J-GEM Follow-Up Observations of The Gravitational Wave Source GW151226

We report the results of optical--infrared follow-up observations of the gravitational wave (GW) event GW151226 detected by the Advanced LIGO in the framework of J-GEM (Japanese collaboration for Gravitational wave ElectroMagnetic follow-up). We performed wide-field optical imaging surveys with Kiso Wide Field Camera (KWFC), Hyper Suprime-Cam (HSC), and MOA-cam3. The KWFC survey started at 2.26 days after the GW event and covered 778 deg$^2$ centered at the high Galactic region of the skymap of GW151226. We started the HSC follow-up observations from 12 days after the event and covered an area of 63.5 deg$^2$ of the highest probability region of the northern sky with the limiting magnitudes of 24.6 and 23.8 for i band and z band, respectively. MOA-cam3 covered 145 deg$^2$ of the skymap with MOA-red filter 2.5 months after the GW alert. Total area covered by the wide-field surveys was 986.5 deg$^2$. The integrated detection probability of all the observed area was $\sim$29%. We also performed galaxy-targeted observations with six optical and near-infrared telescopes from 1.61 days after the event. Total of 238 nearby (<100 Mpc) galaxies were observed with the typical I band limiting magnitude of $\sim$19.5. We detected 13 supernova candidates with the KWFC survey, and 60 extragalactic transients with the HSC survey. Two third of the HSC transients were likely supernovae and the remaining one third were possible active galactic nuclei. With our observational campaign, we found no transients that are likely to be associated with GW151226.


Introduction
Gravitational wave (GW) is a quadrupole wave of space-time distortion propagating with the light speed. Strong GW is emitted by violent gravitational disturbance induced by a coalescence between compact massive objects such as neutron stars (NSs) or black holes (BHs). In order to observe GW directly, new generation GW detectors; Advanced LIGO (aLIGO, Abbott et al. 2016b), Advanced Virgo (aVirgo; Acernese et al. 2015), and KAGRA (Somiya 2012) are being constructed. If the planned sensitivities are achieved, these GW detectors can detect GW signals from an NS-NS merger at a distance of 200 Mpc (Abadie et al. 2010). The GW detection rate is anticipated to be in a range of 0.4-400 events yr −1 for NS-NS merger (Abadie et al. 2010). Uncertainty of the above number primarily comes from the limit of our knowledge on real number of NS binary fraction in a galaxy.
If a compact object merger contains one NS, wide wavelength range of electromagnetic (EM) emission associated with GW is expected (Li & Paczynski 1998;Rosswog 2005;Nakar & Piran 2011;Roberts et al. 2011;Metzger et al. 2010;Metzger & Berger 2012;Barnes & Kasen 2013;Hotokezaka et al. 2013;Berger 2014;Tanaka et al. 2014). The EM emission would tell us important pieces of information about the nature of the GW event; its astrophysical origin, detailed localization, accurate distance, and local environment of the event. Most promising optical-near-infrared emission from GW sources is radioactively-powered emission, so called "kilonova" or "macronova" associated with NS-NS or BH-NS mergers (Metzger & Berger 2012; Barnes & Kasen 2013;Tanaka et al. 2014). A strong tidal force induced by merging process blows out the outer layer of NS, and a wide solid angle outflow from the merger emits a wide range of EM emission due to radioactive decay of the ejecta, that is "kilonova". Neutron rich ejecta of a kilonova produce huge amount of r-process elements, thus the kilonova emission gives important clues to the long standing mystery about the sites of cosmic r-process nucleosynthesis. Moreover, the luminosity and light curve of a kilonova would allow us to constraint the equation of state of NS. To search for EM emission associated with GW events, we organized an EM follow-up observation network J-GEM (Japanese collaboration of Gravitational wave Electro-Magnetic follow-up; Morokuma et al. 2016) by utilizing optical-infrared-radio telescopes of Japan.
The first direct detection of GW was achieved by aLIGO on Sep. 14 2015 (Abbott et al. 2016a). aLIGO performed the first science run (O1) from Sep. 2015 to Jan. 2016. Just before the regular operation of O1, aLIGO detected the GW at Sep. 14 2015 09:50:45 UT (Abbott et al. 2016a). The GW from this event, which was named as GW150914, was emitted by a 36 M⊙-29 M⊙ binary BH coalescence. While many electromagnetic (EM) follow-up observations were performed for GW150914 (Abbott et al. 2016d;Abbott et al. 2016e;Ackermann et al. 2016;Evans et al. 2016a;Kasliwal et al. 2016;Lipunov et al. 2016;Morokuma et al. 2016;Serino et al. 2016;Smartt et al. 2016a;Soares-Santos et al. 2016;Troja et al. 2016), no clear EM counterpart was identified with those observations except for a possible detection of γ-ray emission by F ermi Gamma-ray Burst Monitor (GBM) (Connaughton et al. 2016). However, the F ermi GBM detection was not confirmed by INTEGRAL observations .
Here, we report the EM counterpart search for GW151226 performed in the framework of J-GEM. We assume that cosmological parameters h0, Ωm, and Ω λ are 0.705, 0.27, and 0.73, respectively (Komatsu et al. 2011) in this paper. All the photometric magnitudes presented in this paper are AB magnitudes.

Observations
We performed wide-field survey and galaxy targeted followup observations in and around the probability skymap of GW151226. The 90% credible area of the initial skymap created by BAYESTAR algorithm (Singer et al. 2014) was ∼1400 deg 2 (LSC and Virgo 2015). The final skymap was refined by LALInference algorithm (Veitch et al. 2015) and the 90% area is finally 850 deg 2 (Abbott et al. 2016c). We also made an integral field spectroscopy for an optical transient (OT) candidate reported by MASTER. The specifications of the instruments and telescopes we used for the follow-up observations are summarized in Morokuma et al. (2016).

Wide Field Survey
We used three instruments for the wide-field survey; KWFC (Sako et al. 2012)  The KWFC survey observations were done in r-band on Dec. 28 and 29 and Jan. 1-6 (UT). The total area observed with KWFC was 778 deg 2 far off the Galactic plane. To perform an image subtraction with the archival SDSS (Sloan Digital Sky Survey; Shadab et al. 2015) images, the high probability regions had to be avoided. Each field was observed typically twice or three times. The exposure time is 180 sec each and the seeing was 2.5-3.0 arcsec FWHM.
We carried out an imaging follow-up observations with HSC in the first half nights of Jan. 7, 13, and Feb. 6, 2016 (UT). We observed an area of 63.5 deg 2 centered at (α, δ) = (03:33:45, +34:57:14) spanning over the highest probability region in the initial skymap (BAYESTAR) with 50 HSC fiducial pointings. The fiducial pointings were aligned on a Healpix (Gorski, et al. 2005) grid with NSIDE=64 (a corresponding grid size is 0.84 deg 2 ). To remove artifacts efficiently, we visited each fiducial pointing twice with a 2 arcmin offset. We observed the field in i-band and z-band with an exposure time ranging from 45 sec to 60 sec for each pointing. On Feb. 6, first we surveyed all the fields by single exposure, then observed the whole area again. The seeing ranged from 0.5 arcsec to 1.5 arcsec FWHM.
We also performed survey observations with MOA-cam3 for a part of the skymap in the southern hemisphere from UT Mar. 8 to 11 2016. The total area covered by the MOA-cam3 observations was 145 deg 2 . The "MOA-Red" filter (Sako et al. 2008), which is a special filter dedicated to micro-lens survey with a wide range of transmission from 6200Å to 8100Å was used. The exposure time per field was 120 sec. The seeing was 1.9-4.5 arcsec FWHM.
Since the sky areas observed by the three instruments were not overlapped, the total area covered by the wide-field surveys was 986.5 deg 2 . The integrated detection probabilities of the observed regions for the final skymap (LALInference) were 0.07, 0.09, and 0.13 for HSC, KWFC, and MOA-cam3, respectively. We thus covered a total of ∼29% of the probability skymap of GW151226.
The wide-field survey observations are summarized in Table  1. The survey areas and the probability skymap of GW151226 are shown in Figure 1. An enlarged map of the sky areas observed with HSC is shown in Figure 2.
We selected 309 nearby galaxies from GWGC (Gravitational Wave Galaxy Catalog) (White, Daw & Dhillon 2011) in the skymap regions whose detection probabilities are more than 0.0008. We divided the target galaxies into 4 target groups. The groups 1 to 3 contain northern galaxies accessible from Japan. The number of galaxies of the group 1, 2, and 3 are 77, 76, and 77, respectively. The group 4 contains 79 southern galaxies. We allocated these groups to the above telescopes as target lists.
The summary of the targeted observations is shown in Table  2. The net number of the observed galaxies was 238. The spatial and distance distributions of the observed galaxies are shown in Figure 3 and Figure 4, respectively.

Spectroscopic Follow-up
We carried out a spectroscopic observation of MASTER OT J020906.21+013800.1 (Lipunov et al. 2015) with a fiber-fed integral field spectrograph KOOLS-IFU attached to the 188 cm telescope at Okayama Astrophysical Observatory on UT Dec. 28 2015. The field of view of KOOLS-IFU is 1.8 arcsec per fiber and 30 arcsec in total. The wavelength range and spectral resolving power were 5020-8830Å and 600-850, respectively. The total exposure time was 3600 sec.

KWFC survey
The data reduction of the KWFC data was made using the standard data reduction pipeline developed for Kiso Supernova Survey (KISS; Morokuma et al. 2014). The pipeline functions include bias subtraction, overscan subtraction, overscan trimming, flat-fielding, point spread function (PSF) size measurements, astrometry relative to the USNO-B1.0 catalog (Monet et al. 2003), zeropoint magnitude determination relative to the SDSS, image subtraction using the SDSS images, and detection  of transient candidates in the subtracted images. The 5σ limiting magnitudes of the KWFC observations ranged from 18.0 to 20.5 depending on the sky condition of the Kiso observatory.
The transient candidates detected in the subtracted images include not only astronomical objects but also non-astronomical artifacts, such as cosmic rays, residual of image subtraction due to imperfect image alignment or convolution (see e.g. Bailey et al. 2007;Bloom et al. 2012). Moreover, astronomical objects include minor planets or variable stars in addition to extragalactic transients. Therefore, we first removed the transient candidates around the objects which are registered as star in the SDSS catalog. This effectively removed both variable stars and artifacts around bright stars. Then, all the sources matched with the database of the Minor Planet Center were removed. Finally, the remaining objects were visually inspected to remove artifacts.
As a result, we found 13 extragalactic transient candidates associated with galaxies. The candidates found with the KWFC are summarized in Table 3. Nine out of 13 objects are detected more than twice in our survey. The other four objects (KISS15ah, KISS15ai, KISS16b, and KISS16c) were detected only once. Since KISS15ah and KISS16c are independently discovered by other groups (AT 2016bse and SN 2015bl, respectively), they must be genuine extragalactic transients. Although there is no independent discovery for KISS15ai and KISS16b, they are rather bright (16.6 and 19.6 mag, respectively), and unlikely to be minor planets which are not registered in the database of the Minor Planet Center.
In Table 3, we show estimated absolute magnitudes of 13 Fig. 4. The distance distribution of the observed galaxies. The distance limit (100 Mpc) of the galaxies is determined by GWGC. transient candidates using spectroscopic and photometric redshifts. Except for KISS16f and KISS16b, the candidates were too bright for the expected kilonova emission (e.g. Tanaka et al. 2014), suggesting that they are supernovae (SNe). KISS16f and KISS16b were rather faint, but their host galaxies are located at z = 0.012 and 0.009964, respectively, and thus they were not associated with GW151226. They are likely to be SNe after the peak brightness.

HSC survey
The HSC data were reduced using HSC pipeline version 4.0.1, which had been developed based on the LSST pipeline (Ivezic et al. 2008;Axelrod et al. 2010). The HSC pipeline provides packages for bias subtraction, flat fielding, astrometry, mosaicing, warping, coadding, and image subtraction. The astrometry and photometry were made relative to the Pan-STARRS1 (PS1, Tonry et al. 2012;Schlafly et al. 2012;Magnier et al. 2013) with a 1.5 arcsec (9 pixel) aperture diameter. The limiting magnitudes were estimated by randomly sampling > 10 5 apertures.
The images taken on Feb. 6 were used as the reference images and were subtracted from the images taken on Jan. 7 and 13. Here, we separately adopted images at 2 epochs on Jan. 7, while images on Jan. 13 were coadded. Point sources in the difference images were detected and measured with the HSC pipeline. Since there were many false detections, we screened the detected sources by the following selection procedure. (1) In order to exclude false detections, we selected point sources detected in both of the z-band difference images on Jan. 7 at the same location with a signal-to-noise ratio of > 5σ, ellipticity of >0.8, and FWHM of 0.8-1.3 arcsec. In addition, a small residual of PSF subtraction from the sources (< 3σ) was imposed.
(2) To select objects fading from Jan. 7 to Feb. 6, fluxes of sources in the two z-band difference images on Jan. 7 were required to be positive. (3) To exclude minor planets, first we estimated the maximum distance that an object could move during an interval between zand i-band imaging observations. We found that it is ∼45 arcsec, assuming that the elongation in the z-band difference image was due to the movement of the object during the exposures. For the sources survived after the selections (1) and (2), we checked the i-band difference images. If a source was not detected but another transient source was found at a distance of 0.5-45 arcsec in i-band difference image, we omitted the source as a possible minor planet. We also check the position of sources with MPChecker.
After the above screening, 1256 candidates remained and were visually inspected. First, we removed clear artifacts from the candidate list by visual inspection. Then we identified and removed slowly moving objects which are thought to be distant minor planets not removed by the above criterion (3) by carefully checking the images. Finally, 60 objects remained as extragalactic transient candidates.
The multicolor light curves of the candidates were derived with forced aperture photometry of the difference images with 1.5 arcsec aperture diameter. We corrected the Galactic extinction using Schlegel et al. (1998).
We compared the color-magnitude time variations of variable component of the transient candidates between Jan. 7 and 13 with the color-magnitude evolutions of Type Ia, Ibc, and IIP SNe and kilonova emission and classified the candidates ( Figure 5). For this comparison, we subtracted the brightnesses at 24 days and 30 days after the explosions from the model light curves of SNe and simulated the color-magnitude evolutions of the variable component of SNe. We adopted fiducial kilonova models of NS-NS merger with ejecta mass of 0.01 M⊙ (model "APR4-1215" of Tanaka et al. 2014) and BH-NS merger with ejecta mass of 0.05 M⊙ (model "H4Q3a75" of Tanaka et al. 2014).
By visual inspection and color-magnitude variation study, we found that two third of the HSC transients were probably SNe. One third of the HSC transients were located very close to the centers of the host galaxies and those time variabilities were not typical of SNe. We thus classified these sources as "active galactic nuclei (AGN)". No source whose color-magnitude variation is consistent with the kilonova models was identified by the above procedure. The extragalactic transient candidates found by the HSC survey are summarized in Table 4. Morokuma et al. (2008) derived the number densities of various transient objects as a function of time interval of i ′ band observations from Subaru Suprime-Cam data in the Subaru/XMM-Newton Deep Field (Furusawa et al. 2008). According to Figure 12 of Morokuma et al. (2008), the number density of extragalactic transients (SNe + AGNs) brighter than the variable component i ′ magnitude i ′ vari of 25 mag with 30 days interval observations is ∼30. The variable component i band limiting magnitude and the number density of the extragalactic transients in our work are ∼24 mag (see Table 4) and ∼1 deg −2 , respectively. Scaling the number density of Morokuma et al. (2008) using Figure 13 of (Morokuma et al. 2008), we estimate that it would be 3-4 deg −2 for the limiting magnitude i ′ vari ∼ 24 mag. This is a few times higher than the value of our observation. Part of this discrepancy would come from our detection strategy. We detected the transients based on z-band observation, thus we could systematically undercount blue transients. In addition, since the Galactic latitude b of the HSC observation field is less than ∼30 deg. (see Figure 1), large fraction of the field suffered from Galactic extinction (typical color excess E(B − V ) is ∼0.3-0.7; see Table  4). Considering these factors, we judge that our observation is roughly consistent with Morokuma et al. (2008).

MOA-II survey
The data of MOA-II were reduced in standard manner of CCD data reduction using IRAF. Astrometry of the data was done using Astrometry.net (Lang 2009). Then point source candidates were extracted with SExtractor (Bertin & Arnouts 1996). After excluding known stars using the USNO-B1.0 catalog, we omitted the candidates whose brightness profiles were not consistent with PSF by profile fitting using IRAF task ALLSTAR. We visually inspected the remaining 2953 candidates and selected 39 sources as transient object candidates. Then we checked 2MASS (Skrutskie et al. 2006) and WISE (Wright et al. 2010) images and found that 33 among the 39 candidates were 2MASS sources and one was a WISE source. Using Minor Planet Checker (MPChecker) 1 , we found 3 candidates were asteroids. One of the candidates was a known supernova PSN J14102342-4318437.
After all selections, one candidate with ∼18.0 mag located at (α, δ) = (14: 44:41.06, −44:4:38.4) remained. This source did not seem to be associated with bright galaxies. We observed this source twice with the interval of 180 sec on Mar. 10 2016 and did not detect significant motion of it between the two exposures. It completely disappeared at the third observation performed at the end of Aug. 2016. Though we cannot exclude the possibility that this source is an extragalactic transient, we think that the most plausible explanation is a minor planet not cataloged in MPChecker.
In the above processing, faint objects embedded in galaxies could be systematically lost. To detect such sources, we selected 2143 galaxies between 250 and 620 Mpc in the observed fields using GLADE (Galaxy List for the Advanced Detector Era) 2 . We found 549 point sources within 5 arcsec around these galaxies. Compared to DSS images, we found all the sources were known objects.

Galaxy Targeted Follow-up Data
The data reduction of the instruments used for the galaxy targeted observations -HOWPol, HONIR, MINT, MITSuME, MOA-II, OAO-WFC, and SIRIUS -was made in a standard manner; overscan correction, bias and dark subtraction, and flat-fielding. Then multiple exposure frames were coadded. The photometric calibrations of the optical data were made by comparing the fluxes of the field stars with those listed in the SDSS or GSC2.3 (Guide Star Catalog version 2.3). For the near-infrared bands data calibration, we used the 2MASS point source catalog (PSC) (Skrutskie et al. 2006). The observed galaxies and the limiting magnitudes of our observations are listed in Figure 1 of on-line supplementary data.
We searched for transient point sources in the observed frames taken with the above instruments by comparing them with DSS red frames for R and I bands, and with 2MASS PSC for near-infrared bands. We found transient candidates in I band frames of the galaxies PGC1202981 and UGC 1410 taken with HONIR on Dec. 28 2015. However, the former one was a Galactic variable star and the latter was a known minor planet. We also found a possible transient candidate close to the nucleus of PGC1021744 in a J band image taken with OAO-WFC on Dec. 28 2015. Since the source was slightly fainter than the 5σ limiting magnitude of the image (∼17.2 mag), the detection was quite marginal. We made a follow-up observation for this object with OAO-WFC in the next night. The limiting magnitude of the observation reached 19.2 mag in J band with an exposure time of 2700 sec, but no point source was found at the same position. We thus could not confirm whether the source was a real astronomical transient.
As a conclusion, no extragalactic transient object was found with our galaxy targeted follow-up of GW151226.

Spectroscopic Follow-up Data
The target of the spectroscopy, MASTER OT J020906.21+013800.1, was reported at the unfiltered magnitude of 18.3 in the skymap area of GW151226 on UT Dec. 27 2015 and reported to be brightening (Lipunov et al. 2015). Our integral field spectroscopy found no significant signal from the OT candidate. Given that the radial intensity profile of the object is the Gaussian with FWHM of 3 arcsec, ∼40% of the object flux falls in 3 fibers. The 5σ limiting magnitude was 17.4 at 7400Å. It is noted that the observations with the 3.6 m TNG starting on UT Dec. 28.8247 2015 also did not find any evidence for the OT with the upper limit of r = 21.

Discussion and Conclusion
No optical and near-infrared counterpart of the gravitational wave event GW151226 was identified by the follow-up observations under the J-GEM collaboration. Other teams' trials to find EM counterparts associated with this event also failed (Adriani et al. 2016;Cowperthwaite et al. 2016;Evans et al. 2016b;Racusin et al. 2016;Smartt et al. 2016b). We found 13 SNe candidates in the KWFC survey data, and 60 extragalactic transients in the HSC survey data. About two third of the HSC transients were probably SNe, and the remaining one third were classified as possible AGNs. There was no source which showed the color-magnitude variation consistent with current kilonova models in our dataset. We thus conclude that this work did not find clear candidates of EM counterpart of the gravitational wave source.
Both of the two GW events, GW150914 and GW151226, detected by aLIGO were BH-BH mergers. Inspired by the possible detection of a γ-ray emission associated with GW150914 by F ermi satellite (Connaughton et al. 2016), several physical mechanisms for EM emission from a BH-BH merger event have been proposed (Morsony et al. 2016;Perna, Lazzti & Giacomazzo 2016;Yamazaki, Asano & Ohira 2016). However, all of those theoretical works have difficulties in producing strong EM emission by a BH-BH merger. In addition, questions have been raised for the reality of the γ-ray detection by F ermi both from theoretical side (Lyutikov 2016;Zhang et al. 2016) and observational and data analysis side (Greiner et al. 2016;Savchenko et al. 2016;Xiong 2016). Thus there is still no observational evidence with concrete theoretical background for EM emission from BH-BH merger. In other words, the key ingredient for detection of EM counterpart associated with GW is whether it contains a neutron star. Hence the information of the chirp mass of GW event is crucial for EM follow-up observations. When the chirp mass and distance estimation of a GW event is distributed, EM follow-up teams will be able to make effective observation plans with their available facilities (Singer et al. 2016).
For considering future observation strategies, we summarized the observation epochs and the limiting magnitudes of the J-GEM follow-up of GW151226 in Figure 6. The limiting magnitudes of R, r, I, i, and MOA-red bands taken with HOWPol, HONIR, MINT, MITSuME, MOA-cam3, KWFC, and HSC are plotted with theoretical i band light curves of kilonovae (Tanaka et al. 2014;Tanaka 2016). Our early observations with the small-and mid-sized-telescopes reached the depth of ∼20 mag in optical red bands. The KWFC data around 6-8 days after the GW event were as deep as ∼20.5 mag. The deepest data taken with HSC reached down to ∼24 mag in i band at 12 days after GW151226. According to the theoretical light curves in Figure  6, the depth of our early galaxy targeted observations reached the detection threshold of kilonova emission from a BH-NS merger within a distance of ∼50-100 Mpc. The late KWFC observations at around 7 days after the GW could follow the candidate. The deep HSC observations could follow the light curve of the candidate at most one month after the event.
However, if the event were NS-NS merger, the story would be completely changed. The kilonova emission for NS-NS merger is too faint to detect with our observations. Even if the event distance is 50 Mpc, the maximum magnitude of the optical emission would be much fainter than ∼19 mag at 1 day after the event. Only HSC could detect the optical emission from a kilonova at a distance of 50-100 Mpc if the follow-up observation with HSC was performed within ∼5 days after the event.
HSC has a capability to survey over ∼60 deg 2 with two colors, i and z bands, with the limiting magnitude of ∼24 mag within a half night. Figure 6 shows that quick (<3 days) followup observations with HSC can detect the optical emission of kilonova induced by an NS-NS merger at a distance of ∼200 Mpc. For BH-NS mergers, relatively slow start of the observation is acceptable. The kilonova EM emission from BH-NS merger at a distance of 400 Mpc would be detectable by HSC even after 10 days from the GW event. When aVirgo goes in regular operation and a joint observation of aLIGO and aVirgo starts, the 90% credible area of GW detection would become smaller than ∼50 deg 2 depending on the signal-to-noise ratio of the event (Singer et al. 2014). This size of area matches very well to the area covered by half night observation of HSC, and thus detection of EM emission from kilonova is greatly antici- pated.