Final Moments. I. Precursor Emission, Envelope Inflation, and Enhanced Mass Loss Preceding the Luminous Type II Supernova 2020tlf

SN 2020tlf was first reported to the Transient Name Server by ATLAS (Tonry et al. 2018b) on 2020 September 16, but the earliest detections of the SN are from the Young Supernova Experiment (YSE; Jones et al. 2021) with the PS1 telescope (Kaiser et al. 2002) on 2020 September 5. YSE began monitoring the field in which SN 2020tlf was discovered on 2020 January 18.

YSE data is initially processed by the Image Processing Pipeline (IPP), described in Magnier et al. (2013), including difference imaging and photometry. Those data are passed to the Transient Science Server (Smith et al. 2020), where catalog cross-matching and machine-learning tools are used to identify potential transients in each image. The YSE team performs manual vetting of potential transients to remove artifacts, asteroids, and other contaminating sources, and finally sends new transient discoveries and initial photometric epochs to the Transient Name Server for follow-up by the community. We then load the transient data into YSE's transient management system, "YSE-PZ," which allows us to view Pan-STARRS data with that of other ongoing surveys and schedule follow-up observations. Further detail on this procedure is given in Jones et al. (2021) and references therein.

This process allows for identification and follow-up of fast-rising transients. For SN 2020tlf, we re-measured the pre-explosion photometry using Photpipe (Rest et al. 2005) to ensure highly accurate photometric measurements that took into account pixel-to-pixel correlations in the difference images and host galaxy noise at the SN location. Photpipe is a well-tested pipeline for measuring SN photometry and has been used to perform accurate measurements from Pan-STARRS in a number of previous studies (e.g., Rest et al. 2014; Foley et al. 2018; Jones et al. 2018, 2019; Scolnic et al. 2018). In brief, Photpipe takes as input IPP images that have been re-sampled and astrometrically aligned to match skycells in the PS1 sky tessellation and measures their zero-points by using DoPhot (Schechter et al. 1993) to measure the photometry of stars in the image and comparing to stars in the PS1 DR2 catalog (Flewelling et al. 2016). Then, Photpipe convolves a template image from the PS1 3π survey (Chambers et al. 2017) with data taken between the years 2010 and 2014, using a kernel that consists of three superimposed Gaussian functions, to match the point-spread function (PSF) of the survey image, and it subtracts the template from the image. Finally, Photpipe uses DoPhot again to measure fixed-position photometry of the SN at the weighted average of its location across all images. Further details regarding this procedure are given in Rest et al. (2014) and Jones et al. (2019).

To account for the bright host galaxy of SN 2020tlf, which could cause larger-than-expected pre-explosion photometric noise in the difference image (Kessler et al. 2015; Doctor et al. 2017; Jones et al. 2017), we estimate the noise in the photometry by adding the Poisson noise at the SN location in quadrature to the standard deviation of fluxes measured in random difference-image apertures at coordinates with no pre-SN (or SN) light but approximately the same underlying host galaxy surface brightness as exists at the SN location. These apertures are placed in an annulus at the same elliptical radius from the center as SN 2020tlf to ensure similar surface brightness to the SN location. We find that the SN host galaxy does not contribute significantly to the uncertainty in the photometry (≲15% of the total error budget). We can also rule out contributions from a possible active galactic nucleus in NGC 5731 to fluxes at the SN location given the significant offset of SN 2020tlf from host center.

Based on the above data reduction, we find evidence for a statistically significant (>3σ) pre-explosion flux excess at the SN location (m ≈ 20.7–21.9 mag) in riz-bands from MJD 58971.42–59097.24 (δ t = −127.3 to −1.49 days before first light). However, we find no evidence for similar pre-explosion emission in the YSE g-band images from δ t = −232.1 to −17.49 days before first light. We present the pre-explosion griz-band stacked PS1 images in Figure 2 over the phase range of δ t = −169.7 to −3.7 days before first light (MJD 58929-59095). The multiband, pre-explosion PS1 light curve is displayed in Figure 3. Furthermore, there is no evidence for significant flux in earlier pre-explosion PS1 3π survey imaging of the SN site from 2011 February 28 to 2014 February 21 (δ t = −3478 to −2389 days before first light). For PS1 griyz-bands, we derived 3σ upper limits over this pre-explosion phase range of >22.24, >22.28, >22.02, >21.50, and >21.75 mag, respectively.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Pre-explosion imaging of SN 2020tlf was also acquired by the ZTF (Bellm et al. 2019; Graham et al. 2019) and ATLAS (Tonry et al. 2018b). ZTF g/r-band photometry was obtained through the ZTF forced-photometry service (Masci et al. 2019) and covers a phase range of δ t = −900.4 to −34.5 days before first light. We follow the procedure outlined in the ZTF forced-photometry manual to apply a signal-to-noise threshold (SNT) of 3 to the data, i.e., all photometry with signal-to-noise ratio (S/N) > 3 are considered >3σ detections. After the SNT is applied, we find evidence for tentative pre-explosion ZTF r-band flux (m ≈ 21.2 mag) ranging from δ t = −128.4 to −51.50 days since first light. To further test the validity of these "detections," we downloaded the public difference-image pre-explosion data from the Infrared Processing and Analysis Center ²¹ and performed the same random background aperture analysis on the images as discussed in Section 2.1. We find evidence for >3σ emission in only one epoch of r-band ZTF data at a phase δ t = −56.5 days prior to first light. This ZTF r-band detection is consistent with the PS1 detections and is presented in the pre-explosion light-curve plot (Figure 3(a)). Additionally, there is no evidence for detectable emission of pre-explosion flux in the ZTF g-band images (m ≥ 20.7 mag).

Furthermore, we do not find evidence for significant emission in c/o-band ATLAS pre-explosion photometry during the phase range of δ t = −1714.1 to −6.5 days since first light. Similar to the YSE/PS1 pre-explosion image analysis described above, we model the background noise by placing random apertures near the explosion site and performing aperture photometry of these regions. The flux is then recorded in each of these random background apertures for each pre-explosion epoch and used to create background light curves, i.e., control light curves. To attempt and measure significant pre-SN flux detections at the location of SN 2020tlf, we apply several cuts on the total number of individual as well as averaged data in order to remove bad measurements. Our first cut uses the χ² and uncertainty values of the PSF fitting to clean out bad data. We then obtain forced photometry of eight control light curves located in a circular pattern around the location of the SN with a radius of 17''. The flux of these control light curves is expected be consistent with zero within the uncertainties, and any deviation from that would indicate that there are either unaccounted systematics or underestimated uncertainties.

We search for such deviations by calculating the 3σ cut weighted mean of the set of control light-curve measurements for a given epoch (for a more detailed discussion, see Rest et al. 2021, in preparation). This weighted mean of these photometric measurements is expected to be consistent with zero and, if not, we flag and remove those epochs from the pre-SN light curve. This method allows us to identify potentially bad measurements in the SN light curve without using the SN light curve itself. We then bin the SN 2020tlf light curve by calculating a 3σ cut weighted mean for each night (typically, ATLAS has four epochs per night), excluding the flagged measurements from the previous step. We find that this method successfully removes bad measurements that can mimic pre-SN emission (S. Rest et al. 2021, in preparation). We then calculate the rolling sum of the S/N with a Gaussian kernel of 30 days for the pre-SN and the control light curves and identify any significant flux excess in the rolling sum. The kernel size of 30 days is chosen to maximize the detection of pre-SN emission with similar timescales. We use the peaks in the control light curves as our empirical detection limit: since there is no transient in the control light curves (barring an extremely unlikely coincidence with a transient unrelated to pre-SN emission at the location of SN 2020tlf), any peaks in the control light curves are false positives. We choose as our conservative detection limit a rolling sum value of 20, and we find no evidence of pre-SN activity in SN 2020tlf down to a magnitude limit of m ≳ 20.3 mag, which is consistent with PS1 and ZTF detections.

We started observing SN 2020tlf with the Ultraviolet Optical Telescope (UVOT; Roming et al. 2005) on board the Neil Gehrels Swift Observatory (Gehrels et al. 2004) on 2020 September 9 until 2021 February 18 (δt = 11.0–165.2 days since first light). We performed aperture photometry with a 5'' region with uvotsource within HEAsoft v6.26, ²² following the standard guidelines from Brown et al. (2014). In order to remove contamination from the host galaxy, we employed images acquired at t ≈ 165 days after first light, assuming that the SN contribution is negligible at this phase. This is supported by visual inspection in which we found no flux associated with SN 2020tlf. We subtracted the measured count rate at the location of the SN from the count rates in the SN images following the prescriptions of Brown et al. (2014). We detect bright UV emission from the SN near optical peak (Figure 4) until t ≈ 60 days after explosion. Subsequent non-detections in the w1, m2, and w2 bands indicate significant cooling of the photosphere and/or Fe-group line blanketing.

Zoom In Zoom Out Reset image size

Additional griz-band imaging of SN 2020tlf was obtained through the YSE sky survey (Jones et al. 2021) with the Pan-STARRS telescope (PS1; Kaiser et al. 2002) between 2020 September 8 and 2021 June 26 (δ t = 1.5–292.3 days since first light). The YSE photometric pipeline is based on photpipe (Rest et al. 2005). Each image template was taken from stacked PS1 exposures, with most of the input data from the PS1 3π survey. All images and templates are re-sampled and astrometrically aligned to match a skycell in the PS1 sky tessellation. An image zero-point is determined by comparing PSF photometry of the stars to updated stellar catalogs of PS1 observations (Chambers et al. 2017). The PS1 templates are convolved with a three-Gaussian kernel to match the PSF of the nightly images, and the convolved templates are subtracted from the nightly images with HOTPANTS (Becker 2015). Finally, a flux-weighted centroid is found for each SN position, and PSF photometry is performed using "forced photometry": the centroid of the PSF is forced to be at the SN position. The nightly zero-point is applied to the photometry to determine the brightness of the SN for that epoch.

SN 2020tlf was observed with ATLAS (δ t = −9.40–157.8 days since first light), a twin 0.5 m telescope system installed on Haleakala and Mauna Loa in the Hawai'ian islands that robotically surveys the sky in cyan (c) and orange (o) filters (Tonry et al. 2018b). The survey images are processed as described in Tonry et al. (2018b) and photometrically and astrometrically calibrated immediately (using the RefCat2 catalog; Tonry et al. 2018a). Template generation, image subtraction procedures, and identification of transient objects are described in Smith et al. (2020). PSF photometry is carried out on the difference images, and all sources greater than 5σ are recorded. All sources go through an automatic validation process that removes spurious objects (Smith et al. 2020). Photometry on the difference images (both forced and non-forced) is from automated PSF fitting as documented in Tonry et al. (2018b). The photometry presented here are weighted averages of the nightly individual 30 s exposures, carried out with forced photometry at the position of SN 2020tlf.

We observed SN 2020tlf with the Las Cumbres Observatory Global Telescope Network 1 m telescopes and Las Cumbres Observatory imagers from 2020 September 21 to 2021 March 29 (δ t = 14.34–203.5 days since first light) in ugri-bands. We downloaded the calibrated BANZAI (McCully et al. 2018) frames from the Las Cumbres archive and re-aligned them using the command-line blind astrometry tool solve-field (Lang et al. 2010). Using the photpipe imaging and photometry package (Rest et al. 2005; Kilpatrick et al. 2018), we regridded each Las Cumbres Observatory frame with SWarp (Bertin 2010) to a common pixel scale of 0389 centered on the location of SN 2020tlf. We then performed photometry on these frames with DoPhot (Schechter et al. 1993) and calibrated each frame using PS1 DR2 standard stars observed in the same field as SN 2020tlf in ugri bands (Flewelling et al. 2016).

Observations of SN 2020tlf were obtained with the 1 m Lulin telescope located at Lulin Observatory on 2020 October 9 (δ t = 32.71 days since first light) in BVgr bands. The individual frames were corrected for bias and flat-fielded using calibration frames obtained on the same night and in the same instrumental configuration. Within photpipe, we solved for the astrometric solution in each frame using Two Micron All Sky Survey astrometric standards (Cutri et al. 2003) observed in the same field as SN 2020tlf. Finally, we performed photometry in each frame following the same procedures for Las Cumbres Observatory described above.

For both Las Cumbres Observatory and Lulin photometry, we re-processed the final light curve by calculating the mean astrometric position of SN 2020tlf in all Las Cumbres Observatory and Lulin frames separately. We then performed forced photometry using a custom version of DoPhot at this position using the PSF parameters in each individual frame and solving only for the flux of SN 2020tlf at the time.

The complete light curve of SN 2020tlf is presented in Figure 4, and all photometric observations are listed in Appendix Table A4. In addition to our observations, we include g/r-band photometry from the ZTF (Bellm et al. 2019; Graham et al. 2019) forced-photometry service (Masci et al. 2019), which span from 2020 November 27 to 2021 June 28 (δ t = 81.81–294.5 days since first light).

The Milky Way (MW) V-band extinction and color excess along the SN line of site are A_V = 0.043 mag and E(B −V) = 0.014 mag (Schlegel et al. 1998; Schlafly & Finkbeiner 2011), respectively, which we correct for using a standard Fitzpatrick (1999) reddening law (R_V = 3.1). In addition to MW color excess, we estimate the contribution of galaxy extinction in the local SN environment. We use Equation (9) in Poznanski et al. (2012) to convert the Na i equivalent width (EW) of 0.10 ± 0.010 Å in the first SN 2020tlf spectrum to an intrinsic E(B −V) and find a host galaxy extinction of E(B −V)_host = 0.018 ± 0.003 mag, also corrected for using the Fitzpatrick (1999) reddening law.

In Figure 5, we present the complete series of optical spectroscopic observations of SN 2020tlf from −9 to +257 days relative to the B-band maximum (δ t = 10–270 days relative to first light). A full log of spectroscopic observations is presented in Appendix Table A1.

Zoom In Zoom Out Reset image size

SN 2020tlf was observed with Shane/Kast (Miller & Stone 1993) and Keck/LRIS (Oke et al. 1995) between −9 and +257 days relative to the B-band maximum. For all of these spectroscopic observations, standard CCD processing and spectrum extraction were accomplished with IRAF. ²³ The data were extracted using the optimal algorithm of Horne (1986). Low-order polynomial fits to calibration-lamp spectra were used to establish the wavelength scale, and small adjustments derived from night-sky lines in the object frames were applied. We employed custom IDL routines to flux calibrate the data and remove telluric lines using the well-exposed continua of the spectrophotometric standard stars (Wade & Horne 1988; Foley et al. 2003). Details of these spectroscopic reduction techniques are described in Silverman et al. (2012).

Spectra of SN 2020tlf were also obtained with Keck NIRES and DEIMOS, as well as Binospec on MMT and the Dual Imaging Spectrograph (DIS) on the Astrophysical Research Consortium (ARC) 3.5-m telescope at Apache Point Observatory (APO). All of the spectra were reduced using standard techniques, which included correction for bias, overscan, and flat-field. Spectra of comparison lamps and standard stars acquired during the same night and with the same instrumental setting have been used for the wavelength and flux calibrations, respectively. When possible, we further removed the telluric bands using standard stars. Given the various instruments employed, the data-reduction steps described above have been applied using several instrument-specific routines. We used standard IRAF commands to extract all spectra.

The X-Ray Telescope (XRT; Burrows et al. 2005) on board the Swift spacecraft (Gehrels et al. 2004) started observing the field of SN 2020tlf on 2020 September 9 until 2021 February 18 (δt = 11.0–165.2 days since first light) with a total exposure time of 35.2 ks, (Source IDs 11337 and 11339). We analyzed the data using HEAsoft v6.26 and followed the prescriptions detailed in Margutti et al. (2013), applying standard filtering and screening using the latest CALDB files (version 2021008). We find no evidence for significant X-ray emission in any of the individual Swift-XRT epochs, nor in merged images near optical/UV peak and at all observed phases. From the complete merged image, we extracted an X-ray spectrum using XSELECT ²⁴ at the source location with a 35'' source region (100'' background region) and estimated the count-to-flux conversion by fitting an absorbed simple power-law spectral model with Galactic neutral H column density of 1.25 × 10²⁰ cm⁻² (Kalberla et al. 2005) and spectral index Γ = 2 using XSPEC (Arnaud 1996). Using a merged, 0.3–10 keV XRT image around UV peak (δt = 11.0–23.0 days since first light), we derive 3σ upper limits on the count rate, unabsorbed flux and luminosity of <3.9 × 10⁻³ ct s⁻¹, <1.7 × 10⁻¹³ erg s⁻¹ cm⁻², and <2.6 × 10⁴⁰ erg s⁻¹, respectively. These limits assume no intrinsic absorption from material in the local SN environment, e.g., n_H,host = 0. This n_H,host value is chosen so as to provide the most conservative upper limit on X-ray emission despite the host reddening of E(B −V)_host = 0.018 mag derived from optical spectra (Section 3.1).

We acquired deep radio observations of SN 2020tlf with the Karl G. Jansky Very Large Array (VLA) at δ t = 146–320 days since first light through project SD1096 (PI Margutti). All observations have been obtained at 10 GHz (X band) with 4.096 GHz bandwidth in standard phase referencing mode, with 3C 286 as a bandpass and flux-density calibrator and QSO J1224+21 (in A and B configuration) and QSO J1254+114 (in D-configuration) as complex gain calibrators. The data have been calibrated using the VLA pipeline in the Common Astronomy Software Applications package (CASA; McMullin et al. 2007) v6.1.2 with additional flagging. SN 2020tlf is not detected in our observations. We list the inferred upper limits on the flux densities in Appendix Table A2.

The complete post-explosion, multiband light curve of SN 2020tlf is presented in Figure 4, and pre-explosion gcroiz-band light curves are displayed in Figure 3. We define the time of first light as the average phase between the last photometric detection at the pre-SN flux threshold (M ≈ −12 mag) and the first multicolor detections that rose above that flux threshold (M ≲ −12 mag). This yields a time of first light of ${t}_{\exp }=59098.7\pm 1.5$, which is then used for reference through the analysis. We discuss potential uncertainties on this time when modeling the bolometric light curve (e.g., Section 6). We fit a third-order polynomial to the SN 2020tlf light curve to derive a peak absolute B-band magnitude of M_B = −18.5 ± 0.04 mag at MJD 59117.6 ± 1.5, where the uncertainty on peak magnitude is the 1σ error from the fit, and the uncertainty on the peak phase is the same as the error on the time of first light. Using the adopted time of first light, this indicates a rise time of t_r = 18.9 ± 1.5 days with respect to the B-band maximum.

As shown in Figure 11(b), we compare the r/V-band light-curve evolution of SN 2020tlf to popular SNe II discovered within a few days of explosion, many of which showed photoionization features in the early-time spectra, e.g., SNe 1998S (Leonard et al. 2000; Fassia et al. 2001; Shivvers et al. 2015), 2013fs (Yaron et al. 2017), 2014G (Terreran et al. 2016), 2017ahn (Tartaglia et al. 2021), and 2020pni (Terreran et al. 2021). Compared to these SNe, the peak r/V-band absolute magnitude of SN 2020tlf is more luminous than that of SNe 2013ej, 2013fs, 2017ahn, and 2020pni, but less luminous than SNe 1998S and 2014G at peak. While the r/V-band rise time near maximum light is similar to SN 1998S, SN 2020tlf was discovered at an even earlier phase with a fainter detection absolute magnitude of ∼−13.5 mag. The linear photometric evolution of SN 2020tlf during its photospheric phase is comparable to most of these objects. However, SN 2020tlf has the longest lasting plateau, extending out to ∼110 days after maximum light, suggesting a larger ejecta mass and/or larger stellar radius than other SNe II with early-time signatures of CSM interaction.

We construct a bolometric light curve by fitting the ZTF, PS1, Las Cumbres Observatory, ATLAS, and Swift photometry with a blackbody model that is dependent on radius and temperature. The extremely blue UV colors and early-time color evolution of SN 2020tlf near maximum light impose non-negligible deviations from the standard Swift-UVOT count-to-flux conversion factors. We account for this effect following the prescriptions by Brown et al. (2010). Each SED was generated from the combination of multicolor UV/optical/near-IR (NIR) photometry in the w2, m2, w1, u, b/B, v/V, g, c, o, r, i, and z bands (1500–10000 Å). In regions without complete color information, we extrapolated between light-curve data points using a low-order polynomial spline. We present SN 2020tlf's pre- and post-explosion bolometric light curve in addition to its blackbody radius and temperature evolution in Figure 6. All uncertainties on blackbody radii and temperature were calculated using the co-variance matrix generated by the SED fits. At the time of first spectrum with photoionization emission features, the blackbody radius, temperature, and luminosity are R_BB = (1.5 ± 0.21) × 10¹⁴ cm, T_BB = (3.8 ± 0.65) × 10⁴ K, and L_bol = (3.4 ± 1.4) × 10⁴³ erg s⁻¹, respectively. This R_BB is technically the radius of thermalization (τ > 1), which is much smaller than the photospheric radius (τ = 1; Dessart & Hillier 2005), and the assumption of a pure blackbody is not strictly accurate (see, e.g., Dessart et al. 2015). Consequently, this can lead to the reported luminosities from blackbody fitting to be possible lower limits on the true bolometric luminosity of SN 2020tlf.

Zoom In Zoom Out Reset image size

As shown in Figure 7, we model the post-plateau (t > 120 days after maximum light) bolometric light-curve evolution with energy injection from pure radioactive decay of newly synthesized ⁵⁶Co. The complete analytic formalism behind this model is outlined in Valenti et al. (2008), Wheeler et al. (2015), and Jacobson-Galán et al. (2021). From this modeling, we derive a total ⁵⁶Co mass of M_Co = (2.7 ± 0.070) × 10⁻² M_⊙ and a γ-ray trapping timescale of t_γ = 261.1 ± 9.57 days. The inferred ⁵⁶Co mass is lower than other SNe II with early-time photoionization signatures, e.g., SN 2014G (∼0.06 M_⊙; Terreran et al. 2016) or SN 1998S (∼0.15 M_⊙; Fassia et al. 2001). While the late-time light-curve evolution is consistent with energy injection from the radioactive decay of ⁵⁶Co, there are possibly small, but overall negligible, contributions from additional power sources at these phases such as CSM interaction. Furthermore, the nebular spectra of SN 2020tlf (e.g., Figure 5(b)) show typical O i and Ca ii emission, which is compatible with ⁵⁶Co decay being absorbed by the metal-rich inner ejecta rather than late-time power coming from the outer ejecta ramming into CSM, as observed in SN 1998S.

Zoom In Zoom Out Reset image size

The complete spectroscopic sequence of SN 2020tlf from −8.6 to +257.4 days since maximum light is presented in Figure 5. In the earliest spectrum, SN 2020tlf shows narrow, symmetric emission features of H i, He i, He ii, N iii, and C iii (FWHM < 300 km s⁻¹). As shown in Figure 8, this spectrum is nearly identical to the early-time spectrum of SN 1998S at a phase of +3 days since first detection (−10 days from B-band peak; Fassia et al. 2001). However, the time of first light in SN 1998S is relatively uncertain given that the last non-detection was 8 days prior to the first detection, indicating that the phase of this spectrum could be later than +3 days. Based on our adopted time of first light, SN 2020tlf is at a later phase of +10 days since first light (−8.6 days relative to the B-band peak), despite the overall spectral similarity. This could indicate that the true time of first light for SN 2020tlf is actually later than estimated, that first light emission from SN 2020tlf was detected at earlier phases given the depth of PS1 compared to the instruments used to discover SN 1998S (plus the uncertainty on the time of first light for SN 1998S), or that the environment around each of the two SNe is different, i.e., variations in the properties of the most local CSM or intrinsic extinction from the SN host galaxies.

Zoom In Zoom Out Reset image size

In Figures 8(b)/(c), we present velocity comparison plots of H i and N iii + He ii emission profiles for SN 2020tlf and SN 1998S. The SN 1998S high-resolution spectrum is from Shivvers et al. (2015) and all line velocities can be resolved, unlike in the SN 2020tlf LRIS spectrum. Nonetheless, while line velocities in the SN 2020tlf LRIS spectrum can only be resolved to ≲300–400 km s⁻¹ and ≲200 km s⁻¹ in the APO DIS spectrum at the same phase, the overall similarity of the narrow features in SN 2020tlf compared to SN 1998S indicates that the wind velocities of CSM around the SN 2020tlf progenitor may be comparable to those of the CSM in SN 1998S. To test this, we convolve the high-resolution SN 1998S spectrum to the instrumental resolution of the SN 2020tlf LRIS spectrum and find that the narrow Balmer series emission components in this spectrum, as well as those in the SN 1998S LRIS spectrum, can be modeled with a similar Lorentzian profile velocity (∼300–400 km s⁻¹) as observed in the SN 2020tlf spectrum. Therefore, based on the SN 1998S spectra, it is possible that the H-rich CSM in SN 2020tlf is moving at ∼50 km s⁻¹ (e.g., Figure 8(b)) and other CSM ions such as He ii or N iii (e.g., Figure 8(c)) are moving with wind velocities of ∼90–120 km s⁻¹. We present additional modeling of these photoionization line profiles in Figure 9 using combined Lorentzian profiles. Narrow components of each profile in the LRIS spectrum can only be resolved to FWHM ≲300 km s⁻¹ and FWHM ≲200 km s⁻¹ in the APO DIS spectrum, but the broad components of the profiles resulting from electron scattering (e.g., Chugai 2001; Dessart et al. 2009) are fit using Lorentzian profiles with FWHM ∼ 2000–3000 km s⁻¹. Based on the comparison to SN 1998S and the Lorentzian profile fits, we conclude that the SN 2020tlf progenitor likely had a wind velocity of v_w ∼ 50–200 km s⁻¹. For the N iii + He ii feature shown in Figure 9(b), we explore the possibility of blueshifted, Doppler broadened He ii from the SN ejecta being present in the line profile, in addition to the narrow He ii and N iii profiles derived from the wind. This specific combination of Lorentzian profiles is consistent with the overall profile shape as well as the flux excess on top of continuum emission, blueward of the N iii + He ii feature. Doppler broadened He ii from the SN ejecta has been proposed as an explanation for blue flux excesses in SNe II-P that do not show spectral signatures of CSM interaction (Dessart et al. 2008).

Zoom In Zoom Out Reset image size

In Figure 10, we compare the continuum-subtracted spectrum of SN 2020tlf to other well-studied events with photoionization features such as SNe 1998S, 2014G, 2013fs, 2017ahn, and 2020pni (Fassia et al. 2001; Terreran et al. 2016, 2021; Yaron et al. 2017; Tartaglia et al. 2021). The H i, He ii, and N iii emission lines present in the early-time spectrum of SN 2020tlf are similar to those found in most other objects. SN 2020tlf differs slightly from SN 2013fs in that it does not contain high-ionization lines such as O iv–vi, which indicates a more extended CSM and thus lower ionization temperature for SN 2020tlf (Dessart et al. 2017). SNe 2013fs and 2014G also do not have detectable N iii, unlike SN 2020tlf, SN 1998S, 2020pni and 2017ahn, which have clear N iii emission in the double-peaked N iii + He ii feature. Furthermore, SN 2020tlf does not have significant C iv or N iv emission like most other objects, with the exception of SN 2013fs.

Zoom In Zoom Out Reset image size

We also compare the mid-time spectra (δ t = +40–60 days since peak) of this sample to the second spectrum of SN 2020tlf at +76 days since B-band peak, which was obtained once the SN was visible to ground-based observatories (Figure 11(a)). At this phase, the SN is in its recombination phase, with strong signatures of line blanketing by metals in the H-rich ejecta and a red spectrum. Overall, SN 2020tlf has similar ions to other events, e.g., strong Balmer series, Fe-group, and O i and Ca ii profiles. However, absorption profiles in SN 2020tlf are noticeably narrower than in other objects, which could be due to the later phase and/or larger R_⋆ or lower E_k /M_ej. The SN 2020tlf spectrum is still photospheric at +76 days (+95 days since explosion) and contains a bluer continuum with weaker line blanketing compared to SNe II at similar epochs. This could indicate persistent energy injection from a more extended envelope or additional CSM interaction powering the SN at this phase. Additionally, we compare the IR spectrum of SN 2020tlf at +127 days post-peak to IR spectra of SNe 1998S, 2013ej, 2017eaw at a similar phase in Figure 12. All four SNe show similar ions at this phase such as prominent H emission, Fe-group elements, and Mg i. Additionally, the IR spectrum of SN 2020tlf appears to show evidence for CO emission, similar to that confirmed in SN 2017eaw by Rho et al. (2018).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We present the late-time spectra of SN 2020tlf in Figure 5(b) over a phase range of δ t = 153–277 days since first light. At these phases, SN 2020tlf displays strong emission lines such as Hα, [O i] λλ6300, 6364 [Ca ii] λλ 7291, and 7323 emission. The SN appears to not be fully nebular by the +277 days post-explosion as it still shows Hα and Fe-group element absorption profiles. However, some of these line transitions are optically thick and can exhibit a P-Cygni profile during the nebular phase when the continuum optical depth is low.

To constrain the zero-age main-sequence (ZAMS) mass of the SN 2020tlf progenitor, we compare the late-time spectra to nebular-phase radiative-transfer models that have, in other SN II studies, shown that the [O i] emission profile is a direct tracer of progenitor mass. In Figure 15(a), we compare the nebular-phase models from Jerkstrand et al. (2014) for 12–19 M_⊙ progenitors to SN 2020tlf at +250 days post-explosion. We find that at this phase, the 12 M_⊙ model best reproduces the nebular transitions observed in SN 2020tlf. We also compare the +277 day spectrum of SN 2020tlf to the nebular models from Dessart et al. (2021) that are generated from 9.5 to 15 M_⊙ progenitors at +350 days post-explosion and find that the 10 M_⊙ model is the most consistent with the data. We therefore conclude that the progenitor of SN 2020tlf had a ZAMS of ∼10–12 M_⊙. The estimated SN 2020tlf progenitor mass is comparable to that derived from nebular emission in sample studies of SNe II-P (∼12–15 M_⊙; Silverman et al. 2017), but lower than that of other SNe II with photoionization spectra, e.g., SN 2014G had an estimated progenitor ZAMS mass of 15–19 M_⊙ (Terreran et al. 2016). We note that we cannot completely rule out the possibility that the progenitor of SN 2020tlf was a low-mass (∼9 M_⊙) super-asymptotic giant branch star, as was proposed to be the progenitor of electron-capture SN candidates (e.g., see Hiramatsu et al. 2021). However, based on the observed bolometric light-curve evolution and total synthesized ⁵⁶Ni mass, it is unlikely that SN 2020tlf was an electron-capture SN from such a progenitor star.

SN 2020tlf is the first SN with typical SN II-P/L–like spectral and light-curve behavior that has a confirmed detection of precursor flux. Precursor emission was also identified ∼60 yr prior to SN II, iPTF14hls in archival imaging (Arcavi et al. 2017). However, while the spectral evolution of iPTF14hls resembles a normal SN II, the extremely long-lasting and time variable light-curve evolution indicated that this event, as well as its progenitor star, were very different than standard SN II explosions. The pre-explosion light curve, presented in Figure 3(a), shows >3σ detections in PS1 riz-bands starting from δ t = −130 days and persisting with a consistent flux until first SN light. The lack of precursor detections in bluer bands such as the PS1/ZTF g band or ATLAS c band suggests a moderately cool emission or an extended, low-temperature emitting surface of whatever physical mechanism caused this pre-explosion flux. We construct a pre-explosion bolometric light curve, as well as temperatures and radii, by modeling the SED containing 3σ riz-band detections and g-band upper limits with a blackbody model, the same as that used in Section 5.2. We show the pre-explosion bolometric light curve, blackbody temperatures, and radii in Figure 13(b). It should be noted that the pre-SN bolometric light curve relies on only three optical/NIR bands, and thus contributions from undetected parts of the blueward (or IR) ends of the SED could cause variations from what is observed. Furthermore, the presence of spectral emission lines during the precursor (e.g., Hα) could lead to increased flux in the r band, for example, relative to other bands. We find that the precursor has a bolometric luminosity of ∼10⁴⁰ erg s⁻¹ (∼2 × 10⁶ L_⊙), and has an average blackbody temperature and radius of ∼5000 K and ∼10¹⁴ cm (∼1500 R_⊙), respectively. For reference, we also plot the predicted luminosity, surface temperature, and radius evolution of a 15 M_⊙ RSG progenitor undergoing wave-driven mass loss as presented in Fuller (2017). This model has a consistent emitting radius to the SN 2020tlf precursor emission, but has significantly lower luminosities and temperatures at phases where pre-explosion emission is detected.

Zoom In Zoom Out Reset image size

The pre-SN activity prior to SN 2020tlf is considerably fainter than in other SNe with confirmed precursor emission. In Figure 14, we compare the multicolor SN 2020tlf pre-explosion detections to popular SNe IIn, 2009ip, 2010mc, and 2016bhu, all of which had confirmed precursor emission prior to explosion. As shown in the plot, the SN 2020tlf precursor only reaches ∼−11.5 mag in all bands, while the plotted SNe IIn precursors have absolute r-band magnitudes ranging from −13 to −15.5 mag. Precursor emission from the SN 2020tlf progenitor system is also fainter than the average absolute magnitude of −13 mag found in the sample of ZTF-observed SNe IIn with pre-explosion outbursts presented by Strotjohann et al. (2021). However, as shown in Figure 14, because the limiting magnitude of ZTF (<20.5 mag; Bellm et al. 2019; Graham et al. 2019) is ∼1 mag shallower than YSE (<21.5 mag; Jones et al. 2021), pre-explosion emission in SNe II–like events would not have been detected at the flux level of the precursor of SN 2020tlf. Nevertheless, searches for pre-SN emission from SN II progenitors at closer distances (e.g., ≲50 Mpc) in transient survey archival data (e.g., ZTF, ATLAS, YSE, etc.) will allow us to determine whether more 20tlf-like precursor events are possible.

Zoom In Zoom Out Reset image size

Integrating the pre-explosion bolometric light curve yields a total radiated energy of ∼10⁴⁷ erg over the ∼130 day precursor event. Coincidentally, this derived radiated energy is approximately the binding energy of an H-rich envelope in a typical RSG (Dessart et al. 2010). We explore potential power sources for the precursor emission in the form of CSM interaction-powered and wind-driven emission. For the former, the precursor emission would result from interaction between material ejected in a progenitor outburst and CSM from a previous outburst and/or steady-state wind-driven mass loss, causing a fraction of the kinetic energy to be converted to radiative energy. In this process, the relation between radiated and kinetic energy, as well as CSM properties, goes as:

$$\begin{eqnarray}&&{E}_{\mathrm{rad}}=\displaystyle \frac{\epsilon }{2}{M}_{\mathrm{pre}}{v}_{\mathrm{pre}}^{2}\end{eqnarray} \tag{ 1 }$$

where is the fraction of converted kinetic energy, ${M}_{\mathrm{pre}}$ is the mass ejected in the precursor, and ${v}_{\mathrm{pre}}$ is the velocity of that material. For the observed precursor radiated energy of E_rad ≈ 10⁴⁷ erg, efficiency = 1, and velocities discussed in Section 5.3 (e.g., v_w = 50–200 km s⁻¹), the total mass ejected in the precursor is ${M}_{\mathrm{pre}}\approx 4.3\mbox{--}0.27\,{M}_{\odot }$, respectively. However, if CSM interaction is the mechanism for precursor emission, the conversion efficiency is definitely much less than 100% (Smith et al. 2010), and therefore the derived ${M}_{\mathrm{pre}}$ is at least ≳ M_⊙ for the largest ${v}_{\mathrm{pre}}$ that is consistent with observations. Furthermore, it should be noted that material ejected in a precursor that then collides with preexisting CSM may lead to formation of a semi-static CSM shell of constant density (i.e., s = 0), which is different than the wind-like density CSM that is typically invoked to model events with photoionization spectra (e.g., see Section 6).

If the precursor emission from the SN 2020tlf progenitor was instead from a super-Eddington, continuum-driven wind, we follow the mass-loss prescription outlined in Shaviv (2001a) that goes as:

$$\begin{eqnarray}&&{E}_{\mathrm{rad}}\approx \displaystyle \frac{1}{W}{M}_{\mathrm{CSM}}{c}_{s}c\end{eqnarray} \tag{ 2 }$$

where W is an empirical factor found to be ∼5, c_s is the speed of sound at the base of the optically thick wind (e.g., ∼60 km s⁻¹; Shaviv 2001b), and c is the speed of light. For E_rad ≈ 10⁴⁷ erg, we derive a total amount of material lost in a potential super-Eddington wind to be M_CSM ≈ 2 × 10⁻³ M_⊙. However, it should be noted that this formalism is designed for SN IIn progenitors such as LBVs. Furthermore, a super-Eddington wind is likely unphysical for a 10–12 M_⊙ progenitor mass range as derived from the nebular spectra of SN 2020tlf.

Another possible mechanism to explain the pre-SN activity in SN 2020tlf is stellar interaction between the primary RSG progenitor and a smaller binary companion star. This can manifest as a "common envelope" phase in the progenitor's evolution (Sana et al. 2012), which can result in the merging of primary and binary companions, the result of which is a slightly luminous, short-lived transient (Kochanek et al. 2014). While this scenario has been invoked as an explanation for luminous red novae or intermediate luminosity optical transients, the resulting luminosity produced by this physical mechanism appears to be too faint (∼10²⁻⁴ L_⊙; Pejcha et al. 2017) to match the pre-explosion luminosity in SN 2020tlf (∼10⁶ L_⊙). Therefore, it is more likely that an eruption from the primary progenitor alone is the most likely cause of the pre-SN activity observed in SN 2020tlf.

Pre- and post-explosion panchromatic observations have provided an unprecedented picture of the SN 2020tlf progenitor system. In Figure 20, we attempt to combine inferences made from observation and modeling to create a visualization of the explosion and surrounding progenitor environment. Our model is a snapshot of the SN at the time of first light and contains physical scales and parameters such as distance, velocity, and composition estimates. The illustration also includes progenitor properties derived from precursor emission in the ∼130 days leading up to SBO.

Zoom In Zoom Out Reset image size

As discussed in Section 6, CMFGEN modeling of the SN 2020tlf light curve and photoionization spectrum indicate that the 10–12 M_⊙ (ZAMS; e.g., see Figure 15 and Section 5.3) progenitor star had radius of ∼1100 R_⊙ and was losing mass at an enhanced rate of $\dot{M}={10}^{-2}\,{M}_{\odot }$ yr⁻¹ in the final months before explosion, leading to the creation of dense CSM (shown in sea foam green; Figure 20) at distances r ≲10¹⁵ cm; lower-density CSM extended out to r ≈ 8 × 10¹⁵ cm. These models suggest that the SN 2020tlf progenitor star had a total CSM mass of ∼0.05–0.07 M_⊙ in the local environment at the time of explosion. At the time of the photoionization spectrum (δ t ≈ 10 days post-explosion), SN 2020tlf had a blackbody temperature T ≈ 3.7 × 10⁴ K at the thermalization depth and an emitting radius of ∼2 × 10¹⁴ cm (shown in light blue; Figure 20). The identification of narrow emission lines from photoionized material in the earliest spectrum confirms that the CSM comprised high-ionization species such as He ii, N iii, and C iii–iv, as well as lower ionization species such as H i and He i. As observed in the photoionization spectrum, the wind velocity of the CSM is likely v_w ≈ 50–200 km s⁻¹.

Prior to explosion, the SN 2020tlf progenitor star produced detectable precursor emission for ∼130 days prior to SBO. The observed emission is relatively constant leading up to explosion (∼10⁴⁰ erg s⁻¹), with an average emission radius and temperature of ∼10¹⁴ cm and ∼5000 K, respectively (shown in red; Figure 20). Because the blackbody radius rate of change during the pre-SN activity is ∼1000 R_⊙ over a timescale of ∼30 days, it is likely that the observed pre-SN emission is not derived from the stellar surface; the Kelvin–Helmholtz timescale for a ∼10 M_⊙ progenitor to change in radius at this rate is τ_th ≳ 200 days. As discussed in Section 5.4, this precursor emission could have resulted from the ejection, and subsequent CSM interaction, of >0.3 M_⊙ of stellar material that was most local to the progenitor star (shown in dark blue; Figure 20). However, this estimated mass of precursor material is larger than the CSM mass of ∼0.05–0.07 M_⊙ in the most consistent CMFGEN models. There is also a possibility that the precursor emission arose from a super-Eddington wind that drove off >10⁻³ M_⊙. However, this mass-loss mechanism may be unphysical for the low-mass progenitor of SN 2020tlf.

An open question in understanding the pre-explosion activity of the SN 2020tlf progenitor star is whether material ejected in the detected precursor is the same CSM responsible for the photoionization spectrum at ∼10 days post-explosion. The validity of this conclusion is dependent on what wind velocity we adopt in the range of possible CSM velocities (∼50–200 km s⁻¹) derived in Section 5.3. If the precursor material was ejected with a velocity of v_w ≈ 50–200 km s⁻¹, that specific CSM could reach radii of r ≈ (0.6–2.4) × 10¹⁴ cm in the ∼140 days before the photoionization spectrum was obtained. However, if the material was driven off from the surface of a progenitor star with an extended radius of ∼1100 R_⊙, the distance reached by this material in ∼140 days increases to r ≈ (1.4–3.2) × 10¹⁴ cm. These distances are consistent with the blackbody radius of ∼2 × 10¹⁴ cm at the time of the photoionization spectrum. Therefore, unless the wind velocities are <50 km s⁻¹, it is feasible that the material driven off to cause the precursor emission is the same CSM material that was photoionized by the SN shock wave, resulting in the narrow emission lines present in the early-time spectrum.

The detection of precursor emission, combined with the presence of dense CSM (e.g., see Section 6) around the ∼10–12 M_⊙ progenitor of SN 2020tlf necessitates a physical mechanism for enhanced mass loss and luminosity, together with a likely structural change to the stellar envelope (inflation), in the final year to months before core-collapse. As shown in Section 5.4, powering the precursor emission would require >0.3 M_⊙ of material through CSM interaction and >10⁻³ M_⊙ of material via a super-Eddington wind, the latter of which is much smaller than the CSM mass derived from light-curve and spectral modeling (e.g., Section 6). However, a super-Eddington wind is most likely unphysical given the small progenitor ZAMS mass derived from the nebular spectra; it will also lead to larger CSM densities than those derived from modeling (Section 6). Therefore, in the final ∼year of stellar evolution, a physical mechanism is needed to produce enhanced mass loss (e.g., 0.01 M_⊙ yr⁻¹ derived from modeling) and detectable precursor flux.

As discussed initially in Section 5.4, wave-driven mass loss is one process that occurs in late-stage stellar evolution that could lead to the ejection of material from the progenitor surface, also resulting in detectable pre-explosion emission. The excitation of gravitational waves by oxygen or neon burning in the final years before SN can allow for the injection of energy (e.g., ∼10^46–48 erg) into the outer stellar layers, resulting in an inflated envelope and/or eruptive mass-loss episodes (Meakin & Arnett 2007; Arnett et al. 2009; Quataert & Shiode 2012; Shiode & Quataert 2014; Fuller 2017; Wu & Fuller 2021). While this mass-loss mechanism is a potential explanation for the precursor activity in SN 2020tlf, there are currently no wave-driven models that can match the observed pre-explosion activity. As shown in Figure 13(b), the model for a 15M_⊙ RSG undergoing wave-driven mass loss by Fuller (2017) does not reproduce the bolometric luminosities of the SN 2020tlf precursor, but is consistent in radius in the final ∼130 days before core-collapse. In an updated study of wave-driven models, Wu & Fuller (2021) showed that pronounced pre-SN outbursts could occur in progenitor stars of similar mass to that of SN 2020tlf (e.g., <14 M_⊙). However, the timescales of these mass-loss episodes are inconsistent with relatively constant emission observed in the SN 2020tlf precursor in the final ∼130 days before explosion.

A related, promising explanation for enhanced mass loss is the sudden deposition of energy into the internal layers of a massive star outlined by Dessart et al. (2010). Agnostic to the mechanism for energy injection, these models show that a release of energy (E_dep) that is on the order of the binding energy of the stellar envelope (E_bind) will create a shock front that will propagate outwards, causing a partial ejection of the stellar envelope. As shown in Figures 8 and 9 in Dessart et al. (2010) for an 11 M_⊙ progenitor, energy injection of E_dep ∼ E_bind will produce a detectable pre-SN outburst that is continuous for hundreds of days and matches the observable in the SN 2020tlf precursor, e.g., L ≈ 10⁶ L_⊙, T ≈ 5000 K, and R ≈ 1500 R_⊙. Possible causes for such energy release could be gravitational waves from neon/oxygen burning or even a silicon flash in the final 100–200 days before explosion. For the latter, Woosley & Heger (2015) show that low-mass progenitors (9–11 M_⊙) can produce precursor emission in the final ∼ year before explosion as a result of silicon deflagration in their cores. Specifically, the 10.0C progenitor model listed in Table 3 of Woosley & Heger (2015) has consistent pre-SN properties to those observed in the SN 2020tlf precursor, e.g., L ≈ 10⁴⁰ erg s⁻¹ and R ≈ 10¹⁴ cm. Overall, the simulations from both of these studies are promising scenarios to explain the enhanced mass loss observed in SN 2020tlf.

SN 2020tlf represents the first instance of an SN II where significant variability has been detected in the RSG progenitor star prior to explosion. These observations reveal a clear disjuncture from the findings by other studies that examined the pre-SN activity of SN II progenitors in the final years before core-collapse. For example, the progenitor behavior prior to SN II-P, 2017eaw has been studied extensively using pre-explosion UV/optical/IR imaging in the final decades before explosion (Kilpatrick & Foley 2018; Rui et al. 2019; Tinyanont et al. 2019; Van Dyk et al. 2019). However, the ∼11–13 M_⊙ RSG progenitor of SN 2017eaw only reached a luminosity of ∼4.7 L_⊙ prior to explosion (Kilpatrick & Foley 2018), with IR variability estimated to be at most Δν L_ν ≈ 5000 L_⊙ (Tinyanont et al. 2019); both of these progenitor luminosity estimates are orders of magnitude lower than the precursor recorded prior to SN 2020tlf. Similar quiescent behavior is also observed in sample studies on the long-term variability of SN II progenitors by Johnson et al. (2018) as well as the single object study of SN II-P, ASASSN-16fq by Kochanek et al. (2017). Based on the findings of the former, the SN 2020tlf progenitor lies in the <37% of RSGs that exhibit extended outbursts after O ignition, i.e., ∼1000–100 days before explosion, depending on the progenitor mass. Furthermore, Kochanek et al. (2017) and Johnson et al. (2018) both find that these SN II progenitors show very little variability (e.g., Δν L_ν ≲ 3000 L_⊙) for years to days before core-collapse. Interesting, none of these SNe II showed spectroscopic evidence of interaction with CSM shed by the progenitor during episodes of enhanced mass loss, as detected directly in the earliest spectrum of SN 2020tlf. This may indicate that only RSG progenitors with CSM that is dense enough to be detectable in early-time spectra of young SNe II are also able to produce luminous precursor emission of ∼10⁶ L_⊙, as observed prior to SN 2020tlf.