The SN 2023ixf Progenitor in M101. II. Properties

We follow our first paper with an analysis of the ensemble of the extensive preexplosion ground- and space-based infrared observations of the red supergiant (RSG) progenitor candidate for the nearby core-collapse supernova SN 2023ixf in Messier 101, together with optical data prior to the explosion obtained with the Hubble Space Telescope (HST). We have confirmed the association of the progenitor candidate with the supernova (SN), as well as constrained the metallicity at the SN site, based on SN observations with instruments at Gemini-North. The internal host extinction to the SN has also been confirmed from a high-resolution Keck spectrum. We fit the observed spectral energy distribution (SED) for the star, accounting for its intrinsic variability, with dust radiative-transfer modeling, which assumes a silicate-rich dust shell ahead of the underlying stellar photosphere. The star is heavily dust obscured, likely the dustiest progenitor candidate yet encountered. We found median estimates of the star’s effective temperature and luminosity of 2770 K and 9.0 × 104 L ⊙, with 68% credible intervals of 2340–3150 K and (7.5–10.9) × 104 L ⊙, respectively. The candidate may have a Galactic RSG analog, IRC −10414, with a strikingly similar SED and luminosity. Via comparison with single-star evolutionary models we have constrained the initial mass of the progenitor candidate from 12 M ⊙ to as high as 14 M ⊙. We have had available to us an extraordinary view of the SN 2023ixf progenitor candidate, which should be further followed up in future years with HST and the James Webb Space Telescope.


Introduction
Among the supernovae (SNe) that arise from the collapse of the cores of massive (initial mass M ini  8 M e ) stars, nearly half locally are of Type II (SNe II; Smith et al. 2011)-those which show hydrogen lines in their near-maximum-light optical spectra.The overwhelming majority of SNe II further exhibit extended plateaus in their light curves and are referred to as SNe II-P.We now have compelling observational evidence, supporting the theoretical expectations, that SNe II-P are the explosions of massive stars in the final red supergiant (RSG) phase of their evolution; in particular, direct identifications have been made in more than 20 cases of RSGs as the progenitors of SNe II-P (e.g., Smartt et al. 2009;Smartt 2015;Van Dyk 2017).Most of these direct detections so far have been of relatively low M ini (∼8-10 M e ), consistent with the low-luminosity events resulting from their explosions (e.g., Smartt et al. 2004;Maund et al. 2005;Van Dyk et al. 2023a).Smartt et al. (2009) were the first to point out that a ceiling might exist on the maximum possible M ini for SN II-P progenitors, at ∼17 M e (although Davies & Beasor 2020 have since argued that this limit could extend as high as ≈20 M e ).That some Galactic RSGs appear to have M ini  25-30 M e is perplexing, although the presumed progenitor mass ceiling is still to be adequately challenged observationally (Davies & Beasor 2020 argue that the progenitor sample needs to be at least doubled).Every new example of a possible progenitor identification is therefore welcome and valuable.
The known RSG progenitors of SNe II-P have been characterized so far, mostly (although not exclusively) based on detections in more than one photometric band, such that a spectral energy distribution (SED) can be constructed and analyzed.This has been accomplished primarily via serendipitous preexplosion imaging data obtained with the Hubble Space Telescope (HST), enhanced in a few cases with further detections in data from the Spitzer Space Telescope.Solely ground-based detections have been far less common, e.g., SN 2008bk (Mattila et al. 2008;Van Dyk et al. 2012a), SN 2012A (Tomasella et al. 2013), and the most famous example, of course, SN 1987A (although the progenitor was a blue supergiant, not a red one; Gilmozzi et al. 1987;Sonneborn et al. 1987;West et al. 1987).In the lower-luminosity, lowermass cases the observed SED can be fit reasonably well with a bare photospheric model, such as SN 2008bk (O'Neill et al. 2021) and SN 2018aoq (O'Neill et al. 2019).However, other cases in which the progenitor was inevitably inferred to be of higher luminosity, higher M ini , also required the presence of circumstellar dust to be accounted for in modeling the SED, as for SN 2012aw (Van Dyk et al. 2012b;Fraser et al. 2012;Kochanek et al. 2012) and SN 2017eaw (Kilpatrick & Foley 2018;Rui et al. 2019;Van Dyk et al. 2019).
That this is true is not surprising, given our knowledge of RSGs locally.For Local Group RSGs, the more luminous the RSG, the higher is its inferred mass-loss rate (M  ), and the dustier is its circumstellar environment (Massey et al. 2005;Verhoelst et al. 2009;Bonanos et al. 2010;Mauron & Josselin 2011;Yang et al. 2018).As we pointed out in Soraisam et al. (2023b, Paper I hereafter), RSGs are also known to show pulsationally driven, semiregular variability (e.g., Heger et al. 1997;Kiss et al. 2006;Soraisam et al. 2018;Chatys et al. 2019).Additionally, for instance, in a study of Large Magellanic Cloud RSGs Yang et al. (2018) found not only correlations between luminosity and M  , and luminosity and variability, but also between variability and M;  in short, the more luminous RSG population exhibited a much larger infrared (IR) excess, M  , and variability than did the less luminous objects.
Radial pulsations of RSGs can affect the structure of the star's envelope, which further affects the luminosity evolution of the SN explosion, in that more luminous events tend to decline more steeply on the plateau, and any nonradial pulsations could also lead to observed asymmetries during the plateau phase (Goldberg et al. 2020).Additionally, injection of even a small amount (∼10 46 -10 47 erg) of deposited energy into the envelope, possibly in the form of internal waves excited by late-stage core convection (e.g., Shiode & Quataert 2014;Fuller 2017), can drive ejection of dense and confined circumstellar matter (CSM) prior to explosion, leading to hot, luminous early time emission and short-lived "flashionization" spectral features (Morozova et al. 2020).This energy injection might manifest itself as pre-SN outbursts shortly before explosion, which was problematic for, e.g., SN 2017eaw, for which no such luminous outbursts were detected in the IR (Tinyanont et al. 2019;Van Dyk et al. 2019).However, such outbursts may be more energetic and pronounced in lower-mass (M ini  12 M e ) progenitors (Wu & Fuller 2021) and may affect the optical luminosity more than the IR (Davies et al. 2022).Furthermore, Beasor et al. (2021) concluded that, if some mechanism at late nuclear burning phases led to instabilities and episodic mass loss, then a lower, steady M  at earlier phases (Beasor et al. 2020) would potentially leave the RSG with more envelope to lose in the final years of its life, leading to more CSM at core collapse.
This, then, brings us to SN 2023ixf in the famous nearby "Pinwheel Galaxy" Messier 101 (M101, NGC 5457).Given its proximity and brightness, SN 2023ixf has already achieved a level of fame of its own and has been considered the "SN of the decade," at least in the early years of the 2020s.Much has already been written about this SN, and we briefly summarize here only a portion of that.Within hours of its discovery by Itagaki (2023) on 2023 May 19.727 (UTC is adopted throughout this paper), Perley et al. (2023) classified the SN as Type II, showing a strong blue continuum and prominent optical flash-ionization features of H, He I/II, N III/IV, and C III/IV, all indicative of the presence of CSM.Early photometric and spectroscopic studies of the SN have described the evidence for interaction of the SN shock with a dense, confined (<2 × 10 15 cm) CSM, which further boosted (by 2 mag) the SN's early time optical luminosity (Bostroem et al. 2023;Hiramatsu et al. 2023;Hosseinzadeh et al. 2023;Jacobson-Galán et al. 2023;Smith et al. 2023;Teja et al. 2023;Yamanaka et al. 2023).Zimmerman et al. (2024) further constrained the confined CSM, which led to an extended shock breakout, at <2 × 10 14 cm.Smith et al. (2023), Vasylyev et al. (2023), and Li et al. (2024) presented evidence for asymmetry in the CSM and the SN ejecta.Further signs of early CSM interaction at other wavelengths include the detection of the SN in X-rays (up to 20 keV) at several epochs, starting about 4 days after explosion and beyond (Chandra et al. 2023b;Grefenstette et al. 2023;Mereminskiy et al. 2023;Chandra et al. 2024), as well as, after initial radio and submillimeter nondetections (Chandra et al. 2023a;Berger et al. 2023), detection at centimeter wavelengths ∼29 days postdiscovery (Matthews et al. 2023).Early attempts were made to detect the SN in γ-rays (Ravensburg et al. 2024) and neutrinos (Guetta et al. 2023;Thwaites et al. 2023), with null results.More recent, extended photometric and spectroscopic monitoring (Bianciardi et al. 2023;W. Zheng et al. 2024, in preparation) appears to indicate that SN 2023ixf is a short-plateau SN II-P or an SN II-P/II-Linear hybrid.
A number of constraints on the explosion epoch have been presented by both amateur and professional astronomers, which Yaron et al. (2023) initially summarized.Hosseinzadeh et al. (2023) also performed a careful analysis of the various constraints and narrowed the date of explosion to MJD 60082.75(2023 May 18.75), which we adopted in Paper I and do so here as well.
SN 2023ixf occurred just on the outskirts of the giant H II region NGC 5461 (one of the five largest and brightest in M101; Seyfert 1940), only 24 8 west and 23 0 south of the region's center.It was therefore relatively straightforward to identify a progenitor candidate for SN 2023ixf, especially at Spitzer IR wavelengths (Jencson et al. 2023;Kilpatrick et al. 2023;Niu et al. 2023;Szalai & Van Dyk 2023;Paper I), based solely on the SN's reported absolute position (Itagaki 2023).As Jencson et al. (2023), Kilpatrick et al. (2023), and Paper I found, the candidate was also detectable in various pre-SN near-infrared imaging data.A faint possible counterpart was detectable in the HST F814W (∼I) band as well (Soraisam et al. 2023a;Jencson et al. 2023;Kilpatrick et al. 2023;Niu et al. 2023;Pledger & Shara 2023).All of the available photometric information pointed to an RSG for the candidate.As a companion paper to our analysis of the progenitor candidate variability, we present here our assessment of the overall properties of the candidate.We summarize in Section 2 all of the available data that we have collected and analyzed.In Section 3 we make a precise association between the young SN and the candidate in the HST and near-infrared data.We compile the available host-galaxy distances and adopt a value (and uncertainty) in Section 4. We provide in Section 5 our estimate for the total reddening to the SN, and in Section 6 we place constraints on the metallicity at the SN site, based on a nearby H II region.In Section 7 we describe our model fitting to the candidate's SED, while in Section 8 we analyze that fit to provide an estimate of the candidate's properties.We summarize and discuss our results in Section 9.
We note that M101 is nearly face on (inclination of 8°; Jarrett et al. 2003).We assume throughout a redshift for the galaxy of z = 0.000804 (+241.0km s −1 , via the NASA/IPAC Extragalactic Database).M101 has been the host of SN 1909A andSN 1951H (both unclassified), as well as Type II SN 1970G (at the edge of another giant H II region, NGC 5455; e.g., Winzer 1974;Fesen 1993) and the wellstudied Type Ia SN 2011fe (e.g., Parrent et al. 2012;Tucker et al. 2022).

Observations
Given the proximity of M101 and its nearly face-on orientation, the galaxy and, in particular NGC 5461, have been targets of study for decades by a slew of ground-and spacebased facilities.In turn, for similar reasons SN 2023ixf has already been intently observed by a growing number of investigators.Here we summarize the observational data that we considered for this study.

HST Imaging
The SN 2023ixf site was imaged serendipitously in a number of bands by HST over nearly 24 yr prior to explosion; see Table 1 for a listing.The specific observations analyzed can be accessed via the Mikulski Archive for Space Telescopes (MAST):10.17909/j2ax-4p24.The fields were observed with the Wide Field and Planetary Camera 2 (WFPC2), the Advanced Camera for Surveys Wide-Field Channel (ACS/ WFC), and the Wide Field Camera 3 UVIS channel (WFC3/UVIS).
We processed each band in each data set individually, first by mosaicking the individual frames using Astrodrizzle (STScI Development Team 2012).An additional benefit to making the mosaics in this way is that cosmic-ray (CR) hits are flagged in the Data Quality layer of each frame.We then ran the individual frames through Dolphot (Dolphin 2016) to extract photometric measurements via point-spread function (PSF) fitting.For all of the data we adopted Dolphot parameter values FitSky = 3 and RAper = 8, InterpPSFlib = 1, with the charge transfer efficiency correction set to false for the ACS and WFC3 data (set to true for the WFPC2 data).The CRhit flagging is important, allowing for accurate aperture corrections to the PSF fitting to be estimated and applied.

Spitzer Space Telescope Imaging
In addition to the data obtained by Spitzer with the Infrared Array Camera (IRAC; Fazio et al. 2004) at 3.6 and 4.5 μm that we described in Paper I, during the Spitzer cryogenic mission the SN field was also observed on 2004 March 8 (program ID (PID) 60) at 5.8 and 8.0 μm.The field was also observed with the Multi-Band Imaging Photometer for Spitzer (MIPS; Rieke et al. 2004)

Herschel Space Observatory Imaging
The SN site was also serendipitously captured by the

AKARI Imaging
The site was observed as part of the all-sky survey scanning mode of the AKARI mission with the Far Infrared Surveyor instrument during 2006 April-2007 August at 65, 90, 140, and 160 μm.

WISE/NEOWISE Imaging
The SN site was observed as part of the all-sky survey by the Wide-Field Infrared Survey Explorer (WISE) at 3.4 (W1), 4.6 (W2), 12 (W3), and 22 μm (W4) during the cryogenic mission, from 2009 December to 2010 August; by the threeband mission segment (with W4 no longer useful) from 2010 August to 2010 September; and by the postcryogenic NEOWISE from 2010 September to 2011 February and NEOWISE-Reactivation from 2013 December to the present, in the two shortest-wavelength bands.

GMOS Spectroscopy
Long-slit spectra were pointedly obtained of SN 2023ixf with GMOS-N (Hook et al. 2004) at Gemini-North on 2023 June 3 (MJD 60098.4) as part of program GN-2023A-DD-105 ("Back with a Bang: Gemini Multiwavelength Observations of SN 2023ixf"; PI: J. Lotz).Observations were taken at the parallactic angle (Filippenko 1982) with the 0 75 wide slit using 2 × 2 binning.The B480 grating was used with two central wavelengths of 5500 and 5600 Å, chosen to mitigate against chip-gap effects, with 2 × 120 s exposures at each dither position.A spectrophotometric standard star was observed immediately after the science exposures at a similar airmass.
Raw data were made public immediately via the Gemini Science Archive, and we have reduced them using the Data Reduction for Astronomy from Gemini Observatory North and South (DRAGONS) reduction package (Labrie et al. 2019), using the recipe for GMOS long-slit reductions.This includes bias correction, flat-fielding, wavelength calibration, and flux calibration.
2.6.2.'Alopeke Imaging SN 2023ixf was also observed on 2023 June 9 with 'Alopeke on Gemini-North, as part of the above Directors Discretionary Time program.Raw data were made public immediately via the Gemini Science Archive.'Alopeke (Scott & Howell 2018;Scott et al. 2021) is a resident visiting instrument mounted on the GCAL port of Gemini-North.A dichroic that splits the incoming light at 6740 Å allows 'Alopeke to obtain simultaneous blue and red images using two identical Andor 1K frame-transfer EMCCDs.Selectable plate scales support either speckle (∼0 001 pixel −1 ) or wide-field (WF; ∼0 073 pixel −1 ) imaging, and SN 2023ixf was observed with both modes.
The speckle observations spanned times 07:54-08:15 (airmasses of 1.22-1.24)and included 20 sets of 1000 × 60 ms exposures, using the 5620 and 8320 Å narrowband filters (540 and 400 Å wide, respectively), with an EMGAIN of 1000 and a 256 × 256 region of interest (ROI), yielding an ∼2 5 field of view (FOV).These observations were immediately followed by an observation of the PSF standard star HR 5345 (airmass of 1.21), consisting of three sets of 1000 × 60 ms exposures with an EMGAIN of 20 and 30 in the blue and red, respectively.Data were reduced and final data products produced as described by Howell et al. (2011).
The 'Alopeke WF observations of the SN spanned 08:34-09:05 and were obtained through the Sloan Digital Sky Survey (SDSS) g¢ and i¢ filters.To avoid saturating the SN, 3600 exposures of 0.5 s each were collected using an unbinned 768 × 768 ROI with an EMGAIN of 5. Bias frames were collected immediately afterward, and twilight flats were obtained in the morning.All 3600 science exposures were averaged together, bias subtracted, and flat-fielded to produce the image mosaic we show in Figure 1.Both the raw and fully reduced data for these observations are available in the Gemini archive.

Keck Spectroscopy
SN 2023ixf was observed on 2023 June 7 with the High-Resolution Echelle Spectrometer (HIRES; Vogt et al. 1994) on the 10 m Keck I telescope at the W. M. Keck Observatory.Two spectra were obtained using the C2 Decker with dimensions 0 86 × 14″, which achieve resolution R ≈ 60,000 following the standard setup of the California Planet Search (Howard et al. 2010).For both spectra, an exposure meter was used to obtain a signal-to-noise ratio (SNR) of 95 per reduced pixel on blaze near 5500 Å.The first exposure began at 07:31 and lasted 199 s, while the second exposure began at 07:35 and lasted 192 s.The spectra were converted into a 1D format with a wavelength solution according to the methodology described by Petigura et al. (2017).The nominal wavelength solution is accurate to at least 1 HIRES pixel, or ∼1 km s −1 .We created coadditions of the two exposures for each order.

KAIT Photometry
Follow-up observations of SN 2023ixf were performed by the Katzman Automatic Imaging Telescope (KAIT) as part of the Lick Observatory Supernova Search (Filippenko et al. 2001).Multiband BVRI images and additional "clear" (unfiltered, close to the R band in response; see Li et al. 2003) images were obtained.Here we only focus on the B and V data.The full light-curve data set will be presented elsewhere (W.Zheng et al. 2024, in preparation).
All images were reduced using a custom pipeline14 detailed by Stahl et al. (2019).PSF photometry was obtained using DAOPHOT (Stetson 1987) from the IDL Astronomy User's Library. 15Owing to the small FOV of our images, only one reference star was available for calibration, namely star "m" from Henden et al. (2012, see their Figure 1), with the Landolt magnitudes for the star transformed to the KAIT natural system.Apparent magnitudes were all measured in the KAIT4 natural system, and the final results were transformed to the standard system using the local calibrator and color terms for KAIT4 (see Stahl et al. 2019).

Progenitor Candidate Identification
In Paper I we showed the identification we assigned for the progenitor candidate to a well-detected source in the shortest two Spitzer IRAC bands, as did Jencson et al. (2023), Kilpatrick et al. (2023), andNiu et al. (2023).This was quite straightforward, since the source was almost exactly at the refined absolute position given for the SN (Itagaki 2023).No other IR source was anywhere proximate to this position.However, to be certain, it is best to consider relative astrometry rather than rely on absolute astrometry, tying all of the observations in which the candidate is detectable to a common frame.
We have done this here, first, by astrometrically registering the 'Alopeke WF imaging of the SN from 2023 June to the HST ACS F814W mosaic from 2002 November.We isolated nine stellar objects in common between the two image mosaics and used the package photutils.centroidswithin photutils (Bradley et al. 2022) on the nine fiducials in each data set.We computed average centroids (from the four centroiding methods, centroid_com, centroid_1dg, centroid_2dg, and centroid_quadratic) for the fiducials in the 'Alopeke data, with a 1σ rms uncertainty of 0.70 pixel (50.8 mas, at ∼0 0725 pixel −1 ) and, similarly, for the ACS mosaic, with an rms uncertainty of 0.14 pixel (7.0 mas, at 0 05 pixel −1 ).With the routine geomap within PyRAF (Science Software Branch at STScI 2012), with second-order fitting, from the merged fiducial list we were able to align the two mosaics with an rms uncertainty in the mean astrometry of 0.92 ACS/WFC pixel (46.0 mas).We determined the centroid of the SN in 'Alopeke with an rms uncertainty of 0.19 pixel (13.9 mas), and transform its position with the routine geoxytran to the F814W mosaic.The overall rms uncertainty in the transformation, adding all of the above uncertainties in quadrature, is estimated to be 70.3 mas, or 1.41 ACS pixel.We isolated the red object we indicate in Figure 1 (left panel).
We measured the centroid of this object with an rms uncertainty of 0.06 ACS pixel (2.8 mas).The SN position, transformed onto the F814W mosaic, differs from the centroid of the candidate by 0.56 ACS pixel (28.1 mas).As can be seen, this difference is within the overall uncertainties in the astrometric transformation.We therefore consider it quite likely that the F814W detection is the progenitor candidate.Soraisam et al. (2023a), Pledger & Shara (2023), and Kilpatrick et al. (2023) identified this same object as the progenitor candidate.
Similarly, we register the Gemini NIRI K-continuum band mosaic (see Paper I) to the HST ACS F814W mosaic, to confirm that the object in NIRI is most likely the same as the one in ACS.We matched 20 stars in common between the two data sets and measured centroids for these with photutils, with rms uncertainties of 0.12 pixel (14.3 mas, at 0 1171 pixel −1 ) and 0.09 pixel (4.5 mas) for NIRI and ACS, respectively.The uncertainty in the geomap fit is 0.40 ACS/WFC pixel (20.0 mas), and the uncertainty in the centroid of the object in NIRI is 0.05 pixel (5.6 mas).The total error budget in this registration is then 25.6 mas, or 0.51 ACS pixel.The difference in the object's position transformed to ACS is 0.29 pixel, for a total uncertainty (including that in the F814W centroid) in the transformation of 15.0 mas.The uncertainty in the transformed position is within the total registration uncertainty, and therefore we conclude it is highly likely that the star in the F814W and in the NIRI K-continuum images are one and the same.
We dispensed here with performing a further formal registration of the NIRI data to the Spitzer detection, since we believe it has been convincingly demonstrated that the object in the near-infrared and the mid-infrared vary at essentially the same frequency and in phase (e.g., Jencson et al. 2023;Kilpatrick et al. 2023;Paper I).We consider the likelihood that two stars in immediate proximity to each other would have such strikingly similar and coordinated behavior across photometric bands and not actually be the same object to be extremely low.
We note that, in most of the available HST data, the progenitor candidate was not detected (Table 1).For these data we provide upper limits on a detection, based on the brightnesses of objects in the vicinity of the candidate's location at the formal 5σ detection level from Dolphot (see Van Dyk et al. 2023b for the rationale for this approach).
The star was also not detected in the 2004 Spitzer 5.8 and 8.0 μm data; see Figure 2. We processed the data in these two bands in a similar fashion to that described in Paper I, although instead of using APEX User List Multiframe (Makovoz & Marleau 2005), we simply used APEX Multiframe, in the SNR image input mode, to perform point-response function (PRF) fitting to individual sources in the data.We located the faintest sources that were detected in the overall environment of the progenitor candidate and, from the brightnesses of these sources, we set upper limits on the candidate's detection of 45 and 128 μJy at 5.8 and 8.0 μm, respectively (these limits correspond to SNR = 12 and 29 in each of these two respective bands).
Owing to its close proximity to NGC 5461, and the comparatively inadequate spatial resolution and sensitivity of the Spitzer MIPS, Herschel, and AKARI instruments, and (to a lesser extent) WISE W3 and W4, the progenitor candidate was completely overwhelmed by the luminous emission from that giant H II region at wavelengths 11 μm.We estimated the flux at the exact SN position in these data sets and assumed these fluxes as the upper limits on the progenitor's brightness.None of these limits were particularly constraining: respectively, <0.025, <0.64, and 3.43 Jy at Spitzer 24, 70, and 160 μm; <0.018 and <0.252 Jy at Herschel PACS 70 and 160 μm; <0.296, <0.161, and <0.058 Jy at Herschel SPIRE 250, 350, and 500 μm; and <3.78, <3.29, <6.37, and <1.17 Jy at AKARI 65, 90, 140, and 160 μm.For WISE W3 and W4 we estimated limits that were considerably more constraining: <0.00031 and <0.00125 Jy, respectively.
We did visually inspect all of the hundreds of publicly available WISE and NEOWISE W1 and W2 single exposures up to 2022 May 24, in which the site is distinguishable from its general environment, and the progenitor candidate was not detectable in any of these frames (see also Hiramatsu et al. 2023).Based on the 3σ detections of stars within 60″ of the SN position in each of these data sets, we estimated respective limits on the progenitor candidate detection of >15.83 and >16.33 mag (<0.000144 and <0.000050 Jy) for cryogenic W1 and W2; >16.24 and >15.14 mag (<0.000099 and <0.000151 Jy) for postcryogenic W1 and W2; and >16.42 and >14.88 mag (<0.000084 and <0.000192 Jy) for NEOWISE W1 and W2.
That the star was not detected is not surprising, since we found in Paper I that the peak brightnesses were ∼17.0 and ∼16.5 mag in Spitzer 3.6 and 4.5 μm (respectively) between 2004 March and 2019 October.These limits do, however, rule out any luminous outbursts at these wavelengths in the 2019 October to 2022 May period, i.e., ∼1300 to ∼360 days prior to explosion.This is consistent with the findings by Jencson et al. (2023), based on their J-and K-band data.
We note that Kilpatrick et al. (2023) identified "Source B" in the HST F814W image mosaic, ∼0 1 from the progenitor candidate, their "Source A." We also recognize the presence of what appears to be an "appendage" to the candidate (Kilpatrick et al. 2023 characterized this source as being to the northeast of the candidate, whereas to us, it appears more-or-less due east).However, we call into question here whether this is a real object.First, we visually inspected each of the two frames which comprise the mosaic and find that this other source appears more prominently in one frame than the other.Additionally, although Dolphot, when run on bands F435W, F555W, and F814W all at once, detects it as a separate source, its object type is "4," which indicates that the object is "too sharp" and likely not a good stellar detection.We do find that, when Dolphot is run on F814W and F555W separately (the source is definitely not detectable at F435W), the routine does identify it as a stellar source, albeit too faint for PSF determination in the latter band.If real, we estimate the Kilpatrick et al. (2023) "Source B" is less than a third as bright as the progenitor candidate, so its influence on the photometric measurements of the candidate is relatively minimal, which, as we show in Section 7, would have little effect on the characterization of the progenitor's properties, particularly given the likely variability of the candidate in F814W.
Finally, we can constrain the presence of a neighboring source at the SN location, based on the 'Alopeke speckle observations; see Figure 3.The speckle image exhibits no indication of any fainter, neighboring source at 0 1 to a level ∼5 mag below the SN brightness, out to nearly 9 mag at 1 2 in the F814W (∼I band) equivalent.On June 9, when the speckle observations were obtained, the SN was at I = 11.10 mag (and V = 10.81 mag) from the KAIT photometry.Hence, neighbors can be ruled out to I ≈ 16 mag at 0 1 and ∼20 mag at 1 2. While we cannot from these observations eliminate a much fainter star, in particular, the presumed "Source B," we can say that, at least for the SN in its bright state on the plateau, there would be a negligible effect on the SN light from any neighboring object.

Host-galaxy Distance Estimation
A number of modern, reliable distance estimates exist for M101, either from Cepheid or from tip-of-the-red-giant-branch (TRGB) measurements.For the latter these include distance moduli μ = 29.30± 0.12 mag (distance d = 7.24 ± 0.40 Mpc; Lee & Jang 2012), 29.16 ± 0.13 mag (6.79 ± 0.41 Mpc; Tikhonov et al. 2015), and 29.07 ± 0.06 mag (6.52 ± 0.18 Mpc; Beaton et al. 2019).The Cepheid distances include 28.96 ± 0.11 mag (6.19 ± 0.31 Mpc; Mager et al. 2013), 29.13 ± 0.19 mag (6.70 ± 0.59 Mpc; Nataf 2015, which is a reanalysis of the distance estimate by Shappee & Stanek 2011 with a larger uncertainty), 29.14 ± 0.09 mag (6.71 ± 0.28 Mpc; Foley et al. 2020), and 29.18 ± 0.04 mag (6.85 ± 0.13 Mpc; Riess et al. 2022).Since the TRGB estimates are sensitive to contamination of halo populations by younger stellar populations, and therefore to field choice and availability, our predilection is to select a Cepheid-based distance, and we choose the Riess et al. (2022) distance; to be somewhat more conservative, however, we adopt an uncertainty of 0.1 mag, which is the weighted mean of the uncertainties in the individual Cepheid estimates (as Riess et al. 2022 pointed out, the Cepheids studied by Mager et al. 2013 included stars with long periods that were not adequately sampled and thus could have larger uncertainties in their derived periods; Riess et al. 2022 excluded these long-period Cepheids, and we therefore opted not to include the Mager et al. 2013 result in the distance we adopt here).Therefore, we adopt d = 6.85 ± 0.32 Mpc (29.18 ± 0.10 mag) for the distance to SN 2023ixf.(Since the submission of this paper, Huang et al. 2024 presented a somewhat shorter distance, 6.61 ± 0.18 Mpc, μ = 29.10 ± 0.06 mag, based on Mira variables; however, as those authors pointed out, this distance agrees with the Cepheid-based distance to within the uncertainties.).

Host Reddening Estimation
Next, it is essential, in order to detail the intrinsic nature of the progenitor candidate, to estimate the amount of extinction to SN 2023ixf.The assumption we have made here is that the total interstellar extinction to the SN is the same as to the progenitor candidate.We are not accounting for any circumstellar extinction related to the candidate; that we will do in Section 7. The Galactic foreground contribution to the total extinction is likely to be quite low, A V = 0.024 mag (Schlafly & Finkbeiner 2011, via the NASA/IPAC Extragalactic Database (NED)).The question then centers on the contribution internal to the host galaxy.Blue and red objects are seen throughout the HST image mosaic, shown in Figure 1, in the general vicinity of the progenitor candidate-in fact, a blue star is seen just 0 3 to the southeast of the candidate-and the reddening appears patchy and variable.The reddening to the progenitor candidate is likely not extreme, since, as can be seen in Figure 2, the star appears visually to be located in a region of relatively low (although not negligible) 8 μm emission-the IRAC 8.0 μm From the 8320 Å narrowband filter observations, we can rule out any neighboring object to ∼5 mag and ∼9 mag below the SN brightness at 0 1 and 1 2, i.e., to I ≈ 16 and ≈20 mag, respectively.band probes the strong molecular complex, composed of the 7.7, 8.3, and 8.6 μm polycyclic aromatic hydrocarbon (PAH) features, often associated with interstellar dust (e.g., Gordon et al. 2008).
We can look to constrain the host reddening from spectroscopic data.This has already been attempted by both Lundquist et al. mag.This is about a factor of 2 larger than the NED value, A V = 0.024 mag.Since the overwhelming majority of the detected flux from the progenitor candidate emerges in the IR (see Section 7), the effects of reddening turn out to be quite minor.The Galactic foreground component is comparatively small, regardless of whether we adopt the value inferred from the HIRES spectrum or from Schlafly & Finkbeiner (2011) directly; we choose to adopt the latter for the foreground extinction, thus keeping us consistent with the assumed value in the previous studies of SN 2023ixf.We also attempted to locate the DIB λ5780 feature in the HIRES spectrum.As Phillips et al. (2013) demonstrated, there appears to be a correlation between the EQW of this DIB feature and A V .We modeled the feature assuming a Gaussian shape with an FWHM of 2.1 Å and an EQW corresponding to A V = 0.099 mag (see Phillips et al. 2013, their Equation (6)), the internal extinction value we inferred from the Na I D lines.We also shifted the model in wavelength to match the velocity offset of +21 km s −1 measured from the Na lines.In addition, we produced a model of the feature based on a 3σ upper limit on the feature's EQW.As can be seen in Figure 4, unfortunately, the HIRES spectrum was not of sufficient SNR for the feature to have been detected.
Finally, we can at least qualitatively assess the effect of any host reddening on SN 2023ixf, based on a color comparison between B − V obtained with KAIT and samples of SNe II-P with the bluest colors and with no significant host-galaxy reddening from de Jaeger et al. (2018).We also include the luminous SN 2017eaw, for which Van Dyk et al. (2019) concluded that the reddening was almost entirely from the Galactic foreground, as well as two short-plateau SNe, SN 2006Y and SN 2006ai (Hiramatsu et al. 2021), which also experienced very little host reddening (Anderson et al. 2014).We also show a model of the feature, assuming a Gaussian with FWHM = 2.1 Å and an EQW corresponding to A V = 0.099 mag (see Phillips et al. 2013), shifted by a velocity offset of +21 km s −1 (pink dashed curve).Also shown is a similar model, based on a 3σ upper limit of the EQW (green solid curve).The wavelength axis for both panels is in the host-galaxy rest frame.
We show this comparison in Figure 5.The colors for all of the SNe shown have only been dereddened by removal of the Galactic foreground component in each case.While a fair amount of observational scatter exists in the SN 2023ixf data, the colors of the SN seem to be consistent with the other events with which we compare.de Jaeger et al. (2018) cautioned that the physical origins of observed SN II color differences are not well understood, and factors, such as CSM interaction, possibility of dust destruction and variable reddening with time, further complicate our understanding, such that a significant dispersion in intrinsic SN II colors precludes the existence and development of a "color template."Nevertheless, any color offset for SN 2023ixf, relative to the comparison samples, appears to be relatively minimal, further implying that the host reddening to the SN must be low.
Note that by day ∼ 57, the SN 2023ixf color began to diverge redward relative to the other SNe shown in Figure 5. From more extensive photometric information available for the SN (W.Zheng et al. 2024, in preparation), it is evident that by day ∼ 60 the SN was already falling off the plateau in both B and V.By day ∼ 80 the SN was approaching the exponential decline tail-hence, by that age SN 2023ixf had reached a B − V color (∼1.5 mag) that, for example, the more "photometrically normal" Type II-P SN 2017eaw did not reach until day ∼ 120 (e.g., Van Dyk et al. 2019).Therefore, SN 2023ixf was simply experiencing the expected, normal color evolution, only "accelerated" as a result of the comparatively shorter plateau.We are unsure what impact this had on the circumstellar environment-only that the progenitor likely had less envelope mass than, for instance, the SN 2017eaw progenitor, and hence the H recombination wave passed through the cooling ejected envelope over a shorter timescale.The lower envelope mass is entirely consistent with the existence of extensive CSM around the progenitor star.
The total, Galactic plus internal host, extinction we adopt hereafter for SN 2023ixf therefore is A V = 0.12 mag.We further adopt the Cardelli et al. (1989) reddening law in the optical and near-infrared, and Indebetouw et al. (2005) for the mid-infrared, with R V = 3.1 (appropriate for diffuse interstellar dust).Ultimately, however, it matters little which laws we adopt here, since the total reddening is relatively low.

Supernova Site Metallicity
At the adopted distance of 6.85 Mpc, SN 2023ixf is located ∼8.8 kpc from the center of M101.From the radial gradient in the oxygen abundance (often assumed as a proxy for metallicity) established by Garner et al. (2022)  We compared our measurements with those by Kennicutt & Garnett (1996), who had spectroscopically observed this H II region previously.Any differences we found may arise from the fact that Kennicutt & Garnett (1996) presumably centered on the Hα peak of H1086, whereas, as we show in Figure 7, the GMOS slit passed through the southwestern edge of that H II region.In the first place, the reddening value we estimated is larger than that for H1086 from Kennicutt & Garnett (1996); however, it is reasonable to assume that the reddening may well be variable across the region.Furthermore, Kennicutt & Garnett (1996)  ), the range of O abundances from the nearby emission tends to imply-assuming that emission at that location is representative of the SN 2023ixf site-that the metallicity at the SN site is somewhat subsolar to somewhat supersolar.The O 3 N 2 indicator, in particular, one of the more robust (and nondegenerate) metallicity diagnostics considered here (e.g., Kewley & Ellison 2008), tends to point toward above solar.We will therefore consider solar, subsolar, and supersolar metallicities below when analyzing the properties of the progenitor candidate.We note that Zimmerman et al. (2024) inferred from absorption lines detected in HST ultraviolet (UV) spectra of the SN, likely arising from the star's CSM, that the CSM metallicity was approximately solar.From observations of emission regions near the star those authors also inferred solar to slightly subsolar metallicity.Moreover, we note that any post-SN spectroscopic analysis of the host site's metallicity will likely be at least somewhat contaminated by the light of the SN itself for several years to come.

Spectral Energy Distribution Fitting
We assembled all of the data we have collected for the progenitor candidate, from the optical to the mid-infrared, over a span of years from 1999 through 2019.We demonstrated in Paper I that the progenitor candidate is highly variable in the IR (see also Jencson et al. 2023;Kilpatrick et al. 2023;Niu et al. 2023).Notable gaps exist in the coverage of its light curve, so we cannot strictly rule out the existence of excursions beyond regular variability.However, we have shown in Paper I that the observed data are consistent with periodic behavior.In order to model the observed SED we first established the mean brightness in each of the observed bands.For JHK s , this could be accomplished in two different ways, either by simply averaging the observed data or by using the reconstructed light curves (making sure to restrict this to some integer factor of the ∼1091 day period).For the former, the result is J = 20.39,H = 19.63,and K s = 18.61 mag.For the latter, the values are quite similar: J = 20.39,H = 19.61,and K s = 18.60 mag.Uncertainties (1σ) from the observational data, without any  weighting, are 0.07, 0.06, and 0.04 mag in J, H, and K s , respectively.
For the Spitzer data, we combined all of the "Warm" (noncryogenic) data in each band, effectively averaging over all of those observations-note that the single cryogenic data point in each band was excluded here.At 3.6 μm, 268 individual Warm Spitzer frames contain the progenitor site.We ran MOPEX (Makovoz & Khan 2005) and APEX Multiframe on those frames, following the procedures in Paper I, and extracted a flux density for the progenitor candidate of 31.12 ± 0.12 μJy at SNR = 252 (at the position α = 14 h 03 m 38 572, 54 18 42.11; d = +  ¢  J2000).Similarly, there are 256 frames at 4.5 μm.From that band we extracted a flux density of 25.18 ± 0.09 μJy at SNR = 293 (position α = 14 h 03 m 38 587, 54 18 42.06 d = +  ¢  ).Note that these flux densities have been aperture and pixel-phase corrected;16 however, they were not color corrected17 (although such correction for an ∼3000 K blackbody is effectively unity).
To convert the Spitzer flux densities into Vega magnitudes, we adopted the IRAC zero-points, 272.2 ± 4.1 and 178.7 ± 2.6 Jy, at the nominal channel wavelengths of 3.544 and 4.487 μm, respectively.We list the measured magnitudes (Vega) in all of the bands in Table 3.We also provide in the table both the assumed Galactic foreground and internal host reddening in each band, as well as the resulting reddening-corrected flux densities.
Since the original submission of this paper, Liu et al. (2023) published a detection of the SN progenitor candidate in the SDSS z band at 22.78 ± 0.06 mag, and we have now included this in our analysis.
We incorporated an estimate of the source variability into the flux-density uncertainties as follows (see Riebel et al. 2012, e.g., their Section 2.2).We adopted the amplitudes estimated in Paper I for the J, H, K s , and Spitzer 3.6 and 4.5 μm light curves to estimate the range of flux variation in those bands.Smith et al. (2002, see their Figure 5(a)) presented the relationship between the J-band and V-band amplitudes for their sample of IR-bright Mira variables (which are not too dissimilar from variable RSGs).We fit a line to those data and used it to predict the V-band amplitude for the progenitor candidate, from its J-band amplitude in Paper I; we obtained a value ∼ 2.6 mag, consistent with the source being a long-period variable.This amplitude was used to estimate the range of flux variations for the optical bands, in particular, F814W.We added the corresponding flux variation to the measurement uncertainty in quadrature.We also assumed that the z-band variability amplitude was identical to that in the V band.
We show the reddening-corrected observed SED with the now inflated uncertainties, as described above, in Figure 8.We find that the progenitor candidate SED can be approximated initially by a simple blackbody; an observer within the host galaxy near the star would have assessed it as a rather cool (∼1761 K) and quite luminous (∼1.11 × 10 5 L e , integrating the blackbody and applying our adopted distance to the host) object.The fact that the z-band brightness measurement, obtained from an independent investigation, aligns smoothly (both with and without the inflated uncertainties) with our measurements in the other bands provides us with some confidence in the veracity of our methods and our corresponding conclusions.
We have performed a more rigorous fit of the SED with O-rich dust models from the Grid of Red supergiant and AGB ModelS (GRAMS; Sargent et al. 2011;Srinivasan et al. 2011); see Figure 9.The choice of O-rich dust for the progenitor candidate is justified, given its RSG nature (e.g., Seab & Snow 1989).The GRAMS O-rich models are constructed using the PHOENIX model photospheres at solar metallicity (Kučinskas et al. 2005(Kučinskas et al. , 2006) ) with surface gravity ( [ ]) g log cm s 0.5 2 = --, and with optical constants for O-deficient silicates from Ossenkopf et al. (1992), assuming a spherically symmetric dust shell of inner radius R in (which is effectively set by the condensation temperature of silicate dust) with a constant M  (and, hence, an inverse-square density falloff) and an outer radius 1000 times that of R in .(The SN site may be at subsolar or supersolar metallicity; however, even though the input photospheres are at solar metallicity, as we will see the optical depths of the best-fit models are sufficiently high that the underlying stellar characteristics are virtually impossible to distinguish.) A preliminary fit to the SED with the GRAMS models resulted in an optical depth τ of one at 10 μm; however, the optical depth sampling in this range is quite poor (available models were computed with τ = 0.5, 1, and 2), resulting in unreliable estimates of the effective temperature (the best-fit model corresponds to the lowest temperature available in the grid, consistent with the T eff distribution experiencing a slower falloff toward cooler temperatures).Robust parameter estimates derived from fitting the progenitor candidate SED therefore cannot be obtained unless we increase this sampling.We accomplished this by training an artificial neural network on the GRAMS grid to predict spectra and dust-production rates (DPRs) for arbitrary parameter combinations.We then performed an MCMC procedure using the emcee package (Foreman- Mackey et al. 2013), employing the neural network to predict spectra for the parameter combinations explored by the MCMC.The MCMC fitting does not fit any of the upper limits; however, the best-fit models are consistent with these limits.We treated all of the other available data as upper limits (including the detections at HST F656N, F673N, and F675W) and similarly did not fit them.Another aspect that concerned us was the value of the K s brightness measured from the Gemini NIRI image.In Paper I we photometrically calibrated this brightness to the Two Micron All Sky Survey (Skrutskie et al. 2006), as did other studies (Jencson et al. 2023;Kilpatrick et al. 2023;Qin et al. 2023).However, the filter through which the NIRI observations were made was not K s ; instead, it was a narrower, contiguous K-continuum filter used as an "off-band" for Brγ imaging.(The K-continuum bandpass, however, is within the broader K s bandpass.)For a reddened 1761 K blackbody (referring to Figure 8), a difference ∼ 0.09 mag exists between the synthetic photometry through these two K filters.Additionally, applying synthetic photometry to the suite of GRAMS models, for J − K s ≈ 3.2 mag, the approximate color of the progenitor candidate, the photometry differs between the two filters by 0.31 ± 0.14 mag (molecular absorption features in the input photospheric emission, only moderately diminished by the silicate dust, fall within the K-continuum bandpass; hence, the larger difference compared to a continuum-only blackbody).Furthermore, the MCMC can model the calibration as a nuisance parameter, and from this treatment we found that the uncertainty in the calibration may be 0.40 mag.We initially computed fits with and without K s , and found very little difference between the results.This is not surprising, since the SED for this dusty star is quite well sampled, even in the absence of information in K s .Given the considerable uncertainty in the K s calibration, we chose hereafter to conduct the fitting without this band.
We present in Figure 9 the best-fit GRAMS model spectrum for the progenitor candidate SED.We show in Figure 10 a corner plot with the distribution of the resulting values of L bol , T eff , R in , and τ at 10 μm (τ 10 ).While L bol and τ 10 are well constrained, T eff is not.This is a direct result of the large degree of dust obscuration, which prevents a precise determination of the properties of the photosphere, and the lack of tighter constraints on the dust content owing to the availability of only two mid-infrared detections.The range of acceptable T eff values gradually increases with increasing obscuration, as seen in Figure 10.This degeneracy is responsible for the large range in the predicted T eff .The 68% credible interval for the inner radius of the dust shell is 6.5-13.3R å .It should be noted that with increasing optical depth, emission from the innermost regions of the shell is obscured enough that the models overestimate the inner radius of the dust shell.Even if this were not the case, the range of values for the inner radius is

Progenitor Properties
The best-fit parameters derived from the GRAMS grid are given in Table 4.These parameters describe properties of the photosphere (L bol and T eff ), as well as the dust shell (R in , relative to the effective stellar radius R å , DPR, τ 10 , and the τ of the circumstellar shell at 1 μm, τ 1 ).Here, L bol is obtained by integrating under the entire model SED.For each parameter, we show in the table the MLE from the MCMC sampling, as well as the median (50%) and 68% credible intervals of the values.Based on these results, we find an L bol and DPR of L 9.0 10 yr −1 , respectively, for the progenitor candidate.We quote here the median values, as they are more representative of the typical range of parameter values than the MLE.As discussed in the previous section, T eff is not well constrained; we find a median of 2770 K, with a 68% credible interval of 2340-3150 K.The stellar radius R å is in the approximate range 921-2012 R e , with median 1389 R e .
GRAMS models assume an expansion velocity of 10 km s −1 , typical for AGB stars and RSGs (the common range observed for RSGs is 10-20 km s −1 ; e.g., Decin et al. 2006;Humphreys & Jones 2022;Decin et al. 2024).Kilpatrick et al. (2023), instead, assume 50 km s −1 , which is more appropriate for hypergiants (e.g., VY CMa) or for superwinds associated with enhanced mass loss prior to explosion (e.g., Shivvers et al. 2015).Unfortunately, the lack of Spitzer data near the epoch of explosion makes it challenging to verify directly the superwind scenario (see Section 9 below).Since the DPR scales linearly with this parameter, we note that the DPR quoted in Table 4 must be multiplied by 5 to compare it with the DPR value from Kilpatrick et al. (2023), 1.3 ± 0.1 × 10 −8 M e yr −1 .The latter value is therefore ∼20× lower than ours, partly because they assumed graphitic dust (see Section 9.2 for more about this assumption), whose higher opacity results in a lower dust mass, in order to reproduce the observed mid-infrared flux.
Since several studies employed the dust radiative-transfer code DUSTY (Ivezic & Elitzur 1997;Ivezic et al. 1999;Elitzur & Ivezić 2001) in their analysis of the progenitor candidate (Kilpatrick et al. 2023;Qin et al. 2023;Neustadt et al. 2024;Xiang et al. 2024), we have also modeled the star in a similar fashion.The primary goal of doing so is not for direct comparison with these other investigations per se, but as a confirmation of our results with the GRAMS modeling.Here we have assumed for the central source the PHOENIX model photospheres employed for the GRAMS models (rather than the Gustafsson et al. 2008 MARCS atmospheres used by the other studies).Also, similar to our GRAMS modeling, we have assumed the Ossenkopf et al. (1992) "warm" silicates for the dust and the same spherically symmetric dust-shell configuration; additionally, we assumed the "modified Mathis-Rumpl-Nordsieck" (Mathis et al. 1977) dust-grain distribution.The grid of models we considered was quite coarse, since our intent here was only to present an approximate comparison to our GRAMS results and not necessarily provide a robust fit to the data: the model photosphere steps in T eff were 200 K, the steps in T in (the inner temperature of the dust shell) were 100 K, and the optical depth at 0.55 μm, τ V , was considered in steps of unity.We also employed only a simple chi-squared, χ 2 , goodness of fit.We similarly excluded the observed K s data point in the fitting.The results are shown in Figure 11.The best-fitting model is obtained from inputs T eff = 2500 K, T in = 1300 K, and τ V = 10, although a number of other models also providing reasonable fits have input ranges of T eff = 2300-2700 K, T in = 900-1300 K, and τ V = 9-11.Models outside of these ranges provided significantly poorer fits.The luminosity of the best-fitting model is L bol = 8.2 × 10 4 L e , which is somewhat less than the MLE and median values, however, certainly within the credible interval, for the GRAMS models.The T eff is somewhat lower than the GRAMS MLE and median, although certainly within the credible interval.The τ V = 10 corresponds to τ 1 = 2.9 and τ 10 = 0.9, which are both somewhat larger than the GRAMS values, although the latter optical depth is just within the credible interval.The inferred R in /R å , resulting from T in , L bol , and T eff , is 3.4, with R å ≈ 1534 R e .Whereas R in /R å is significantly smaller than the credible interval from the GRAMS modeling, R å itself agrees with the GRAMS results.Despite the inputs and assumptions for both the GRAMS and DUSTY modeling being essentially the same, we found some differences as described here.Ueta & Meixner (2003) compared 2DUST, the source routine behind the GRAMS models, with DUSTY and found overall good agreement in the results, although the differences in the treatment of geometry, particularly at the inner shell edge, likely accounted for some minor discrepancies.Overall, though, whether we had singularly relied on GRAMS or DUSTY, we ultimately would have arrived at quite similar T eff and L bol estimates for the star.
In Figure 12 we present a Hertzsprung-Russell (H-R) diagram with the locus of the progenitor candidate defined by the 68% credible intervals of the values for T eff and L bol from Table 4.For comparison we show single-star theoretical evolutionary tracks from Stanway & Eldridge (2018) at solar (Z = 0.020), subsolar (Z = 0.010), and supersolar (Z = 0.040) metallicities.One can see that the higher-metallicity supersolar tracks provide better agreement with the progenitor candidate's location in the H-R diagram, given that these models have a correspondingly cooler Hayashi limit.Based on the end points of these models (we note that the termini for these BPASS tracks are at the end of carbon burning), we can infer that M ini for the progenitor candidate ranges from 12 M e to as high as 14 M e , depending on metallicity.These model stars all terminate within a range of effective radii R eff ≈ 740-1028 R e .
We also considered BPASS binary models, at solar, subsolar, and supersolar metallicity, for which the primary in the system terminates with the appropriate range in L bol and with T eff 3700 K. Furthermore, we imposed the criteria from Eldridge et al. (2013Eldridge et al. ( , 2017) ) for the primary to end as an SN II-P: total primary mass M M 1.5 prim   , CO core mass 1.38 M e , ONe core mass 0.1 M e , total H mass > 1 M e , and the ratio of total H-to-He mass > 1.05 M e .We eliminated from consideration model systems, all with initial periods <4 days, for which the orbital separation wound up well inside the primary's envelope at terminus, with the remainder being wide binaries with initial masses for the primary of M 12 prim = -14 M e (up to 15 M e for the supersolar models) and initial periods between ∼1000 and ∼10,000 days (orbital separations from ∼1150 to ∼5968 R e ) over a wide range of initial companion-to-primary mass ratios q M M comp prim

=
. The evolution of the primary was essentially unaffected by the presence of the companion, and the primary's track was the same as if it were a single star.The final separations are generally in the range of 1.1-7.7 R å .However, for just a few allowed model tracks, the primary is in a longperiod system, and the star appears to lose more mass during its lifetime than it would if it were single (based on the BPASS M   (Ivezic & Elitzur 1997;Ivezic et al. 1999;Elitzur & Ivezić, 2001).Left: the reddeningcorrected observed SED (black squares), together with the best-fitting DUSTY model (solid red curve), with input parameters T eff = 2500 K, T in = 1300 K, and τ V = 10, and a sample of other good-fitting models (gray curves; see text).The T eff = 2500 K PHOENIX photosphere (dotted purple curve) at the inferred L bol (see text) is also shown for comparison.The error bars include a combination of the measurement uncertainties, as well as the range in flux in each band as a result of source variability.Unlike Figure 9, only flux upper limits in the Spitzer IRAC 5.8 and 8.0 μm and WISE W3 bands are shown, since these are the most constraining.The model predictions are consistent with these limits.The blue curve represents a DUSTY model computed with a central source comprised of a T eff = 2500 K RSG photosphere, plus a hot 24,500 K source (a Castelli & Kurucz 2003 B1V model atmosphere, meant to approximate a putative binary companion) with ∼0.14× the inferred L bol of the best-fitting RSG model (see right panel), assuming the same input parameters.Right: the input SED to DUSTY for the putative binary system (solid blue curve), consisting of a T eff = 2500 K RSG photosphere (dotted red curve, in this panel) as the primary and the less luminous, hot (24,500 K) companion (dashed magenta curve).prescription), while the companion gains some mass, presumably through some level of interaction with the companion.Although the details of these models may not fully apply, they are at least suggestive that a (close) binary companion could have induced additional mass loss from the primary at some point prior to explosion, which could be relevant for the SN 2023ixf progenitor.

Summary
In this paper we have analyzed the combination of the ground-based near-infrared and Spitzer data from Paper I, together with the available HST data, on the SN 2023ixf progenitor candidate in M101.The only other hybrid, spaceand ground-based SN II-P progenitor identifications of which we are aware have been for the nearby Type II SN 2003gd (Van Dyk et al. 2003;Smartt et al. 2004) and SN 2004et (Li et al. 2005;Crockett et al. 2011).Quite notably, unlike what has been the case in the past for a large number of progenitor identifications, the HST pre-SN data were of comparatively little value in understanding the progenitor candidate's nature, since the overwhelming majority of the star's emission prior to explosion was emerging in the IR.We confirmed that the progenitor candidate at HST F814W is most likely associated with the SN, via observations of SN 2023ixf with the 'Alopeke instrument at Gemini-North, and also found that the HST detection is most likely the counterpart of the detection at 2 μm (additionally, the near-infrared detections are the counterparts of the Spitzer detections).
We have confirmed that the reddening internal to the host galaxy is likely quite low, via measurements of features, in particular Na I D, in a high-resolution Keck HIRES spectrum of the SN; this conclusion is also supported by the early time B − V color of the SN, after correction for the Galactic foreground.We have adopted a total visual extinction A V = 0.12 mag, including the Galactic foreground.We have also assessed a likely value for the metallicity at the SN site, based on a spectrum of nebular emission at the outskirts of the cataloged H II region H1086 ∼8 3 northwest of the site, and concluded from various strong-line indicators that it was likely somewhat subsolar (Z = 0.010) to supersolar (Z = 0.040).
We have employed dust radiative-transfer GRAMS models to fit the observed SED of the progenitor candidate, corrected first for the total reddening.We find that the star is heavily dust obscured (high τ) from a likely dusty circumstellar shell or shells.The properties of the star correspond to a median of T eff = 2770 K and L bol = 9.0 × 10 4 L e , with 68% credible intervals of 2340-3150 K and (7.5-10.9)× 10 4 L e , respectively.The 68% credible interval for the DPR is (4.0-12.8)× 10 −8 M e yr −1 .We have also performed DUSTY modeling of the observed SED and found consistent values for L bol and T eff ; whether we approached the modeling via GRAMS or DUSTY, we arrived at the same overall picture of the progenitor candidate's nature.
As we had stated in Section 8, the low values of T eff resulting from the modeling are not well constrained, as a result of the large dust obscuration of the star.An illustrative example applies to the extreme dusty RSG, VY Canis Majoris, for which T eff from modeling results in ∼2800 K (e.g., Monnier et al. 1999), whereas the luminosity and diameter estimates from interferometry imply ∼3500 K (e.g., Wittkowski et al. 2012).
We have placed the median values, together with the 68% credible intervals, for T eff and L bol on an H-R diagram and have compared these to the end points of BPASS single-star models at both solar and subsolar metallicities, concluding that the progenitor candidate likely had M ini = 12-14 M e , depending on metallicity.This mass range is consistent with the results of hydrodynamical modeling of the bolometric light curve by Bersten et al. (2024).Binary models are also possible; however, the overwhelming majority of these models correspond to long-period, wide binaries, in which the primary evolves essentially as a single star, with little effect on it from its binary companion.

Comparison with Previous Studies
Our inferred range in this paper for M ini , based on the SED modeling, is significantly lower than what we estimated in Paper I, based on the RSG period-luminosity relation: We can also compare with the estimates of M ini for the progenitor candidate in the previous studies by other investigators.Pledger & Shara (2023) found a low M ini ≈ 8-10 M e ; however, their estimate was entirely based on the available HST data, whereas we have shown that the vast majority of the flux from the star must be emerging in the IR.We do note that our value at HST F814W agrees, to within the uncertainties, with that of Pledger & Shara (2023; as well as with Soraisam et al. 2023a;Kilpatrick et al. 2023).Kilpatrick et al. (2023) also concluded that M ini was comparatively low, at 11 M e .Those authors modeled the candidate assuming MARCS atmospheres as input and employing DUSTY, which led them to a hotter T eff ≈ 3920 K and a lower L bol ≈ 10 4.74 L e than the results of our modeling.
Although Kilpatrick et al. (2023) also used Dolphot, even after converting their HST measurements of detections from AB to Vega magnitudes, our measurements were generally systematically brighter (by as much as ∼1.7 mag, in the case of the 1999 F675W observation); interestingly, our F814W measurement agrees with theirs, to within the uncertaintiessee Table 1 (see Kilpatrick et al. 2023, their Table 1).We cannot provide any definitive explanation for these differences, since there are a number of input factors that may have led to them.We also cannot explain the differences in the upper limits to detections, although these differences are not extreme-as we explained above, we have based our limits on 5σ Dolphot detections in the overall vicinity of the SN site, whereas their threshold was not explicitly spelled out in their paper.Note that we detected a source at the SN position in the 2014 F673N observation, whereas they did not; and we considered the 2014 F502N data, and they had not.In Paper I we already described the differences of our measurements in the IR with those by Kilpatrick et al. (2023), which may also work to contribute to the differences in L bol and M ini .
We also found differences between our Dolphot photometry and that by Niu et al. (2023).The values in that study all tended to be fainter than ours-at the extreme, their detection at F675W differed by ∼0.9 mag-and this also applied to the upper limits (although those differences, again, were not severe).Niu et al. (2023) also considered far less of the available archival HST data than we (or Kilpatrick et al. 2023) did.We stand generally by our Dolphot results, since our methods and procedures have been developed over the years in consultation with the Dolphot author and following the techniques of large HST photometric surveys (see Van Dyk 2017).
The Spitzer flux densities measured by Niu et al. (2023) are on average ∼13 and ∼12 μJy at 3.6 and 4.5 μm, respectively (or ∼0.5 mag in both bands), fainter than the values we measured.We note that Niu et al. (2023) performed PSF-fitting photometry with DoPHOT on the postprocessed (mosaicked) Basic Calibrated Data; no details were provided regarding calibration of the photometry or how the PSF was modeled.In general, attempting point-source-fitting photometry of the IRAC mosaics is not recommended, since the mosaicking process both blurs the undersampled point sources and loses the pixel-phase information. 18PRF fitting with APEX of the individual (corrected) Basic Calibrated Data frames, as we have implemented both in Paper I and in this paper, can, with the proper corrections applied, provide photometric measurements with errors < 1%. 19oth Kilpatrick et al. (2023) and Niu et al. (2023) use purecarbon best-fit models.A justification for carbonaceous dust grains in RSG circumstellar shells arises from the detection of PAHs in their mid-infrared spectra and the presence of continuum emission in the 3-8 μm range, in excess of that predicted by silicate models (Verhoelst et al. 2009).The availability of C is due to the dissociation of CO by chromospheric UV photons (Beck et al. 1992).Verhoelst et al. (2009) employed a dust composition with a small mass fraction (5%) of amorphous C to fit the SEDs in their sample, which would require 0.1% of the CO to be dissociated.Pure-C dust models require a much larger dissociation fraction (∼10%), which is not supported by models of chromospheric dissociation (see, e.g., Beck et al. 1992, their Figure 1).Moreover, this estimate requires the unrealistic assumption that all of the dissociated C is locked into dust.Verhoelst et al. (2009) admitted that an alternate solution to the problem is to assume that large (>0.1 μm) silicate grains can form in these atmospheres, which would reproduce the flux at wavelengths shorter than 8 μm.Subsequent work (e.g., Höfner 2008) has shown that this is indeed possible, and such grains have been detected in Galactic RSGs (Scicluna et al. 2015(Scicluna et al. , 2020)).In light of this simpler solution to the problem, the justification for pure-C models is unfounded.
Our results are similar to those of Jencson et al. (2023), who also used GRAMS models (although, with a different fitting method), as well as a "superwind" model, and found T eff ≈ 3500 K and L bol ≈ 10 5.1 L e , which, together with the inferred M ini = 17 ± 4 M e , are essentially consistent with our values, to within the uncertainties.Jencson et al. (2023) only included the HST photometry from Pledger & Shara (2023) in their SED fitting.We described in Paper I that our Spitzer measurements at 4.5 μm agreed well with those by Jencson et al. (2023), although disagreement exists at 3.6 μm, which might have contributed to differences in the fitting and the final progenitor candidate properties.Note that Choi et al. (2016) pointed out that, at nominally solar metallicity, the predicted slopes of the MESA Isochrones and Stellar Tracks RSG tracks are too shallow compared to the observations; hence, unlike Kilpatrick et al. (2023) and Jencson et al. (2023), we did not use those tracks here.
The Niu et al. (2023) models are constructed using assumptions and parameter values very similar to ours (thick dust shells and comparable L bol , T eff , R in , and τ).The main discrepancies between our results arise from four sources: (1) Niu et al. (2023) included the F675W detection (which, we have already pointed out, is fainter than ours) in their SED fitting, which we did not.(2) Those authors rejected two of their silicate models based on the temperatures not being consistent with the stellar track end points; even for Galactic RSGs with spectro-interferometric constraints on the radius and temperature of the sources, RSGs are often found at temperatures beyond the termini of single-star evolutionary tracks, and so this should not be considered an effective way to exclude parameter space (see, e.g., Wittkowski et al. 2017, their Figure 10).Indeed, the 68% credible intervals we find for the T eff and L bol of the progenitor lie close to those of two stars, V602 Car and HD 95687, in Wittkowski et al. (2017).(3) Niu et al. (2023) accounted for the variability by adding a 0.5 mag uncertainty to the optical fluxes, which is much lower than the value we use, therefore assigning a higher relative weight to the optical data points when computing their best fit.This inherently restricts the range of IR fluxes that can be probed by the best-fit model.(4) Niu et al. (2023) rejected the remaining silicate models in favor of the pure-C models, because the predicted 8 μm flux is not lower than their estimated upper limit (which, incidentally, is lower than ours, as a result of the different photometric techniques and assumptions), and those authors estimated a low probability of this being the case.As discussed above, we believe that the justification for a pure-C model is far weaker than for silicate models.Had those authors chosen the silicate models, their results, in terms of the range of possible M ini might well have been more consistent with our inferred range.
Rather unintentionally, our estimates for M ini wound up being roughly consistent with the ∼15 M e which Szalai & Van Dyk (2023) hastily rendered shortly after the SN discovery.

Further Thoughts
As something of a confirmation of our modeling results, we show in Figure 13 a direct comparison of the progenitor candidate SED with that of the Galactic RSG IRC −10414, a luminous late-M star and an OH, H 2 O, and SiO maser source (see Gvaramadze et al. 2014, and references therein).Additionally, it is also known to be variable with period ∼ 768 days. 20The IRC −10414 SED is compiled from Messineo & Brown (2019) and part of a larger set (S. Van Dyk 2024, in preparation).The two SEDs are strikingly similar, to within the uncertainties.Gvaramadze et al. (2014) inferred T eff = 3300 K and Messineo & Brown (2019) found a similar, but somewhat cooler, 3110 ± 170 K for the Galactic star.The bolometric magnitude we adopt here is M bol ≈ −7.75, and assuming M bol ( e ) = 4.74 mag (Mamajek et al. 2015), this corresponds to L bol ≈ 10 5.00 L e , consistent with the estimated luminosity from the GRAMS model fitting.(Note that Gvaramadze et al. 2014 estimated a higher luminosity for the star, L bol ≈ 10 5.2 L e , based on different assumptions.)Interestingly, like the luminous Galactic RSGs Betelgeuse (α Ori; Noriega-Crespo et al. 1997) and μ Cep (Cox et al. 2012), IRC −10414 is associated with an interstellar bow shock (Gvaramadze et al. 2014).
We can make a comparison of the SN 2023ixf progenitor candidate with the two confirmed dusty RSG progenitors of SN 2012aw (Van Dyk et al. 2012b;Fraser et al. 2012) and SN 2017eaw (Kilpatrick & Foley 2018;Rui et al. 2019;Van Dyk et al. 2019) We show the overall comparison in Figure 14.It is evident that the SN 2023ixf progenitor candidate would be the dustiest progenitor we as a community have encountered so far.The optical emission from the star is highly suppressed, relative to the other progenitors in the comparison.Correspondingly, as the UV/optical light was reprocessed, far more emission emerged in the IR out to 4.5 μm, and we would further expect a much higher luminosity than the other stars at wavelengths redward of that (as the SED modeling implies).The total bolometric luminosities of the SN 2023ixf progenitor candidate and the SN 2012aw and SN 2017eaw progenitors are quite similar, although the difference in the shapes of the three SEDs appears to be stark.How can these otherwise similarly luminous SN progenitors be so different?How this larger total dust opacity, given the likelihood of similar dust stoichiometry (e.g., Verhoelst et al. 2009), arose for the SN 2023ixf candidate is unknown, but we can speculate that these differences correspond to a more massive and potentially larger circumstellar environment set up around SN 2023ixf, possibly as the result of a higher M  , stronger wind, due to whatever cause or causes, or binary interaction.
Our comparison accentuates the heterogeneity in the characteristics of SN progenitors, even those at similar luminosities.This logically parallels that, although general trends exist in observed properties, variations exist between known RSGs, for example in Local Group galaxies.Differences in individual stellar evolution, up to the end of the star's life, are at play, and this could be further affected by the presence and proximity of a binary companion.This comparison emphasizes that each identified SN II progenitor is valuable and should be individually considered going forward, and that broad-brush conclusions, based on the ensemble, should be made taking all of these considerations into account.
That the progenitor star set up a dense, confined CSM seems extraordinarily likely.How the CSM arose remains uncertain currently.Hiramatsu et al. (2023), for instance, presented modeling of the pseudobolometric light curve over the first month and concluded that enhanced mass loss from the progenitor must have occurred during the final 1-2 yr before explosion, either through a single eruption or a continuous massloss process at M 0.01  » -1.0 M e yr −1 , leading to the formation of CSM of 0.3-1.0M e .However, Neustadt et al. (2024) analyzed a roughly 15 yr long optical data set (spanning 5600-400 days before explosion) obtained with the Large Binocular Telescope and found no evidence for preexplosion outbursts during this period (however, because of the sparse coverage of their data, they could not directly exclude short-lived outbursts).Ransome et al. (2023) also found no significant detections in long-baseline, multiband Pan-STARRS light curves prior to explosion.Flinner et al. (2023) also ruled out luminous UV eruptions 15-20 yr prior to explosion.Dong et al. (2023) examined over ∼5 yr of preexplosion data from the DLT40, Zwicky Transient Facility, and ATLAS surveys and also concluded that the probability of precursor outbursts is low.Again, being insensitive to short outbursts, those authors set upper limits of 100 and 200 days for the duration of any such outbursts in the case of a peak brightness of M r ≈ −9 mag and −8 mag, respectively, leading to a maximal amount of ejected preexplosion matter of 0.015 M e .The formation mechanism, or mechanisms, for the CSM remains elusive.We note, however, that we have shown that it could be plausible to set up the CSM with just a steady, low wind velocity and a moderate M  over the course of the RSG phase, without the need to invoke a superwind or eruption: the range in values for R in for the dust shell are essentially consistent with the estimates of the confined CSM dimensions obtained from the SN itself (e.g., Bostroem et al. 2023;Jacobson-Galán et al. 2023).(Furthermore, Zimmerman et al. 2024 argued that radiative acceleration can explain the observed high-velocity flash-feature profiles without invoking a recent pre-SN stellar eruption to accelerate matter.)We caution that wind velocities  100 km s −1 might even hamper significant dust production, since matter needs to remain long enough within the dust-formation zone.
It is intriguing to suggest that IRC −10414 may serve as a Galactic analog for the progenitor of SN 2023ixf, not that we are stating here that the progenitor candidate was a maser source-we obviously have no way of confirming that possibility-nor that IRC −10414 is only decades, years, or days, from core collapse.However, in the context of this possible analog, particularly notable is the flux excess in the HST bands short-ward of 8000 Å (F658N, F673N, and F675W), relative to the overall fit to the observed SED of the candidate including the flux at F814W, and to the IRC −10414 SED (see Figure 13).The first two narrow bands are sensitive to Hα (+[N II]) and [S II], respectively, while the broader F675W (WFPC2 ∼R) would be sensitive to detecting both.The excess could imply that line emission is present within the PSF of the progenitor candidate, not necessarily from the star itself, but from its immediate environment.It is interesting to note that Gvaramadze et al. (2014) detected strong Hα + [N II], and comparatively weaker [S II], from the arc-like bow shock at ∼15″ (∼0.14 pc) from IRC −10414; with FWHM ≈ 1.5 ACS pixel for the candidate profile, at 6.85 Mpc this is ∼2.5 pc, so an analogous bow shock would be within the PSF.If this were a bow shock, it could imply that the SN 2023ixf progenitor candidate possibly had been a runaway star.It is indeed curious that this candidate, as well as the progenitors of SN 2012aw and SN 2017eaw, were all isolated spatially from any obvious stellar clustering, which is not what we would necessarily expect for such an initially massive star.
Other explanations for a possible flux excess in those HST filters include an ionized circumstellar shell around the progenitor candidate, analogous to what Wright et al. (2014) found around the RSG W26 in the massive Galactic cluster Westerlund 1, or a photoionized confined circumstellar shell, such as that around Betelgeuse (Mackey et al. 2014).Since RSGs are clearly too cool to photoionize matter around them, the possible source of UV photons, in the case of the candidate, could either be a neighboring hot, blue star or from a hot binary companion.As can be seen in Figure 1, a blue star is ∼0 24 to the southwest of the progenitor; however, this is ∼7.9 pc away, which may well be too distant.Alternatively, possibly some shock excitation process was at work in the star's circumstellar environment.
As we pointed out in Section 8, binary progenitor systems are theoretically possible and, in the case of the wide binaries, the larger mass ratios (0.4) for the models would correspond to secondaries (companions) which would be hot enough to ionize any circumbinary environment.The main obstacle here is that the BPASS binary orbital separations span ∼1.1-7.7 R å , whereas for our dust modeling the preferred R in are 6.5-13.3R å , i.e., the binary system would most likely exist entirely within the dust shell, based on our results.Any ionization of the circumstellar environment would likely be on the inner parts of the shell and would be heavily obscured (based on the resulting optical depth from our best-fit GRAMS model, A Hα ≈ 4 mag).
Another curious aspect of the excess optical emission is that the progenitor candidate is no longer detectable in the ACS F658N observation from 2018 March, more than 14 yr after and a factor ∼ 2.4 deeper than the ACS F658N observation in 2004 February.The candidate would have decreased in brightness by 0.9 mag.(We note that the candidate was not detected in the WFPC2 F656N imaging in 1999 March and June; however, those observations were not as deep and as sensitive as the ACS ones; that excess light was detected contemporaneously in F675W indicates that some Hα emission existed at the time).Either whatever source of excitation that was responsible was curtailed or obscuring dust was suddenly present, possibly in some episodic event (this would definitely require the Hα emission to have been local to the circumstellar environment).Whatever obscuration there might have been had little effect on the mid-infrared emission from the candidate, as detected with Spitzer up to 2018, as well as the near-infrared detected from 2007 through 2013 (see Paper I).
One additional aspect regarding the progenitor as a binary follows on from a point raised by Kilpatrick et al. (2023).As we mentioned above, a putative companion would have been completely obscured prior to explosion.We agree with those authors that, among the deep nondetections at F336W, F435W, and F555W, the latter is the most constraining-with additional CSM obscuration of A F555W ≈ 6 mag based on the GRAMS  Fraser et al. 2012;Kochanek et al. 2012, SN 2017eaw (diamonds;Van Dyk et al. 2019; see also Kilpatrick & Foley 2018;Rui et al. 2019), andSN 2018aoq (triangles;O'Neill et al. 2019).For the latter two SN progenitors, we have adjusted the published luminosities by more recent measurements of the distances to the host galaxies (see text).model, the limit would be M F555W  −8.8 mag.However, the secondaries of all of the allowed model binary systems are easily less luminous than this limit.
An additional constraint on the presence of a binary companion, as suggested by the reviewer of this paper, is to determine at what companion luminosity would the observed SED be detectably modified by the additional UV flux from the hot star.We could not undertake this test with the GRAMS models, since those are all precomputed and packaged.We could, however, use DUSTY, by including progressively hotter and more luminous blue stars to the central RSG source, until the model SED diverged significantly from the observed one.
The assumption is that the hot companion is within R in , as both the SED and binary models imply, and can indeed effectively be treated as an additional central flux source within DUSTY (this, admittedly, may be an oversimplification).We adopted our best-fitting model parameters, as shown in Figure 11 (left panel), and the allowed BPASS models.The companions were approximated by Castelli & Kurucz (2003) main-sequence model stellar atmospheres; see Figure 11 (right panel).We found that a companion would have to be quite massive (q  0.8), hot (T eff  24,500 K), and luminous (10 4.1 L e , L 0.139 prim  ) for it to have had any appreciable effect on the observed SED.As can be seen in Figure 11, the increased UV contribution from the companion noticeably increases the IR emission at 3 μm; at this companion luminosity the brightness at Spitzer IRAC 4.5 μm would be significantly higher than the actual observed ensemble brightness, including variability, in this band.The total luminosity from the dusty system would also be 14% higher than what we have inferred from the observed progenitor candidate.Therefore, if the RSG progenitor had a companion, we can surmise that it would have had to be cooler, less luminous, and less massive than this.From the BPASS models a surviving companion would have luminosities of, for instance, −5.3 and −4.8 mag in WFC3/UVIS F275W and F336W, respectively.Zimmerman et al. (2024), via an analysis of early time UV spectra obtained of the SN with HST, estimated densities and dimensions of different regimes within the star's CSM.Those authors determined that a higher-density (∼5 × 10 −13 g cm −3 ), confined (2 × 10 14 cm; see also Martinez et al. 2024) region of CSM above the stellar photosphere (they estimated the star's radius at ∼5 × 10 13 cm, whereas our estimate is a factor of 2 larger, ∼1 × 10 14 cm; see Section 8) extended the shock breakout.Beyond that the CSM density drops (to 10 −15 g cm −3 ) and continues to gradually decline.We have found from our SED modeling that at ∼10 15 cm (or, ∼10 R å ) the CSM was cool enough for dust to form significantly and the resulting dust shell extended outward as the density progressively declined.That Zimmerman et al. (2024) concluded that the density beyond the dense, confined CSM and shock breakout regime must fall off as ∝r −2 lends support to our assumption of a similar density distribution for the dust shell.Following Grefenstette et al. (2023), Zimmerman et al. (2024) assumed a gas density ∼ 4 × 10 −16 g cm −3 at ∼10 15 cm, whereas we found the dust density to be ∼10 −17 to ∼10 −18 g cm −3 at this radius (R in ); these two density estimates are consistent for a reasonable assumed gas-to-dust ratio of 200.We note that the modeling by Martinez et al. (2024) results in a constraint on an extended CSM, at ∼8 × 10 14 cm, which is consistent with the confidence intervals on R in ; however, we posit that the lowerdensity dusty CSM extends well beyond that (see also Li et al. 2024).In Figure 15 we show a cartoon schematic of the approximate progenitor CSM geometry, with the asymmetric (Smith et al. 2023;Vasylyev et al. 2023;Li et al. 2024), confined CSM indicated above the photosphere and the inner radius of the simplistically spherical dust shell represented, as well as the range of possible binary orbits; this again illustrates Figure 15.A cartoon schematic of the SN 2023ixf progenitor candidate prior to explosion, to approximate scale, based on observational and modeling inference.Here we represent the stellar photosphere at its median effective radius R å ≈ 1390 R e .Immediately above that is the dense, asymmetric (Smith et al. 2023;Vasylyev et al. 2023;Li et al. 2024), confined CSM inferred by Zimmerman et al. (2024).The inner radius R in of the (assumed) spherically symmetric dust shell, which we infer here from our SED modeling, is shown at its median value of 10 R å ≈ 1 × 10 15 cm.The shell is assumed to extend to 1000 R in (not shown), with density decreasing as ρ ∝ r −2 .Additionally, we show the range of allowed model binary orbital radii, which are all within the primary's dust shell (see text).
that any binary companion was most likely within the dust shell, and those with the smallest orbits were even potentially within the denser, inner CSM.
The SN should be observed, with either HST or the James Webb Space Telescope (JWST), when its brightness has decreased enough that the SN's image does not saturate the detectors and when enough fiducial stars around the site also can be imaged at sufficiently high SNR, such that a robust astrometric alignment can be made with the pre-SN imaging data, to more securely associate the SN with the progenitor candidate.Qin et al. (2023) had already attempted this with adaptive optics from the ground, although the SN image was still heavily saturated at the time.Of course, years from now, when the SN has faded to a sufficiently low level (given the indications of CSM interaction, this could be quite a long time), it should be observed again with either of the two space telescopes, to confirm that the progenitor candidate is indeed the actual progenitor.Furthermore, with much of the pre-SN CSM dust destroyed, deep observations in the blue and UV should be performed to detect or place constraints on a binary companion (although, the presence of any remaining preexisting dust or freshly formed dust in the SN shock could compromise this set of observations).
With the SN 2023ixf progenitor candidate we have been able to obtain an unprecedented portrait of the potential star that exploded.For one thing, we have a first ever spectacular view of the semiregular variability of the progenitor in the years prior to core collapse.For another, we can construct a wellsampled SED from the optical through the mid-infrared for the star, which details the final state of the star.Sometime in the not so distant future, an SN will occur in a very nearby host galaxy -possibly even once again M101-with unheralded data coverage by both HST and JWST, such that a spectacular SED can be constructed for the progenitor star.Until that time, the well-characterized progenitor candidate for SN 2023ixf will have to suffice.
Jencson et al. (2023);Kilpatrick et al. (2023),Niu et al. (2023), and Paper I all detailed the variability of the candidate in the IR.In Paper I we found a fundamental period of the variability ∼1091 days, implying a long-period, semiregular nature.Based on the HST data alone,Pledger & Shara (2023) estimated a low M ini ≈ 8-10 M e , whereas from modeling of the combined HST and Spitzer data Kilpatrick et al. (2023) found a higher bolometric luminosity, L bol = 10 4.74 L e , and M ini = 11 M e .Jencson et al. (2023) constrained the luminosity and initial mass at even higher values, 10 5.1 L e and M ini = 17 ± 4 M e , respectively.Niu et al. (2023) also found similar values.In Paper I, from the RSG period-luminosity relation we provided an estimate, also on the high side, of M ini = 20 ± 4 M e .

Figure 1 .
Figure 1.Left: a portion of the HST ACS/WFC F435W + F555W + F814W color-composite image mosaic from 2002 November 16, with the progenitor candidate indicated by solid tick marks.(The chip gap can be seen toward the top of the panel.)A putative "Source B" is indicated with dashed tick marks (Kilpatrick et al. 2023; see also their Figure 1).Right: a portion of the i-band image stack created from Gemini 'Alopeke observations in the WF mode on 2023 June 9, with SN 2023ixf clearly visible.Both panels are shown to the same scale and orientation.North is up and east is to the left.

Figure 2 .
Figure 2. Left: a portion of the Spitzer IRAC 5.8 μm image mosaic from 2004, with the SN progenitor candidate location (based on the absolute position) indicated by tick marks.The candidate is not detected in these data.Right: same as the left panel, but at 8.0 μm.Both panels are shown to the same scale and orientation.North is up and east is to the left.

Figure 3 .
Figure3.Reconstructed speckle image (inset) and differential magnitude detection limits Δm of SN 2023ixf obtained with the 'Alopeke speckle camera at Gemini-N on 2023 June 9.From the 8320 Å narrowband filter observations, we can rule out any neighboring object to ∼5 mag and ∼9 mag below the SN brightness at 0 1 and 1 2, i.e., to I ≈ 16 and ≈20 mag, respectively.
(2023) and Smith et al. (2023), based on high-spectralresolution observations of the SN, effectively using SN 2023ixf as a bright light bulb behind a dusty screen.Both of these studies concluded that the host reddening is relatively low, at E(B − V ) = 0.031 mag, based on the equivalent widths (EQWs) of the Na I D absorption lines.From our Keck HIRES spectrum we performed a similar measurement of the Na I D EQW; this portion of the spectrum is shown in Figure 4. We have identified the features we assign, based on their wavelengths, to both the Galactic component and the host-galaxy component.For the host contribution we measure an Na I D2 EQW of 0.177 ± 0.001 Å and for D1, 0.122 ± 0.003 Å.These measurement values are overall quite similar to those that Lundquist et al. (2023) quoted, although the values measured by Smith et al. (2023) are somewhat larger than ours.We found that the two lines were offset relative to the host-galaxy redshift by +21.4 km s −1 (interestingly, Smith et al. 2023 estimated this offset as +7 ± 1 km s −1 ).Following these two studies, we applied the relations provided byPoznanski et al. (2012) to convert EQW to reddening and found an average (from the three relations, for D1, D2, and D1 + D2) of ( ) , overall this is consistent with the estimates made byLundquist et al. (2023) andSmith et al. (2023), as well as byJacobson-Galán et al. (2023).If we assume that the dust in M101 is similar to Galactic diffuse interstellar dust, then we adopt R V = 3.1, for a visual extinction internal to the host of A 0also measured the EQW of the lines we associated with the Galactic foreground contribution and found that the D2 EQW = 0.020 ± 0.004 Å and the D1 EQW = 0.018 ± 0.001 Å.Again, followingPoznanski et al. (2012), we computed that ( )

Figure 4 .
Figure 4. Left: a portion of the Keck HIRES coadded spectrum of SN 2023ixf from 2023 June 7, showing the locations of the Na I D absorption lines that we attribute to Galactic foreground and to the host galaxy, as indicated.Right: a portion of the same spectrum in the region of the diffuse interstellar band (DIB) λ5780 feature.We also show a model of the feature, assuming a Gaussian with FWHM = 2.1 Å and an EQW corresponding to A V = 0.099 mag (seePhillips et al. 2013), shifted by a velocity offset of +21 km s −1 (pink dashed curve).Also shown is a similar model, based on a 3σ upper limit of the EQW (green solid curve).The wavelength axis for both panels is in the host-galaxy rest frame.

Figure 5 .
Figure 5. (B − V ) 0 colors for SN 2023ixf from Lick KAIT observations (solid circles), from Hiramatsu et al. (2023, H23, open circles) and from Bianciardi et al. (2023, B23, dotted-dashed curve).Also shown are the two samples of SNe II-P, one with the bluest colors and the other with no significant host-galaxy reddening, from de Jaeger et al. (2018).Additionally, we show the colors for the Type II-P SN 2017eaw in NGC 6946 (Van Dyk et al. 2019), which likely experienced only Galactic foreground reddening, and two short-plateau SNe, SN 2006Y and SN 2006ai (Hiramatsu et al. 2021), which also experienced minimal host reddening (Anderson et al. 2014).All of the SNe shown, therefore, were dereddened only by the Galactic foreground contribution.
measured higher EQWs for [O II] and [O III], whereas our EQWs for [S II] are higher by a factor of ∼1.5, while our EQWs for [N II] are comparable to those of Kennicutt & Garnett.We measured an EQW for Hβ of 185 Å, while the Kennicutt & Garnett (1996) value is 292 Å.We analyzed the various line ratios from our measurements using the strong-line diagnostics from Curti et al. (2020).The resulting individual abundance values are shown in Table 2.Note that Kennicutt & Garnett (1996) estimated the value of the R 23 indicator as 5.02, whereas we computed it to be 3.63.Similarly, following the Curti et al. (2020) calibration with the Kennicutt & Garnett (1996) value, the O abundance of the H1086 region would be ( found 8.67 for the emission closer to the edge.In short, we estimated from the Gemini spectrum a range of values the diagnostic (and including the uncertainties), mirroring the same spread in O abundance found by Garner et al. (2022) and Esteban et al. (2020).Assuming the most recent value for the solar O abundance of ( et al. 2022; note that those authors justified their discrepancy with the 8.69 ± 0.04 value from Asplund et al. 2021

Figure 6 .
Figure 6.Gemini GMOS-N spectrum of emission from the outskirts of H II region #1086 (H1086; Hodge et al. 1990) nearby (∼8 3) to SN 2023ixf.See Figure 7.The spectral continuum has been normalized, however the spectrum as shown has not been reddening corrected.The locations of the various strong lines we measured, in order to estimate the oxygen abundance from the emission site, are indicated.

Figure 7 .
Figure 7. Outline of the Gemini GMOS-N 0 75 slit (dashed lines) for the 2023 June 3 spectral observations, overlaid on a portion of the HST ACS/WFC F658N image mosaic from 2004 February 10.The position of SN 2023ixf is indicated with a circle.Along with the SN, the slit position intersected the outskirts of H II region #1086 (H1086; Hodge et al. 1990), the center of which is ∼8 3 northwest of the SN.

Figure 8 .
Figure8.The observed SED of the SN 2023ixf progenitor candidate (blue squares), together with detections dominated by emission from the SN environment (magenta triangles; see text), all corrected for both Galactic foreground and host-galaxy reddening.The SED can be approximated by a cool blackbody (dashed red curve), ∼1761 K.

Figure 9 .
Figure 9.The reddening-corrected observed SED (red circles), together with the best-fit GRAMS model (solid yellow curve) and posterior samples from a Markov Chain Monte Carlo (MCMC) procedure (gray curves).A PHOENIX model (dashed line) at the median estimate for T eff , 3500 K, is also shown for comparison.The error bars include a combination of the measurement uncertainties, as well as the range in flux in each band as a result of source variability.Flux upper limits are shown for various HST bands, the Spitzer IRAC 5.8 and 8.0 μm bands, and the WISE W1, W2, and W3 bands (red triangles); note that we have shown the three HST detections short-ward of 0.8 μm as triangles (upper limits) as well.The model predictions are consistent with these limits.

Figure 10 .
Figure10.Posterior distribution of parameters (luminosity L bol , effective temperature T eff , inner radius of the dust shell R in , and 10 μm optical depth τ 10 ) obtained from the MCMC sampling (see text).The scatter plots show the variation of each parameter against every other parameter, with contours containing ∼39% and 84% of the probability mass (corresponding to 1σ and 2σ contours in 2D).The histograms illustrate the distributions of the individual parameters, with vertical dashed lines representing the 68% credible interval for each parameter.

Figure 11 .
Figure11.Analysis of the SN 2023ixf progenitor candidate SED using DUSTY(Ivezic & Elitzur 1997;Ivezic et al. 1999;Elitzur & Ivezić, 2001).Left: the reddeningcorrected observed SED (black squares), together with the best-fitting DUSTY model (solid red curve), with input parameters T eff = 2500 K, T in = 1300 K, and τ V = 10, and a sample of other good-fitting models (gray curves; see text).The T eff = 2500 K PHOENIX photosphere (dotted purple curve) at the inferred L bol (see text) is also shown for comparison.The error bars include a combination of the measurement uncertainties, as well as the range in flux in each band as a result of source variability.Unlike Figure9, only flux upper limits in the Spitzer IRAC 5.8 and 8.0 μm and WISE W3 bands are shown, since these are the most constraining.The model predictions are consistent with these limits.The blue curve represents a DUSTY model computed with a central source comprised of a T eff = 2500 K RSG photosphere, plus a hot 24,500 K source (a Castelli & Kurucz 2003 B1V model atmosphere, meant to approximate a putative binary companion) with ∼0.14× the inferred L bol of the best-fitting RSG model (see right panel), assuming the same input parameters.Right: the input SED to DUSTY for the putative binary system (solid blue curve), consisting of a T eff = 2500 K RSG photosphere (dotted red curve, in this panel) as the primary and the less luminous, hot (24,500 K) companion (dashed magenta curve).

Figure 12 .
Figure 12.H-R diagram showing the locus of the progenitor candidate based on the median (solid triangle), together with the 68% credible intervals of values (shaded region), from Table 4.For comparison we also display single-star theoretical evolutionary tracks from Stanway & Eldridge (2018) at solar (Z = 0.020; dashed line), subsolar (Z = 0.010; dashed-dotted line), and supersolar (Z = 0.040; solid line) metallicities.

Figure 13 .
Figure13.A direct comparison of the absolute SED for the SN 2023ixf progenitor candidate (black points; at the assumed distance, see Section 4) with that of the Galactic RSG IRC −10414 (dashed brown curve, with 1σ uncertainties as the gray region;Gvaramadze et al. 2014;Messineo & Brown 2019;Healy et al. 2024;  S. Van Dyk 2024, in preparation).The implications of the analyses represented in both panels is that the progenitor candidate had luminosity L bol ≈ 10 5.0 L e in the years prior to explosion.
, as well as of SN 2018aoq (O'Neill et al. 2019), which has been shown not to require circumstellar dust in the progenitor SED fitting.For SN 2012aw, Kochanek et al. (2012) presented a different treatment of the dust and constrained the progenitor luminosity to 10 4.8 < (L bol /L e ) < 10 5.0 .Van Dyk et al. (2019) estimated the luminosity of the SN 2017eaw progenitor as 1.2( ± 0.2) × 10 5 L e , and M ini ≈ 15 M e .Here we have adjusted the distance to the SN 2017eaw host, NGC 6946, from 7.73 ± 0.78 Mpc (Van Dyk et al. 2019) to the more recent and likely superior estimate, 7.12 ± 0.38 Mpc (Anand et al. 2021).The adjusted luminosity would then be 1.02 × 10 5 L e (which also then reduces the estimated M ini to 13-14 M e ).Here we have also adjusted the distance to the SN 2018aoq host, NGC 4151, from 18.2 Mpc in O'Neill et al. (2019) to, again, the more recent Cepheid-based distance of 15.8 ± 0.4 Mpc measured by Yuan et al. (2020).The SN 2018aoq luminosity in O'Neill et al. (2019) is L bol ≈ 10 4.7 L e .Adjusted, this is L bol ≈ 10 4.58 L e .

Figure 14 .
Figure 14.A comparison of SN II-P We show the reddening-and distance-corrected SED for the SN 2023ixf progenitor (circles), together with the SEDs for the progenitors of SN 2012aw (squares; Van Dyk et al. 2012b; Fraser et al. 2012; Kochanek et al. 2012, SN 2017eaw (diamonds; Van Dyk et al. 2019; see also Kilpatrick & Foley 2018; Rui et al. 2019), and SN 2018aoq (triangles; O'Neill et al. 2019).For the latter two SN progenitors, we have adjusted the published luminosities by more recent measurements of the distances to the host galaxies (see text).
Note.Columns: HST observation band, mean UTC observation date, HST instrument, total exposure time, Dolphot magnitude or inferred 5σ upper limit, and HST proposal ID number.Uncertainties in the Dolphot magnitudes are in parentheses.Upper limits are estimates from formal 5σ detections via Dolphot.

Table 2
Strong-line Metallicity Diagnostics

Table 3
Adopted Brightnesses of the Progenitor Candidate Note.Columns: observational band, bandpass effective wavelength for a 1761 K blackbody, observed magnitude, Galactic and internal host extinctions at the effective wavelength, Vega mag zero-point, extinction-corrected flux density, and uncertainty in the flux density.Uncertainties in m obs and A λ (host) are in parentheses.

Table 4 GRAMS
Best-fit Parameters for the Progenitor Candidate Note.The table lists, for each parameter, the MLE and the 16th, 50th, and 84th percentiles (p16, median, and p84, respectively).