ZTF20aajnksq (AT 2020blt): A Fast Optical Transient at z ≈ 2.9 with No Detected Gamma-Ray Burst Counterpart

The ZTF Uniform Depth Survey (D. A. Goldstein et al. 2020, in preparation) covers 2000 deg² twice per night in g-, r-, and i-bands using the 48-inch Samuel Oschin Schmidt telescope at the Palomar Observatory (P48). The ZTF observing system is described in Dekany et al. (2020). The pipeline for ZTF photometry makes use of the image subtraction algorithm of Zackay et al. (2016) and is described in Masci et al. (2019).

AT 2020blt was discovered at r = 19.57 ± 0.14 mag (all magnitudes given in AB) in an image obtained on 2020 January 28.28, ¹⁹ at the position α = 12^h47^m0487, δ = + 45^d12^m023 (J2000). One and a half hours later, the source had faded to r = 20.01 ± 0.16 mag.

AT 2020blt passed a filter that searches the ZTF alert stream (Patterson et al. 2019) for young and fast transients. More specifically, the filter identifies transients that:

• 1.  
  have an upper limit from the previous night that is at least one magnitude fainter than the first detection,
• 2.  
  have no historical detections in the Catalina Real-Time Transient Survey (Drake et al. 2009; Djorgovski et al. 2011; Mahabal et al. 2011), ZTF, or the predecessor to ZTF the Palomar Transient Factory (Law et al. 2009),
• 3.  
  have a real-bogus score drb > 0.9, which is associated with a false positive rate of 0.4% and a false negative rate of 5% (Duev et al. 2019),
• 4.  
  have a Galactic latitude ∣b∣ > 15,
• 5.  
  have two detections separated by at least half an hour (to remove asteroids), and
• 6.  
  have no stellar counterpart (sgscore < 0.76; Tachibana & Miller 2018).

AT 2020blt fulfilled the criteria listed above: the last upper limit from the high-cadence survey was 0.74 days prior to the first detection with an upper limit of r > 20.73 mag. There is no source within 15'' of the position of AT 2020blt in ZTF r-band and g-band reference images, with a 5σ limiting magnitude in the PSF-fit reference image catalog of r = 23.17 mag and g = 22.77 mag.

Motivated by the fast rise and lack of a detected host-galaxy counterpart in ZTF reference images ²⁰ we immediately triggered a series of follow-up observations (Section 2.2) which were coordinated through the GROWTH "Marshal" (Kasliwal et al. 2019). All observations will be made available on WISeREP, the Weizmann Interactive Supernova Data Repository (Yaron & Gal-Yam 2012).

2.2.1. Optical Imaging

In searches for extragalactic fast transients, the primary false positives are stellar flares in the Milky Way (Kulkarni & Rau 2006; Rau et al. 2008; Berger et al. 2013; Ho et al. 2018). At optical frequencies, stellar flares can be distinguished from afterglow emission by color. At peak, stellar flares have typical blackbody temperatures of ∼10,000 K (Kowalski et al. 2013), so optical filters will be on the Rayleigh–Jeans tail and colors will obey f_ν ∝ ν ^+ 2 (g − r = −0.17 mag). By contrast, in optical bands synchrotron emission will obey g − r > 0. For example, a spectral index of f_ν ∝ ν^−0.7 (Sari et al. 1998) corresponds to g − r = 0.24 mag.

To measure the color of AT 2020blt, we triggered target-of-opportunity (ToO) programs on the IO:O imager of the Liverpool Telescope ²¹ (LT; Steele et al. 2004) and with the Spectral Energy Distribution Machine ²² (SEDM; Blagorodnova et al. 2018; Rigault et al. 2019) on the automated 60-inch telescope at the Palomar Observatory (P60; Cenko et al. 2006). LT image reduction was provided by the basic IO:O pipeline. P60 and LT image subtraction was performed following Fremling et al. (2016), using PS1 images for griz and SDSS for u band.

LT griz observations on January 29.17 and P60 gri observations 2 hr later confirmed that AT 2020blt had red colors. Furthermore, forced photometry (Yao et al. 2019) on P48 images revealed two g-band detections that were below the 5σ threshold of the nominal ZTF pipeline, which give g − r = 0.77 ± 0.16 mag. Photometry was corrected for Milky Way extinction following Schlafly & Finkbeiner (2011) with E(B − V) = A_V /R_V  = 0.034 mag, using R_V  = 3.1 and a Fitzpatrick (1999) extinction law. The full light curve of AT 2020blt is shown in the left panel of Figure 1, and the photometry is listed in Table 1.

Zoom In Zoom Out Reset image size

Table 1. Summary of Observations of AT 2020blt

Optical Photometry
Obs. Date	Δt (days)	Instrument	Filter	Mag
Jan 27.54	−0.64	P48+ZTF	r	>21.36
Jan 28.28	0.10	P48+ZTF	r	19.60 ± 0.08
Jan 28.35	0.17	P48+ZTF	r	19.97 ± 0.08
Jan 28.39	0.21	P48+ZTF	g	20.74 ± 0.14
Jan 28.51	0.33	P48+ZTF	g	20.70 ± 0.13
Jan 29.17	0.99	LT+IO:O	g	21.52 ± 0.21
Jan 29.17	0.99	LT+IO:O	r	21.29 ± 0.18
Jan 29.17	0.99	LT+IO:O	i	21.15 ± 0.25
Jan 29.17	0.99	LT+IO:O	z	20.82 ± 0.40
Jan 29.21	1.03	LT+IO:O	g	21.83 ± 0.21
Jan 29.21	1.03	LT+IO:O	r	21.00 ± 0.16
Jan 29.21	1.03	LT+IO:O	i	21.15 ± 0.27
Jan 29.29	1.11	P60+SEDM	r	21.03 ± 0.19
Jan 29.29	1.11	P60+SEDM	g	21.30 ± 0.26
Jan 29.45	1.27	P48+ZTF	i	21.52 ± 0.32
Jan 29.47	1.29	P48+ZTF	i	21.45 ± 0.25
Jan 29.51	1.33	P48+ZTF	r	21.58 ± 0.26
Jan 29.53	1.35	P200+WaSP	g	22.53 ± 0.10
Jan 29.54	1.36	P48+ZTF	g	21.69 ± 0.32
Jan 29.55	1.37	P200+WaSP	r	21.90 ± 0.07
Jan 29.55	1.37	P200+WaSP	i	21.56 ± 0.05
Jan 31.11	2.93	LT+IO:O	r	>22.04
Jan 31.12	2.94	LT+IO:O	i	>22.24
Feb 01.53	4.35	Gemini-N+GMOS	r	25. 20 ± 0. 05
Optical Spectrum with LRIS on Keck I
Obs. Date	Δt (days)	Observing Setup		Exposure Time
Jan 30.64	2.44	1"-wide slit, 400/3400 grism, 400/8500 grating, D560 dichroic		900 s
0.3–10 keV X-ray Observations with Swift/XRT
Obs. Date	Δt (days)	Count Rate	Flux
Jan 29.70	2.1	$({3.96}_{-1.08}^{+1.30})\times {10}^{-3}\,{{\rm{s}}}^{-1}$	${1.33}_{-0.36}^{+0.44}\times {10}^{-13}\,\mathrm{erg}\,{{\rm{s}}}^{-1}\,{\mathrm{cm}}^{-2}$
Jan 31.04	3.4	<3.95 × 10−3 s−1	<1.33 × 10−13 erg s−1 cm−2
VLA Radio Observations at 10 GHz
Obs. Date	Δt	Time On-source	Flux Density	Flux at 10 GHz
Feb 07.24	10.08	0.7 hr	52.1 ± 6.5 μJy	(5.21 ± 0.65) × 10−18 erg s−1 cm−2
Feb 23.54	26.38	0.7 hr	<15 μJy	<1.5 × 10−18 erg s−1 cm−2
Apr 29.98	92.82	0.7 hr	<21 μJy	<2.1 × 10−18 erg s−1 cm−2

Note. Time given relative to t₀ as defined in Section 3.1. Optical magnitudes have been corrected for Milky Way extinction. P48 values were measured using forced photometry (Yao et al. 2019). X-ray uncertainties are 1σ and upper limits are 3σ. Radio upper limits are 3× the image rms. Uncertainties on radio measurements are given as the quadrature sum of the image rms and a 5% uncertainty on the flux density due to flux calibration.

Download table as:  ASCIITypeset image

To monitor the light curve, we triggered a ToO program ²³ with the Wafer-Scale Imager for Prime (WaSP) on the 200-inch Hale telescope at the Palomar Observatory (P200) and obtained 2 × 180 s exposures in each of g-, r-, and i-bands. The WaSP reductions were performed using a pipeline developed for Gattini-IR, described in De et al. (2020). The measurement established a rapid fade rate of 2.3 magnitudes in 1.3 days and confirmed the red colors (g − r = 0.63 ± 0.12 mag).

For a final photometry measurement, we triggered a ToO observation with the Gemini Multi-Object Spectrograph (GMOS; Hook et al. 2004) on the Gemini-North 8-meter telescope on Maunakea. ²⁴ In 8 × 200 s exposures on February 01.53, calibrating against PS1 DR1 (Chambers et al. 2016), we detected the source at r = 25.20 ± 0.05 mag (Singer et al. 2020). Data were reduced using DRAGONS (Data Reduction for Astronomy from Gemini Observatory North and South), a Python-based reduction package provided by the Gemini Observatory. In Section 3.1 we model the full optical light curve of AT 2020blt and compare it to GRB afterglows in the literature.

We triggered ToO observations ²⁵ using the Low Resolution Imaging Spectrometer (LRIS; Oke et al. 1995) on the Keck I 10-m telescope. The observation details are listed in Table 1. The spectrum was reduced with LPipe (Perley 2019) and is shown in Figure 2. The spectrum showed features consistent with the Lyman break (rest-frame 912 Å) and Lyα absorption (rest-frame 1216 Å) at $z={2.90}_{-0.04}^{+0.05}$ (luminosity distance 25 Gpc ²⁶ ), although the S/N is low due to the short exposure time and the fact that the observation started close to morning twilight. We searched for narrow lines consistent with z = 2.9 with no convincing detections. The redshift sets the rest-frame UV magnitude at the time of discovery as M_{1626 Å} = −25.91 mag, assuming a distance modulus of 46.99 mag and a central wavelength of the ZTF r-band filter of 6340 Å.

Zoom In Zoom Out Reset image size

We triggered ToO observations ²⁷ with the X-Ray Telescope (XRT; Burrows et al. 2005) on board the Neil Gehrels Swift observatory (Gehrels et al. 2004). We obtained two epochs of 4 ks exposures and reduced the data using the online tool ²⁸ developed by the Swift team (Evans et al. 2007). In the first epoch (January 29.70; Δt = 2.1 days) a source was detected at the position of AT 2020blt with a 0.3–10 keV count rate of $({3.96}_{-1.08}^{+1.30})\times {10}^{-3}\,{{\rm{s}}}^{-1}$. Assuming a neutral hydrogen column density n_H  = 1.69 × 10²⁰ cm⁻² and a photon index Γ = 2 the unabsorbed flux density is ${1.33}_{-0.36}^{+0.44}\times {10}^{-13}\,\mathrm{erg}\,{{\rm{s}}}^{-1}\,{\mathrm{cm}}^{-2}$. The source was not detected in the second epoch (January 31.04; Δt = 3.4 days) with a 3σ confidence upper limit of  < 3.95 × 10⁻³ s⁻¹. We used webpimms ²⁹ with Γ = 2 and the same value of n_H from the first observation to convert the upper limit on the count rate to an upper limit on the flux density of < 1.33 × 10⁻¹³ erg s⁻¹ cm⁻². A log of our X-ray observations is provided in Table 1, and we model the X-ray to radio SED in Section 3.2.

On February 3 we triggered our ToO program on the Karl G. Jansky Very Large Array (VLA; Perley et al. 2011) for fast-rising and luminous transients. ³⁰ We obtained an X-band observation on February 7.24 (start time; Δt = 10.08 d) in C configuration, using 3C286 as the bandpass and flux density calibrator and J1219 + 4829 as the phase calibrator. We calibrated the data using the automated pipeline available in the Common Astronomy Software Applications (CASA; McMullin et al. 2007), and performed additional flagging manually before imaging. Imaging was performed using the CLEAN algorithm (Högbom 1974) implemented in CASA. The cell size was one-tenth of the synthesized beamwidth, and the field size was the smallest magic number (10 × 2ⁿ ) larger than the number of cells needed to cover the primary beam. A source was detected at the position of AT 2020blt with a flux density of 52.1 ± 6.5 μJy. In the next X-band image (February 23.54; Δt = 26.38 d) the source was not detected with an rms of 5 μJy. In the final observation (Apr 29.98; Δt = 92.82 d) the source was not detected with an rms of 7 μJy. A log of our radio observations is provided in Table 1. In Section 3.2 we model the X-ray to radio SED and in Section 3.3 we put the radio luminosity in the context of GRB afterglows.

The third Interplanetary Network (IPN ³¹ ) consists of six spacecraft that provide all-sky full-time monitoring for high-energy bursts. The most sensitive detectors in the IPN are the Swift Burst Alert Telescope (BAT; Barthelmy et al. 2005) the Fermi Gamma-ray Burst Monitor (GBM; Meegan et al. 2009), and the KONUS instrument on the Wind spacecraft (Aptekar et al. 1995).

We searched the Fermi GBM Burst Catalog, ³² the Fermi-GBM Subthreshold Trigger list ³³ (with reliability flag not equal to 2), the Swift GRB Archive, ³⁴ and the Gamma-Ray Coordinates Network archives ³⁵ for an associated GRB between the last ZTF nondetection (January 27.54) and the first ZTF detection (January 28.28). There were no GRBs coincident with the position and time of AT 2020blt. ³⁶

The position of AT 2020blt was visible ³⁷ to GBM only 65% of the time: 27% of the time it was occulted by the Earth, and 8% of the time GBM was not observing due to a South Atlantic Anomaly passage. By contrast, KONUS-Wind is in interplanetary space, not Earth orbit, and therefore had complete coverage. KONUS-Wind found no detection with a 90% confidence upper limit on the peak flux of 1.7 × 10⁻⁷ erg cm⁻² s⁻¹ for a typical long-GRB spectrum ³⁸ (Ridnaia et al. 2020). At the distance of AT 2020blt, this corresponds to an upper limit on the isotropic gamma-ray luminosity of L_γ,iso < 1.3 × 10⁵² erg s⁻¹.

Overall, the IPN detects bursts with a 50–300 keV fluence of 1–3 × 10⁻⁶ erg cm⁻² at 50% efficiency. Following Cenko et al. (2013) we take 10⁻⁶ erg cm⁻² as a nominal fluence threshold and obtain a limit on the isotropic gamma-ray energy release of E_γ,iso < 7 × 10⁵² erg. We put the limit on E_γ,iso in the context of classical GRBs in Section 3.

As shown in Figure 1, the light curve of AT 2020blt has a clear break well-described by a broken power law. Optical afterglows with "classical" breaks like this were commonly observed prior to the Swift era (Kulkarni et al. 1999; Harrison et al. 2001; Klose et al. 2004; Zeh et al. 2006), so it was a surprise when relatively few such breaks were detected in the X-ray afterglows of Swift GRBs (Gehrels et al. 2009). Suggestions for why breaks are rarely detected include that observations do not extend long enough after the burst time (Dai et al. 2008), that the breaks are present in the data but missed in fitting (Curran et al. 2008), that bursts are viewed from a range of viewing angles (Zhang et al. 2015), and that Swift GRBs are on average more distant (Gehrels et al. 2009) and less energetic (Kocevski & Butler 2008). Furthermore, in X-ray as well as optical bands, the search for breaks can be complicated by the presence of flares or rebrightening episodes (e.g., Kann et al. 2010).

To make a direct comparison to afterglows in the literature with breaks (Zeh et al. 2006; Kann et al. 2010; Wang et al. 2018) we fit the light curve using a conventional smooth broken power law, modifying it to take into account the fact that we do not know the burst time t₀:

$$\begin{eqnarray}&&\begin{array}{l}m(t)=-2.5{\mathrm{log}}_{10}\\ \times \ {\left({10}^{-0.4{m}_{c}}\left[\displaystyle \frac{{\left(t-{t}_{0}\right)}^{{\alpha }_{1}n}}{{t}_{b}}+\displaystyle \frac{{\left(t-{t}_{0}\right)}^{{\alpha }_{2}n}}{{t}_{b}}\right]\right)}^{-1/n}.\end{array}\end{eqnarray} \tag{ 1 }$$

In Equation (1), m(t) is the apparent magnitude as a function of time, n parameterizes the smoothness of the break (where n = ∞  is a sharp break), α₁ is the power-law index before the break, α₂ is the power-law index after the break, t_b is the time of the break, and m_c is the magnitude at the time of the break assuming n = ∞. Note that the original equation also includes terms for the underlying supernova and the host galaxy, which we take to be zero—a reasonable assumption given that we do not observe any flattening in the optical light curve and the SN would not be detectable in our optical observations at this redshift.

First we fit Equation (1) to the r-band light curve, because it has the most extensive temporal coverage and we cannot necessarily assume constant colors across the optical light curve. Using the Levenberg–Marquardt algorithm implemented in scipy we find parameters that are very poorly constrained: m_c  = 20.99 ± 5.01 mag, t₀ = 2458876.69 ± 0.42, t_b  = 1.00 ± 1.84 days, α₁ = 0.52 ± 2.81, and α₂ = 2.59 ± 0.26. The fit has a reduced χ²/ν = 3.6/ν, where ν = 1 is the number of degrees of freedom (number of data points minus number of fitted parameters).

In Section 2 we show that constant colors are a reasonable assumption at optical frequencies. So, to obtain more precise parameters we fit Equation (1) to the g-, r-, and i-band light curves simultaneously, assuming constant g − r and r − i offsets. The result is m_c,r  = 20.96 ± 0.53 mag, m_c,g  = 21.50 ± 0.53 mag, m _c,i  = 20.66 ± 0.53 mag, t₀ = 2458876.68 ± 0.01, t_b  = 1.00 ± 0.10 days, α₁ = 0.54 ± 0.26, and α₂ = 2.62 ± 0.01. The smoothing parameter is still poorly constrained: n = 4 ± 67. Note that we do not assume a single spectral index across the optical band because the g-band flux is slightly attenuated by the Lyα absorption feature and the Lyman forest (from the spectrum we estimate that it is attenuated at the 15% level, corresponding to about 0.2 mag). The fit has a reduced χ²/ν = 11.8/ν = 1.31, where ν = 10 is the number of degrees of freedom. Throughout the paper, we use the parameters resulting from this multiband fit, which results in a best-fit light curve shown in the left panel of Figure 1.

The best-fit t₀ is January 28.18 ± 0.01, which is 2.4 hr before the first detection and 15.6 hr after the last nondetection. The best-fit t_j  = 1.00 ± 0.10 d after t₀ (observer-frame) is typical of optical afterglows with breaks (e.g., Zeh et al. 2006; Kann et al. 2010; Wang et al. 2018). In Figure 4 we show the resulting value of Δα = 2.08 ± 0.26 compared to the distribution in Zeh et al. (2006) and Kann et al. (2010). The value of Δα appears to be large compared to afterglows in the literature, but not unprecedented.

Zoom In Zoom Out Reset image size

The origin of breaks in afterglow light curves is still debated. A leading hypothesis is that a break results from a collimated jet (Rhoads 1997; Sari et al. 1999). The traditional argument is that while Γ(t) ≫ θ⁻¹, the emission cannot be distinguished from an isotropic outflow, because relativistic beaming confines the viewing angle to a small region that is expanding too quickly to interact sideways. As Γ(t) decreases to Γ(t) ∼ θ⁻¹, two effects become important: the jet begins expanding sideways (Rhoads 1997), and the edge of the jet becomes visible (Mészáros & Rees 1999). However, "textbook" achromatic breaks are rarely observed, and simulations suggest that breaks can be chromatic (van Eerten et al. 2011) and that sideways expansion can take place significantly later than when the edge of the jet becomes visible (Panaitescu et al. 1998; Granot & Piran 2012).

If the break in the light curve of AT 2020blt is a jet break—and we caution that it is rare to see breaks that actually behave in the way one would expect for jet breaks, e.g., Liang et al. (2008)—we can use the timing of the break to estimate the opening angle of the jet θ₀. For a constant-density ISM we have from Sari et al. (1999) that

$$\begin{eqnarray}&&{t}_{\mathrm{jet}}\approx 6.2{({E}_{52}/{n}_{1})}^{1/3}{({\theta }_{0}/0.1)}^{8/3}\,\mathrm{hr},\end{eqnarray} \tag{ 2 }$$

where E₅₂ is the kinetic energy release of the explosion in units of 10⁵² erg, θ₀ is in radians, and n₁ is the ambient density in units 1 cm⁻³. Using our rest-frame value t_jet = 0.26 ± 0.03 d (6.24 ± 0.62 hr) we have

$$\begin{eqnarray}&&1.0\pm 0.1={({E}_{52}/{n}_{1})}^{1/3}{({\theta }_{0}/0.1)}^{8/3}\,\mathrm{hr}.\end{eqnarray} \tag{ 3 }$$

From the GRB search in Section 2.6 we have an approximate upper limit on the isotropic gamma-ray energy release of E₅₂ < 7. The E₅₂/n₁ term has a much weaker dependence than the opening angle term, and can be reasonably estimated to be unity based on typical GRB environment densities (Chandra & Frail 2012). We find an opening angle of θ₀ = 0.10 ± 0.04 = 5.7 ± 2.3 degrees, typical of opening angles inferred from optical jet breaks (Panaitescu & Kumar 2001; Zeh et al. 2006; Wang et al. 2018).

If the break is due to the material spreading sideways, then the temporal index after the break is F_ν (t) ∝ t^−p (Sari et al. 1999), where p is the power-law index of the electron energy distribution. Using the value of α₂ above we have p = α₂ = 2.62 ± 0.01, which is large but consistent with values expected for shock acceleration (Jones & Ellison 1991) given the uncertainties, and furthermore is within the normal range of values inferred from optical afterglows in the literature (Wang et al. 2018).

If the steepening is due solely to detecting the edge of the jet, the expected post-break slope ³⁹ is the slope of a spherically expanding outflow (t^{−3(p−1)/4};  Sari et al. 1999) with two additional powers of Γ ∝ t^−3/8. The resulting temporal slope is F_ν (t) ∝ t^−3p/4, so 3p/4 = α₂ = 2.62 ± 0.01. The value of p = 3.49 ± 0.01 is larger than what is predicted for shock acceleration (Jones & Ellison 1991) so in what follows we assume p = 2.62 ± 0.01.

The spectrum of an afterglow is determined by the kinetic energy of the explosion, the ambient density, and the fraction of the energy in magnetic fields _B and relativistic electrons _e (Sari et al. 1998). The spectrum is characterized by several break frequencies: the cooling frequency ν_c , the characteristic frequency ν_m , and the self-absorption frequency ν_a . The spectral index in any region of the spectrum therefore depends on physical properties of the explosion and on the position of the observing frequency relative to the break frequencies.

In Section 3.1 we found p = 2.62 ± 0.01 based on the post-break light-curve power-law index. Because the g-band measurement is attenuated by Lyα and Lyman forest absorption, we use the r − i color from the WASP observation (r − i = 0.34 ± 0.09 mag) to estimate that the optical spectral index β_opt = 1.3 ± 0.4. Such a steep spectral index is only ever observed as a result of absorption (Cenko et al. 2009; Greiner et al. 2011): at this redshift even i-band is well into the far-ultraviolet, so it takes relatively little extinction to significantly alter the flux and color. If β_opt were the "true" (unextincted) spectral index, that would indicate that the cooling frequency ν_c lies below the optical bands (Sari et al. 1999). Taking ν_c  < 10¹⁴ Hz at t_d  ≈ 0.5 d we have, following Sari et al. (1998)

$$\begin{eqnarray}&&26\gt {\epsilon }_{B}^{-3/2}{E}_{52}^{-1/2}{n}_{1}^{-1}.\end{eqnarray} \tag{ 4 }$$

Using _B  < 3 × 10⁻⁴ and E₅₂ = 1 (Section 3.3) we find a very large CSM density of n > 7 × 10³ cm⁻³, which would violate the assumption that we made in estimating the jet opening angle of E₅₂/n₁ ≈ 1. Furthermore, as shown in the right panel of Figure 1, the optical to X-ray spectral index is β_opt,X  ≈ 0.7, inconsistent with β_opt. It seems most natural that the optical spectral index is steepened by dust attenuation, and β_opt,X is the "true" spectral index, rather than β_opt.

The value of β_opt,X is consistent with (p − 1)/2 but not with p/2, so the cooling frequency ν_c lies above the X-ray band (Sari et al. 1999). For adiabatic evolution we have (Sari et al. 1998)

$$\begin{eqnarray}&&{\nu }_{c}=2.7\times {10}^{12}{\epsilon }_{B}^{-3/2}{E}_{52}^{-1/2}{n}_{1}^{-1}{t}_{d}^{-1/2}\,\mathrm{Hz}.\end{eqnarray} \tag{ 5 }$$

where t_d is the time (in days) after the explosion. Taking ν_c  > 10¹⁸ Hz and t_d  = 0.5 days (time of the X-ray observation; rest-frame), we have

$$\begin{eqnarray}&&2.6\times {10}^{5}\lt {\epsilon }_{B}^{-3/2}{E}_{52}^{-1/2}{n}_{1}^{-1}.\end{eqnarray} \tag{ 6 }$$

Using _B  < 3 × 10⁻⁴ and E₅₂ = 1 (Section 3.3) we find n < 0.7 cm⁻³, which is typical for GRBs (Panaitescu & Kumar 2001; Chandra & Frail 2012).

In Section 3.1 we showed that the optical light curve of AT 2020blt is fairly typical for classical GRBs. However, the radio light curve of AT 2020blt (left panel of Figure 1) decays steeper than F_ν  ∝ t^−2.3, which is unusual for GRBs with detected radio afterglows in general (Chandra & Frail 2012), including PTF11agg (Cenko et al. 2013).

Early fast-evolving emission in GRB radio afterglow light curves can arise from reverse shocks or diffractive scintillation in the interstellar medium (Laskar et al. 2013; Perley et al. 2014; Laskar et al. 2016; Alexander et al. 2017; Laskar et al. 2018; Alexander et al. 2019). To determine whether scintillation could be the origin in this case, we use the NE2001 model of the ISM (Cordes & Lazio 2002). For context, scintillation results from small-scale inhomogeneities in the ISM, which change the phase of an incoming wave front. As the Earth moves, the line of sight to a background source changes, so the net effect is an observed change in flux. The effect is greatest for sources observed at a frequency ν_obs that is close to the transition frequency ν₀, which separates strong scattering (ν_obs < ν₀) from weak scattering (ν_obs > ν₀).

Using the NE2001 map, we determine that the line of sight toward AT 2020blt has a transition frequency ν₀ = 7.12 GHz, which is close to our observing frequency. Furthermore, we can estimate the timescale for flux changes. Using D = 100 pc as the characteristic scale height of the ISM, and λ = 3 cm as our observing wavelength, the Fresnel length is ${r}_{f}=\sqrt{\lambda D}\,\approx {10}^{10}$ cm. Assuming that Earth moves at v = 30 km s⁻¹, we obtain t ∼ r_f /v ≈ 1 hr. In conclusion, the flux could easily change by an order of magnitude due to scintillation over the large time window (16 days, observer-frame) between our observations. Note that the timescale is close to our time on-source (∼30 minutes), so there could be some damping of the scintillation over the course of our observation. However, the signal-to-noise of the data is not high enough for us to search for scintillation within the observation.

If, on the other hand, the rapid change in flux is due to a truly steep power-law decay in the radio emission, there would be implications for the ambient density and the value of _B . In particular, the characteristic frequency ν_m must lie below the radio band. For adiabatic evolution we have (Sari et al. 1998)

$$\begin{eqnarray}&&{\nu }_{m}=5.7\times {10}^{14}{\epsilon }_{B}^{1/2}{\epsilon }_{e}^{2}{E}_{52}^{1/2}{t}_{d}^{-3/2}\,\mathrm{Hz}.\end{eqnarray} \tag{ 7 }$$

Requiring ν_m  < 10 GHz (39 GHz rest-frame) at t_d  = 2 days (time of the radio detection; rest-frame) and adopting _e  = 0.1 (Kumar & Zhang 2015; Beniamini & van der Horst 2017) we find

$$\begin{eqnarray}&&0.02\gt {\epsilon }_{B}^{1/2}{E}_{52}^{1/2}.\end{eqnarray} \tag{ 8 }$$

Assuming E₅₂ = 1, we find _B  < 3 × 10⁻⁴, which is also typical for GRBs based on high-energy and optical afterglow modeling (Kumar & Zhang 2015; Beniamini & van der Horst 2017). So, although an early steep-decaying radio light curve is unusual for GRBs with detailed radio observations, we have no reason to believe that the radio behavior of AT 2020blt is unusual for the population of GRBs as a whole.

Here we consider the possibility that AT 2020blt was a classical GRB viewed slightly outside the jet opening angle. Beniamini & Nakar (2019) argued that the vast majority of GRBs observed so far must have been observed close to or within the jet core, implying that GRB emission is not produced efficiently away from the core. So, as discussed in Section 1, there is a natural expectation for X-ray and optical afterglows without detected GRB emission (Mészáros & Rees 1997; Rhoads 1997; Nakar & Piran 2003). The slightly off-axis model has been invoked to explain low-luminosity GRBs or X-ray flashes (Ramirez-Ruiz et al. 2005) as well as plateaus observed in X-ray afterglow light curves (Eichler & Granot 2006; Beniamini et al. 2020a).

One signature of a slightly off-axis afterglow could be an early shallow decay and a large value of Δα (Ryan et al. 2020; Beniamini et al. 2020b). This can be understood as follows. In on-axis events, the early stage of the light curve is set by two competing effects: the shock is decelerating, but the beaming cone is widening to include more material. In a slightly off-axis event, there is a third effect, which is that the beaming cone widens to include material of increasing energy per solid angle—hence a shallower decay.

As shown in Figure 4, we did observe a large value of Δα in AT 2020blt. A larger number of events would help to test this hypothesis: the luminosity function of the early afterglow should be different from the luminosity function of directly on-axis afterglows, and the distribution of limits on E_iso would eventually make it unlikely that the afterglows are drawn from the same population as classical GRBs. With more events, we could hope to make the first measurement of the optical beaming factor in GRB afterglows (Nakar & Piran 2003).

Here we consider the possibility that AT 2020blt was a "dirty fireball," i.e., a jet with lower Lorentz factor (Γ ∼ 10) that did not produce any GRB emission, as proposed for PTF11agg (Cenko et al. 2013). The basis for the dirty fireball argument for PTF11agg was the rate: at the time, it seemed that the rate of PTF11agg-like events may have been significantly higher than the rate of classical GRBs (Cenko et al. 2013), although this was later shown to not be the case (Cenko et al. 2015; Ho et al. 2018). Taking a similar approach to Cenko et al. (2015) and Ho et al. (2018), we searched high-cadence (6× per night) ZTF survey data (Bellm et al. 2019b) from 2018 March 1 to 2020 May 12 to estimate the areal exposure in which an event like AT 2020blt would have passed our filter.

We folded the light curve of AT 2020blt through all r-band exposures in the ZTF high-cadence fields, varying the burst time by 0.01 day intervals, to see over what duration the transient would have had two r-band detections above the limiting magnitude, with a first detection over one magnitude brighter than the last nondetection. We found a total exposure of 855 field-nights. We assume a 100% detection efficiency, so our result is somewhat of a lower limit, particularly at these fainter magnitudes; the efficiency as a function of limiting magnitude has not yet been characterized for ZTF. The 92 high-cadence survey fields included in our search have a combined footprint of 3307 deg² after removing the overlap between fields. So, we estimate the all-sky rate of transients similar to AT 2020blt to be

$$\begin{eqnarray}\begin{array}{rcl}{ \mathcal R } & \equiv & \displaystyle \frac{{N}_{\mathrm{rel}}}{{A}_{\mathrm{eff}}}\\ & = & \displaystyle \frac{1}{30,734\,{\deg }^{2}\,\mathrm{days}}\times \displaystyle \frac{365.25\,\mathrm{days}}{\mathrm{year}}\times \displaystyle \frac{41,253\,{\deg }^{2}}{\mathrm{sky}}\\ & = & 490\,{\mathrm{yr}}^{-1},\end{array}\end{eqnarray} \tag{ 9 }$$

with a 68% confidence interval from Poisson statistics of 85–1611 yr⁻¹. For comparison, the all-sky rate of Swift GRBs out to z = 3 has been estimated to be ${1455}_{-112}^{+80}\,{\mathrm{yr}}^{-1}$ (Lien et al. 2014). The Swift GRB rate is larger than the rate of optical afterglows, since only a subset of GRBs show bright optical afterglow emission (Cenko et al. 2009). So, within the uncertainties, the rate of optical afterglows in ZTF is compatible with the GRB rate. We therefore concur with the conclusion in Cenko et al. (2015) and Ho et al. (2018) that there is no evidence for an afterglow-like phenomenon that is significantly more common than classical GRBs.

The light curve of a dirty fireball will take longer to rise to peak because it is set by the time it takes the shock to sweep up material of mass 1/Γ₀ times the ejecta mass (the "deceleration" time). For a uniform-density medium, the expression (observer-frame) is

$$\begin{eqnarray}&&{t}_{\mathrm{dec}}=30\,{E}_{53}^{1/3}{n}^{-1/3}{{\rm{\Gamma }}}_{0,2.5}^{-8/3}\,\sec .\end{eqnarray} \tag{ 10 }$$

So, an outflow with Γ₀ = 100 will have an afterglow that rises to peak in 300 s, but an outflow with Γ₀ = 10 will have an afterglow that rises to peak in 1.2 days. The power-law index of the rising light curve will remain the same, however. We searched for GRBs within three days prior to the discovery of AT 2020blt, but found no coincident bursts. So, we have no evidence for such a long rise time in AT 2020blt.