It is difficult to establish the properties of massive stars that explode as supernovae1,2. The electromagnetic emission during the first minutes to hours after the emergence of the shock from the stellar surface conveys important information about the final evolution and structure of the exploding star3,4,5,6. However, the unpredictable nature of supernova events hinders the detection of this brief initial phase7,8,9. Here we report the serendipitous discovery of a newly born, normal type IIb supernova (SN 2016gkg)10, which reveals a rapid brightening at optical wavelengths of about 40 magnitudes per day. The very frequent sampling of the observations allowed us to study in detail the outermost structure of the progenitor of the supernova and the physics of the emergence of the shock. We develop hydrodynamical models of the explosion that naturally account for the complete evolution of the supernova over distinct phases regulated by different physical processes. This result suggests that it is appropriate to decouple the treatment of the shock propagation from the unknown mechanism that triggers the explosion.
On 2016 September 20 (dates are given in ut throughout), amateur astronomer V.B. was testing a camera mounted on his 40-cm Newtonian telescope. He pointed the telescope at NGC 613, a spiral galaxy at a distance of 26.4 Mpc, because at that time it was located near the zenith. Over approximately 1.5 h he imaged the galaxy with a clear filter while a supernova was being born, using 20-s exposures to avoid saturation caused by the bright city sky. An initial series of 40 images obtained during 20 min showed no sign of the supernova. From the combined image we obtained a 5σ detection limit of 19.4 mag converted to the V band (see Methods). When observations resumed, after an interval of 45 min, the supernova became visible. During the remaining 25 min of observations the supernova doubled its flux (Fig. 1, Methods). The deep detection limit and frequent sampling during the initial rise constitute an unprecedented set of observations for a supernova discovery. A linear fit to the discovery photometry yields a remarkably fast rise rate of 43 ± 6 mag d−1. An extrapolation of this linear rise back in time to the proposed brightness of the progenitor of V ≈ 24 mag (Methods) suggests that SN 2016gkg exploded some time between 2:50 ut and 5:35 ut. This constraint of less than 3 h on the explosion epoch is one of the most stringent available11,12,13,14,15.
The prompt discovery and announcement of SN 2016gkg10 triggered extensive monitoring that began less than one day later, including Swift X-ray, ultraviolet and optical observations16,17. This permitted excellent coverage of the subsequent evolution of the supernova. Consequently, the cooling peak, which lasted for about three days, is one of the best observed so far. Follow-up spectroscopy provided a classification of SN 2016gkg as a type IIb supernova16. Our photometric and spectroscopic monitoring, which began less than one day after discovery, is described in Methods. In addition to these observations, the Hubble Space Telescope (HST) archive contained images of the site of the supernova, obtained with the WFPC2 camera in 2001. An object was identified at the location of the supernova in all three available optical bands (F450W, F606W and F814W). Our analysis of these data and inferences about the progenitor system are given in Methods16,18.
To interpret the physical process that governed the behaviour of the supernova at the time of discovery, we plot the luminosity versus the rise rate and compare these data with available early optical observations of other supernovae and transients (Fig. 2; see ref. 9 for a similar graph in the near-ultraviolet range). The low luminosity and fast rise seen in SN 2016gkg place the discovery data in a clearly different location on the diagram compared with later observations and with data for other objects. This suggests a different physical origin for the initial rise of SN 2016gkg.
A natural explanation for the above comparison is that the first signal of SN 2016gkg corresponds to the long-sought shock-breakout phase19,20,21. Hydrodynamical simulations indicate that, although the shock-breakout signal is predominant in the X-ray and ultraviolet ranges, it has a clear manifestation in the optical range as well, characterized by an extremely rapid brightening at relatively low luminosity6,22. To test this interpretation, we performed numerical modelling of the supernova observations.
We divided the supernova modelling into two main stages23. First, the overall evolution of the supernova was modelled by adjusting the explosion energy, ejecta mass and 56Ni mass to reproduce the main light-curve peak and expansion velocities from spectral lines. In this way, we found that a model with an energy of E ≈ 1.2 × 1051 erg, an ejecta mass of Mej ≈ 3.4M⊙ (where M⊙ is the mass of the Sun) and a 56Ni mass of MNi ≈ 0.085M⊙ provided a good match to the observations (see Methods for alternative solutions). Once these parameters were constrained, the second step consisted of modelling the post-shock cooling peak. It is well known that to reproduce such a peak, an extended H-rich envelope has to be attached to the usual progenitor structure from stellar-evolution calculations. We varied the radius and mass of the envelope and arrived at a radius of Renv ≈ 320R⊙ (where R⊙ is the radius of the Sun) and a mass of Menv ≈ 0.01M⊙ for the extended envelope (see refs 16 and 17 for other estimates). Our preferred model is shown in Fig. 3 (more details in Methods). The derived parameters are in close agreement with those of normal type IIb supernovae. In particular, the progenitor of SN 2016gkg was slightly more massive and had a more extended envelope than the well-studied type IIb supernova SN 2011dh23.
Although our model was designed to match the observed cooling and radioactive peaks, it also explains without any modification the quick rise of the discovery data as being caused by emission from the breakout of the shock. Regardless of the parameters adopted, no physical process other than the breakout of the shock can produce such a fast rise. We verified this by exploring a set of hydrostatic progenitor structures and explosion parameters and comparing the rise slopes of our light-curve models during the breakout of the shock and the post-shock cooling phases (Methods). We identified the explosion energy as the dominant factor that determines the slope of the rise to the cooling peak. However, even with an energy value far beyond what is allowed by the rest of the observations, we were unable to reach a cooling peak rise rate near that of the shock-breakout phase. This result implies that different processes govern the initial rise and the cooling phase, which provides strong support for our interpretation of the early rise as the manifestation of the breakout of the shock.
A closer look at the discovery data reveals that the observed rise is more gradual than that of the model. This difference could be caused by limitations of our radiative-transfer treatment24,25, but it could also be indicative of the presence of some circumstellar material (Methods). More detailed analysis of the shock-breakout signal could potentially provide important information about the outermost progenitor structure and the physical processes that occur during the emergence of the shock. The serendipitous nature of the discovery observations and the sampling that was required highlight the difficulty of systematizing this type of finding, which has been the goal of several recent and future transient surveys, including KISS7, HiTS8, HSC-SHOOT9, KEGS (http://www.mso.anu.edu.au/kegs/) and ZTF26. We note that the chance probability of this discovery is of the order of 10−6 assuming a duration of 1 h and one supernova per century per galaxy. If we consider other factors, such as the sky conditions of the observing site and the location of the supernova away from bright host-galaxy regions, then this probability decreases by one order of magnitude.
Supernova discovery data
SN 2016gkg was discovered on 2016 September 2010 by amateur astronomer V.B. using a 406-mm Skywatcher Newtonian f = 4.4 reflector equipped with a ZWO ASI1600 MM-C camera and a clear (L) filter. Observations comprised four series of images (40, 17, 20 and 21 images), each with an exposure time of 20 s. Images were bias- and dark-subtracted, flat-fielded and aligned using the MaxIm DL software. The supernova is visible in the last three series of images at right ascension α = 01 h 34 min 14.46 s and declination δ = −29° 26′ 25.0″ (J2000), whereas a stack of the first 40 images shows no evidence of the supernova. Clear-band photometry was calibrated to standard V-band magnitudes on the basis of five nearby stars in the AAVSO Photometric All-Sky Survey (APASS) catalogue (https://www.aavso.org/apass). The location and catalogue magnitudes of the comparison stars are given in Extended Data Table 1 and Extended Data Fig. 1. We decided to transform magnitudes to the V band because of the dense follow-up coverage in that band (including Swift/UVOT observations) compared with other bands that also lie within the range of the clear filter. In addition, as explained further below, the clear band is centred similarly to the V band, which reduces the transformation error to a minimum. The results of the photometry measurements described in this section are listed in Extended Data Table 2. We note that the airmass during the complete observing time ranged between 1.00 and 1.03; thus, we expect no large systematic error in the photometry produced by differences in colour between the supernova and the comparison stars.
We estimated the detection limit on the stack of 40 images from the first series. Following ref. 29, we set the 5σ detection limit at the magnitude level at which point sources are detected with a 50% probability. To find this magnitude, we first calibrated the V-band zero point of the combined image using the DAOPHOT photometry of the comparison stars. Then, we added artificial point sources in an area around the supernova location, with varying apparent magnitudes in the range of 19.0 mag < V < 20.0 mag. We did this in groups of 1,000 artificial stars within each 0.1-mag bin. Finally, we used the daofind task in the DAOPHOT package of IRAF (the Image Reduction and Analysis Facility, distributed by the National Optical Astronomy Observatory, which is operated by the Association of Universities for Research in Astronomy (AURA) under cooperative agreement with the National Science Foundation (NSF); see http://iraf.noao.edu) to recover the artificial stars and thus determined the recovery fraction at each magnitude bin. We set the detection threshold Q in daofind as a factor of 3–4 times the background noise level. This value was based on the image scale value of p = 0.85 that results from the measured point-source full-width at half-maximum (FWHM) intensity of 2.1 pixels in the combined image (see figure 2 in ref. 29). By applying threshold values between 3 and 4, we obtained consistent fractions of recovered artificial sources as a function of magnitude. The resulting detection limit was V ≈ 19.4 mag, which is about 4.5 mag fainter than the brightness at both the cooling and nickel peaks.
To increase the signal-to-noise ratio in the last three series of images, we combined them into several groups. On the one hand, we produced a single combined image per series. On the other hand, we computed 11 averaged images from groups of 5 or 6 individual exposures (the number varied in each group owing to the different number of images in each series). Extended Data Fig. 2 shows a mosaic of the rising supernova as seen in the series of combined images. We performed point-spread function (PSF) photometry of the supernova using the DAOPHOT package. Photometric zero points were computed for each individual and combined image, using the five comparison field stars. The results are listed in Extended Data Table 2 and shown in Fig. 1. The quick initial rise is evident from both the individual-image and combined-image photometry. A linear fit to the resulting light curves yielded slopes of 37.3 ± 5.3 mag d−1 using the photometry from individual exposures, 43.4 ± 6.1 mag d−1 using 11 data points, and 47.2 ± 7.8 mag d−1 using 3 data points. The reduced χ2 is about unity in all cases, which indicates that no significant departures from a linear rise in magnitudes was seen. The relatively lower slope obtained from individual measurements, although statistically compatible, can be explained by the fact that the supernova is initially very near the detection limit of individual images. This introduces a bias towards brighter measurements in DAOPHOT, which in turn causes the slope to be slightly smaller. Combined images into groups of 5 or 6 show the supernova comfortably above the detection limit. The corresponding slope is thus more reliable in these cases.
We tested our results by performing aperture photometry. We set the aperture size to approximately the value of the FWHM of the source. Although the results presented a slight systematic difference during the second series (when the supernova is the faintest), in the sense that aperture photometry yielded brighter magnitudes, the main result was confirmed: the supernova brightened by about 40 mag d−1. Other possible sources of error in the photometry are contamination by background host-galaxy light and unaccounted variability in the shape of the PSF across the image. The host galaxy is faint and relatively smooth at the supernova location (see Extended Data Fig. 1) and so is expected to be accurately subtracted during the measurement of the supernova count rate. Moreover, the image quality is very stable throughout the complete observations (see FWHM values in Extended Data Table 2). Therefore, any contamination would be approximately constant in time and hence would not affect the observed slope. Regarding the PSF variability, we found its shape to be consistent across the image. Also, the supernova and comparison stars are all located near the centre of the image, where the shape of the PSF is very stable.
We also tested the possible effect on the measured slope that would be caused by the supernova varying in colour during the observations. At very early times, the temperature evolution can be very fast, thus changing the supernova colour. Because we transform the measured counts in the clear band to the V band, a rapid change of colour can affect the derived magnitudes. Beyond a constant error introduced by the fact that the comparison stars are expected to be redder than the supernova, the effect on the slope would arise solely from the change in supernova colour. The supernova colour during this phase is unknown. However, our models (see Methods section ‘Hydrodynamical model’) show that the supernova peaks at a temperature of about 2 × 105 K soon after explosion and cools off to about 5 × 104 K a few hours later. Assuming black-body spectra of varying temperature in the range (20–200) × 103 K, we estimated the possible variation in the calibrated V-band supernova photometry. For this purpose we calculated an approximate clear-band transmission function, including contributions from the manufacturer’s filter transmission, the quantum efficiency of the detector, two aluminium reflecting surfaces and the atmospheric transmission. This filter passband is available online (see Methods section ‘Data availability’). Synthetic (clear − V) colours varied by less than 0.05 mag in the range of temperatures considered above. Such a variation is less than 10% of the observed change in magnitude, which is below the uncertainty in the fitted slope. This means that the observed rise cannot be caused by an increasing amount of flux entering the optical range as the supernova cools down. We also estimated the size of the error introduced by the fact that the comparison stars are substantially redder than the supernova during this phase (see Extended Data Table 1). On the basis of those colours we assumed that the stars are of a spectral type between G2 and K3 and adopted ATLAS99 atmosphere models to represent their spectral energy distributions (SEDs). Synthetic colour differences between these stars and the supernova were below 0.1 mag for any assumed supernova temperature and stellar spectrum. The error is small because the clear passband, although much wider, is centred near the central wavelength of the V band. The extra flux introduced by the supernova on the blue side of the filter passband is compensated by extra flux from the stars on the red side. The correction from clear to V is thus accurate enough to the purposes of the current analysis.
V.B. and J.L.S. obtained new images of SN 2016gkg on September 21, with clear, B, V and I bands. Both observers used identical telescopes and cameras. We performed aperture photometry of the supernova and comparison stars and converted the results to the standard system. The clear-band data were transformed to V magnitudes. Extended Data Table 3 provides these magnitudes.
Follow-up B, V, R and I multiband images of SN 2016gkg were obtained with both the Katzman Automatic Imaging Telescope (KAIT)30 and the 1-m Nickel telescope at Lick Observatory. All images were reduced using a custom pipeline31. Aperture photometry was then obtained from a customized interactive data language (IDL) tool using the IDL Astronomy User’s Library. Note that, owing to the small field of view for both KAIT and the Nickel telescope, we were able to use only one appropriate nearby star from the APASS catalogue as the reference star for calibration. Its magnitudes were first transformed into the Landolt system using an empirical prescription (see http://www.sdss.org/dr7/algorithms/sdssubvritransform.html#lupton2005) and then transformed to the KAIT and Nickel natural systems. Apparent magnitudes were all measured in the KAIT4/Nickel2 natural system31,32. The final results were transformed to the standard system using local calibrators and colour terms for KAIT4 as given in table 4 of ref. 31, and updated Nickel colour terms as given in ref. 32. Extended Data Table 3 lists the resulting standard-system magnitudes from KAIT and the Nickel telescope.
Extended Data Fig. 3 shows the resulting light curves of SN 2016gkg compared with those of the well-observed type IIb supernovae SN 1993J and SN 2011dh. In this figure we also added the early-time V-band (or transformed) photometry from Atlas, ASAS, LCOGT and Swift.
Spectroscopic observations of SN 2016gkg were performed using the Kast spectrograph on the Lick Observatory 3-m Shane telescope on 2016 September 24, November 3, December 4 and December 23, the Low Resolution Imaging Spectrometer (LRIS)33 on the 10-m Keck-I telescope on 2016 September 28 and 2017 January 2, and the Deep Imaging Multi-Object Spectrograph (DEIMOS)34 on the 10-m Keck-II telescope on 2016 October 25 with both the 600 and 1,200 lines mm−1 gratings. Data were obtained at the parallactic angle35 to ensure accurate relative spectrophotometry; moreover, LRIS is equipped with an atmospheric dispersion corrector. Standard data reduction (including bias subtraction, flat-fielding and spectral extraction) was performed within IRAF. The spectra were flux calibrated via observations of spectrophotometric standard stars at similar airmasses to those for the supernova observations. The spectra are shown in Extended Data Fig. 3. Also displayed for comparison are spectra of the type IIb supernovae SN 1993J36,37 and SN 2011dh38; these spectra were all obtained from WISeREP39. We find, generally, that the SN 2016gkg spectra bear a stronger resemblance to those of SN 2011dh than to those of SN 1993J.
We produced synthetic spectra using the SYNOW code (see ref. 40 and references therein) with the aim of estimating expansion velocities as a function of phase. SYNOW provides approximate continuum and line-strength levels, using simple assumptions. It can, however, provide a robust estimate of the velocity at the photosphere from the overall fit to the observed spectrum including lines of several ionic species. Each SYNOW spectrum was computed consistently with the rest of the epochs, by keeping a smooth variation of the photospheric temperature and velocity with epoch. The derived photospheric velocities, shown in Extended Data Fig. 4b, were used to compare with the hydrodynamical models.
To analyse the supernova photometry and photospheric velocity evolution, we used a one-dimensional Lagrangian hydrodynamics code that assumes flux-limited radiation diffusion for optical photons and a one-group approximation for the non-local deposition of rays produced by radioactive decay41. The code simulates the supernova explosion by injecting energy in a hydrostatic structure and self-consistently following the shock-wave propagation inside the star, the shock breakout and the subsequent expansion of the supernova ejecta during the photospheric phase. For this work the code was adapted to include light-travel time effects and limb-darkening corrections, following the prescription of ref. 4, which are relevant only at the earliest epochs (the first 2 h) of the supernova evolution.
As initial configurations (pre-supernova models), we used hydrostatic structures from single stellar-evolution calculations. Specifically, models with pre-supernova masses of 3.3M⊙ (HE3.3), 4M⊙ (HE4), 5M⊙ (HE5) and 6M⊙ (HE6), which correspond to initial main-sequence masses of 13M⊙, 15M⊙, 18M⊙ and 20M⊙, respectively, were tested42. All of these configurations are compact hydrogen-free structures with radii of R < 3R⊙, and were evolved from He burning until core collapse, assuming solar initial abundances42. However, these pre-supernova models were later modified to take into account the presence of a tenuous hydrogen envelope (of at most 1M⊙). This model construction is necessary to reproduce successfully the two-peak morphology of the light curves and the spectra of type IIb supernovae (see more details in ref. 23).
Our first step in the modelling was to find a set of parameters, such as explosion energy (E), ejected mass (Mej), and the mass of synthesized 56Ni (MNi) and its distribution, that provided a good representation of the main bolometric light-curve peak and the photospheric velocity (vph) evolution. Owing to large uncertainties in distance, reddening, bolometric corrections, photospheric velocities, and so on, we do not attempt a statistical fit to the observations. However, our conclusions are not affected by this. The model that provides the best overall agreement with the bolometric light curve and expansion velocities, shown with a solid line in Extended Data Fig. 4a, b, corresponds to the HE5 model for an explosion energy of E = 1.2 × 1051 erg, a 56Ni mass of MNi = 0.085M⊙ and an ejecta mass of Mej = 3.4M⊙, assuming the formation of a compact remnant (Mcut) of 1.6M⊙. Extended Data Fig. 4 also shows models with lower and higher mass that resulted in worse fits to the data. Note that the HE4 model, which corresponds to Mej = 2.5M⊙, E = 1 × 1051 erg, MNi = 0.087M⊙ and Mcut = 1.5M⊙, provides a possible solution, although slightly worse than that of HE5. Indeed, models with intermediate parameters—between those of HE4 and HE5, with Mej = (2.5–3.4)M⊙, E = (1–1.2) × 1051 erg and MNi = (0.085–0.087)M⊙— are also valid. Therefore, these values can be considered as the ranges of validity for the physical parameters. An important conclusion of this analysis is that the progenitor of SN 2016gkg needs to be a relatively low-mass He star, as is commonly the case for stripped-envelope supernovae43,44,45.
Once the global parameters were set, we focused on the modelling at epochs before the onset of radioactive-decay domination. At such times, the light curve is extremely sensitive to the extent (radius) and mass of the H-rich envelope. Therefore, we modified the initial structure in the HE5 model by smoothly attaching a low-mass H-rich envelope in hydrostatic and thermal equilibrium. Note that the HE4 model could also have been used in this analysis without changing the conclusions. Different configurations were tested with various progenitor radii and envelope masses. By comparing with the observations, we were able to find our preferred configuration, denoted as the preferred model and shown in Fig. 3, corresponding to an object with an H-rich envelope of radius R = 320R⊙ and a mass of Menv = 0.01M⊙. The preferred model provides a satisfactory match to the observations. Although the agreement during the post-shock cooling peak is not perfect, the model consistently reproduces the complete supernova evolution. Extended Data Fig. 4c shows this model with a solid line, compared with models of larger (smaller) progenitor radii, which overestimate (underestimate) the luminosity during the post-shock cooling phase. From these comparisons we conclude that reasonable ranges of validity for the radius and mass of the envelope are R = (300–340)R⊙ and Menv = (0.01–0.09)M⊙. Note that this analysis was based on the V-band light curve instead of the bolometric luminosity. The reason for using the V-band data was that the earliest observations were done with that band or with a clear filter that was transformed to the V band. In addition, a bolometric correction at such early epochs is highly uncertain. Therefore, we computed theoretical V-band photometry assuming a black-body SED at the thermalization temperature23.
Previous analyses of the post-shock cooling emission of SN 2016gkg arrived at different values of R and Menv, for example, R ≈ 257R⊙ (ref. 16), R = (50–125)R⊙ (ref. 17) and R = (44–131)R⊙ (ref. 18). In these analyses an analytical approach was used, which is simpler than the hydrodynamical modelling presented here. Moreover, such analytic approaches have been proven to be wrong in the interpretation of the progenitor radius for the similar type IIb supernova SN 2011dh23,46. By contrast, a similar hydrodynamical model for the post-shock cooling emission47 of SN 2016gkg yielded a radius of R ≈ 200R⊙ and an envelope mass of Menv ≈ 0.02M⊙, which are smaller than our values. Although the hydrodynamical code used47 is similar to ours, the initial structures are not: instead, parametric models were used that allow the modification of the initial density profiles to take into account the existence of a tenuous extended envelope. This method, contrary to ours, does not ensure hydrostatic and thermal equilibrium of the external envelope. This could be a cause of the differences in the derived parameters. Several interesting conclusions can be derived from Extended Data Fig. 4, as follows.
(1) Our model indicates that the optical light curves of type IIb supernova have three peaks, rather than the double peaks usually referred to in the literature. These are the shock-breakout (SBO) peak, the post-shock cooling peak and the nickel-powered peak (see also Fig. 3).
(3) The progenitor radius has a much more noticeable effect on the decline rate than on the rise rate during the cooling peak.
(4) The larger the progenitor radius, the more luminous the cooling peak becomes and the later it appears.
To further test our SBO interpretation of the early-time data, we extensively explored parameters other than the radius that could increase the slope of the cooling peak to a similar rate to that of the SBO peak (that is, around 40 mag d−1). We found that the explosion energy has the strongest effect on increasing this slope (Extended Data Fig. 4d). However, within the range of E values allowed by the modelling of the complete supernova evolution, the cooling-peak rise is always appreciably slower than the SBO rise. This is true even in an extreme case, with an explosion energy of 5 × 1051 erg (about four times larger than our preferred model), as shown in Extended Data Fig. 5. Our analysis demonstrates that the initial rise is always steeper than the rise to the post-shock cooling peak, and that there is always a local maximum in the light curve between both phases, provided that realistic pre-supernova structures are assumed. This result gives support to our SBO interpretation.
Remarkably, we have demonstrated a model that is able to reproduce three distinct light-curve phases with very dissimilar timescales (note the logarithmic timescale in Fig. 3), assuming a standard set of parameters that fit normal type IIb supernovae. Nevertheless, a close look at the earliest phases shows that the observed rise is slower than that of the models during the SBO phase (see Extended Data Fig. 5). We thus tested if this situation could be ameliorated by adding some extra surrounding material to the initial density structure. We did not assume this material to be in hydrostatic equilibrium. It could be material ejected by the progenitor before the supernova explosion, probably in the form of a dense wind. Extended Data Fig. 5 shows a model with such circumstellar material (CSM; dashed line). Clearly, the presence of this material slows the rise during the SBO, in better concordance with the observations, without affecting the light curve at later epochs (at times of 1 day or more), owing to the small amount of matter involved. Specifically, the model presented here corresponds to a mass of 0.002M⊙ distributed out to 3 × 1013 cm, assuming a steep power-law density profile with an index of 10. This corresponds to an average mass-loss rate of 6 × 10−4M⊙ yr−1, for a wind velocity of vwind = 100 km s−1. However, we found that the exact properties of this material are not very relevant. In fact, assuming a range of other density structures with different slopes and extensions produces similar results. In particular, for a constant wind profile (with an index of r−2), we found almost the same results, but in this case the mass-loss rate increased by almost two orders of magnitude. On the basis of our tests, we can say that the mere presence of this material is enough to slow down the SBO rise, with very little dependence on its exact nature. In this context, it is noteworthy that recently there has been increasing evidence from early-time observations of normal supernovae (photometry and spectroscopy) of the existence of surrounding material in the vicinity of the progenitor, possibly produced by a dense wind or an eruption that occurred shortly before the explosion13,15,48,49.
It has been noted that non-thermal processes could have a role in shaping the light curve during the SBO phase25. This second-order effect could smooth the SBO peak. However, according to those calculations, the initial rise rate remains basically unchanged. This suggests that, even if deriving detailed CSM properties from our models would be premature, the conclusion of the SBO signal detection would remain well founded. A deeper analysis of these effects is thus left for future study.
Previous SBO claims
Some supernovae have been associated in the past with possible SBO emission. The outstanding cases of SN 2006aj and SN 2008D11,12,50,51 are worth noting. SN 2006aj was connected with a γ-ray burst (GRB) and SN 2008D was preceded by an X-ray flash. The X-ray transient in both cases was interpreted by some as being produced by the SBO. However, the peculiar characteristics of both events and the lack of a model that fully describes the hard and soft emission cast some doubt on this interpretation51,52,53. The early-time optical data for these supernovae are shown in Fig. 2. The initial optical rise of SN 2008D has a slope and luminosity similar to those of SN 2016gkg during the cooling peak. Interestingly, Swift/UVOT V-band data for SN 2006aj11 obtained within 1 h after the associated GRB show a similarly steep rise of 49 ± 16 mag d−1 as that of SN 2016gkg at discovery, although with a much larger luminosity. Re-analysis of the early-time data based on the 2015 measurements available in the Swift Optical/Ultraviolet Supernova Archive (SOUSA; ref. 54 and P. Brown, private communication) provides a smaller slope of 21 ± 12 mag d−1. Nevertheless, the behaviour of this emission may still be interpreted as an SBO, although emerging from some CSM instead of the stellar surface. It should also be noted that the rise slope of the SBO in this case may be reduced by a declining contribution from the GRB afterglow.
More recently, the case of KSN 2011d, discovered by the Kepler mission, was considered an SBO detection on the basis of an excess in the early-time optical light curve relative to a simple analytic model14. However, modifying the data binning and comparison function shows55 that there is no statistical significance for an SBO in KSN 2011d.
The SN 2016gkg site is contained in publicly available archival HST images, obtained with the Wide-Field Planetary Camera 2 (WFPC2) in bands F450W, F606W and F814W on 2001 August 21, as part of programme GO-9042. We also obtained images of the supernova itself on 2016 October 10 with HST, with the Wide-Field Camera 3 (WFC3) UVIS channel in band F555W, as part of the Target of Opportunity (ToO) programme GO-14115. The observations consisted of 24 dithered frames, each of 10-s duration; the short exposure time mitigated against possible saturation by a potentially bright supernova (we knew that the supernova brightness was increasing at the time, but not to what level). The frames were combined into a final mosaic of 240-s total exposure using AstroDrizzle56 in DrizzlePac57 within PyRAF. Adopting 13 stars in common between the WFPC2 image mosaic at F606W and the WFC3/UVIS mosaic at F555W, we astrometrically registered the two datasets with a root-mean-square (r.m.s.) uncertainty of 0.42 WFPC2/WF pixels (0.042″; see the registered images in Extended Data Fig. 6). After measuring the centroid of the supernova in the WFC3 mosaic, we found that the supernova position on the WFPC2 mosaic is (1552.47, 196.39). On the WFPC2 mosaic we measured a centroid for the progenitor candidate of (1552.63, 196.06). This is a difference of 0.37 pixels, which is within the astrometric uncertainty. We therefore consider this progenitor candidate to be solidly identified; it will not be until the candidate has vanished well after the supernova explosion that its identity will be confirmed with little doubt. We note that there is more ambiguity in the identifications of the candidates in refs 16 and 18 than in ours.
We measured photometry for the progenitor candidate running Dolphot58 on the individual ‘c0m’ WFPC2 frames after masking cosmic rays with AstroDrizzle. Although the star is relatively isolated, with little apparent background, we set FitSky = 3 (rather than FitSky = 1), because the star is only 10 pixels from the edge of the WF4 chip and much of the sky annulus would sit off the edge. We therefore also set RAper = 8, as well as InterpPSFlib = 1 using the TinyTim PSF library. We enabled charge transfer efficiency corrections by setting WFPC2UseCTE = 1 in Dolphot. This resulted in brightnesses on the Vega system of mF450W = 24.07 ± 0.16 mag, mF606W = 24.04 ± 0.07 mag and mF814W = 23.58 ± 0.14 mag. The authors of ref. 16 found 23.60 ± 0.14 mag, 23.72 ± 0.08 mag and 23.25 ± 0.14 mag, respectively, for their ‘object A’ (which we have shown is the progenitor candidate). In addition, in ref. 16, results from ref. 18 were converted from STMAG to VEGAMAG, yielding 23.42 mag, 23.10 mag and 23.32 mag, respectively, and it is noted that this object in the Hubble Source Catalogue has 23.85 ± 0.08 mag at F450W and 23.34 ± 0.05 mag at F606W, all VEGAMAG. Our measurements are brighter by 0.3–0.4 mag compared to those of ref. 16. After contacting the authors of ref. 16, we now recognize that the source of the differences is the cosmic-ray masking procedure (they used LACosmic whereas we used AstroDrizzle) and the combination of the FitSky/RAper parameters in Dolphot (we used 3/8 whereas they used 1/4).
From our photometry of the progenitor candidate, and assuming a distance modulus of 32.11 ± 0.38 mag and an extinction (Milky Way only) of AV = 0.053 mag (see Methods section ‘Supernova site extinction and metallicity’), we obtained absolute magnitudes of MF450W = −8.1 ± 0.4 mag, MF606W = −8.1 ± 0.4 mag and MF814W = −8.6 ± 0.4 mag. We performed χ2 fits of the resulting SED to stellar atmosphere models from ATLAS959 and found best-fit values of and luminosity . The fitted SED is shown in Extended Data Fig. 6. Assuming a black body, this corresponds to a progenitor radius of . Such a radius is slightly smaller than (but still compatible with) what we estimated from the hydrodynamical modelling (see refs 16, 18 and 47 for previous estimates).
We attempted to find a consistent progenitor picture based on the information from the light-curve modelling and the pre-explosion photometry. The location of the pre-explosion object in the Hertzsprung–Russell diagram (HRD) is not compatible with the endpoints of single stellar evolutionary tracks16, unless some non-standard enhanced mass loss is assumed. The relatively low progenitor mass derived from the light-curve modelling, however, goes against the possibility of large mass loss produced by an isolated star wind. A more natural scenario is that of a close binary system in which the primary star explodes as a supernova after transferring mass to its companion. This type of system allows for strong mass loss even in the case of relatively low-mass progenitors. Here we present a possible such scenario for SN 2016gkg. Our proposed model is not supposed to be a unique solution.
We used a code described60 and applied61 previously to SN 2011dh. This code has detailed and updated physical ingredients (see ref. 61 and references therein). When stars are detached, it works as a standard Henyey code. When the donor star undergoes Roche-lobe overflow62, the code computes the mass-transfer rate simultaneously with the structure of the donor star in an implicit Henyey-like, numerically stable algorithm. We neglected rotation of the components and assumed that the orbit is circularized and synchronized. We assumed that the accreting star retains a fraction β of the material transferred by the donor component—a free parameter the value of which is kept constant throughout the entire evolution. Here we considered values of β = 0.0, 0.25 and 0.50 (that is, non-conservative evolution). The material lost from the system is assumed to have the specific angular momentum of the companion star.
We found good agreement with the observations by assuming a binary progenitor with solar abundance, initial masses of 19.5M⊙ and 13.5M⊙ and an orbital period of 70 days. The primary star explodes as an supernova, with a final mass and radius of M = 4.61M⊙ and R = 183R⊙ and a final orbital period of 631 days. The total amount of hydrogen that remains in the primary is 6 × 10−3M⊙, contained in the outer approximately 0.06M⊙ of the star. The surface abundance is Xsurf = 0.21. The model stays inside the error box of Extended Data Fig. 6d for the final 14,000 yr of evolution. All of these values are almost independent of the uncertain value of β.
Qualitatively, the evolution of the progenitor of SN 2016gkg is very similar to that of SN 2011dh. For both objects, binary models are more plausible candidates than single stars, because isolated objects need very specific mass-loss rates to account for the final luminosity and effective temperature (L and Teff) indicated by the pre-supernova observations. In binary systems, the donor (progenitor) star spends almost all of its final nuclear burning stages (carbon, neon, oxygen and silicon), which last for several thousand years, under Roche-lobe-overflow conditions. This places the progenitor in a well-defined region of the HRD (Extended Data Fig. 6), inside the error box in (L, Teff) for SN 2016gkg. Thus, binary systems provide a natural reference frame for interpreting the evolution of the progenitor of SN 2016gkg.
We note that in ref. 18 a very different progenitor was proposed, with 15M⊙ + 1.5M⊙ and Pini = 1,000 days, which yielded a pre-supernova mass of 5.2M⊙. The evolutionary track in ref. 18 is completely different from ours, because it undergoes core He burning as a red supergiant, and after He exhaustion executes very large loops. This scenario requires some fine tuning of the initial conditions for the pre-supernova model to be at the observed location in the HRD; as discussed above, ours does not.
Remarkably, for the computed systems, most of the mass accreted by the companion star is gained before core He burning. Thus, there is plenty of time for the accreted mass to accommodate to the stellar structure. This implies that the companion star remains close to the zero-age main sequence in the HRD, while being over-luminous for its mass. Similarly to those for SN 2011dh, our calculations predict the existence of a hot companion to the progenitor that should remain after the explosion. The position of the companion star in the HRD depends on its final mass and therefore on β. At the moment of the explosion, the companion star is still undergoing core hydrogen burning. The presence of this object may be tested with future observations, once the supernova fades sufficiently.
Supernova site extinction and metallicity
To estimate a colour excess for SN 2016gkg, we compared its (B − V) colours with those of SN 2011dh, which shows very similar spectral evolution (see Extended Data Fig. 3). Adopting only the Galactic reddening of E(B − V) = 0.017 mag for SN 2016gkg, and a total reddening of E(B − V) = 0.074 mag for SN 2011dh38, the colour curves of both supernovae match very well. This indicates that the host-galaxy reddening for the former supernova is negligible.
To test this, we inspected the optical spectra for signatures of dust extinction by looking for interstellar absorption lines. In our highest-resolution spectrum, obtained with DEIMOS on 2016 October 25, we detected the Na i D doublet both from the Milky Way and at the redshift of the host galaxy. The equivalent width of the D1 + D2 lines was 0.16 + 0.13 Å for the Galactic component and 0.43 + 0.26 Å for the host-galaxy component. This may indicate a larger host extinction than that from the Milky Way (as assumed in ref. 17). However, the Na i D equivalent width has been shown to be a poor indicator of dust extinction63. Following ref. 63, we instead studied the diffuse interstellar band at 5,780 Å. Such a feature is not detected in our DEIMOS spectrum, with a limiting equivalent width of around 0.01 Å. This is indicative of a low host-galaxy extinction, AV < 0.05 mag.
On the basis of the colour comparison with SN 2011dh and the study of spectral lines, we decided to neglect host-galaxy extinction; our results are not affected by this assumption.
We can also estimate the metallicity more directly from the spectrum of an H ii region, at α = 01 h 34 min 14.53 s, δ = −29° 26′ 16.4″ (J2000), which is about 8.6″ nearly due north of the supernova site64. We had included the H ii region in the slit, while observing the supernova with Keck/DEIMOS on 2016 October 25 with the lower-resolution grating. Owing to the short DEIMOS slit length, for strong emission lines (in this case, H) not much spatial area exists on the spectral image for accurately estimating the overall night-sky value, which probably introduces systematic uncertainty in the strong-line flux. Nonetheless, we measured the Balmer decrement from the observed spectrum and estimate AV = 3.5 mag for the nebula. The line-of-sight Galactic foreground contribution to the extinction is comparatively low, AV = 0.053 mag (ref. 65). A high extinction is plausible, given the conspicuous presence of a counterpart of the H ii region in archival Spitzer Space Telescope data. The nebula corresponds to a luminous source at both 3.6 μm and 4.5 μm, and one of the brightest sources at 24 μm, in the outer disk of the galaxy. We corrected the spectrum for this extinction, and for an assumed recession velocity of 1,481 km s−1 (from the NASA/IPAC Extragalactic Database, NED), and show the corrected spectrum in Extended Data Fig. 6.
We enlisted the various strong-line indicators used to estimate the metallicity of extragalactic H ii regions; these lines are labelled in Extended Data Fig. 6e. We measured their fluxes from the corrected spectrum and list them in Extended Data Table 4. Unfortunately, the spectrum did not go blueward enough that we could use the well-calibrated R23 indicator or [N ii]/[O ii], which depends on the intensity of the [O ii] λ3,727 line66. Instead, we had to rely on other indicators. From the indicators R3, N2 and S2, as defined in ref. 67, and using the ‘S calibration’ (equation (6) in ref. 67), we find 12 + log(O/H) = 8.65. Considering the indicators R3, N2 and O3N2 from ref. 68, derived with the online tool at http://www.arcetri.astro.it/metallicity/, we arrive at 12 + log(O/H) = 8.7.
Culling all of these estimates and assuming a solar abundance, 12 + log(O/H) = 8.69 ± 0.05 (ref. 69), the metallicity at the SN 2016gkg site appears to be consistent with solar, and we have adopted this throughout.
The datasets analysed during this study are available from http://fcaglp.unlp.edu.ar/~gaston/data/sn2016gkg/.
We have opted not to make the supernova light-curve modelling code or the binary evolution code available because they have not been prepared for portability and lack the necessary documentation for general use. However, all optical spectra will be made available at WISeREP39.
We are grateful to P. Brown for providing information about the photometry of the early Swift/UVOT data of SN 2006aj. M.C.B. acknowledges support from the Agencia Nacional de Promoción Científica y Tecnológica (ANPCyT) through grant PICT-2015-3083 ‘Progenitores de Supernovas de Colapso Gravitatorio’ and from the Munich Institute for Astro- and Particle Physics (MIAPP) of the DFG cluster of excellence ‘Origin and Structure of the Universe’. M.C.B., G.F. and O.G.B. acknowledge support from grant PIP-2015-2017-11220150100746CO of CONICET ‘Estrellas Binarias y Supernovas’. G.F. further acknowledges support from ANPCyT grant PICT-2015-2734 ‘Nacimiento y Muerte de Estrellas Masivas: Su relación con el Medio Interestelar’. K.M. acknowledges support from JSPS KAKENHI grant 17H02864. Partial support for this work was provided by NASA through programmes GO-14115 and AR-14295 from the Space Telescope Science Institute, which is operated by AURA, Inc., under NASA contract NAS 5-26555. M.O. acknowledges support from grant PI UNRN40B531. A.V.F. is also grateful for financial assistance from the Christopher R. Redlich Fund, the TABASGO Foundation and the Miller Institute for Basic Research in Science (University of California Berkeley). We thank the University of California Berkeley undergraduate students S. Channa, G. Halevy, A. Halle, M. de Kouchkovsky, J. Molloy, T. Ross, S. Stegman and S. Yunus for their effort in collecting Lick/Nickel data, and T.d.J. for help with some of the Keck observations. The Lick and Keck Observatory staff provided excellent assistance. A major upgrade of the Kast spectrograph on the Shane 3-m telescope at Lick Observatory was made possible through gifts from William and Marina Kast as well as the Heising-Simons Foundation. Research at Lick Observatory is partially supported by a gift from Google. KAIT and its on-going operation were made possible by donations from Sun Microsystems, Inc., the Hewlett-Packard Company, AutoScope Corporation, Lick Observatory, the NSF, the University of California, the Sylvia and Jim Katzman Foundation and the TABASGO Foundation. Some of the data presented here were obtained at the W. M. Keck Observatory, which is operated as a scientific partnership among California Institute of Technology, the University of California and NASA; the observatory was made possible by financial support from the W. M. Keck Foundation. O.G.B. is a member of the Carrera del Investigador Científico de la Comisión de Investigaciones Científicas de la Provincia de Buenos Aires (CIC), Argentina.