Swope Supernova Survey 2017a (SSS17a), the Optical Counterpart to a Gravitational Wave Source

References

Acknowledgments

We thank the LIGO/Virgo Collaboration, and all those who have contributed to gravitational wave science for enabling this discovery. We thank J. McIver for alerting us to the LVC circular. We thank J. Mulchaey (Carnegie Observatories director), L. Infante (Las Campanas Observatory director), and the entire Las Campanas staff for their extreme dedication, professionalism, and excitement, all of which were critical in the discovery of the first gravitational wave optical counterpart and its host galaxy as well as the observations used in this study. We thank I. Thompson and the Carnegie Observatory Time Allocation Committee for approving the Swope Supernova Survey and scheduling our program. We thank the University of Copenhagen, DARK Cosmology Centre, and the Niels Bohr International Academy for hosting D.A.C., R.J.F., A.M.B., E.R., and M.R.S. during the discovery of GW170817/SSS17a. R.J.F., A.M.B., and E.R. were participating in the Kavli Summer Program in Astrophysics, “Astrophysics with gravitational wave detections.” This program was supported by the the Kavli Foundation, Danish National Research Foundation, the Niels Bohr International Academy, and the DARK Cosmology Centre.

The UCSC group is supported in part by NSF grant AST–1518052, the Gordon & Betty Moore Foundation, the Heising-Simons Foundation, generous donations from many individuals through a UCSC Giving Day grant, and from fellowships from the Alfred P. Sloan Foundation (R.J.F.), the David and Lucile Packard Foundation (R.J.F. and E.R.) and the Niels Bohr Professorship from the DNRF (E.R.). A.M.B. acknowledges support from a UCMEXUS-CONACYT Doctoral Fellowship. M.R.D. is a Hubble and Carnegie-Dunlap Fellow. M.R.D. acknowledges support from the Dunlap Institute at the University of Toronto. B.F.M. is an unpaid visiting scientist at the University of Chicago and an occasional consultant to the NASA/IPAC Extragalactic Database. J.X.P. is an affiliate member of the Institute for Physics and Mathematics of the Universe. J.D.S. acknowledges support from the Carnegie Institution for Science. Support for this work was provided by NASA through Hubble Fellowship grants HST–HF–51348.001 (B.J.S.) and HST–HF–51373.001 (M.R.D.) awarded by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., for NASA, under contract NAS5–26555.

This paper includes data gathered with the 6.5 meter Magellan Telescopes located at Las Campanas Observatory, Chile. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. Figure 4A is based on observations made with the NASA/ESA Hubble Space Telescope, obtained from the Data Archive at the Space Telescope Science Institute (https://archive.stsci.edu; Program 14840), which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5–26555. These observations are associated with programs GO–14840.

The data presented in this work and the code used to perform the analysis are available at https://ziggy.ucolick.org/sss17a/ .

Materials and Methods Supplementary Text Figs. S1 to S6 Tables S1 to S2 References (36-65)

Materials and Methods

To improve the chances of detecting an EM counterpart to a GW source, we generate a prioritized list of targets, which we then attempt to combine into as few images as possible, and then produce an observing schedule to be executed by an on-site observer.

We start with a catalog of nearby galaxies specifically created for the purpose of targeted-galaxy GW searches (?). The catalog contains 53,161 galaxies, is nearly complete to 40 Mpc (in terms of stellar mass), and still has high completeness to 100 Mpc. In addition to positions and distances, the catalog includes $B$-band magnitudes, from which we calculate a $B$-band luminosity that is a proxy for some combination of total stellar mass (more massive galaxies are more luminous) and star-formation rate (SFR; bluer galaxies are more vigorously forming stars). Both are useful for EM searches of GW sources since more massive galaxies have more BNS systems and the theoretical merger rate depends on the SFR (?, ?).

Our algorithm depends on the three-dimensional localization probability maps produced by LVC. In the case of GW170817, we used the revised BAYESTAR map (?, ?). This map contains a two-dimensional (plane of the sky) probability and a distance (and distance uncertainty) for pixels in a Hierarchical Equal Area isoLatitude Pixelization (HEALPix) map. Within this map, there were 49 galaxies in the catalog and in the 90th-percentile localization volume. Since GW170817 had an associated SGRB (?, ?), which might imply a face-on system and thus a larger distance (?), we assumed a single distance for every pixel of the probability map with an inflated uncertainty, $D=43\pm 12$ Mpc.

where $D_{\rm gal}$ is the catalog distance for a galaxy and $m_{B}$ is its $B$-band magnitude.

Combining the two-dimensional location, distance, and galaxy luminosity, we obtain our final probability:

where $P_{\rm gal}$ is the probability that a particular galaxy hosts the GW event, $k$ is a normalization factor such that all probabilities sum to 1, $P_{\rm 2D}$ is the two-dimensional probability for a particular galaxy, $D_{\rm gal}$ and $\sigma_{D,{\rm~{}gal}}$ are the distance and distance uncertainty for the galaxy, and $D_{\rm LVC}$ and $\sigma_{D,{\rm~{}LVC}}$ are the distance and distance uncertainty for the GW source.

We then attempted to schedule the 1000 highest-probability galaxies for observations with the Swope telescope, the Magellan Clay telescope with the LDSS-3 imaging spectrograph (?) and the Magellan Baade telescope with the FourStar near-infrared camera (?). Using all three telescopes for the search enabled us to cover a range of optical and near-IR wavelengths over which a counterpart could be detected. Our scheduling algorithm takes an input of object positions, using the computed probabilities as priorities, and takes into account observational constraints such as the total observing time for each object (exposure plus overhead) to produce a schedule. The algorithm maximizes a merit function based on a total net priority that includes the object’s observing constraints in addition to its computed probability, and it attempts to place the highest net-priority targets at each target’s lowest airmass. For the GW170817 search, all exposure times were identical and every target was setting, which reduced the algorithm to scheduling the highest-probability unobserved targets above our airmass limit of 3 at any given time.

Using our scheduling algorithm, we were able to schedule $\sim 100$ galaxies with non-negligible probability, indicating roughly the number of observations we could perform before all high-probability galaxies set below our airmass limit. We then examined the positions of the 100 highest-probability galaxies to determine if multiple galaxies could be observed simultaneously. We visually examined the galaxy positions and were able to group 46 galaxies into 12 separate telescope pointings, improving our efficiency by a factor of 3.8 for the grouped galaxies, and 52% for the 100 highest-priority galaxies.

We added special targets corresponding to the positions of the multi-galaxy fields and assigned them a high (and equal) priority to guarantee that they would be observed. By including the multi-galaxy fields, we freed up time to observe 34 additional high-probability galaxies. After swapping galaxies for multi-galaxy fields covering those same galaxies in our target list, we recomputed an observing schedule. We did not attempt to further optimize the pointings.

Table S2 contains a list of observed galaxies, their probabilities, and the order of observation for each telescope.

While NGC 4993 was included in a multi-galaxy field, skipping this step would have delayed our imaging of NGC 4993 by only $\sim$6 minutes. That is, NGC 4993 was in the ninth image obtained, but was the twelfth-highest probability galaxy.

We applied the same algorithms to produce observing schedules for Magellan Clay/LDSS-3 and Magellan Baade/FourStar. Maps displaying the locations of galaxies observed by these telescopes are shown in Fig. S1 & S2, respectively.

S2: Initial Transient Search

In preparing to search the LIGO/Virgo localization region for potential optical counterparts, we obtained template images of our galaxy fields from the National Optical Astronomy Observatory (NOAO) public archive (?). Our preference was to obtain deep template images covering as many galaxies in our sample over as wide an area as possible and in a similar photometric band to the Swope $i$ filter. Therefore, we searched the NOAO archive for Dark Energy Camera (DECam) (?) $r$- and $i$-band images covering as large a region as many of our galaxy fields as possible.

With the DECam reference images, we immediately identified the 100 highest-probability galaxies using ds9 (?) region files. As each Swope field was observed, we reduced the images using a reduction pipeline (described below). We then loaded each file into ds9 to visually inspect the galaxies associated with each region.

Transient discovery by visual inspection is heavily biased toward bright, isolated sources, and so we used multiple kinds of scaling to inspect sources around each galaxy. This process included inspection of faint sources in the outskirts of each galaxy using typical linear scaling and sources deeply embedded in each galaxy with significantly higher clip values in order to resolve faint, nuclear sources. As we inspected each image, we blinked them rapidly with the DECam images in order to identify any differences between the images. Using this method, we were able to visually inspect a single image every $\sim$1 minute, which was appreciably faster than the total time allocated to expose on each field, read out the instrument, and slew to each successive field. However, accounting for image reductions, we visually inspected the initial images roughly $20$ minutes after they were obtained.

The image containing SSS17a was the ninth image obtained, reduced, and inspected. As we demonstrate in Figs. 3 & 4, this field contained only two galaxies and the transient source was immediately apparent upon comparison with template imaging.

After the identification of this source at 23:59 UT (Fig. S6), we queried the Minor Planet Center (?) and Transient Name Server (?) databases to confirm that SSS17a was not a known asteroid or SN. Upon confirmation that SSS17a was an unknown source, our priorities immediately changed to characterization through follow-up observations. After additional observations, we continued our search to observe the remaining galaxies in our list. The fields observed are presented in Figure S3.

S3: Astrometry

We determined the location of SSS17a from the centroid of our point-spread function (PSF) fit in the discovery $i$ band image and the world-coordinate system (WCS) solution derived for that image. The astrometric uncertainty associated with the location of SSS17a is the $1$-$\sigma$ uncertainty on the PSF centroid and on the WCS solution added in quadrature. Our total astrometric uncertainty for SSS17a, which was detected at $17.476$ mag in a relatively bright galaxy, was about 0.23\hbox{{}^{\prime\prime}}.

S4: Swope Photometry

Subsequent to discovery, we imaged the site of SSS17a in $BV\!gri$ bands with the Swope telescope from 2017 August 17-24, at which point it became too faint to detect in $30$ minute $i$ band images. After that time, we imaged the same field with long exposure times ($20$–$60$ minute) in order to construct deep templates for difference imaging.

We performed standard reductions on all of our Swope imaging using photpipe (?), a well-tested pipeline used in the Swope Supernova Survey and several major time-domain surveys (?, e.g., Pan-STARRS1). We used photpipe to correct the Swope images for bias, flat-fielding, cross-talk, and overscan. We performed astrometric calibration and corrections for geometric distortion in photpipe by resampling each image onto a corrected grid with SWarp (?), then applied a WCS derived from the locations of stars in the 2MASS Point Source Catalog (?). We used photpipe to perform difference imaging with hotpants (?, ?), which uses a spatially varying convolution kernel to match the image and template PSF before image subtraction. As an example, we display the discovery image after image subtraction in Fig. S4. Finally, we performed photometry using DoPhot, which is optimized for PSF photometry on the difference images (?). Our $BV\!gri$ photometry was calibrated using Pan-STARRS1 (PS1) standard stars (?, ?, ?, ?) observed in the same field as SSS17a and transformed into the Swope natural system using Supercal transformations (?).

The Swope photometry is presented in Fig. S5 and listed in Table S1.

S5: Extensive Transient Search

From 2017 September 4-7, we obtained follow-up $120$ second $i$-band images of all fields observed with the Swope telescope on 2017 August 17. We performed difference imaging in the same manner using photpipe to search for new optical transients that had either faded since 2017 August 17, or appeared over the 18–20 day interval from the first to second epoch of observation. Apart from SSS17a, we did not detect any transients in any of our Swope fields.

We calculated a $5$-$\sigma$ limiting magnitude for each Swope image by examining sources with background fluxes within $10\%$ of the median sky background. We then calculated the signal-to-noise ratio for each of those sources as estimated from the count rate and $1$-$\sigma$ uncertainty in the count rate. We fit a cubic spline to the signal-to-noise ratio versus magnitude and estimated the magnitude at which the signal-to-noise ratio was equal to $5.0$. This $5$-$\sigma$ limit was typically around $20.0$ mag in the $60$ second images from 2017 August 17, and $20.8$ mag in the $120$ second images from 2017 September 4-7.

Treating this $5$-$\sigma$ limit as a bolometric magnitude at the distance of NGC 4993, we would expect to detect all sources down to $M=-13.0$ to $-12.2$ mag. This limiting magnitude rules out the presence of most supernovae, even those observed within hours (?, ?, ?) or weeks after explosion (?). Very low-luminosity transients such as classical novae and intermediate luminosity red transients (?) are faint enough to be missed by our optical survey, but neither of these classes of sources are thought to be strong sources of GW emission.

Our choice of filter was also designed to target EM emission associated with predictions from kilonova models, which are expected to have very high optical opacities and thus very red colors (?). If the theoretical predictions were incorrect and the intrinsic color of EM counterparts are blue, it is still unlikely that any source would be so blue to be detected in a bluer band (e.g., $g$) and not $i$. SSS17a is well-fit by kilonova models (?, ?), but with an added component that is hot ($>10,000$ K) within hours of the LVC trigger and cooled rapidly over several days.

S6: 1M2H Slack Conversation

At the time of the GW170817 alert, D. Coulter, R. Foley, and M. Siebert were at the Dark Cosmology Centre in Copenhagen, Denmark, while C. Kilpatrick and C. Rojas-Bravo were in Santa Cruz, California. Meanwhile, B. Shappee, J. Simon, and N. Ulloa were at Las Campanas Observatory with M. Drout supporting from Pasadena, California. Both because of the multiple locations and for its speed, we used Slack to communicate. In Fig. S6 below, we present our conversation, which includes the timeline of our evolving strategy and discovery of SSS17a. All times displayed are Pacific Daylight Time.