Massive Stars: an Infrared Exploration Across the HR Diagram

Home , 2MASS, Imai (star), Mira variable, RSGC1, Stephenson 2, Superbubble

Massive Stars: An Infrared Exploration Across the HR Diagram

A DISSERTATION SUBMITTED TO THE FACULTY OF THE GRADUATE SCHOOL OF THE UNIVERSITY OF MINNESOTA BY

Michael Scott Gordon

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF Doctor of Philosophy

Advisors: Terry J. Jones, Roberta M. Humphreys, Robert D. Gehrz

There are many who have earned my gratitude for their contribution to my time in graduate school. I oﬀer here my sincerest thanks and dedicate this work in part to each of you: to Mom, Dad, and Mom-mom for creating a world where I could pursue any dream, and through whose eyes I am always the best possible version of myself. to Robyn, for being my closest conﬁdante, for never needing me to explain myself, and for never letting me go. to Micaela, for being my strongest pillar from the very beginning of this journey. to Skyler and Breanne, for letting me crash into your lives and your many years of tough love. to Karlen, for always being my comrade-in-arms on every adventure. to Dylan for being a constant source of humor and perspective. to my college friends, Randy, Mike S, Mike H, Josh, and Sean for keeping me honest. to MIfA grads past and present, for sharing the long road of frustration. to Dr. Gehrz, for teaching me that science is tenacity. to Dr. Humphreys, for teaching me that science is passion. to Dr. Jones, for teaching me that science is fun.

i Contents

Acknowledgementsi

List of Tables iv

List of Figuresv

1 Introduction1 1.1 YSGs and Post-RSG Evolution...... 1 1.2 Galactic RSGs...... 3 1.3 VY CMa & the Southwest Clump...... 4 1.4 Obscured RSGs in M33...... 5

2 Luminous and Variable Stars in M31 and M33. The Yellow and Red Supergiants and Post-Red Supergiant Evolution7 2.1 Introduction...... 8 2.2 Sample Selection and Observations...... 9 2.2.1 Target List...... 9 2.2.2 Observations...... 11 2.3 The Intermediate-Type or Yellow Supergiants...... 13 2.3.1 Spectral Characteristics...... 13 2.3.2 Multi-Wavelength Photometry...... 16 2.3.3 Extinction Correction and the Spectral Energy Distributions.... 19 2.3.4 Two New Hypergiants in M31...... 22 2.4 Red Supergiants...... 25 2.5 Discussion...... 33 2.5.1 Circumstellar Dust and Mass Loss...... 33 2.5.2 HR Diagrams...... 35 2.6 Conclusion...... 36

ii 3 Searching for Cool Dust: Infrared Imaging of OH/IR Supergiants and Normal RSGs 40 3.1 Introduction...... 40 3.2 Observations and Data Reduction...... 42 3.2.1 SOFIA/FORCAST: Far-IR Imaging (11 37 µm)...... 42 − 3.2.2 Adaptive Optics Mid-IR Imaging (8 10 µm)...... 44 − 3.2.3 IRAS, AKARI, WISE, and ISO-SWS (2 100 µm)...... 45 − 3.2.4 Herschel/PACS (70, 160 µm)...... 45 3.3 Results & Discussion...... 49 3.3.1 DUSTY modeling...... 49 3.3.2 VX Sgr...... 52 3.3.3 S Per...... 56 3.3.4 RS Per and T Per...... 59 3.3.5 NML Cyg...... 63 3.4 Conclusions...... 66

4 Thermal Emission in the Southwest Clump of VY CMa 70 4.1 Introduction...... 70 4.2 Observations & Data Reduction...... 72 4.3 Results & Discussion...... 76 4.3.1 DUSTY modeling...... 76 4.3.2 Scattered vs. Thermal Emission...... 79 4.3.3 Mass Estimates...... 80 4.4 Conclusions...... 81

5 Obscured Red Supergiants in M33 83 5.1 Introduction...... 83 5.2 Target Selection...... 84 5.3 Observations & Data Reduction...... 86 5.4 Modeling & Analysis...... 88 5.5 Object X & M33-8...... 92 5.6 Conclusions...... 98

6 Conclusions & Future Work 99

References 104

iii List of Tables

2.1 Luminous Stars: Journal of Observations...... 11 2.2 Luminous Stars: Spectroscopically-Conﬁrmed YSGs...... 16 2.3 Luminous Stars: Photometry of YSGs and YSG candidates...... 18 2.4 Luminous Stars: Extinction and Luminosities of YSGs and YSG Candidates 21 2.5 Luminous Stars: YSGs with Mass-Loss Indicators...... 24 2.6 Luminous Stars: Photometry of Candidate RSGs...... 28 2.7 Luminous Stars: Extinction and Luminosities of Candidate RSGs...... 30 2.8 Luminous Stars: RSG Candidates with Evidence for CS Dust...... 33 3.1 Searching for Cool Dust: Summary of Observations...... 43 3.2 Searching for Cool Dust: New Mid-Infrared Photometry...... 48 3.3 Searching for Cool Dust: DUSTY Model Parameters and Mass-Loss Rates.. 52 4.1 VY CMa and the SW Clump: Photometry of the SW Clump...... 75 4.2 VY CMa and the SW Clump: DUSTY Model Parameters...... 78

iv List of Figures

1.1 Evolutionary Tracks in the HRD...... 2 1.2 HST Composite of VY CMa...... 5 2.1 M33 with Spectral Targets...... 12 2.2 YSG spectra with Hα emission...... 14 2.3 Sample A- and F-type supergiant spectra from M31 and M33...... 15 2.4 SEDs of two F-type supergiants in M31...... 22 2.5 SEDs of two warm supergiants in M33...... 23 2.6 Spectra of new LBV candidates...... 26 2.7 SEDs of new LBV candidates...... 27 2.8 SEDs of two red supergiants in M31...... 31 2.9 SEDs of two red supergiants in M33...... 32 2.10 Luminosity vs. mass lost for RSGs in M31 and M33...... 35 2.11 HR Diagram of M31...... 37 2.12 HR Diagram of M33...... 38 3.1 SOFIA/FORCAST imaging of S Per...... 47 3.2 DUSTY model with “enhanced” mass loss...... 51 3.3 Optical and IR SED of VX Sgr...... 54 3.4 Radial profiles of VX Sgr...... 55 3.5 Optical and IR SED of S Per...... 57 3.6 Radial profiles of S Per with FORCAST...... 58 3.7 Radial profiles of S Per with MIRAC...... 59 3.8 Optical and IR SED of RS Per...... 60 3.9 Radial profiles of RS Per...... 61 3.10 Optical and IR SED of T Per...... 62 3.11 Radial profiles of T Per with FORCAST...... 63 3.12 Radial profiles of T Per with MIRAC...... 63 3.13 Optical and IR SED of NML Cyg...... 64

v 3.14 Radial profiles of NML Cyg with FORCAST...... 65 3.15 Radial profiles of NML Cyg with MIRAC...... 66 3.16 Luminosity vs. mass-loss rates...... 68 4.1 ALMA Contours of VY CMa...... 72 4.2 Near-IR imaging of VY CMa...... 73 4.3 Near-IR imaging of the SW Clump...... 73 4.4 Photometry of VY CMa and the SW Clump...... 77 5.1 CMD of Spitzer/IRAC sources...... 85 5.2 M33 with IRTF Targets...... 86 5.3 SEDs of Obscured RSG Candidates...... 87 5.4 SEDs and IRTF Spectra of RSG Candidates...... 89 5.5 IRTF Spectrum of J013403.84+303752.9...... 90 5.6 IR Spectrum of WX Ser...... 91 5.7 CMD of IRTF Sample...... 92 5.8 R-band and Hα Images of Object X...... 93 5.9 SED and IR Spectrum of Object X...... 94 5.10 MODS Spectrum of Object X...... 94 5.11 Ca II Absorption in Object X...... 95 5.12 V-band and Hα Images of M33-8...... 96 5.13 SED and IR Spectrum of M33-8...... 97 6.1 NGC 7419 at 2 µm...... 100 6.2 LMIRCam Radial Profile of B139...... 101 6.3 RSGC1 in FORCAST FOV...... 102

vi Chapter 1

Introduction

The main goal of this dissertation is to understand how stellar winds, dusty ejecta, and overall mass-loss histories affect the end-stage evolution of supergiant stars. To what extent mass loss and periods of enhanced stellar outflow can influence the terminal state of the most massive stars remains an outstanding question in the fields of stellar physics, chemical enrichment of the Galaxy, and supernova research. Here, we focus on characterizing the circumstellar (CS) ejecta around supergiants through a combination of observing techniques. Over the last several years, using the Large Binoc- ular Telescope (LBT), MMT, NASA’s Infrared Telescope Facility (IRTF), the Very Large Telescope (VLT), and the Stratospheric Observatory for Infrared Astronomy (SOFIA), we have performed high-resolution imaging, spectroscopy, and polarimetry—methods that provide us with keen insight on mass-loss histories and the 3D morphology of the material ejected by the Local Group’s most fascinating stars.

1.1 YSGs and Post-RSG Evolution

Mass loss in massive stars has a profound eﬀect on their late-stage evolution and eventual fate. For many years, the standard view of massive star evolution progressed from main- sequence to red supergiant (RSG) to terminal supernova (SN) explosion. However, recent observational evidence has uncovered a “Red Supergiant Problem” (Smartt et al., 2009): namely, there appears to be an upper limit of 18 M for Type II SN supernova progenitors ∼ (Smartt, 2015). Therefore, the 20 40 M RSGs presumably end their lives in some other − manner. Ultimately, the fate of a massive star is governed not only by its zero-age main- sequence (ZAMS) mass but also by its mass-loss history (de Jager, 1998; Heger et al., 2003). Massive star models over the last two decades allude to a population of RSGs with

1 Chapter 1. Introduction 2

high mass-loss rates that evolve back to warmer temperatures on the Hertzsprung-Russell Diagram (HRD). These “post”-RSG stars would undergo periods of enhanced mass loss during their late-stage evolution, possibly becoming luminous blue variables (LBVs) or Wolf- Rayet (WR) stars before their terminal state (Meynet et al., 2013). Figure 1.1 shows Geneva Group (Ekstr¨omet al., 2012; Meynet et al., 2015) evolutionary track models for diﬀerent

ZAMS masses. The higher mass models (M & 20 M ) loop through the yellow supergiant (YSG) region of the HRD, sometimes in multiple passes, before eventually terminating on the red supergiant branch.

Figure 1.1 Evolutionary tracks in the HRD for supergiants. Solid lines show the track until the red supergiant phase. The blue supergiant phase is highlighted in red toward the RSG phase, and in blue for evolution back from the RSG phase. The models are from Ekstr¨omet al.(2012). Adapted from Meynet et al.(2015).

As these models are quite sensitive to parameters such as metallicity, mass loss, and rotational velocity, the precise locations of the YSG loops and the RSG branches serve as observational tests of supergiant evolution. The YSG stars (deﬁned roughly as 4800 K <

Teﬀ < 7200 K and log L/L & 4.0, spanning spectral types of late-A to early G) are transient Chapter 1. Introduction 3

in nature as they cool and expand oﬀ the main sequence towards the RSG branch on the HRD (Stothers & Chin, 2001; Neugent et al., 2012). YSGs share the instability strip on the HRD with subgiant RR Lyrae variables and the brighter giant and supergiant Classical

Cepheids. Stars between 9 M and 40 M will pass through this instability strip from blue to red after leaving the main sequence; however, the more luminous yellow super and hypergiant stars are believed to have already spent time as RSGs and are instead progressing back towards the blue (Stothers & Chin, 2001). It is this latter class of YSGs, those evolving past the RSG stage, that are key in characterizing the interplay between mass loss and a supergiant’s terminal state. In Chapter2(Gordon et al., 2016), we present a comprehensive spectroscopic survey of the yellow and red supergiants in M31 and M33. We use evidence of circumstellar dust in their long-wavelength spectral energy distributions (SEDs) and spectroscopic indicators of mass loss and winds to identify candidates for post-RSG evolution. After studying almost 500 supergiant stars, we conclude that 40% of the observed yellow supergiants are likely ∼ in a post-RSG state. Comparison with evolutionary tracks shows that these mass-losing,

post-RSGs have initial masses between 20 and 40 M . The results from this work are signiﬁcant as the observations demonstrate that as many as half of the known supergiants in the local universe shed a signiﬁcant amount of mass during and after the red supergiant phase. This mass loss implies that these stars will likely follow an evolutionary path not properly discussed in the literature.

1.2 Galactic RSGs

While the spectroscopic survey discussed in Chapter2(Gordon et al., 2016) was successful in characterizing the mass-loss rates of dusty RSGs, high-resolution imaging is required to glean the mass-loss histories. That is, have the dust and stellar wind been constant and smooth over the lifetime of the stars, or has the mass ejection been more episodic and explosive in nature? To answer this question, we obtained high-resolution imaging of Galactic supergiants with adaptive optics on MMT and LBT, combined with mid-infrared imaging and photometry on SOFIA. With these imaging techniques, we were able to observe gas and dust ejecta both close in to the central star and at larger distances. The resulting radial proﬁles are valuable probes on timescale for the ejecta. These snapshots in diﬀerent infrared (IR) bands (1 5 µm with LBT/LMIRCam; 9 10 µm with MMT/MIRAC; − − 11 40 µm with SOFIA/FORCAST) allow us to recreate the mass-loss histories of the − observed RSGs from as recently as 100 years to as old as 10,000 years in the past. We Chapter 1. Introduction 4

incorporate radiative-transfer modeling code DUSTY (Ivezic et al., 1997) to determine the mass, temperature, and distribution of dust around each star. In Chapter3(Gordon et al., 2018a), we explore the circumstellar ejecta morphology around ﬁve Galactic RSGs—the OH/IR stars NML Cyg, VX Sgr, and S Per, as well as two RSGs without masers, RS Per and T Per. Though our images lacked the spatial resolution to observe CS ejecta close in to the central star—the point-spread function (PSF) of SOFIA/FORCAST is 300 at 20 µm—the observed radial proﬁles and DUSTY models ∼ suggest that constant mass-loss rates do not produce enough dust to explain the IR excess emission present in each of these RSGs’ SEDs. Rather, these stars have likely undergone periods of enhanced mass loss in the last few thousand years.

1.3 VY CMa & the Southwest Clump

We apply similar techniques in Chapter4(Gordon et al., 2018b, in review) to the extreme RSG VY Canis Majoris, one of the brightest IR sources in the sky. With multiple arcs and knots present in HST images (Smith et al., 2001; Humphreys et al., 2005, 2007), VY CMa has a complex circumstellar nebula environment, with many kinematically distinct ejection events evident in the surrounding nebulosity. Shenoy et al.(2013) explored the CS material in the near-IR with LBT/LMIRCam, concentrating speciﬁcally on the peculiar “Southwest Clump” (hereafter, SW Clump) emission feature (see Figure 1.2). Due to both its high surface brightness at 5 µm (LBT/LMIRCam; Shenoy et al., 2013) and its observed 30% fractional polarization at 3.1 µm (MMT/MMT-Pol; Shenoy et al., 2015), the SW Clump was determined to be optically thick from scattering, rather than from thermal emission. The −3 −2 clump was estimated to have a total mass between 5 10 and 2.5 10 M , depending on × × the adopted gas-to-dust ratio (100:1 vs. 500:1). This ejection event can be contrasted with 4 −1 the “normal” mass-loss rate for VY CMa of 10 M yr (Danchi et al., 1994; Humphreys ∼ et al., 2005; Decin et al., 2006) suggesting that the SW Clump likely represents a single mass-loss episode from a localized region of VY CMa’s stellar atmosphere. Despite the clump’s large mass and relatively warm color temperature of 275 K (see §4.3.1), recent sub-millimeter observations with ALMA found no evidence for the SW Clump in thermal emission at 180–700 GHz (O’Gorman et al., 2015; Vlemmings et al., 2017). Given the mass estimates of the SW Clump from the LMIRCam and MMT-Pol observations in Shenoy et al.(2013, 2015), this non-detection in thermal emission in the ALMA bands may have implications for the dust grain properties in the far-IR. Chapter 1. Introduction 5

Figure 1.2 Left: HST/WFPC2 composite of VY CMa (F410m, F547m, F656n ﬁlters; adapted from Smith et al., 2001). Right: F1042m (1 µm) HST image with prominent outﬂow arcs and knots labeled (adapted from Humphreys et al., 2007).

In Chapter4, we present observations of the SW Clump with LBT/NOMIC at 9– 12 µm. We isolate the photometry of the SW Clump from the stellar SED and model the circumstellar ejecta with DUSTY. The SED model separates the dusty thermal emission in the clump from scattered stellar light, which allows us to extend the thermal SED into the longer wavelength ALMA bands. We ﬁnd that the non-detection of the SW Clump by ALMA is consistent with the temperature and mass of the clump measured by Shenoy et al. (2013).

1.4 Obscured RSGs in M33

Unfortunately, our spectroscopic survey of the most luminous stars in M31 and M33 (Chap- ter2 of this work; Humphreys et al., 2013, 2014; Gordon et al., 2016) does not include supergiants that may be heavily obscured and faint in the optical. Since the most luminous warm and cool supergiant populations are more likely to have the highest mass-loss rates (Mauron & Josselin, 2011; Gordon et al., 2016), it is probable that those stars will be highly obscured by their own circumstellar ejecta and thus represent a previously-unknown Chapter 1. Introduction 6

population of hidden RSGs. To complete the upper portion of the HR diagram requires investigating bright infrared sources with faint or undetected optical counterparts. As a follow-up to the spectroscopic survey on mass-losing RSGs, we completed an IR survey of optically-obscured RSGs in M33 at the NASA Infrared Telescope Facility (IRTF) on Mauna Kea, Hawaii. With the infrared imaging, guiding, and spectroscopy capabilities at the IRTF, we have been able to measure directly the mass-loss rates of some of the dustiest supergiants in the local universe. Preliminary results from this project are summarized in Chapter5. The resulting catalog of mass-losing RSGs will be valuable in understanding the late-stage evolution of cool, massive stars, as well as identifying what fraction of RSGs are hiding inside their own dust shells. Chapter 2

Luminous and Variable Stars in M31 and M33. The Yellow and Red Supergiants and Post-Red Supergiant Evolution

Adapted from Gordon et al.(2016)

Abstract Recent supernova and transient surveys have revealed an increasing number of non-terminal stellar eruptions. Though the progenitor class of these eruptions includes the most luminous stars, little is known of the pre-supernova mechanics of massive stars in their most evolved state, thus motivating a census of possible progenitors. From surveys of evolved and unstable luminous star populations in nearby galaxies, we select a sample of yellow and red supergiant candidates in M31 and M33 for review of their spectral characteristics and spectral energy distributions. Since the position of intermediate and late-type supergiants on the color-magnitude diagram can be heavily contaminated by foreground dwarfs, we employ spectral classification and multi-band photometry from optical and near-infrared surveys to confirm membership. Based on spectroscopic evidence for mass loss and the presence of circumstellar dust in their SEDs, we find that 30 40% of the yellow supergiants are likely in a post-red supergiant − state. Comparison with evolutionary tracks shows that these mass-losing, post-

RSGs have initial masses between 20 40 M . More than half of the observed − red supergiants in M31 and M33 are producing dusty circumstellar ejecta. We

7 Chapter 2. Luminous and Variable Stars in M31 and M33 8

also identify two new warm hypergiants in M31, J004621.05+421308.06 and J004051.59+403303.00, both of which are likely in a post-RSG state.

2.1 Introduction

For many decades, the standard model of stellar evolution for massive stars ( 9 M ) was ≥ characterized as the progression from main sequence OB star to red supergiant (RSG) to terminal supernova (SN) explosion. We now know that the evolutionary paths of massive stars, as well as their terminal state, depends strongly on mass loss and their mass-loss

histories. It was recognized some time ago that stars above some initial mass ( 40 50 M ) ∼ − do not evolve to the RSG stage (Humphreys & Davidson, 1979); however, due to mass loss—possibly eruptive (Humphreys & Davidson, 1994)—these massive stars then return to hotter temperatures, perhaps becoming Luminous Blue Variables (LBVs) or Wolf-Rayet (WR) stars prior to their terminal state. Stellar interior models also show that as stars shed their outer layers, the mass fraction of the He core increases, and when it exceeds 60 70% − of the total, the star will evolve to warmer temperatures (Giannone, 1967). Lower mass supergiants that enter the RSG stage will either end their lives as Type II-P SNe or in some cases evolve back to warmer temperatures before the terminal explosion. Smartt et al.(2009) recently identiﬁed what he called “the red supergiant problem”—the lack of Type II-P and Type II-L SN progenitors with initial masses greater than 18 M .

RSGs between 18 and 30 M would presumably end their lives in some other manner. They might migrate on the HR diagram to warmer temperatures before their terminal explosions. The RSG stage is a well-established high mass-losing phase. Mass-loss rates can be −6 −1 anywhere from 10 M yr in RSGs (Gehrz & Woolf, 1971; Mauron & Josselin, 2011) −4 −1 to as high as 10 M yr in extreme stars like VY CMa and in the warm hypergiants (Humphreys et al., 2013). What fraction of the RSGs return to warmer temperatures, the physical characteristics of candidate post-RSGs, and their locations on the HR Diagram are thus crucial to our understanding the ﬁnal stages of the majority of massive stars. Due in part to their position on the HR Diagram (HRD), few post-RSGs are known. As yellow supergiants (YSGs), with spectral types from late A to K, they occupy a transient state between the blue and red supergiants and may either be evolving from the main sequence to cooler temperatures, or back to warmer temperatures from the RSG stage. Both populations represent a relatively short transition state. In the Galaxy, the warm hypergiants, close to the upper luminosity boundary in the HRD with high mass-loss rates, enhanced abundances, and dusty circumstellar (CS) environments, are excellent candidates Chapter 2. Luminous and Variable Stars in M31 and M33 9

for post-RSG evolution. These stars contrast with the intermediate-type yellow supergiants which have normal spectra in their long-wavelength spectral energy distributions (SEDs)— that is, no evidence for circumstellar dust or mass loss in their spectra. de Jager(1998) has suggested that all of the mass-losing, high-luminosity F and G-type supergiants are in a post-RSG state. The Galactic hypergiant IRC +10420 has long been acknowledged as a post-RSG (Jones et al., 1993; Oudmaijer et al., 1996). With its complex CS environment, large infrared excess, high mass-loss rate, and mass-loss history, it is in many ways the best example (Humphreys et al., 1997, 2002; Oudmaijer, 1998; Shenoy et al., 2016). Others include HR 5171A and HR 8752 (Nieuwenhuijzen et al., 2012). In M33, the peculiar Variable A, a high luminosity F-type hypergiant (Humphreys et al., 2006), with its apparent transit in the HR Diagram to cooler temperatures due to a high mass-loss episode is another candidate for post-RSG evolution. Humphreys et al.(2013) (hereafter, Paper I) has identiﬁed several additional warm hypergiants in M31 and M33 with dusty ejecta, strong stellar winds, and high mass-loss rates similar to their Galactic counterparts that are likely post-RSGs. As part of our larger program on the luminous and variable stars in M31 and M33 Humphreys et al.(2013, 2014), in this paper we present a more comprehensive survey of the yellow and red supergiants. We use the presence of circumstellar dust in their long- wavelength SEDs and spectroscopic indicators of mass loss and winds to identify candidates for post-RSG evolution. We likewise use the presence of a large infrared excess in the SEDs of the red or M-type supergiants to identify those with high mass loss. One of the greatest observational challenges is to separate the member supergiants from the signiﬁcant foreground population of yellow and red dwarfs and halo giants in the Galaxy. In the next section, we describe our target selection, foreground contamination, and observations. In §2.3 we discuss the yellow supergiant population and our selection of the post-RSG candidates. The SEDs and the role of circumstellar dust on the luminosities of the red supergiants are presented in §2.4, and in §2.5 we present estimates of the mass loss for the dusty RSGs. In the last section, we discuss the resulting HRDs for the yellow and red supergiants, and compare the candidate post-RSG population with evolutionary track models.

2.2 Sample Selection and Observations

2.2.1 Target List

Our targets were primarily selected from the published surveys of M31 and M33 for yellow and red supergiants (Drout et al., 2009; Massey et al., 2009; Drout et al., 2012). Their Chapter 2. Luminous and Variable Stars in M31 and M33 10 red and yellow candidates were all chosen from the Local Group Galaxies Survey (LGGS; Massey et al., 2007b). Although their adopted magnitude limit and color range for the YSGs (V < 18.5 and 0.4 B V 1.4) corresponds to that of F- and G-supergiants, the ≤ − ≤ same color range will include a large fraction of foreground contamination from Galactic dwarfs and halo giants (Massey et al., 2006; Drout et al., 2012). To establish membership for the yellow candidates, the Drout et al.(2009) M31 survey relied on radial-velocity measurements. However, as their Figure 10 illustrates, even restricting the candidates to the velocity range expected for M31 includes substantial foreground contamination. Drout et al.(2009) therefore used a relative velocity—the measured velocity compared to the expected velocity of the star at its position in M31—to establish probable membership. They identified 54 rank-1 (highly probable) and 66 rank-2 (likely) yellow supergiants in M31 from a sample of 2901 targets. The 96% foreground contamination clearly demonstrates the difficulties of color and magnitude criteria for determining membership. For the candidate YSGs in M33 (Drout et al., 2012), the authors again relied on the relative radial velocities, but also added a measurement of the strong luminosity sensitive O I λ7774 blend in A and F-type supergiants, which greatly increased the probability that the stars were supergiant members. With these criteria, they identified 121 rank-1 YSGs and 14 rank-2 in M33. Fortunately for the RSG candidates, the two-color B V vs. V R diagram has been − − demonstrated as an effective metric for distinguishing red dwarfs and supergiants (Massey et al., 2009; Drout et al., 2012), from which the authors identify 437 RSG candidates in M31 and 408 in M33. For the M31 candidates, 124 had additional radial-velocity information for membership determination, and 16 were spectroscopically confirmed as M-type supergiants. For M33, the 408 candidate RSGs from Drout et al.(2012) were reduced to 204 (189 rank-1, 15 rank-2) likely RSGs using radial-velocity criteria. In addition to the 120 and 135 YSG candidates from the Drout/Massey catalogs of M31 and M33, respectively, we include 18 confirmed YSGs from Humphreys et al.(2014) (hereafter, Paper II), and seven warm hypergiants from Paper I, 39 Hα emission stars with intermediate colors from the survey by Valeev et al.(2010), and seven H α emission sources from an unpublished survey by K. Weis (see Paper II). With these catalogs, we assembled a final target list of 124 and 165 candidate YSGs (after cross-identification among the listed works) for spectroscopy from M31 and M33. We did not obtain follow-up spectroscopy of the RSG candidates; our discussion of them instead relies on published photometry and analysis of their SEDs for circumstellar dust (§2.4). Chapter 2. Luminous and Variable Stars in M31 and M33 11

Table 2.1 Journal of Observations

Target Date Exp. Time Grating, Tilt (UT) (minutes)

M31A-Blue 2013 Sep 25 120 600l, 4800A˚ M31A-Red 2013 Sep 26 90 600l, 6800A˚ M31B-Blue 2013 Oct 12 120 600l, 4800A˚ M31B-Red 2013 Oct 9 90 600l, 6800A˚ M33-Blue 2013 Oct 7 120 600l, 4800A˚ M33-Red 2013 Oct 7 90 600l, 6800A˚ M33-Blue 2014 Nov 29 120 600l, 4800A˚ M33-Red 2014 Nov 16 90 600l, 6800A˚ M31A-Blue 2015 Sep 20 120 600l, 4800A˚ M31A-Red 2015 Sep 20 90 600l, 6800A˚ M31B-Blue 2015 Sep 20 120 600l, 4800A˚

2.2.2 Observations

Our spectra of the YSG candidates were obtained with the Hectospec Multi-Object Spectro- graph (Fabricant et al., 1998, 2005) on the MMT at Mount Hopkins over several observing sessions in 2013, 2014, and 2015. The Hectospec has a 1◦ field of view with 300 fibers each subtending 100. 5 on the sky. We used the 600 line mm−1 grating with a tilt of 4800A˚ yielding 2500A˚ coverage with 0.54A˚ pixel−1 resolution and R of 2000. The same grating with ≈ ∼ a tilt of 6800A˚ was used for the red spectra with similar coverage and resolution and R of 3600. A total integration time for each field was 90 minutes for the red and 120 minutes ∼ for the blue. The journal of observations is in Table 2.1. Due to the large angular size of M31 on the sky, observations of this galaxy were split across two fields, labeled A and B in Table 2.1, centered at 00:43:36.5 +41:32:54.6 and 00:41:15.9 +40:40:31.2, respectively. Weather conditions at Mount Hopkins during the 2015 season prevented observations on the last set of supergiant candidates in the red filter setting. The M33 targets were observed in a single field, shown in Figure 2.1. The spectra were reduced using an exportable version of the CfA/SAO SPECROAD package for Hectospec data.1 The spectra were bias subtracted, flat-fielded, and wavelength calibrated. Due to crowding, sky subtraction was performed using pre-selected sky fibers off the field of each galaxy. These sky fiber positions were chosen from Hα maps in regions where nebular contamination would be minimized. Flux calibration was done in IRAF using standard stars Feige 34 and 66 from Hectospec observations during the 2013-2015 seasons.

1 External SPECROAD was developed at UMN by Juan Cabanela for use on Linux or MacOS X systems outside of the CfA. It is available online at http://astronomy.mnstate.edu/cabanela/research/ESPECROAD/. Chapter 2. Luminous and Variable Stars in M31 and M33 12

30°50'00" 0 0

0 40'00" 2

30'00"

1:35:00 34:30 00 33:30 00 32:30

2000 Figure 2.1 V-band mosaic of M33 observed from the Kitt Peak 4m telescope on 2000 Oct 4 (Massey et al., 2006). The circles show the positions of the MMT Hectospec ﬁbers for some of the targets in our spectroscopic sample; yellow for the YSG candidates, red for the RSG candidates.

In M31, we obtained spectra for 113 of the 120 YSG candidates from Drout et al.(2009) plus follow-up spectra for the 10 previously conﬁrmed warm supergiant and hypergiant stars from Papers I and II (6 of which were cross-listed in the Drout catalog) for a total of 117 spectra. In M33, 71 of the 135 YSG candidates from Drout et al.(2012) were observed, plus 14 conﬁrmed supergiants from Papers I and II (4 cross-listed in the Drout catalog), and 37 Hα-emission sources from Valeev et al.(2010) (15 cross-listed in the Drout and Humphreys catalogs) for a total of 103 spectra. For all of these sources, as well as the remaining YSG candidates from the Drout catalogs for which we did not observe spectra, we obtain photometry from published catalogs, discussed in §2.3.2. Chapter 2. Luminous and Variable Stars in M31 and M33 13 2.3 The Intermediate-Type or Yellow Supergiants

2.3.1 Spectral Characteristics

Our primary goal for the spectroscopy of the YSG candidates is to search for evidence of mass loss and winds from emission lines and P Cygni profiles when present. Since foreground contamination is an obstacle for identifying yellow and red supergiant members in external galaxies, the same spectra can be used for spectral and luminosity classification. We refine the classification of the YSG candidates with established luminosity and spectral type indicators in the blue and red spectra. The blends of Ti II and Fe II at λλ4172-8 and λλ4395-4400 are strong luminosity criteria in the blue when compared against Fe I lines that show little luminosity sensitivity such as λ4046 and λ4271. The O I λ7774 triplet in the red spectra is also a particularly strong luminosity indicator in A- to F-type supergiants. This feature is present in all of the candidate YSGs from M33, since Drout et al.(2012) identified all probable members based on this criterion. The Sr II λλ4077, 4216 lines are especially useful for temperature classifications of the yellow supergiants. Comparing the relative strength of Sr II λ4077 to the nearby Hδ feature, and the relative strengths of the Fe II λ4233 and Ca I λ4226 lines, for example, provide a quick diagnostic for all F-type supergiants. Later type supergiants (G-type) are identified by the growth of the G-band, a wide absorption feature around λ4300 A˚ due to CH. Luminosity criteria for G-type stars is similar to that in F supergiants. The Mg I triplet λλ5167, 72, 83 is strong in the later type dwarfs and allows for filtering foreground contaminants from our sample. In M31, we identify 75 yellow supergiants, including the previously-confirmed stars from Papers I and II. Seventy of the 113 observed stars from Drout et al.(2009) are confirmed YSGs, and 42 are foreground dwarfs or subgiants. Therefore, the Drout et al.(2009) M31 catalog was 35% contaminated by foreground stars. The remaining 8 rank-1/rank-2 ∼ candidates Drout et al.(2009) for which we did not obtain spectra are analyzed in §2.3.3 for evidence of mass loss in their SEDs along with the confirmed YSGs. We identify 86 yellow supergiants in M33, which also includes the warm supergiants and hypergiants discussed in Papers I and II. Sixty-two of the 71 observed candidates from Drout et al.(2012) are spectroscopically confirmed as yellow supergiants. The remaining 9 observed sources were identified as foreground dwarfs. Thus, the M33 catalog was only 7% contaminated by ∼ foreground stars. Since their M33 survey used the luminosity-sensitive O I λ7774 line in addition to relative velocities, the cleaner sample is not surprising. Chapter 2. Luminous and Variable Stars in M31 and M33 14

Twelve YSGs in M31—including the hypergiants M31-004322.50, M31-004444.52, M31- 004522.58, and hypergiant candidate J004621.08+421308.2 (see §2.3.4)—and 18 in M33 (including hypergiants B324, Var A, N093351, and N125093) exhibit spectroscopic evidence for stellar winds. The notable spectral features include P Cygni profiles in the hydrogen emission lines, broad wings in Hα or Hβ emission indicative of Thomson scattering, and [Ca II]/Ca II triplet emission. Example spectra are shown in Figure 2.2 highlighting two stars with Hα emission indicative of stellar winds and circumstellar outflows. We find that approximately 17% of the observed YSGs in M31 and 21% in M33 demonstrate evidence for mass loss in their spectra. We discuss the evidence for circumstellar dust ejecta in their SEDs in §2.3.3. Representative A- and F-type supergiants from both galaxies are illustrated in Figure 2.3.

12000 M31 M33 D-004259.95 D-013229.20 13000 A5 A8 11000

12000

10000

11000

9000 10000

8000 9000

7000 8000

6000 7000

5000 6000

6530 6540 6550 6560 6570 6580 6590 6530 6540 6550 6560 6570 6580 6590

Wavelength (Å) Wavelength (Å)

Figure 2.2 Two yellow supergiant spectra with Hα emission. The broad wings in both spectra are indicative of Thomson-scattering. D-004259.95 (left) has strong P Cygni absorption, while D-013229.20 (right) has double-peaked emission suggesting bipolar outﬂow or a rotating circumstellar disk. Spectra are plotted in arbitrary counts for display.

Table 2.2 is a list of the conﬁrmed YSGs in both galaxies, 75 in M31 and 86 in M33, with their spectral types. Notes to the table include comments on the evidence for winds and mass loss in their spectra and references to cross-identiﬁed objects. The rank is included for stars from the Drout surveys. Chapter 2. Luminous and Variable Stars in M31 and M33 15

M31 D-003943.43 F0

Hβ Hα

M31 D-003949.86 F8

M33 [N II] 060906 A5

[S II] M33 130270 A8

O I

4000 4500 5000 5500 6000 6500 7000 7500 8000 Wavelength ( ) Wavelength ( )

Figure 2.3 Sample A- and F-type supergiant spectra from M31 and M33. Spectra are not ﬂux calibrated or rectiﬁed.

After publication of this work (Gordon et al., 2016), P. Massey published a spectroscopic survey of supergiants in M31 and M33 (Massey et al., 2016). Of the 75 YSGs in M31 Chapter 2. Luminous and Variable Stars in M31 and M33 16

Table 2.2 Spectroscopically-Conﬁrmed YSGs

Star Name RA DEC Sp Type Notes Alt Desig/Refa Rankb

M31 M31-004247.30 J004247.30 +414451.0 F5 Paper II 2 M31-004322.50 J004322.50 +413940.9 A8-F0 warm hypergiant Paper I M31-004337.16 J004337.16 +412151.0 F8 Paper II 2 M31-004350.50 J004350.50 +414611.4 A5 P Cyg H em Paper II 2 J004410.62 +411759.7 F2 2 M31-004424.21 J004424.21 +412116.0 F5 Paper II 2 J004427.76 +412209.8 F5 neb em 2 J004428.99 +412010.7 F0 2 M33 M33C-4640 J013303.09 +303101.8 A0-2 weak He I, Fe II em Paper II J013303.40 +303051.2 F5 neb em V-021266 1 J013303.60 +302903.4 F8-G0 G-band 1 J013311.70 +302258.9 F0-2 V-028576 J013410.61 +302600.5 F5-8 V-119710 M33-013442.14 J013442.14 +303216.0 F8 Paper II 1 J013446.93 +305426.5 A2 1

aV- or N preﬁx indicates the source identiﬁcation is from Valeev et al.(2010). bRanks from Drout et al.(2009, 2012) specify if the source was a (1) “highly likely” or (2) “possible” supergiant. (This table is available in its entirety in a machine-readable form in the online journal, or in searchable form at: http://vizier.cfa.harvard.edu/viz-bin/VizieR-3?-source=J/ApJ/825/50/table2. A portion is shown here for guidance regarding its form and content.)

identified here, 40 had consistent spectral types in Massey et al.(2016). Similarly, of the 86 YSGs identified in M33, 22 were given spectral types in their work. Differences in spectral classification are typically within the same spectral type, e.g. F5 vs. F8. Reduced spectra for the confirmed YSGs and foreground stars observed from 2013 to 2015 can be found at http://etacar.umn.edu/LuminousStars/M31M33/.

2.3.2 Multi-Wavelength Photometry

For each source in our target list, we cross-identify the visual photometry from the LGGS (Massey et al., 2006) with the near- and mid-infrared photomery from 2MASS (Skrutskie et al., 2006) at J, H, and Ks, the Spitzer/IRAC surveys of M31 (Mould et al., 2008) and M33 (McQuinn et al., 2007; Thompson et al., 2009) at 3.6, 4.5, 5.8 and 8 µm, and WISE (Wright et al., 2010) at 3.4 (W1), 4.6 (W2), 12 (W3), and 22 (W4) µm. For cross- identification between the 2MASS and IRAC coordinates, we use a search radius of 000. 5. The WISE satellite has angular resolutions of 600. 1, 600. 4, 600. 5, and 1200. 0 in the four bands, which presents some issues for cross-identification in the crowded M31 and M33 fields. We Chapter 2. Luminous and Variable Stars in M31 and M33 17 selected a 600 search radius for matching the LGGS/2MASS coordinates to WISE, which is consistent with the FWHM of the WISE PSF at 3.4 µm (Wright et al., 2010). Since the longer-wavelength photometry has such a large beamsize, we recognize that some of our matched candidates may contain multiple sources or be contaminated by PAH emission. To mitigate this, the prime candidates for circumstellar dust (as characterized by infrared excess in the WISE bands, §2.3.3) were each checked visually in the 2MASS Ks-band images, and the photometry was rejected if the sources were likely composites. The resulting multi- wavelength photometry for the spectroscopically-confirmed yellow supergiants, as well as the YSG candidates, are summarized in Table 2.3. Mould et al.(2008) and McQuinn et al.(2007) additionally provide catalogs of infrared variable sources in M31 and M33, respectively. We checked each source for variability against those catalogs, as well as the DIRECT survey (Kaluzny et al., 1998; Macri et al., 2001), and find low-amplitude fluctuations (< 0.1 mag) in the IRAC bands, most likely associated with Alpha Cygni variability. For M33 sources, we also check against the Hart- man et al.(2006) optical survey and find similarly low-level flux variability in g0, r0, and i0 band observations. YSGs and YSG candidates that show variability in either the optical or infrared are indicated in Table 2.3. Chapter 2. Luminous and Variable Stars in M31 and M33 18 d ID. M31- m Var µ LGGS m 22 µ m 12 µ 4.6 c m µ m 3.4 µ m 8.0 µ . A portion is shown here for guidance regarding its m 5.8 µ 4.5 b m µ bands in Mould et al. ( 2008 ) for M31 or McQuinn et al. ( 2007 ) for M33, . IRAC IRAC . Photometry of YSGs and YSG candidates survey ( Kaluzny et al. , 1998 ; Macri et al. , )2001 and from Hartman et al. ( 2006 ) for M33. WISE DIRECT Table 2.3 m photometry from Spitzer/ µ m photometry from µ U B V R I J H K 3.6 a Indicates the source was identified as variable in the 3.6, 4.5, 5.8, and 8.0 D- indicates the source was listed in Drout et al. ( 2009 , 2012 ) with the name specifying the RA coordinate of its 3.4, 4.6, 12, and 22 (This table is available in its entirety in a machine-readable form in the online journal, or in searchable form at: Star Name d a c b form and content.) or variable in the optical from the http://vizier.cfa.harvard.edu/viz-bin/VizieR-3?-source=J/ApJ/825/50/ysg or M33C- indicate apaper. star The name complete given RA in and Paper DEC I designations or are provided II. in The Table shorthand 2.2 namingand convention in is the for electronic ease version. of matching to other tables in this M31 D-003907.59D-004009.13M31-004247.30 18.0D-004255.16 17.6 17.1 19.1 16.7M31-004337.16 16.9 18.5 16.3 16.4 17.6M33 18.6 19.1 15.8 16.0 17.2D-013231.94 17.2 18.7 15.3 15.6 16.8 17.0 17.8 14.8Var 15.3 16.1 A 16.6 17.3 14.9 15.0 15.8M33C-4640 16.1 18.5 16.8 15.0 15.4D-013439.98 15.8 18.1 16.1 ... 14.9 15.4 17.4 15.7M33C-013442.14 ... 15.5 17.0 15.5 18.4 16.4 17.6 16.6 14.3 18.2 17.1 ... 14.4 20.1 17.3 17.3 17.0 ...... 19.8 16.8 16.9 16.9 14.3 14.5 18.8 16.5 16.4 16.7 ...... 18.2 16.1 16.0 13.7 ...... 13.6 17.7 ... 15.7 15.2 ... 15.4 14.6 14.8 ...... 12.7 15.2 ... 13.7 15.3 ... 14.3 15.0 ...... 14.6 14.6 13.2 15.1 12.3 15.0 ... 14.6 ... 16.2 14.1 14.5 ... 8.8 ... 13.3 14.6 11.4 16.6 ... 12.1 ... 11.6 12.2 8.6 11.7 9.1 ...... 8.3 11.4 13.8 ... V 14.4 ... 10.2 15.2 13.0 ... 14.1 13.2 10.4 15.0 16.6 9.4 ... 12.6 12.0 8.1 15.9 6.5 8.8 8.8 12.8 8.9 7.4 V Chapter 2. Luminous and Variable Stars in M31 and M33 19

2.3.3 Extinction Correction and the Spectral Energy Distributions

To determine whether the yellow supergiants have excess free-free emission in the near- infrared (1 2 µm) due to stellar winds and/or an excess at longer wavelengths due to − circumstellar dust, we must ﬁrst correct the SEDs for interstellar extinction. Many of these targets are likely embedded in their own circumstellar ejecta or warm circumstellar dust. Additionally, we have noticed in our previous work that the extinction can vary considerably across the face of these galaxies, especially in M31. Drout et al.(2009) assumed a ﬁxed E(B V ) = 0.13 reddening law for all YSGs in M31, and Drout et al.(2012) similarly − adopted E(B V ) = 0.12 for sources in M33. We instead proceed more conservatively and − calculate the extinction for each source individually. For those stars with spectral types, we compare the observed B V color to the intrinsic − colors of supergiants from Flower(1977) and calculate AV from the standard extinction curves (Cardelli et al., 1989) with R = 3.2. This procedure is uncertain for stars with strong emission lines in their spectra, so we also estimate the visual extinction using two other methods: the reddening-free Q-method (Hiltner & Johnson, 1956; Johnson, 1958) for nearby OB-type stars in the LGGS within 2 300 of each target, assuming that their UBV − colors are normal, and the relation between the neutral hydrogen column density (NH ) and 2 the color excess, EB−V (Savage & Jenkins, 1972; Knapp et al., 1973).

We measure NH from the recent H I surveys of M31 (Braun et al., 2009) and M33 (Gratier et al., 2010). Since we do not know the exact location of the stars along the line-of-sight with respect to the neutral hydrogen, we follow Paper II and deﬁne the total

AV as the foreground AV ( 0.3 mag) plus half of the measured NH . Since the H I ≈ surveys have spatial resolutions of 3000 and 1700 for M31 and M33, respectively, we favor the extinction estimates from the two other methods when available. We use the Q-method and

NH measurements for the supergiant candidates for which we did not obtain spectra. The

results from these diﬀerent methods, the adopted AV , and the resulting extinction-corrected

MV are summarized in Table 2.4. For the spectroscopically-conﬁrmed YSGs with known spectral types, we calculate the

bolometric luminosities by applying bolometric corrections from Flower(1996) to MV . Bolo- metric corrections for stars in this temperature range are small, typically 0.2 mag. For ≤ | | the candidate supergiants without spectra, we integrate the SED from the optical to the

2MASS Ks band (2.2 µm). If the SEDs show an infrared excess, and thus evidence of circumstellar dust, we integrate the SED out to the IRAC 8 µm band and/or the 22 µm WISE band if available and if not obviously contaminated by nebulosity in the beam. Those

2 21 −2 −1 for R = 3.2, NH /AV = 1.56 × 10 atoms cm mag (Rachford et al., 2009). Chapter 2. Luminous and Variable Stars in M31 and M33 20 sources are indicated with an asterisk in Table 2.4. Figures 2.4 and 2.5 are example SEDs from supergiants in M31 and M33. The observed visual, 2MASS, and IRAC magnitudes are shown as filled circles, and the WISE data as open circles. The optical and 2MASS extinction-corrected photometry are shown as grey boxes. We fit a blackbody to the extinction-corrected optical photometry to model the contribution from the central star. If the flux in the near-infrared 2MASS and IRAC bands exceeds the expected Rayleigh-Jeans tail of the stellar component, we identify this as an infrared excess. Many of the supergiants in our sample, show the characteristic upturn redward of 8 µm due to PAH emission (Draine & Li, 2007). Due to the large beamsize of WISE, it is likely that some sources are contaminated by H II region PAH emission in the mid-infrared. However, if a source also has an apparent excess in the near-infrared 2MASS and IRAC bands, the infrared photometry provides evidence of mass loss. An infrared excess in the 1 2 µm 2MASS bands is characteristic of free-free emission in stellar − winds, with the 3.6 to 8 µm IRAC data providing evidence for warm CS dust. Free-free emission is generally identified as constant Fν in the near-infrared, often extending out to 5 µm (see Figure 2.7). The IRAC photometry can be used to estimate the mass of the dusty circumstellar material (see §2.5.1). We note that the data provided in the broadband visual (LGGS), near- (2MASS) and mid-infrared (IRAC and WISE) photometry were not all observed simultaneously. The resulting SEDs, then, do not represent a single snapshot in time. Twenty-six YSG candidates have indicators for free-free emission in the near-IR 2MASS photometry and/or CS dust emission in the mid-IR IRAC or WISE bands. D-004009.13, as its SED in Figure 2.4 shows, likely has both nebular contamination and dust. Combining both the spectroscopic and photometric data, we find a total of 32 sources in M31 with evidence for mass loss either from the stellar wind features in their spectra or free-free/CS dust emission in their SEDs. Six show evidence for both; the warm hypergiants: M31- 004322.50, M31-004444.52, M31-004522.58, the new hypergiant J004621.08+421308.2 (see §2.3.4), and two F-type supergiants: M31-004424.21, M31-004518.76. We do not have spectra for two of the sources with evidence of free-free emission, D-003745.263 and D- 003936.96, so we cannot confirm membership in M31. Therefore, of the 75 confirmed YSGs in M31, 30 (or 40%) are likely post-RSG candidates, plus two sources that require follow-up spectroscopy to confirm supergiant status. Five sources (D-003711.98, D-003725.57, D- 003907.59, D-004102.78, D-004118.69) are likely contaminated with nebular PAH emission

3 This source is identiﬁed as an F5 supergiant by Massey et al.(2016). D-003936.96 and the six others for which we did not observe spectra remain unclassiﬁed. Chapter 2. Luminous and Variable Stars in M31 and M33 21

Table 2.4 Extinction and Luminosities of YSGs and YSG Candidates

∗ Star Name Sp Type AV (colors) AV (stars) AV (NH ) Adopted AV MV MBol M31 D-003926.72 A2 0.7 0.7 1.4 0.7 -7.1 -7.3 D-003936.96 ...... 0.9 0.5 0.9 -7.2 -7.5* D-003948.85 F2-5 1.2 ... 1.0 1.2 -8.3 -8.4 M31-004247.30 F5 0.6 ... 0.9 0.6 -8.7 -8.9* M31-004522.58 A2 0.4 1.4 1.1 0.4 -6.4 -7.2* M33 M33C-4640 A0-2 0.6 0.6 0.6 0.6 -8.1 -8.3 N045901 F5 1.2 0.9 0.5 1.2 -8.5 -8.4* V071501 A5 0.6 ...... 0.6 -7.0 -7.0 N125093 F0-2 ...... 0.8 0.8 -8.8 -8.9* D-013439.73 A5-8 0.6 0.9 0.6 0.6 -8.0 -8.0

∗ indicates the presence of an IR excess, and thus MBol was calculated by integrating the SED out to the mid-infrared. (This table is available in its entirety in a machine-readable form in the online journal, or in searchable form at: http://vizier.cfa.harvard.edu/viz-bin/VizieR-3?-source=J/ApJ/825/50/ysg. A portion is shown here for guidance regarding its form and content.)

from nearby H II regions. In M33, 22 stars show evidence for free-free emission and/or CS dust emission in their SEDs. Combining the spectroscopic and photometric data indicators, we find a total of 30 sources in M33 with evidence for mass loss. Eight have both the spectroscopic stellar wind and circumstellar dust features; the warm hypergiants: Var A, M33-013442.14, N093351, N125093, and the supergiants: V002627, D-013233.85, V021266, V130270, V104958. Three of the sources have not been spectroscopically confirmed as members of M33 (D-013345.50, D-013349.85, D-013358.05).4 Therefore, of the 86 confirmed YSGs in M33, we identify 27 (or 31%) as likely post-RSG candidates, plus three sources requiring follow-up spectroscopy. ∼ Thirty-three sources show evidence for nebular PAH contamination. In Table 2.5 we list all the stars, both confirmed and unconfirmed YSGs, that show at least one indicator of circumstellar ejecta: spectroscopic evidence of a stellar wind, free- free emission and/or thermal dust emission in the SED. If the stellar spectrum contains nebular emission markers such as [O III], we indicate in the table that the source is likely contaminated with nebular emission. Estimates of total mass lost are discussed in §2.5.1.

4 These three sources are also in the Massey et al.(2016) catalog, but without spectral types. Chapter 2. Luminous and Variable Stars in M31 and M33 22

D-004009.13, F2-5

AV = 1.8 10-14 MBol = -8.6 ¢ 2

m -15 / 10 W ¡ λ F λ 10-16

10-17

M31-004337.16, F8

AV = 0.8 10-14 MBol = -8.4 ¢ 2

m -15 / 10 W ¡ λ F λ 10-16

10-17 0. 2 0. 5 1 2 5 10 20 λ (µm)

Figure 2.4 SEDs of two F-type supergiants in M31. The observed visual, 2MASS, and IRAC magnitudes are shown as filled circles, and the WISE data as open circles. The extinction-corrected photometry is plotted as filled squares. The SEDs of both stars show evidence for circumstellar dust. The dotted grey lines represent blackbody fits to the optical components of the SEDs, with color temperatures of 7500 K and 6700 K for top and bottom. While D-004009.13 shows the characteristic PAH upturn in the W3 and W4 bands, its SED has an excess between 3 and 8 µm; therefore, its infrared excess is very likely a combination of free-free emission, CS dust emission, and H II region contamination from PAH emission. MBol is calculated by integrating the SED through the IRAC bands.

2.3.4 Two New Hypergiants in M31

J004621.05+421308.06 was considered a candidate LBV by Massey et al.(2007a) and King et al.(1998). Due to its position in the northern arm of M31, it was outside both of our two Chapter 2. Luminous and Variable Stars in M31 and M33 23

D-013254.37b, A5

AV = 0.6 10-14 MBol = -7.7 ¢ 2

m -15 / 10 W ¡ λ F λ 10-16

10-17

D-013327.98, A8

AV = 0.3 10-14 MBol = -7.2 ¢ 2

m -15 / 10 W ¡ λ F λ 10-16

10-17 0. 2 0. 5 1 2 5 10 20 λ (µm)

Figure 2.5 SEDs of two warm supergiants in M33. The symbols are the same as in Figure 2.4. The upturn at 12 22 µm in both stars is most likely due to nebulosity. − While D-013254.37b has evidence for free-free emission as an excess in 2MASS H and Ks bands, it is unlikely that this star has dusty CS ejecta. Though D-013327.98 has no 2MASS photometry, the obvious infrared excess in the IRAC bands suggests the presence of signiﬁcant mass loss through dust. The color temperatures of the blackbody ﬁts to the optical photometry are 8500 K and 7900 K for the top and bottom panels.

Hectospec ﬁelds for M31. Consequently, we obtained a long-slit spectrum of this target on 22 November 2014 with the MODS1 spectrograph on the Large Binocular Telescope. The instrument setup and observing procedure are described in Paper I. Its blue and red spectra are shown in Figure 2.6. J004621.05+421308.06 has the absorption-line spectrum of a late A- type supergiant with strong Balmer emission lines with deep P Cygni proﬁles and the broad Chapter 2. Luminous and Variable Stars in M31 and M33 24

Table 2.5 YSG and YSG Candidates with Evidence for Stellar Winds and CS Dust

a Star Name Sp Type Wind IR excess Comments Mass Lost (M ) M31 D-004009.13 F2-5 ... yes CS dust 0.07±0.01 × 10−2 M31-004322.50 A8-F0 yes yes CS dust 0.08±0.01 × 10−2 M31-004337.16 F8 ... yes CS dust 0.17±0.02 × 10−2 M31-004424.21 F5 yes yes CS dust 1.42±0.18 × 10−2 M31-004444.52 F0 yes yes CS dust 1.74±0.22 × 10−2 M33 D-013231.94 F2 yes yes CS dust 0.76±0.09 × 10−2 Var A F8 yes yes CS dust 2.36±0.28 × 10−2 D-013349.86 F8 ... yes CS dust 0.87±0.10 × 10−2 M33-013357.73 A0 yes yes CS nebula, H II PAH ... N093351 F0 yes ... 2.06±0.24 × 10−2

aTotal mass lost through circumstellar ejecta estimated from IRAC /WISE photometry. See §2.5.1. (This table is available in its entirety in a machine-readable form in the online journal, or in searchable form at: http://vizier.cfa.harvard.edu/viz-bin/VizieR-3?-source=J/ApJ/825/50/table5.A portion is shown here for guidance regarding its form and content.)

wings characteristic of Thomson scattering. The outflow velocity is -223 km s−1, measured from three P Cygni absorption minima. Numerous Fe II and [Fe II] emission lines are present. Like the warm hypergiants discussed in Paper I, its red spectrum shows the Ca II and [Ca II] line emission indicative of a low-density circumstellar nebula. The O I λ8846 line is also in emission. The SED shown in Figure 2.7 reveals a prominent circumstellar dust envelope in the near- and mid-infrared not observed in the LBVs (Paper II). These properties are shared with the warm hypergiants and probable post-red supergiants experiencing high mass loss. Their photospheres are not due to the cool dense winds formed by an LBV in eruption, but represent the stellar surface. We therefore suggest that J004621.05+421308.06 belongs with the class of warm hypergiants. J004051.59+403303.00 has been described by Massey et al.(2007a) and by Sholukhova et al.(2015) as a candidate LBV. Our blue and red spectra from 2013 shown in Figure 2.6 are similar to the blue spectrum published by Massey et al.(2006) and the blue and red spectra in Sholukhova et al.(2015) suggesting little spectroscopic change in the last 10 years. The spectra show prominent P Cygni profiles in the Balmer lines with broad wings and in the Fe II multiplet 42 lines. The mean outflow velocity measured from the absorption minimum in six P Cygni profiles is -152 km s−1 similar to the LBVs and hypergiants (Papers I and II). There are no other Fe II or [Fe II] emission lines. The relative strengths of the Mg II λ4481 and He I λ4471 lines suggest an early A-type supergiant. Its spectrum is similar to the warm Chapter 2. Luminous and Variable Stars in M31 and M33 25 hypergiant M31-004444.52 in Paper I, but it does not have the [Ca II] emission lines in the red. It also resembles J004526.62+415006.3 in 2010 which was later shown to be an LBV entering its maximum light or dense-wind stage (Sholukhova et al., 2015; Humphreys et al., 2015). Therefore, based on this spectrum, its nature is somewhat ambiguous. Its SED in Figure 2.7 shows an excess in the near-infrared due to free-free emission, as evidenced by constant flux (Fν) out to 5 µm. The WISE photometry at 12 and 22 µm may be due to circumstellar dust from silicate emission, but is more likely contaminated by PAH emission from a nearby H II region and nebulosity. Thus, J004051.59+403303.00 may be a mass-losing post-red supergiant like several of the stars discussed in this paper or a candidate LBV. Future spectroscopic and photometric variability will be necessary to confirm that it is an LBV, but even so, given its luminosity, it is very likely in a post-RSG state similar to the less-luminous LBVs.

2.4 Red Supergiants

With signiﬁcant mass loss, red supergiants can evolve back to warmer temperatures, we examine the SEDs of the RSGs to identify what fraction of these cool supergiants are in a mass-losing state and determine their positions on the HRD. Additionally, we can roughly estimate the total mass lost through circumstellar ejecta from the stars’ infrared photometry. The RSGs currently experiencing episodes of high mass loss may eventually evolve to become post-RSG warm supergiants, LBVs, or WR stars. We cross-identify the visual photometry for the RSGs from the LGGS with 2MASS, IRAC , and WISE. The multi-wavelength photometry is summarized in Table 2.6 for the 437 RSG candidates in M31 (Massey et al., 2009) and the 204 (189 rank-1 plus 15 rank-2) in M33 (Drout et al., 2012). Chapter 2. Luminous and Variable Stars in M31 and M33 26

J004621.05+421308.06 Ca II

Hγ [Ca II] O I Hδ Fe II

Hα

4000 4500 5000 6000 6500 7000 7500 8000 8500 9000

J004051.59+403303.00

Hβ O I Mg II

4000 4500 5000 6500 7000 7500 8000

Wavelength (Å) Wavelength (Å)

Figure 2.6 Top: Warm hypergiant J004621.05+421308.06, LBT MODS1 2014. Absorption-line spectrum appears as a late A-type star with strong Balmer emission lines. Ca II and [Ca II] emission lines indicate the presence of a low-density circumstellar nebula. Bottom: LBV candidate J004051.59+403303.00, MMT Hectospec 2013. Emission-line spectrum shows prominent P Cygni profiles on Balmer lines with broad wings. Measured outflow velocity of -152 km s−1 is similar to the LBVs and hypergiants discussed in Papers I and II. It is most likely in a post-RSG, mass-losing state. Spectra are not flux calibrated or rectified. Chapter 2. Luminous and Variable Stars in M31 and M33 27

J004621.05+421308.06

AV = 1.2 10-14 MBol = -8.0 ¢ 2

m -15 / 10 W ¡ λ F λ 10-16

10-17

J004051.59+403303.00

AV = 0.6 10-14 MBol = -8.4 ¢ 2

m -15 / 10 W ¡ λ F λ 10-16

10-17 0. 2 0. 5 1 2 5 10 20 λ (µm)

Figure 2.7 SEDs of warm hypergiant candidates J004621.05+421308.06 and J004051.59+403303.00. The symbols are the same as in Figure 2.4. The SED of J004621.05+421308.06 reveals a prominent CS dust envelope in the IRAC and WISE bands. The WISE photometry of J004051.59+403303.00 is suggestive of silicate dust emission, but is most likely due to contamination from a nearby H II region and nebulosity. The dotted line is a curve of constant Fν, which is evidence for free-free emission in wind. Chapter 2. Luminous and Variable Stars in M31 and M33 28 m ID. M- µ m 22 LGGS µ m 12 µ 4.6 c m µ m 3.4 µ . A portion is shown here for guidance m 8.0 µ m 5.8 µ 4.5 b m µ . IRAC Photometry of Candidate RSGs . WISE Table 2.6 m photometry from Spitzer/ µ m photometry from µ U B V R I J H K 3.6 a 3.6, 4.5, 5.8, and 8.0 D- indicates the source was listed in Drout et al. ( 2012 ) with the name specifying the RA coordinate of its 3.4, 4.6, 12, and 22 (This table is available in its entirety in a machine-readable form in the online journal, or in searchable form at: b c a Star Name indicates the source wastables listed in in thisMassey paper. et The al. ( 2009 ). complete RA The and shorthand DEC naming designations convention are is provided for in ease the of electronic matching version. to other http://vizier.cfa.harvard.edu/viz-bin/VizieR-3?-source=J/ApJ/825/50/rsg regarding its form and content.) M31 M-003703.64 ...M-003723.56 22.1 21.6M-003724.48 20.8 19.9 18.9 ...M-004120.25 19.0 17.9 23.1 17.7 21.7M-004444.66 16.9 20.9 16.6 19.6 22.5 15.7 18.9 15.7M33 18.5 20.9 14.9 17.6 15.7 17.5 19.0 14.8 16.3D-013217.79 16.2 18.0 14.9 20.5 ... 15.0 16.9D-013224.33 ... 14.0 19.6 15.1 15.8 20.9 13.7 18.4D-013312.26 14.8 21.4 17.8 16.3 19.1 13.0 14.8 19.6D-013421.55 ... 17.2 17.6 ... 18.7 16.6 14.2 16.0 16.2D-013502.06 ... 17.8 13.2 15.7 15.1 21.0 16.7 15.9 21.2 14.3 ... 13.8 19.9 16.1 ...... 12.9 19.3 18.5 15.8 15.6 18.1 13.3 17.9 14.1 11.8 12.4 16.8 15.7 12.6 17.3 15.8 ... 15.4 ... 12.4 16.5 14.1 14.8 14.6 13.4 15.9 16.0 12.4 14.4 16.6 ... 15.5 14.5 14.5 15.1 13.6 14.2 15.8 ... 12.4 15.7 14.6 12.7 14.7 12.7 14.3 ... 15.8 ... 12.9 12.3 9.4 12.4 ... 9.2 ... 15.8 ... 9.4 9.1 9.3 12.1 15.7 16.0 13.5 12.3 ... 15.9 12.9 14.0 12.3 12.4 15.9 8.8 14.1 12.0 9.5 16.0 12.3 8.7 12.8 8.7 9.0 Chapter 2. Luminous and Variable Stars in M31 and M33 29

Since we lack spectral type information, we cannot correct for extinction using intrinsic colors. Massey et al.(2009) applied a constant AV = 1 to the entire sample of M31 RSGs, and Drout et al.(2012) adopted a ﬁxed reddening law of E(B V ) = 0.12 for all RSGs − in M33. Here, we follow the methodology for the YSGs and estimate AV from the Q- method for nearby O and B stars and from the neutral hydrogen column density along the line-of-sight to each RSG as described in §2.3.3. Unfortunately, roughly 60% of the RSG candidates lacked nearby O and B stars, so we are forced to adopt the less accurate extinction from the neutral hydrogen. The results from the two methods are summarized in Table 2.7. Bolometric luminosities are calculated by integrating the optical through 2MASS

Ks. Similar to the YSG sources, if the SEDs show an infrared excess in the 2MASS or IRAC photometry of the RSG candidates, we integrate the SED out to the IRAC 8 µm band and/or the 22 µm WISE band if uncontaminated. Those sources are indicated with an asterisk in Table 2.7. Several sources were found to have anomalous photometry in the optical or infrared, possibly due to crowding in the field or source mismatch from the LGGS. Some of these objects with unusually high bolometric luminosities may actually be foreground stars, but without spectra, we cannot confirm membership. These stars are omitted from the HR diagrams in §2.5.2. They are included in the catalogs for completeness and are indicated with a dagger in Table 2.7. Figures 2.8 and 2.9 are example SEDs for RSGs in both galaxies. The symbols follow the same pattern as Figure 2.4. Since these stars are cooler, the peak of the optical thermal component from the star shifts redward in the SED. This makes any infrared excess in the 2MASS bands less discernable than in the YSGs; however, the CS dust component at wavelengths longer than 3.6 µm can still be readily distinguished in most of the RSG candidate sources. For this reason, we divide our SEDs into rankings. Rank-1 SEDs have an infrared excess in the IRAC and/or WISE bands most probably due to CS dust emission. Rank-2 SEDs either have missing IRAC photometry but show an IR excess in WISE, or have an IR excess in the IRAC bands but which is somewhat uncertain due to the characteristic PAH upturn in the WISE bands and are thus possibly contaminated by nebulosity. Figure 2.8 demonstrates two rank-1 RSGs, with SEDs showing excess emission above the color-temperature fits to the optical data. The bottom panel of Figure 2.9 illustrates one of the more ambiguous sources in M33. D-013353.91 has a clear infrared excess at 8 µm, while the infrared photometry D-013506.97 can be easily confused with nebulosity. We consider D-0133506.97 as a rank-2 mass-losing RSG candidate. Of the 437 RSG candidates in M31 from Massey et al.(2009), 231 (129 rank-1 and 102 rank-2) show evidence for circumstellar dust emission in the mid-IR IRAC or WISE bands. Chapter 2. Luminous and Variable Stars in M31 and M33 30

Table 2.7 Extinction and Luminosities of Candidate RSGs

∗ Star Name AV (stars) AV (NH ) Adopted AV MV MBol

M31 M-003739.41 ... 0.8 0.8 -5.9 -6.7 M-003739.88 0.9 0.7 0.9 -6.7 -8.2* M-003907.69 1.8 1.4 1.8 -7.5 -8.3 M-003907.98 1.6 1.4 1.6 -7.4 -8.0* M-004638.17 1.2 1.1 1.2 -6.1 -8.3* M33 D-013339.28† 0.7 ... 0.7 -8.4 -9.8* D-013340.80 0.8 0.6 0.8 -6.1 -7.0 D-013349.09 0.9 0.6 0.9 -6.3 -8.0* D-013349.99 1.9 1.0 1.9 -7.1 -8.2* D-013438.95 0.4 0.5 0.4 -5.5 -7.5*

∗ indicates the presence of an IR excess, and thus MBol was calculated by integrating the SED out to the mid-infrared. †Photometry of sources is anomalous. For some stars in crowded ﬁelds, there may be either a source mismatch between the optical and infrared or the photometry may be contaminated by multiple sources in the aperture. M-004539.99, D-013312.26, and D-013401.88 (included in full table online) are likely foreground stars. Marked sources are omitted from the HR diagrams for above reasons. (This table is available in its entirety in a machine-readable form in the online journal, or in searchable form at: http://vizier.cfa.harvard.edu/viz-bin/VizieR-3? -source=J/ApJ/825/50/rsg. A portion is shown here for guidance regarding its form and content.) Chapter 2. Luminous and Variable Stars in M31 and M33 31

M-003930.30

AV = 0.8 10-14 MBol = -7.2 ¢ 2

m -15 / 10 W ¡ λ F λ 10-16

10-17

M-004108.42

AV = 1.5 10-14 MBol = -8.2 ¢ 2

m -15 / 10 W ¡ λ F λ 10-16

10-17 0. 2 0. 5 1 2 5 10 20 λ (µm)

Figure 2.8 SEDs of two red supergiants in M31. The symbols are the same as in Fig- ure 2.4. Grey dotted lines represent blackbody ﬁts with color temperatures of 4100 K and 3100 K for top and bottom, respectively. The SEDs of both stars show evidence for circumstellar dust. MBol is calculated for both of these sources by integrating the SEDs through the infrared.

Thus, 53% of the candidate M31 RSGs have CS dust.5 An additional 110 candidate RSGs ∼ are likely contaminated with nebular emission. In M33, 126 of the 204 candidate RSGs from Drout et al.(2012), have indicators for CS dust emission in the infrared. Again dividing the 126 sources with infrared excess into ranks from their SEDs, we ﬁnd 53 rank-1 (highly probable) RSGs and 73 rank-2 (likely)

5 Of the 231 candidate RSGs with evidence for mass loss, 152 have spectral classiﬁcations from Massey et al. (2016). Chapter 2. Luminous and Variable Stars in M31 and M33 32

D-013353.91

AV = 1.2 10-14 MBol = -8.6 ¢ 2

m -15 / 10 W ¡ λ F λ 10-16

10-17

D-013506.97

AV = 0.9 10-14 MBol = -7.6 ¢ 2

m -15 / 10 W ¡ λ F λ 10-16

10-17 0. 2 0. 5 1 2 5 10 20 λ (µm)

Figure 2.9 SEDs of two red supergiants in M33. D-013353.91 has an obvious infrared excess at 8 µm, and the relatively constant irradiance through the WISE bands is strong evidence for circumstellar dust emission. The bottom source, D-013506.97, may also have dusty ejecta, but the sharp rise in the W3 and W4 bands is most likely contamination from nebulosity. The color temperatures of the blackbody ﬁts to the optical photometry are 3300 K and 3500 K for the top and bottom panels.

candidates.6 Thus, 60% of our RSG candidates in M33 have evidence for dusty ejecta. ∼ Forty-three sources show the PAH upturn in their SEDs and are likely contaminated with nebulosity. Table 2.8 summarizes the results from both galaxies. We ﬁnd that more than half of

6 Unlike the RSG candidates in M31, these sources in M33 did not have spectral classiﬁcations in Massey et al. (2016). Chapter 2. Luminous and Variable Stars in M31 and M33 33

Table 2.8 RSG Candidates with Evidence for CS Dust

a b Star Name LGGS Rank Mass Lost (M ) M31 M-003930.30 J003930.30+404353.4 1 0.13±0.02 × 10−2 M-004024.52 J004024.52+404444.8 1 0.05±0.01 × 10−2 M-004036.08 J004036.08+403823.1 1 1.07±0.13 × 10−2 M-004031.00 J004031.00+404311.1 1 0.39±0.05 × 10−2 M-004304.62 J004304.62+410348.4 1 0.51±0.06 × 10−2 M33 D-013354.32 J013354.32+301724.6 1 0.16±0.02 × 10−2 D-013401.88 J013401.88+303858.3 1 1.48±0.18 × 10−2 D-013416.75 J013416.75+304518.5 2 0.10±0.12 × 10−2 D-013454.31 J013454.31+304109.8 1 0.37±0.04 × 10−2 D-013459.81 J013459.81+304156.9 2 0.14±0.02 × 10−2

aRank 1 indicates that an infrared excess in the SED is most probably due to CS dust emission. Rank 2 indicates that features in the SED are likely caused by thermal dust emission but may be due to PAH contamination. bTotal mass lost through circumstellar ejecta estimated from IRAC /WISE photometry. See §2.5.1. (This table is available in its entirety in a machine-readable form in the online journal, or in searchable form at: http://vizier.cfa.harvard.edu/viz-bin/VizieR-3?-source= J/ApJ/825/50/table8. A portion is shown here for guidance regarding its form and content.)

the RSG candidates in M31 and M33 exhibit evidence for mass loss. This high fraction is not surprising since M supergiants have been known for decades to have mass loss and dusty circumstellar ejecta (Woolf & Ney, 1969). Our results are consistent with the 45% ∼ found by Mauron & Josselin(2011) in the Milky Way and LMC using the IRAS 60 µm band. Since we have included our rank-2 SEDs in this census, the fractions reported here may represent an overestimate.

2.5 Discussion

2.5.1 Circumstellar Dust and Mass Loss

Thermal emission from dust appears in the mid-infrared IRAC data from 3.6 to 8 µm and is also present in the WISE photometry at longer wavelengths. With some assumptions about the dust grain parameters, we can estimate the mass of the circumstellar material from the mid-infrared ﬂux (see Paper I):

2 4D ρλFλ Mdust = (2.1) 3 (λQλ/a) Bλ (T ) Chapter 2. Luminous and Variable Stars in M31 and M33 34 where D is the distance to the source (here, the average distance to M31/M33), Fλ is the

mid-infrared ﬂux, a is the grain radius, ρ is the grain density, Qλ is the absorption eﬃciency

factor for silicate dust grains, and Bλ (T ) is the blackbody emission at temperature T . For many of the YSG and RSG candidates with an infrared excess, the ﬂux is fairly constant across the mid-infrared, which implies that the dust is emitting over a range of temperatures and distances around the central star. Using the Suh(1999) prescription for silicate dust grains and mass loss around AGB stars, we assume dust grain size of a = 0.1 µm at a density ρ = 3 g cm−3 and an average temperature of 350 K. The absorption

(and emission) efficiency factor Qλ is maximized at the 9.8 µm Si–O vibrational mode (Woolf & Ney, 1969) and the 18 µm O–Si–O bending mode (Treffers & Cohen, 1974), so the flux at these wavelengths would be the ideal tracers of thermal dust emission. Since we lack photometry precisely centered on the silicate features, we calculate the dust mass using the flux at 8 µm, or at 12 µm (W3) if no IRAC data exists for the sources. We assume a nominal gas-to-dust ratio of 100, which allows for the calculation of the total mass lost in each source. The results are summarized in Table 2.5 for the YSGs and Table 2.8 for the RSGs, where the error is calculated as the standard error propagation on the average distance to M31/M33 (< 5%) and the photometric errors for measured flux by IRAC /WISE (< 3% / < 9%; Hora et al., 2004; Wright et al., 2010). For both YSGs and RSGs in M31 and M33, we find a range of at least a factor of 10 for the mass of the circumstellar material. −3 −2 Most of the supergiants have shed 10 10 M , which is consistent with Paper I. ∼ − Mauron & Josselin(2011) apply the de Jager et al.(1988) mass-loss prescription to Galactic −6 −1 RSGs to calculate an average mass-loss rate of 10 M yr from IRAS 60 µm flux. ∼ For our dusty RSGs, we can approximate a timescale probed by the IRAC photometry and thus compare our total integrated mass loss to typical RSG mass-loss rates. If we assume an average dust condensation distance of 250 AU and an outflow velocity of 20 km/s, ∼ we estimate 100 years for the dust condensation time—a rough timescale for the dust we ∼ observe at 8 µm. Considering that the circumstellar ejecta most likely contains dust over a range of temperatures ( 150 400 K), as well as the possibility of episodic mass loss ∼ − −5 −4 −1 in the more massive RSGs, an average mass-loss rate of 10 10 M yr is consistent − −3 −2 with the total mass-lost estimates of 10 10 M over the dust condensation timescale. − For the RSG populations in both galaxies, we plot bolometric luminosity vs. total mass lost in Figure 2.10. The de Jager et al.(1988) formulation predicts an increasing mass-loss rate with luminosity, and we find a similar trend with total mass lost as traced by dust. We separate the rank-1 RSGs, those candidates with clear indication of mass loss in their SEDs, Chapter 2. Luminous and Variable Stars in M31 and M33 35 from the rank-2 RSG candidates. The rank-2 sources yield ejecta masses on the lower end of the RSG sample. These RSG candidates likely have circumstellar dust, but the infrared excess was not as obvious as in the rank-1 SEDs, thus the lower derived mass-loss estimate. Since the highest luminosity RSGs also have the highest mass loss, these dusty RSGs may evolve back to warmer temperatures to become the intermediate-type post-RSGs discussed in this paper.

MBol MBol 6.5 7.0 7.5 8.0 8.5 9.0 9.5 10.0 6.5 7.0 7.5 8.0 8.5 9.0 9.5 10.0

1.5

2.0 ] t s o L

s s

a 2.5 M [

¯ M / M

g 3.0 o l

3.5 M31 M33

4.4 4.6 4.8 5.0 5.2 5.4 5.6 5.8 6.0 4.4 4.6 4.8 5.0 5.2 5.4 5.6 5.8 6.0 log L/L log L/L ¯ ¯

Figure 2.10 Bolometric luminosity vs. total mass lost based on dust measurements for RSG candidates in M31 and M33. Closed circles are the rank-1 RSGs, those with clear evidence for mass loss in their SEDs. Open circles are the less certain mass losers, the rank-2 RSG candidates. We note that the higher luminosity RSGs tend to have lost more mass, consistent with the de Jager et al.(1988) prescription of mass-loss in RSGs.

2.5.2 HR Diagrams

The HR Diagrams for the yellow and red supergiants populations are shown in Figures 2.11 and 2.12. The temperatures for the YSGs are derived from the (B V )0 colors using the − transformations in Flower(1996) for intermediate-type supergiants. As described in §2.3.3, Chapter 2. Luminous and Variable Stars in M31 and M33 36 their luminosities are calculated based on the bolometric corrections given in Flower(1996) or by integrating the SED for those stars with emission line spectra or with circumstellar dust. There are several temperature scales in the literature for red supergiants. Since we do not have spectral types for the RSGs, we adopt the temperatures from Massey et al.(2009) for the M31 RSGs and from Drout et al.(2012) for M33 simply for the purpose of placing them on the HR diagram to compare with the YSG population. Their M31 temperature scale is based on a color–temperature relationship from (V K)0 from MARCS atmosphere − models (Gustafsson et al., 2008), while their temperatures for the M33 stars are based

on (V R)0 color transformations from Levesque et al.(2006) in the LMC, which has a − metallicity similar to M33. The bolometric luminosities for the RSGs are determined from integrating their SEDs (§2.4). Stellar evolution tracks from non-rotating models from Ekstr¨omet al.(2012) are shown on the HR diagrams for ZAMS masses of 15, 25, and 40 M . In both galaxies, the post-RSG candidates are preferentially more abundant at higher luminosities. Comparison with the evolutionary tracks suggests that most of the progenitor main-sequence stars have masses & 20 M . Likewise, the dusty RSGs dominate the higher luminosities. This is most obvious for the M33 population with a smaller sample. This is not surprising, as we know from Figure 2.10 that the mass lost in the RSGs correlates with luminosity. We note the presence of several “warm” RSGs in both galaxies. These red supergiant candidates, with temperatures upwards of 4000 K, fall in the temperature range of the yellow supergiants. The temperature scales are somewhat uncertain, and without spectra of these objects, we cannot conﬁrm that some of them may actually be yellow supergiants. Labeled sources in Figures 2.11 and 2.12 are the warm hypergiants from Paper I, as well as the two new hypergiant candidates, J004051.59+403303.00 and J004621.05+421308.06, discussed in §2.3.4.

2.6 Conclusion

We identify 75 spectroscopically-conﬁrmed yellow supergiants in M31, including the three warm hypergiants from Paper I and 86 in M33 including the 14 previously-known YSGs from Papers I and II. The majority have normal absorption-line spectra, but a signiﬁcant fraction, 30 in M31 and 27 in M33, show evidence for mass loss via stellar winds and/or circumstellar dust in their SEDs. Since the RSG stage is a well-established high mass-losing state, we consider these stars to be excellent candidates for post-red supergiant evolution. Chapter 2. Luminous and Variable Stars in M31 and M33 37

6.0 10 M31-004444.52 40 M ¯

5.5 9 25 M ¯ J004051.59

¯ 8 l L

J004621.08 o / 5.0 B L M g

o M31-004322.50 l 15 M M31-004522.58 ¯ 7

4.5

12000K 9000K 7000K 5000K 4000K 3500K 4.0 4.3 4.2 4.1 4.0 3.9 3.8 3.7 3.6 3.5

log Teff/K

Figure 2.11 HR Diagram of M31. Red circles represent our RSG sample, black circles are the YSGs. Closed symbols are sources with evidence of mass loss, either in their spectra (for the YSGs) or their SEDs (for both the YSGs and RSGs). Non-rotating stellar evolution tracks for three mass bins from Ekstr¨omet al.(2012) are shown for comparison. The stars with mass loss, the post-RSG candidates, appear to dominate the upper portion of the HR diagram. The main-sequence progenitors of these supergiants likely had masses & 20 M . Labeled sources are previously-conﬁrmed hypergiants in M31 as well as the two new hypergiant candidates, J004051.59+403303.00 and J004621.05+421308.06.

Thus, about 30 40% of the observed YSGs are likely in a post-RSG state. The post-RSG − 5 candidates are more common at luminosities above 10 L . Most appear to have initial ∼ masses of 20 40 M , and may be the evolutionary descendants of the more massive red − supergiants that do not explode as supernovae (Smartt et al., 2009). The eventual fate of these stars may be either as “less-luminous” LBVs or WR stars before their terminal explosion; however, in his most recent review, Smartt(2015) argues for an upper limit of

18 M of supernova progenitors, and that more massive stars collapse directly to black ≈ holes.

The less-luminous LBVs (Mbol 8 to -9.5 mag) have high L/M values of 0.5, compared ≈ ∼ Chapter 2. Luminous and Variable Stars in M31 and M33 38

6.0 B324 10

Var A 40 M ¯

5.5 9 N125093 N093351 25 M ¯

¯ 8 l L o / 5.0 B L M g o l 15 M ¯ 7

4.5

12000K 9000K 7000K 5000K 4000K 3500K 4.0 4.3 4.2 4.1 4.0 3.9 3.8 3.7 3.6 3.5

log Teff/K

Figure 2.12 HR Diagram of M33. Symbols are the same as in Figure 2.11. We see clearly that the brighter, more massive RSGs all have evidence for mass loss. Labeled sources are previously-confirmed hypergiants in M33. Var A shown with Mbol 9.5 as discussed in ≈ − Paper I. to the B- and A-type supergiants in the same part of the HR diagram. The most likely explanation is that the LBVs have shed a significant fraction of their mass in a previous state and are now close to their Eddington limit. Consequently, they have also been considered as evidence for post-RSG evolution (Humphreys & Davidson, 1994; Vink, 2012), and would have passed through the YSG region of the HR diagram in their evolution to warmer temperatures. Thus, the mass-losing yellow supergiants may be thought of as the progenitor class of the less-luminous LBVs. We identify two new warm hypergiant candidates in M31, J004621.05+421308.06 and J004051.59+403303.00. The spectra of both stars show strong P Cygni absorption profiles in the Balmer emission lines with broad Thomson scattering wings. J004621.05+421308.06 also has strong Ca II and [Ca II] emission indicative of a circumstellar nebula plus dusty circumstellar ejecta. Both stars are very likely in a post-RSG state. J004051.59+403303.00 Chapter 2. Luminous and Variable Stars in M31 and M33 39 is also considered a candidate LBV. If so, it would be one of the less-luminous LBVs, but future spectroscopy and photometry is necessary for confirmation. The red supergiant sample yielded 231 stars in M31 (53%) and 126 in M33 (60%) with observable dusty emission. Therefore, a large fraction of RSGs are in a mass-losing state. Consistent with Mauron & Josselin(2011) and the de Jager prescription, we find that mass loss correlates with luminosity along the RSG branch. The IRAC 8 µm band provides a reasonable estimate of the total dust mass lost over a timescale of about a century, and we estimate that the RSGs in both galaxies tend to have dusty ejecta on the order of −3 −2 10 10 M assuming a warm dust component of 350 K. If more than 50% of RSGs are − indeed experiencing sufficient mass loss to produce CS dusty ejecta, a large fraction of stars along the red supergiant branch may evolve back towards the blue to become the warm post-RSG stars before their terminal state as supernovae or black holes. We note that our target selection was derived from optical surveys. Therefore, our survey of the most luminous stars in M31 and M33 does not include supergiant stars that may be obscured. Since the most luminous warm and cool supergiant populations are more likely to have the highest mass-loss rates, it is probable that those sources will be highly obscured in the optical by their own circumstellar ejecta. To complete the upper portion of the HR diagram requires a further search through the IRAC data to find the brightest infrared sources.

Research by M. Gordon and R. Humphreys on massive stars is supported by the National Science Foundation AST-1109394. We thank Perry Berlind and Michael Calkins at the MMT for their excellent support and operation of the Hectospec. This paper uses data from the MODS1 spectrograph built with funding from NSF grant AST-9987045 and the NSF Telescope System Instrumentation Program (TSIP), with additional funds from the Ohio Board of Regents and the Ohio State University Oﬃce of Research. This publication also makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation, and from the Wide-ﬁeld Infrared Survey Explorer, which is a joint project of the University of California, Los Angeles, and the Jet Propulsion Laboratory/California Institute of Technology, funded by the National Aeronautics and Space Administration. Chapter 3

Searching for Cool Dust: Infrared Imaging of OH/IR Supergiants and Normal RSGs

Adapted from Gordon et al.(2018a)

Abstract New MMT/MIRAC (9–11 µm), SOFIA/FORCAST (11–37 µm), and Herschel/PACS (70 and 160 µm) infrared (IR) imaging and photometry is presented for three famous OH/IR red supergiants (NML Cyg, VX Sgr, and S Per) and two normal red supergiants (RS Per and T Per). We model the observed spectral energy distributions (SEDs) using radiative transfer code DUSTY. Azimuthal average profiles from the SOFIA/FORCAST imaging, in addition to dust mass distribution profiles from DUSTY, constrain the mass-loss histories of these supergiants. For all of our observed supergiants, the DUSTY models suggest that constant mass-loss rates do not produce enough dust to explain the observed infrared emission in the stars’ SEDs. Combining our results with Shenoy et al. (2016) we find mixed results with some red supergiants showing evidence for variable and high mass-loss events while others have constant mass loss over the past few thousand years.

3.1 Introduction

The evolution and fate of massive stars depends on mass loss and their mass loss histories.

The majority of massive stars ( 9 M ) will pass through the red supergiant (RSG) stage, ≥ long recognized as an important end product of stellar evolution. Recently, Smartt et al.

40 Chapter 3. Searching for Cool Dust 41

(2009) and Smartt(2015) have suggested that RSGs with initial masses greater than 18 M do not explode as supernovae, but may evolve back to warmer temperatures before the terminal explosion or collapse directly to black holes. The RSG stage is also a high mass losing stage, and to what extent mass loss can affect the terminal state of the RSGs is now an open question. Even though the mass-loss mechanism for RSGs is still debated, we can measure the mass lost from the thermal infrared (IR) emission from dust in the circumstellar ejecta surrounding the RSGs. In the first paper of this series (Shenoy et al., 2016), we examined the cold dust in the mid- to far-IR and the mass-loss histories of the famous hypergiants µ Cep, VY CMa, IRC +10420, and ρ Cas, whose mass-loss rates are among the highest observed. In this paper, we present similar observations of three strong IR and maser sources, the OH/IR red supergiants NML Cyg, VX Sgr, and S Per, plus the normal red supergiants RS Per and T Per. OH/IR stars, characterized by strong winds and OH maser emission, are bright IR sources due to thermal dust emission by their own circumstellar ejecta. The more typical red supergiants, without OH or H2O maser emission, also show high mass-loss rates that increase as a function of luminosity (Reimers, 1975; de Jager et al., 1988; Mauron & Josselin, 2011). In this study, we analyze the mass loss in these five red supergiants through observations in the mid-IR with SOFIA/FORCAST (Herter et al., 2012) 11 37 µm imaging, − combined with publicly-available Herschel 1 (Pilbratt et al., 2010) PACS (Poglitsch et al., 2010) images. We also include sub-arcsecond resolution 8 12 µm observations of NML Cyg − (Schuster et al., 2009), S Per, and T Per made with MMT/MIRAC (Hoffmann et al., 1998; Hinz et al., 2000). Finally, we present spectral energy distribution (SED) models from the radiative transfer code DUSTY (Ivezic et al., 1997). These SED models, in combination with azimuthal profiles from FORCAST and MIRAC, provide estimates on mass-loss rates, ejecta dust temperatures, and mass-loss histories.

1 Herschel is an ESA space observatory with science instruments provided by the European-led Principal Investigator consortia and with important participation from NASA. The Herschel data used in this paper are from the Level 2 (ﬂux-calibrated) images provided by the Herschel Science Center via the NASA/IPAC Infrared Science Archive (IRSA), which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with NASA. Chapter 3. Searching for Cool Dust 42 3.2 Observations and Data Reduction

3.2.1 SOFIA/FORCAST: Far-IR Imaging (11 37 µm) − The targets were observed with SOFIA/FORCAST during Cycles 3 & 4 (OBS IDs: 03 0082, 04 0013; PI: R. M. Humphreys). FORCAST is a dual-channel mid-IR imager covering the 5 to 40 µm range. Each channel uses a 256 256 pixel blocked-impurity-band (BiB) array and × provides a distortion-corrected 3.02 3.02 field of view with a scale of 000. 768 pix−1. FORCAST × achieves near-diffraction limited imaging, with a PSF FWHM of 300. 7 in the longest filters. ∼ We elected to image in single-beam mode to maximize throughput. The observations were obtained in the standard two-position chop-and-nod mode with the direction of the nod matching the direction of the chop (NMC). The data were reduced by the SOFIA Science Center using the FORCAST Redux pipelines version 1.0.3 (S Per), 1.0.5 (VX Sgr), 1.0.7 (NML Cyg), and 1.1.0 (RS Per, T Per). After correction for bad pixels and droop effects, the pipeline removed sky and telescope background emission by first subtracting chopped image pairs and then subtracting nodded image pairs. The resulting positive images were aligned and merged. The details of the FORCAST pipeline are discussed in the Guest Investigator Handbook for FORCAST Data Products, Rev. A3.2 Bright point sources cause cross-talk in the horizontal direction on the FORCAST array. To mitigate this effect, chop angles were selected so that the cross-talk pattern from one chop position did not overlap with the other chop position. Additionally, the FORCAST pipeline applies a correction that reduces the effect, although some of the pattern remains for some targets. The effect is strongest for the brightest IR targets, NML Cyg and VX Sgr; it is less so for S Per and is not present in the images of RS Per or T Per. However, one effect that may appear in some of the fainter targets (especially T Per) is possible coma introduced from the NMC chopping pattern.3 This effect may explain the asymmetries in the surface brightness profile of T Per, shown in Figure 3.11 and discussed in §3.3.4. For each of the stars, observations of the asteroid 2 Pallas were used for PSF calibration in the same filters with the four-position slide in either the mirror position (for the short wavelength channel) or the open position (for the long wavelength channel). The color temperature of the asteroid Pallas ( 160 K) is far less than the effective temperatures of ∼ the target stars (&3200 K). This color difference is not ideal for a PSF calibrator, since the cooler source will peak at longer wavelengths, possibly resulting in a broader profile. However, Pallas was the only source observed in each SOFIA cycle under the same conditions

2 Available at https://www.sofia.usra.edu/researchers/data-products 3 See description of optical aberrations in §1.3.1 of the SOFIA Observer’s Handbook, available at https://www.sofia.usra.edu/science/proposing-and-observing/sofia-observers-handbook-cycle-6 Chapter 3. Searching for Cool Dust 43

Table 3.1 Summary of Observations

Target Instrument Date Filtera Int Time PSF FWHMb (UT) (µm) (s) (00)

VX Sgr FORCAST 2015 06 13 F111, F197, F253 77, 81, 232 2.8, 2.9, 2.8 F315, F348, F371 224, 354, 496 3.1, 3.4, 3.6 S Per MIRAC4 2006 11 05 8.9, 9.8 420, 60 0.3, 0.4 FORCAST 2015 02 04 F197, F253 164, 374 2.7, 3.2 F315, F348, F371 374, 705, 1430 3.4, 3.6, 3.8 RS Per FORCAST 2016 02 18 F197, F315, F371 29, 270, 1062 2.6, 3.1, 3.6 T Per MIRAC4 2009 10 02 8.9, 9.8 50, 70 0.32, 0.35 FORCAST 2016 09 17 F197, F315 160, 484 2.6, 3.1 NML Cygc MIRAC3 2006 07 23 8.9, 9.8, 11.9 260, 260, 220 0.3, 0.4, 0.6 FORCAST 2015 09 11 F197, F253 65, 233 2.6, 2.8 F315, F348, F371 320, 582, 342 3.2, 3.4, 3.6

aThe effective wavelengths of the SOFIA/FORCAST filters are: F197 = 19.7 µm, F253 = 25.3 µm, F315 = 31.5 µm, F348 = 34.8 µm, F371 = 37.1 µm. bFor SOFIA, FWHM measured on PSF calibrator source 2 Pallas. Different cycles on SOFIA may have very slight differences in apparent spatial resolution for the same filter sets. For MIRAC, FWHM measured on β And. cNML Cyg MIRAC3/BLINC observations originally published in Schuster et al.(2009). Chapter 3. Searching for Cool Dust 44 and at each wavelength studied in this work. We analyzed another calibrator (α Aur) at 11.1 and 31.5 µm and measured a similar FWHM at each wavelength. We present the profiles of Pallas in the figures below for consistency, acknowledging the possibility that we have overestimated the size of the PSF at the longer wavelengths. The FORCAST pipeline coadds the merged images. We use the standard deviation of the mean of fluxes extracted from the merged images (prior to coadding) as the 1-σ uncertainty of the fluxes in the coadded images of each of our targets. This uncertainty is negligible compared to the 6% uncertainty that we adopt for the flux calibration, per the GI Handbook § 4.1 (Herter et al., 2013). The band-passes of the selected FORCAST filters are 2 such that only small color corrections are required. Based on the Fν ν spectral shapes of ∝ our targets in the relevant ranges, we have applied color corrections of 1.004, 1.071, 1.004, 1.044, 1.025, and 1.025 to fluxes extracted from the F111, F197, F253, F315, F348, and F371 images respectively. An example of the FORCAST images is shown in Figure 3.1. S Per was imaged in five of the FORCAST filters, and the outermost contours shown in each panel represent 1-σ above the background. Each contour moving into towards the center is another σ above the previous one (e.g., for the 37.1 µm image on the far right, the four contours moving from the outermost to the center are 1, 3, 5, 7-σ of the background noise.) Aperture photometry was performed using the open-source Astropy (Astropy Collab- oration et al., 2013) affiliated photutils4 package. Apertures span between 1500 and 2000, chosen to encompass the extended emission around each object. FORCAST photometry is reported in Table 3.2 and included in the SEDs in §3.3. Photometric error is reported as measured uncertainty in the sky background apertures.

3.2.2 Adaptive Optics Mid-IR Imaging (8 10 µm) − NML Cyg, S Per, and T Per were observed with the mid-IR adaptive optics system on the MMT using the Mid-Infrared Array Camera and Bracewell Infrared Nulling Cryostat (MIRAC3/4/MIRAC-BLINC; Hoﬀmann et al., 1998; Hinz et al., 2000; Skemer et al., 2008). The observations of NML Cyg are described in Schuster et al.(2009), and discussed here in §3.3.5. S Per was observed on UT 2006 Nov 05 and T Per on UT 2009 Oct 02 at 8.9 and 9.8 µm. MIRAC achieved Strehl-ratios close to 0.95, providing diﬀraction-limited imaging and stable PSFs (e.g. Biller et al., 2005). MIRAC4 employed a Si:As array with 256 256 × pixels, and observations were made with a standard chop-and nod sequence to remove

4 photutils provides tools for detecting and measuring the photometry of astronomical sources. The software is still in development, with documentation available at https://photutils.readthedocs.io/ Chapter 3. Searching for Cool Dust 45

IR background emission. Cross-talk in the array electronics introduced faint artifacts in the horizontal and vertical directions which is not completely removed by chop-and-nod subtraction. As described in Shenoy et al.(2016), the horizontal cross-talk is mitigated during the data reduction with a code from Skemer et al.(2008). For consistency with the FORCAST photometry, we perform aperture photometry on the MIRAC images using photutils. We report the results in Table 3.2, include the photometry in the SEDs below, and as input to DUSTY.

3.2.3 IRAS, AKARI, WISE, and ISO-SWS (2 100 µm) − To populate the mid-IR SEDs, we include IRAS photometry (and AKARI photometry when available) from point-source catalogs in the literature for RS Per and VX Sgr (Smith et al., 2004), S Per and T Per (Abrahamyan et al., 2015), and NML Cyg (Schuster, 2007). The Abrahamyan et al.(2015) catalog cross-correlates IRAS point sources with WISE, the latter of which presents some issues due to its large beam-size (up to 1200 at 22 µm; Wright et al., 2010). For stars embedded in nebulosity or crowded ﬁelds, the WISE photometry can be systematically too bright. Additionally, optical photometry is compiled from the Extended Hipparchos Compila- tion catalog (XHIP; Anderson & Francis, 2012), the SKY2000 Master Catalog (Myers et al., 2015), or the AAVSO Photometric All Sky Survey (APASS; Henden, 2016) (see SEDs in §3.3). These optical data, as well as the published photometry from IRAS and AKARI, are dereddened using the extinction law from O’Donnell(1994). The values for interstellar extinction AV chosen for each source are listed in Table 3.3 and in the SED captions below. We also compile spectra from ISO-SWS (de Graauw et al., 1996) for all targets except T Per. S Per and RS Per spectra are from the Japanese guaranteed observing time program REDSTAR1 (PI T. Tsuji; Aoki et al., 1998), and NML Cyg and VX Sgr were observed with the AGBSTARS program (Justtanont et al., 1996; Speck et al., 2000). The color and extinction-corrected spectra are displayed in the SEDs below and are provided as near- to mid-IR photometric input to DUSTY.

3.2.4 Herschel/PACS (70, 160 µm)

We also include in our analysis the publicly-available 70 and 160 µm observations made with Herschel/PACS. VX Sgr, NML Cyg, and S Per were observed as part of the Herschel key program Mass-loss of Evolved StarS (MESS; Groenewegen et al., 2011). The Herschel Chapter 3. Searching for Cool Dust 46

Interactive Processing Environment (HIPE; Ott, 2010)5 was used to download the images, but photometry was performed using photutils for consistency with the SOFIA images. Apertures span between 45 and 7000 to encompass extended emission around each object. Since the PACS pixels are large on sky, we did not have enough pixels in traditional sky annuli to model the background. Instead, we first mask the star and its nebulosity, and then model the background across the field of view as a 2D polynomial. For each of the PACS fields, these background models were fairly flat but had high RMS variation. As summarized in Table 3.2, this uncertainty was as high as 40% for VX Sgr and S Per. ∼ The width of the PACS bandpasses requires color corrections to be applied to the 70 and 160 µm photometry from the images. In Shenoy et al.(2016), the necessary corrections were estimated by convolving the “blue” (70 µm) filter response functions to the sources’ ISO LWS spectra. However, lacking spectra for all of the sources in this work, we instead fit the β mid- to far-IR photometry from SOFIA and IRAS with a power-law of the form Fν = ν to represent the targets’ SEDs at the PACS wavelengths. The results are modest corrections of 1.003 and 1.04 for the two bandpasses. PACS photometry is reported in Table 3.2 and included in the SEDs in §3.3. Photometric error is reported as measured uncertainty in the sky background models.

5 HIPE is a joint development by the Herschel Science Ground Segment Consortium, consisting of ESA; the NASA Herschel Science Center; and the HIFI, PACS, and SPIRE consortia. Chapter 3. Searching for Cool Dust 47

2000 20" 15" 10" 58°35'05" levels above σ (etc.) above the m σ

51s 1 0 . 7 0 3 0

2 A I F 52s O S 2h22m53s m

51s 8 0 . 4 0 3 0

2 A I F 52s O S 2h22m53s m

51s 5 0 . 1 0 3 0

2 A I F 52s O S 2h22m53s m

51s 3 0 . 5 0 2 0

2 A I F 52s O S 2h22m53s m

models for several of the RSGs in this work. The contours shown here represent diﬀerent 51s 7 0 . 9 0 1 0

/FORCAST imaging of S Per. Much of the circumstellar material in each of the RSGs studied was unresolved 2 A I F DUSTY 52s O S SOFIA 15" 10" 05" 2h22m53s

58°35'20"

0 0 0 2 Figure 3.1 the background r.m.s. noise.background. For each panel, the contours from the outermost towards the center are 1, 3, 5, 7- by SOFIA. However, themade radial by proﬁles diﬀerent shown in Figures 3.4 – 3.15 are still diagnostically useful for distinguishing the predictions Chapter 3. Searching for Cool Dust 48 c 2 . 30 11 m 1 µ ± ± ± 160 c 1 3 . 96 116 32 23 4 m ······ µ ± ± ± 70 1 6 21 b . . 10 153 4 9 m 287 652 ± µ ± ± ± 15 37.1 4 97 b . 75 697 9 m 123 2809 ··· ············ ± µ ± ± 34.8 0 9 b . . 6915 835 109 2 2 m 110 2849 ± ± µ ± ± ± 2 . 21 7 31.5 b 8524 1090 146 m 328 3626 ··· ··· ± ± µ ± 25.3 b New Mid-Infrared Photometry 8 2 . . 41 3930 9128 1400 187 m 4 1 µ ± ± ± ± ± 7 . 82 342 9 19.7 4868 b Table 3.2 98 2250 m ··· ··· ··· µ ± 11.1 3740 1 . a 48 3 160 m ± µ ± ± 4 . 9.8 4 11 a . 63 3780 52 340 1 m ········· ······ ± ± µ ± 8 (Jy) (Jy) (Jy) (Jy) (Jy) (Jy) (Jy) (Jy) (Jy) (Jy) . 3735 d SOFIA/FORCAST NML Cyg MIRAC photometry originally presented in Schuster et al. ( 2009 ). MMT/MIRAC HERSCHEL/PACS Name 8.8 a b c d T Per 8 NML Cyg RS Per S Per 316 VX Sgr Chapter 3. Searching for Cool Dust 49 3.3 Results & Discussion

3.3.1 DUSTY modeling

To estimate the mass-loss rates, mass-loss histories, and dust density distributions, we used the DUSTY radiative-transfer code (Ivezic et al., 1997) to model the observed SEDs and azimuthal average intensity profiles at each of the MIRAC and FORCAST wavelengths, in a manner similar to that used in Shenoy et al.(2016). DUSTY solves the 1D radiative- transfer equation for a spherically-symmetric dust distribution around a central source. We provide as input the chosen optical properties, chemistry, size distribution of the dust grains, and a dust temperature, which fixes the inner boundary of the surrounding dust shell (the dust condensation radius, r1). We generate a grid of models for each star with fixed stellar effective temperatures based on the spectral type of each target, fixed shell

extent (1000 r1), and fixed dust condensation temperature (1000 K). Our grid consists × of varying optical depths of the circumstellar material (0.01 < τV < 50) and different dust density distribution functions, described below. For a given set of inputs, DUSTY outputs a model SED and radial profiles of the dust shell at requested wavelengths. As noted in Shenoy et al.(2016), the spherical symmetry assumed by DUSTY fails to model the azimuthal complexities observed in the asymmetric outflows of massive stars such as VY CMa (Smith et al., 2001; Humphreys et al., 2005, 2007; Shenoy et al., 2013) and IRC +10420 (Humphreys et al., 1997; Tiffany et al., 2010; Shenoy et al., 2015). However, DUSTY allows for a consistent analysis of the dust, SEDs, and intensity profiles of the targets in this work and those in Shenoy et al.(2016).

At a given wavelength, an output optical depth τλ from the model, and thus its grain

opacity κλ, speciﬁes the dust mass density ρ (r) throughout the shell. If we assume a

constant expansion rate vexp of the outﬂowing dust shell, we can estimate the mass loss rate as: 2 M˙ (t) = gd 4πr ρ (r) vexp

where radius r is a probe on timescale t since r = vexp t, and gd is the gas-to-dust ratio. For consistency with Shenoy et al.(2016), we assume gd = 100:1 (Knapp et al., 1993); however, this can be as high as 200:1 for supergiants (Decin et al., 2006; Mauron & Josselin, 2011). For the dust optical properties, we use the “cool” circumstellar silicates from Ossenkopf et al.(1992), and assume the grain radii follow a Mathis, Rumpl, Nordsieck (MRN) size −3.5 distribution n (a) a da (Mathis et al., 1977) with amin = 0.005 µm and amax = ∝ 0.25 µm. In general, the mass density distribution of the outflows can be modeled with DUSTY Chapter 3. Searching for Cool Dust 50 as a power-law ρ (r) r−q. An index of q = 2 is the case of constant mass-loss rate and ∝ constant expansion velocity for the shell, while q < 2 indicates a gradual decline in the mass-loss rate over the dynamical age of the expanding shell. A steeper power-law index q > 2 represents a mass distribution with more recent high mass loss, and less significant mass loss in the past. For each of our targets, the fundamental research question is how well the stars’ SEDs and radial profiles in the mid-IR can be modeled with a constant mass-loss rate scenario in DUSTY. For each of the targets in our sample, we perform three Monte Carlo experiments. In the first, we force DUSTY to use the constant mass-loss rate distribution ρ (r) r−2, and by ∝ varying the optical depth of the CS material, recover the best-fitting, constant mass-loss SED in the near- to mid-infrared. For the second set of simulations, we allow the power-law index of the mass distribution to vary between 1 and 3 with a step size of 0.2, deliberately excluding the r−2 case, while also allowing the optical depth to vary. For both set of DUSTY models, we evaluate the best fit based on a reduced χ2 measurement of the extinction- corrected SED and the DUSTY output spectrum. We then compare the DUSTY-predicted intensity profiles to the observed radial profiles in the SOFIA wavelengths (and MIRAC, when available). The image and profile models output from DUSTY do not account for the optics of the telescopes, so the intensity profiles are convolved with an azimuthal average of the PSF and are displayed in the figures below. For the third and final set of DUSTY models, we select the best-fitting r−2 model, and re-run DUSTY with those same parameters, this time enhancing the dusty density profile by

a factor of ten at 50 condensation radii (50 r1). An example model is shown in Figure 3.2. × These “enhanced,” piecewise-defined models explore the possibility of an extreme mass-loss event in a star’s past, similar in some respects to the models from the second experiment described above where the density distribution can be shallower than r2. The latter DUSTY models imply a smoothly-changing mass-loss rate over the lifetime of the star, whereas the “enhanced” models simulate a single eruptive event in the mass-loss history. Similar piece-wise defined density profiles were used in Shenoy et al.(2016) to model the SED of IRC +10420. Though a factor of ten enhancement in mass-loss rate is likely an extreme case, we apply this model to explore how well the scenario of a constant mass-loss rate with a single one-off eruptive event reproduces the observed IR SED of our target stars. Finally, we can estimate an average mass-loss rate for the non-constant mass-loss models (q = 2) by integrating the density distribution ρ (r) and multiplying by the gas-to-dust mass 6 ratio (100:1) to compute the total mass of the shell M. We assume an average expansion

velocity to estimate the dynamical age of the shell ∆t = r2/vexp where r2 is the outer Chapter 3. Searching for Cool Dust 51

t (yr) 10 100 1000

10 19

10 20 ] 3 m c 10 21 g [ ) r (

10 22

10 23

102 103 104 Distance from star (AU)

Figure 3.2 Example of an “enhanced” DUSTY model, where the dust density ρ (r) follows −2 an r distribution out to 50 r1 (where r1 is the condensation radius corresponding to a ∼ × dust condensation temperature of 1000 K), at which point we simulate a discrete mass-loss event by enhancing the dust density distribution by a factor of 10. The density units are artiﬁcially scaled in this example, but the distance at which the enhancement is placed for all the models is roughly 1000 AU from the central star. If we assume an outﬂow velocity of 25 km/s, this distance corresponds to a high mass-loss event 200 years ago. ∼

radius of the shell predicted by a given model. The expansion velocity, vexp, is assumed to be 25 km s−1 unless specified in the sections for individual stars below. The average mass-loss rate is then M˙ = M/∆t. The specific parameters for the DUSTY models for each h i target in our program, as well as the output DUSTY models and computed average mass-loss rates, are summarized in Table 3.3, and the best-fitting SEDs to the observed photometry are shown in the figures below. Note that in Table 3.3, the first row for each target star represents the best-fitting DUSTY simulations forced to evaluate the models in the constant mass-losing, r−2 dust profile case. The second row represents the best-fitting SED with non-constant mass-loss. The columns on the left reflect the input values, and the right-hand columns are the recovered output Chapter 3. Searching for Cool Dust 52

Table 3.3 DUSTY Model Parameters and Mass-Loss Rates

a b c Model AV Teﬀ Tdust τV r1 M˙ K K AU M˙ /yr

Inputs Outputs VX Sgr r−2.0 6.7 86 4.5 10−5 2.0 3200 1000 × r−1.6 3.7 76 2.2 10−5 × S Per r−2.0 1.4 45 2.6 10−5 3.1 3500 1000 × r−1.6 1.2 43 2.4 10−5 × RS Per r−2.0 0.3 50 3.5 10−5 1.7 3600 1000 × r−1.6 0.3 50 3.9 10−5 × T Per r−2.0 0.1 30 7.5 10−6 2.1 3700 1000 × r−1.6 0.1 31 8.1 10−6 × NML Cyg r−2.0 41 133 4.8 10−4 4.0 3300 1000 × r−1.8 37 128 4.2 10−4 ×

aDUSTY output models are ﬁt to extinction-corrected SEDs with these values of AV . b Dust temperature at condensation radius, r1. cM˙ is computed as an average mass-loss rate over the lifetime of the shell. The outﬂow velocity is assumed to be 25 km/s unless noted in the sections for the individual stars.

parameters from the best-fitting models for each target and each simulation set (constant vs. non-constant mass-loss rates). We do not include the enhanced, piecewise-defined models here, since the parameters were fixed to the r−2 model for each star. Throughout the text we will refer to the three different models as constant (r−2), non-constant (r−q, q = 2), and 6 enhanced (r−2, e) mass-loss rates.

3.3.2 VX Sgr

VX Sgr has a marginally-resolved, nearly symmetric extended circumstellar envelope in its HST visual images (Schuster et al., 2006). Additionally, Vlemmings et al.(2005) has Chapter 3. Searching for Cool Dust 53 identified a dipole magnetic field in its ejecta mapped by its H2O masers, which may be a clue to its mass loss mechanism. VX Sgr is also a semi-regular variable that behaves like a fundamental mode pulsator (i.e. a Mira variable), which is rare for such a luminous star. It has been observed to vary by several magnitudes with corresponding changes in its apparent spectral type from M4 to M10. During one of its Mira-like episodes, Humphreys & Lockwood(1972) noted a decline of 0.5 mag out to 10 µm over a few months. Therefore, we have chosen optical photometry ∼ to align in light-curve phase with the 2MASS and IRAC photometry compiled in Smith et al.(2004). To constrain the 2–10 µm regime of the SED, we also include the photometric average over the 3.6-yr cycle of the COBE DIRBE project (Price et al., 2010). The SED is shown in Figure 3.3, with observed data plotted as open symbols and extinction-corrected photometry in solid. The constant mass-loss DUSTY model is overplotted with a dashed blue line, and the best-fitting power-law model with index q = 1.6 is shown in dashed-dotted green. Note that both models fit the 10 µm silicate feature in the ISO spectrum, but the q = 1.6 model better simulates the cool thermal dust emission out to 100 µm. However, the observed IR flux from 30 70 µm falls in between the two ∼ − models. Displayed in dotted red is the “enhanced” DUSTY model, which is the r−2.0 constant mass-loss model with a factor of 10 enhancement in dust density at 50 r1. This model × appears to over-estimate the thermal dust emission, implying too much dust is produced to match the observations of VX Sgr. The derived mass-loss rates for the models are summarized in Table 3.3. The outflow velocity vexp adopted for this calculation is 24.3 km/s from the AGB/supergiant CO-line −5 survey by De Beck et al.(2010). The mass-loss rates from DUSTY (2 5 10 M /yr) − × −5 are somewhat lower than the measurements from CO-line profiles (6.1 10 M /yr; De × Beck et al., 2010). However, the most obvious explanation for this is due to the assumed gas-to-dust ratio. Where we have assumed 100:1 for consistency with Shenoy et al.(2016), De Beck et al.(2010) allow the gas-to-dust ratio to vary when fitting the observed outflow velocities (using GASTRoNOoM; Decin et al., 2006). Mass-loss rates scale linearly with the gas-to-dust ratio, so if we had applied a ratio of 200:1, perhaps more appropriate for RSGs (Decin et al., 2006; Mauron & Josselin, 2011), our estimated mass-loss rate would be more consistent with the derived measurement from CO-line profiles. In Figure 3.4, we compare the observed radial profiles to the PSF calibrator (2 Pallas) and the DUSTY output image models. The ejecta around VX Sgr is only marginally resolved above the PSF at 19.7 µm, but the envelope is more easily distinguished from the PSF at longer wavelengths. The DUSTY model profiles are convolved with the PSF in each band, and Chapter 3. Searching for Cool Dust 54

10 8 Anderson 2012 Smith 2004 Price 2010 10 9 ISO This Work

10 10 ) 2 11 m 10 / W ( F

10 12

10 13

r 2.0 r 1.6 10 14 r 2.0, e

0.5 1 2 5 10 20 50 100 ( m)

Figure 3.3 Optical and IR SED of VX Sgr. The solid points represent the extinction corrected photometry (AV = 2.0), and open points are observed photometry. Triangles are optical data from the Extended Hipparchos Compilation catalog (XHIP; Anderson & Francis, 2012). Squares are 2MASS and IRAS photometry, compiled in the COBE DIRBE Point Source Catalog (Smith et al., 2004). Diamonds are the 3.6-yr average of the DIRBE photometry in Price et al.(2010). Circles are PSF photometry from this work on SOFIA/FORCAST and Herschel/PACS images. The dashed blue line is the DUSTY model with fixed constant mass-loss, and the dashed-dotted green line is the best-fitting DUSTY model with a non-constant mass-loss dust density profile and power-law index q = 1.6. The dotted red line represents the “enhanced” DUSTY models with a bump up in mass-loss rate 200 years ago. An example density profile for this last model is shown in Figure 3.2. ∼ we note that the constant mass-loss rate model aligns more closely with the observed surface brightness profiles of VX Sgr. Both the shallower r−1.6 and enhanced models over-estimate the amount of thermal dust emission observed in the FORCAST images. The azimuthal profiles combined with the SED modeling suggest that the mass-loss rate of VX Sgr is fairly constant with perhaps a period of elevated mass-loss in the past, as illustrated by the infrared excess emission in the 20 160 µm photometry. The model intensity − profiles, though, all predict a higher surface brightness for the extended emission than what Chapter 3. Searching for Cool Dust 55

R (AU) R (AU) 0 200 400 600 800 1000 0 200 400 600 800 1000

1.0 19.7 m 25.3 m PSF PSF 2.0 2.0 0.8 (r) r (r) r (r) r 1.6 (r) r 1.6 0.6 (r) r 2.0, e (r) r 2.0, e

0.4

Normalized Flux 0.2

0.0

1.0 31.5 m 37.1 m PSF PSF 2.0 2.0 0.8 (r) r (r) r (r) r 1.6 (r) r 1.6 0.6 (r) r 2.0, e (r) r 2.0, e

0.4

Normalized Flux 0.2

0.0

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.00.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 (arcsec) (arcsec) Figure 3.4 Radial profiles of VX Sgr at four SOFIA wavelengths (black dots). The dotted gray line shows the PSF measured at each wavelength from images of the asteroid Pallas. The dashed blue line represents the modeled profile from DUSTY assuming a constant mass- loss rate, which implies a dust density distribution of ρ (r) r−2. The dashed-dotted green ∝ line is the best-fitting model to the SED from DUSTY, excluding the non-constant case, and the dotted red line is the “enhanced” DUSTY model. All models are convolved with the Pallas PSF. We note that the circumstellar ejecta around VX Sgr is only marginally resolved above the FORCAST PSF at 19.7 µm, and all three models over-estimate the amount of dust emission expected at the FORCAST wavelengths.

was actually observed at SOFIA wavelengths (Figure 3.4). One possible explanation for not observing this emission is that VX Sgr has experienced a sudden decline in mass-loss rate in very recent times. Exploring this possibility is beyond the scope of this paper, but we plan to make high-resolution 5 12 µm observations using LMIRCam and NOMIC on the − LBT (Skrutskie et al., 2010; Hoﬀmann et al., 2014). At FWHM spatial resolutions of 000. 12 and 000. 29 at 5 and 12 µm, respectively, we can explore the dust shell at 200 AU scales ∼ and combine these observations with our SOFIA data and DUSTY modeling. We also note that although the envelope is spherically-symmetric in HST optical images

(Schuster et al., 2006), the H2O masers around VX Sgr appear to align with the equatorial Chapter 3. Searching for Cool Dust 56 plane of the star’s dipole magnetic field. Vlemmings et al.(2005) suggest that this alignment could create an overdensity in the circumstellar material in this plane, as modeled by Matt et al.(2000). While we do not see evidence for asymmetry in the FORCAST images, we note again that DUSTY assumes spherical symmetry in its models. As discussed further with S Per and NML Cyg below, it is likely that DUSTY may fail to accurately model stars with known asymmetric outflows and profiles.

3.3.3 S Per

S Per (Sp. Type M3-4e Ia) is an OH/IR source and a member of the Per OB1 association

(Humphreys, 1978), with a distance of 2.3 0.1 kpc as determined by VLBI H2O maser ± astrometry (Asaki et al., 2010). Schuster et al.(2006) present HST images showing that the star is embedded in an elongated circumstellar envelope with a position angle of 20◦ ∼ E of N with a FWHM of 000. 1 (240 AU). Schuster et al.(2006) speculated that the shape ∼ could be due to bipolarity in the star’s ejecta or a ﬂattened circumstellar halo, and they

note this elongated structure is also seen in OH and H2O maser observations on the same scale and with similar orientation (Richards et al., 1999; Vlemmings et al., 2001). Fitting elliptical Gaussians to S Per’s MIRAC4 images yields a mean position angle of 19◦ 2◦ E ± of N, matching closely the orientation seen in the HST images. The observed SED is shown in Figure 3.5 along with the three DUSTY models. Both the shallower q = 1.6 and enhanced DUSTY models accurately reconstruct the near- to mid-IR flux, while the constant mass-loss rate model underestimates the thermal dust emission. We note here one possible complication in our analysis. DUSTY simulations assume spherical symmetry in CS material, which could lead to underestimating the density, and thus optical depth, of the ejecta relative to the observed compact envelope seen in the Schuster et al. (2006) WFPC2 images of S Per. Our models for this star, then, may not best represent the stellar outflows and dusty envelope. The observed azimuthal average radial profiles from SOFIA/FORCAST for S Per are presented in Figure 3.6. S Per has resolvable extended emission above the PSF; however, the q = 2 and q = 1.6 profiles, once convolved with the large PSF beam of FORCAST, are virtually indistinguishable. The enhanced DUSTY model, though, predicts too much emission close to the central star. In Figure 3.7, we illustrate the surface brightness profiles at higher spatial resolution with MIRAC. Here, the observed surface brightness profile is clearly resolved above the PSF at the shorter wavelengths; however, the DUSTY models underestimate the shape of the stellar envelope. Adding a period of enhanced mass loss, the DUSTY model in dotted Chapter 3. Searching for Cool Dust 57

10 9 Anderson 2012 Abrahamyan 2015 ISO This Work 10 10

10 11 ) 2 m / W (

F 10 12

10 13

r 2.0 1.6 10 14 r r 2.0, e

0.5 1 2 5 10 20 50 100 ( m)

Figure 3.5 Optical and IR SED of S Per. The solid points represent the extinction corrected photometry (AV = 3.1), and open points are observed photometry. Triangles are optical data from the Extended Hipparchos Compilation catalog (XHIP; Anderson & Francis, 2012). Squares are from the 2MASS, WISE, IRAS, and AKARI point-source catalogs, compiled in Abrahamyan et al.(2015). Circles are PSF photometry from this work on SOFIA/FORCAST and Herschel/PACS images. The dashed-dotted green line is the best-ﬁtting model to the SED from DUSTY, excluding the non-constant case, and the red dotted line is the “enhanced” model.

red, produces too much emission at the shorter wavelengths. Unfortunately, then, the radial profile models do not provide any conclusive evidence that the mass-loss history of S Per is constant vs. non-constant. Note that the deviations from a smooth profile in the 31.5 µm and the two MIRAC figures are due to the asymmetry in the outflows. From the SED, though, we glean that S Per may have had a higher mass-loss rate in the past, but we acknowledge that DUSTY is not ideal for simulating stellar ejecta of stars with known bipolar/asymmetric envelopes. As reported in Table 3.3, the two DUSTY models predict mass-loss rates between ∼ −5 2 3 10 M /yr. Richards et al.(1999) summarizes results from previous literature to − × Chapter 3. Searching for Cool Dust 58

R (AU) R (AU) 0 2000 4000 6000 8000 0 2000 4000 6000 8000

1.0 19.7 m 25.3 m PSF PSF 2.0 2.0 0.8 (r) r (r) r (r) r 1.6 (r) r 1.6 0.6 (r) r 2.0, e (r) r 2.0, e

0.4

Normalized Flux 0.2

0.0

1.0 31.5 m 37.1 m PSF PSF 2.0 2.0 0.8 (r) r (r) r (r) r 1.6 (r) r 1.6 0.6 (r) r 2.0, e (r) r 2.0, e

0.4

Normalized Flux 0.2

0.0

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.00.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 (arcsec) (arcsec) Figure 3.6 Radial profiles of S Per at the SOFIA/FORCAST wavelengths, similar to Figure 3.4. S Per has resolvable extended envelope emission above the PSF flux; however, the predicted profiles from DUSTY are too similar in shape to distinguish one over the other as a best-fitting model to the observed surface brightness profile. Still, the enhanced DUSTY model perhaps over-predicts the amount of thermal emission that would be observed close in to the central star.

−6 show a range of published mass-loss rates from as low as 7 10 M /yr (OH 1612 MHz; × −4 Jura & Kleinmann, 1990) to as high as 2 10 M /yr (CO-line profiles; Knapp & Morris, × 1985). With such a large range of published values, each measuring mass-loss rates with a different observational technique, we can only conclude that we have derived a rate within published bounds. Fok et al.(2012) also performed DUSTY modeling on a number of Galactic RSGs, including S Per, T Per, and RS Per. However, they used a different mode, the “dusty AGB” radiatively-driven wind mode, and a higher gas-to-dust ratio of 200:1. The radiatively- driven wind mode in DUSTY is provided for modeling AGB star envelopes and is not necessarily appropriate for RSGs (e.g., Heras & Hony, 2005). Groenewegen(2012) analyzed the systematic difference in mass-loss rates computed using DUSTY in this mode as compared to Chapter 3. Searching for Cool Dust 59

R (AU) R (AU) 0 2000 4000 6000 8000 0 2000 4000 6000 8000

1.0 8.9 m 9.8 m PSF PSF 2.0 2.0 0.8 (r) r (r) r (r) r 1.6 (r) r 1.6 0.6 (r) r 2.0, e (r) r 2.0, e

0.4

Normalized Flux 0.2

0.0

0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 (arcsec) (arcsec) Figure 3.7 Radial profiles of S Per from the two MIRAC filters. The observed profile shape has bumps and ridges due to the asymmetry of the envelope in the NE–SW direction. The PSF shown is of β And. the default where the user supplies the density distribution as a power-law function. Groe- newegen(2012) found that the mass-loss rates computed with the radiatively-driven wind mode differ significantly from those obtained from the default mode in which the equation of radiative transfer alone is solved when applied in the context of RSGs. Thus, our results, with fixed single-component power-law distributions, are not directly comparable to the Fok et al.(2012) models. Nonetheless, Fok et al.(2012) yields a best-fitting model with −5 M˙ = 1 10 M /yr. We note that Gehrz & Woolf(1971) derived a mass-loss rate for × −5 S Per of 2.7 10 M /yr using an independent analysis of the 3.6–11.4 µm SED. These × results are consistent with our measurements.

3.3.4 RS Per and T Per

Departures from circular symmetry have been reported for both RS Per and T Per based on H-band interferometric imaging with CHARA by Baron et al.(2014) at an angular scale of 1.3 mas, which the authors attribute to surface asymmetries or spots. We do not see much evidence for asymmetry in the FORCAST images, and the azimuthal proﬁles in Figure 3.9 do not show signiﬁcant excess emission above the PSF, though the angular scales for FORCAST are much larger than Baron et al.(2014) observed with the CHARA array. RS Per has the 10 µm silicate emission feature in its SED but is not a known maser source. It is likely a normal red supergiant that may just be entering a more active phase with enhanced mass loss, perhaps driven by surface activity like that seen in the OH/IR supergiants and VY CMa. T Per, another member of the Perseus OB1 association, similarly Chapter 3. Searching for Cool Dust 60 shows no evidence for SiO maser emission (Jiang et al., 1999). Both stars exhibit long-period variability of 4200 and 2500 days for RS Per and T Per, respectively (AAVSO; Kiss et al., ∼ 2006).

10 8 Myers 2015 Abrahamyan 2015 10 9 ISO This Work

10 10

10 11 ) 2 m / 12

W 10 ( F

10 13

10 14

r 2.0 10 15 r 1.6 r 2.0, e 10 16 0.5 1 2 5 10 20 50 100 ( m)

Figure 3.8 Optical and IR SED of RS Per. The solid points represent the extinction corrected photometry (AV = 1.7), and open points are observed photometry. Triangles are optical data from the SKY2000 Master Catalog (Myers et al., 2015). Squares are from the 2MASS, WISE, IRAS, and AKARI point-source catalogs, compiled in Abrahamyan et al. (2015). Circles are PSF photometry from this work on SOFIA/FORCAST images.

The SED of RS Per is shown in Figure 3.8. Both models fit the 10 µm silicate feature, as well as the ISO spectrum, from 2 11 µm. However, the constant mass-loss, q = 2 profile − underestimates the flux for the mid- to far-IR at wavelengths larger than 20 µm, while the shallower dust density distribution q = 1.6 profile and the enhanced r−2.0 profile better match the longer wavelength SED. For T Per, shown in Figure 3.10, both the constant mass-loss and enhanced models produce insufficient thermal dust emission at the longer wavelengths. As summarized in Table 3.3, the mass-loss rates derived for RS Per and T Per are Chapter 3. Searching for Cool Dust 61

R (AU) R (AU) 0 2000 4000 6000 8000 0 2000 4000 6000 8000

1.0 19.7 m 31.5 m PSF PSF 2.0 2.0 0.8 (r) r (r) r (r) r 1.6 (r) r 1.6 0.6 (r) r 2.0, e (r) r 2.0, e

0.4

Normalized Flux 0.2

0.0

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 1.0 37.1 m (arcsec) PSF 2.0 0.8 (r) r (r) r 1.6 0.6 (r) r 2.0, e

0.4

Normalized Flux 0.2

0.0

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 (arcsec) Figure 3.9 Radial profiles of RS Per at the SOFIA/FORCAST wavelengths, similar to Figure 3.4. The figures reveal extended emission above the PSF at 2 400 from the central − star. The shallower mass-loss model (ρ (r) r−1.6), though a better fit to the photometry ∝ in Figure 3.8, appears too similar to the constant mass-loss model after convolution with the PSF to conclude which mass-loss history is a better fit.

−5 −6 4 10 and 8 10 M /yr, respectively. Fok et al.(2012) estimates a mass-loss rate of × × −6 −7 3.0 10 M /yr for RS Per and 5 10 M /yr for T Per using DUSTY “AGB mode.” As × × discussed in §3.3.3, this radiatively-driven wind mode is less appropriate for RSGs, and so our results are not directly comparable. Additionally, we note that the SED fits provided in their work do not extend longward of 30 µm, so we cannot qualitatively gauge which SED modeling mode (our power-law profiles vs. their radiatively-driven wind models) would fit best with our new FORCAST photometry through 40 µm. Particularly in the case of T Per, we note that this long-wavelength IR photometry is crucial in constraining the models. While the r−1.6 DUSTY model is clearly the better fit to the observed mid-IR SED, the radial profiles of RS Per in Figure 3.9 reveal that the SOFIA images lack the spatial resolution necessary to favor one model over the other as a better fit to the extended envelope emission. The radial profiles of T Per in Figures 3.11 and 3.12 reveal a much clearer extended profile above the PSF at 31.5 µm and around the 10 µm silicate feature Chapter 3. Searching for Cool Dust 62 with MIRAC. However, the two power-law models predict very similar output profiles. The enhanced model accurately recovers the extended emission around the 10 µm silicate feature, but the MIRAC images in the two other wavebands do not resolve emission extended above the PSF.

10 8 Henden 2016 Abrahamyan 2015 This Work 10 9

10 10 ) 2 11 m 10 / W ( F

10 12

10 13

r 2.0 r 1.6 10 14 r 2.0, e

0.5 1 2 5 10 20 50 100 ( m)

Figure 3.10 Optical and IR SED of T Per. The solid points represent the extinction corrected photometry (AV = 2.1), and open points are observed photometry. Triangles are optical data from the AAVSO Photometric All Sky Survey (APASS; Henden, 2016). Squares are from the 2MASS, WISE, IRAS, and AKARI point-source catalogs, compiled in Abrahamyan et al.(2015). Diamonds are MSX photometry in Egan et al.(2003). Circles are PSF photometry from this work on SOFIA/FORCAST and MMT/MIRAC images.

We note a curious ripple in the 31.5 µm FORCAST profile in Figure 3.11. As mentioned in §3.2.1, some optical aberrations may be introduced to the images due to chopping patterns of the secondary mirror on SOFIA. One possible source of the profile shape could be coma effects that stretch out T Per along one axis. When generating azimuthally-averaged surface brightness profiles, this asymmetry would cause the profile to deviate from a smooth power- law. Chapter 3. Searching for Cool Dust 63

R (AU) R (AU) 0 2000 4000 6000 8000 0 2000 4000 6000 8000

1.0 19.7 m 31.5 m PSF PSF 2.0 2.0 0.8 (r) r (r) r (r) r 1.6 (r) r 1.6 2.0 2.0 0.6 (r) r , e (r) r , e

0.4 Normalized Flux 0.2

0.0

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 (arcsec) (arcsec)

Figure 3.11 Radial profiles of T Per at the SOFIA/FORCAST wavelengths, similar to Figure 3.4. T Per has obvious extended emission above the PSF at 31.5 µm, though neither model reproduces this observed profile. The shape of the observed radial profile suggests a geometry more complicated than a single power-law distribution can generate. T Per may have undergone multiple eruptive mass-loss events, the outflows are asymmetric, or we are seeing coma effects in our images (see discussion in §3.3.4).

R (AU) R (AU) R (AU) 0 2000 4000 6000 8000 0 2000 4000 6000 8000 0 2000 4000 6000 8000

1.0 8.9 m 9.8 m 11.9 m PSF PSF PSF 2.0 2.0 2.0 0.8 (r) r (r) r (r) r (r) r 1.6 (r) r 1.6 (r) r 1.6 2.0 2.0 2.0 0.6 (r) r , e (r) r , e (r) r , e

0.4

Normalized Flux 0.2

0.0

0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 (arcsec) (arcsec) (arcsec) Figure 3.12 Radial profiles of T Per from the three MIRAC filters. T Per has clearly resolved extended emission above the PSF (β And) at λ > 9.8 µm. Although the shallower r−1.6 model is a better fit at the longer SOFIA wavelengths (Figure 3.11), the enhanced mass-loss rate DUSTY model reproduces the shape of the observed profile at 10 µm in the ∼ MIRAC images.

3.3.5 NML Cyg

HST visual images of NML Cyg revealed a peculiar bean-shaped asymmetric nebula only 00 0. 2 across and coincident with the distribution of its H2O masers. Schuster et al.(2006) ≈ showed that its circumstellar envelope is shaped by photodissociation from the powerful nearby association Cyg OB2 inside the Cygnus X superbubble which is relatively void of gas and dust. This conﬁguration allows the UV radiation from the numerous luminous hot Chapter 3. Searching for Cool Dust 64 stars in Cyg OB2 to travel the 80 pc to NML Cyg unimpeded. Subsequent adaptive ≈ optics mid-IR imaging at 8.8, 9.8 and 11.7 µm with MIRAC3 on the MMT (Schuster et al., 2009) spatially resolve the physical structures near the star ( 240 AU) responsible for its ∼ 10 µm silicate-absorption feature and an asymmetric excess at 000. 3 000. 5 from the star − due to thermal emission from hot dust. This excess is also oriented toward the Cyg OB2 association and is attributed to the destruction of NML Cyg’s dusty wind by the hot stars in Cyg OB2.

10 8 Schuster 2007 2MASS/WISE 10 9 ISO This Work

10 10

10 11 ) 2 m / 12

W 10 ( F

10 13

10 14

r 2.0 10 15 r 1.8 r 2.0, e 10 16 0.5 1 2 5 10 20 50 100 ( m)

Figure 3.13 Optical and IR SED of NML Cyg. The solid points represent the extinction corrected photometry (AV = 4.0; see Schuster, 2007), and open points are observed photometry. Triangles are from observations in Schuster(2007) using HST/WFPC2 at V , Hα, and R; MMT/MIRAC3 at 9 12 µm; the OBO Bolometer at 1 12 µm; and Spitzer/IRAC and − − IRAS photometry. Squares are from the 2MASS and WISE point-source catalogs. Circles are PSF photometry from this work on SOFIA/FORCAST and Herschel/PACS images.

As illustrated in the SED in Figure 3.13, the 10 µm silicate feature is seen in absorption, rather than emission. Discussed in detail in Schuster et al.(2006, 2009), this absorption is due to NML Cyg’s thick circumstellar envelope obscuring the central star. Indeed, the

DUSTY models predict a large optical depth (τV > 40 for both models) to fit the silicate Chapter 3. Searching for Cool Dust 65 feature in absorption as well as the mid- to far-IR photometry. At the FORCAST wavelengths, 19.7 to 37.1 µm, NML Cyg appears as a point source, with no evidence of cold dust hidden or protected from the UV radiation in Cyg OB2. Additionally, no preferential extension towards Cyg OB2 is evident at the angular resolution of FORCAST. The radial profiles in Figure 3.14 do not show any obvious excess emission above the PSF. While we note that the shape of the observed azimuthal profile seems similar to the r−2 model, the models seem to greatly over-predict the surface brightness flux for the large estimated optical depth.

R (AU) R (AU) 0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000

1.0 19.7 m 25.3 m PSF PSF 2.0 2.0 0.8 (r) r (r) r (r) r 1.8 (r) r 1.8 0.6 (r) r 2.0, e (r) r 2.0, e

0.4

Normalized Flux 0.2

0.0

1.0 31.5 m 37.1 m PSF PSF 2.0 2.0 0.8 (r) r (r) r (r) r 1.8 (r) r 1.8 0.6 (r) r 2.0, e (r) r 2.0, e

0.4

Normalized Flux 0.2

0.0

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.00.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 (arcsec) (arcsec) Figure 3.14 Similar to Figure 3.4 but for observations of NML Cyg. For each of the observed wavebands, we do not resolve any obvious excess emission above the PSF.

As noted in Schuster et al.(2009), the MIRAC images do indeed appear asymmetric along the NW–SE axis. For these images, we generate isophotes and separate our radial profiles into two axes. The major axis (NW–SE) is shown with solid points in Figure 3.15, and the minor axis (NE–SW) is shown with open circles. Here, we see that the constant mass-loss, q = 2, model does fit the observed major-axis brightness profile for the 9.8 µm image, though the q = 1.8 profile is similar in shape for all three MIRAC bands. We note that DUSTY cannot take into account the external radiation field from Cyg OB2, so the Chapter 3. Searching for Cool Dust 66

R (AU) R (AU) R (AU) 0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000

8.9 m 9.8 m 11.9 m 1.0 8.9 m off-axis 9.8 m off-axis 11.9 m off-axis PSF PSF PSF 0.8 (r) r 2.0 (r) r 2.0 (r) r 2.0 (r) r 1.8 (r) r 1.8 (r) r 1.8 0.6

0.4 Normalized Flux 0.2

0.0

0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 (arcsec) (arcsec) (arcsec) Figure 3.15 Radial profiles of NML Cyg from the three MIRAC filters, reproduced from Schuster et al.(2009). Due to the observed asymmetry from interactions with Cyg OB2, we have divided the surface brightness measurements into two different axes. Isophotal analysis confirms the elongation along the major (NE–SW) axis, shown with solid points. The minor axis is shown with open circles. model profile shapes are not necessarily conclusive as to the mass-loss history of NML Cyg. NML Cyg has one of the highest mass-loss rates of any red supergiant/hypergiant— −5 −4 from 6.4 10 M /yr (Morris & Jura, 1983) to as high as 1.6 10 M /yr (Lucas et al., × × 1992). We calculate even higher mass-loss rates from the DUSTY models, though, at 4 5 − × −4 −1 10 M /yr with average outflow velocity 23 km s (Schuster et al., 2009). It is likely, however, that the complex morphology of NML Cyg cannot be well-modeled with a single- component power-law dust mass distribution, similar to the difficulties in modeling S Per with its known asymmetric profile.

3.4 Conclusions

With mid-infrared imaging from MMT/MIRAC and SOFIA/FORCAST, we observed three OH/IR red supergiants, NML Cyg, VX Sgr, and S Per and the normal red supergiants RS Per and T Per. We present new photometry at 9–11 µm with MIRAC, at 20–40 µm with FORCAST, and at 70 and 160 µm from the Herschel/PACS archive. These data, in combination with published optical and near- to mid-IR photometry, are used to constrain DUSTY model SEDs. VX Sgr: Though a symmetric extended circumstellar envelope is resolved in HST images (Schuster et al., 2006), we have only marginally resolved a cooler dust component at 19.7– 37.1 µm with FORCAST. From DUSTY, we conclude that the mass loss around VX Sgr cannot necessarily be well modeled by smooth, constant mass loss. The IR excess emission combined with DUSTY modeling show evidence for a higher mass-loss rate in the past for −5 −1.6 VX Sgr with an average estimated mass-loss rate of 2 10 M /yr for the ρ (r) r × ∝ Chapter 3. Searching for Cool Dust 67

−5 model, and 5 10 M /yr for the constant mass-loss case. × S Per: Azimuthal profiles of S Per reveal an IR excess above the PSF emission from the central star in MIRAC imaging, and to a lesser extent in the FORCAST wavelengths. However, the radial profiles produced from DUSTY are not significantly different when convolved with the optics of SOFIA. Both the r−1.6 dust density distribution model and the enhanced mass-loss model fit the near- to mid-IR SED out to 160 µm, implying the possibility of a higher mass-loss rate in the past. S Per is known to posses an asymmetric outflow close to the central star and lack an extended spherical nebula; therefore, the 1D spherically-symmetric radiative-transfer code DUSTY may not be the most accurate method for reconstructing the mass-loss history of S Per. We estimate an average mass-loss rate of −5 2 3 10 M /yr. ∼ − × RS Per and T Per: SED models of both stars suggest that a constant mass-loss rate is insufficient to generate enough 20–40 µm emission to match the observed mid-IR photometry. The FORCAST images for both RSGs, and the MIRAC 10 µm image for T Per, show modest excess emission above the flux from the PSF. The SED for RS Per appears to be best fit with a shallow power-law distribution in dust density of ρ (r) r−1.6, suggesting ∝ it had a higher mass-loss rate in the past. Over the lifetimes of the observed dust shells, −5 −6 we estimate average mass-loss rates of 4 10 M /yr for RS Per and 8 10 M /yr for × × T Per. NML Cyg: We do not observe any circumstellar envelope around NML Cyg at 31.5 and 37.1 µm. Though the DUSTY constant mass-loss models appear to fit the near- to far-IR SED accurately, we cannot conclude from these data alone whether the mass loss around NML Cyg is smooth and constant, or decreasing over time. Additionally, at the resolution of FORCAST at 20–40 µm, we do not observe any evidence for an optically-thick, cool dust −4 shell. The DUSTY models provide an estimate for a mass-loss rate of 4 5 10 M /yr. ∼ − × Finally, as described in Schuster et al.(2006, 2009), there is an external heat source affecting the temperature structure of the circumstellar envelope surrounding NML Cyg. Since DUSTY can only model sources with central internal heating, discrepancies in the mass-loss rates from different measurements probing various parts of the envelope are expected. In Figure 3.16, we summarize the results from this work and Shenoy et al.(2016). We plot the estimated mass-loss rates as a function of luminosity with three mass-loss rate prescriptions from Mauron & Josselin(2011). Reimers(1975) and Kudritzki & Reimers(1978) measured mass-loss rates for O-rich dust-enshrouded RSGs in the LMC and fit an empir- ical relation to luminosity. Reimers’ law is largely consistent with the later formulations, NJ90 (Nieuwenhuijzen & de Jager, 1990) and observations by van Loon et al.(2005) on Chapter 3. Searching for Cool Dust 68

2.5 − Van Loon NJ90 3.0 − Reimers VY CMa

3.5 NML Cyg − IRC +10420

yr] 4.0 / − RS Per VX Sgr M S Per

˙ 4.5 M[ − log

5.0 T Per −

5.5 − µ Cep

6.0 − 4.6 4.8 5.0 5.2 5.4 5.6 5.8 log Lbol [L ]

Figure 3.16 Average mass-loss rates as a function of luminosity. Stars with open circles are the sources discussed in Shenoy et al.(2016). Luminosity values (and luminosity error bars) are drawn from the literature (see §3.4). Overplotted are mass-loss prescriptions for dusty RSGs from van Loon et al.(2005), Nieuwenhuijzen & de Jager(1990) (NJ90), and Reimers Law (Reimers, 1975; Kudritzki & Reimers, 1978). Each of the models displayed are calculated with ﬁxed stellar eﬀective temperature of 3750 K for consistency.

5 dusty RSGs in the LMC at lower luminosities L 2 10 L . Mauron & Josselin(2011) . × apply these mass-loss rate prescriptions to a number of Galactic RSGs. In Figure 3.16, we reproduce their implemented formulae at a fixed stellar effective temperature of 3750 K for consistency with their figures. We adopt their luminosities for all of the sources except

IRC +10420 (log L/L 5.7 5.8, De Beck et al., 2010; Tiffany et al., 2010). The error ≈ − bars shown in the y-axis (mas-loss rate) are the standard deviation of the derived mass-loss rates from the five best-fitting DUSTY models, and the error bars in the x-axis (luminosity) are from the literature. We note that the mass-loss rates for the stars in our sample are largely consistent with the Van Loon and NJ90 prescriptions (analytical forms given in Mauron & Josselin, 2011). As noted in Shenoy et al.(2016), µ Cep has a curiously low mass-loss rate for its luminosity Chapter 3. Searching for Cool Dust 69 class, and the hypergiants VY CMa and IRC +10420 are well-known, extremely luminous hypergiants whose mass-loss rates are among the highest observed. DUSTY modeling, in combination with imaging from SOFIA/FORCAST and MIRAC, are powerful tools for exploring the mass-loss histories and dust density profiles of luminous supergiants. Although the spatial resolution of FORCAST yielded a PSF too wide for us to compare DUSTY output radial profiles, the photometry from 20–40 µm represents a crucial dataset in constraining the thermal dust properties of RSGs. The models shown here and in Shenoy et al.(2016) hint at possible variable mass-loss rates among the most luminous red supergiants while others have constant mass-loss histories over the past few thousand years. We plan to follow up observations of the supergiants discussed here with high-resolution imaging at 5 12 µm with LMIRCam and NOMIC on the LBT. Surface − brightness profiles with better spatial resolution at these shorter wavelengths, when coupled with SED modeling, will allow us to characterize mass-loss events from the last few hundred years.

We thank Rubab Khan for discussion on χ2-minimization of DUSTY models. This work has used unpublished data from Michael Schuster’s PhD thesis, which is available through the SAO/NASA Astrophysics Data System (ADS) at http://adsabs.harvard.edu/abs/ 2007PhDT...... 28S. Financial support for this work was provided by NASA through awards SOF 03 0082 and SOF 04 0013 to R. M. Humphreys issued by USRA. RDG acknowledges support from NASA and the United States Air Force. Chapter 4

Thermal Emission in the Southwest Clump of VY CMa

Adapted from Gordon et al.(2018b, in review)

Abstract We present high spatial resolution LBTI/NOMIC 9–12 µm images of VY CMa and its massive outﬂow feature, the Southwest (SW) Clump. Combined with high-resolution imaging from HST (0.4–1 µm) and LBT/LMIRCam (1–5µm), we isolate the spectral energy distribution (SED) of the clump from the star itself. Using radiative-transfer code DUSTY, we model both the scattered light from VY CMa and the thermal emission from the gas and dust in the clump to estimate the optical depth, mass, and temperature of the SW Clump. The SW Clump is optically thick at 8.9 µm with a brightness temperature of 200 K. ∼ With a dust chemistry of equal parts silicates and metallic iron, as well as −5 assumptions on grain size distribution, we estimate a dust mass of 5.4 10 M . × Compared to the typical mass-loss rate of VY CMa, the SW Clump represents an extreme, localized mass-loss event from . 300 years ago.

4.1 Introduction

The extreme red supergiant VY Canis Majoris is one of the brightest infrared sources in the sky. HST imaging and long-slit spectroscopy from 0.4 to 1 µm reveal a complex circumstellar nebula environment with multiple arcs and knots (Smith et al., 2001; Humphreys et al., 2005, 2007). Ejected in separate mass-loss events over the past 1000 years, these features are ∼ structurally and kinematically distinct from the surrounding nebulosity.

70 Chapter 4. VY CMa and the SW Clump 71

Shenoy et al.(2013) extended the exploration of VY CMa’s ejecta into the near- to mid- infrared with higher spatial resolution than previous studies with ground-based 1–5 µm, adaptive optics imaging using LMIRCam (Skrutskie et al., 2010) on the Large Binocular 0 Telescope (LBT). The dominant IR source in the 2.2, 3.8, and 4.8 µm (Ks,L , and M band) images is the peculiar “Southwest Clump,” which is optically thick in the HST/WFPC2 images at 1 µm (Smith et al., 2001). Shenoy et al.(2013) determined that the high surface brightness of the SW Clump requires optically-thick scattering out to 5 µm rather than thermal emission from dust grains since the expected blackbody equilibrium temperature for material 100 from the central star is quite low ( 170 K). ∼ . Scattering as the dominant component of the SW Clump has been conﬁrmed using high- resolution imaging polarimetry in the near-IR. Using MMT-Pol (Packham et al., 2012) on the MMT, Shenoy et al.(2015) observed 30% fractional polarization in the clump at ∼ 3.1 µm, which requires optically-thick scattering from low albedo dust grains. In earlier work, Shenoy et al.(2013) estimates a lower-limit on the total mass within the clump of −3 5 10 M , though this value assumes a gas-to-dust ratio of 100:1. If this ratio is as × high as 500:1, suggested by Decin et al.(2006), then the lower limit would increase to −2 2.5 10 M . In any case, this ejecta event can be contrasted with VY CMa’s “normal” × −4 −1 mass-loss rate of 10 M yr (Danchi et al., 1994; Humphreys et al., 2005; Decin et al., ∼ 2006), suggesting that the SW Clump represents a single mass-loss episode from a localized region of VY CMa’s stellar atmosphere. Recent sub-millimeter observations with ALMA reveal dusty concentrations within 10 ∼ R? of VY CMa (Richards et al., 2014; O’Gorman et al., 2015; Vlemmings et al., 2017), adopting the Wittkowski et al.(2012) measurement R? = 1420 R . O’Gorman et al.(2015) found a cold clump to the southeast, “Clump C,” located closer to VY CMa than the

SW Clump—400 AU (61 R?) vs. 1500 AU (230 R?). While O’Gorman et al.(2015) estimate −4 a dust mass lower limit of 2.5 10 M for Clump C, similar to the SW Clump, there is × no evidence for the SW Clump in the ALMA images at 321 and 658 GHz (Bands 7 and 9; O’Gorman et al., 2015) or at 178 GHz (Band 5; Vlemmings et al., 2017) in thermal emission (see Figure 4.1). Similarly, Kami´nskiet al.(2013) did not observe the SW Clump in thermal emission with the Submillimeter Array (SMA), though it was observed in line maps of H2S (300.5 GHz) and CS (293.9 GHz). Given the mass estimates of the SW Clump from the LMIRCam and MMT-Pol observations in Shenoy et al.(2013, 2015), the non-detection in thermal emission in the ALMA bands may have implications for the dust grain properties in the far-IR. In this study, we present LBT/NOMIC (Hoﬀmann et al., 2014) 8.9, 10.3, and 11.9 µm Chapter 4. VY CMa and the SW Clump 72

Figure 4.1 HST/WFPC2 F1042M image of VY CMa, overplotted with ALMA contours at 321 GHz (white) and 658 GHz (black). The corresponding synthesized beams are shown in the bottom left. Note that the SW Clump in the lower right was not detected in the radio. Adapted from O’Gorman et al.(2015). imaging and photometry of VY CMa and its SW Clump. While the earlier LMIRCam observations reveal the scattered light of the dusty clump, NOMIC imaging provides measurements of the thermal emission of the dusty grains. We model the spectral energy distributions of both VY CMa and the SW Clump separately using the radiative-transfer code DUSTY (Ivezic et al., 1997) to show that the thermal emission at 8–12 µm is consistent with a non-detection by ALMA at 400–1000 µm.

4.2 Observations & Data Reduction

We observed VY CMa with NOMIC on UT 2017 January 12 with a single 8.4 m primary mirror on the LBT. The Nulling Optimized Mid-Infrared Camera (NOMIC; Hoﬀmann et al., Chapter 4. VY CMa and the SW Clump 73

2014) is part of the Large Binocular Telescope Interferometer (LBTI; Hinz et al., 2016) system. It uses a 1024 1024 Si:As array with a pixel scale of 0.01800 pix−1 and provides a × ﬁeld-of-view of 1200 1200. Images were made at 8.9 µm (∆λ = 0.76 µm), 10.3 µm (∆λ = × 6.0 µm), and 11.9 µm (∆λ = 1.13 µm) with individual exposure times of 27.5 milliseconds for a total of 90 seconds in each ﬁlter ( 3200 individual frames). The exposure times were ∼ ∼ short to mitigate saturation from the central star, and the telescope was nodded between two positions on the NOMIC chip to ease background subtraction in data reduction. The reduced 8.9 µm image is shown in Figure 4.2 on the right, aligned with the HST/WFPC2

1 µm image from Smith et al.(2001) and the LBT/LMIRCam K s-band image from Shenoy et al.(2015). Figure 4.3 shows the same three frames zoomed in on the SW Clump.

HST 1 m LBT 2 m LBT 9 m

-25°46'01" 01"

02" 02" 0 2 0 0 0 0 2 0 03" 03"

04" 0.5" 0.5" 0.5" -25°46'04"

7:22:58.5 58.4 58.3 7:22:58.5 58.4 58.3 7:22:58.5 58.4 58.3

2000 2000 2000

Figure 4.2 Left: HST/WPFC2 1 µm (F1042M; Smith et al., 2001). Center: LBT/LMIRCam 2.2 µm (Ks-band; Shenoy et al., 2015). Right: LBT/NOMIC 8.9 µm (this work). The white bands in the NOMIC image are artifacts due to column saturation around the central star.

HST 1 m LBT 2 m LBT 9 m -25°46'02.5" 02.5"

03.0" 03.0" 0 2 0 0 0 0 2 03.5" 03.5" 0

04.0" -25°46'04.0"

7:22:58.35 58.30 58.25 7:22:58.35 58.30 58.25 7:22:58.35 58.30 58.25

2000 2000 2000

Figure 4.3 Same as in Figure 4.2 but zoomed in on the SW Clump feature. The ellipse in the bottom right corner is the FWHM of the PSF of the NOMIC 8.9 µm image. Chapter 4. VY CMa and the SW Clump 74

Sirius was observed at similar airmass and with the same nod locations on the NOMIC array for both flux and point-spread function (PSF) calibration. The PSFs were modeled in each wavelength at each nod position using the Astropy (Astropy Collaboration et al., 2013) fitting functions for a two-dimensional Gaussian. For flux calibration, we used photometry of Sirius with Gemini/T-ReCS from Skemer & Close(2011). The T-ReCS and NOMIC filters have similar central wavelengths but different filter bandwidths, so we scale our measured counts into “synthetic filters” to effectively interpolate the Sirius photometry into the NOMIC filter sets. This filter correction permits flux calibration of our VY CMa images. For each nod position, the 3200 frames in each filter are mean-combined with a sigma ∼ clipping threshold of three standard deviations from the average in each pixel. The two nod position images are subtracted from each other, VY CMa is masked out, and the background RMS in each NOMIC amplifier is modeled separately using the Astropy-affiliated photutils1 package. To quantify the flux in the SW Clump relative to VY CMa’s SED, we need to subtract the contribution from the central star itself. In a manner similar to Shenoy et al.(2013) we scale the amplitude of the PSF models from the Sirius images to match the profile of VY CMa. Since the central star is partly saturated, the “wings” of the PSF are used in the scaling to both locate the centroid and scale the amplitude. While centroiding on a saturated source can be uncertain, at the distance of the SW Clump from the star, the flux contribution from the PSF was minimal (. 10% of the flux in the clump at 8.9 µm). We recalculate the SW Clump photometry from Shenoy et al.(2013) on the LMIRCam 0 Ks,L , and M band images for consistent treatment with the NOMIC images. We generate an aperture around the SW Clump using the Ks image to define a region which extends to 1 σ above the background. This aperture is roughly elliptical (0.61 0.4400 beam) and centered × 1.500 from the central star inclined at 45◦ CCW from north. Photometry is performed ∼ with photutils, and the same aperture is used in all the LMIRCam and NOMIC images. Additionally, we apply this aperture to the HST/WFPC2, PSF-deconvolved images from Smith et al.(2001) in the F410M, F547M, and F1042M medium-width continuum filters, and the narrow Hα filter (F656N). In the HST optical images, several arcs, knots, and clumpy features are resolved within the large aperture, so the measured photometry is likely an overestimate. Additionally, without radial velocity measurements of each of these resolved sub-clumps, we cannot determine which of these features are actually coincident

1 photutils provides tools for detecting and measuring the photometry of astronomical sources. The software is still in development, with documentation available at https://photutils.readthedocs.io/. Chapter 4. VY CMa and the SW Clump 75

Table 4.1 Photometry of the SW Clump

† Telescope Instrument Date Obs Filter λ0 Flux Sys. Error (UT) (µm) (Jy) (Jy)

HST WFPC2 22 Mar 1999 F410M 0.4 6.6 10−3 1.9 10−3 × × HST WFPC2 22 Mar 1999 F547M 0.5 0.1 0.04 HST WFPC2 22 Mar 1999 F656N 0.7 0.4 0.15 HST WFPC2 22 Mar 1999 F1042M 1.0 2.1 0.91 LBT LMIRCam 16 Nov 2011 Ks 2.2 6.8 1.2 LBT LMIRCam 16 Nov 2011 L0 3.8 15.0 4.1 LBT LMIRCam 16 Nov 2011 M 4.8 29.1 11.8 LBT NOMIC 12 Jan 2017 8.9 8.9 186.9 52.7 LBT NOMIC 12 Jan 2017 10.3 10.3 318.8 92.5 LBT NOMIC 12 Jan 2017 11.9 11.9 389.4 113.0 ALMA 16 Aug 2013 Band 9 456 0.75∗ ··· ··· ALMA 16 Aug 2013 Band 7 934 1.6 10−2∗ ··· × ··· ALMA 16 Oct 2016 Band 5 1680 8.1 10−4∗ ··· × ···

†Systematic error reported as the standard deviation of the flux in a grid of apertures with different center positions (see discussion in text). Photometric error in the flux-calibrated NOMIC images is estimated at < 10%. ∗Fluxes represent the 3-σ upper limits estimated from the RMS noise in ALMA data scaled to the SW Clump aperture size.

with the SW Clump mass-loss event. However, the aperture photometry in the optical is performed in the same manner as for the IR images for consistency. The SED models described in §4.3.1 do not weight the optical photometry to determine the best fit. The photometry is summarized in Table 4.1. Since the SW Clump is diffuse and we are uncertain of its total spatial extent, we generate a grid of apertures, all with the same total area, but allowing the center to move 0.100 in all directions. The error value in Table 4.1 is the standard deviation of this aperture grid and represents here our measure of systematic uncertainty in the flux. Also included are flux limits for three ALMA bands. VY CMa was observed as ALMA Science Verification data on UT 2013 August 16-19 (321, 658 GHz; Richards et al., 2014; O’Gorman et al., 2015) and on UT 2016 October 16 (178 GHz; Vlemmings et al., 2017). As the SW Clump was undetected in these bands, we instead report a flux limit as 3σ the root-mean-square (RMS) noise in each image, where the measured × RMS in the ALMA images is scaled to the beam-size of our photometric aperture. For example, with the synthesized ALMA beam at 178 GHz of 0.5 0.200 with an RMS noise ∼ × Chapter 4. VY CMa and the SW Clump 76 of 0.1 mJy beam−1 (Vlemmings et al., 2017) and our 0.61 0.4400 aperture beam (2.7 × × ALMA beam-size), then the detection limit assuming the total flux of the SW Clump is distributed evenly over the beam would be 0.1 mJy beam−1 2.7 beams 3σ limit × × 0.8 mJy. These limits are included in Table 4.1. ≈ The observed SEDs of both VY CMa and the SW Clump are shown in Figure 4.4. The closed circles represent photometry of VY CMa compiled from the literature, including the HST/WFPC2 observations at 0.4–1 µm plus the ESO 3.6 m telescope 1–20 µm IR photometry from Smith et al.(2001), and the 20–40 µm SOFIA/FORCAST and 60–150 µm Herschel/PACS photometry from Shenoy et al.(2016). The open circles are the extinction- corrected optical and near-IR photometry for foreground (interstellar) AV = 1.5 (Shenoy et al., 2015) and a traditional extinction curve (Cardelli et al., 1989). The black squares are the photometry from this work on the SW Clump using the elliptical aperture region discussed above in the WFPC2, LMIRCam, and NOMIC images. The 3σ ALMA detection limits are shown as downward arrows in the far-IR to millimeter.

4.3 Results & Discussion

4.3.1 DUSTY modeling

To study the thermal properties of the SW Clump, we model the SEDs of both VY CMa and the SW Clump using the DUSTY radiative-transfer code (Ivezic et al., 1997). DUSTY solves the one-dimensional radiative-transfer equation for either a spherically-symmetric dust distribution around a central source or through a slab of dusty material. For modeling the SED of VY CMa itself, we employ the spherical mode of DUSTY following previous work in Shenoy et al.(2016), which analyzed the mass-loss histories around hypergiant stars µ Cep, IRC +10420, ρ Cas, and VY CMa. Shenoy et al.(2016) fit a variety of dust density distributions to each star, and they found that for VY CMa, a density profile of ρ (r) r−1.5 ∝ best explained the mid-infrared emission in the star’s SED. For our spherical DUSTY model, we adopt this dust density distribution as well as the chemistry from Shenoy et al.(2016)—a 50-50 mixture of astronomical silicates from Draine & Lee(1984) and metallic iron from Harwit et al.(2001). We assume the grain radii follow −3.5 an MRN size distribution n (a) a (Mathis et al., 1977) with amin = 0.005 µm and ∝ amax = 0.5 µm. With an effective temperature of 3490 K (Wittkowski et al., 2012) and an assumed dust condensation temperature of 1000 K, DUSTY generates the model SED shown at the top of Figure 4.4 in red. We use the “slab” mode in DUSTY for the SW Clump, which reproduces the scattered Chapter 4. VY CMa and the SW Clump 77

VY CMa, Shenoy+ 2016 10 13 SW Clump

10 15 2 10 17 m c W

10 19

10 21

1 10 100 1000 m

Figure 4.4 Photometry of VY CMa and the SW Clump. Closed circles represent photometry of VY CMa compiled from the literature. Open circles are the interstellar extinction- corrected optical and near-IR photometry. Black squares are the photometry from this work on the SW Clump from WFPC2 (0.4–1 µm), LMIRCam (2–5 µm), and NOMIC (9–12 µm). The 3σ ALMA detection limits are shown as downward arrows in the far-IR. The red line represents the best-fitting DUSTY model from Shenoy et al.(2016) using a spherical dust distribution with a density profile of ρ (r) r−1.5. The blue line is the best fit slab model ∝ (this work) to the SW Clump. The dotted line is the scattered light component for the slab model, and the dashed line indicates thermal emission, which begins to dominate at 5 µm ∼ to longer wavelengths. Note that the model SED for the clump is fainter than the ALMA limits in the far-IR and millimeter regime and thus consistent with their non-detection.

(reﬂected) and thermal emission from a central source on some planar geometry. To reproduce the SW Clump emission, we generate a grid of models varying the optical depths of the SW Clump material at 8.9 µm (0.01 < τ8.9 < 5). Since the SW Clump is located within a circumstellar nebula of dusty material, the radiation incident on the clump will include light partially extinguished from the star as well as radiation from hot dust between the star and the clump. Therefore, we approximate the central source seen from the clump as a blackbody with eﬀective temperature between 1000 and 2000 K while maintaining the total Chapter 4. VY CMa and the SW Clump 78

Table 4.2 DUSTY Model Parameters & Observed Temperatures

Inputs Outputs Observed a b c d e f g Teﬀ τ8.9 fsc, 5µm Td Tcolor TBB Tbright Tcolor

1600 1.03 92% 207 277 165 205 275

aEffective temperature of the input blackbody to model a “pseudo-photosphere” interior to the SW Clump. bFractional contribution of scattered light to total SED flux at 5 µm (M-band). cDust temperature in the SW Clump measured from the model at the slab boundary facing the star. See §4.3.2 for discussion of the various temperature quantities. dColor temperature from the model SED calculated from a ratio of the 8.9 and 11.9 µm flux. e Blackbody equilibrium temperature for a 270,000 L central source (Wittkowski et al., 2012) at the distance of the SW Clump. fBrightness temperature for the measured flux in the SW Clump. gColor temperature in the SW Clump aperture calculated from the 8.9 and 11.9 µm images.

bolometric flux. A blackbody in this temperature range would roughly peak between 2 and 5 µm. To select a best-fitting model, we evaluate a reduced χ2 measurement of the observed SW Clump photometry and the DUSTY output spectrum. Unlike for the spherical case, we scale the slab DUSTY model SEDs by the solid angle subtended relative to the central star, which for our elliptical aperture is 0.5 sr. The best-fitting model is shown in Fig- ∼ ure 4.4 with the DUSTY input/output parameters summarized in Table 4.2. Our model fitting demonstrates that an optical depth around unity at 8.9 µm is required for the slab to emit the observed flux in the mid-IR. The luminosity of the SW Clump relative to the SED of VY CMa itself serves as an independent check on the aperture area we derived from the 2 µm LMIRCam. The bolometric flux of the clump, estimated by integrating the model curve through the far-IR, is about 3% of the total luminosity of VY CMa. Our clump aperture subtends a solid angle Chapter 4. VY CMa and the SW Clump 79 of 0.5 sr relative to the star, which is 4% of the full sphere. Thus, our aperture area is ∼ ∼ consistent with the observed photometry. We note here two of our greatest uncertainties in constraining the DUSTY models: the SED of circumstellar material between the star and the SW Clump and the geometry of the SW Clump. As discussed above, the SW Clump is not illuminated by the 3490 K photosphere from VY CMa, but rather a combination of attenuated light from the central star and emission from hot dusty material between the star and the clump. We have provided as input to the DUSTY slab a simple blackbody with T= 1600 K to approximate this incident SED, but the actual SED incident on the slab will certainly be more complicated. We note that a few hundred degree variation in the input blackbody temperature does not significantly alter the shape of the model SEDs from 5 µm out to longer wavelengths. The actual extent of the SW Clump is not fully resolved in the NOMIC images. The

photometric aperture was defined from the LMIRCam Ks image for consistency with Shenoy et al.(2013), but as we see in Figures 4.2 and 4.3, the shape of the SW Clump in scattered emission at 2 µm is not the same as in thermal emission at 8.9 µm. Additionally, DUSTY assumes isotropic scattering from dust grains without consideration of a phase angle de- pendence on the scattering efficiency. This may in part explain the different spectral shape observed from 1–5 µm in Figure 4.4. Finally, the 3D morphology of the clump is unknown, so we assume for our models that the slab is in the plane of the sky and is tilted at 45 The assumptions made on the geometry lead to uncertainty in the solid angle subtended by the SW Clump relative to the central star; though, as described above, the fraction of the total sphere subtended by our aperture is consistent with the fraction of flux in the SW Clump relative to VY CMa’s SED.

4.3.2 Scattered vs. Thermal Emission

Shenoy et al.(2013) used the BHMIE code (Bohren & Huffman, 1983) to calculate the extinction and scattering efficiencies of dust grains in the SW Clump using Mie theory to determine the fractional contribution of scattering and thermal emission in the SED. At 5 µm, Shenoy et al.(2013) estimates that 75% of the flux in the SW Clump is due to ∼ scattered light from VY CMa. We can make a similar calculation since DUSTY also separates the scattered and thermal components of the SEDs, shown in Figure 4.4 with dotted and dashed lines, respectively. We derive a fractional contribution from scattering at 5 µm of 92%. At wavelengths longer than 10 µm, the emission is purely thermal. In Table 4.2, we present several distinct temperature measurements. DUSTY provides estimates on the dust temperature as a function of optical depth through the slab. For the Chapter 4. VY CMa and the SW Clump 80 surface of the slab facing the star, the dust temperature Td is 207 K. As an independent check on consistency, we can roughly measure the dust temperature directly from the flux- calibrated NOMIC images. The analagous quantity to the DUSTY temperature at the slab’s surface would be an observed brightness temperature in the IR, calculated for a blackbody as: 2 hc −1 2hc Tbright = ln 1 + 5 (4.1) kλ Iλλ

For the total flux in our SW Clump aperture at 8.9 µm, this yields Tbright = 205 K, similar to the “physical” dust temperature from DUSTY at the clump’s surface. We can also measure a color temperature from both the observed photometry and the model SED. The NOMIC 8.9 and 11.9 µm filters bracket the 10 µm silicate emission feature, and therefore sample the continuum emission from dust in the SW Clump. The ratio of the observed photometry yields a color temperature of 275 K. We recover a similar measurement from the model SED of 277 K, which is not surprising since the χ2-fitting performed on our grid of DUSTY models guarantees a recovered SED with a similar spectral shape to the observed photometry in the IR. Also included for comparison in Table 4.2 is the blackbody-equilibrium temperature evaluated at a distance of 1500 AU from a source with the bolometric luminosity of VY CMa (270,000 L ; Wittkowski et al., 2012).

4.3.3 Mass Estimates

Since the optically-thick τ8.9 = 1.03 DUSTY model recovers the observed total flux in the SED, we can estimate the total mass in the SW Clump from its optical depth, the grain size distribution, and the extinction efficiency. Optical depth is defined as:

Z amax 2 τλ Qλ n(a) πa da (4.2) ≡ amin where Qλ is the extinction efficiency factor, and n(a) is the MRN grain size distribution discussed in §4.3.1. Since our input grains to the DUSTY models are 50% silicates and 50% iron, our efficiency is simply the average of the efficiency functions from Draine & Lee (1984) and Harwit et al.(2001) (here, Q 0.05 at 8.9 µm). We assume both the extinction ≈ efficiency and the internal mass density of the grains (ρ) are constant with grain size, and we define the column density as: Z amax 4 m = ρ n(a) πa3 da (4.3) amin 3 where ρ is a typical grain mass density of 3 gm cm−3. Combining these two equations and multiplying by the total area in the clump—2.6 1032 cm2 for our aperture at a distance × Chapter 4. VY CMa and the SW Clump 81

−5 of 1.2 kpc (from VLBA parallax; Zhang et al., 2012)—we derive a mass of 5.4 10 M in × dust. Adopting a gas:dust ratio of 100:1 (for consistency with Shenoy et al., 2013), yields a −3 total mass (gas+dust) in the SW Clump of 5.4 10 M . × −3 This result is consistent with the lower limit estimate of M 5 10 M from imaging & × polarimetry at 3.1 µm by Shenoy et al.(2013). If, however, we adopt the higher gas:dust ratio of 500:1 from Decin et al.(2006) for VY CMa, our mass estimate for the SW Clump be- −2 −4 −1 comes 2.7 10 M . Compared to the typical mass-loss rate of VY CMa of 10 M yr × ∼ (Danchi et al., 1994), such a large mass in a discrete feature likely represents an extreme, localized mass-loss event. For comparison, O’Gorman et al.(2015) estimates a dust mass for the Clump C feature of −4 2 10 M , which they cite as a lower limit since their calculation is in the optically-thin ∼ × regime. With additional Band 5 (178 GHz) ALMA data, Vlemmings et al.(2017) updates −3 this dust mass to > 1.2 10 M . Clump C is then almost two orders of magnitude × more massive than the SW Clump. Richards et al.(2014) and O’Gorman et al.(2015) also identiﬁed a second radio-bright continuum source at or near the center of the star that they −5 call the VY component. This source has a mass estimate of 3 10 M , which is about ∼ × half of our dust mass estimate for the SW Clump.

4.4 Conclusions

High-resolution imaging with NOMIC has allowed us to isolate the peculiar SW Clump feature from the IR emission of VY CMa. The resulting SED is a powerful tool in characterizing the thermal properties of the clump relative to the central star. Through DUSTY modeling, we confirm that the clump is optically-thick from 9–12 µm and has a brightness −5 temperature of 200 K. With a dust mass of 5.4 10 M , the SW Clump is comparable ∼ × in mass to the radio-bright Clump C and “VY” component identified in Richards et al. (2014) and O’Gorman et al.(2015). At a distance of 1500 AU, the SW Clump represents a recent mass-loss event from ∼ VY CMa. If we assume a conservative value for the velocity of 25 km s−1, typical for red supergiants, then the clump would have been ejected . 300 years ago. However, in the ejecta of VY CMa, Decin et al.(2006) finds a velocity as high as 60 km s −1 for dust grains with size 0.25 µm. With this wind speed, the SW Clump and Clump C could each have ∼ been ejected just over a century ago. Finally, we note that our models and estimates on thermal emission from the dust in the SW Clump are consistent with the non-detection at ALMA. The SED models predict far-IR Chapter 4. VY CMa and the SW Clump 82

ﬂux at or below the 3σ ALMA detection limits for our SW Clump aperture beam. However, we note that our models are not at all constrained beyond the 11.9 µm NOMIC photometry. Therefore, high-resolution imaging of the SW Clump in the 20–100 µm regime is required ∼ to characterize fully the thermal emission from this fascinating mass-loss event.

M Gordon acknowledges support from the University of Minnesota Graduate School’s Doc- toral Dissertation Fellowship. Chapter 5

Obscured Red Supergiants in M33

Abstract Since the target selection for Gordon et al.(2016) was drawn from optical surveys, our catalog of the most luminous stars in M31 and M33 does not necessarily include supergiants that may be heavily obscured and faint in the optical. Since the most luminous warm and cool supergiant populations are more likely to have the highest mass-loss rates, it is probable that those stars will be highly obscured by their own circumstellar ejecta. To complete the upper portion of the HR diagram requires investigating bright infrared sources with faint or undetected optical counterparts. As a follow-up to the optical spectroscopic survey on mass-losing RSGs, we obtained infrared spectra on IRTF/SpeX of bright Spitzer/IRAC sources in M33 with strong excess emission in their SEDs. We ﬁnd higher than average mass-loss rates for these RSG candidates −5 of 5 10 M /yr through preliminary modeling with radiative-transfer code ∼ × DUSTY. We additionally discuss two possible η Car analogs, Object X and M33-8, and outline the future observations and analysis for this project.

5.1 Introduction

In Gordon et al.(2016), we found that 50% of observed RSGs in M31 and M33 show ∼ evidence for mass loss in their SEDs. Thus, a large fraction of supergiants may evolve back to the blue to become the warm post-RSGs. However, the target selection for that paper was drawn from optical surveys. Therefore, our survey of the most luminous stars in M31 and M33 does not include supergiants that may be obscured in the optical. Since the most luminous warm and cool supergiant populations are more likely to have the highest mass-loss

83 Chapter 5. Obscured RSGs in M33 84 rates (Mauron & Josselin, 2011) it is probable that those stars will be highly obscured by their own circumstellar ejecta. To complete the upper portion of the HR diagram requires investigating bright infrared sources with faint or undetected optical counterparts. What fraction of RSGs in the local universe are obscured? Of the heavily-reddened sources in M33, which are RSGs, extreme-AGB stars, or emission-line stars at diﬀerent stages of evolution? In this project, we seek to explore the bright infrared sources in M33, spectroscopically identify these sources, study their mass loss and mass-loss histories, and place them on the HR diagram. This IR spectroscopic survey is a natural extension of our optical spectroscopic survey of the luminous stars in M31 and M33 with the LBT and MMT (Humphreys et al., 2013, 2014; Gordon et al., 2016, see also §2). Here, we summarize our ongoing and future work observing obscured RSG candidates with longslit spectroscopy at the NASA Infrared Telescope Facility (IRTF) on Mauna Kea, Hawaii.

5.2 Target Selection

We have selected infrared-bright sources from two Spitzer/IRAC surveys (McQuinn et al., 2007; Khan et al., 2015) and matched them to 2MASS and WISE photometry. The dusty, obscured RSG candidates will have an infrared excess from thermal dust emission, so we select all stars with K [4.5] > 1.0. We performed cross-identification on two optical − catalogs, the CFHT survey of M33 (Hartman et al., 2006) and the LGGS (Massey et al., 2007b), to select only the infrared sources with either no detected optical counterpart, or with strong IR excess emission. Figure 5.1 shows a color-magnitude diagram of the IRAC sources in M33 from McQuinn et al.(2007). The blue squares are highly-reddened stars with the K [4.5] criterion from above. A subset of those blue squares are all the sources − with no optical counterparts or have evidence for CS ejecta in their SEDs (the red squares). The resulting 58 red, obscured sources are likely RSGs or other evolved stars. For comparison to our previous work on RSGs in M31, we revisit the MMT/Hectospec target selection from Gordon et al.(2016). In Figure 5.2, we adapt the Hectospec fiber positions from Figure 2.1 and overlay the IR bright Spitzer sources from (McQuinn et al., 2007; Khan et al., 2015). The blue circles correspond to the blue squares in Figure 5.1. Note that some of these sources overlap with the Hectospec optical survey sources. Some of these stars would have been flagged as having an optical counterpart from the LGGS and are removed from our final target list (the red squares in Figure 5.1), while some were re-observed in this work to serve as controls. Chapter 5. Obscured RSGs in M33 85

14 K

M 12

3 2 1 0 1 2 3 4 K - [4.5]

Figure 5.1 CMD of Spitzer/IRAC sources in M33 (pphotometry from McQuinn et al., 2007; Khan et al., 2015). Blue squares are highly-reddened sources that meet our color criterion K [4.5] > 1.0. Red squares are all the obscured sources with either no optical − counterparts in the LGGS (Massey et al., 2007b) or the CFHT survey of M33 (Hartman et al., 2006), or have signiﬁcant IR excess in their SEDs. These 58 obscured sources (red) are our target RSG candidates.

The infrared SEDs of two candidate RSGs are shown in Figure 5.3 with blackbody ﬁts to the 2MASS photometry for estimates on color temperature. All of the targets show evidence for warm dusty circumstellar ejecta as infrared excess in their SEDs. With bright photometry in the K-band ( 13 15 mag), these sources are prime candidates for follow-up ∼ − IR spectroscopy. One of the goals for this project is to use low-resolution infrared spectra to identify whether the sources are indeed dusty, M-type supergiants or instead obscured AGB/extreme-

AGB stars or warm hypergiants based on the presence of CO, TiO, H2O, or atomic absorption features. As many of these sources are likely embedded in emission nebulosity, long-slit spectroscopy is required to separate the stars from the background nebulosity. Chapter 5. Obscured RSGs in M33 86

30°50'00" 0 0

0 40'00" 2

30'00"

1:35:00 34:30 00 33:30 00 32:30

2000

Figure 5.2 Similar to Figure 2.1 with an overlay of the bright IR targets from Spitzer/IRAC shown in blue.

5.3 Observations & Data Reduction

Observations were conducted at the IRTF with SpeX, a 0.7 5.3 µm medium-resolution − spectrograph and imager built at the University of Hawaii, Institute for Astronomy (Rayner et al., 2003). SpeX uses a Teledyne 2048 2048 Hawaii-2RG array, which provides a spatial × resolution of 000. 10 per pixel in spectrographic mode. Considering our targets are optically- obscured, IR guiding is one of the key elements for the success of this project. One of the guiding modes in SpeX allows for spillover light from the slit to be used for auto-guiding adjustments. This method proved essential for this project, although it did limit the targets Chapter 5. Obscured RSGs in M33 87

10 11 J013403.84+303752.9 Tcolor = 2200 K K = 13.94 mag )

2 10 12 m c 1 c e

s 13

g 10 r e ( F

10 14 W1 W2 W3 W4 J H K 3.6 4.5 8.0

10 11 J013343.62+303343.6 Tcolor = 2020 K K = 15.47 mag )

2 10 12 m c 1 c e

s 13

g 10 r e ( F

10 14 W1 W2 W3 W4 J H K 3.6 4.5 8.0

0.2 0.5 1 2 5 10 20 ( m)

Figure 5.3 SEDs of two RSG candidates in our target list. Infrared excess in the IRAC and WISE bands indicates the presence of warm dusty circumstellar ejecta. Color temperatures are estimated by the blackbody ﬁts to the 2MASS and IRAC photometry. WISE photometry is shown with open circles since the large beam-size (up to 1200 at 22 µm; ∼ Wright et al., 2010) creates some uncertainty in the mid-IR photometry due to crowding in the ﬁeld or unresolved nebulosity in the beam. to brighter than 16 mag in J.1 ∼ We observed 40 of the 58 red optically-obscured sources from our M33 Spitzer sample, in addition to another 30 RSGs from three Galactic clusters. We also observed the peculiar

1 SpeX can guide on objects as faint as 18 mag in J if located in the ﬁeld oﬀ the slit (http://irtfweb.ifa. hawaii.edu/~spex/spex_manual/SpeX_manual_06Oct17.pdf). Chapter 5. Obscured RSGs in M33 88

“Object X,” the brightest mid-IR source in M33 (Khan et al., 2011, discussed in §5.5). All the M33 sources were observed with the 0.7 2.52 µm prism, which provides a resolving − power of R 200 with a 000. 3 slit, while some of the brighter RSGs in cluster NGC 7419 ∼ were observed with the short wavelength cross-disperser, SXD mode, which has the same wavelength coverage but a higher resolving power of R 2000. The data were collected over ∼ ﬁve nights on 2016 Sept 11, 2016 Nov 18, and 2017 Oct 16-18. The integration time for each target was between 6 and 12 minutes depending on the brightness of each source as well as sky conditions. Each observation was dithered along the slit to aid in sky subtraction. The data were reduced using Spextool (Vacca et al., 2003; Cushing et al., 2004), which wavelength calibrates the spectra using both comparison lamps and sky lines. Spextool also ﬂux calibrates the spectra and performs telluric corrections. Figure 5.4 shows a montage of two highly-reddened sources from our M33 sample (top and center), along with an RSG candidate from M31 for comparison (bottom; Massey et al., 2007b; Gordon et al., 2016). The SEDs in the left panels show IR excess emission in all three sources, and note also that these stars did have optical photometry from the KPNO 4m LGGS survey (Massey et al., 2006). The optical and near-IR photometry has not been corrected for interstellar or CS extinction. At the time of this writing, several sources are being reprocessed to better separate them from their surrounding nebulosity.

5.4 Modeling & Analysis

Similar to work described in §3(Gordon et al., 2018a), we use the DUSTY radiative-transfer code to model the observed SEDs and derive the mass-loss rates of these RSGs. As in §3, we generate a grid of models for each star at fixed effective temperatures, circumstellar shell extent (1000 r1), and dust condensation temperature (1000 K). For each set of model × inputs, DUSTY outputs a model SED, which we match to the observed IR photometry through χ2-minimization. While we explored a variety of power-law dust density distributions and mass-loss histories for the Galactic RSGs, we restrict our DUSTY models to constant mass- loss rates (ρ (r) r−2) for our preliminary analysis of the IRTF M33 sample stars. ∝ An example of the modeling is illustrated in Figure 5.5 for RSG candidate J013403.84+303752.9 (SED shown in Figure 5.3). For this initial pass through the DUSTY grids, the model spectrum shown was the best fit to the near- to mid-IR photometry, as well as the Ks-band IRTF spectrum. Also shown are some of the distinguishing features for RSGs—CO band-heads and deep molecular TiO in absorption. The mass-loss rate derived from this particular Chapter 5. Obscured RSGs in M33 89

10 15 2 m 16 W 10

J01340369+3042022 10 17

2 10 15 m W 10 16 J01341224+3053141

10 15 2 m W 10 16 M-004124.81

1 10 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 m m

Figure 5.4 SEDs and IRTF SpeX Ks-band spectra of two RSG candidates in M33 (top, middle), and of RSG candidate M-004124.81 (bottom) from Gordon et al.(2016) (LGGS J004124.81+411206.1 from Massey et al., 2007b, for comparison). The gray bands represent the relative opacity of telluric contamination in the target spectrum. The optical and near-IR photometry has not been corrected for interstellar or CS extinction. For labeled absorption features typical for RSGs, see Figure 5.5.

−5 model was 3 10 M /yr for a 3300 K effective temperature photosphere. In fact, the × −5 average mass-loss rate for our sample of stars is 5 10 M /yr, though we re-iterate that ∼ × these results are still preliminary. This high mass-loss rate can be compared to a general −7 −4 range of 10 10 M /yr from RSG surveys of the Galaxy, M31, or in the LMC (e.g., ∼ − Massey, 1998; Levesque et al., 2005; Massey et al., 2009; Mauron & Josselin, 2011, and certainly others). Considering the population studied here was selected based on strong IR excess emission and highly-reddened SEDs, it is not surprising, then, that these RSGs have somewhat larger mass-loss rates than average. One of the modeling parameters we explored in Gordon et al.(2018a)( §4) is the outflow chemistry—whether, for example, the dust particles are Draine & Li(2007) “astronomical silicates,” cold or warm oxygen-rich silicates (Ossenkopf et al., 1992), or containing a large fraction of atomic iron as in VY CMa (Harwit et al., 2001; Shenoy et al., 2016). The particle chemistry in the outflow models can have an appreciable effect on the mid- to far-IR SED, and the input dust density distribution (assumed constant for these first-pass results) can Chapter 5. Obscured RSGs in M33 90

1e 15 5 Teff = 3300 K DUSTY model TiO 4 CO TiO )

m 3 / 2 m / W

( 2

F CO

0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 ( m)

Figure 5.5 IRTF SpeX Ks-band spectrum of J013403.84+303752.9. The labeled absorption features are molecular band heads for TiO and CO—typical markers in evolved stellar photospheres. The gray bands represent the relative opacity of the atmosphere (telluric contamination in the target spectrum). Overplotted is a DUSTY model for a 3300 K star −5 with a mass-loss rate of 3 10 M /yr. ∼ × drastically alter the mass-loss rate over the lifetime of the outflowing dusty shell. We will experiment more with the DUSTY modeling on this dataset in the months following the publication of this thesis. A potential contaminant for dusty M-type supergiants in M33 is a population of AGB or “extreme” AGB stars, which are photometrically identical to RSGs. However, SpeX provides sufficient spectral resolution at 2 µm to distinguish the two stellar populations. Figure 5.6 shows a low-resolution IR spectrum of WX Ser from Jones et al.(1988) observed with UKIRT. WX Ser is a classical Mira variable but is also an OH/IR star. Its spectrum has very strong CO and water absorption bands, the latter of which can be seen from 1.7 2.0 µm and from 2.4 3.0 µm. For our RSG candidates in M33, we were using these − − two deep absorption patterns as criteria for flagging AGB or otherwise non-supergiant stars from our sample. While we are still processing some of the spectra from the five observing nights, so far more than two-thirds of our sample are RSGs with no obvious AGB stars in the list. The remaining sources are still being classified, but current work on two of the sources is described in §5.5 below. Figure 5.7 shows the sample of observed M33 stars plotted on a CMD. The bolometric Chapter 5. Obscured RSGs in M33 91

1e 10 1.2 WX Ser M8e

TBB = 1750 K 1.0 )

m 0.8 / 2 m / 0.6 W (

F 0.4

0.2

1.5 2.0 2.5 3.0 3.5 ( m)

Figure 5.6 IR Spectrum of Mira variable WX Serpentis. WX Ser is also an OH/IR star with a IR excess in its SED due to a dusty wind. Note the strong, deep absorption band at 1.9 µm, which is due to the water vapor in the photosphere of the star. This deep absorption can be contrasted with the IR spectrum of the RSG in Figure 5.5. Adapted from Jones et al.(1988). luminosities were calculated by integrating the IR SEDs and scaling the sources to a distance of 794 kpc (average distance to M33 from Cepheids distance scale; Scowcroft et al., 2009; Humphreys et al., 2013). The SEDs were not corrected for interstellar extinction, though in the IR this correction would be minimal. Since many sources lack optical photometry, these luminosities are still somewhat uncertain. As these stars peak in the near-IR, though, the values in Figure 5.7 may well represent underestimates. However, considering the large beam-size for WISE—and even the 200 PSF FWHM for Spitzer/IRAC2 —the IR SED will be contaminated with surrounding nebulosity. Regardless of the PSF size, the pixel scale for IRAC is 100. 22, which at 1000 kpc encompasses a physical diameter of 6 pc. Therefore, ∼ the IRAC photometry is most likely an overestimate of the ﬂux from each source, since the beam will capture nebulosity that may not be associated with a single RSG, or potentially the ﬂux from multiple stars in close proximity, as is the case for at least one pair of sources in our IRTF sample.

Also shown in Figure 5.7 is the AGB luminosity limit of Mbol 7.1 (Costa & Frogel, ∼ − 1996). With the exception of one source, all of the M33 sample stars exceed this limit and are likely RSGs. This luminosity criterion, combined with the lack of deep water absorption

2 PSF and PRF information available in the IRAC handbook available online at https://irsa.ipac.caltech. edu/data/SPITZER/docs/irac/iracinstrumenthandbook/5/. Chapter 5. Obscured RSGs in M33 92

107

106 L / L

105

0.2 0.4 0.6 0.8 1.0 1.2 1.4 [H] [Ks]

Figure 5.7 Color-luminosity diagram of the IRTF M33 sample stars. The dashed line is the AGB luminosity limit (Mbol 7.1; Costa & Frogel, 1996). With one exception, ∼ − all of our M33 sources exceed this limit. The bolometric luminosities were calculated by integrating the near- to far-IR SEDs and adopting a distance to M33 of 794 kpc (Scowcroft et al., 2009). The brightest source, M33-8 (2MASS J01340022+3040475), shown in red, is discussed further in §5.5. in any of our IRTF spectra, suggest that our sample sources represent a clean population of RSGs. The brightest source in Figure 5.7, J01340022+3040475 (in red), is discussed further in §5.5 below.

5.5 Object X & M33-8

h m s ◦ 0 00 Object X (M33-1; α2000 = 01 33 24. 1, δ2000 = +30 25 34. 8) was ﬁrst identiﬁed by Khan et al.(2011) as the brightest IR source in M33 in all four IRAC bands. Object X, shown 5 in R and Hα in Figure 5.8, has a bolometric luminosity of 5 10 L and is optically ∼ × variable on short timescales of tens of days, indicating that it is stellar in nature. Khan et al.(2011) traced the position through archival photographic plates and found no optical source, which suggests that the star has only recently emerged (since 1991) from an ∼ Chapter 5. Obscured RSGs in M33 93 obscured state. Based on these historical data and its current optical and IR SED, Khan et al.(2011) proposed that Object X is a self-obscured evolved star of 30 M . ∼

30°25'50"

40" 0 0 0 2

30"

20"

1:33:25.0 24.0 23.0 1:33:25.0 24.0 23.0

2000 2000

Figure 5.8 R-band and Hα images of Object X (M33-1; Khan et al., 2011). M33 mosaic images observed from the KPNO 4m by P. Massey.

In Khan et al.(2013) and Khan et al.(2015), several self-obscured stars in nearby galaxies were identified from their SEDs and were named “η Carinae analogs.” Of the nine candidates, Khan et al.(2013) concluded that five, including Object X, were single stars. Considering how rare these objects are, they represent a unique laboratory for studying a short-lived evolutionary stage of massive stars. Here, we represent preliminary analysis of the IR and optical spectra of Object X. The SED and SpeX spectrum from the M33 self-obscured sample in the above study are presented in Figure 5.9. The Ks-band spectrum is largely featureless. With no identi- fiable CO or TiO molecular absorption, Object X is likely not an M-type star. Follow-up spectroscopy was performed in the optical in October 2016 using the Multi-Object Double Spectrographs (MODS; Pogge et al., 2006) on the LBT. Due to its faintness, Object X was observed with the combined light from both mirrors for separate exposures in the red and blue for total integration times of 150 and 180 minutes, respectively. The red spectrum ( 6400 9500 A)˚ in Figure 5.10 shows strong Ca II absorption lines, which are luminosity ∼ − and temperature sensitive indicators of warm supergiants. A zoomed-in view of the Ca II lines is shown in Figure 5.11 and reveals asymmetric line profiles due to emission on the red side—evidence for an outlowing wind. Comparison with the Ca II triplet absorption in known yellow supergiants from Humphreys et al.(2013); Gordon et al.(2016) suggests Chapter 5. Obscured RSGs in M33 94 that Object X is late F- to early G-type star. Also highlighted is the O I λ 7774 A,˚ which is another luminosity-class indicator for supergiants. As Figure 5.8 shows, Object X is surrounded by nebulosity. This nebular emission was isolated from the Object X spectrum by subtracting the background on both sides of the stellar profile of the MODS data. The optical spectrum from the H II region is shown in blue in Figure 5.10.

10 14

CO CO

2 16 10 Br m

W TiO

10 18 Obj X

1 10 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 m m

Figure 5.9 SED and SpeX Ks-band spectrum of Object X, observed with the IRTF on 2016 Sept 12. As in Figure 5.4, the gray bands represent telluric contamination. The spectrum is mostly featureless in the IR, with faint Brδ emission, and possibly CO band- head emission. However, Object X is embedded in nebulosity (see Figure 5.8), so the emission spectrum might be contaminated by nebular features.

8e-18

Ca II ] A / 2 m c / O I s

/ 4e-18 g r e [ F

6500 7000 7500 8000 8500 9000 9500 Wavelength [A]

Figure 5.10 MODS red spectrum of Object X (in red). The nebular contamination from the nearby H II region (in blue) was subtracted from the Object X spectrum. Ca II triplet and O I λ 7774 A˚ absorption are strong temperature and luminosity-class indicators, suggesting that Object X is likely an embedded yellow supergiant.

Another object from this program to highlight in brief is M33-8 (J01340022+3040475; h m s ◦ 0 00 Khan et al., 2013, α2000 = 01 34 00. 22, δ2000 = +30 40 47. 5), the extremely bright (L ∼ Chapter 5. Obscured RSGs in M33 95

Ca II Em. Em. Em.

8450 8500 8550 8600 8650 8700 Wavelength [A]

Figure 5.11 Zoomed-in view of Figure 5.10 on the Ca II absorption. This view highlights the emission seen on the red side of each absorption line. This asymmetry is suggestive of an outﬂowing wind. Adapted from Humphreys et al.(2018).

6 7.9 10 L ) source from the CMD in Figure 5.7. This source is shown in V-band and × Hα in Figure 5.12, the latter of which reveals that M33-8 is embedded in or surrounded by nebulosity. The IRTF SpeX spectrum is shown at the top of Figure 5.13 alongside its optical and near-IR SED. Compared to the RSGs in our survey (Figure 5.4) with a dearth of emission lines, this spectrum is quite unusual. While the strong Paschen and Brackett series emission lines could be indicative of an H II region, the very strong He I λλ 1.083 and 2.058 µm emission, as well as Fe II (and [Fe II]) in emission is suggestive of an embedded hot star. Additionally, this source was identiﬁed as a resolved point source in 2MASS (Cutri et al., 2003), and Khan et al.(2013) ﬂagged it as a possible η Car analog,

similar to Object X. For comparison, we include in Figure 5.13 the J, H, and Ks-band spectra of η Car from Allen et al.(1985), which are remarkably similar. At the time of this writing, we have only begun our analysis of this curious star. We will be following up the IR observations with optical spectra on LBT/MODS. If we can identify spectral-type and luminosity-class indicators in the optical spectra, we can determine the evolutionary Chapter 5. Obscured RSGs in M33 96 state of this star. Combined with the near-IR spectrum presented here, we will able to determine if this object is indeed one source rather than a tight association of multiple hot stars. Finally, the near- to mid-IR SED can be used to reconstruct the mass-loss history of any outﬂowing circumstellar material.

30°41'00"

0 40'50" 0 0 2

40"

30"

1:34:01.0 00.0 33:59.0 1:34:01.0 00.0 33:59.0

2000 2000

Figure 5.12 V-band and Hα images of M33-8 (2MASS J01340022+3040475). The Hα image reveals an area of enhanced nebulosity around the star; however, we cannot say at this stage in the analysis if any of that nebular emission is associated with this star. Chapter 5. Obscured RSGs in M33 97 2.6 and γ ) are the m is from µ 2.4 bottom

2.2 r B He I

a 2.0 P

r B Adapted from Allen, Jones, Hyland (1985) m 1.8 Br 9 Br 10 + Fe II Br 11 + [Fe II] 1.6 Br 12 + [Fe II] Br + 1.4 ). The photometry from MIPS at 70 and 160

a top P 1.2

[Fe II] a P He I 1.0 r a C

100 M33-8 ]—all indicators of an outﬂowing wind of collisionally-excited material. The Pa II 10 Car from Allen et al. ( 1985 ). Note the strong Paschen and Brackett emission features, as well m η , and [Fe II , Fe I 1 SED and IR spectrum of M33-8 (J01340022+3040475; m lines are slightly blended in the M33-8 spectrum. The two spectra have been arbitrarily scaled for this visual UKIRT spectra of µ s 3 4 5 0 1 9 083 1 1 1 1 1 . 1 0 0 0 0 0 0 λ 1

1 1 1 1 1

m W m W

2 2 comparison. J, H, and K as the strong linesHe of He Figure 5.13 Khan et al. ( 2013 ), which was estimated as an upper limit due to the large beam-size. Shown for comparison ( Chapter 5. Obscured RSGs in M33 98 5.6 Conclusions

We present here preliminary results on IRTF/SpeX observations of M33 sources with either strong excess emission in their IR SEDs and/or have no counterpart in optical surveys. Even with low spectral resolution (R 200) PRISM mode on SpeX, we are able to distinguish ∼ potential RSGs from AGB or extreme AGB sources. Thus far in the analysis, 40 out of ∼ the 58 obscured sources identified from Spitzer/IRAC appear to indeed be RSGs, with no obvious AGB contaminants in the population. Included is our initial exploration into using DUSTY radiative-transfer models in conjuc- tion with the near- to mid-IR SEDs to estimate mass-loss rates. The preliminary measure- −5 ments from the models suggest a rather high average mass-loss rate of 5 10 M /yr for ∼ × the population of objects studied. If this mass-loss rate remains consistent as we progress further in the analysis, then these stars could prove to be in a highly evolved, post-RSG state, similar to the supergiants studied in Gordon et al.(2016)( §2). Finally, we summarize our efforts in identifying two particularly curious objects from the Khan et al.(2013) catalog of possible η Car analogs in M33. Object X, with strong Ca II and O I triplet absorption, is possibly a very high luminosity yellow supergiant. With evidence for additional Ca II lines in emission, Object X is likely the progenitor star for the obscuring CS dust evident in its SED. We have distinguished its spectrum from that of the surrounding H II region using LBT/MODS, and a preliminary velocity measurement of -133 km/s for the H II region is consistent with the expected velocity for its distance from the center of M33. Object X is possibly more than one star, although imaging in the near-IR with HST would be necessary to resolve this bright source. We have briefly explored here our initial thoghts on M33-8, another η Car-like star. Its IR spectrum has very strong hydrogen and helium lines in emission, in addition to ionized iron and forbidden iron transitions which suggest the presence of dusty outflows. We will studying this source carefully over the next few months after follow-up optical spectroscopy is taken on the LBT.

This work uses data obtained with the MODS spectrographs built with funding from NSF grant AST-9987045 and the NSF Telescope System Instrumentation Program (TSIP), with additional funds from the Ohio Board of Regents and the Ohio State University Oﬃce of Research. Chapter 6

Conclusions & Future Work

Our observing program over the last few years has enhanced our understanding of the final stages and pre-supernova evolution of massive stars. The Red Supergiant Problem—the possibility that a large fraction of massive stars above some initial mass may not explode as supernovae—increases the importance of studying mass loss in evolved stars. High mass- loss episodes not only influence the terminal state of an RSG, but also return a significant amount of processed material to the interstellar medium. While a majority of massive stars will pass through the RSG branch of the HR diagram, our results on post-RSG candidates, in addition to evolutionary track modeling from the literature, suggest that at least 30% of stars above 20 M will evolve back to warmer temperatures. The role of mass loss in ∼ this evolution remains an open question in the field. However, the observations and DUSTY modeling explored in the chapters above will set limits on the frequency of high mass-loss events in the RSG and post-RSG stages of massive stars and the impact of mass-loss history on the terminal state of both the red and yellow supergiant populations. While Chapter5 discusses our ongoing work on analyzing IRTF/SpeX 2 µm spectra of obscured supergiants in M33, in this closing section we will outline other current and future observations on red supergiants in cluster environments. Beginning in 2017, we have begun a multi-wavelength exploration into three Galactic RSG clusters with the LBT, VLT, h m s ◦ 0 00 and SOFIA. These three clusters, RSGC1 (α2000 = 18 36 29 , δ2000 = 06 52 48 ; Figer − h m s ◦ 0 00 et al., 2006), RSGC2 (Stephenson 2; α2000 = 18 39 21 , δ2000 = 06 01 44 ; Stephenson, − 1990; Davies et al., 2007), and NGC 7419 (see Figure 6.1; Beauchamp et al., 1994; Marco & Negueruela, 2013) each contain co-eval populations of 12, 23, and 5 RSGs, respectively. ∼ With high-resolution infrared imaging of clustered RSGs and conservative estimates on outflow velocities, we can directly determine the mass-loss histories from the past 100 1000 − years.

99 Chapter 6. Conclusions & Future Work 100

60°52' N294

B950 N230 50' N207 B435 0 0

0 N153 2 B696 N126 N98 N96 B921 B139 48' N47 N44 N55 N34 N30 N18

46' 22:54:48.0 36.0 24.0 12.0 00.0

2000

Figure 6.1 2MASS Ks-band image of red supergiant cluster NGC 7419. The numbers correspond to the original IDs from Beauchamp et al.(1994). The stars labeled with “B” are the ﬁve RSGs in the cluster. The other sources are members of NGC 7419, but are either AGB stars or other subgiants in the cluster.

While SEDs provide only a single snapshot in time, imaging plus photometry allows for investigating both the timescale and magnitude of the stars’ mass-loss, as we demonstrated in Chapter3(Gordon et al., 2018a) on the Galactic RSGs and OH/IR stars. By comparing the surface brightness of circumstellar material at different radii from the central source with LBT/LMIRCam from 1 5 µm and with SOFIA/FORCAST at wavelengths as long − Chapter 6. Conclusions & Future Work 101 as 40 µm, we can directly observe if the mass-loss history is dominated more by episodic mass ejections or smoother circumstellar wind. Finally, analyzing RSGs in co-eval clusters is important since any observed differences among the stars cannot be related to metallicity or age effects. Thus, these clusters provide excellent populations for constraining mass-loss in evolved supergiants at a fixed age and in a fixed environment.

104

103 Counts

102

0.00 0.05 0.10 0.15 0.20 0.25 0.30 R (arcsec)

Figure 6.2 LBT/LMIRCam radial proﬁle of RSG B139 in NGC 7419 in the Ks band. The red dashed line represents the observed PSF. Note that the stellar proﬁle has an obvious excess above the PSF, which indicates possible resolved CS material.

In December 2017, we obtained imaging of several RSGs in the NGC 7419 cluster with LBT/LMIRCam at 2 µm. An example surface brightness profile is shown in Figure 6.2, comparing RSG B139 to the observed PSF. Note that the CS material is clearly resolved above the PSF an almost an order of magnitude brighter than the first Airy ring peak at 000. 2. The analysis of this star and others in the cluster is still preliminary—due to weather- ∼ related issues, the sampling rate for the AO system on LBT was different between the PSF calibrator and the target star. We are currently investigating what effect, if any, this would have on surface brightness measurements of the CS material. We will be following up these observations with similar imaging of stars in the other clusters, RSGC1 and RSGC2. We have obtained time on SOFIA with FORCAST for long wavelength (11 40 µm) − photometry. Capable of imaging most of the cluster in a single pointing, the 30 30 field- × of-view is a very efficient match to RSGC1 (see Figure 6.3). With the spatial resolution Chapter 6. Conclusions & Future Work 102 on SOFIA, we will not be able to resolve individual structure in the CS ejecta, but we can sample the colder dust, if any, around each of the RSGs in the cluster. Combining the new photometry in the mid-IR with published SEDs in the optical and near-IR, as well as further DUSTY modeling, we will be able to develop a complete picture of the mass-loss histories in each of these clusters. The SOFIA program will likely be completed by the end of Fall 2018.

-6°51'00" FORCAST

F11

F08 52'00" F09F05 F13 F07 F01 F12 F02

0 F14

0 F15 0

2 53'00" F06 F10 F04

54'00" F03

55'00" 18:38:04 00 37:56 52 48

2000

Figure 6.3 2MASS Ks-band image of red supergiant cluster RSGC1. The overlay indicates the 30 30 ﬁeld-of-view of SOFIA/FORCAST. ∼ ×

Finally, one common issue in high-resolution imaging is that the central star can be a factor of 104 brighter than its ejected dust. To mitigate this contrast issue, we were Chapter 6. Conclusions & Future Work 103 awarded VLT/NACO time for imaging polarimetry observations for RSGC1 and RSGC2. Polarized intensity images at 1.6 µm allows for separating the stellar PSF and scattered light from extended nebular material, providing unprecedented spatial resolution of ejecta in the near-IR. While the program was partially completed, weather and instrument issues in Fall 2017 resulted in observations of only two out of the proposed twenty targets. We will be processing the completed VLT program alongside the LBT observations to better analyze the eﬀect of PSF modeling/subtraction on extracting the faint thermal dust emission from the IR images.

In August 2018, following the publication of this thesis, I will be continuing my work with SOFIA Science Center as a post-doctoral research fellow at NASA Ames. My science there will be a natural extension of my thesis, focusing on IR instrumentation, CS ejecta from massive stars, and the interstellar medium. References

Abrahamyan, H. V., Mickaelian, A. M., & Knyazyan, A. V. 2015, Astronomy and Com- puting, 10, 99

Allen, D. A., Jones, T. J., & Hyland, A. R. 1985, ApJ, 291, 280

Anderson, E., & Francis, C. 2012, Astronomy Letters, 38, 331

Aoki, W., Tsuji, T., & Ohnaka, K. 1998, A&A, 340, 222

Asaki, Y., Deguchi, S., Imai, H., et al. 2010, ApJ, 721, 267

Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, A&A, 558, A33

Baron, F., Monnier, J. D., Kiss, L. L., et al. 2014, ApJ, 785, 46

Beauchamp, A., Moﬀat, A. F. J., & Drissen, L. 1994, ApJS, 93, 187

Biller, B. A., Close, L. M., Li, A., et al. 2005, ApJ, 620, 450

Bohren, C. F., & Huﬀman, D. R. 1983, Absorption and scattering of light by small particles

Braun, R., Thilker, D. A., Walterbos, R. A. M., & Corbelli, E. 2009, ApJ, 695, 937

Cardelli, J. A., Clayton, G. C., & Mathis, J. S. 1989, ApJ, 345, 245

Costa, E., & Frogel, J. A. 1996, AJ, 112, 2607

Cushing, M. C., Vacca, W. D., & Rayner, J. T. 2004, PASP, 116, 362

Cutri, R. M., Skrutskie, M. F., van Dyk, S., et al. 2003, 2MASS All Sky Catalog of point sources.

Danchi, W. C., Bester, M., Degiacomi, C. G., Greenhill, L. J., & Townes, C. H. 1994, AJ, 107, 1469

104 References 105

Davies, B., Figer, D. F., Kudritzki, R.-P., et al. 2007, ApJ, 671, 781

De Beck, E., Decin, L., de Koter, A., et al. 2010, A&A, 523, A18

de Graauw, T., Haser, L. N., Beintema, D. A., et al. 1996, A&A, 315, L49

de Jager, C. 1998, A&A Rev., 8, 145

de Jager, C., Nieuwenhuijzen, H., & van der Hucht, K. A. 1988, A&AS, 72, 259

Decin, L., Hony, S., de Koter, A., et al. 2006, A&A, 456, 549

Draine, B. T., & Lee, H. M. 1984, ApJ, 285, 89

Draine, B. T., & Li, A. 2007, ApJ, 657, 810

Drout, M. R., Massey, P., & Meynet, G. 2012, ApJ, 750, 97

Drout, M. R., Massey, P., Meynet, G., Tokarz, S., & Caldwell, N. 2009, ApJ, 703, 441

Egan, M. P., Price, S. D., & Kraemer, K. E. 2003, in Bulletin of the American Astronomical Society, Vol. 35, American Astronomical Society Meeting Abstracts, 1301

Ekstr¨om,S., Georgy, C., Eggenberger, P., et al. 2012, A&A, 537, A146

Fabricant, D., Fata, R., Roll, J., et al. 2005, PASP, 117, 1411

Fabricant, D. G., Hertz, E. N., Szentgyorgyi, A. H., et al. 1998, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 3355, Optical Astronomical In- strumentation, ed. S. D’Odorico, 285–296

Figer, D. F., MacKenty, J. W., Robberto, M., et al. 2006, ApJ, 643, 1166

Flower, P. J. 1977, A&A, 54, 31

—. 1996, ApJ, 469, 355

Fok, T. K. T., Nakashima, J.-i., Yung, B. H. K., Hsia, C.-H., & Deguchi, S. 2012, ApJ, 760, 65

Gehrz, R. D., & Woolf, N. J. 1971, ApJ, 165, 285

Giannone, P. 1967, ZAp, 65, 226

Gordon, M. S., Humphreys, R. M., & Jones, T. J. 2016, ApJ, 825, 50 References 106

—. 2018b, in review

Gordon, M. S., Humphreys, R. M., Jones, T. J., et al. 2018a, AJ, 155, 212

Gratier, P., Braine, J., Rodriguez-Fernandez, N. J., et al. 2010, A&A, 522, A3

Groenewegen, M. A. T. 2012, A&A, 540, A32

Groenewegen, M. A. T., Waelkens, C., Barlow, M. J., et al. 2011, A&A, 526, A162

Gustafsson, B., Edvardsson, B., Eriksson, K., et al. 2008, A&A, 486, 951

Hartman, J. D., Bersier, D., Stanek, K. Z., et al. 2006, MNRAS, 371, 1405

Harwit, M., Malfait, K., Decin, L., et al. 2001, ApJ, 557, 844

Heger, A., Fryer, C. L., Woosley, S. E., Langer, N., & Hartmann, D. H. 2003, ApJ, 591, 288

Henden, A. 2016, Journal of the American Association of Variable Star Observers (JAAVSO), 44, 84

Heras, A. M., & Hony, S. 2005, A&A, 439, 171

Herter, T. L., Adams, J. D., De Buizer, J. M., et al. 2012, ApJ, 749, L18

Herter, T. L., Vacca, W. D., Adams, J. D., et al. 2013, PASP, 125, 1393

Hiltner, W. A., & Johnson, H. L. 1956, ApJ, 124, 367

Hinz, P. M., Angel, J. R. P., Woolf, N. J., Hoﬀmann, W. F., & McCarthy, D. W. 2000, in Proc. SPIE, Vol. 4006, Interferometry in Optical Astronomy, ed. P. L´ena& A. Quirrenbach, 349–353

Hinz, P. M., Defr`ere,D., Skemer, A., et al. 2016, in Proc. SPIE, Vol. 9907, Optical and Infrared Interferometry and Imaging V, 990704

Hoﬀmann, W. F., Hinz, P. M., Defr`ere,D., et al. 2014, in Proc. SPIE, Vol. 9147, Ground- based and Airborne Instrumentation for Astronomy V, 91471O

Hoﬀmann, W. F., Hora, J. L., Fazio, G. G., Deutsch, L. K., & Dayal, A. 1998, in Proc. SPIE, Vol. 3354, Infrared Astronomical Instrumentation, ed. A. M. Fowler, 647– 658 References 107

Hora, J. L., Fazio, G. G., Allen, L. E., et al. 2004, in Proc. SPIE, Vol. 5487, Optical, Infrared, and Millimeter Space Telescopes, ed. J. C. Mather, 77–92

Humphreys, R., Gordon, M. S., & Khan, R. 2018, in American Astronomical Society Meeting Abstracts, Vol. 231, American Astronomical Society Meeting Abstracts #231, 348.13

Humphreys, R. M. 1978, ApJS, 38, 309

Humphreys, R. M., & Davidson, K. 1979, ApJ, 232, 409

—. 1994, PASP, 106, 1025

Humphreys, R. M., Davidson, K., Grammer, S., et al. 2013, ApJ, 773, 46

Humphreys, R. M., Davidson, K., Ruch, G., & Wallerstein, G. 2005, AJ, 129, 492

Humphreys, R. M., Davidson, K., & Smith, N. 2002, AJ, 124, 1026

Humphreys, R. M., Helton, L. A., & Jones, T. J. 2007, AJ, 133, 2716

Humphreys, R. M., & Lockwood, G. W. 1972, ApJ, 172, L59

Humphreys, R. M., Martin, J. C., & Gordon, M. S. 2015, PASP, 127, 347

Humphreys, R. M., Weis, K., Davidson, K., Bomans, D. J., & Burggraf, B. 2014, ApJ, 790, 48

Humphreys, R. M., Smith, N., Davidson, K., et al. 1997, AJ, 114, 2778

Humphreys, R. M., Jones, T. J., Polomski, E., et al. 2006, AJ, 131, 2105

Ivezic, Z., Groenewegen, M. A. T., Men’shchikov, A., & Szczerba, R. 1997, MNRAS, 291, 121

Jiang, B. W., Deguchi, S., & Ramesh, B. 1999, PASJ, 51, 95

Johnson, H. L. 1958, Lowell Observatory Bulletin, 4, 37

Jones, T. J., Hyland, A. R., Fix, J. D., & Cobb, M. L. 1988, AJ, 95, 158

Jones, T. J., Humphreys, R. M., Gehrz, R. D., et al. 1993, ApJ, 411, 323

Jura, M., & Kleinmann, S. G. 1990, ApJS, 73, 769 References 108

Justtanont, K., de Jong, T., Helmich, F. P., et al. 1996, A&A, 315, L217

Kaluzny, J., Stanek, K. Z., Krockenberger, M., et al. 1998, AJ, 115, 1016

Kami´nski,T., Gottlieb, C. A., Young, K. H., Menten, K. M., & Patel, N. A. 2013, ApJS, 209, 38

Khan, R., Stanek, K. Z., & Kochanek, C. S. 2013, ApJ, 767, 52

Khan, R., Stanek, K. Z., Kochanek, C. S., & Bonanos, A. Z. 2011, ApJ, 732, 43

Khan, R., Stanek, K. Z., Kochanek, C. S., & Sonneborn, G. 2015, ApJS, 219, 42

King, N. L., Walterbos, R. A. M., & Braun, R. 1998, ApJ, 507, 210

Kiss, L. L., Szab´o,G. M., & Bedding, T. R. 2006, MNRAS, 372, 1721

Knapp, G. R., & Morris, M. 1985, ApJ, 292, 640

Knapp, G. R., Rose, W. K., & Kerr, F. J. 1973, ApJ, 186, 831

Knapp, G. R., Sandell, G., & Robson, E. I. 1993, ApJS, 88, 173

Kudritzki, R. P., & Reimers, D. 1978, A&A, 70, 227

Levesque, E. M., Massey, P., Olsen, K. A. G., et al. 2005, ApJ, 628, 973

—. 2006, ApJ, 645, 1102

Lucas, R., Bujarrabal, V., Guilloteau, S., et al. 1992, A&A, 262, 491

Macri, L. M., Stanek, K. Z., Sasselov, D. D., Krockenberger, M., & Kaluzny, J. 2001, AJ, 121, 870

Marco, A., & Negueruela, I. 2013, A&A, 552, A92

Massey, P. 1998, ApJ, 501, 153

Massey, P., McNeill, R. T., Olsen, K. A. G., et al. 2007a, AJ, 134, 2474

Massey, P., Neugent, K. F., & Smart, B. M. 2016, AJ, 152, 62

Massey, P., Olsen, K. A. G., Hodge, P. W., et al. 2007b, AJ, 133, 2393

—. 2006, AJ, 131, 2478 References 109

Massey, P., Silva, D. R., Levesque, E. M., et al. 2009, ApJ, 703, 420

Mathis, J. S., Rumpl, W., & Nordsieck, K. H. 1977, ApJ, 217, 425

Matt, S., Balick, B., Winglee, R., & Goodson, A. 2000, ApJ, 545, 965

Mauron, N., & Josselin, E. 2011, A&A, 526, A156

McQuinn, K. B. W., Woodward, C. E., Willner, S. P., et al. 2007, ApJ, 664, 850

Meynet, G., Ekstrom, S., Maeder, A., et al. 2013, in Lecture Notes in Physics, Berlin Springer Verlag, Vol. 865, Lecture Notes in Physics, Berlin Springer Verlag, ed. M. Goupil, K. Belkacem, C. Neiner, F. Ligni`eres,& J. J. Green, 3–642

Meynet, G., Kudritzki, R.-P., & Georgy, C. 2015, A&A, 581, A36

Morris, M., & Jura, M. 1983, ApJ, 267, 179

Mould, J., Barmby, P., Gordon, K., et al. 2008, ApJ, 687, 230

Myers, J. R., Sande, C. B., Miller, A. C., Warren, Jr., W. H., & Tracewell, D. A. 2015, VizieR Online Data Catalog, 5145

Neugent, K. F., Massey, P., Skiﬀ, B., & Meynet, G. 2012, ApJ, 749, 177

Nieuwenhuijzen, H., & de Jager, C. 1990, A&A, 231, 134

Nieuwenhuijzen, H., De Jager, C., Kolka, I., et al. 2012, A&A, 546, A105

O’Donnell, J. E. 1994, ApJ, 422, 158

O’Gorman, E., Vlemmings, W., Richards, A. M. S., et al. 2015, A&A, 573, L1

Ossenkopf, V., Henning, T., & Mathis, J. S. 1992, A&A, 261, 567

Ott, S. 2010, in Astronomical Society of the Paciﬁc Conference Series, Vol. 434, Astronomi- cal Data Analysis Software and Systems XIX, ed. Y. Mizumoto, K.-I. Morita, & M. Ohishi, 139

Oudmaijer, R. D. 1998, A&AS, 129, 541

Oudmaijer, R. D., Groenewegen, M. A. T., Matthews, H. E., Blommaert, J. A. D. L., & Sahu, K. C. 1996, MNRAS, 280, 1062 References 110

Packham, C., Jones, T. J., Warner, C., et al. 2012, in Proc. SPIE, Vol. 8446, Ground-based and Airborne Instrumentation for Astronomy IV, 84463R

Pilbratt, G. L., Riedinger, J. R., Passvogel, T., et al. 2010, A&A, 518, L1

Pogge, R. W., Atwood, B., Belville, S. R., et al. 2006, in Proc. SPIE, Vol. 6269, Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, 62690I

Poglitsch, A., Waelkens, C., Geis, N., et al. 2010, A&A, 518, L2

Price, S. D., Smith, B. J., Kuchar, T. A., Mizuno, D. R., & Kraemer, K. E. 2010, ApJS, 190, 203

Rachford, B. L., Snow, T. P., Destree, J. D., et al. 2009, ApJS, 180, 125

Rayner, J. T., Toomey, D. W., Onaka, P. M., et al. 2003, PASP, 115, 362

Reimers, D. 1975, Memoires of the Societe Royale des Sciences de Liege, 8, 369

Richards, A. M. S., Yates, J. A., & Cohen, R. J. 1999, MNRAS, 306, 954

Richards, A. M. S., Impellizzeri, C. M. V., Humphreys, E. M., et al. 2014, A&A, 572, L9

Savage, B. D., & Jenkins, E. B. 1972, ApJ, 172, 491

Schuster, M. T. 2007, PhD thesis, University of Minnesota

Schuster, M. T., Humphreys, R. M., & Marengo, M. 2006, AJ, 131, 603

Schuster, M. T., Marengo, M., Hora, J. L., et al. 2009, ApJ, 699, 1423

Scowcroft, V., Bersier, D., Mould, J. R., & Wood, P. R. 2009, MNRAS, 396, 1287

Shenoy, D., Humphreys, R. M., Jones, T. J., et al. 2016, AJ, 151, 51

Shenoy, D. P., Jones, T. J., Packham, C., & Lopez-Rodriguez, E. 2015, AJ, 150, 15

Shenoy, D. P., Jones, T. J., Humphreys, R. M., et al. 2013, AJ, 146, 90

Sholukhova, O., Bizyaev, D., Fabrika, S., et al. 2015, MNRAS, 447, 2459

Skemer, A. J., & Close, L. M. 2011, ApJ, 730, 53

Skemer, A. J., Close, L. M., Hinz, P. M., et al. 2008, ApJ, 676, 1082

Skrutskie, M. F., Cutri, R. M., Stiening, R., et al. 2006, AJ, 131, 1163 References 111

Skrutskie, M. F., Jones, T., Hinz, P., et al. 2010, in Proc. SPIE, Vol. 7735, Ground-based and Airborne Instrumentation for Astronomy III, 77353H

Smartt, S. J. 2015, PASA, 32, e016

Smartt, S. J., Eldridge, J. J., Crockett, R. M., & Maund, J. R. 2009, MNRAS, 395, 1409

Smith, B. J., Price, S. D., & Baker, R. I. 2004, ApJS, 154, 673

Smith, N., Humphreys, R. M., Davidson, K., et al. 2001, AJ, 121, 1111

Speck, A. K., Barlow, M. J., Sylvester, R. J., & Hofmeister, A. M. 2000, A&AS, 146, 437

Stephenson, C. B. 1990, AJ, 99, 1867

Stothers, R. B., & Chin, C.-w. 2001, ApJ, 560, 934

Suh, K.-W. 1999, MNRAS, 304, 389

Thompson, T. A., Prieto, J. L., Stanek, K. Z., et al. 2009, ApJ, 705, 1364

Tiﬀany, C., Humphreys, R. M., Jones, T. J., & Davidson, K. 2010, AJ, 140, 339

Treﬀers, R., & Cohen, M. 1974, ApJ, 188, 545

Vacca, W. D., Cushing, M. C., & Rayner, J. T. 2003, PASP, 115, 389

Valeev, A. F., Sholukhova, O. N., & Fabrika, S. N. 2010, Astrophysical Bulletin, 65, 140

van Loon, J. T., Cioni, M.-R. L., Zijlstra, A. A., & Loup, C. 2005, A&A, 438, 273

Vink, J. S. 2012, in Astrophysics and Space Science Library, Vol. 384, Eta Carinae and the Supernova Impostors, ed. K. Davidson & R. M. Humphreys, 221

Vlemmings, W., Diamond, P. J., & van Langevelde, H. J. 2001, A&A, 375, L1

Vlemmings, W. H. T., van Langevelde, H. J., & Diamond, P. J. 2005, A&A, 434, 1029

Vlemmings, W. H. T., Khouri, T., Mart´ı-Vidal,I., et al. 2017, A&A, 603, A92

Wittkowski, M., Hauschildt, P. H., Arroyo-Torres, B., & Marcaide, J. M. 2012, A&A, 540, L12

Woolf, N. J., & Ney, E. P. 1969, ApJ, 155, L181

Wright, E. L., Eisenhardt, P. R. M., Mainzer, A. K., et al. 2010, AJ, 140, 1868

Zhang, B., Reid, M. J., Menten, K. M., & Zheng, X. W. 2012, ApJ, 744, 23