Abstract
We present detailed submillimeter- through centimeter-wave observations of the extraordinary extragalactic transient AT2018cow. The apparent characteristics—the high radio luminosity, the rise and long-lived emission plateau at millimeter bands, and the sub-relativistic velocity—have no precedent. A basic interpretation of the data suggests coupled to a fast but sub-relativistic () shock in a dense () medium. We find that the X-ray emission is not naturally explained by an extension of the radio-submm synchrotron spectrum, nor by inverse Compton scattering of the dominant blackbody UV/optical/IR photons by energetic electrons within the forward shock. By , the X-ray emission shows spectral softening and erratic inter-day variability. Taken together, we are led to invoke an additional source of X-ray emission: the central engine of the event. Regardless of the nature of this central engine, this source heralds a new class of energetic transients shocking a dense medium, which at early times are most readily observed at millimeter wavelengths.
Export citation and abstract BibTeX RIS
1. Introduction
1.1. The Transient Millimeter Sky
Although the sky is regularly monitored across many bands of the electromagnetic spectrum (as well as in gravitational waves and energetic particles) the dynamic sky at millimeter to submillimeter wavelengths (0.1–10 mm) remains poorly explored. There has only been one blind transient survey specific to the millimeter band11 (Whitehorn et al. 2016); millimeter facilities are usually only triggered after an initial discovery at another wavelength. Even when targeting known transients, the success rate for detection is low, and only a few extragalactic transients12 have well-sampled, multifrequency light curves to date. This sample includes supernovae (SNe; Weiler et al. 2007; Horesh et al. 2013), tidal-disruption events (TDEs; Zauderer et al. 2011; Yuan et al. 2016), and gamma-ray bursts (GRBs; de Ugarte Postigo et al. 2012; Laskar et al. 2013; Perley et al. 2014; Urata et al. 2014).
The paucity of millimeter transient studies can be attributed in part to costly receiver and electronics systems and the need for excellent weather conditions, but it also reflects challenges intrinsic to millimeter-wave transients themselves: most known classes are either too dim (SNe, most TDEs) to detect unless they are very nearby, or too short-lived (GRBs) to detect without very rapid reaction times (<1 day, and even in these circumstances the emission may only be apparent from low-density environments; Laskar et al. 2013).
An evolving technical landscape, together with rapid follow-up enabled by high-cadence optical surveys, present new opportunities for millimeter transient astronomy. Lower-noise receivers and ultra-wide bandwidth capability have greatly increased the sensitivity of submm facilities (e.g., the Submillimeter Array or SMA; Ho et al. 2004), and the Atacama Large Millimeter Array (ALMA), a flagship facility, recently began operations. Optical surveys are discovering new and unexpected classes of transient events whose millimeter properties are unknown—and possibly different from previously known types—motivating renewed follow-up efforts.
1.2. AT2018cow
AT2018cow was discovered on 2018 June 16 UT as an optical transient (Prentice et al. 2018; Smartt et al. 2018) by the Asteroid Terrestrial-impact Last Alert System (ATLAS; Tonry et al. 2018). It attracted immediate attention because of its fast rise time ( days), which was established by earlier non-detections (Fremling 2018; Prentice et al. 2018), together with its high optical luminosity () and its close proximity ( Mpc).
UV/optical/IR (UVOIR) observations (Perley et al. 2018; Prentice et al. 2018) revealed unprecedented photometric and spectroscopic properties. Long-lived luminous X-ray emission was detected with the Neil Gehrels Swift Observatory/X-ray Telescope (Swift/XRT; Rivera Sandoval & Maccarone 2018), International Gamma-Ray Astrophysics Laboratory (INTEGRAL) (Ferrigno et al. 2018; Savchenko et al. 2018), and the Nuclear Spectroscopic Telescope Array (NuSTAR) (Grefenstette et al. 2018; Margutti et al. 2018a). Early radio and submillimeter detections were reported by Northern Extended Millimeter Array (NOEMA) (de Ugarte Postigo et al. 2018), JCMT (Smith et al. 2018), Arcminute Microkelvin Imager (AMI) (Bright et al. 2018), and by us using the Australia Telescope Compact Array (ATCA) (Dobie et al. 2018a, 2018b). The source does not appear to be a GRB, as no prompt high-energy emission was detected in searches of Swift/Burst Alert Telescope (BAT) (Lien et al. 2018), Fermi/GBM (Dal Canton et al. 2018), Fermi/Large Area Telescope (LAT) (Kocevski & Cheung 2018), and AstroSat Cadmium Zinc Telluride Imager (CZTI) (Sharma et al. 2018).
Perley et al. (2018) suggested that AT2018cow is a new member of the class of rapidly rising ( days) and luminous () blue transients, which have typically been found in archival searches of optical surveys (Drout et al. 2014; Pursiainen et al. 2018; Rest et al. 2018). The leading hypothesis for this class was circumstellar interaction of a supernova (SN; Ofek et al. 2010), but this was difficult to test because most of the events were located at cosmological distances, and not discovered in real time. AT2018cow presented the first opportunity to study a member of this class up close and in real time, but its origin remains mysterious despite the intense ensuing observational campaign. Possibilities include failed SNe and TDEs, but although AT2018cow shares properties with both of these classes, it is clearly not a typical member of either (Kuin et al. 2018; Perley et al. 2018; Prentice et al. 2018).
Given the unusual nature of the source, we were motivated to undertake high-frequency observations. We began an extensive monitoring campaign with the SMA at 230 and 340 GHz and carried out supporting observations with the ATCA from 5–34 GHz. To our surprise AT2018cow was very bright and still rising at submillimeter wavelengths (and optically thick in the centimeter band) days after the discovery. Our SMA observations represent the first millimeter observation of a transient in its rise phase.
This finding led us to seek Director's Discretionary Time (DDT) with ALMA at even higher frequencies, which enabled us to resolve the peak of the spectral energy distribution (SED). A technical highlight of the ALMA observations was the detection of the source at nearly a terahertz frequency (Band 9). We present the submillimeter, radio, and X-ray observations in Section 2, and our modeling of the radio-emitting ejecta in Section 3. In Section 4, we put our velocity and energy measurements in the context of other transients (Section 4.1), attribute the high submm luminosity of AT2018cow to the large density of the surrounding medium (Section 4.2), and discuss some problems with the synchrotron model parameters (Section 4.3). In Section 5, we attribute the late-time X-ray emission to a powerful central engine. We look ahead to the future in Section 6.
2. Observations
All observations are measured (observer frame) from the zero-point MJD 58285 (following Perley et al. 2018), which lies between the date of discovery (MJD 58285.441) and the last non-detection (58284.13; Prentice et al. 2018). The full set of flux-density measurements at radio and millimeter wavelengths is presented in Table 1. At we find excellent agreement between the SMA and the ALMA data, showing that the flux scales are consistent.
2.1. Radio and Submillimeter Observations
2.1.1. The SMA
AT2018cow was regularly observed with the SMA under its DDT/Target of opportunity program. Observations took place over the period of UT 2018 June 21–UT 2018 August 3 (–49 days) in the Compact configuration, with an additional epoch on UT 2018 August 31 ( days). All observations contained 6–8 antennas and covered a range of baseline lengths from 16.4–77 m. A majority of these observations were short and were repeated almost nightly by sharing tuning and calibration data with other science tracks. The SMA has two receiver sets each with 8 GHz of bandwidth in each of two sidebands (32 GHz total), covering a range of frequencies from 188–416 GHz. Each receiver can be tuned independently to provide dual-band observations. Additionally, the upper and lower sidebands are separated (center to center) by 16 GHz, allowing up to four simultaneous frequency measurements. During some observations, the receivers were tuned to the same local oscillator frequency, allowing the lower and upper sidebands to be averaged together, improving the signal-to-noise ratio. For all observations, the quasars 1635+381 and 3C 345 were used as primary phase and amplitude gain calibrators, respectively, with absolute flux calibration performed by nightly comparison to Titan, Neptune, or (maser-free) continuum observations of the emission-line star MWC349a. The quasar 3C 279 and/or the blazar 3C 454.3 was used for bandpass calibration. Data were calibrated in IDL using the MIR package. Additional analysis and imaging were performed using the MIRIAD package. Given that the target was a point source, fluxes were derived directly from the calibrated visibilities, but the results agree well with flux estimates derived from the CLEANed images when the data quality and UV coverage was adequate.
2.1.2. The ATCA
We obtained six epochs of centimeter-wavelength observations with the ATCA (Frater et al. 1992). During the first three epochs, the six 22 m dishes were arranged in an east–west 1.5A configuration, with baselines ranging from 153–4469 m. During the latter three epochs, five of the six dishes were moved to a compact H7513 configuration, occupying a cardinally oriented "T" with baselines ranging from 31–89 m. Full-Stokes data were recorded with the Compact Array Broadband Backend (Wilson et al. 2011) in a standard continuum CABB Filter Bank (CFB) 1M setup, simultaneously providing two 2.048 GHz bands each with 2048 channels. Observations were obtained with center frequencies of 5.5 and 9 GHz, 16.7 and 21.2 GHz, and 33 and 35 GHz, with data in the latter two bands typically being averaged to form a band centered at 34 GHz. The flux-density scale was set using observations of the ATCA flux standard PKS 1934−638. For observations below 33 GHz, PKS 1934−638 was also used to calibrate the complex time-independent bandpasses, and regular observations of the compact quasar PKS 1607+268 were used to calibrate the time-variable complex gains. For the higher-frequency observations, a brighter source (3C 279) was used for bandpass calibration (except for epochs 1 and 4), and the compact quasar 4C 10.45 was used for gain calibration. In the H75 configuration, we only report results from observations at 34 GHz, from baselines not subject to antenna shadowing. For all 34 GHz observations, data obtained with the sixth antenna located 4500 m from the center of the array were discarded because of the difficulty of tracking the differential atmospheric phase over the long baselines to this antenna. The weather was good for all observations, with negligible wind and <500 μm of rms atmospheric path-length variations (Middelberg et al. 2006).
The data were reduced and calibrated using standard techniques implemented in the MIRIAD software (Sault et al. 1995). To search for unresolved emission at the position of AT2018cow, we made multifrequency synthesis images with uniform weighting. Single rounds of self-calibration over 5–10 minute intervals were found to improve the image quality in all bands. For data at 5.5 and 9 GHz, point-source models of all strong unresolved field sources were used for self-calibration. For data at the higher frequencies, self-calibration was performed using a point-source model for AT2018cow itself, as no other sources were detected within the primary beams, and AT2018cow was detected with a sufficient signal-to-noise ratio. We report flux densities derived by fitting point-source models to the final images using the MIRIAD task imfit.
2.1.3. ALMA
AT2018cow was observed with ALMA as part of DDT during Cycle 5 using Bands 3, 4, 7, 8, and 9. Observations were performed on 2018 June 30 ( Bands 7 and 8), 2018 July 08 ( Bands 3 and 4), and on 2018 July 10 ( Band 9).14
The ALMA 12 m antenna array was in its most compact C43-1 configuration, with 46–48 working antennas and baselines ranging from 12–312 m. The on-source integration time was 6–8 minutes for Bands 3–8, and 40 minutes for Band 9. The Band 3–8 observations used two-sideband (2SB) receivers with 4 GHz bandwidth each centered on 91.5 and 103.5 GHz (Band 3), 138 and 150 GHz (Band 4), 337.5 and 349.5 GHz (Band 7), and 399 and 411 GHz (Band 8). The Band 9 observations used double-sideband (DSB) receivers with 8 GHz bandwidth (2 times larger than that for the Band 3–8 observation, by using 90° Walsh phase switching) centered on 663 and 679 GHz. All calibration and imaging was done with the Common Astronomical Software Applications (CASA; McMullin et al. 2007). The data in Bands 3–8 were calibrated with the standard ALMA pipeline, using J1540+1447, J1606+1814, or J1619+2247 to calibrate the complex gains, and using J1337−1257 (Band 7), J1550+0527 (Band 3/4), or J1517−2422 (Band 8) to calibrate the bandpass response and apply an absolute flux scale. Band 9 observations were delivered following manual calibration by the North American ALMA Science Center, using J1540+1447 for gain calibration, and J1517−2422 for bandpass- and flux calibration. We subsequently applied a phase-only self-calibration using the target source (for Bands 3–8), performed a deconvolution, imaged the data, and flux-corrected for the response of the primary beam. AT2018cow is unresolved in our ALMA data, with a synthesized beam that ranges from () in Band 3 to () in Band 9. The signal-to-noise ratio in the resulting images ranges from ∼500 in Bands 3 and 4 to ∼80 in Band 9. Details about the ALMA Band 9 data reduction can be found in Appendix A.
2.2. X-Ray Observations
2.2.1. Swift/XRT
Swift (Gehrels et al. 2004) has been monitoring AT2018cow since June 19, with both the Ultraviolet-Optical Telescope (Roming et al. 2005) and the XRT (Burrows et al. 2005). The transient was well detected in both instruments (e.g., Rivera Sandoval et al. 2018).
We downloaded the Swift/XRT data products (light curves and spectra) using the web-based tools developed by the Swift-XRT team (Evans et al. 2009). We used the default values, but binned the data by observation. To convert from count rate to flux, we used the absorbed count-to-flux rate set by the spectrum on the same tool, . This assumes a photon index of and a Galactic NH column of .
Table 1. Flux-density Measurements for AT2018cow
(days) | Facility | Frequency (GHz) | Flux Density (mJy) |
---|---|---|---|
5.39 | SMA | 215.5 | 15.14 ± 0.56 |
5.39 | SMA | 231.5 | 16.19 ± 0.65 |
6.31 | SMA | 215.5 | 31.17 ± 0.87 |
6.31 | SMA | 231.5 | 31.36 ± 0.97 |
7.37 | SMA | 215.5 | 40.19 ± 0.56 |
7.37 | SMA | 231.5 | 41.92 ± 0.66 |
7.41 | SMA | 330.8 | 36.39 ± 2.25 |
7.41 | SMA | 346.8 | 30.7 ± 1.99 |
8.37 | SMA | 215.5 | 41.19 ± 0.47 |
8.37 | SMA | 231.5 | 41.44 ± 0.56 |
8.38 | SMA | 344.8 | 26.74 ± 1.42 |
8.38 | SMA | 360.8 | 22.79 ± 1.63 |
9.26 | SMA | 243.3 | 35.21 ± 0.75 |
9.26 | SMA | 259.3 | 36.1 ± 1.0 |
9.28 | SMA | 341.5 | 22.85 ± 1.53 |
9.28 | SMA | 357.5 | 25.84 ± 2.5 |
10.26 | SMA | 243.3 | 36.6 ± 0.81 |
10.26 | SMA | 259.3 | 31.21 ± 0.92 |
10.26 | SMA | 341.5 | 19.49 ± 1.47 |
10.26 | SMA | 357.5 | 17.42 ± 2.8 |
11.26 | SMA | 243.3 | 22.14 ± 1.05 |
11.26 | SMA | 259.3 | 20.02 ± 1.28 |
13.3 | SMA | 215.5 | 35.67 ± 0.81 |
13.3 | SMA | 231.5 | 32.94 ± 1.01 |
14.36 | SMA | 344.8 | 26.85 ± 2.22 |
14.36 | SMA | 360.8 | 26.13 ± 2.77 |
14.37 | SMA | 215.5 | 42.05 ± 0.5 |
14.37 | SMA | 231.5 | 38.71 ± 0.58 |
15.23 | SMA | 225.0 | 30.82 ± 2.41 |
15.23 | SMA | 233.0 | 28.64 ± 4.0 |
15.23 | SMA | 241.0 | 27.41 ± 3.21 |
15.23 | SMA | 249.0 | 15.4 ± 4.74 |
17.29 | SMA | 234.6 | 36.57 ± 1.55 |
17.29 | SMA | 250.6 | 34.04 ± 1.81 |
18.4 | SMA | 217.5 | 52.52 ± 0.55 |
18.4 | SMA | 233.5 | 49.32 ± 0.65 |
19.25 | SMA | 193.5 | 59.27 ± 1.49 |
19.25 | SMA | 202.0 | 56.03 ± 1.5 |
19.25 | SMA | 209.5 | 55.09 ± 1.39 |
19.25 | SMA | 218.0 | 54.54 ± 1.33 |
20.28 | SMA | 215.5 | 50.6 ± 1.69 |
20.28 | SMA | 231.5 | 49.16 ± 1.84 |
20.28 | SMA | 267.0 | 41.69 ± 1.62 |
20.28 | SMA | 283.0 | 37.84 ± 1.63 |
24.39 | SMA | 215.5 | 55.57 ± 0.53 |
24.39 | SMA | 231.5 | 53.2 ± 0.6 |
24.4 | SMA | 333.0 | 23.98 ± 1.39 |
24.4 | SMA | 349.0 | 28.46 ± 1.37 |
26.26 | SMA | 215.6 | 38.83 ± 1.2 |
26.26 | SMA | 231.6 | 34.1 ± 1.33 |
31.2 | SMA | 230.6 | 36.76 ± 1.12a |
31.2 | SMA | 246.6 | 31.41 ± 1.42a |
35.34 | SMA | 215.5 | 21.59 ± 0.89 |
35.34 | SMA | 231.5 | 20.63 ± 1.04 |
36.34 | SMA | 215.5 | 24.32 ± 1.19 |
36.34 | SMA | 231.5 | 20.79 ± 1.42 |
39.25 | SMA | 217.0 | 18.34 ± 1.65 |
39.25 | SMA | 233.0 | 19.74 ± 1.76 |
39.26 | SMA | 264.0 | 17.61 ± 2.79 |
39.26 | SMA | 280.0 | 8.27 ± 2.93 |
41.24 | SMA | 217.0 | 12.58 ± 1.5 |
41.24 | SMA | 225.0 | 8.91 ± 1.9 |
41.24 | SMA | 233.0 | 15.08 ± 1.73 |
41.24 | SMA | 241.0 | 9.64 ± 2.13 |
44.24 | SMA | 230.6 | 9.42 ± 1.61 |
44.24 | SMA | 234.6 | 8.04 ± 2.51 |
44.24 | SMA | 246.3 | 10.43 ± 2.13 |
44.24 | SMA | 250.6 | 10.06 ± 3.24 |
45.23 | SMA | 217.0 | 8.28 ± 2.24 |
45.23 | SMA | 233.0 | 10.55 ± 2.39 |
45.23 | SMA | 264.0 | 8.35 ± 3.27 |
45.23 | SMA | 280.0 | 5.7 ± 3.49 |
47.24 | SMA | 230.6 | 11.47 ± 2.81 |
47.24 | SMA | 234.6 | 10.81 ± 4.39 |
47.24 | SMA | 246.6 | 11.65 ± 3.76 |
47.24 | SMA | 250.6 | 5.6 ± 5.37 |
48.31 | SMA | 217.5 | 7.63 ± 1.11 |
48.31 | SMA | 233.5 | 5.73 ± 1.32 |
76.27 | SMA | 215.5 | 1.33 ± 0.55 |
76.27 | SMA | 231.5 | 0.61 ± 0.63 |
76.27 | SMA | 335.0 | −2.27 ± 1.87 |
76.27 | SMA | 351.0 | −0.32 ± 1.76 |
10.48 | ATCA | 5.5 | <0.15 |
10.48 | ATCA | 9.0 | 0.27 ± 0.06 |
10.48 | ATCA | 34.0 | 5.6 ± 0.16 |
13.47 | ATCA | 5.5 | 0.22 ± 0.05 |
13.47 | ATCA | 9.0 | 0.52 ± 0.04 |
13.47 | ATCA | 16.7 | 1.5 ± 0.1 |
13.47 | ATCA | 21.2 | 2.3 ± 0.3 |
13.47 | ATCA | 34.0 | 7.6 ± 0.5 |
17.47 | ATCA | 5.5 | 0.41 ± 0.04 |
17.47 | ATCA | 9.0 | 0.99 ± 0.03 |
19.615 | ATCA | 34.0 | 14.26 ± 0.21 |
28.44 | ATCA | 34.0 | 30.59 ± 0.2 |
34.43 | ATCA | 34.0 | 42.68 ± 0.19 |
81.37 | ATCA | 34.0 | 6.97 ± 0.09 |
14.03 | ALMA | 336.5 | 29.4 ± 2.94 |
14.03 | ALMA | 338.5 | 29.1 ± 2.91 |
14.03 | ALMA | 348.5 | 28.49 ± 2.85 |
14.03 | ALMA | 350.5 | 28.29 ± 2.83 |
14.14 | ALMA | 398.0 | 26.46 ± 2.65 |
14.14 | ALMA | 400.0 | 26.21 ± 2.62 |
14.14 | ALMA | 410.0 | 25.69 ± 2.57 |
14.14 | ALMA | 412.0 | 25.95 ± 2.6 |
22.02 | ALMA | 90.5 | 91.18 ± 4.6 |
22.02 | ALMA | 92.5 | 92.31 ± 4.6 |
22.02 | ALMA | 102.5 | 93.97 ± 4.7 |
22.02 | ALMA | 104.5 | 93.57 ± 4.7 |
22.04 | ALMA | 138.0 | 85.1 ± 4.3 |
22.04 | ALMA | 140.0 | 84.58 ± 4.2 |
22.04 | ALMA | 150.0 | 80.62 ± 4.0 |
22.04 | ALMA | 152.0 | 79.71 ± 4.0 |
23.06 | ALMA | 671.0 | 31.5 ± 6.3 |
Notes. Time of detection used is mean UT of observation. SMA measurements have formal uncertainties shown, which are appropriate for in-band measurements on a given night. However, for night-to-night comparisons, true errors are dominated by systematics and are roughly 10%–15% unless indicated otherwise. ALMA measurements have roughly 5% uncertainties in Bands 3 and 4, 10% uncertainties in Bands 7 and 8, and a 20% uncertainty in Band 9. ATCA measurements have formal errors listed, but also have systematic uncertainties of roughly 10%.
aSystematic uncertainty 20% due to uncertain flux calibration.2.2.2. NuSTAR
NuSTAR (Harrison et al. 2013) comprises two co-aligned telescopes, Focal Plane Module A (FPMA) and FPMB. Each is sensitive to X-rays in the 3–79 keV range, with slightly different response functions. NuSTAR observed AT2018cow on four epochs, and a log of these observations as well as the best-fit spectral model parameters is presented in Table 2.
Table 2. NuSTAR Flux Measurements for AT2018cow and the Spectral Model Parameters
Epoch | OBSID | Exp. Time (ks) | (days) | Flux () | Photon Index | /DOF | |||
---|---|---|---|---|---|---|---|---|---|
3–10 keV | 10–20 keV | 20–40 keV | 40–80 keV | ||||||
1 | 90401327002a | 32.4 | 7.9 | 4.94 ± 0.04 | 4.41 ± 0.10 | 12.21 ± 0.39 | 21.46 ± 4.29 | ⋯ | 421/443 |
2 | 90401327004 | 30.0 | 16.8 | 5.21 ± 0.04 | 4.99 ± 0.10 | 7.70 ± 0.33 | 12.80 ± 4.79 | 1.39 ± 0.02 | 424/412 |
3 | 90401327006 | 31.2 | 28.5 | 1.58 ± 0.03 | 1.45 ± 0.06 | 1.74 ± 0.21 | ⋯ | 1.51 ± 0.04 | 174/169 |
4 | 90401327008 | 33.0 | 36.8 | 1.10 ± 0.02 | 0.92 ± 0.05 | 1.02 ± 0.20 | ⋯ | 1.59 ± 0.05 | 134/135 |
Notes. Fluxes were measured with a model-independent method.
aOBSID 90401327002 is best described by a bkn2pow model with parameters , , , , . All reported values are for this model.Download table as: ASCIITypeset image
NuSTAR data were extracted using nustardas_06Jul17_v1 from HEASOFT 6.24. Source photons were extracted from a circle of 60'' radius, visually centered on the object. We note that such a large region, appropriate for NuSTAR data, includes the transient as well as the host galaxy. Background photons were extracted from a non-overlapping circular region with 120'' radius on the same chip. Spectra were grouped to 20 source photons per bin, ignoring energies below 3 keV and above 80 keV.
Spectra were analyzed in XSPEC (v12.10.0c), using NuSTAR CALDB files dated 2018 August 14. Rivera Sandoval et al. (2018) report a low absorbing column density (N), hence we ignore this component in fitting. We opt for a simple phenomenological model to describe the spectrum. We do not fit for a cross-normalization constant between NuSTAR FPMA and FPMB. Epoch 1 (OBSID 90401327002) spectra are not consistent with a simple power law or a broken power law, hence we fit it with the bkn2pow model (obtaining spectral breaks at 9.0 ± 0.3 keV and ). Spectra of the remaining three epochs are well fit by a simple, unabsorbed power law.
We calculate the flux directly from energies of individual source and background photons detected, converted into flux using the Ancillary Response Files generated by the NuSTAR pipeline. We use a bootstrap method to estimate the error bars: we draw photons from the data with replacement, and calculate the source flux from this random sample. By repeating this process 10,000 times for each OBSID and each energy range, we calculate the 1σ error bars on the fluxes. This method gives answers consistent with xspec flux and cflux measurements for bright sources (see for instance Kaspi et al. 2014), but has the advantage of giving flux measurements without the need to assume a spectral model for the source. We find that the source is not well detected in the 40–80 keV band at the third and fourth epochs.
3. Basic Properties of the Shock
3.1. Light Curve
The radio and X-ray light curves are shown in Figure 1. The 230 GHz light curve rises (the first observation of a millimeter transient in its rise phase) and then shows significant variability, presumably from inhomogeneities in the surrounding medium. We have tentative evidence that the rise is at least in part due to a decreasing peak frequency: at –6 days, the flux is marginally higher at 231.5 GHz than at 215.5 GHz, and at –8 days, it seems that the peak may have been within the SMA observing bands. However, the position of the peak is ill-constrained; future early observations would benefit from observations at more frequencies.
By , the radio flux has diminished both due to the peak frequency shifting to lower frequencies, and to a decay in the peak flux. Specifically, the peak of the 15 GHz light curve is 19 mJy around 47 days (A. Horesh 2018, personal communication), substantially less luminous than the peak of the 230 GHz or the 34 GHz light curve. As we discuss in Section 4.2, this diminishing peak flux suggests that the interaction itself is diminishing, and enables us to constrain the size of the "circum-bubble" of material.
The X-ray light curve seems to have two distinct phases. We call the first phase ( days) the plateau phase because the X-ray emission is relatively flat. The second phase, which we call the decline phase, begins around days. During this period, the X-ray emission exhibits an overall steep decline, but also exhibits strong variation (by factors of up to 10) on shorter timescales (see also Kuin et al. 2018; Perley et al. 2018; Rivera Sandoval et al. 2018).
We use the shortest timescale of variability in the 230 GHz light curve to infer the size of the radio-emitting region, and do the same for the X-ray emission in Section 5. On Days 5–6, the 230 GHz flux changed by order unity in one day, setting a length scale for the source size of (170 au). We find no evidence for shorter-timescale variability in our long SMA tracks from the first few days of observations (Figure 2).
Download figure:
Standard image High-resolution imageTogether with the 230 GHz flux density () and the distance () we infer an angular size of and a brightness temperature of
where . This brightness temperature is close to the typical rest-frame equipartition brightness temperatures of the most compact radio sources, K (Readhead 1994).
3.2. Modeling the Radio to Submillimeter SED
The shape of the radio to submillimeter SED (Figure 3), together with the high brightness temperature implied by the luminosity and variability timescale (Section 3.1), can only be explained by nonthermal emission (Readhead 1994). The observed spectrum is assumed to arise from a population of electrons with a power-law number distribution in Lorentz factor , with some minimum Lorentz factor and electron energy power index p:
Download figure:
Standard image High-resolution imageAs argued below, we expect an adiabatic strong shock moving into a weakly magnetized, ionized medium at a nonrelativistic speed. First-order Fermi acceleration gives , where r is the compression ratio of the shock. A strong matter-dominated shock has r = 4, hence p = 2 (Blandford & Eichler 1987). However the back-reaction of the accelerated particles decelerates the gas flow, weakening the gas dynamic subshock and reducing the compression ratio from the strong shock r = 3, so typical are obtained in both simulations and astrophysical data (Jones & Ellison 1991). Quasi-perpendicular magnetized and relativistic shocks are more subtle, since some particles cannot return along field lines after their first shock crossing, but the limiting value is (Pelletier et al. 2017).
Equation (3) provides an expression for . Behind the shock (velocity v) some fraction of the total energy density goes into accelerating electrons. Conserving shock energy flux gives
The value of is large for relativistic shocks, e.g., in GRBs. But we will see that for this source (, ), the bulk of the electrons are just mildly relativistic (). For ordinary SN shocks is always nonrelativistic (). Thus, in the parameter estimations below, we follow SN convention and assume that the relativistic electrons follow a power-law distribution down to a fixed (Chevalier 1982, 1998; Kulkarni et al. 1998; Frail et al. 2000; Soderberg et al. 2005). We apply only to this relativistic power law, not to the nonrelativistic thermal distribution of shock-heated particles at lower energy.
We now describe each of the break frequencies that characterize the observed spectrum. First, the characteristic synchrotron frequency emitted by the minimum energy electrons is:
where is the gyrofrequency,
and qe is the unit charge, B is the magnetic field strength, me is the electron mass, and c is the speed of light.
Next, there is the cooling frequency , the frequency below which electrons have lost the equivalent of their total energies to radiation via cooling. In general, the timescale for synchrotron cooling depends on the Lorentz factor as . Thus, electrons radiating at higher frequencies cool more quickly. Separately, electrons could also lose energy by Compton upscattering of ambient (low energy) photons—the so-called Inverse Compton (IC) scattering. In Section 5, we find that IC scattering dominates at early times and that synchrotron losses dominate at later times, and that the transition is at .
At , electrons with cool principally by synchrotron radiation to in a time t, where
For , Compton cooling on the UVOIR flux exceeds the synchrotron cooling rate by a factor , and is correspondingly lowered. The cooled electrons emit around the characteristic synchrotron frequency
Next, the self-absorption frequency is the frequency at which the optical depth to synchrotron self-absorption is unity. The rise at 34 GHz obeys a power law (as shown in Figure 1), consistent with the optically thick spectral index we measure (Figure 3). This indicates that the self-absorption frequency is above the ATCA bands (). Figure 3 also shows that the emission in the SMA bands is optically thin at , constraining the self-absorption frequency to be .
On Day 22, we resolve the peak of the SED with our ALMA data. We denote the peak frequency and the flux at the peak frequency Fp, and find and . Motivated by the observation of optically thick emission at , we assume that , and adopt the framework in Chevalier (1998) (hereafter referred to as C98) to estimate properties of the shock at this epoch. These properties are summarized in Table 3, and outlined in detail below.
Table 3 . Quantities Derived from Day 22 Measurements, Using Different Equipartition Assumptions
Parameter | ||
---|---|---|
(GHz) | 100 | 100 |
(mJy) | 94 | 94 |
r (1015 cm) | 7 | 6 |
v/c | 0.13 | 0.11 |
B (G) | 6 | 4 |
erg) | 4 | 35 |
ne ( cm−3) | 3 | 41 |
(GHz) | 2 | 8 |
Note. In the text unless otherwise stated we use .
Download table as: ASCIITypeset image
Following Equations (11) and (12) in C98, the outer shock radius Rp can be estimated as
and the magnetic field can be estimated as
where, as in Equation (2), p is the electron energy index. Note that C98 use γ for the electron energy power index. We use p instead and γ for the Lorentz factor. The constant in cgs units, and the constants c5 and c6 are tabulated as a function of p on page 232 of Pacholczyk (1970). D is the distance to the source, MeV is the electron rest mass energy, and is the ratio of energy density in electrons to energy density in magnetic fields (in C98 this ratio is parameterized as α, but we use α as the optically thin spectral index of the radio SED.) Finally, f is the filling factor: the emitting region is approximated as a planar region with thickness s and area in the sky , and thus a volume , which can be characterized as a spherical emitting volume .
On Day 22, we measure where , which corresponds to p = 3.2. Later in this section we show that our submillimeter observations lie above the cooling frequency, and therefore that the index of the source function of electrons is ps = 2.2. However, the C98 prescription considers a distribution as it exists when the electrons are observed, from a combination of the initial acceleration and the energy losses (to cooling). So, we proceed with p = 3.2, and discuss this unusual regime in Section 4.3. The closest value of p in the table in Pacholczyk (1970) is p = 3, so we use this value to select the constants (and note that, as stated in C98, the results do not depend strongly on the value of p.) With this, Equations (8) and (9) reduce to Equations (13) and (14) in C98, respectively, reproduced here:
Next we estimate the total energy U. For p = 3, Equations (10) and (11) can be combined into the following expression for ,
Following C98 we take f = 0.5, but the dependence on this parameter is weak. In choosing and there are several normalizations (or assumptions) used in the literature. As a result the inferred energy can vary enormously (see Section 4.1 for further details). For now, we follow Soderberg et al. (2010) in setting (in other words, that energy is equally partitioned between electrons, protons, and magnetic fields). With all of these choices, we find that at , and . We find that the total energy . Assuming 10% uncertainties in Fp and and a 50% uncertainty in p, a Monte Carlo with 10,000 samples gives uncertainties of 0.15–0.3 dex in these derived parameters. Our results are robust to departures from equipartition given the large penalty in the required energy (Readhead 1994).
The mean velocity up to is . We can write a general expression for v/c (taking , noting that ):
Furthermore, from the t2 rise at 34 GHz (Figure 1) we can infer that the radius increases as and therefore that the velocity is constant. As shown in Figure 5, this is consistent with later-time, low-frequency observations by Margutti et al. (2018b). We put this derived energy and velocity into the context of other energetic transients in Section 4.1.
Next, we estimate the density of the medium into which the forward shock is propagating. The ejecta expands into the medium with velocity v1, producing a shock front (a discontinuity in pressure, density, and temperature) with shock-heated ejecta immediately behind this front. Conservation of momentum across this (forward) shock front requires that
where P is pressure (not to be confused with p used as the power-law index for the electron energy distribution). The subscript 1 refers to the upstream (ambient) medium and the subscript 2 refers to the downstream medium (the shocked ejecta). Far upstream, the pressure can be taken to be 0, and in the limit of strong shocks (for a monatomic gas) . Thus, this can be simplified to
If the medium is composed of fully ionized hydrogen, and the number densities of protons and electrons are equal (np = ne). Using Equation (15) together with Equation (11), as well as the relations and ,
We find that the number density of electrons at is . We note that the strong jump conditions used here assume , and that there is a correction for the contribution of a relativistic () component. Chevalier (1983) quantified this correction using the factor w, the ratio of the relativistic pressure to the total pressure. In the most extreme case (w = 1) the correction is small, only a factor of 1.14 in ne. This is negligible compared to our uncertainties.
At such a high density, the optical depth to free–free absorption (FFA) might be expected to have a significant effect on the shape of the spectrum at low radio frequencies (Lundqvist & Fransson 1988). From Lang (1999), we have
which, with our measured values of ne and R on Day 22, gives the characteristic value
However, in AT2018cow the gas through which the shock is propagating is not at normal H ii-region temperatures of . The UV and X-ray photons emitted at early times will completely ionize and Compton heat any surrounding gas: for gas at the density and radius given in Table 3, the lifetime to photoionization of a neutral hydrogen atom is less than 0.01 s, while the recombination time is years.15 Compton heating of the electrons increases their temperature at the rate
where the Klein–Nishina correction .16 Even though the blackbody () luminosity at is 100 times larger than the coeval X-ray luminosity (Perley et al. 2018), the Compton heating is dominated by the 10–100 X-ray flux, and we find, for , gas at the density and radius given in Table 3 has
Given the spectral evolution shown in Perley et al. (2018), the Compton temperature (at which Compton heating balances Compton cooling) is on Day 3, hardening to on Day 20 since the blackbody UV flux drops as , while the hard X-ray flux drops much more slowly. At these high Compton-heated temperatures , the FFA optical depth given by Equation (18) only rises above unity below frequencies of 300 MHz, accessible to facilities like Low-Frequency Array (LOFAR).
Next, we estimate the luminosity from free–free emission of the ionized gas (Lang 1999):
where ni is the number density of ions, Z is the atomic number, and is the Gaunt factor, a quantum mechanical correction. Assuming that the gas is completely ionized out to the light-travel sphere at 22 days (), we have ne = ni in the region of interest. We also take Z = 1. With the inferred density (Table 3) we find , so the contribution to the observed X-ray luminosity is negligible.
Finally, we estimate the different break frequencies, beginning with . Using Equation (3), taking and using our inferred from Day 22, . Next, using Equation (5) and our measured value of B, . Equation (4) thus gives , substantially below our peak frequency. The spectral index at is (Rybicki & Lightman 1986), which we show as dotted lines in Figure 3. Clearly, the lowest-frequency fluxes are in excess of extrapolation. This naturally occurs if the source is inhomogeneous (e.g., magnetic field and/or particle energy density decreasing outwards). It can also arise even for a perfectly homogeneous source because the energy spectrum of the radiating electrons is not a pure power law: note that , so even beyond the Maxwellian-like peak at , the spectrum is convex, steepening with energy above . These both produce self-absorbed spectra flatter than (see Section 6.8 in Rybicki & Lightman (1986) and de Kool et al. (1989) for model calculations).
The cooling frequency due to synchrotron radiation is determined by Equations (6) and (7). We find , giving a cooling frequency . The relative contributions to electron cooling from synchrotron radiation and IC scattering are determined by the ratio between the radiation energy density and the magnetic energy density. On Day 22, the bolometric luminosity as measured in the UVOIR is (Perley et al. 2018), so the radiation density is . The magnetic energy density on the same day is Thus, synchrotron radiation is the dominant cooling mechanism, with a roughly 10% contribution from Inverse Compton (IC) scattering.
At this epoch, the cooling timescale for an electron with , which is roughly 80 for an electron radiating at 100 GHz. So on Day 22 the cooling timescale is shorter than the timescale on which we are observing the source. This means that continuous re-acceleration of the electrons is required, which could be provided by ongoing shock interaction.
As stated in Section 3.1, it seems that during the rise phase (–8 days) was above or within the SMA observing bands. Using the peak observed flux and frequency as a lower limit on the peak flux and peak frequency, respectively, we consistently find that , albeit with a decreasing ne ( at ).
4. Implications of Shock Properties
4.1. AT2018cow in Velocity–energy Space, and a Discussion of Epsilons
It is challenging to directly compare the energy of AT2018cow to that of other classes of radio-luminous transients, because there are several conventions that produce discrepant results. In particular, the energy partition fractions and are important for determining the total amount of energy in the shock, but are difficult to measure.
In the classical GRB literature, has been consistently measured to be , within a factor of 2, while values of have much wider spreads, with a median value of but a distribution spanning four orders of magnitude (Kumar & Zhang 2015). Kumar & Barniol Duran (2010) constrain for a CSM density of , and an even smaller value for higher densities. One of the best-observed GRB afterglows is GRB 130427A, and from modeling the evolving spectrum Perley et al. (2014) find and .
There are different approaches to modeling for the handful of low-luminosity GRBs (LLGRBs) discovered to date. For LLGRB 980425/SN 1998bw, Kulkarni et al. (1998) invoked equipartition (). For LLGRB 031203/SN2003 lw, Soderberg et al. (2004) used models from Sari et al. (1998) and Granot & Sari (2002) together with the cooling frequency inferred from X-ray observations to estimate and . For LLGRB 060218/SN2006aj, Soderberg et al. (2006) used the same prescription as was used in SN 1998bw. Finally, for LLGRB 100316D/SN2010bh, Margutti et al. (2013) set and allowed to vary from 0.01–0.1.
For Type II and Type Ibc radio SNe, approaches range from using the SN 1998bw convention (i.e., Soderberg et al. 2005; Horesh et al. 2013) to (e.g., Chevalier & Fransson 2006; Soderberg et al. 2006; Salas et al. 2013) to for the relativistic supernova SN 2009bb (Soderberg et al. 2010).
In this work, we follow the convention in Soderberg et al. (2010) so that we can compare our velocity–energy diagram to the corresponding diagram (Figure 4) in that paper. To put all transients on the same scale, we take the peak frequency, peak luminosity, and peak time for each event, and run them through the same equations that we used to infer the shock properties of AT2018cow. Note that we do not vary the values of , p, but these are all very small corrections, whereas the effect of and the ratio in estimating the energy is large. The details of how we selected the peak values for each event are in Appendix C. When possible, we use the peak of the SED at a particular epoch. However, for most events, we use the peak flux density corresponding to a certain frequency, because well-sampled SEDs are rare.
Our rederived velocity–energy diagram is shown in Figure 4. AT2018cow has an energy comparable to mildly relativistic (LLGRBs; e.g., SN 1998bw) outflows and energetic SNe (e.g., SN 2007bg). We display vertical axes for two different conventions ( and ) to show how this affects the inferred energy. Note that these values are not evaluated for a consistent epoch. However, for AT2018cow, we have reason to believe that the values of velocity and energy do not change significantly over the course of our observations. For other sources, it would be necessary to have a well-sampled SED over multiple epochs in order to trace the evolution of these values, and this is rare in the literature.
Download figure:
Standard image High-resolution image4.2. A Luminous Millimeter Transient in a Dense Environment
Here we compare the radio luminosity of AT2018cow to that of other transients, as observed at the spectral peak frequency at time . As shown in Figure 5, AT2018cow stands out as being several times more luminous than SN 1998bw, and having a late peak at high frequencies. Over time, the peak luminosity diminishes to the value reported in the low-frequency radio observations of Margutti et al. (2018b), supporting our inference that the velocity is not changing significantly.
Download figure:
Standard image High-resolution imageOn this diagram we also indicate lines of constant velocity (see Figure 3 of Soderberg et al. 2010 and Figure 4 of C98) and lines of constant mass-loss rate scaled by velocity , as a diagnostic of density (see Figure 10 of Jencson et al. 2018). Note that these lines assume .
We now derive relations between the observational coordinates of the diagram in Figure 4 and physical quantities: the ordinate, peak radio luminosity Lp, is simply a power of the energy per unit radius U/R. We get an expression for U/R using Equations (12) and (10):
This translation between Lp and U/R is shown on the left and right axis labels of Figure 4.
We now show that the abscissa of Figure 5, , is very nearly proportional to the square root of the swept up mass per unit radius M/R, or equivalently, if the surrounding medium was from a pre-explosion steady wind of speed vw, . A steady spherical wind of ionized hydrogen with velocity vw has ), so we can reparameterize the density in terms of the mass-loss rate:
Notice the weak ( power) dependence on Lp, and the quadratic dependence on , which means the lines of constant are nearly vertical in Figure 5.
AT2018cow lies along the same velocity line as SN 2003bg and SN 2003L, but ne is a factor of a few to an order of magnitude larger.17 Similarly, SN 1998bw lies along a similar velocity line to SN 2006aj, but ne (using our prescription) is 40 , while the density inferred for 2006aj using the same prescription is 3 . For SNe Ibc in general, Chevalier & Fransson (2006) attributed the large spread in radio luminosity to a spread in circumstellar density, with the example that SN 2002ap (not shown in Figure 5 due to its relatively low luminosity) is roughly three orders of magnitude less luminous than SN 2003L, and its inferred ambient density is also a factor of 3 smaller. In SN 2003L and SN 2003bg, the high density was attributed to a stellar wind.
This is not the whole story: as we showed above, high peak radio luminosity just corresponds to high U/R, i.e., high energy and/or small radius. Since U is the converted energy, it represents only a lower limit to the actual driving kinetic energy (becoming equal to it as the explosion transitions from free expansion to the Sedov phase). A higher-density medium more quickly converts the piston's energy to thermal energy than does a low-density medium. Thus, for a large fixed explosion energy, a denser medium will indeed lead to larger peak radio luminosities. But the direct correlation is with the (thermalized) energy per unit radius, U/R. Similarly, Equation (11) shows that (except for a very weak dependence), . Thus, higher peak frequencies are directly indicative of a higher magnetic field, or equivalently, pressures. Therefore, AT2018cow's high and high Lp are quite likely mostly a consequence of it being energetic, and observed early, when the high wind density at small radii led to high pressure, and enhanced U/R. As we discuss below, this suggests that many other SNe could have shown similar bright mm-submm fluxes, had they been observed at those wavelengths in their first week.
On Day 22, the inferred density is at a radius of (Table 3). From this we can infer . The mass swept up to radius r is . In Section 3, we argue that the blast wave reaches the edge of the surrounding bubble around . If so, given our inferred velocity, the radius of the circum-bubble is , and the mass of the circum-bubble . The mass-loss rates for hot stars (, Lamers & Leitherer 1993) and red supergiants (, van Loon 2010) range from 10−4–10−6 (Smith et al. 2018). Thus, is for a hot star progenitor, and ∼100 yr for a red supergiant. Therefore, the circum-bubble could either have been formed by normal mass loss in a red supergiant, or end-of-life enhanced mass loss from a hot star or red supergiant (see, e.g., Smith et al. 2017 and references therein).
UVOIR observations of AT2018cow place strong constraints on the nature of the surrounding medium (Perley et al. 2018). The high luminosity and fast rise can be interpreted as shock heating of a dense shell of material at or 10 au, qualitatively consistent with the inference of dense material given the properties inferred from the radio shock. On the other hand, early spectra show no narrow emission lines indicative of a shock and the light curve declines steeply after peak, both of which suggest that this dense material must also be quite limited in extent, with little material at larger radii. While the radio observations also suggest that a cutoff in the density distribution may exist, the shock does not reach it for almost 20 days, a quite different timescale than the optical peak (reached in less than 3 days) or early spectroscopy. This might be due to the shock being produced by breakout from the shell, which re-energized the (much slower) SN shock. Or there could be deviations from spherical symmetry (for example, with the optical heating a quasi-spherical shell but the radio shock passing through a denser toroidal component or clouds along a bipolar jet).
Thus, an energetic shock propagating into a dense environment could produce a radio SED that peaks at submillimeter wavelengths at early times. However, as illustrated in the left panel of Figure 6, searches at high frequencies at early times have been rare, and primarily limited to transients with relativistic jetted outflows (GRBs, TDEs). We suggest that these searches be expanded to other classes of transients: luminous SNe such as SN 2003L and SN 2007bg, and luminous TDEs such as ASASSN14li, all exploded into dense media and exhibited luminous centimeter-wavelength emission at . As time goes on, the SED peak shifts to lower frequencies and diminishes in brightness, so these events could have been bright millimeter transients at . This is supported by Figure 5, which shows that SN 2007bg, SN 2003bg, and SN 2003L could have appeared similar to AT2018cow had they been observed earlier at higher frequencies.
Download figure:
Standard image High-resolution image4.3. Novel Features of the Synchrotron Model Parameters
The ordering of the break frequencies, , is an unusual regime for long-wavelength observations. For a relativistic shock (GRBs), the typical orderings are (the fast cooling regime) and (the slow cooling regime; Sari et al. 1998). For nonrelativistic shocks, the ordering in most cases seems to be at measured frequencies above 1.4 GHz, but can also be is typically considered unimportant for long-wavelength observations (Nakar & Piran 2011).
The low cooling frequency is a consequence of a large magnetic field strength, (reduced even further for by Compton cooling on the UVOIR flux, which dominates over synchrotron cooling). This in turn presumably arises from the injection of a large amount of energy into a small volume of material, consistent with the low velocity we measure. From Equation (11) we see that Bp scales as . Changing from 1/3–0.01 could increase by a factor of 8, still much lower than our observed frequencies. This regime is selectively probed by submillimeter observations, because a low (high Bp) gives rise to a that falls in this wavelength regime. We note that in the same framework, the relativistic TDE Swift J1644+57 (whose long-wavelength SED also peaked in the submillimeter for the first few weeks) would also have had a cooling frequency below much of the observed frequency range ().
Since is below any of our measured frequencies, the injection spectrum (the spectrum of the electrons prior to cooling) has a shallower power-law index than what we measure, . This suggests that on Day 22, when , This is not unreasonable for Fermi acceleration from a strong shock. Typical young Galactic SN remnants have in the radio, flattening to at higher frequencies (Urošević 2014).
5. Origin of the X-Ray Emission and Emergence of a Compact Source
During the plateau phase, the fluence in the Swift/XRT bands is . Integrating the Swift/XRT light curve until , we find . Using the NuSTAR spectra index , extrapolating this to would increase this energy by a factor of 3.
The total X-ray energy emitted in the first 3 weeks is thus greater than the total energy in the shock inferred from radio observations on Day 22, . If a significant proportion of the X-rays is produced by IC emission, then our assumption of clearly results in an underestimate of the total energy, and an assumption of would be more appropriate. As shown in Figure 4, U would be increased by a factor of 9 with the assumption and , just barely comparable to the total energy emitted in X-rays.
The luminosity of the UVOIR source declines , where (Perley et al. 2018). Assuming a constant expansion speed for the shock of (see Table 3 and related discussion in Section 3) the photon energy density of the UVOIR source, //. Assuming that the magnetic field pressure scales with the ram pressure of the shock () and assuming , from Equation (11), the magnetic energy density . For , , , with only a rather weak dependence on the epsilons. This ratio is equal to unity around , marking the transition from a regime dominated by Compton cooling to a regime dominated by synchrotron cooling.
This ratio is much lower than the observed ratio , and the X-ray spectral index is also substantially flatter than the radio spectral index. We conclude that the X-ray emission during the plateau phase does not naturally arise from IC scattering of the UVOIR source by the electrons in the post-shocked region (which also generate the radio to submillimeter emission via synchrotron radiation): IC from the radio-mm emitting region alone underpredicts the X-ray luminosity, predicts an X-ray luminosity declining much more rapidly than observed in the first 20 days, and predicts too steep a spectrum. It also does not naturally arise from an extension of the radio-submm synchrotron spectrum: the X-ray emission is some 25 times brighter than that extrapolation (see Figure 7), and has a much flatter () spectral index. Further speculative modeling of the source of the X-ray emission during the plateau phase is beyond the scope of this paper.
Download figure:
Standard image High-resolution imageDuring the decline phase , the timescale of these fluctuations is around , while the diameter we infer for the radio-emitting region (see Section 3) is . Thus, the X-ray emission must arise in a different and more compact source than the radio-emitting shell.
From the plateau phase to the decline phase, the X-ray emission softens, as shown in the bottom panel of Figure 1 and reported by Kuin et al. (2018). From the NuSTAR data, measuring the flux using cflux, we infer a hardness ratio LX(10–200 keV)/LX(0.3–10 keV) ≈ 26 on Epoch 1, similar to what is inferred by Kuin et al. (2018) using a joint BAT/XRT analysis. From the NuSTAR data, we find a hardness ratio of LX(10–200 keV)/LX(0.3–10 keV) ≈ 4–5 on Epochs 3 and 4. This is consistent with other studies, which found negligible spectral evolution in the Swift 0.3–10 keV band, but significant spectral evolution at higher energies (Kuin et al. 2018; Margutti et al. 2018b; Rivera Sandoval et al. 2018).
These two changes—the onset of variability five times faster than the light-travel time across the radio-emitting shell, and the striking change in the spectrum—lead us to conclude that beyond 20 days the X-ray emission arises from a different and more compact source than during the plateau phase. In the decline phase we are, arguably, probing regions closer to the central engine of the event.
The peculiarities of the UVOIR spectrum have led some to propose that AT2018cow is a TDE of a white dwarf by an black hole (Kuin et al. 2018; Perley et al. 2018). Given its off-nucleus location (1.7 kpc; Perley et al. 2018) in a star-forming galaxy, and the similarities of the radio-emitting shock to those of other SNe, it seems more natural to suppose that AT2018cow originated in a stellar cataclysm. Ultimately, however, our radio observations only require a shock wave propagating into a dense medium, which could very plausibly arise in both TDE and SN models. The radio observations do little to distinguish them. In either picture, the striking late-time change in the X-ray behavior suggests the emergence of a central engine. In the TDE case, this could be an accretion disk around a black hole. In the stellar explosion case, this could be a natal black hole accreting (fallback) matter from the debris, or a magnetar. The emergence could then be due to a channel between the interior and the surface opened up by a collimated outflow (a "jet" or stifled jet's cocoon breakout; Nakar 2015), or to gaps in the photosphere opened by Rayleigh–Taylor instabilities.
We now briefly explore the magnetar model, which has been proposed for cosmological long-duration GRBs (e.g., Thompson et al. 2004) and superluminous SNe (e.g., Kasen & Bildsten 2010). Prentice et al. (2018) invoked a magnetar model to explain the UVOIR observations of AT2018cow and found a best-fit magnetic field strength of and a best-fit spin period of . Note that the magnetar itself need not be directly visible: the X-rays we see could be due to the emergence of bubbles of the magnetar-powered wind nebula (Kasen et al. 2016). The spin-down luminosity of a magnetar with period P is , where is the angular frequency. The spin-down timescale is . We set , and find and , assuming that all the spin-down power goes into X-ray production. For a constant spin-down rate, this would correspond to an initial spin period of 26 ms, similar to the result in Prentice et al. (2018) for the model fit to the griz light curve.
With P and in hand, using the standard dipole formula, we find a lower limit on the magnetic field strength of G, which is consistent with the value found in Prentice et al. (2018). Our modeling of the forward shock led to a lower limit to the energy of erg, depending on the value of . If this was supplied by a magnetar, then the initial period of the magnetar is . We end this discussion by noting that the spin-down luminosity in the dipole model (with constant B-field) is ∝ t−2, which is roughly consistent with the slope of the decay of the X-ray light curve. This is, however, not easily distinguished from the slope expected from accretion in a TDE (Phinney 1989) or fallback (Michel 1988).
If this is a stellar explosion, then the features in the UVOIR spectra and the rise time point to an extended progenitor ( Perley et al. 2018), comparable in size to the largest red supergiants. This is not consistent with the compact, stripped stars invoked as progenitors for other classes of engine-driven explosions like GRBs and SLSNe (although see Smith et al. (2012) for a possible exception). As discussed in Perley et al. (2018) and Section 4.2, a more likely scenario is that the progenitor experienced a dramatic, abrupt episode of mass loss shortly before the explosion, and the UVOIR photosphere lies within this "brick wall," which the SN blast wave struck and re-thermalized.
Regardless of the nature of the central engine, we have the following model: the fastest-moving ejecta races ahead at into a dense circum-bubble of radius, Rb. In the post-shocked gas, electrons are accelerated into a power-law spectrum and magnetic fields are amplified. We attribute decay of the resulting radio emission at to the fast-moving ejecta reaching the edge of this circum-bubble and infer a radius , and a mass of . Within the radio-emitting shell is a long-lived engine, which may inflate a bubble of plasma and magnetic fields (Bucciantini et al. 2007). There is also slower ejecta heated by a central source (or radioactivity) that expands and emits UVOIR radiation. The photosphere of this component recedes with time, and at early times its large Compton optical depth obscures direct emission from the vicinity of the central engine. At later times, this central region emerges.
6. Conclusions and Outlook
Persuasive arguments can and have been made for both SN and TDE origins for AT2018cow. Our extensive radio through submillimeter observations enable us to draw definitive conclusions about the outer blast wave launched following the event, independent of the origin of the event. The blast wave is sub-relativistic () and plows into a dense medium ( at ). The energy contained within this blast wave is . In contrast to the UVOIR luminosity, which declines as over the period of , the mm-submm and X-ray luminosities are both relatively constant over the same period (with ∼50% variations over timescales of a few days, comparable to the light-travel time across the radio-emitting shell, see Figure 1).
The initially attractive idea of attributing the X-rays to IC scattering of the rapidly declining flux of UVOIR photons by relativistic electrons (which give rise to the radio and submillimeter flux) is not naturally consistent with the slow decline of the X-ray and radio-submm flux, the X-ray luminosity, or the X-ray spectral index. Thus, we are forced to invoke an additional source of X-ray emission during both the plateau phase, and during the decline phase after 20 days, when the X-rays begin to fade and show dramatic variations now on timescales several times shorter than the light-travel time across the radio-emitting shell. These are suggestive of power from a central engine, which could be consistent with either a stellar explosion or a TDE. Future X-ray monitoring may be useful in differentiating between central engine models; in particular, a power-law decay () is a distinct signature of the magnetar model, although difficult in practice to distinguish from the expected from fallback or a TDE.
The radio source is remarkable even on purely observational grounds. The peak radio luminosity (nearly ) greatly exceeds that of the most radio-luminous SNe and "normal" TDEs, and is surpassed only by relativistic jetted transients (GRBs and TDEs). The source remains luminous at submillimeter wavelengths for nearly a month, with a self-absorption frequency at .
The source is strongly detected at nearly a terahertz (ALMA Band 9; 671 GHz) even 3 weeks post discovery. We note that the Band 9 flux is higher than the extrapolation based from lower frequency bands (Figure 3, middle panel), and intriguingly connects to the NIR nonthermal component suggested by Perley et al. (2018). However, we readily admit that the apparent excess in the Band 9 flux is only 2σ, and also note that the case for the NIR nonthermal component is not secure.
Finally, it is worth re-iterating that AT2018cow is a mere 60 Mpc away. The proximity hints at an extensive population of which AT2018cow is the prototype. The key distinction between AT2018cow and other fast transients is the strong millimeter and submillimeter emission. In general, it is apparent from Figure 5 that an energetic shock propagating into a dense medium will exhibit strong millimeter emission during the first weeks. Many other SNe would likely have had bright emission at mm-submm wavelengths, had they been observed early at those wavelengths. Combining velocities measured at very early times at such short wavelengths, with much later observations at low frequencies could reveal the slowing of the shock associated with the transition from free expansion to the Sedov phase, constraining the total energy in relativistic ejecta. Taken together, these two developments, given that we are now squarely in the era of industrial optical time domain astronomy (e.g., PS-1, PS-2, ASAS-SN, ATLAS, ZTF, and soon BlackGEM), argue for a high-frequency facility dedicated to the pursuit of transients.
The code used to produce the results described in this paper was written in Python and is available online in an open-source repository.18
The authors are grateful to the staff at the SMA, the CSIRO Astronomy and Space Science (CASS), and ALMA for rapidly scheduling and executing the observations and reducing the data. It is a pleasure to thank John Carpenter for his guidance and his assistance with the ALMA observations. Thank you to Dale Frail, Raffaella Margutti, and Roger Chevalier for providing feedback on the manuscript, and Gregg Hallinan, Dillon Dong, and Jacob Jencson for helpful discussions. Finally, we would like to thank the anonymous referee for thoughtful suggestions that greatly improved the clarity of the paper.
A.Y.Q.H. was supported by a National Science Foundation Graduate Research Fellowship under Grant No. DGE-1144469. This work was supported by the GROWTH project funded by the National Science Foundation under PIRE Grant No. 1545949. This research was funded in part by the Gordon and Betty Moore Foundation through Grant GBMF5076 to E.S.P., and A.Y.Q.H., E.S.P., and S.R.K. benefited from interactions with Dan Kasen, David Khatami, and Eliot Quataert funded by that grant. T.M. acknowledges the support of the Australian Research Council through Grant FT150100099. D.D. is supported by an Australian Government Research Training Program Scholarship.
The SMA is a joint project between the Smithsonian Astrophysical Observatory and the Academia Sinica Institute of Astronomy and Astrophysics and is funded by the Smithsonian Institution and the Academia Sinica. The ATCA is part of the Australia Telescope National Facility, which is funded by the Australian Government for operation as a National Facility managed by CSIRO. This paper makes use of the following ALMA data: ADS/JAO.ALMA#2017.A.00047.T. ALMA is a partnership of ESO (representing its member states), NSF (USA), and National Institutes of Natural Sciences (NINS) (Japan), together with National Research Council (NRC) (Canada), Taiwan Ministry of Science and Technology (MOST), and Academia Sinica Institute of Astronomy and Astrophysics (ASIAA) (Taiwan), and Korea Astronomy and Space Science Institute (KASI) (Republic of Korea), in cooperation with the Republic of Chile. The Joint ALMA Observatory is operated by ESO, Associated Universities, Inc. (AUI)/NRAO and National Astronomical Observatory of Japan (NAOJ). The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. This work made use of data supplied by the UK Swift Science Data Centre at the University of Leicester.
Software: Astropy (Astropy Collaboration et al. 2013, 2018), IPython (Pérez & Granger 2007), Matplotlib (Hunter 2007), NumPy (Oliphant 2006), and SciPy (Jones et al. 2001).
Appendix A: ALMA Band 9 Calibration
For ALMA Band 9, due to the relatively low signal-to-noise of the data, all eight spectral windows were combined to derive a combined phase solution, which was then mapped to each individual spectral window. As a result, no in-band analysis of the spectral index was done and a single flux value was derived for the Band 9 imaging. When fitting a Gaussian function to the Band 9 image, the source is represented as a point source, suggesting that the image is of good enough quality to derive a meaningful flux densities. A phase-only self-calibration did not provide good solutions, it decreased the phase coherence, and it resulted in an image that could no longer be fitted with a point source. Therefore, we did not apply a self-calibration to the Band 9 data. To verify that changing weather conditions did not affect the phase coherence in the data, we split the data in three different time bins and imaged each time bin separately. The change in flux density between the different time bins was within 6%. Similarly, imaging the data from the short, intermediate and long baselines by splitting the data in three bins in the UV range (12−90 m, 90−170 m, and 170−312 m) showed a difference in flux density <13%, despite the sparser antenna distribution and poor UV coverage in the long-baseline bin. Also, the XX and YY polarization images were similar to within 3%. To examine the reliability of the absolute flux calibration of the Band 9 data, we imaged the two secondary calibrators, the quasars J1540+1447 and J1606+1814, using the same flux and bandpass calibrator as for AT2018cow. Figure 8 shows that the Band 9 flux densities of these two secondary calibrators are in reasonable agreement with values from the ALMA calibrator catalog in the lower bands if there is no spectral curvature, although uncertainties in absolute flux calibration may have led us to slightly overpredict our derived Band 9 values. In all, our tests are consistent with the ALMA Band 9 flux density being accurate to within a 20% uncertainty, which is standard for high-frequency ALMA observations.
Download figure:
Standard image High-resolution imageAppendix B: Full SMA Light Curves
Figure 9 shows the full set of SMA light curves, grouped by frequency.
Download figure:
Standard image High-resolution imageAppendix C: Selection of Peak Frequency and Peak Luminosity for Other Transients
- 1.
- 2.
- 3.
- 4.
- 5.
- 6.
- 7.
- 8.
- 9.
- 10.
- 11.
- 12.
- 13.
- 14.
- 15.Swift J1644+57 (z = 0.354): For Figures 4 and 5, we use the reported on Day 15 (corrected to Day 18 in Eftekhari et al. 2018). For Figure 6, we use the 225 and 230 GHz light curves from the SMA (Zauderer et al. 2011; Berger et al. 2012), adding 3.04 days to the Zauderer et al. (2011) points because (as described in Eftekhari et al. 2018) subsequent analysis of the BAT data revealed emission earlier than had been previously noticed. We use 4.9 GHz data from Berger et al. (2012), Zauderer et al. (2013), and Eftekhari et al. (2018).
Footnotes
- 11
The authors searched for transient sources at 90 and 150 GHz. They found a single candidate event, which intriguingly showed linear polarization.
- 12
Here we use "transient" as distinct from "variable": millimeter observations are used to study variability in protostars (e.g., Herczeg et al. 2017) and more commonly for active galactic nuclei (e.g., Dent et al. 1983). There have also been millimeter detections of galactic transient sources, primarily stellar flares (e.g., Bower et al. 2003; Fender et al. 2015).
- 13
- 14
Band 9 observations were also performed on 2018 July 09, but these data were of too poor quality to use as a result of weather conditions.
- 15
For much lower temperatures , the Case B (high-density limit) recombination coefficient is (Draine 2011), and the timescale is . For , . This timescale becomes even longer for the expected higher temperatures.
- 16
- 17
The radial density profile inferred for SN 2003bg is (Soderberg et al. 2006) and the radial density profile inferred for SN 2003L is (Soderberg et al. 2005). In both cases, is the shock radius at . For SN 2003L, we infer t0 using the result that at , and that in the parameterization (Model 1 in Soderberg et al. 2006).
- 18