\&$|M), CNPQ, FAPERJ, FINEP (Brazil), MINECO (Spain), DFG (Germany), and the collaborating institutions in the DES, which are Argonne Lab, UC Santa Cruz, University of Cambridge, CIEMAT-Madrid, University of Chicago, University College London, DES-Brazil Consortium, University of Edinburgh, ETH Zürich, Fermilab, University of Illinois, ICE (IEECCSIC), IFAE Barcelona, Lawrence Berkeley Lab, LMU München and the associated Excellence Cluster Universe, University of Michigan, NOAO, University of Nottingham, Ohio State University, OzDES Membership Consortium, University of Pennsylvania, University of Portsmouth, SLAC National Lab, Stanford University, University of Sussex, and Texas A|$\&$|M University. Based on observations at Cerro Tololo Inter-American Observatory, National Optical Astronomy Observatory (NOAO Prop. ID 2012B-0001; PI: J. Frieman), which is operated by the Association of Universities for Research in Astronomy (AURA) under a cooperative agreement with the National Science Foundation.

FACILITY: VST ESO programmes: 099.D-0191, 099.D-0568, 0100.D-0022.

Footnotes

REFERENCES

APPENDIX A: BRIGHT STARS SPIKES MASKING

Figure A1.

Example of bright star mask.

Open in new tabDownload slide

APPENDIX B: TRANSIENTS

In Figs B1 and B2 detailed images of the 53 transients candidates are shown. Each cut-out, extracted from the original image at the epoch of brightest magnitude, is centred on the transient and has a size of 1 arcmin x1 arcmin. In Table B1 is reported the transients photometry catalogue. In case of missing detection in a particular epoch, we report the limiting magnitude of the image preceded by the symbol > .

Figure B1.

First 28 transient candidates found with the VST in |$99\, {\rm deg^2}$| covering the GW170814 error area. Each image has a size of 1 arcmin x 1 arcmin and is extracted from the original images at the epoch of the brightest magnitude.

Open in new tabDownload slide

Figure B2.

Images of transient candidates from c29 to d17 at the brightest magnitude. The size is 1 arcmin x 1 arcmin.

Open in new tabDownload slide