Spitzer Space Telescope Infrared Observations of the Binary Neutron Star Merger GW170817

We present Spitzer Space Telescope 3.6 and 4.5 micron observations of the binary neutron star merger GW170817 at 43, 74, and 264 days post-merger. Using the final observation as a template, we uncover a source at the position of GW170817 at 4.5 micron with a brightness of 22.9+/-0.3 AB mag at 43 days and 23.8+/-0.3 AB mag at 74 days (the uncertainty is dominated by systematics from the image subtraction); no obvious source is detected at 3.6 micron to a 3-sigma limit of>23.3 AB mag in both epochs. The measured brightness is dimmer by a factor of about 2-3 times compared to our previously published kilonova model, which is based on UV, optical, and near-IR data at<30 days. However, the observed fading rate and color (m_{3.6}-m_{4.5}>0 AB mag) are consistent with our model. We suggest that the discrepancy is likely due to a transition to the nebular phase, or a reduced thermalization efficiency at such late time. Using the Spitzer data as a guide, we briefly discuss the prospects of observing future binary neutron star mergers with Spitzer (in LIGO/Virgo Observing Run 3) and the James Webb Space Telescope (in LIGO/Virgo Observing Run 4 and beyond).

At 10 days the kilonova spectral energy distribution (SED) peaked in the NIR, with a blackbody temperature of 1300 K, and hence an expected substantial contribution into the mid-IR (Chornock et al. 2017;Nicholl et al. 2017;Kasliwal et al. 2017). Here, we present the full set of Spitzer Space Telescope IR observations of GW170817, obtained at arXiv:1805.08192v1 [astro-ph.HE] 21 May 2018 43, 74, and 264 days post-merger, which extend the kilonova observations to 3.6 and 4.5 µm (see Lau et al. 2017); we uncover clear detections at 4.5 µm. In Section 2 we present the observations and our data analysis, image subtraction, and photometry procedures. We compare the results to our kilonova models from Villar et al. (2017) in Section 3. Motivated by the results, in Section 4 we discuss the prospects for IR observations of future events with Spitzer and the James Webb Space Telescope (JWST).
All magnitudes presented in this Letter are given in the AB system and corrected for Galactic reddening with E(B −V ) = 0.105 mag (Schlafly & Finkbeiner 2011). All uncertainties are reported at the 1σ level. We assume negligible reddening contribution from the host galaxy (Blanchard et al. 2017), and a luminosity distance to NGC 4993 of 40.7 Mpc (Cantiello et al. 2018).

OBSERVATIONS AND DATA ANALYSIS
We obtained public Spitzer (Werner et al. 2004) observations of GW170817 taken on 2017 September 29, 2017 October 30, and 2018 May 8 with the InfraRed Array Camera (IRAC; Fazio et al. 2004) in the 3.6 and 4.5 µm bands during the "warm" Spitzer mission (Director's Discretionary Time Program 13202; PI: Kasliwal); see Table 1. Each visit consisted of 466 frames with exposure times of 30 sec per frame, for a total on-source time of ≈ 3.9 hours in each band. We processed the images using standard procedures in Mopex (Makovoz & Marleau 2005) to generate mosaic images. Mopex cleans the images of cosmic rays and applies appropriate distortion corrections before drizzling the images. We used a drizzling factor of 0.8 and an output pixel scale of 0.4 . We compare the native astrometry to seven 2MASS point sources in the field, and find that the astrometric solution is good to about 1 pixel. We performed image subtraction with the HOTPANTS package (Alard 2000;Becker 2015), using the 2018 May 8 observations as a template in each band. We note that at 264 days post-merger, the emission from the relativistic ejecta (which dominates in the radio and X-ray bands) has m 3.6 = 25.9 and m 4.5 = 25.7 mag Margutti et al. 2018;Xie et al. 2018), more than an order of magnitude below the expected brightness of the kilonova emission, and well below the detection level of the Spitzer data. A composite 3.6 and 4.5 µm image, and the subtracted 4.5 µm image at 43 days are shown in Figure 1. A point source is apparent in the subtracted image.
Although HOTPANTS computes and utilizes a spatiallyvariable convolution kernel, and is therefore able to match dissimilar point spread functions (PSFs), we find that the location of GW170817 is heavily contaminated by residual artifacts from the bright host galaxy. To remove the remaining contamination, we first mask the source location with a region the size of the expected PSF (≈ 5 pixels). We then smooth the masked image with a Gaussian kernel, interpolating across the masked region. We use a kernel standard deviation of one pixel (but find that the kernel width has little effect on our results). We then subtract the smoothed image from the original data to isolate the point source. The resulting final 4.5 µm images from 43 and 74 days are shown in Figure 1 and clearly reveal the presence of a point source at the location of GW170817.
We measure the brightness of the source using both fixed aperture photometry and PSF-fitting assuming a Gaussian PSF. We injected fake point sources around the host galaxy at a similar offset to that of GW170817 to quantify the systematic uncertainties of the subtraction methods and photometry. For the observations with a detected source at the location of GW170817 (4.5 µm), we specifically injected fake sources of the same measured magnitude. For each injected source, we executed the same method of smoothing and subtraction from a masked image. We used the spread in the recovered magnitudes as our overall uncertainty. For the observations without a significantly detected source (3.6 µm), we injected sources with a range of fluxes to determine 3σ upper limits. The results are summarized in Table 1. We additionally confirmed that the 4.5 µm detection is not an artifact of the subtraction process or the IRAC PSF by searching for sources at the same relative location as the GW170817 counterpart around a nearby saturated star within the field of view, following the same procedure. We did not find any significant sources around the star.
The detected source at 4.5 µm has 22.9 ± 0.3 mag at 43 days and 23.8 ± 0.3 mag at 74 days post-merger. The source is detected with a signal-to-noise ratio (SNR) of ≈ 10 at 43 days and ≈ 5 at 74 days. However, the final uncertainties are dominated by systematic effects, as determined from the spread in magnitudes for the injected fake point sources. We do not detect a source at 3.6 µm in either epoch to a 3σ limit of 23.3 mag.  We compare the observations to our three-component kilonova model, which was previously used to fit all available UVOIR photometry ); see Figure 2. Each component is characterized by a unique gray opacity roughly corresponding to its lanthanide fraction (Tanaka et al. 2018), and is independently described by a blackbody SED. The blackbody SEDs cool as a function of time until they reach a minimal "temperature floor", at which point we assume that the photosphere recedes into the ejecta, at a constant temperature. At late times ( 10 days), our three-component model predicts that the light curve is dominated by the intermediate r-process component and that this component has reached its temperature floor of ≈ 1300 K, somewhat cooler than the lowest lanthanide ionization temperature (e.g., Kasen et al. 2013).
We find that our model over-predicts the Spitzer measurements at 43 and 74 days by about a factor of ≈ 3 (1.2 mag) and ≈ 2.5 (1 mag), respectively ( Figure 2). However, the decline rate between the two measurements is in good agreement with the model prediction. Similarly, the temperature  . For comparison we also show the Ks-band (2.2 µm) data and model (gray; Cowperthwaite et al. 2017;Drout et al. 2017;Kasliwal et al. 2017;Smartt et al. 2017;Tanvir et al. 2017;Troja et al. 2017;Utsumi et al. 2017;Villar et al. 2017). implied by the flat or red color in the 3.6 and 4.5 µm bands ( 1200 K) is consistent with the temperature floor in our model. We observe a similar late-time deviation from our model in the K s -band (2.2 µm) at 20 days. Assuming a blackbody SED with T = 1200 K we find that the bolometric luminosity implied by the 4.5 µm detections is ≈ 6 × 10 38 erg s −1 and ≈ 2 × 10 38 erg s −1 at 43 and 74 days, respectively. This is consistent with the drop off in bolometric luminosity starting at ≈ 10 days, when the estimated bolometric luminosity is ≈ 2 × 10 40 erg s −1 (Cowperthwaite et al. 2017;Arcavi 2018;Waxman et al. 2017) Relaxing some of the assumptions in our model may eliminate the brightness discrepancy. For example, at the time of the Spitzer observations the kilonova is likely transitioning into the nebular phase, and the blackbody SED approximation may break down. Using the parameters of the dominant intermediate-opacity component of our model , we find that at 43 days the optical depth is τ ≈ 1, sug-gesting that the ejecta are becoming optically thin. Additionally, the shape of the late-time light curve is also dictated by the time-dependent thermalization efficiency of the merger ejecta (Barnes et al. 2016). A steeper decline of the thermalization efficiency at 20 days will better capture the lower observed fluxes in the K s and Spitzer bands. The thermalization is highly dependent on the nuclear mass models assumed (see e.g., Rosswog et al. 2017), and is uncertain by almost an order of magnitude at 1 month.
We also consider the possibility that the observed IR emission is due to reprocessing of bluer kilonova emission by newly formed dust. The warm temperature implied by our observations requires carbon-based dust, due to its high condensation temperature (T c ≈ 1800 K; Takami et al. 2014). We fit a modified blackbody to the Spitzer photometry at day 43, assuming m 3.5 ≈ m 4.5 (following Equations 1 and 2 of Gall et al. 2017). We find that the carbon dust mass required to reproduce the observed luminosity is ≈ 5 × 10 −7 M . How-ever, Gall et al. (2017) explored a range of theoretical kilonova wind models and found that at most ∼ 10 −9 M of carbon dust can be produced. We therefore conclude that the observed IR emission is not due to dust reprocessing.

IMPLICATIONS FOR IR OBSERVATIONS OF FUTURE BNS MERGERS
The Spitzer detections of GW170817 at 43 and 74 days post-merger indicate that future BNS mergers should be observed at IR wavelengths. Indeed, taking our models at face value, at least in the well-characterized regime at 20 days, it appears that the peak of the kilonova emission shifts into the NIR/MIR bands at 10 days. This suggests that significant effort should be focused on robust characterization of the IR emission of kilonovae. This will provide numerous benefits, including more accurate determination of the bolometric luminosity and therefore total r-process ejecta mass, improved measurements of the r-process opacity at long wavelengths, observational constraints on the late-time thermalization efficiency, and continued insight into BNS mergers as sites of cosmic r-process production.
Advanced LIGO/Virgo (ALV) Observing Run 3 (O3) is expected to begin in early 2019 and span a full year, with an expected BNS merger detection distance of ≈ 120 Mpc. The timing of O3 overlaps favorably with Spitzer Cycle 14 (the final Spitzer cycle), and the sensitivity should be sufficient to detect events with a similar IR luminosity to GW170817 in the first ≈ 40 days to ≈ 120 Mpc. For example, observations of about 9 hours on-source can achieve 5σ limiting magnitudes 1 of m 3.6 ≈ 25.5 mag and m 4.6 ≈ 25 mag, assuming no significant contamination from host galaxy subtraction.
Beyond O3 (2021 and later), the ALV network is expected to achieve design sensitivity, with typical BNS merger detections to ≈ 200 Mpc and a maximal detection distance of ≈ 450 Mpc (for favorably oriented and positioned BNS mergers). The timing is ideal for overlap with JWST, which will be able to provide NIR and MIR spectra. In the NIR, NIRSpec can produce low-resolution (R ≈ 100) spectra at 0.6 − 5.3 µm; this resolution is sufficient for kilonovae given the typical velocities of ∼ 0.1 − 0.3c. In particular, spectra with SNR 50 can be obtained near peak for a GW170817like kilonova to 450 Mpc in just 1 hour of on-source time. At later times, BNS mergers could be tracked to ≈ 40 days at ≈ 200 Mpc with SNR ≈ 10 in about 6 hours of on-source time.
We do not yet know the full range of brightnesses and SEDs of kilonovae, as well as the potential contribution of dust reprocessing, but the discussion above illustrates that NIR/MIR characterization of kilonovae can be achieved with Spitzer in ALV O3 and with JWST when ALV reaches design sensitivity. This can be achieved with a modest time investment, but will require target-of-opportunity response to BNS mergers.

CONCLUSIONS
We present Spitzer IR observations of the kilonova associated with GW170817 spanning to 264 days post-merger. We detect the kilonova at 4.5 µm at 43 and 74 days post-merger with a brightness of ≈ 22.9 and ≈ 23.8 mag, respectively. We do not identify a confident detection at 3.6 µm, to a 3σ upper limit of 23.3 mag. The inferred color of the kilonova indicates that the ejecta has cooled to 1200 K at these late times. These magnitudes are fainter than an extrapolation of our model to the UVOIR data at 30 days, highlighting the need for improved models at late times (for example, the details of the ejecta thermalization). Finally, we show that future BNS mergers with kilonovae similar to GW170817 will be detectable with Spitzer to 120 Mpc at 40 days postmerger, and will be accessible to NIR and MIR spectroscopy with JWST to ≈ 450 Mpc at peak and to ≈ 100 − 200 Mpc at 40 days post-merger (and to later times with JWST imaging).
The Berger Time-Domain Group at Harvard is supported in part by the NSF through grant AST-1714498, and by NASA through grants NNX15AE50G and NNX16AC22G. This work is based in part on observations made with the Spitzer Space Telescope, which is operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with NASA.