A Search for Optical Emission from Binary Black Hole Merger GW170814 with the Dark Energy Camera

Since the first binary black hole (BBH) merger detection in 2015 September (Abbott et al. 2016a), mergers of two black holes have become a mainstay of gravitational-wave (GW) astrophysics. The first five observed BBHs, found only by the Hanford and Livingston Laser Interferometer Gravitational-Wave Observatory (LIGO) detectors, offered significant astrophysical insight into the BBH mass distribution and event rates (Abbott et al. 2016a, 2016b, 2016c, 2016d, 2017b, 2017c). For electromagnetic (EM) follow-up, however, the two LIGO detectors alone place poor constraints on the sky position, typically a few hundred deg².

To date, no compelling optical counterparts to BBH mergers have been identified. However, Stalder et al. (2017) found optical candidate ATLAS17aeu in their follow-up of GW170104 and hypothesize a chance coincidence. Additionally, a weak gamma-ray burst in coincidence with GW150914 was reported in Connaughton et al. (2016), but its association with GW150914 is still under dispute. There are three (not mutually exclusive) reasons for non-detections: (1) the probable sky regions of previous BBH detections were not searched comprehensively, (2) the BBH emission could not be identified or distinguished from background transients, and/or (3) optical emission from BBH mergers is non-existent or below the detectable threshold at the times of the existing observations. Theoretical models have been proposed that could produce EM signals (e.g., Loeb 2016; Perna et al. 2016; de Mink & King 2017; Stone et al. 2017; McKernan et al. 2018), but these models are highly speculative. With little theoretical guidance, there is a need for more complete searches for BBH EM emission while also controlling false-positive event rates. Detection of BBH EM counterparts would be of immense scientific value (e.g., Phinney 2009), as it could constrain the formation environments of BBHs, the behavior of matter in strong field gravity, and cosmological parameters such as the Hubble constant.⁶⁴

Thus far, a number of optical follow-up campaigns have been conducted to search for BBH counterparts (e.g., Cowperthwaite et al. 2016; Morokuma et al. 2016; Smartt et al. 2016a, 2016b; Soares-Santos et al. 2016; Lipunov et al. 2017; Stalder et al. 2017; Yoshida et al. 2017; Smith et al. 2018). However, the large probable sky areas of the "double-coincident" LIGO detections (Hanford and Livingston detectors only) curtailed searches for EM counterparts from BBH mergers. For example, Soares-Santos et al. (2016) observed 102 deg² of the GW150914 high-probability sky region with the optical imager Dark Energy Camera, (DECam; Flaugher et al. 2015), corresponding to 38% of the initial LIGO sky map probability. After accounting for the lack of existing images (templates) for difference imaging, a shift in the sky map in a reanalysis of LIGO data, and other efficiency losses, only 3% of the probable GW150914 sky area was searched and analyzed. Similarly, the DECam follow-up campaign of GW151226 reported in Cowperthwaite et al. (2016) covered 29 deg², just ∼2% of the final GW151226 high-probability region. In contrast, with the three-detector network including the Virgo interferometer, the smaller 28 deg² 90% localization region of neutron-star merger GW170817 enabled 81% DECam coverage of the final LIGO-Virgo sky map and identification of the EM counterpart (Abbott et al. 2017a; Soares-Santos et al. 2017). These searches were all performed in the i- and z-bands, requiring two tilings of the search area. We note that these DECam searches attempted to tile maximal sky map probability, but for the nearby events such as GW170817, targeting based on galaxy catalogs can be successful (e.g., Arcavi et al. 2017; Coulter et al. 2017; Valenti et al. 2017).

On 2017 August 14, the LIGO-Virgo Collaboration (LVC), with the addition of the Virgo detector, made the first "triple-coincident" detection of GWs from a BBH event, GW170814, and provided a much tighter constraint on the sky position of the source than those of previous BBH detections (Abbott et al.2017d; The LIGO Scientific Collaboration & the Virgo Collaboration 2018). The detection of GW170814, with its 87 deg² 90%-localization region, enabled our team to perform a comprehensive search of the sky area for BBH merger optical counterparts and significantly improve our sensitivity to BBH merger EM emission models.

We report on our search for optical counterparts to GW170814 using DECam. In Section 2, we describe the parameters and cadence of our follow-up observations, which extended to 12 days after the GW170814 trigger and covered 225 deg². Section 3 describes the analysis. Finally, Section 4 presents the results of the analysis, which we then comment on in Sections 5 and 6.

On 2017 August 14 at 10:30:43 UTC, the LVC reported a signal consistent with the inspiral and merger of two black holes of masses ${30.5}_{-3.0}^{+5.7}\,{M}_{\odot }$ and ${25.3}_{-4.2}^{+2.8}\,{M}_{\odot }$ at a luminosity distance of ${540}_{-210}^{+130}$ Mpc and redshift⁶⁵ $z={0.12}_{-.04}^{+.03}$ (Abbott et al. 2017d). LIGO and Virgo sent out a Bayestar sky map 2 hr after the trigger (Singer & Price 2016; LIGO-Virgo Collaboration 2017a) and we captured our first DECam image of the probability region at 06:00 UTC on August 15, 19.5 hr after the GW detection. DECam is an optical imager, installed on the Blanco 4 m telescope at the Cerro Tololo Inter-American Observatory. It has a 3 deg² field of view and is equipped with several broadband optical/near-infrared (NIR) filters (u, g, r, i, z, Y, VR), making it well suited to search for faint transients over large sky areas (Flaugher et al. 2015).

We imaged the high-probability area of the Bayestar sky map in the i-band with 90 s exposures, corresponding to a 5σ point-source depth of ≈23 mag. Our strategy of imaging the most probable regions was similar to that used in Soares-Santos et al. (2016) and Cowperthwaite et al. (2016) (who surveyed in the i- and z-bands), but in order to maximize the sky area coverage we only surveyed in the i-band. Our search covered 225 deg², corresponding to 90% of the initial Bayestar map, 90% of the LALInference sky map that was released with the original GW170814 LIGO-Virgo publication (Abbott et al. 2017d), and 86% of the final map from the O1-O2 GW catalog (The LIGO Scientific Collaboration & the Virgo Collaboration 2018). These estimates account for chip gaps on the camera but not masking of bright stars. A preliminary LALInference map accounting for calibration uncertainties was sent after our first night of observations, causing a shift in the search region on the second night of observations and onward (LIGO-Virgo Collaboration 2017b). Figure 1 shows our tiling over the LIGO-Virgo sky maps. Observations of the region of interest were taken in epochs that began roughly 0.8, 1.8, 2.8, 5.8, 8.8, 9.8, 10.8, and 11.8 days after the GW event, and each epoch's tiling spanned about 4 hr. The 5σ limiting point-source magnitude in the i-band was approximately 24 over all the tiles on the first night of observations. The first DECam image was taken on 2017 August 15 at 06:05:31 UTC. Our cadence was chosen to have a dense sampling in time, but observing conditions and follow up of GW170817 introduced larger gaps between the third and eighth nights.

Zoom In Zoom Out Reset image size

We processed the images from our search using the Dark Energy Survey's (DES) transient detection pipeline as in Soares-Santos et al. (2016) and K. Herner et al. (2019, in preparation). The pipeline consists of a single-epoch processing stage (Bernstein et al. 2017; Morganson et al. 2018) followed by a stage that takes the difference of search images and template images to identify sources with fluctuating brightness (DiffImg; Kessler et al. 2015). Template images were available from existing DES data as the LIGO-Virgo sky maps were contained in the DES footprint. The sources detected in the pipeline are used to generate candidate light curves: a candidate requires at least two detections by the pipeline, and for each candidate a light curve is constructed from a point-spread function (PSF)-fitted flux at each observation. The pipeline also removes persistent point sources in the DES Y1 catalog that are brighter than 20.5 mag in any band.

We split the data into two samples because of a shift in the GW sky map after the first night of observations. This shift prompted a change in the patch of sky that we targeted, creating inhomogeneity in the data sample as the cadence of observations was not uniform over the full area that we imaged. The red and orange "hexes" in Figure 1 show which fields were observed the first night versus only on later nights, respectively. The first data set, D1, includes the ${N}_{{\rm{D}}1}=42368$ candidates that were first observed ∼0.8 days after the GW trigger when we were targeting the Bayestar sky map. D2 contains the ${N}_{{\rm{D}}2}=17192$ candidates observed for the first time after acquiring the preliminary LALInference sky map. Over the full GW170814 follow-up campaign, the median number of observations per candidate is eight and five for D1 and D2, respectively.

To identify candidates of interest, we apply selection requirements (or "cuts") to the full set of candidates produced by DiffImg. We present these criteria in Section 3.2 and have chosen them to (a) minimize contamination from both astrophysical transients such as supernovae (SNe) and asteroids as well as artifacts in the data, and (b) identify "fast transients" that quickly decline after the merger. SNANA simulations (Kessler et al. 2009) of Type Ia and core-collapse SN light curves (using the SALT-II Ia light curve model of Guy et al. (2010) and Ibc, IIp, IIn core-collapse templates from Kessler et al. 2010) provide guidance on cuts to remove SNe. A full optimization and exploration of the cuts is not explored here and is left for future analyses. We choose these cuts using a control sample of candidates that are away from the highest probability regions of the LALInference sky map, as described in Section 3.1. The number of candidates remaining in the control sample after applying cuts is used to infer the number of candidates expected in the full sample. This inference is detailed in Section 3.3.

Our analysis also makes use of two auxiliary tools. The first is a machine-learning (ML) algorithm, autoScan, trained on DES difference images that produces a score for each difference-imaging detection between 0 and 1, with high scores corresponding to high-confidence point-source-like sources (Goldstein et al. 2015). The second is a set of object-fitting algorithms that classify persistent DES sources as galaxies or stars depending on the sources' spatial extent (Drlica-Wagner et al. 2018; Sevilla-Noarbe et al. 2018). We use these tools to identify high-confidence point-source detections in our search and to match these detections with stars and galaxies.

3.1. Control Sample

3.2. Selection Requirements

3.3. Expectation of Number of Candidates in Full Sample

Table 2 shows the effect of the cuts on the full sample, which includes the control sample. It also shows the initial i-band magnitudes and sky positions for the events passing all cuts. After analyzing the blinded sample, one more candidate is found, leaving a total of two candidates passing cuts in the control and blinded samples, with ID numbers 1 and 2, respectively. Finding one candidate passing cuts in the blinded sample is consistent with the 2.4 expected background events derived from the control sample presented in Section 3.3. The light curves for both events and their sky positions are shown in Figure 3.

Table 2.  Top: Candidates Remaining in the Full Data Sample after Applying Cuts

Cuts	Nseqa	Nonlyb	NLOc
1. Raw sample	59,560	⋯	⋯
2. First epoch ML > 0.7	1206	1206	258
3. Unmatched or host z < 0.30	730	31,119	8
4. Second obs. S/N ≥ 2.0	663	44,181	4
5. >2.0 sigma decline	45	5570	65
6. Nobs ≥ 4	31	50,029	2
7. Late-time S/N < 6	4	27,571	21
8. No late-time brightening	2	36,499	4
9. Visual inspection	2	⋯	2

Candidate #	R.A.	Decl.	mi
1	4235047	−4032632	22.5
2	4763365	−3636045	21.9

Notes. Bottom: sky coordinates and initial i-band magnitude m_i of the two candidates passing all cuts.

^aNumber of candidates remaining after applying each cut sequentially. ^bNumber of candidates after applying an individual cut. ^cNumber of candidates if a cut is "left out" but all the rest are applied.

Download table as:  ASCIITypeset image

Upon visual inspection of the two candidates, neither is an obvious subtraction artifact or cosmic ray. Here we do not show examples of subtraction artifacts and cosmic rays that would be cut by visual inspection, because visual inspection did not end up removing any candidates in this analysis. However, the template images for both candidates contain a bright source at the position of the candidates. The template, search, and difference images from the first epoch of observations of each candidate are shown in Figure 2.

Zoom In Zoom Out Reset image size

A deeper search through the DES high-quality object catalog ("Y3 Gold") reveals that Candidate 1 is associated with an object that is classified as either a galaxy at z ∼ 0.9, or a star, depending on the classifier used. A multi-epoch, multi-object-fitting algorithm classifies the object's PSF as a candidate star, whereas the single-object fit categorizes the object as a galaxy (Drlica-Wagner et al. 2018; Sevilla-Noarbe et al. 2018). Notably, the object is too faint to meet the brightness cutoff for inclusion in our star veto catalog and it is not vetoed by our host galaxy redshift cut (cut 3) because we only include the highest confidence galaxies in the host-galaxy matching. Fitting each band to a constant flux for all archival observations of the object (DES Years 1–4) results in a χ²/DOF of 48.6/17 = 2.9 and p-value $p({\chi }^{2}\geqslant 48.6| \mathrm{DOF}=17)=7\times {10}^{-5}$, indicating previous variability of the source. These archival fluxes are shown in Figure 3. Spectroscopic observations of this source could clarify if the object is a star or galaxy.

Zoom In Zoom Out Reset image size

Candidate 2 is also associated with a DES Y3 Gold object and is classified as a star by both classifiers; constant-flux fits to archival observations yield a χ²/DOF of 25.7/14 = 1.8 and p-value of $p({\chi }^{2}\geqslant 25.7| \mathrm{DOF}=14)=0.03$ (see Figure 3). However, the star is also too faint (by 0.16 mag) to meet the brightness cutoff for the star veto catalog of our pipeline, and hence was not removed by the 20.5 mag persistent-point-source cut in Section 2.

Although our search identified two interesting candidates, it is unlikely that either candidate is associated with GW170814. Neither candidate is located in the 90% confidence region of the LALInference sky map, and both are associated with existing objects in DES catalogs that are inconsistent with our expectations of the GW source. Candidate 2 is likely the transient behavior of a variable star and is consistent with the number of background candidates expected in the blinded sample. Candidate 1 could also be stellar variability, or it could be a signal associated with a distant galaxy. Assuming that it is a galaxy, the DNF photometric redshift of the object is z = 0.95 ± 0.12, far beyond the possible redshift of GW170814 at that sky position. The 99% upper limit on the GW distance along the Candidate 1 line of sight is 615 Mpc, whereas the galaxy distance is ${6380}_{-980}^{+1010}$ Mpc assuming the LCDM cosmology parameters of Planck Collaboration (2016). We note, however, that photometric redshifts can occasionally have catastrophic failures.

An alternative explanation for the persistent emission from the two candidates is that one or both of these candidates is associated with a quasar. If either is a quasar, it is unlikely to be at the low redshifts of interest for GW170814 (Pâris et al. 2017). Spectroscopic follow up of the persistent sources associated with Candidates 1 and 2 could resolve the question of whether or not we have miscategorized them.

We conclude that these two candidates are not associated with GW170814, and thus we find no EM counterpart associated with the BBH merger over the 225 deg² region that we surveyed with 86% sky map coverage. We have not yet computed the efficiency, which is needed to set rate limits on BBH merger emission, but this rate-limit analysis is underway using SNANA simulations similar to those used in Soares-Santos et al. (2016). Our rate-limit analysis will also re-evaluate the cuts to maximize possible BBH model efficiency while minimizing SN background events. Qualitatively though, the analysis presented here covers 86% of the GW sky map and searches for events with rapidly declining light curves. The non-detection of an EM counterpart in our sample results in stringent limits on fast-declining optical models brighter than i ∼ 23 mag 1–2 days after the BBH coalescence. This search is not sensitive to models that fade faster than the time between the first two observations due to Cut 4. Assuming a flat-in-frequency optical spectrum from 4000 to 7000 Å and the GW170814 median distance, this i ∼ 23 mag limit corresponds to a luminosity limit of L_optical ∼ 5 × 10⁴¹ erg s⁻¹.

Our results constrain the space of models put forth in e.g., Stone et al. (2017) and de Mink & King (2017). For example, Stone et al. (2017) posited that BBH mergers occurring in the gaseous environments of active galactic nuclei disks could be accompanied by gas accretion onto the final merged black hole that powers luminosities of order L ∼ 10⁴⁰ erg s⁻¹ lasting a few years, but highly super-Eddington accretion might result in a brighter and shorter-lived transient than our analysis is sensitive to. Our search also narrows the feasibility of models from de Mink & King (2017), which predict emission with luminosities of approximately L ∼ 10⁴² erg s⁻¹ occurring on fast timescales. The search performed here is tailored to remove longer-lived transients, and therefore it does not constrain long-lived BBH counterparts, such as the SN association suggested in Loeb (2016) and later discussed in Woosley (2016) and D'Orazio & Loeb (2018).

Aside from identifying interesting candidates, our search for counterparts to GW170814 is a test-bed for future BBH follow-up analyses where the sky-map credible areas will be small enough to be completely tiled in less than one night using DECam. For a real-time search for future counterparts, we consider resources to spectroscopically follow ∼10 candidates, which we would want to identify within roughly two days of the GW trigger. In this scenario, we only apply the first five cuts in Table 2, as the remaining cuts depend on observations beyond two days. Through cut 5 (>2σ decline), our search finds 45 candidates. Of these, we find that four candidates (including Candidates 1 and 2) are associated with DES-catalog objects that are either galaxies beyond our redshift cut (cut 3), or stars, and are thus uninteresting as black-hole-merger counterparts. Excluding these four candidates, our real-time search would find 41 candidates over 225 deg² or ∼16 candidates per 87 deg² (the 90% credible area of the final GW170814 sky map). For comparison, Cowperthwaite & Berger (2015) predicted ∼19 Type Ia SNe detected at z < 0.25 over a 7 day, 87 deg² search. This suggests that the first five cuts are adequate to find interesting spectroscopic targets over a region the size of the GW170814 sky map.

Future work will incorporate simulations of BBH and SN light curves to assess the efficiency and false alarm rate of our search. If several BBH events are followed up with no EM counterpart found, a combined analysis will be needed to set limits on BBH EM emission.

We have presented an optical search for counterparts to gravitational wave GW170814 using DECam. Our search covered 225 deg², corresponding to 86% of the final LIGO-Virgo sky map. Our difference-imaging pipeline produces 59560 light curves from the search images that are analyzed with the criteria in Section 3.2. After applying these cuts to the i-band light curves, two candidates remain. These two candidates are most likely not associated with GW170814: one is a high-confidence variable star, and the other is either a variable star or a transient associated with a high-redshift galaxy well beyond the expected GW170814 redshift.

With no candidates associated with GW170814, our analysis disfavors fast-declining optical emission from BBH mergers 1–2 days after merger with i ≲ 23 mag. Future work will assess the efficiency and false-positive rate in optical BBH searches such as this one using simulations of BBH and SN light curves. Additionally, we will consider updates to our star veto catalog and galaxy catalog to account for fainter stars and objects with uncertain star or galaxy classification.

Tens of BBH signals are expected in the LVC's third operating run, and some are likely to have localization regions of similar size to that of GW170814. Based on the search and analysis presented here, we are preparing to search for additional BBH merger signals and quickly identify candidates for spectroscopic follow up. With future BBH optical searches and forward modeling of background and foreground signals, we will set increasingly stringent limits on BBH EM emission. Although BBH mergers may remain electromagnetically dark, the future of BBH astrophysics is bright.

Z.D. would like to thank Reed Essick for useful discussions and would also like to thank the Dark Cosmology Center and Neils Bohr Institute, where he was studying at the time of the GW170814 alert. Z.D. is supported by NSF Graduate Research Fellowship grant DGE-1144082. Z.D. and D.E.H. were partially supported by NSF CAREER grant PHY1151836 and NSF grant PHY-1708081. H.-Y.C. was supported by the Black Hole Initiative at Harvard University, through a grant from the John Templeton Foundation. R.J.F. is supported in part by NASA grant NNG17PX03C, NSF grant AST-1518052, the Gordon & Betty Moore Foundation, the Heising-Simons Foundation, and by a fellowship from the David and Lucile Packard Foundation. This work was completed in part with resources provided by the University of Chicago Research Computing Center and support from the Kavli Institute for Cosmological Physics at the University of Chicago through NSF grant PHY-1125897 and an endowment from the Kavli Foundation.

Funding for the DES Projects has been provided by the DOE and NSF (USA), MEC/MICINN/MINECO (Spain), STFC (UK), HEFCE (UK). NCSA (UIUC), KICP (U. Chicago), CCAPP (Ohio State), MIFPA (Texas A&M), CNPQ, FAPERJ, FINEP (Brazil), DFG (Germany), and the Collaborating Institutions in the Dark Energy Survey.

The Collaborating Institutions 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 (IEEC-CSIC), IFAE Barcelona, Lawrence Berkeley Lab, LMU München and the associated Excellence Cluster universe, University of Michigan, NOAO, University of Nottingham, Ohio State University, University of Pennsylvania, University of Portsmouth, SLAC National Lab, Stanford University, University of Sussex, Texas A&M University, and the OzDES Membership Consortium.

Based in part on observations at Cerro Tololo Inter-American Observatory, National Optical Astronomy Observatory, which is operated by the Association of Universities for Research in Astronomy (AURA) under a cooperative agreement with the National Science Foundation.

The DES Data Management System is supported by the NSF under grant No. AST-1138766 and AST-1536171. The DES participants from Spanish institutions are partially supported by MINECO under grants AYA2015-71825, ESP2015-66861, FPA2015-68048, SEV-2016-0588, SEV-2016-0597, and MDM-2015-0509, some of which include ERDF funds from the European Union. IFAE is partially funded by the CERCA program of the Generalitat de Catalunya. Research leading to these results has received funding from the European Research Council under the European Union's Seventh Framework Program (FP7/2007–2013) including ERC grant agreements 240672, 291329, and 306478. We acknowledge support from the Australian Research Council Centre of Excellence for All-sky Astrophysics (CAASTRO), through project number CE110001020, and the Brazilian Instituto Nacional de Ciência e Tecnologia (INCT) e-Universe (CNPq grant 465376/2014-2).

This manuscript has been authored by Fermi Research Alliance, LLC under contract No. DE-AC02-07CH11359 with the U.S. Department of Energy, Office of Science, Office of High Energy Physics.