Source Classification

We classify the candidates either as stellar or non-stellar based on their photometric properties. We focus on identifying two tell-tale signatures of the SED of a luminous star obscured by warm circumstellar dust -- low optical fluxes or flux-limits compared to the mid-IR luminosities and signs of the SEDs turning over between $8\,\micron$ and $24\,\micron$. Towards longer wavelengths, emission from warm circumstellar dust should peak between the IRAC $8\,\micron$ and MIPS $24\,\micron$ bands. It is almost impossible for mass lost from a single star to to both have a significant optical depth and a dust temperature cold enough to peak at wavelengths longer than $\sim24\,\micron$. Such systems are almost certainly star clusters with significant amounts of cold dust. Therefore, any SED that appears to have a steep slope between $8$ and $24\,\micron$ is considered to be a likely cluster, rather than a single dust obscured star. Frequently, these sources are also too luminous to be a single star. At the shorter wavelengths, we expect a dusty star to have relatively lower luminosity compared to its mid-IR luminosity and redder optical colors.

We examine the HST $B-V$/$V$ and the $V-I$/$V$ color magnitude diagrams (CMDs) for each source for which HST data is available. The presence of a very red optical counterpart or the absence of a luminous star supports the existence of significant dust obscuration. On the other hand, the presence of a blue or bright optical counterpart makes it likely that the source is a star cluster, a background galaxy/AGN, or a foreground star. We first search for bright and/or red optical sources within the $0\farcs3$ matching radius that can be the obvious counterpart of the bright and red IRAC source. Next, if multiple bright and/or red optical matches are found, we identify the best astrometric match to the IRAC location. Finally, if no reasonable match is found, we adopt the flux of the brightest of the nearby sources as a conservative upper limit on the optical luminosity of the candidate.

To demonstrate these, we discuss the case of M81-12 in detail. M81-12 has a steeply rising optical and mid-IR SED (Figure1) with two distinct peaks -- one in the near-IR, between the $R$-band and 3.6µ, and another in the mid-IR between 8 and 24µ. Figure4 shows the HST optical CMD for sources near the location of M81-12. Besides the sources within the $0\farcs3$ matching radius, it also shows all sources within $0\farcs3 - 2\farcs0$ of the candidate using a different symbol to emphasize the absence of any other unusual nearby sources. We detect a very red ($B=23.95$, $V=21.98$, $I=19.07$, $B-V\simeq2$, $V-I\simeq2.9$) HST counterpart with an excellent astrometric match ($<0\farcs1$, Figure2) to the IRAC position. This source is the brightest, red HST point source within $2\farcs0$ of the IRAC location (Figure4) and so we define it to be the counterpart used in the SED. The LBT $V$ and $R$ band light curves show a variable source with the correlated irregular variability ($\sim$0.4mag, Figure5) typical of many evolved massive stars (e.g., Kourniotis et al.2014). Based on the SED shape and the unambiguous detection of a red, variable optical counterpart, we conclude that M81-12 is a massive, dust-obscured, single star.

In addition to ObjectX (M33-1), we identified 17 additional dust obscured stars and classified 16 others as non-stellar. We left one source (N7793-12) unclassified due to a lack of sufficient optical data (it falls on an HST/ACS chip gap). It could well be a dusty star, but we do not discuss it further.