iPTF SEARCH FOR AN OPTICAL COUNTERPART TO GRAVITATIONAL-WAVE TRANSIENT GW150914

The direct detection of gravitational waves (GWs) marks the dawn of a new era (Abbott et al. 2016b). It is widely agreed that the detection and study of the anticipated electromagnetic (EM) counterparts will vastly enrich the science returns for the field of GW astronomy. The photometric discovery of the EM counterpart will give a precise location and a spectrum of the host galaxy will give a precise redshift. This will enable a more accurate measurement of basic astrophysical properties such as the luminosity and energetics of this strong-field gravity event. If the spectrum is timely, it may also solve the long-standing mystery of the unknown sites of r-process nucleosynthesis.

The inherent challenge is that the two advanced GW interferometers, due to the low frequency of operation, give very poor on-sky localization (Abbott et al. 2013; Kasliwal & Nissanke 2014; Singer et al. 2014; Berry et al. 2015). Nevertheless, the prospect of finding EM counterparts by searching large sky areas is promising as the search methodology is steadily improving—from early efforts in the enhanced LIGO S6 run (Aasi et al. 2014), to proof-of-concept localizations of coarse Fermi gamma-ray bursts (Singer et al. 2013, 2015), to a score of EM facilities promptly responding to GW150914 (Abbott et al. 2016a).

At the time of the GW150914 trigger, there was no information disclosed on the nature of the event, i.e., whether it was a binary black hole merger or binary neutron star merger or something else (GCN 18330). Many facilities undertook a search for an EM counterpart (e.g., Connaughton et al. 2016; Evans et al. 2016; Smartt et al. 2016; Soares-Santos et al. 2016). Months later, after offline analysis, the event was identified as a binary black hole merger (GCN 18858).

Here, we present the intermediate Palomar Transient Factory (iPTF) follow-up effort. iPTF uses the Samuel Oschin 48 inch telescope on Palomar mountain equipped with the CFH12K camera with a field of view of 7.1 deg² (Law et al. 2009). Our motivation was to look for an optical counterpart powered by free neutron decay (Metzger et al. 2015), or heavy element radioactive decay (Metzger & Fernández 2014; Kasen et al. 2015). We describe the sky area coverage, candidate identification, spectroscopic classification, and panchromatic follow-up. We conclude with our plans for a way forward.

On UT 2015 September 16 03:17, the iPTF Target of Opportunity Marshal automatically responded to the gravitational-wave trigger alert G184098 (later named GW150914). It immediately notified the team via phone calls and SMS alerts that there had been a GW trigger. It also computed that due to the Sun-angle constraint and elevation constraints, iPTF would only be able to access 2.5% of the enclosed probability in the initial map by tiling 126 deg² just before sunrise at high airmass (Figure 1). This total area calculation takes into account the two non-working CCDs and the gaps between the CCDs. The small containment probability was because the southern mode of the updated ("LIB") localization was too far south to be observable from Palomar, whereas most of the northern mode rose only after 12° morning twilight. Clouds did not cooperate and the Palomar 48 inch dome remained closed the first night after trigger. However, the next night (UT September 17), we imaged 18 fields covering this area with exposures of 1 min using the R-band filter (see details in Table 3; GCN 18337). The scheduling and choice of tiles was further optimized applying the algorithm described in Rana et al. (2016). A second epoch with a baseline separation of 30 minutes (±1 minutes) was obtained for 13 fields.

Zoom In Zoom Out Reset image size

Within minutes of obtaining the data, our automated real-time image subtraction pipeline started loading candidates into our database. We have two independent, real-time pipelines—one running at the National Energy Research Scientific Computing Center (NERSC) using the HOTPANTS image subtraction algorithm (Nugent et al. 2015) and the other running at the Infrared Processing and Analysis Center (IPAC) using the PTFIDE algorithm (F. Masci et al. 2016, in preparation). Due to the dynamic nature of the optical sky, the candidate list was dominated by false-positive transients unrelated to the gravitational-wave trigger. A total of 127,676 candidates were loaded into the NERSC database and 32,576 in the IPAC database. Our automated machine-learning-aided filtering algorithms rejected the moving objects in our solar system, variable stars in the Milky Way as well as subtraction artifacts. A list of 13 candidates was presented on a dynamic web portal for human vetting.

We have been refining our software algorithms that quickly sift through the large number of candidates during our Fermi Gamma-ray Burst Monitor afterglow search effort (Singer et al. 2015). The EM–GW challenge has some similarities and some differences. The similarities are that we need to continue to reject foreground asteroids/variable stars and background supernovae/active galactic nuclei. The differences are that compared to a gamma-ray burst afterglow, the EM–GW counterpart may be relatively fainter and/or slower and/or redder. Knowing that the EM counterpart is relatively nearby due to the advanced LIGO sensitivity helps further reduce false positives.

The following are some rejection criteria:

• 1.  
  Movement in detections in two epochs separated by at least 15 minutes suggesting the candidate is an asteroid.
• 2.  
  Past history of eruption in PTF/iPTF data (baseline of six years) suggesting the candidate is an old transient.
• 3.  
  Previously known radio source or X-ray source suggesting the candidate is an active galactic nucleus.
• 4.  
  Previously known optical or infrared point source underneath the position suggesting the candidate is a stellar flare.

The following criteria lead to flags for follow-up spectroscopy, additional photometry, and/or multi-band follow-up:

• 1.  
  Host galaxy (within 100 kpc of transient) with spectroscopic redshift < 0.05 (or photometric redshift < 0.1)—this is motivated by advanced LIGO's sensitivity limit to binary neutron star mergers.
• 2.  
  Photometric evolution on hour timescale (> 0.2 mag) or day timescale (> 0.5 mag) or one-week timescale (> 1 mag)—this serves as a strong discriminant against old supernovae. We note that this flag was not applied for GW150914 as all candidates of interest were spectroscopically classified within two hours.
• 3.  
  Hostless candidates with no counterpart in deep iPTF reference co-adds—even though these are unlikely to be local, we flag these events as they are relatively rare.

To quantify the relative efficacy of each criterion, we discuss the most severe cuts in order of severity by applying each criterion independently. Of the 127,676 candidates in our NERSC pipeline, only 1007 candidates (0.8% selection) are selected as being coincident with a galaxy within 200 Mpc, hence this is the most severe cut. 5803 candidates (4.5%) are selected as passing our machine-learning cuts (we now have three generations of machine-learning algorithms; see details in Brink et al. 2013; Rebbapragada et al. 2014). 15,624 candidates (12.2%) are selected as having two detections separated by 30 minutes in the same night. 78,951 candidates (62% selection) are selected as not having an optical point source in the reference image. Similarly, in our IPAC pipeline, we had a total of 32,576 candidates. Of these, 24,699 did not match a star (75.8% selection), 5302 had two detections (16.2% selection), and 1964 passed our machine-learning cut (6.0% selection).

In practice, these criteria are not all applied simultaneously and the candidates selected for human vetting are the result of a more complex database query. For example, prior to human vetting, we do not require coincidence with a nearby galaxy and we do not require any light curve properties. For the five fields where a second epoch was not completed on the same night, we did a manual search requiring a local universe match, and found two candidates that were both rejected as known asteroids. After human vetting of 13 candidates, 5 candidates were rejected as they showed past history of variability in the PTF data. In summary, our team flagged eight candidates for further follow-up in our marshal database (see Table 1). Next, we describe the prompt follow-up that was undertaken to investigate whether any of the candidates was associated with GW150914 (GCN 18341).

Table 1.  Candidates Flagged for Follow-up

Name	RA (J2000)	DEC (J000)	Discovery Time	Mag (R-band)	Minutes to Spectrum	Classification	Redshift
iPTF15cyo	8h 19m 5618	+13 d 52' 420	2015 Sep 17 05:54:55.6	17.75 ± 0.01	71	SN Ia (SN1996X-like, +23d)	0.029
iPTF15cyp	8h 21m 4368	+16 d 12' 420	2015 Sep 17 05:56:31.6	19.48 ± 0.05	125	Nuclear	0.028
iPTF15cys	8h 11m 5559	+16 d 43' 101	2015 Sep 17 06:05:16.6	17.84 ± 0.03	46	SN Ia (SN2004eo-like, +22d)	0.05
iPTF15cym	7h 52m 3567	+16 d 45' 596	2015 Sep 17 05:46:17.1	19.88 ± 0.20	113	SN II (SN1999M-like, +5d)	0.055
iPTF15cyq	8h 10m 0086	+18 d 42' 181	2015 Sep 17 05:57:16.3	20.05 ± 0.10	39	SN II (SN2004et-like, +47d)	0.063
iPTF15cyn	7h 59m 1493	+18 d 12' 549	2015 Sep 17 05:47:20.5	20.34 ± 0.28	124	Nuclear	0.062
iPTF15cyt	7h 38m 5935	+21 d 45' 432	2015 Sep 17 06:08:09.3	19.65 ± 0.09	82	Nuclear	0.078
iPTF15cyk	7h 42m 1487	+20 d 36' 434	2015 Sep 17 05:38:38.3	20.28 ± 0.12	97	SLSN I (LSQ12dlf-like, +16d)	0.539

Download table as:  ASCIITypeset image

Since Hawaii is west of Palomar Observatory, sunrise was three hours later and we were able to obtain spectra of all eight candidates in less than two hours from discovery (Figure 2). We emphasize that iPTF has routinely been obtaining spectroscopic classification on the same night as discovery, totaling 165 transients with spectra within 12 hr, thus far. We observed with the DEep Imaging Multi-Object Spectrograph (DEIMOS; Faber et al. 2003) mounted on the Keck II telescope. We used the low-resolution 600 ZD grating, giving spectral coverage between 4650 and 9600 Å with a resolution of 3.5 Å (FWHM). Our spectra are shown in Figure 2. A priori, since we searched 126 deg² to a depth of 20.5 mag, we expect $\approx 3.2$ supernovae using the rates in Li et al. (2011; and assuming that supernovae are brighter than −17 mag for 1 month, i.e., a volume out to z = 0.075).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We cross-matched our spectra with a library of supernovae spectra augmenting the superfit software (Howell et al. 2005). Our classifications are in Table 1. We found two Type Ia supernovae (SNe Ia), two hydrogen-rich core-collapse supernovae (SNe II), three nuclear candidates (e.g., weak AGNs where the spectrum is dominated by the host galaxy), and one hostless transient with initially unclear classification (iPTF15cyk). Offline processing of the three nuclear candidates also shows past history of photometric variability in the PTF data, which is consistent with the AGN hypothesis. The spectrum of iPTF15cyk was dominated by a blue continuum, with narrow lines suggesting a redshift of 0.539 (which would imply a very luminous transient). Since the nature of the GW source was unclear, we decided to obtain additional spectroscopic and multi-wavelength follow-up.

We observed iPTF15cyk and the necessary calibrators with the Karl G. Jansky Very Large Array (VLA; Perley et al. 2009) in its D and DnC configurations. The observations were performed in the C-band ($\approx 6\;{\rm{GHz}}$ central frequency) under our Target of Opportunity program (VLA/15A-339; PI: Corsi). VLA data were reduced and imaged using the Common Astronomy Software Applications (CASA) package. In Table 2, we report the $3\sigma$ upper limits derived for iPTF15cyk using the full 2 GHz bandwidth (GCN 18914).

Table 2.  Panchromatic follow-up of Superluminous Supernova iPTF15cyk

Facility	Epoch	Frequency	Flux Limit
	UTC		erg cm−2 s−1 Hz−1
Swift/XRT	2015 Sep 18 18:12	2 keV	$\lt 4.5\times {10}^{-32}$
VLA	2015 Oct 15 11:20:32-12:05:27	5.43 GHz	<2.6 $\times {10}^{-28}$
VLA	2015 Dec 06 04:52:11-05:00:07	5.43 GHz	<2.3 $\times {10}^{-28}$
VLA	2016 Jan 20 01:55:31-02:59:40	5.43 GHz	<2.3 $\times {10}^{-28}$

Download table as:  ASCIITypeset image

Given the host galaxy redshift, iPTF15cyk could be a superluminous supernova (SLSN) since the absolute magnitude at discovery was −22 mag. Radio and X-ray emission from superluminous SNe may arise from interaction with the circumstellar medium (CSM; see, e.g., Ofek et al. 2013). In an alternate model, SLSNs could be powered by the spin-down of a nascent magnetar inside the supernova ejecta (Kasen & Bildsten 2010), which may also produce X-ray emission (Metzger et al. 2014).

However, such emission is likely to be very sensitive to the exact properties of the CSM including density profile and homogeneity. In dense CSM environments, free–free absorption can suppress the radio emission at early times. Thus, chances for a detection are maximized by observing after maximum light (Ofek et al. 2013). Hence, we observed iPTF15cyk thrice between 1 month and 4 months after discovery.

We also observed the location of iPTF15cyk with the Swift satellite (Gehrels et al. 2004) beginning at 18:12 UT on 2015 September 18 (${\rm{\Delta }}t=4.3\;{\rm{days}}$ after the GW trigger). We do not detect any emission with the on board X-Ray Telescope (XRT; Burrows et al. 2005) to a 3σ limit of $\lt 3.2\times {10}^{-3}$ ct s⁻¹. Assuming a power-law spectrum with a photon index of ${\rm{\Gamma }}=2$, this corresponds to an upper limit on the unabsorbed flux (0.3–10.0 keV) of ${f}_{X}\lt 1.3\times {10}^{-13}$ erg cm⁻² s⁻¹.

Simultaneously, we obtained images of the field with the Ultra-Violet Optical Telescope (UVOT; Roming et al. 2005) on board Swift in the V, B, U, UVW1, UVM2, and UVW2 filters. No emission is detected at the location of iPTF15cyk. For a 30 aperture we place the following magnitude limits (AB system) at this time: $V\gt 19.29;$ $B\gt 19.81;$ $U\gt 20.62;$ ${UVW}1\gt 21.61;$ ${UVM}2\gt 22.27;$ and ${UVW}2\gt 22.42$. These limits were derived using the revised UV zero points and time-dependent sensitivity from Breeveld et al. (2011).

Additional spectroscopic follow-up of iPTF15cyk showed that it was a hydrogen-poor SLSN I (Quimby et al. 2011) at z = 0.539 (Figure 4), similar to LSQ12dlf at +16 d (Nicholl et al. 2014). The radio and X-ray upper limits were consistent with this classification. Given the high redshift, we concluded that this event was unrelated to GW150914. We note that the odds of finding a superluminous supernova were lower than the odds of finding other core-collapse or thermonuclear supernovae. The snapshot rate is only ∼0.2 using the volumetric rate in Quimby et al. (2013; and assuming that SLSNs are brighter than −21 mag for 1 month). Moreover, we have a total of only six events with z > 0.5 (out of 2650 spectroscopically classified supernovae) in the six years of operating PTF/iPTF.

Zoom In Zoom Out Reset image size

The post-detection era promises to be one of routine GW detections of binary neutron star mergers. With routine detections, the joint probability of $\sim \tfrac{1}{3}$ that the Sun $\left(\sim \tfrac{2}{3}\right)$, clouds $\left(\sim \tfrac{2}{3}\right)$, and latitude $\left(\sim \tfrac{3}{4}\right)$ simultaneously cooperate to identify the optical counterpart is not discouraging. Furthermore, given the location of Palomar Observatory in Southern California, relative to the location of the advanced LIGO interferometers, the time lag to respond is inherently less than an hour as we do not need to wait for the Earth to rotate (Kasliwal & Nissanke 2014). Most of the GW150914 localization was not accessible from the northern night sky. However, based on our simulations (Singer et al. 2014), iPTF would include the true position of the GW source for an average of ≈ 1 out of 2 events assuming a total of 100 iPTF observations (see Figure 5; each observation is two 60 s exposures of 7.1 deg²).

Zoom In Zoom Out Reset image size

As advanced LIGO ramps up in GW sensitivity, we are undertaking both hardware and software upgrades to improve EM sensitivity. In 2017, we plan to commission the Zwicky Transient Facility (ZTF²² ; Kulkarni 2012; Bellm 2014), a 47 deg² camera on the Palomar 48 inch, with a 12 times higher volumetric survey speed than iPTF. This increase in survey speed enables a faster cadence and deeper search for the optical counterpart (e.g., 22 mag in 10 minutes). The larger field of view may also be more robust to a shifting localization (e.g., for GW150914, our enclosed probability went from 2.5% in the initial map to 0.2% in the final map; see Abbott et al. 2016a). We are continuing to improve our software algorithms, e.g., better candidate filtering, image co-addition, and more optimal image subtraction (Zackay et al. 2016). We are continuing to complete our census of the local universe (CLU; D. Cook et al. 2016, in preparation) as this 200 Mpc galaxy catalog serves as the most severe filter for false positives (see examples in Nissanke et al. 2013). We have recently made some hardware changes to enable observing with the I-band filter, and possibly provide some constraints on kilonova models predicting red emission (Metzger & Fernández 2014; Kasen et al. 2015).

Among the various models for EM emission from binary neutron star mergers, free neutron decay gives the most luminous optical counterpart (Figure 3; Metzger et al. 2015). Varying free neutron mass and opacity suggests that this counterpart may fade quickly, as much as 4 mag in 24 hr. Thus, we are also systematizing our follow-up with the Global Relay of Observatories Watching Transients Happen (GROWTH²³ ) program. The combination of a longitudinally distributed network of telescopes as well as multi-wavelength follow-up (VLA and Swift) should effectively filter candidates on a 24 hr timescale. Obtaining a timely light curve, spectra, and spectral energy distribution will unravel both the astrophysics and the astrochemistry of the EM counterpart. With this first gravitational-wave detection, the 21st century gold rush (Kasliwal 2013) has officially begun!

Based on observations obtained with the Samuel Oschin Telescope 48 inch and the 60 inch Telescope at the Palomar Observatory as part of the iPTF project, a scientific collaboration among the California Institute of Technology, Los Alamos National Laboratory, the University of Wisconsin-Milwaukee, the Oskar Klein Center, the Weizmann Institute of Science, the TANGO Program of the University System of Taiwan, and the Kavli Institute for the Physics and Mathematics of the universe. M.M.K., R.L., and Y.C. acknowledge support from the National Science Foundation PIRE program grant 1545949. A.A.M. acknowledges support from the Hubble Fellowship HST-HF-51325.01. P.E.N. and Y.C. acknowledge support from the DOE under grant DE-AC02-05CH11231, Analytical Modeling for Extreme-Scale Computing Environments. The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. A.C. and N.P. acknowledge support from NSF CAREER award 1455090. Part of the research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with NASA. M.M.K. thanks Brian Metzger for providing us theoretical light curves for neutron-powered precursors. We thank the referee for constructive feedback.

We simulated optimal P48 scheduling for GW events in order to quantify the tradeoff between observing time and contained probability during O1. We took all N_i = 250 simulated BNS merger sky maps, enumerated as i = 1, ..., N_i, from the "2015" scenario in Singer et al. (2014). We shifted the GMSTs of the events by random integer multiples of 1 day so that the sky maps remained unchanged in equatorial coordinates but the event times were uniformly distributed throughout the planned time span of O1, 2015 October 1 through 2016 January 1.

We divided each day into N_j = 864 time blocks labeled j = 1, ..., N_j, each 100 s long. We divided the sky into 12,288 equal-area HEALPix pixels (Górski et al. 2005), each spanning an area of 3.4 deg². (This is about half of the FOV of iPTF; the next coarser resolution would be about twice the FOV of iPTF because HEALPix pixel densities are related by powers of 4.)

Assuming ideal weather conditions, we calculated an observability mask as a function of time and sky position in the form of a ${\text{}}{N}_{j}$ × ${\text{}}{N}_{k}$ element bitmap for each one-day period following a simulated GW event. A time and position was considered observable under the following criteria:

• 1.  
  The altitude of the Sun is $\lt -12^\circ$.
• 2.  
  The hour angle is $\lt {6.5}^{{\rm{h}}}$.
• 3.  
  The airmass is $\lt 2.5$ (e.g., the altitude is $\gt 23\buildrel{\circ}\over{.} 5$).
• 4.  
  The separation from the center of the Moon is $\gt 10^\circ \times {({\rm{Moon}}{\rm{phase}}/180^\circ )}^{2}$.

For each event i, we planned an optimal one-night observing schedule by casting it as a canonical problem in graph theory: finding the maximum matching of an edge-weighted bipartite graph. The left nodes of the graph were the N_j time blocks, and the right nodes were the N_k pixels. The edge jk was present if pixel k was observable at time j. The weight of edge jk was the probability from the sky map that the GW source is contained in pixel k, _Pi(k). We found the global optimum observing plan using the LEMON library²⁴ . We ranked the resulting schedules from most to least probable, yielding the sequence $\{{P}_{i}^{\prime }(k)\}{}_{k}$.

In Figure 5, we plot the expected number of GW sources that would be successfully imaged as a function of the number of iPTF fields, if we followed up N = 1, 2, 3, or 4 GW candidates selected randomly out of the Singer et al. (2014) sample. (The number of fields plotted on the horizontal axis is actually half of the number of scheduled tiles in order to correct for the fact that the HEALPix tiles are half of the size of the iPTF FOV.)

Unsurprisingly, we find that we need to follow up at least two events to have a good chance of containing the position of one source: the bimodality of two-detector O1 sky maps means that only half of the GW localization region is observable for any one event. The improvement in contained probability reaches a point of diminishing returns at 100–150 iPTF fields per GW event. The growth of the containment probability with each added field generally flattens out by 150 fields. Assuming two 100 s exposures per field, this implies a maximum of 5.5–8.3 hr, or an entire night of iPTF P48 observations, per GW event.

In the specific case of GW150914, the Sun-angle and declination constraints severely constrained the visibility window. Our detailed observing log is presented in Table 3.