Red Supergiants as Supernova Progenitors

Abstract

The inevitable fate of massive stars in the initial mass range of ≈8–$30M_{\odot }$ in the red supergiant (RSG) phase is a core-collapse supernova (SN) explosion, although some stars may collapse directly to a black hole. We know that this is the case, since RSGs have been directly identified and characterized for a number of supernovae (SNe) in pre-explosion archival optical and infrared images. RSGs likely all have some amount of circumstellar matter (CSM), through nominal mass loss, although evidence exists that some RSGs must experience enhanced mass loss during their lifetimes. The SNe from RSGs are hydrogen-rich Type II-Plateau (II-P), and SNe II-P at the low end of the luminosity range tend to arise from low-luminosity RSGs. The typical spectral energy distribution (SED) for such RSGs can generally be fit with a cool photospheric model, whereas the more luminous RSG progenitors of more luminous SNe II-P tend to require a greater quantity of dust in their CSM to account for their SEDs. The SN II-P progenitor luminosity range is $log(L_{\mathrm{bol}}/L_{\odot })\sim 4.0$–5.2. The fact RSGs are known up to $log(L_{\mathrm{bol}}/L_{\odot })\sim 5.7$ leads to the so-called “RSG problem”, which may, in the end, be a result of small number of available statistics to date.

1. Introduction

It had long been speculated that red supergiants (RSGs) were the progenitors of supernovae (SNe), catastrophic explosions arising from the gravitational-collapse endpoints of stars with initial, zero-age main sequence masses $M_{\mathrm{ini}}\gtrsim 8M_{\odot }$ [1] (see Ekström & Georgy in this volume). As such, the resulting SNe should be dominated by strong hydrogen lines in their spectra [2], given the massive RSG hydrogen envelope, i.e., they are SNe of Type II [3], and the light curves should exhibit an extended “plateau” in their luminosity [4,5,6]. More modern theoretical analyses have further strengthened this connection, through detailed modeling of both the SN photometric and spectroscopic evolution, e.g., [7,8,9,10,11,12,13].

One must consider the observed properties of SNe in the context of their progenitors. They are inextricably linked. Substantial heterogeneity exists in both the overall shapes and the peak luminosities of Type II SNe (SNe II) [14], with implications for diversity in progenitor properties at the time of explosion. Classically, SN II light curves were separated into II-Plateau (II-P) and II-Linear (II-L) [15], with possible spectroscopic differences exemplified by a weaker P-Cygni profile in the H$\alpha$ line [16]. More recently, the light-curve shape distinction has become less clear with significantly more data available [14,17,18], although large samples appear to indicate that SNe with higher expansion velocities have higher luminosities, higher radioactive ⁵⁶Ni masses, shorter (≲100-day) plateau durations, and more rapidly declining light curves [19,20]. At the other extreme are low-luminosity SNe II, with peak luminosities at least an order of magnitude lower than average, underluminous exponential declines in the light-curve tail, ∼10% of the average ⁵⁶Ni mass, and significantly lower expansion velocities [21,22].

Recent sophisticated modeling has impressively simulated SN II-P light curves through radiative transport of shock-deposited and radioactively powered energy through the RSG stellar ejecta [23], successfully reproducing the observations of the radiative breakout of the shock wave through the outer RSG envelope [24].

Neutrino heating, together with neutrino-driven—albeit chaotic— turbulence, appears to be the main mechanism driving explosion in massive stars, although an understanding of stellar evolution to core collapse has not yet converged [25]. Whereas empirical (and possibly theoretical) correlations may exist between initial progenitor mass; ejecta expansion velocity; and, therefore, explosion energy [26,27,28,29,30] (however, see [31,32]), recent 3D core-collapse SN models point more toward explosion energy as a function of, say, progenitor mantle binding energy and, more importantly, a core “compactness parameter” [33] rather than initial progenitor mass [25]. We point out that explosions have also been modeled as jet-driven [34,35,36].

We know that many RSGs are often-irregular, long-period variables (LPVs; e.g., [37,38,39]). We also know that many, if not all, RSGs have at least some circumstellar matter (CSM, e.g., [40]) beyond their photospheres, often dusty [41,42], as a result of mass loss during this phase of stellar evolution [43,44]. We observe evidence for a CSM layer above the photosphere via short-lived so-called “flash” emission features in early-time optical spectra for a number of SNe II [45]. The interaction of the SN shock with the dense CSM can lead to enhanced luminosity in the early-time light curves [46,47]. In fact, if the CSM is sufficiently dense, the breakout of the SN shock can be significantly delayed [48,49].

A question then emerges: What is the origin of the CSM around RSGs, and, then, at what rate and at what interval does it emerge before explosion, especially when the CSM is particularly dense? Theoretical evolutionary models include baseline empirical prescriptions of mass loss [50,51] (∼${10}^{-6}{M}_{\odot }$${\mathrm{yr}}^{-1}$; different, higher mass-loss rate prescriptions have been presented elsewhere [52]; see also van Loon in this volume and [53]). However, these rates are not nearly high enough to account for the amount and densities of the CSM that are inferred from a number of SN observations, which, in several, cases also imply short-time-scale (days to decades) mass-loss episodes prior to explosion.

Enhanced mass loss (∼${10}^{-4}$–${10}^{-2}{M}_{\odot }$${\mathrm{yr}}^{-1}$) can be driven pulsationally by internal convective shocks [54] induced by nuclear flashes [55] or convectively driven wave heating [56,57], which, toward the end of the RSG phase during late-stage nuclear burning, could lead to higher mass-loss rates [58] or eruptive outbursts during the final year or years prior to explosion [59,60,61], when turbulence and pulsations are most vigorous. (A “superwind” could develop from internal dynamical instabilities; however, this may only be the case for the most massive RSGs [62].) Such a precursor event has been observed for at least one normal SN II-P [63]. Other recent studies have provided alternatives to outbursts, e.g., a late-time effervescent zone [64] or a wave-driven chromospheric layer [65] to account for the immediate dense CSM above RSGs. Binary interaction may offer yet another mechanism [66].

Several indirect observational constraints on SN progenitors include the modeling of SN light curves via progenitor population synthesis [67], probing late-time ejecta [68,69], analyzing circumstellar gas ionized by the early ultraviolet/X-ray radiation from the SN blast [45,70,71], and inferring ages and turnoff masses from the SN’s local stellar and interstellar environment ([72,73,74,75]; however, see [76]). Of course, the surest means to understand the nature of the progenitor is to directly identify and characterize it as it was some time before it exploded. The first SN II to have its progenitor identified in this way was SN 1987A, although the SN was peculiar in its properties. In addition, ironically, the progenitor was identified as blue, not red, supergiant Sanduleak $-{69}^{\circ }$ 202 [77,78] (the unanticipatedly unusual progenitor could possibly be explained by a merger of two stars [79]). The first SN II-P progenitor to be characterized from pre-SN imaging as an RSG was SN 2003gd [80,81]. Subsequent identifications up to a certain point in time have been reviewed and discussed elsewhere [82,83,84,85]. In the remainder of this article, we provide a further and more current review.

2. Direct Identification of RSGs as SN Progenitors

A summary of all of the direct identifications of SN II-P progenitors to date is given in

Table 1. Direct identification of RSGs as SN progenitors.

SN	Host	$log({\mathit{L}}_{\mathbf{bol}}/{\mathit{L}}_{\odot })$	${\mathit{T}}_{\mathbf{eff}}$ (K)	Refs.
2003gd	NGC 628	4.3	∼3500	[80,81]
2004A	NGC 6207	∼4.2–4.7	∼2800–5900	[93]
2004et	NGC 6946	4.54	∼4300–6150	[86,87]
2005cs	NGC 5194	∼4–4.5	∼2950–4500	[94,95]
2006my	NGC 4651	∼4.2–4.8, <$5.1$	≲3750	[87,96,97]
2008bk	NGC 7793	4.5–4.6	∼3500–3700	[88,89,90]
2008cn	NGC 4603	≲5	≳3160	[98,99]
2009hd 1	NGC 3627	≲5.05	≲5200	[100]
2009ib	NGC 1559	5.1	∼3400	[101]
2012A	NGC 3239	4.7	∼4250–3400	[91,92]
2012aw	NGC 3351	4.8–5.0	∼3400–3900	[102,103,104]
2012ec	NGC 1084	5.15	<4000	[105]
2013ej 1	NGC 628	4.46–4.85	∼3400–4000	[106]
2017eaw	NGC 6946	4.9, 5.07	∼2500–3300	[107,108]
2018zd 2	NGC 2146	∼4.5–5.1	∼3500 3	[109]
2018aoq	NGC 4151	4.7	∼3500	[110]
2020jfo	NGC 4303	4.1 ? 4	∼2900	[111,112]
2022acko	NGC 1300	4.3–4.5	∼3500–3565	[113]
2023ixf	NGC 5457	4.7–5.5	∼2770–4000	[114,115,116,117,118,119,120,121,122,123]
2024ggi	NGC 3621	4.9	∼3290	[124]
2024abfl	NGC 2146	∼4.6 ? 4	∼3600 ? 4	[125]

¹ Possible II-L. ² Peculiar II-P, possible SAGB progenitor. ³ If the progenitor is similar to SAGB candidate MSX SMC 055, possibly $T_{\mathrm{eff}}\sim 2500$ K [126]. ⁴ Question mark, “?”, denotes uncertainty and necessity to confirm further.

. Like SN 1987A, the host galaxies of SN 2004et [86,87], SN 2008bk [88,89,90], and SN 2012A [91,92] were nearby enough (≲10 Mpc) that the progenitors could be identified in ground-based image data (this was strikingly true for SN 2008bk, whose host is at 3.4 Mpc, and data of high quality in multiple photometric bands were available). All of the remaining examples involve images taken prior to each SN with the high spatial resolution of the Hubble Space Telescope (HST). The SN 2003gd progenitor identification was a hybrid of HST data at shorter optical wavelengths and ground-based images at longer wavelengths. A number of cases also included infrared data from the Spitzer Space Telescope.

The precise location of the progenitor is generally established based on high-resolution images of the SN itself, usually when the SN is still bright, using either the HST or adaptive optics-assisted ground-based observations (AO). A notable exception is SN 2022acko, for which the SN was serendipitously imaged by the James Webb Space Telescope (JWST); those data were astrometrically aligned with the pre-SN HST imaging to pinpoint the progenitor site [113].

In Figure 1, we show one example of direct progenitor identification for SN 2018aoq in NGC 4151 (see also [110]). The leftmost panel shows the RSG progenitor in the HST F814W (∼I) band, the site of which was serendipitously imaged more than 2 years prior to its explosion. The precise location of the progenitor was pinpointed using an image of the SN itself we obtained with the HST in 2018 in the F555W band (∼V). We revisited the site again in 2020 with the HST in the F814W band, confirming that the RSG was, indeed, the progenitor based on its disappearance [127].

In Table 1, we provide both the host galaxy name and the inferred bolometric luminosity of the RSG progenitor relative to the Sun ($log[L_{\mathrm{bol}}/L_{\odot }]$). In the literature on each of the progenitors, the initial mass($M_{\mathrm{ini}}$), is generally inferred for the stars. However, we resist including these here and list just the luminosity instead. This is since the mapping from luminosity to initial mass, via theoretical stellar evolutionary tracks, can be fraught with additional uncertainties and open to interpretation [128], particularly given that the various available model tracks [129,130,131] for a given $M_{\mathrm{ini}}$ do not all terminate at the same locus on the Hertzsprung–Russell diagram, due, at least in part, to different input assumptions made in the modeling.

Many of the cases involve detection of the progenitor only in one band, with upper limits on detection in other bands, especially early on in the pursuit of direct progenitor identifications, e.g., SN 2005cs [94,95]. Typically, these one- or two-band detections have since been converted to luminosity estimates via bolometric corrections for RSGs [132,133,134]. When more than two photometric bands are available, model fits can be applied to the resulting observed spectral energy distributions (SEDs), then integrated over wavelengths to obtain luminosity estimates.

In practice, a limit exists on how far away the host can be while still being able to detect or resolve an individual luminous star. For the HST, that limit is ≲20–30 Mpc, beyond which crowding and confusion become insurmountable issues; for ground-based optical and near-infrared (and even the Spitzer telescope in the mid-infrared range), that limit is far more restrictive. In nearly equal numbers to the progenitor detections are non-detections, many as a result of the depth (or lack thereof) of the pre-SN imaging, given either the exposure time, the host distance, or both, as listed in

Table 2. Upper limits on progenitor detection.

SN	Host	$log({\mathit{L}}_{\mathbf{bol}}/{\mathit{L}}_{\odot })$	Refs.
1999an	IC 755	…	[135,136]
1999br	NGC 4900	…	[135,136]
1999em	NGC 1637	…	[137]
1999ga 1	NGC 2442	…	[138]
1999gi	NGC 3184	≲$5.1$	[139]
2001du	NGC 1365	…	[136,140,141]
2002hh	NGC 6946	…	[82]
2003ie 2	NGC 4051	…	[82]
2004am 3	NGC 3034	…	[142]
2004dg	NGC 5806	≳$4.5$, ≲$5.1$	[82]
2004dj 3	NGC 2403	…	[143]
2006bc	NGC 2397	≳$4.8$, ≲$5.1$	[82]
2007aa	NGC 4030	≳$4.5$, ≲$5.1$	[82]
2009H	NGC 1084	…	this work
2009N	NGC 4487	≳$4.5$, ≲$5.1$	this work
2016cok	NGC 3627	…	[144]
2018ivc 4	NGC 1068	≳$4.5$, ≲$5.1$	[145]
2019mhm	NGC 6753	…	[146]
2020fqv	NGC 4568	…	[147]
2020pvb 5	NGC 6993	…	[148]
2021yja	NGC 1325	…	[149]

¹ Possible II-L. ² Peculiar II-P?. ³ In stellar cluster. ⁴ Possible II-L or IIb. ⁵ Possible “IIn-P”.

. In a few of these cases, the resulting upper limits provided useful constraints on the progenitor’s luminosity, which are provided in the table.

We show absolute SEDs for several of the detected progenitors listed in Table 1 in Figure 2. The observed SEDs were all corrected both for assumptions of galactic foreground and internal host galaxy reddening, as well as for the assumed distance to each of the hosts. As one can see, first, the SEDs were not all been equally sampled across wavelengths, since we are completely beholden to whatever available archival imaging data serendipitously included the SN site. Secondly, with the caveat that the statistics are still quite small, the SEDs have generally different shapes relative to one another; the variety of shapes most likely arise from differing amounts of intrinsic reddening in (i.e., dustiness of) the circumstellar environments of the progenitors. This is not entirely surprising, since we do not expect the apparent SEDs of RSGs to be the same, given the rich variety of SED shapes for, e.g., galactic RSGs [42]. We should not expect all massive stars to have the same properties at the end of their lives; therefore, each RSG is an individual. Lastly, the detected progenitors span a range of at least an order of magnitude in luminosity, as can be seen as well in Table 1, from $log(L_{\mathrm{bol}}/L_{\odot })\sim 4$ to 5.5. This is, to the first order, consistent with the expected range of $M_{\mathrm{ini}}$ (∼8–$30M_{\odot }$) for massive stars that terminate as RSGs (however, we discuss this further in Section 3).

We must stress here that we have been using the term “progenitor” rather cavalierly so far. A star identified at the precise location of an SN is only a candidate progenitor until the star has demonstrably vanished. This requires patient vigilance to determine whether the SN has faded sufficiently below the luminosity of the candidate, with the wait time occasionally exacerbated by any lingering light as a result of CSM interaction. For many of the SNe in Table 1, the candidates have been shown to have disappeared, reinforcing the likelihood that these were the actual progenitors [99,127,150,151,152,153,154]. One important caveat to this is that we cannot outright eliminate the possibility that dust has formed around the SN and obscured or otherwise dimmed the light. We expect dust to form over the time span of years in the SN ejecta from massive star explosions [155]. Additionally, unresolved binaries or multiple-star systems certainly can appear as single stars at the distances of the host galaxies. We would expect this to occur quite frequently, since the multiplicity fraction is ≳70% for (primary) stars with $M_{\mathrm{ini}}\gtrsim 8M_{\odot }$ [156]. It is also possible that an identified progenitor is a chance superposition of two or more stars, again, given the host distances. In fact, the yellow star associated with SN 2008cn [98] appears, in retrospect, to have been a combination of a blue and a red (probably supergiant) star [99]. Similarly, the progenitor of SN 2013ej also appears to be the near-superposition of a blue and a red star, although their photocenters are just slightly displaced from one another [106].

2.1. Progenitors of Low-Luminosity SNe

Several of the SNe in Table 1 for which an RSG progenitor has been identified were low-luminosity events, including SN 2003gd [157], SN 2004A [93], SN 2005cs [158], SN 2008bk [89,159], SN 2018aoq [160], and SN 2022acko [161] (likely also SN 2006my [162]). Interestingly, the SED for these progenitors can be fit essentially with a bare photosphere, with effective temperatures in the range of $T_{\mathrm{eff}}\sim 3400$–3700 K. This is typified by, e.g., the case of the SN 2018aoq progenitor (see Figure 1), for which the multi-band SED was fit by a 3500 K stellar atmosphere model with minimal, if any, additional CSM extinction [110].

We also illustrate this for the SN 2008bk progenitor in Figure 3. Correcting the photometry [89] for the contributions of fainter stars, as seen with the HST, within the ground-based point-spread function (PSF) to the overall brightness of the progenitor in each band (see also [90]), we present the SED for the progenitor in the figure. For comparison with the corrected SED, we show MARCS stellar atmospheres at subsolar metallicity ($Z=0.010$ [163]) at $T_{\mathrm{eff}}=3600$, 3600, and 3700 K, as well as the SEDs at the endpoints of two BPASS model single-star evolutionary tracks [131]. One can see that all of the photospheres provide excellent fits to the full observed SED. In addition, the endpoints of the 7.7 and $8M_{\odot }$ tracks are roughly consistent with the SED to 2.2 µm. The $T_{\mathrm{eff}}$ and $log(L_{\mathrm{bol}}/L_{\odot })$ for the two model endpoints are 3555 K and 4.51 and 3550 K and 4.59, respectively. The main point here is that any effect of extinction as a result of CSM dust on the stellar SED is minimal, at least to wavelengths ≲ 3.5 µm. The luminosity implied for the SN 2008bk progenitor is consistent with the overall limiting luminosity for low-luminosity SNe II-P, i.e., $log(L_{\mathrm{bol}}/L_{\odot })\lesssim 4.7$. Fits to the SEDs of these progenitors generally—and specifically in the cases of SN 2008bk and SN 2018aoq—appear to imply that $M_{\mathrm{ini}}\sim 8$–$10M_{\odot }$.

One notable and unresolved case is the low-luminosity SN 2018zd. This event was a collapse either as a result of an electron-capture (EC) reaction in a degenerate ONeMg core of an $M_{\mathrm{ini}}\sim 7$–$9.5M_{\odot }$ super-asymptotic giant branch (SAGB) star [109] or of a more standard Fe core in an $M_{\mathrm{ini}}\sim 12M_{\odot }$ RSG [165]. The ambiguity arises from a large uncertainty in the host distance (∼6–18 Mpc). Assuming a distance on the short side, ≤10 Mpc, the detected star at the SN position, indeed, had properties more consistent with a those of an SAGB than an RSG [109].

We point out that the SN 2022acko progenitor may have been straddling the SAGB/RSG boundary as well, with $log(L_{\mathrm{bol}}/L_{\odot })\sim 4.3$–4.5 [113]. Although SN 2005cs was only somewhat underluminous on the light-curve plateau, it suffered a large fall from the plateau to an extraordinarily underluminous radioactive tail, likely resulting from a very low ⁵⁶Ni mass (∼$3\times {10}^{-3}{M}_{\odot }$) synthesized in the explosion [158]. The luminosity estimates for the progenitor, $log(L_{\mathrm{bol}}/L_{\odot })\sim 4$–4.5 [94,95], are also indicative of a similarly lower-mass progenitor.

What, if anything, can we glean from the upper limits to progenitor detections presented in Table 2? These limits were all estimated in optical photometric HST bands, e.g., F555W or F814W. A previously unpublished example for SN 2009H is shown in Figure 4. What has been generally done in the literature is to infer a luminosity constraint by, e.g., adjusting the brightness limits in these bands with RSG bolometric corrections [132,133,134], in a similar way as for a number of the actual detections (e.g., [139]) (see Section 2). However, this method may have led to luminosity underestimates [166,167]. We show the upper limits from Table 2 in Figure 5 and compare these to the intrinsic SEDs of several of the direct detections (Figure 2), with the intent of imposing luminosity constraints from this comparison.

What is readily obvious from Figure 5 is that the upper limits for several of the SNe do not provide meaningful constraints at all on the progenitor luminosities (contrary to what has been inferred in the literature for these events). However, some of the limits do, in a limited way. Based on the blue end of the SEDs of the progenitor detections and the luminosities associated with those stars (see Table 1), we can infer at least relatively useful, although approximate, luminosity upper limits, and for several cases, we can also place approximate lower limits. We provide these estimates in Table 2. The lowest lower-luminosity constraints arise from the observed limits being indicative of a progenitor at least as luminous as the detected low-luminosity progenitors (e.g., SN 2008bk; $log[L_{\mathrm{bol}}/L_{\odot }]\gtrsim 4.5$). The upper luminosity constraints arise from comparison with, e.g.,  the SN 2023ixf progenitor SED; therefore, the progenitor was likely no more luminous than that ($log[L_{\mathrm{bol}}/L_{\odot }]\lesssim 5.1$; see Section 2.3). The optically based limits have little constraining power (since most of the emission from an RSG is at longer wavelengths) and, had detection limits been available, particularly in the near-IR range, we would likely have been able to place far more stringent luminosity constraints, again, treating the known progenitor SEDs as templates.

With regards to the low-luminosity SNe II-P, the limits on detection of a progenitor only likely pertain to SN 2009N (see Figure 6), which had spectral properties similar to low-luminosity events, yet had photometric similarities to more intermediate-luminosity SNe II-P [168].

The nearby (∼14 Mpc) SN 2020jfo appears to be something of an unusual puzzle. The light-curve properties of the SN are consistent with more luminous SNe, including a short plateau duration of ≲65 days [111,112,169,170]; however, a single-band progenitor detection at HST F814W was underluminous relative to what would be expected for luminous, relatively massive RSGs, which could also imply enhanced CSM dust [111]. Furthermore, analyses of the late-time spectra were indicative of a star with a moderate initial mass of ∼$12M_{\odot }$[111,112,169,170]. However, the progenitor may also have been detected with the Spitzer telescope, and, as measured at these wavelengths, the total SED could be fit with a quite cool (∼2900 K), low-luminosity ($log[L_{\mathrm{bol}}/L_{\odot }]\sim 3.9$) blackbody; even including a consideration of CSM dust raised the luminosity estimate only somewhat to $log(L_{\mathrm{bol}}/L_{\odot })\sim 4.1$, which further implies a low initial mass [112]. Based on all of this, it is unclear whether SN 2020jfo should really be considered a low-luminosity SN, even though the inferred properties of its progenitor, at this point, at least, are consistent with those of other low-luminosity SN progenitor stars (see also [171]).

In summary, all in all, low-luminosity SNe II-P appear to arise from relatively lower-luminosity RSGs, for which dust in the CSM is likely relatively limited in quantity. One can infer that these RSGs are also at the low end of the possible range in initial mass for progenitors, i.e., $M_{\mathrm{ini}}\sim 8$–$30M_{\odot }$. Some of these progenitors may have been in a transitional state between EC-SN and Fe core-collapse SN.

2.2. Progenitors of More Luminous SNe

The remaining SNe in Table 1 with identified progenitors appear to have arisen from RSGs with higher luminosities ($log(L_{\mathrm{bol}}/L_{\odot })\gtrsim 4.7$) based on the analyses of the stars’ observed properties (although, this demarcation is somewhat arbitrary and not yet well-defined). Since the CSM dust production rate appears to be correlated with $L_{\mathrm{bol}}$ [41,42] (the mass of CSM dust may also increase as the RSG progenitor approaches explosion [58]), the effect of dust on the observed SED is expected to be larger for higher-luminosity progenitors.

If one examines Figure 2, then this is effectively what one sees. The presence of and necessity for including CSM dust became particularly apparent when attempting to fit the observed SED of the SN 2012aw progenitor [104]. The reddened overall shape of the SED, compared to that of a bare photosphere, is evident, as the dust in the CSM reprocesses the UV/blue photospheric emission into the infrared range. This was even more conspicuous in the detailed multiband SED of the SN 2017eaw progenitor [107,108]. The lone attempt to characterize the SN 2024ggi progenitor so far resulted in a luminosity $log(L_{\mathrm{bol}}/L_{\odot })\sim 4.9$ [124]. All of these RSG progenitors had luminosities at explosion approaching 100,000 Suns.

As one can see from Figure 2 the SED for the SN 2004et progenitor is unusual relative to the SEDs of the other SN progenitors (including for low-luminosity SNe). It looks quite unlike the expected SED of an RSG — the luminosity of the star in the blue is far higher than that of any of the other progenitors presumed to be RSGs. The SN 2004et progenitor was characterized solely based on ground-based image data. Subsequent observations of the field with the HST revealed that other, fainter stars within the PSF may have contaminated the light from the observed progenitor itself. However, even accounting for this contamination only reduces the observed flux in each band by ∼$0.3$ mag or so, and the SED is still better characterized by a luminous star hotter than the M type, i.e., a yellow supergiant [87]. What may be the case is that the detected star is actually a composite of more than one star. We will not really know until the SN has significantly faded; however, this may well be a long wait, since SN 2004et continues to exhibit signs of CSM interaction, as well as the formation of dust [172].

The progenitor requiring the largest amount of CSM dust to date in its characterization is SN 2023ixf, and we reserve a separate discussion for the next subsection (Section 2.3).

Have there been SN II-P progenitors more luminous than $log(L_{\mathrm{bol}}/L_{\odot })\approx 5.0$? The modeling of the two-band detection of the SN 2009ib progenitor in HST archival data led to an estimate of $log(L_{\mathrm{bol}}/L_{\odot })=5.12\pm 0.14$[101], although the SN itself was assessed possibly to be of low luminosity (or, at least, at the transition from low to more normal luminosity). Based on a single HST band (F814W) detection, however, the progenitor of SN 2012ec was estimated to have had $log(L_{\mathrm{bol}}/L_{\odot })=5.15\pm 0.19$ [105], which would imply that it was one of the most luminous (and potentially most massive) known RSG progenitors, which, unlike for SN 2009ib, is not entirely inconsistent with the observed properties of the SN itself [105,173].

Whereas the overwhelming majority of the identified progenitors were well-isolated spatially, we have already noted that the SN 2008cn progenitor candidate [98] was likely a blend of the RSG with a superposed (or companion) blue star [99] (we note that the progenitor candidate was distinctly variable with a period $\sim 64$ days [98]) and that the SN 2013ej progenitor was partially confused with the presence of a blue neighbor [106]. Therefore, it was very difficult to measure and accurately characterize the brightness of the star before explosion, despite indications that both were quite luminous. Although data were scant, the observed properties of SN 2008cn indicated that it was a luminous SN II-P [98]. SN 2013ej was somewhat unusual in that it was a slow-rising [174] and fast-declining, luminous SN II-P or II-L [175,176,177,178] that was also powered, at later times, by the interaction of the SN shock with the CSM [179]. The observed properties of SN 2009hd, particularly its fast decline, pointed to it being a possible SN II-L; substantial uncertainty existed as to whether the progenitor was actually identified due to the level of confusion in its environment, even at HST resolution, and the fact the SN observations indicated that the line of sight to the star is heavily dust-obscured [100]. Nevertheless, the likelihood is that it was a luminous RSG.

As discussed in Section 2.1, approximate constraints on the luminosity of a progenitor can be estimated from the observed limits on the progenitor’s detection. Lower luminosity constraints can be placed based on comparison of the detection limits with the least dusty SEDs of the lower-luminosity progenitors, and upper luminosity constraints can be established based on comparison with the most dusty SED, i.e., for SN 2023ixf. In three cases of moderate- to higher-luminosity SNe II (SN 2004dg [180], SN 2007aa [14,19], and SN 2018ivc [145]), we can infer that the star was at least as luminous as the SN 2008bk progenitor ($log[L_{\mathrm{bol}}/L_{\odot }]\gtrsim 4.5$) and, in one case, (SN 2006bc), at least as luminous as the SN 2012aw progenitor ($log[L_{\mathrm{bol}}/L_{\odot }]\gtrsim 4.8$).

2.3. The SN 2023ixf Progenitor as a Special Example

SN 2023ixf occurred in the nearly face-on Messier 101 (NGC 5457) at a distance of 6.85 Mpc, in close proximity within the host to giant H ii region NGC 5471. As such, the SN generated tremendous excitement very soon after discovery, at 14.9 mag, with classification as an SN II, and immediately became the target of intense scrutiny by a number of investigators working across the full gamut of wavelengths. Given the closeness and prominence of the host and interest in the nearby H ii region, a veritable treasure trove of archival data, both ground- and space-based, were available. In particular, the host had been observed over numerous epochs with Spitzer. (In fact, due to the relative scarcity of pre-explosion optical imaging data in which the star was detected, HST contributed comparatively little to our knowledge of this SN progenitor.) It became readily evident that the star (which could be straightforwardly isolated in Spitzer data with a high degree of certainty based on its equatorial coordinates alone) was highly variable. In fact, it was possible to describe the star classification as a semi-regular, LPV (period $P\approx 1000$–1100 days [115,118]), not unlike many known RSGs in the local group. This, in itself, was an astonishing discovery.

The existence of exquisite multi-epoch photometry at various bands over years before explosion also helped rule out any dramatic changes in the star’s luminosity up to days the SN, effectively ruling out the occurrences of any pre-SN outbursts, eruptions, or other catastrophic late-time mass loss [115,118,121,181,182,183]. Therefore, the origin of the dense CSM inferred around SN 2023ixf must have occurred via some other mechanism.

With the wealth of pre-explosion data, the proximity and brightness of the SN, and the relative isolation and ease of identification of the progenitor candidate, a number of investigators undertook the construction of appraisals of the progenitor’s properties. Beyond the general agreement that resulted in the analysis of the star’s variability, as mentioned above, and the overall recognition that the CSM was especially dusty, the approaches of these various investigations and their results were, by and large, disparate for a number of reasons. From the start, the measurement techniques for obtaining photometry from the existing HST, Spitzer, and ground-based near-IR data significantly diverged: For the HST data, most studies employed the Dolphot package v2.0 [184], while one [116] used DAOPHOT [185] and yet another used a tool developed by the authors [123]. In principle, everyone should have obtained the same numbers; however, even those using Dolphot arrived at different values based on different parameter settings, pre-processing steps, and so on. For measurements from the Spitzer image data, one study applied aperture photometry to the pipeline-processed mosaics [120]; another applied DAOPHOT PSF fitting to these mosaics [115], another employed DoPHOT [186] on the mosaics [117], while yet another [118] used the MOPEX [187] and APEX MultiFrame [188] packages v18.5.6 as advised by the Spitzer project. Needless to say, again, a range of output values was obtained. In Figure 2, we show the SED constructed by VanDyk [122] from HST F814W to Spitzer 4.5 µm, for which the uncertainties included the estimated amount of variability in each band.

Next, the approaches to model fitting to the various SEDs differed. Several of the studies [114,117,119,120,121,123] used the dust radiative-transfer code (DUSTY) [189], together with MARCS atmospheres [163] as the input light source; another [122] used PHOENIX atmospheres [190] instead. Two studies [115,122] used GRAMS models [191,192], in lieu of or in addition to DUSTY. About half of the studies adopted silicate-rich dust for the CSM, whereas roughly the other half adopted carbon-rich dust (although the CSM around known RSGs tends to have little to no C-rich dust [42]).

The net result of all of these various assumptions and approaches is an astonishing level of disagreement on the bolometric luminosity and effective temperature of the star years to days before it exploded (see Figure 7). This might have been expected if little data had been available to construct the star’s SED. However, in this case, we had an astounding level of multi-band data available, from the UV to the mid-infrared band. Given that, we should have been able to sharply define this one star’s characteristics; therefore, it is frustrating that the results have not converged. Technically, from a statistical standpoint, the various values for $T_{\mathrm{eff}}$ and $log(L_{\mathrm{bol}}/L_{\odot })$ do agree with each other in the ensemble, to within all of their estimated uncertainties; one could compute a weighted mean and uncertainty in each of $3478\pm 392$ K and $5.09\pm 0.17$. However, it is disappointing that, given the very high quality of the available data for the star, we have to place such large overall uncertainties on these inferred values.

How we [118] approached this analysis was to carefully measure the fluxes or estimate upper limits of detection based on the available HST, Spitzer, and ground-based near-IR imaging data over all of the available epochs, then perform a variability analysis of the resulting light curves. We estimated a period of $1091\pm 71$ days for the star over years before explosion. From this period, assuming two different $T_{\mathrm{eff}}$ values, we inferred bolometric luminosities from an established RSG period–luminosity (P-L) relation, with both values being on the high side ($log[L_{\mathrm{bol}}/L_{\odot }]\sim 5.3$–5.4). Next, we averaged the various epochs in the IR to arrive at global values, including the amplitude of the flux variability into the overall uncertainty in each band (for the single HST band in which the star was detected, we adopted a representative amplitude based on other cool variable giants). Then, we formally fit the SED with both GRAMS and DUSTY, assuming Si-rich dust, and obtained consistent results (and likely more stringent ones than from the P-L relation) of $log(L_{\mathrm{bol}}/L_{\odot })=4.88$–5.04 and $T_{\mathrm{eff}}=2340$–3150 K (68% credible intervals) [122]. As can be seen in Figure 7, these are on the lower-luminosity and (considerably) cooler side. Lest it be a consideration that our estimated properties were fanciful and unrealistic, we found a direct galactic analog, RSG IRC −10414, with strikingly similar properties [193,194]. (We did not necessarily intend, in our analysis, to suggest that IRC −10414 is also nearing explosion.) Interestingly, we concluded that the SN 2023ixf progenitor, like IRC −10414 (and Betelgeuse [195] and $\mu$ Cep [196]), may have had an interstellar bow shock.

Even with the surprising disparity in the inferred properties of the SN 2023ixf progenitor (with the overall disagreement, it is unfortunate that the community will likely just randomly pick from the various results when referencing this progenitor in the future), we were able to characterize this candidate star in extraordinary and unprecedented detail. We will likely not have the opportunity for a similar or superior breakthrough until an SN occurs in a host galaxy with multi-band JWST coverage. Although the level of certainty is exceedingly high that the candidate was, indeed, the progenitor, this still requires confirmation when the SN has sufficiently faded (see the discussion in Section 1).

3. Discussion

The key observationally driven advances in our understanding of SN II-P progenitors and their link to RSGs have only been made possible by the direct identification of progenitors in pre-explosion archival images. The statistics are still amazingly too small to draw any broad conclusions about SN II-P progenitors, despite previous attempts at this. Accumulating sufficient statistics on SN progenitors requires both time and patience; thus, every new example is highly welcomed and informative. Although upper limits of detection have provided some coarse indications of progenitor characteristics, only actual progenitors and progenitor system identifications will continue to advance this line of study in finer detail.

As a community, we have generally found that low-luminosity SNe arise from lower-luminosity ($log[L_{\mathrm{bol}}/L_{\odot }]\sim 4$–4.7; and lower $M_{\mathrm{ini}}$) RSGs [197]. Given a normal initial mass function within host galaxies in the local Universe, we would expect these progenitors to be predominant. These RSGs appear to have less (and less dusty) CSM prior to explosion, and in a few cases, the stars may dwell near the boundary of RSG and SAGB, perishing as either a core-collapse or electron-capture event. Whereas low-luminosity SNe II-P possess a certain degree of observational homogeneity, both in photometric and spectroscopic properties [22,198], higher-luminosity SNe are more heterogeneous in their properties, preventing us from clearly distinguishing what is “normal” from what is more extreme. If anything, a continuum of properties likely exists, starting at the low-luminosity floor [10,14,17,18,19]. We likely see this reflected in the diversity of observed progenitor characteristics for RSGs with $log(L_{\mathrm{bol}}/L_{\odot })\gtrsim 4.7$. Suffice it to say that the mass, extent, and dust content of CSM is substantially higher for progenitors of more luminous explosions.

Target-of-opportunity programs with both the HST and AO, observing young SNe II-P to identify progenitor candidates, have been ongoing for the last two decades and necessitate the virtual equivalent of a long-term program because of the relatively low rate of viable candidates. (JWST has yet to be purposely used in this capacity, although it has serendipitously it has [113].) Over time we, as a community, have successfully amassed constraints on dozens of interesting SNe and their progenitors. Every new progenitor identification is precious, and each has greatly advanced our understanding of SN explosion physics, stellar evolution, and progenitor mass-loss history, particularly by putting the characteristics of each SN in context with the properties of its directly identified progenitor.

The “Red Supergiant Problem”

This brings us to one last topic on RSGs as SN progenitors. As one can see from Figure 8, the luminosities derived for the SN progenitors, both from direct identifications and from limits on detection, all tend to top out at $log(L_{\mathrm{bol}}/L_{\odot })\lesssim 5.2$ (with the possible exception of the upper range of the luminosity estimate for the SN 2023ixf progenitor) and generally fall well below the empirical Humphreys–Davidson limit [199] on the luminosities of local-group cool supergiants ($log(L_{\mathrm{bol}}/L_{\odot })\sim 5.7$; see also Humphreys and Jones & Humphreys in this volume). The apparent fact that no SN II-P progenitor has, so far, been found closer to this limit, even though such luminous RSGs exist in the local Universe, is a puzzle. Why are we not finding highly luminous RSG SN progenitors? This so-called “red supergiant problem” has been debated intensively in the literature [82,84,128,200,201], and various explanations have been offered to help solve this conundrum: from CSM dust [202] to bolometric correction [134], statistics [201,203], and alternative evolutionary pathways for the highest-luminosity (most initially massive) RSGs [204]. The fact that the fate of the most massive RSGs may depend on the core compactness parameter [205] can lead to mass ranges that terminate in failed explosions, leading directly to black-hole formation [33,206].

The search was then underway for RSGs that simply failed to explode [207,208] or experienced very low-energy SNe [209]. A promising candidate source, “NGC 6946-BH1”, was isolated [210], although according to analysis of recent JWST observations, its origin remains ambiguous [211,212]; possibilities include a stellar merger [212] manifested as an intermediate-luminosity optical transient [213] or luminous red nova [214].

Possibly a simpler means of explaining, or attempting to explain away, the RSG problem is by taking into account the full SEDs of RSGs, from the optical to the mid-IR. Since many of the SN progenitor identifications are in a single band, the question asked is, how much divergence is the inferred luminosity from the actual luminosity; it has essentially been argued that the inference is underluminous, and taking this into account, one could statistically eliminate the existence of an “RSG problem [166,167]”, although the assumed luminosities for identified progenitors differ to a fair extent from those we have presented here.

We would argue that the RSG problem remains to be adequately challenged observationally. The sample still needs to be at least doubled [128]. Therefore, every new SN II with a progenitor identification further augments the luminosity (mass) distribution.