A publishing partnership

The following article is Open access

The Red Supergiant Progenitor Luminosity Problem

, , and

Published 2025 January 21 © 2025. The Author(s). Published by the American Astronomical Society.
, , Citation Emma R. Beasor et al 2025 ApJ 979 117DOI 10.3847/1538-4357/ad8f3f

Download Article PDF
DownloadArticle ePub

You need an eReader or compatible software to experience the benefits of the ePub3 file format.

0004-637X/979/2/117

Abstract

Analysis of pre-explosion imaging has confirmed red supergiants (RSGs) as the progenitors to Type II-P supernovae (SNe). However, extracting an RSG's luminosity requires assumptions regarding the star's temperature or spectral type and the corresponding bolometric correction, circumstellar extinction, and possible variability. The robustness of these assumptions is difficult to test since we cannot go back in time and obtain additional pre-explosion imaging. Here, we perform a simple test using the RSGs in M31, which have been well observed from optical to mid-IR. We ask the following: By treating each star as if we only had single-band photometry and making assumptions typically used in SN progenitor studies, what bolometric luminosity would we infer for each star? How close is this to the bolometric luminosity for that same star inferred from the full optical-to-IR spectral energy distribution (SED)? We find common assumptions adopted in progenitor studies systematically underestimate the bolometric luminosity by a factor of 2, typically leading to inferred progenitor masses that are systematically too low. Additionally, we find a much larger spread in luminosity derived from single-filter photometry compared to SED-derived luminosities, indicating uncertainties in progenitor luminosities are also underestimated. When these corrections and larger uncertainties are included in the analysis, even the most luminous known RSGs are not ruled out at the 3σ level, indicating there is currently no statistically significant evidence that the most luminous RSGs are missing from the observed sample of II-P progenitors. The proposed correction also alleviates the problem of having progenitors with masses below the expected lower-mass bound for core collapse.

Export citation and abstractBibTeXRIS

Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.

1. Introduction

The most robust test of stellar evolutionary theory for massive stars comes from linking observed supernovae (SNe) directly to their detected progenitors. It has long been suggested that all single stars with initial masses between 8 and 30 M (e.g., G. Meynet & A. Maeder 2000) will pass through the red supergiant (RSG) phase, before ending their lives as Type II-P core-collapse supernovae (CCSNe). It was also expected that above some initial mass (∼30 M) the cores of stars are so massive that infall after core collapse will quench a successful SN and will lead to the creation of a black hole. However, both the analysis of SN progenitor detections or upper limits (e.g., S. J. Smartt et al. 2009) and some evolutionary models (e.g., T. Sukhbold et al. 2018) now suggest the upper mass limit for II-Ps may in fact be lower (∼20 M).

Following the first case of SN 1987A, serendipitous pre-explosion observations of several SN progenitors (see S. J. Smartt 2015, and references therein) provided important and direct information about the endpoints of stellar evolution for massive stars. Namely, it was through the direct imaging method that RSGs were confirmed as the long-expected progenitors of Type II-P SNe (e.g., J. R. Maund et al. 2004). However, this evidence also pointed to an apparent discrepancy between the presumed initial mass distribution of RSGs in the field compared to RSGs dying as SNe. S. J. Smartt et al. (2009) argued that the apparent discrepancy between the initial mass distribution of RSGs in nearby field populations, as compared to the initial mass distribution of Type II-P progenitors, was clear evidence that some RSGs do not die as bright SNe II-P. In particular, RSGs with inferred initial masses of 20–30 M are clearly seen in nearby stellar populations, but the maximum observed SN II-P progenitor mass is only ∼17 M.

The unknown fate of stars in the 20–30 M range has famously been termed the "red supergiant problem," and has led to an explosion of theories that may explain the phenomenon. One leading idea is that these stars fail to produce bright SN explosions as their cores collapse into a black hole. Theoretical studies also seem to be converging on an upper mass limit for II-Ps of around 20 M (see, e.g., T. Sukhbold et al. 2018). However, the inferred distribution of RSG progenitor initial mass has some unresolved issues. For one, even if the distance to the object is known, there is considerable uncertainty stemming from the assumption of spectral type (SpT) for each progenitor star. B. Davies & E. R. Beasor (2018) showed that assuming a constant SpT of M0 for the RSG progenitors of SN II-Ps could lead to an underestimation of the initial mass of 5 M. By taking this into account, the significance of the missing RSGs was reduced to <2σ. In addition, converting luminosities to initial masses is heavily dependent on the mass–luminosity relation chosen (see Table 2 within B. Davies & E. R. Beasor 2018) furthering the uncertainty. One way to circumvent these uncertainties is to skip a step altogether. Rather than comparing the inferred mass distribution of the field RSGs to the inferred masses of the progenitor RSGs, one can simply compare the measured luminosity distributions directly. In this way, considerable uncertainties inherent to stellar evolutionary models can be removed from the problem.

When comparing field RSGs to SN progenitors in terms of luminosity, it appears there is a dearth of Type II-P progenitors above = 5.2, in tension with the empirical luminosity limit for RSGs of = 5.5 (B. Davies et al. 2018). In B. Davies & E. R. Beasor (2020a), it was shown, however, that the absence of high luminosity progenitors can, in part, be explained by the steepness of the L distribution and low number statistics, and they estimated that the significance of the "missing" progenitors is between 1σ and 2σ. Further complications arise when taking into account the effect of circumstellar dust, which may have been destroyed by the SN (e.g., J. J. Walmswell & J. J. Eldridge 2012).

However, even when eliminating systematic errors from the conversion of luminosity to initial mass, there is still a glaring uncertainty that mars the pre-explosion imaging technique; it is impossible to know the true shape of the progenitor spectral energy distribution (SED) from only a single-band detection. This means we are forced to make assumptions about the intertwined values for reddening (foreground and circumstellar) and temperature. Here, we conduct an experiment to characterize the potential systematic effects, by directly comparing the inferred luminosities from single Hubble Space Telescope (HST) I-band (F814W) photometry—the filter in which serendipitous pre-explosion imaging is most frequently found—to the bolometric luminosity found when integrating under the entire SED (a method which is free from model assumptions) for the same sample of stars. To do this, we use a sample of RSGs from M31 (S. L. E. McDonald et al. 2022) that have well-sampled SEDs from optical to the mid-IR, and in addition, have HST F814W-band observations from the Panochromatic Hubble Andromeda Treasury (PHAT) survey (J. J. Dalcanton et al. 2012). With this direct comparison, we will demonstrate the potential systematic effects on the luminosities determined for SNe when only a single-band observation is available.

In Section 2 we detail the methods used to determine the luminosities of each RSG in the sample. In Section 3 we compare the luminosities determined for each RSG via the bolometric correction (BC) and SED methods and discuss potential implications for SN progenitors. We also apply the results of this work to the red supergiant problem. In Section 4 we summarize the results and make suggestions for future work on RSG progenitors.

2. Method

We begin by matching the stars in the catalog developed by S. L. E. McDonald et al. (2022, hereafter M22) to point sources in the PHAT survey using astropy routine SkyCoord. In Figure 1 we show the positions of the stars, where the pink circles show sources from M22, and those that correspond to a PHAT source are highlighted with a blue circle. In some cases we found multiple PHAT sources associated with RSGs from the M22 sample, implying possible binary/multiple systems or close neighbors as projected on the sky. For simplicity, we ignore any objects with multiple matching PHAT sources. After excluding these multiples, we found 139 sources within PHAT that are likely associated with RSGs in the M22 sample (see the bottom panel of Figure 1).

Figure 1. Refer to the following caption and surrounding text.

Figure 1. Top panel: sources in the M22 catalog (pink-filled circles) cross-matched with sources in PHAT (purple open circles). Bottom panel: luminosity distribution for the cross-matched sample.

Standard image High-resolution image

We now compare the luminosities of each of the RSGs derived via two methods. The SED-derived luminosities were calculated in M22 by using IDL routine int _tabulated and adopting an M31 distance modulus of 24.4 (I. D. Karachentsev et al. 2004). In this method, the luminosity is obtained by simply integrating under the observed SED. The only assumption necessary for this method is that any of the flux lost at shorter wavelengths due to absorption by circumstellar material (CSM) is re-emitted at longer wavelengths, and so when integrating from optical-to-mid-IR wavelengths, all of the stars' flux is recovered (see B. Davies et al. 2018).

The second method we use to obtain a luminosity is that which is commonly used for pre-explosion imaging of SN progenitors. Pre-explosion imaging relies on serendipitous detections or upper limits from surveys that often used only a single band, most commonly the HST F814W (I-band) filter (see, e.g., Table 4 within B. Davies & E. R. Beasor 2018). A number of assumptions are therefore needed to derive a luminosity. We begin by converting the apparent I-band magnitude to an absolute magnitude by applying a distance modulus of 24.4 (see above) and dereddening the photometry according to the extinction value from PHAT (see M22). We then apply the same bolometric correction as adopted in the study by S. J. Smartt et al. (2009; BCF814W = 0.9) to convert the I-band magnitude to a bolometric magnitude. Finally, we convert this to luminosity.

The comparison between the SED-derived luminosities and the I-band-derived luminosities (using BCF814W = 0.9 mag; S. J. Smartt et al. 2009) is shown in the top panel of Figure 2. B. Davies & E. R. Beasor (2018) argued that RSGs tend to evolve to later spectral types of M5 before their death. We therefore also show the results when assuming a later spectral type (and thus smaller BCF814W), as suggested by B. Davies & E. R. Beasor (2018), in the bottom panel of Figure 2, using BCF814W = 0.0 mag. We demonstrate the differences again in Figure 3, where we show the ratio L(I band)/L(SED) for each BC value.

Figure 2. Refer to the following caption and surrounding text.

Figure 2. Comparison between luminosities derived by integrating under the observed SED and the luminosities derived from the HST F814W photometry. Top panel: using BCF814W = 0 from S. J. Smartt et al. (2009). Bottom panel: using BCF814W = 0.9 from B. Davies & E. R. Beasor (2018).

Standard image High-resolution image
Figure 3. Refer to the following caption and surrounding text.

Figure 3. Histograms showing the values for the log ratio of the luminosity determined from the I-band method using the bolometric corrections for early-type (BC = 0.9) and late-type (BC = 0) RSGs.

Standard image High-resolution image

3. Results and Discussion

Figure 2 demonstrates that when treating the RSGs in M31 the same way that SN progenitors are often treated, i.e., assuming a SpT and hence BC, there can be significant disagreement between this derived luminosity and the luminosity as derived more accurately by the SED method.

When adopting the same assumptions applied to preexplosion photometry in the original "RSG problem" papers (S. J. Smartt et al. 2009; S. J. Smartt 2015; BCF814W = 0.9), we find that luminosities from single-band photometry systematically underestimate the bolometric luminosity of the star by about a factor of 2 (i.e., by an average of 0.3 dex in Figure 2). This translates to a shift in the derived initial mass of ∼4 M (see, e.g., Figure 4 within S. J. Smartt et al. 2009). Importantly, this indicates that the distribution of initial masses for SN II-P progenitors summarized by S. J. Smartt et al. (2009) should be shifted to higher mass, which will modify both the lower bound and upper bound for the inferred range of masses of SN II-P progenitors. On the other hand, when applying a BC appropriate for stars of a later spectral type (BCF814W = 0), the average offset between the two methods is much smaller (−0.08 dex), although there is still dispersion on the order of 0.2 dex. We now discuss the potential reasons for this discrepancy between SED-derived luminosities and F814W-derived luminosities, as well as the effect this may have on progenitor detections.

Figure 4. Refer to the following caption and surrounding text.

Figure 4. Required BCI values to make the luminosities from the I-band method match the SED method.

Standard image High-resolution image

3.1. Potential Causes of Discrepancy

There are likely two main causes for the discrepancy between the SED method and the single-band method: the uncertainty of the SpTs for each star and stellar variability. 4 In this case, we know that most of the RSGs in the M31 sample are not close to the time of explosion, and therefore are expected to span a broader range of SpTs than SNe progenitors; this is because SN progenitors should all be located in the coolest and most luminous endpoints of their evolutionary tracks, whereas the population of RSGs in M31 sample a range of points along those tracks. For this reason, the assumed value of BC = 0.0 mag causes a slight overestimate of the luminosity compared to the SED value (Figure 3). While we do not know their individual SpTs, we do know that the range of SpTs is broad (M0–M5; see, e.g., R. M. Humphreys et al. 1988; P. Massey et al. 2016), while SN II-P progenitors likely all have later SpTs (B. Davies & E. R. Beasor 2018). If we allow BCF814W to vary, while still using a single BC for the entire RSG population in M31, we find that a BC of 0.2 mag (corresponding to an average SpT = M4; see B. Davies & E. R. Beasor 2018) yields the smallest offset between the two methods, implying the sample of RSGs in M31 have preferentially late spectral types despite not all being close to SN. We suggest the relation between SpT and BC may need further refinement; this will be addressed in a future work. While this reduces the average offset, there is still a dispersion at a level of 0.2 dex around the 1:1 relation.

We now investigate the range of BC values that would be required to achieve a perfect match between the luminosities determined via the SED and the single-band methods. When leaving BCF814W as a free parameter for each individual star, it is possible to find agreement between the two luminosity methods using BC values ranging between −1 and 1 (see Figure 4). BC values less than 0 would imply a very late spectral type (B. Davies & E. R. Beasor 2018) or a local reddening correction larger than the adopted value.

The other potential cause for the large spread is stellar variability. It is known that RSGs are variable (e.g., M. D. Soraisam et al. 2018) by up to 1 mag in the optical, decreasing to around 0.1 mag at mid-IR wavelengths. For luminosity estimates that rely on single-band photometry, this could potentially have a large effect on the derived Lbol. Since multi-epoch SN progenitor detections are still rare, it is usually not possible to tell how variable a star was, or at what point in its variability cycle the star was when the observations were taken. Some studies do search for variability in SN progenitors (W. V. Jacobson-Galán et al. 2022; J. E. Jencson et al. 2023; J. M. M. Neustadt et al. 2024). For the stars in M31, the photometry used in the SED method is all ground-based, and not taken contemporaneously with the HST PHAT data. The inherent variability within the RSGs could potentially influence the dispersion in luminosity observed within the sample, but the overall average should remain unaffected, as it is unlikely that the detections have been biased toward any particular phase of the variability cycle. In the same way as above, we now investigate the level of variability that would be required to allow the luminosities from each method to match perfectly (see Figure 5). We find that by adding variability of up to ±1 mag to the F814W photometry it is possible to reconcile the inconsistencies between the SED and the I-band luminosity method.

Figure 5. Refer to the following caption and surrounding text.

Figure 5. Level of I-band variability required to minimize the discrepancy between the BC and SED luminosity calculation methods.

Standard image High-resolution image

As such, we suggest that the offset is likely caused by a combination of uncertain BCs and stellar variability. It is also important to note that the RSGs in M31 represent a best-case scenario compared to most SN progenitor detections since the distance and foreground extinction are well known. More specifically, the line-of-sight interstellar extinction toward progenitors is determined from a combination of Milky Way reddening derived from dust maps and possible host galaxy extinction determined from, e.g., Na i D lines seen in the SN spectrum. However, host galaxy reddening determined from observations of SN light does not account for possible CSM dust around the progenitor that may have since been destroyed in the SN explosion itself. The uncertainty introduced by this is inherently in one direction—i.e., it potentially makes the progenitor more luminous if there is CSM dust.

3.2. Impact on SN Progenitor Luminosities

Linking progenitor stars to their explosive deaths is a clear and direct test of stellar evolutionary theory. However, the results here highlight the difficulty in deriving fundamental progenitor properties from single-band imaging. Namely, both the precision and accuracy in deriving a progenitor luminosity and initial mass from optical photometry are significantly worse than previously assumed.

The situation is, however, even worse than demonstrated above. In this work, we are working under idealized conditions where the foreground extinction and distance to M31 are well known (J. J. Dalcanton et al. 2012), and despite this, there is still significant uncertainty in the luminosities derived from single bands. In real situations with SN progenitors, the uncertainties on distance and extinction are likely far higher, in addition to the uncertainty surrounding the SpT and any intrinsic variability.

Indeed, it is difficult to determine SpTs (and hence the appropriate BCs) for stars without spectra. S. J. Smartt et al. (2009) assumed all RSG progenitors would be the earliest spectral type, M0, and hence would require a BCF814W = 0.9 (taken from J. H. Elias et al. 1985). Here, we have shown that when treating the M31 RSGs the same way, this assumption artificially causes an underestimation of a star's luminosity by an average of 0.3 dex.

Real RSGs have a range of later M types, especially in the case of SN progenitors, and this discrepancy is improved somewhat when using a BC for a later SpT. B. Davies & E. R. Beasor (2018) showed that RSGs likely evolve to later spectral types as they approach SN, and suggest that, instead, for SN progenitors we should assume a later SpT (in the absence of any further knowledge). When reanalyzing Type II-P progenitor luminosities, a smaller BC appropriate for late-type (M5+) RSGs B. Davies & E. R. Beasor (2018) showed that on average the inferred luminosity (and hence initial mass) increased slightly. Here, for the RSGs in M31, we have shown that while the average offset between the SED and the I-band method is improved with a more appropriate BC, there is still scatter on the order of 0.2 dex (see bottom panel of Figure 2). While the population of stars in M31 is likely not directly comparable to SNe progenitors, since not all of the RSGs will be close to their time of death, the results here suggest the errors on luminosities determined using single-band photometry are likely larger than previously assumed.

Variability in SN progenitors may also have a significant impact on the luminosity determined. Here, we have shown by assuming I-band variability up to ±1 mag, the dispersion of the luminosities can be reduced.

Combined, the intrinsic variability of the RSGs and the uncertainty in bolometric correction for the progenitor sources likely result in an error in luminosity that is larger than currently assumed. This has an impact at the high and the low end of the progenitor luminosity function and may go some way to explain the lack of high-mass SN II progenitors as well as the presence of stars with luminosities below = 4.5 (M. Fraser et al. 2014). The lower luminosity progenitors may form a more severe conflict with theory than the missing high-mass progenitors. For a single star to undergo core collapse, its initial mass must be above ∼8 M(e.g., A. Heger et al. 2003). Below this initial mass, single stars are expected to die as a planetary nebula and white dwarf. Binary evolution processes such as mass transfers or mergers can give a tail of CCSNe at later ages than a single 8.5 M star (and hence, we may see CCSN explosions at initial masses lower than M = 8.5 M; E. Zapartas et al. 2017, 2019, 2021), but the progenitor stars themselves should have a luminosity similar to a single star of Mini > 8 M at the time of explosion since they have gained mass throughout their evolution. There have been progenitor detections that, when using standard assumptions to estimate luminosity, appear to have luminosities below this limit (e.g., SN 2003gd and SN 2005cs; see Figure 6). The results found here alleviate this tension since they imply the luminosities for these low-mass progenitors likely have significantly larger errors than previously assumed. In addition, both of these progenitors have their luminosities estimated from F814W photometry only, and as discussed in Section 3.2.1 by progenitors with F814W and mid-IR photometry, relying on F814W photometry only may lead to the luminosity being significantly underestimated.

Figure 6. Refer to the following caption and surrounding text.

Figure 6. Cumulative luminosity distributions for Type II progenitors with standard errors (top panel) and additional systematic uncertainty of 0.2 dex added to I-band observations (bottom panel). In each case, the symbols represent the following: filled circles—multiwave band; open squares—F555W; open diamonds—F814W; and open crosses—K band. We also show the luminosity measurements from S. J. Smartt et al. (2009) via the green symbols. One might expect that the older estimates from S. J. Smartt et al. (green) should all be fainter by 0.36 dex as compared to the revised estimates (blue). While the newer estimates are corrected by this factor, they also have updated photometry, distance, and extinction estimates as detailed in B. Davies & E. R. Beasor (2018), so they are not all offset by the same amount. Finally, we show the luminosities that would have been estimated for SN 2017eaw, SN 2023ixf, and SN 2024ggi if only I-band photometry were available using the BCs for both an M0 and an M5 RSG via the red diamonds.

Standard image High-resolution image

3.2.1. Progenitors with Mid-IR Pre-explosion Photometry

Only a handful of SN progenitors have pre-explosion detections in the mid-IR beyond 2 μm. An interesting case is SN 2017eaw. The progenitor was identified as being a dusty RSG from multiwavelength pre-explosion photometry (C. D. Kilpatrick & R. J. Foley 2018; S. D. Van Dyk et al. 2019), including near- and mid-IR data from Spitzer out to 4.5 μm and two epochs of HST/F814W. C. D. Kilpatrick & R. J. Foley (2018) derived a luminosity from SED fitting of = 4.9, assuming a distance to NGC 6946 of 6.7 Mpc (μ = 29.13; N. A. Tikhonov 2014) and total foreground extinction of E(BV) = 0.34 mag. S. D. Van Dyk et al. (2019) found a larger luminosity of = 5.1 but assumed a correspondingly larger distance of 7.73 Mpc from their own TRGB analysis. We now take only the HST/F814W photometry and compare the luminosity that would have been found for the progenitor had there been no other photometry available.

For SN 2017eaw, there were two available epochs of F814W data from 2004 and 2016. We begin by taking the 2004 photometry from S. D. Van Dyk et al. (2019), mF814W = 22.60 mag, and determine the luminosity adopting the same values for the distance and extinction, and find = 4.8 and 4.5 using BCF814W = 0 mag (M5) and BCF814W = 0.9 mag (M0), respectively. When using the photometry from 2016 of mF814W = 22.8 mag, the corresponding luminosity estimates decrease to = 4.7 and 4.4. This results in offsets from the SED-derived luminosities of 0.3–0.4 dex for an M5 BC and significantly larger offsets of 0.7–0.8 dex for an M0 BC. The F814W photometry reported by C. D. Kilpatrick & R. J. Foley (2018) is ∼0.05 mag brighter at both epochs. Conducting the same exercise, we find offsets from their SED-derived luminosities of 0.1–0.2 dex for an M5 BC and 0.5–0.6 dex for an M0 BC.

Recently, SN 2023ixf was detected in the nearby spiral M101 (K. Itagaki 2023), a galaxy that has been imaged many times across multiple wavelengths. Several works have examined the extensive pre-explosion data set—including optical imaging from HST and ground-based resources, ground-based near-IR imaging, and Spitzer mid-IR imaging—and derived SED-based luminosities of the RSG progenitor spanning a wide range: = 4.74 ± 0.07 at the low end (C. D. Kilpatrick et al. 2023), mid-range values of = 4.8–5.0 (J. M. M. Neustadt et al. 2024; S. D. Van Dyk et al. 2024; D. Xiang et al. 2024a), and high-end estimates of ≈ 5.1 (J. E. Jencson et al. 2023; Z. Niu et al. 2023; Y.-J. Qin et al. 2024; C. L. Ransome et al. 2024). Notably, J. L. Pledger & M. M. Shara (2023) examined only the HST data, inferring an I-band absolute magnitude of , which corresponds to = 3.6 (M0) or = 3.9 (M5). On the other end, M. D. Soraisam et al. (2023) obtained = 5.37 ± 0.12 by applying the RSG period–luminosity relation of M. D. Soraisam et al. (2018) to the observed pre-explosion periodicity. Despite the large range in luminosity estimates, there has been general agreement that the progenitor was highly dust-enshrouded and undergoing enhanced pre-explosion mass loss, essentially guaranteeing that luminosities derived solely from optical data will be vast underestimates.

We now repeat the exercise of taking only the HST/F814W photometry to derive a luminosity estimate for SN 2023ixf. We adopt mF814W = 24.31 ± 0.05 mag (taken on 2002 November 16) and a total foreground extinction of E(BV) = 0.04 mag from S. D. Van Dyk et al. (2024), and a distance of 6.85 ± 0.32 Mpc (μ = 29.18; A. G. Riess et al. 2022). Thus, we find ≈ 3.5 (M0 BC) or = 3.9 (M5 BC), consistent with the comparable analysis of J. L. Pledger & M. M. Shara (2023). As expected, the offsets from the SED-based luminosities are remarkably large, 1.2–1.6 dex for an M0 BC and 0.8–1.2 dex for an M5 BC.

Finally, SN 2024ggi was recently discovered in the nearby galaxy NGC 3621 (J. Tonry et al. 2024), again with extensive pre-explosion coverage in the optical, near- and mid-IR. So far, D. Xiang et al. (2024b) have analyzed this data set and derived an SED-based luminosity of . They also determined MF814W = −6.21 ± 0.08 mag (assuming total foreground extinction E(BV) = 0.19 mag; J. Zhang et al. 2024), and distance modulus μ = 29.14 ± 0.06 mag (R. B. Tully et al. 2013). Based on the I-band magnitude alone, we would thus infer = 4.0 (M0) and 4.4 (M5), once again giving substantial offsets from the SED-based estimates of 0.9 and 0.5 dex, respectively.

Overall, these three SN progenitor cases—where we have sufficient multiwavelength photometry to place the best available constraints on the SED and hence the true luminosity—illustrate that the method of applying a reasonable assumed bolometric correction to single-band photometry generally does a poor job of approximating the true luminosity. In particular, underestimating the luminosity by a factor of 10 or more is not unusual using this method. 5 This should raise concern about luminosity and initial mass estimates for other SN progenitors where such multiwavelength information is not available.

3.3. The Red Supergiant Problem

Since S. J. Smartt et al. (2009) postulated that the highest mass (or luminosity) RSGs may not end their lives as SNe, various evolutionary scenarios have been proposed to account for the RSG problem, such as enhanced mass-loss (C. Georgy 2012), direct collapse to black hole without a bright SN (T. Sukhbold et al. 2018), or massive RSGs producing SN subtypes such as SNe IIn or II-L instead of SNe II-P (N. Smith et al. 2009, 2011). It has also been suggested that the "missing" RSGs may be an artifact of unaccounted extinction (N. Smith et al. 2011; J. J. Walmswell & J. J. Eldridge 2012) or incorrect bolometric corrections (B. Davies & E. R. Beasor 2018). Whatever the explanation, the statistical significance of the result has, however, remained contentious in the literature (see, e.g., B. Davies & E. R. Beasor 2020a, 2020b; C. S. Kochanek 2020).

Here we have demonstrated that when RSG luminosities are determined by single F814W-band photometry, they are both (1) systematically underestimated by a factor of 2 when adopting standard BC values, and (2) have larger errors than previously claimed, due to the added uncertainty surrounding a progenitors' variability and SpT. The work presented here indicates that the potential additional errors on luminosities determined from HST I-band detections could be as high as 0.2 dex.

We now explore the effect this additional uncertainty may have on the statistical significance of the RSG problem. To do this, we repeat the analysis presented in B. Davies & E. R. Beasor (2020a), including SN progenitors that have been discovered more recently, and include an additional error on any SN progenitor detected in the F814W filter. To include the additional error on each of these Lbol measurements, we randomly sample from a Gaussian centered on zero with a standard deviation of 0.2 dex, and we add this to the normal errors. We repeat this 104 times and the luminosity that is used in the analysis is taken to be the average with errors drawn from the standard deviation. The results of each run are shown in Figure 6.

When including this additional uncertainty, the significance of the RSG problem is even more dubious. As demonstrated by B. Davies & E. R. Beasor (2018, 2020b), using later spectral types for progenitors yields higher luminosities/initial masses. Compared to the RSGs in M31, we show that when using the assumed BC from S. J. Smartt et al. (2009) (BCI = 0), the luminosities of the RSGs in M31 are underestimated by an average of 0.3 dex. We now suggest that even if a later BC is assumed, the error bars on the derived luminosity are likely far larger than typically appreciated, due to unseen variability and an uncertain BC versus spectral type relation. In Figure 7 we show the constraints on the minimum Llo and maximum Lhi progenitor luminosities within the 68%, 95%, and 99.7% confidence limits. In the left-hand figure, we show the results when considering only the standard errors, while on the right, we show the effect on the probability planes when including the larger errors. These figures show that it is not possible to rule out even the highest luminosity RSGs known ( > 5.8) at the 3σ level (see Figure 7).

Figure 7. Refer to the following caption and surrounding text.

Figure 7. 2D probability planes from the CLD fitting for the upper and lower ends of the luminosity distribution. The left-hand plot shows the results when considering the standard errors, while the right-hand plot shows the results when including an extra error on the F814W detections of 0.2 dex.

Standard image High-resolution image

There is considerable additional model-dependent uncertainty when converting these luminosities to initial masses, so we have performed our analysis in terms of stellar luminosity. However, for comparison with previous results, we now convert the upper luminosity limits to progenitor mass estimates using the same mass–luminosity (M–L) relation as in S. J. Smartt et al. (2009). For standard errors, we find Mhi = 20, 23, and 25 M at the 68%, 95%, and 99.7% confidence levels, respectively. When considering the appropriate extra error on the F814W luminosity estimates, Mhi increases to 21, 24, and >25 M. 6 This upper mass limit is consistent with the observed upper mass limit for RSGs in local galaxies (e.g., B. Davies et al. 2018; S. L. E. McDonald et al. 2022) and does not support the conclusion that high-mass progenitors are missing from the progenitor sample. Using Figure 3 in S. J. Smartt (2015) we convert Llo to Mlo, finding a best-fit value of 8 M, slightly higher than the Mlo found in previous studies. This result is consistent with the canonical theoretical predictions of stellar evolution.

Figure 6 also demonstrates the discrepancy between luminosity estimates for the three well-studied progenitor cases discussed in Section 3.2.1. For each of these progenitors, we show the luminosity that is estimated from the full SED (including mid-IR photometry) with the filled blue circles. We also plot the luminosity that would have been found in each case if only F814W photometry were available, using the BC correction for both early-type (BC = 0.9) and late-type (BC = 0) RSGs, via the red open diamonds. In each of these cases the luminosity that would have been estimated from I-band photometry only is vastly underestimated compared to the SED luminosity, even when we consider the color luminosity diagram (CLD) with additional errors. Finally, we also plot the luminosity estimates for progenitors in the S. J. Smartt et al. (2009) study (i.e., using BC = 0.9) via the green symbols. These are systematically lower than the revised luminosities (blue).

4. Summary

We have examined the population of RSGs in M31 in order to inform the ways that luminosities and initial masses are inferred for more distant SN progenitors from pre-explosion photometry. To do this, we compare RSG luminosities measured from the full optical/IR SED to the luminosity inferred from a single filter detection and an assumed BC. This comparison yields two important effects: (1) luminosities inferred from single-band (usually I-band) detections are systematically lower than the full SED luminosity by about 0.3 dex, and (2) there is a significant spread in luminosity, larger than is accounted for in previous estimates of the uncertainty for SN progenitors. Together, these two effects act to eliminate the RSG problem.

This confirms the results presented in B. Davies & E. R. Beasor (2018)—that the bolometric corrections used in S. J. Smartt et al. (2009) and S. J. Smartt (2015) tend to underestimate the luminosity of SN progenitors. For the RSGs in M31, using BCs for later spectral types does reduce the systematic offset compared to SED-derived luminosities, but there is still dispersion between the SED-derived luminosities and the I-band-derived luminosities. We suggest that this dispersion is likely due to a range of unknown spectral types, extinction, and variability. When this additional error is taken into account for the sample of SN progenitors, the statistical significance of the RSG problem is further reduced.

We also use the three very nearby and well-studied examples of SN 2017eaw, SN 2023ixf, and SN 2024ggi to demonstrate the systematic underestimation of luminosity that occurs when only I-band photometry is available, compared to an SED with mid-IR data points. In each of these cases, the luminosity derived from the more complete SED yielded a higher luminosity than when using only the I-band observation. For SN 2023ixf, this effect corresponded to a difference in luminosity of over 1.4 dex, or a difference in the implied initial mass of more than 10 M. If we only had access to the I-band photometry, the luminosity would have been vastly underestimated.

We, therefore, conclude that any progenitor luminosity derived without mid-IR data is likely to be highly uncertain, perhaps underestimated by as much as 1 dex, and should not be used in an attempt to constrain the progenitor luminosity function. Given the above results, we reiterate that the statistical evidence for the missing high-mass progenitors, the RSG problem, does not exist.

Notably, these results also alleviate the tension at the lower end of the progenitor luminosity function, whereby some progenitors have been detected at luminosities that imply initial masses in the range of roughly 6–8 M. This would have been problematic since single stars in this initial mass range are too low mass to end their lives as CCSNe. This tension is eliminated if previously derived initial masses were systematically underestimated, as shown above.

Acknowledgments

The authors would like to thank the anonymous referee for constructive comments that helped improve the paper. We thank discussions with the community at the MIAPbP Interacting Supernovae workshop that motivated this work. This research was supported by the Munich Institute for Astro-, Particle and BioPhysics (MIAPbP), which is funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy—EXC-2094—390783311. This work makes use of the IDL astrolib, as well as Python packages Astropy (Astropy Collaboration et al. 2013, 2018, 2022), numpy (C. R. Harris et al. 2020), pandas (W. McKinney 2010), and matplotlib (J. D. Hunter 2007).

Footnotes

  • 4  

    We note that most RSGs do not appear to be significantly extincted by their circumstellar dust, see, e.g., E. R. Beasor & N. Smith (2022).

  • 5  

    We note that the three SN progenitors discussed in this work may not be representative of the entire RSG progenitor sample, as 2023ixf and 2024ggi are early "flash" SNe, implying they are particularly dusty objects. It is not clear how many progenitors had mid-IR coverage (e.g., Spitzer) but remained undetected.

  • 6  

    The M–L relation used previously does not extend as high as the luminosities of = 5.8.

10.3847/1538-4357/ad8f3f
undefined