PHAT STELLAR CLUSTER SURVEY. I. YEAR 1 CATALOG AND INTEGRATED PHOTOMETRY

Large, high-quality samples of stellar clusters provide key data for studies of a wide variety of astrophysical topics, including cluster evolution, stellar evolution, star formation, and galaxy evolution. However, current cluster samples suffer from serious limitations. For example, Milky Way clusters suffer from severe dust attenuation within the Galactic plane, resulting in disk cluster samples that are complete only within a small region (radius of ∼1 kpc) around the Sun (e.g., Dias et al. 2002; Piskunov et al. 2008). This limits the variety of objects and galactic environments explored, as exemplified by the dearth of massive, intermediate-age clusters known within the Milky Way (e.g., Davies et al. 2011). Although infrared surveys of the Galactic plane are improving this situation (e.g., Dutra et al. 2003; Mercer et al. 2005; Borissova et al. 2011), current samples of Galactic clusters do not probe the full stellar cluster parameter space, limiting our ability to study mass, age, and environmental dependencies of evolutionary processes.

One solution to the incompleteness of Galactic samples has been to extend cluster studies to other galaxies. Extragalactic cluster samples have grown immensely over the past decade, thanks in part to the power of Hubble Space Telescope (HST) imaging. These analyses probe a variety of galactic environments, ranging from starbursting galaxy mergers (e.g., Whitmore et al. 1999) to quiescent spirals (e.g., Larsen 2002), producing provocative results concerning the environmental dependence of cluster formation and evolution (e.g., Larsen & Richtler 2000; Goddard et al. 2010). However, even with HST, clusters in most distant galaxies appear as marginally resolved single objects. This increases the difficulty and uncertainty associated with basic steps in cluster analysis, such as object identification, photometry, and the derivation of ages and masses. As a result, the interpretation of underlying cluster evolutionary processes has considerable associated uncertainty, even leading to cases of conflicting interpretations derived from the same data set (e.g., in M83; Chandar et al. 2010b; Bastian et al. 2012; Fouesneau et al. 2012).

Closer to the Milky Way, large cluster catalogs exist for the Large and Small Magellanic Clouds (LMC and SMC; Bica et al. 1999, 2008; Bica & Dutra 2000; Hunter et al. 2003). In the Clouds, the ability to resolve clusters into individual stars has resulted in a number of important results in stellar evolution (e.g., Chiosi et al. 1989; Frogel et al. 1990; Girardi et al. 2009), cluster evolution (e.g., Gieles & Bastian 2008; Chandar et al. 2010a), and cluster formation (e.g., Mackey et al. 2008). However, there are limitations associated even with this excellent sample of objects. On a practical level, while these cluster samples do not suffer from the same incompleteness issues as those from the Milky Way, their piecemeal assembly by many different groups has imprinted a complex and little-understood selection function. On a more fundamental level, there are additional limits to the applicability of stellar and cluster evolution results derived from these interacting, relatively low-mass galaxies. The LMC and SMC do not probe the range in galactic environments that are characteristic of a majority of the baryonic universe. More than 75% of all stars in the universe today have metallicities within a factor of two of the solar value (Gallazzi et al. 2008), higher than those probed by the Clouds. No cluster sample comparable to those in the Clouds is currently available in a large spiral galaxy.

The neighboring galaxy M31 (the Andromeda galaxy) is a prime target for stellar cluster studies. The galaxy's proximity allows resolution of individual bright stars in clusters and the robust detection of clusters down to faint (<10⁴ L_☉) luminosities. M31 also provides access to a range of widely varying environments across the extent of the star-forming disk. Andromeda's role as a valuable laboratory was realized long ago, and decades of work have gone into exploring its cluster population. We defer the detailed review of existing cluster catalogs until Section 5, but beginning with Hubble (1932), previous work has mainly focused on M31's globular cluster population through ground-based imaging (e.g., Galleti et al. 2004; Peacock et al. 2010). The proximity of M31 also enables high-resolution spectroscopic follow-up of bright clusters (e.g., Colucci et al. 2009; Strader et al. 2011), as well as low-resolution spectroscopy of intrinsically faint clusters that are inaccessible in distant galaxies (e.g., Caldwell et al. 2009). Numerous studies have utilized HST's excellent spatial resolution to study massive clusters (e.g., Barmby & Huchra 2001; Perina et al. 2009a) and their individual resolved stars (e.g., Rich et al. 2005; Mackey et al. 2006; Perina et al. 2009b, 2011), as well as to identify and study less massive disk clusters (e.g., the Hodge–Krienke Catalogs; Krienke & Hodge 2007, 2008; Hodge et al. 2009, 2010, hereafter the HKC). While extremely valuable, these previous space-based observing programs have been limited to focused studies of a small number of targeted regions, as opposed to a wide-ranging survey to obtain a broad sampling of the galaxy and an overall assessment of the M31 cluster population.

The Panchromatic Hubble Andromeda Treasury (PHAT; Dalcanton et al. 2012) is poised to revolutionize the study of stellar clusters in M31. This ongoing HST multi-cycle program will image one-third of the M31 disk at high spatial resolution, with wavelength coverage from the ultraviolet through the near-infrared. In terms of cluster studies, this survey provides a number of distinct advantages over existing work. High spatial resolution imaging allows us to resolve clusters into individual stars, permitting detailed characterization of their stellar populations through analysis of their color–magnitude diagrams (CMDs). Figure 1 demonstrates the data quality provided by PHAT for a previously unidentified cluster, showing the considerable gain over existing ground-based surveys. The wide wavelength coverage enables accurate age dating, as well as the ability to probe a broad range of stellar effective temperatures, from massive main-sequence stars to evolved supergiants and asymptotic giant branch stars. As a result, these objects provide a wealth of valuable knowledge in terms of calibrating and refining stellar evolution models at high metallicity. Finally, these images allow the detection of faint clusters, reliably probing more than ∼2 mag further down the luminosity function than was previously possible using ground-based data sets. The PHAT survey will increase the HST spatial coverage of the M31 searched for clusters by a factor of ∼3 (1800 arcmin² in total compared to 650 arcmin² surveyed in the HKC; 390 arcmin² in the current Year 1 data set). However, this metric underestimates the scientific gain provided by high-quality, uniform PHAT data products, as opposed to the heterogeneous archival data used in the HKC work. Fundamentally, the PHAT survey represents the shift from a discrete, targeted mode of cluster study to a broad survey mode, allowing for comprehensive analysis of cluster evolutionary processes and its environmental dependencies.

Zoom In Zoom Out Reset image size

The stellar clusters identified as part of the PHAT survey will constitute the most comprehensive sample of clusters available for any large spiral galaxy. The large range of galactocentric radius (0–20 kpc) included in the survey spans a wide range of star formation intensities and gas densities. The diversity of galactic environments will be important for testing models of cluster formation and evolution. In addition, the simultaneous accessibility of objects over a >3 order of magnitude range in cluster luminosity provides a top-to-bottom view of the cluster population, given that we sample a continuous range of objects that extend from those equivalent to Galactic open clusters up to massive globular clusters.

This paper is the first in a series utilizing the PHAT data set for studies of stellar clusters. Here, we present the first installment of an HST-based cluster catalog, which will serve as the basis for extensive study of Andromeda's cluster population. Catalog updates and improvements will be published over the course of this four-year observing program. In this edition, we publish positions, sizes, and integrated photometry for the Year 1 cluster sample. Age and mass determinations derived from the integrated photometry will be presented in M. Fouesneau et al. (2012, in preparation). Additional studies, including analysis of structural parameters, resolved star content, and integrated spectroscopy of the cluster sample will follow in subsequent work.

We summarize the PHAT observations in Section 2, while in Section 3 we describe our cluster identification procedures, present results from completeness testing, and introduce the Year 1 cluster catalog. Next, we describe and test our photometry methodology in Section 4, followed by a comparison between the PHAT cluster catalog and existing catalogs in Section 5. We present a basic characterization of the cluster catalog contents in Section 6, followed by discussions of luminosity functions, the cluster size distribution, and objects of interest in Section 7. We conclude with a summary and description of future work in Section 8. Throughout this work, we assume a distance modulus for M31 of 24.47 (785 kpc; McConnachie et al. 2005), for which 1'' corresponds to a physical size of 3.81 pc.

This paper includes clusters identified in Year 1 PHAT imaging data, taken before 2011 June. A full description of the PHAT observational design is available in Dalcanton et al. (2012), but we briefly summarize relevant details below. PHAT observations are grouped into 23 area units known as "bricks," each made up of 18 mosaiced HST fields of view in a rectangular 6 × 3 arrangement that covers a ∼12' × 65 region of sky. Data are obtained simultaneously at different field centers with the Advanced Camera for Surveys (ACS; using the F475W and F814W filters) and Wide Field Camera 3 (WFC3; using the F275W, F336W, F110W, and F160W filters) instruments in two epochs, separated by ∼6 months. During each epoch, imaging is obtained by the cameras in two side-by-side, half-brick (3 × 3) arrays. Between the epochs, the orientation of the cameras change by 180° due to the annual roll angle variation of HST. As a result, the half-brick mosaic obtained by the ACS camera during the first epoch is now imaged by the WFC3 camera in the second epoch, and vice-versa, completing six-filter imaging across the brick. In all, we obtain ∼130 minutes of exposure time at each of the brick's 18 field centers.

The Year 1 imaging used in this work includes four full bricks (designated B01, B09, B15, and B21) and the western halves of two additional bricks (B17W and B23W). These data sample locations along the major axis of M31 from the center out to a projected radius of ∼20 kpc. The full PHAT survey footprint, along with the locations of the Year 1 bricks are presented in Figure 2.

Zoom In Zoom Out Reset image size

Combining imaging data across multiple pointings from three separate cameras (ACS/WFC, WFC3/UVIS, and WFC3/IR) requires astrometry with higher precision than that obtained using the standard telescope telemetry and data processing pipeline. Images of neighboring fields have sufficient overlap to enable us to derive an astrometric solution across a full (or half) brick. These astrometric solutions are obtained separately for each camera using photometric catalogs derived using DOLPHOT,¹⁸ a modified version of HSTPhot (Dolphin 2000) that has been updated to include specialized ACS and WFC3 modules. Affine distortion corrections (in addition to those already known for each camera from the IDCTABs) are required to obtain consistent brick-wide astrometric solutions. An additional correction brings the brick-wide astrometric solutions onto a global astrometric frame, defined using Canada–France–Hawaii Telescope observations tied to Two Micron All Sky Survey (2MASS) catalogs (Skrutskie et al. 2006). The global astrometric alignment agrees with that of the 2MASS reference system within an absolute level of ∼60 mas.

Once aligned, we use the multidrizzle task within PyRAF (Koekemoer et al. 2002) for cosmic-ray rejection and image creation along with lacosmic (van Dokkum 2001) for supplemental cosmic-ray flagging. We note that cosmic-ray correction is particularly difficult in the case of the WFC3/UVIS data due to a large number of cosmic-ray artifacts (particularly in F275W images) and the availability of only two frames for artifact detection in regions of non-overlapping field coverage. Pixel scales of the resulting images are (0.04, 0.05, and 0.065 arcsec pixel⁻¹) for the (WFC3/UVIS, ACS/WFC, WFC3/IR) cameras, where the WFC3/IR images are up-sampled from their native plate scale (0.128 arcsec pixel⁻¹) to take advantage of the higher effective resolution afforded by the survey's sub-pixel dither strategy.

The goal of the PHAT cluster survey is to identify and analyze a sample of gravitationally bound star clusters in M31. In this paper, we undertake the first step toward achieving this goal: the visual inspection of Year 1 PHAT imaging to identify candidate bound clusters. As shown in Figure 1, clusters appear in PHAT imaging as composite objects composed of a centrally concentrated overdensity of individual resolved stars and a broad unresolved light component. The ability to resolve these objects into individual stars allows for the clean separation between genuine stellar clusters and contaminants such as background galaxies or single stars.

Although PHAT imaging facilitates robust identifications of stellar overdensities, determining whether or not these clusterings are gravitationally bound is challenging. Quantitative assessment of an object's boundedness requires age, mass, and spatial profile information (e.g., Gieles & Portegies Zwart 2011). Further, determining the boundedness of young objects (≲10 Myr) is made even more difficult because dynamical evolution has had little time to evolve stellar structures from an initial hierarchical, scale-free spatial distribution (for further discussion, see, e.g., Bastian et al. 2011 and references therein), blurring the distinction between bound and unbound stellar groupings. In future work, we will utilize age, mass, and structural characteristics to assess boundedness for each object (e.g., using the Π statistic; Gieles & Portegies Zwart 2011), but that analysis is beyond the scope of the initial classification work presented here.

Given the current limitations in assessing an object's boundedness, we adopt a liberal approach for cluster candidate identification. We prioritize sample completeness over purity and therefore include all cluster-like stellar overdensities as part of the object catalog. We acknowledge that the cluster catalog presented here will likely include both bound and unbound groupings of stars, particularly among the youngest objects. As a result, throughout this work all objects are formally considered cluster candidates, although for brevity we will refer to them simply as clusters.

With the intent of the cluster search well defined, we proceed to a description of our search methodology and a presentation of the cluster search results in Section 3.1. We conclude our discussion of cluster identification by characterizing the completeness characteristics of the sample in Section 3.2.

3.1. By-eye Search

We undertake a systematic by-eye search of the Year 1 PHAT images following a precedent set by previous M31 cluster studies (e.g., Barmby & Huchra 2001; Krienke & Hodge 2007). Drawing on this rich history of visual cluster identification, we make a concerted effort to improve upon previous work through the use of uniform analysis techniques, redundancy, cross-validation, and improved characterization of the resulting selection function.

When using any method of cluster identification, manual or automated, it is important to understand the biases and limitations inherent to the technique. We use artificial cluster tests in order to assess the completeness characteristics of our search methodology, which we discuss in detail in Section 3.2. We are also developing automated methods of cluster identification to further reduce subjectivity of the identifications and enable more robust completeness testing in future work with this data set (K. A. G. Olsen et al. 2012, in preparation). The cluster sample presented here will act as an important comparison sample to help refine these automated search techniques for use with the PHAT data set.

Our by-eye search consists of two stages: an initial search of the available imaging for all viable cluster candidates, followed by a re-evaluation of each preliminary candidate in a systematic manner. The completeness and accuracy of this process are enhanced by the redundancy of eight experienced astronomers conducting each stage of the search.

To perform the initial image search, the survey area is subdivided using the footprint of the WFC3-IR camera (23 × 21), providing 18 contiguous, minimally overlapping search fields within each brick. Each field is searched by three or four team members in a "blind" manner, meaning that these individuals are not provided with the locations of clusters previously identified by other PHAT team members or previous surveys. To identify cluster candidates, searchers use a suite of images that include a two-band optical color image, all six available single-band images spanning from the UV to the NIR, and two star-subtracted optical images used to identify diffuse emission that makes up a cluster's unresolved light. These images are accessed using a custom image viewer that allows users to switch between the available spatially aligned, full resolution images, as well as the ability to alter brightness and contrast levels for optimal visualization.

Once candidate objects are selected in the initial visual search, we cross-match and combine the identifications of all team members. Based on this preliminary catalog, we perform initial photometry and create cutout images for each cluster candidate. The preliminary catalog is reviewed independently by all eight team members on an interactive Web site, and each individual assigns scores for each candidate, providing an assessment of the likelihood that an object is a cluster. Scores (S_by-eye) are based on a scale of 1-to-3, where S_by-eye = 1 represents a definite cluster, S_by-eye = 2 signifies a likely cluster, and S_by-eye = 3 represents an unlikely or non-cluster object. We present examples of the cluster scoring system in Figure 3.

Zoom In Zoom Out Reset image size

We average the scores from all team members for each object and use these average rankings to divide the candidates into three subsamples: clusters, possible clusters, and unlikely objects. We choose thresholds of S_by-eye < 2.0 for clusters, 2.0 ⩽ S_by-eye < 2.5 for possible clusters, and S_by-eye ⩾ 2.5 for unlikely objects. We discard unlikely objects from the catalog, while retaining clusters and possible clusters in two separate catalogs. The average scores for each cluster are provided as part of the cluster catalogs as an assessment of candidate quality. To determine the reliability of these average scores, we built a ranking experiment into our classification procedure. During the course of the ranking work, 24 objects appeared twice within the preliminary catalog. As a result, these clusters were each ranked two separate times by each team member. When the resulting average scores of the duplicate entries are compared, we find the standard deviation of the 1-to-3 ranking differences to be 0.27, showing good consistency and repeatability for the scores provided by our search team.

The Year 1 cluster search yielded a catalog of 601 high-scoring clusters. Table 1 presents positions for each object, as well as other descriptive information (radii, photometric measurements) that will be described in Section 4. Cutout images for each object are presented in Figure 4. In addition, the spatial distribution of the clusters are shown in Figure 2. Tabulated information and image cutouts for 237 possible clusters are presented in Table 2 and Figure 5, respectively. Information about the possible clusters is provided for completeness, but due to the uncertain nature of their classifications, we exclude these objects from further analysis.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

View figure set (31 panels)

Tarball of figure set images

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

View figure set (12 panels)

Tarball of figure set images

Table 1. PHAT Year 1 Cluster Catalog

PC ID	R.A.	Decl.	Rap	Reff	F275W	σ	F336W	σ	F475W	σ	F814W	σ	F110W	σ	F160W	σ	Sby-eye	ApCora	Alternate Name
	(J2000)	(J2000)	('')	('')
1	11.638274	42.193887	1.57	0.68	17.54	0.02	17.68	0.03	18.82	0.02	18.34	0.04	17.95	0.07	17.73	0.12	1.00	−0.12	Hodge10-85
2	11.637139	42.209936	2.51	0.64	15.60	0.02	15.91	0.02	17.33	0.01	17.23	0.04	17.39	0.19	17.30	0.35	1.43	−0.00	Hodge10-84
20	11.630550	42.200631	1.10	0.48	⋅⋅⋅	⋅⋅⋅	22.55	0.25	21.97	0.63	19.83	0.07	19.16	0.23	18.63	0.39	1.00	−0.13	⋅⋅⋅
21	11.631591	42.199991	1.00	0.52	19.60	0.02	19.82	0.01	20.89	0.10	20.98	0.28	22.88	1.80	⋅⋅⋅	⋅⋅⋅	1.57	−0.21	⋅⋅⋅
22	11.630849	42.201656	0.75	0.46	23.38	0.36	22.52	0.11	22.22	0.06	21.39	0.67	22.47	1.72	⋅⋅⋅	⋅⋅⋅	1.71	−0.31	⋅⋅⋅

Notes. The sample presented here consists of objects classified as clusters, with S_by-eye < 2.0. ^aAperture Corrections are provided such that m_Total = m_Aperture + ApCor.

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

Table 2. PHAT Year 1 Possible Cluster Catalog

PC ID	R.A.	Decl.	Rap	Reff	F275W	σ	F336W	σ	F475W	σ	F814W	σ	F110W	σ	F160W	σ	Sby-eye	ApCora	Alternate Name
	(J2000)	(J2000)	('')	('')
4	11.664001	42.192391	2.10	0.83	19.37	0.06	19.43	0.05	20.38	0.02	19.80	0.14	19.42	0.22	18.63	0.10	2.43	−0.09	⋅⋅⋅
6	11.668817	42.197452	1.13	0.32	21.50	1.01	20.93	0.13	21.86	0.29	21.97	1.72	21.57	0.69	21.71	3.33	2.29	−0.01	⋅⋅⋅
7	11.675393	42.193555	0.80	0.29	⋅⋅⋅	⋅⋅⋅	25.59	2.19	22.63	0.14	20.96	0.13	20.38	0.11	19.73	0.08	2.29	−0.07	⋅⋅⋅
11	11.648244	42.227958	1.00	0.38	25.42	1.35	23.98	0.24	22.56	0.15	20.12	0.12	18.77	0.03	17.82	0.07	2.43	−0.08	⋅⋅⋅
12	11.661360	42.190571	1.05	0.31	21.39	0.16	21.02	0.05	21.38	0.04	20.74	0.06	20.82	0.13	21.39	1.90	2.00	−0.02	⋅⋅⋅

Notes. The sample presented here consists of objects classified as possible clusters, with 2.0 ⩽ S_by-eye < 2.5. ^aAperture corrections are provided such that m_Total = m_Aperture + ApCor.

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

During the cluster search, we also identified 370 putative background galaxies. We did not explicitly search for these objects, therefore this catalog does not constitute a complete sample of objects. However, the potential usefulness of these identifications (e.g., as astrometric references, multi-wavelength source catalog cross-correlation) warrants its release. We present this catalog of objects in Appendix A.

3.2. Catalog Completeness

To characterize the completeness of our cluster sample, we conduct artificial cluster tests that mimic the selection procedure of the clusters. We use artificial clusters that span the range of ages and masses we expect to find in the cluster sample, while the size distribution is chosen to sample the minimum, average, and maximum sizes of the true sample (see Section 7.2). Ages and masses are chosen from a logarithmic grid of values, while sizes are drawn from the set of three characteristic values. Specifically, we select ages ranging from 4 million to 10 billion years, masses ranging from 10² to 10⁵ M_☉, and profiles that have effective radii (R_eff; equivalently, half-light radii) of 1, 3 or 7 pc (0.26, 0.79, or 1.84 arcsec).

We create artificial clusters by populating a Padova isochrone (Girardi et al. 2010) of the appropriate age using a Kroupa (2001) stellar initial mass function. Next, the stars are spatially distributed according to a King (1962) profile, assuming no mass segregation. Finally, the size of the cluster and magnitudes of the individual stars are scaled appropriately to account for the distance of M31, assuming zero Galactic foreground or other internal M31 extinction. For each test field, we insert 46 artificial clusters into individual raw (FLT) images using new functionality developed for the DOLPHOT photometry package. The clusters are randomly positioned within each image to prevent search bias (as would result from a regular grid pattern), though we ensure that clusters are well separated within the image so that they do not overlap or interfere with the photometry of other artificial clusters. We drizzle the resulting images together to create searchable images in the optical F475W and F814W passbands. In all, we created three fields of artificial clusters for each of the four full bricks in the Year 1 data set, resulting in a total of ∼550 simulated objects. In addition to the completeness tests that follow, these artificial clusters are also used for quality assurance of our photometric results, as discussed in Section 4.2.

After creating the artificial cluster images, we identify clusters in the same way we searched the original images, with all eight team members searching each field and rating the reliability of the clusters. By comparing the resulting cluster identifications to the full list of inserted clusters, we can estimate the completeness limits of the sample based on various cluster input parameters; the results of this analysis are shown in Figure 6. The left panel shows the best estimate of the sample's characteristic cluster recovery fraction, computed using a subsample of 110 "typical" simulated clusters with R_eff of 3 pc distributed within the outer three bricks (B09, B15, and B21). We estimate a 50% completeness limit of m_ap ∼21.2. Considering distance and Galactic foreground dust reddening (E(B − V) = 0.062; Schlegel et al. 1998), this translates to an absolute magnitude limit of M_ap ∼ −3.5.

Zoom In Zoom Out Reset image size

The true completeness of the sample as a whole, however, is a complicated function of cluster luminosity, size, and location within M31. Using the full complement of simulated clusters, we explore the dependence of completeness on the latter two cluster properties. The middle panel of Figure 6 shows that variations in galactocentric position result in a ∼1 mag difference in completeness limits. Detection limits are brighter for clusters located in the bulge-dominated inner galaxy due to luminous, crowded background fields. Cluster size also plays a role, in which increasing size at constant luminosity results in reduced surface brightness and lower detection efficiency. The right panel of Figure 6 shows that there is a ∼1 mag difference in completeness when comparing clusters with R_eff of 1 versus 7 pc. To account for these multiple dependencies, we will undertake a larger and more rigorous set of completeness tests in future work to better characterize the subtleties of the completeness function.

We use aperture photometry to measure the six-band integrated fluxes of each cluster. Aperture photometry consists of two main analysis tasks: defining an aperture (center and size) and determining the local background flux level. We build upon photometry procedures used in M31 by Barmby & Huchra (2001) and Krienke & Hodge (2007), with several refinements related to the assessment of local background levels and accounting for light that lies outside the photometric aperture. A detailed description of photometric analysis procedures and results is provided in Section 4.1, followed by artificial cluster validation analysis in Section 4.2. Photometry zero points for the ACS and WFC3 cameras were obtained from the STScI Web site¹⁹ and are listed in Table 3. All photometry is presented in the Vega magnitude system using the native HST passbands; we do not perform passband conversions.

4.1. Aperture Photometry Procedure

The first step in aperture photometry is to define an aperture center. Centers are estimated by centroiding on an F475W image that is smoothed using a 03 (6 pixel) FWHM Gaussian kernel. For certain clusters, particularly low-luminosity objects whose light is dominated by a small number of bright sources, the flux-weighted positions determined by the automated procedure do not always accurately reflect the cluster center. For this reason, we visually inspect central positions and manually adjust incorrect determinations.

We adopt a photometric aperture size that provides the largest signal-to-noise ratio for flux measurement by enclosing a maximum amount of cluster light while including as little background light as possible. The clusters considered in this study vary by a factor of ∼10 in radius, and consequently, the chosen photometric aperture radii vary by the same amount. We define circular apertures using growth curve analysis to determine an appropriate radius. The aperture limit is defined at the radius where the cluster profile drops below the level of the noise in the background, equivalent to the point at which the curve of growth turns over and the increase in cumulative flux as a function of radius stops. An illustrative example of a cluster image and growth curve is provided in Figure 7. The aperture radii (R_ap) are determined by visual inspection of the growth curves for each cluster and are reported with the photometric results. While it would be preferable to adopt an algorithmic approach for defining R_ap, the relatively noisy character of the local background significantly complicates automated determinations. We perform aperture definition and growth curve analysis on the F475W image, which provides the best combination of signal to noise and contrast between cluster and field populations for a wide range of cluster ages. Apertures of the same angular size are used for the five other images. Photometric aperture radii range between 05 and 6'' for the Year 1 sample.

Zoom In Zoom Out Reset image size

Aperture photometry depends greatly on the determination of the background flux level. Following traditional photometric terminology, we also refer to non-cluster background light as the "sky" or the "sky background." For the PHAT cluster sample, the sky is made up of two components: individual resolved stars and unresolved light. Traditionally, the background is determined using the mode of sky region pixel values. However, the fact that resolved stars are a true component of the background light calls for an alternate statistical treatment.

We define 10 annular sky regions that encircle the photometric aperture, extending radially from 1.2 × R_ap to 3.4 × R_ap. Each of the annuli are equal in area to the photometric aperture, to accurately measure the dispersion of the sky background based on an equal number of pixel samples. Figure 7 shows an example of the aperture layout, where we denote the inner and outer extent of the sky measurement annuli. Next, we calculate the total integrated flux within each of the sky regions. We perform iterative 2σ rejection on these 10 fluxes, thereby excluding regions that contain bright stars or other objects not representative of the typical sky background. We adopt the mean of the non-rejected sky fluxes as the sky background value and the standard deviation of these fluxes as a measure of the uncertainty in the sky background determination. We propagate the uncertainty from the sky level determination into the overall cluster photometry by adding it in quadrature with the cluster's flux measurement uncertainty.

In agreement with the HKC cluster studies, the uncertainty in the sky background determination dominates the overall uncertainty in cluster magnitudes. When compared to previous ground-based cluster photometry in M31 or other HST-based extragalactic cluster photometry, our magnitude uncertainties appear larger. Our inclusion of sky level uncertainty into the reported values accounts for these larger overall magnitude uncertainties. Uncertainty in the determination of cluster fluxes, comparable to the errors reported in other cluster catalogs, never rise above 0.01 mag for any object in the Year 1 catalog.

The resulting magnitudes measured within R_ap, hereafter referred to as aperture magnitudes, are presented in Table 1 for the cluster sample and Table 2 for the possible cluster sample. These aperture magnitudes represent high signal to noise, spatially matched measurements of cluster light, and are optimal for calculating cluster colors. We plot the photometric uncertainties as a function of magnitude for the cluster sample in Figure 8. To assess relative reliability in the six passbands, we count the number of well-determined (σ < 0.5 mag) photometric measurements in each band and record the results in Table 4. F475W imaging provides the highest-quality measurements, as shown by that filter's small photometric uncertainties, followed by the F336W image which probes a similar wavelength regime. The F814W image has increased levels of photometric error due to the reduced contrast between cluster and field populations. The quality of the measurements is lower in the three remaining filters due to intrinsic wavelength-dependent limitations; the F275W measurements suffer from low signal to noise for all but the youngest, bluest clusters; while the F110W and F160W measurements suffer from high sky background levels and poor cluster-field contrast. In addition to the problem of faint signal, we note that ∼10% of F275W magnitudes (and F336W magnitudes, but at a lower level) are affected by cosmic-ray artifacts. While a vast majority of these defects are adequately corrected for by our image processing, we caution that a few percent of the UV magnitudes might still be adversely affected.

Zoom In Zoom Out Reset image size

Table 4. Passband Photometric Quality Comparison for Cluster Sample

Passband	N(Valid Measurements)	N(Well-determined Measurements)
F275W	552 (91.8%)	447 (74.4%)
F336W	590 (98.2%)	566 (94.2%)
F475W	600 (99.8%)	597 (99.3%)
F814W	593 (98.7%)	514 (85.5%)
F110W	518 (86.2%)	358 (59.6%)
F160W	472 (78.5%)	313 (52.1%)

Notes. Valid measurements denote magnitudes that result from positive fluxes (signal measured above sky level) and suffer no other failures (e.g., image artifacts). Well-determined measurements denote magnitudes where σ < 0.5 mag.

Download table as:  ASCIITypeset image

In addition to the aperture magnitudes, we also provide estimates of the effective radius (R_eff; equivalently, the half-light radius) for each cluster in Tables 1 and 2. These estimates are obtained by measuring the cluster flux profile and interpolating this curve to find the radius that contains half of the light within R_ap. We use the F475W cluster light profiles to make these assessments, given their good data quality. We discuss the resulting sizes in Section 7.2 and use them to calculate aperture corrections in the next section. We recognize that the R_eff estimates will systematically underestimate the true cluster sizes due to the fact that they are calculated using aperture magnitudes that fall short of measuring the full luminosity of the cluster. However, we expect the impact of this underestimation to be small due to the steepness of the cluster light profile.

4.1.1. Aperture Corrections

The aperture magnitudes presented above measure a majority of the cluster light. However, these values do not account for light that lies below the noise level of the sky background, beyond the limits of our photometric aperture in the faint outer wings of the cluster profile. We correct for this missing flux by using R_eff determinations to approximate the cluster's luminosity profile shape, then extrapolate this profile past the limits of our photometric aperture to make an estimate of the cluster's total light. The magnitude difference calculated between the original aperture magnitudes and this total light estimate are equivalent to an aperture correction.

To make this profile-based extrapolation, we require a cluster profile shape and a normalization for that profile. We adopt a King (1962) profile, assuming a concentration (c = R_tidal/R_core) that matches the characteristic profile of PHAT clusters, as obtained during preliminary cluster profile fitting (c = 7). Next, we use a cluster's R_eff to set the radial scaling of that characteristic King profile. Finally, we normalize the scaled profile such that the integrated flux within R_ap matches the cluster's aperture magnitude measurement. Once normalized, we calculate the fraction of flux that lies outside R_ap and transform this value into an aperture correction in magnitudes, which is presented in Tables 1 and 2 for all clusters. These corrections may be applied to the aperture magnitudes to obtain total magnitudes.²⁰ These total magnitudes are optimal for the estimation of total cluster magnitudes and luminosities. While the aperture corrections were derived in the F475W passband, they may be used for all filters under the simplifying assumption of flat radial color profiles in the outer parts of the cluster.

The amplitude of the aperture corrections are presented in Figure 9. Over the sample of clusters, the corrections vary from 0.0 to −0.6 mag, with a median correction of −0.1 mag. These corrections are negligible for the brightest clusters, where a majority of the light is detectable above the noise level of the sky background. The corrections become larger for fainter clusters due to their low cluster-to-field flux contrast. The simplifying assumption of a universal cluster profile shape that varies only as a function of R_eff provides suitable accuracy for this correction, as shown by artificial cluster tests that follow in Section 4.2. Full cluster profile and structural parameter fitting is currently underway (M. Fouesneau et al. 2012, in preparation), and these results could be used to further refine these profile extrapolations and improve upon our estimates of total cluster light in future work. However, we expect little overall improvement in photometry as a result of increased aperture correction precision because these corrections are comparable in size to the amplitude of the photometric uncertainties for most clusters.

Zoom In Zoom Out Reset image size

4.2. Artificial Cluster Photometry Experiments

We use artificial clusters to assess the uncertainties and biases associated with our photometry procedures. For these tests, we only consider clusters that lie in the outer three bricks, have R_eff of 3 pc, and were detected as part of our artificial cluster search in Section 3.2. Analysis of this sample of 76 artificial clusters provides an evaluation of photometric accuracy for "typical" clusters in the Year 1 sample. We process the simulated objects using photometric procedures identical to those described in the previous section.

The results of these tests are shown in Figure 10. The artificial cluster tests show good agreement (ΔF475W ≲ 0.1 mag) between the input and output magnitudes for bright clusters (F475W <19). For less luminous clusters, aperture magnitudes are fainter than input magnitudes by up to ∼0.4 mag at F475W ∼21.5. However, total magnitudes are more successful in recovering input magnitudes accurately, showing a smaller faintward bias of ∼0.2 mag at F475W ∼21.5. This bias represents a ∼0.5σ deviation when compared to the ∼0.4 mag scatter in the photometry. Finally, we acknowledge that these photometric experiments are slightly idealized, for instance due to our consideration of only a single family of cluster profile shapes. However, this testing confirms that our photometry techniques provide accurate assessments of cluster properties and their associated uncertainties.

Zoom In Zoom Out Reset image size

As discussed in the introduction, there is a long history of stellar cluster studies in the Andromeda galaxy. Decades of effort have produced a wealth of knowledge on this topic. To place our findings in context, in this section we cross-reference our cluster identifications with existing catalogs, allowing us to reference previous work on the same objects and compare the results of our cluster analysis to existing catalogs.

To compile a list of known clusters located within the Year 1 footprint, we began by searching the Revised Bologna Catalog²¹ (RBC; Galleti et al. 2004, last updated 2009 December to v4.0). This excellent resource has aggregated all known cluster identifications from early catalogs (e.g., Vetešnik 1962; Sargent et al. 1977; Crampton et al. 1985; Battistini et al. 1987, 1993; Barmby et al. 2000, among many others) as well as more recent works (e.g., Kim et al. 2007; Huxor et al. 2008; Caldwell et al. 2009). Other than the RBC, we searched the HKC and other works published since the most recent RBC revision (Vansevičius et al. 2009; Peacock et al. 2010; Fan et al. 2010). Our search of the HKC produced additional objects for cross-matching, however no new objects were recovered from the three other catalogs. The Peacock et al. (2010) catalog is composed solely of edits and reclassifications from an earlier version of the RBC (v3.5). Similarly, the objects studied in Fan et al. (2010) are derived directly from v4.0 of the RBC. Finally, we find no overlap between any part of the PHAT footprint (existing or planned coverage) and the southwest region of M31 studied by Vansevičius et al. (2009). In the discussion that follows, we consider cluster classifications with respect to those provided in the RBC v4.0.

The RBC and HKC contain a total of 146 published clusters, 32 cluster candidates, and 84 other non-cluster classifications (foreground stars, background galaxies, and H ii regions) that lie within the Year 1 PHAT survey footprint. We cross-match these previously known objects with all identifications made as part of the Year 1 search (cluster, possible clusters, and unlikely objects). Further, we re-examined the PHAT data at positions of previously cataloged objects that were not matched to PHAT candidates to ensure the complete re-analysis of all existing catalog entries that lie within the Year 1 footprint. In total, we classify 132 of the previously known objects as clusters, 12 as possible clusters, and reject the remaining 118 objects. We note that all confirmed Year 1 PHAT clusters were identified independently as part of our blind cluster search. Tables 1 and 2 provide cross-matched names of clusters using the naming convention of Barmby et al. (2000), consisting of the identifier from the RBC, followed by the identifier of the next most significant cluster catalog, where such exists. Additional details concerning the comparison and reclassification of existing cluster identifications are provided in Appendix B. There we provide catalog-specific commentary on the validity of the previously published classifications, as well as object-by-object classification revisions for the RBC and the HKC.

A comparison between ground-based and space-based M31 cluster catalogs reveals the importance of high spatial resolution imaging in cluster identification work. In Figure 11, we present a histogram of the number of cluster as a function of apparent magnitude for the Year 1 cluster sample of 601 clusters and the 132 previously known objects confirmed as clusters in the PHAT data. It is immediately apparent from the plot that HST imaging has allowed the PHAT cluster survey to identify hundreds of low-luminosity clusters that could not be identified from ground-based data. As shown in the inset of Figure 11, the sample completeness associated with previously known clusters discovered from ground-based data (green dashed histogram) drops precipitously at m_ap > 18 (M_F475W > −6.5). This falloff in completeness reflects the difficulty in differentiating between single unresolved stars and compact clusters in low-resolution images. In contrast, high spatial resolution imaging from HST enables the identification of clusters 2–3 mag fainter than previous ground-based surveys. Within this fainter luminosity range, Figure 11 shows an order of magnitude increase in the number of objects identified in the PHAT cluster catalog when compared to previous HST-based cluster survey work (HKC; blue dotted histogram in the inset). This improvement results from the order of magnitude increase in spatial coverage provided by PHAT when compared to the limited number of previous, targeted HST observations that fall within the Year 1 survey footprint (35 arcmin² versus 390 arcmin² in PHAT Year 1).

Zoom In Zoom Out Reset image size

In addition to the catalog comparison presented here, in Appendix C we compare photometry results presented in this work to those of existing catalogs. This analysis acts as quality assurance for the photometry presented here and provides the reader with an assessment of the inherent differences between the sets of photometric results.

The Year 1 cluster sample, derived from ∼1/4 of the total expected PHAT survey data, provides the first glimpse of what can be expected from the full balance of the PHAT stellar cluster survey. In Section 5, we showed that our catalog represents a considerable increase in the number and diversity of clusters known in M31. Our excellent HST-based imaging should lead to factor of >4 increases in the number of known clusters within the PHAT survey footprint.

To obtain a better sense for the type of clusters we have identified in the Year 1 sample, we present color and magnitude distributions from the catalog photometry. For this analysis, we select a subset of objects with well-determined photometry, where the uncertainties in the F336W, F475W, and F814W magnitudes are each less than 0.5 mag. This quality cut results in a subsample of 482 well-characterized objects that we use to explore the properties of the catalog.

We plot a color–color diagram for clusters with well-determined photometry in Figure 12. This diagram aids our ability to assess the cluster age distribution. To provide reference points to guide the eye, we overlay the stellar evolution model predictions from the Padova group (Girardi et al. 2010). These model tracks assume solar metallicity and are reddened to account for foreground Galactic extinction (E(B − V) = 0.062; Schlegel et al. 1998). On this plot, ages increase as we follow the evolutionary track from the upper left to the bottom right. In addition to the initial cluster selection based on photometric uncertainties, we define a second subset of clusters based on cluster luminosity. The red points denote the well-determined subsample's 92 most luminous clusters, with F475W <19.5.

Zoom In Zoom Out Reset image size

The first conclusion we draw from Figure 12 is that the Year 1 cluster sample includes a wide range of ages. Clusters populate the full length of the model evolutionary track, with a large number of objects populating an intermediate-age range (300 Myr to 3 Gyr). However, the mapping from position in the color–color diagram to age suffers from well-known degeneracies with extinction and metallicity. For example, old (12–14 Gyr) metal-poor ([Fe/H] ≲ −1.0) globular clusters inhabit the same position on this diagram as ∼100 Myr old, solar metallicity clusters with A_V of ∼1.5 mag. The green region in the right panel of the figure denotes the shared color–color region where this particular age–metallicity degeneracy exists.

Second, we observe that the cluster color distribution is affected by the effects of stochastic stellar mass function sampling in low-mass (<10⁴ M_☉) clusters. The red region in the right panel of Figure 12 highlights color outliers that have anomalous red F475W−F814W colors. While young clusters (<50 Myr) suffering from large amounts of dust attenuation (A_V > 2 mag) can populate this region of the diagram, these red colors are more frequently caused by the presence of a small number of bright evolved supergiant stars that can bias integrated cluster colors. We discuss this behavior in greater detail in Section 7.3.1. In addition to the red outliers, stochastic effects can also cause integrated cluster colors to appear bluer than model predictions. The fluctuation in the small number of evolved stars sometimes results in the complete absence of supergiant cluster members, meaning that the cluster's integrated light is emitted exclusively by main-sequence stars. Such clusters fall onto a linear sequence in color–color space we refer to as the cluster main sequence. We highlight this feature in blue in the right panel of Figure 12.

Next, we plot a cluster CMD in Figure 13 to assess the cluster mass range probed by the Year 1 sample. As in Figure 12, we plot foreground reddened, solar metallicity Padova stellar models for reference. The PHAT clusters span ∼8 mag in F475W luminosity, translating to >3 orders of magnitude in cluster mass. This range indicates that the PHAT cluster sample hosts a wide variety of clusters, spanning systems that contain hundreds of solar masses up to those with a million solar masses. The most luminous clusters, however, all appear to have red colors (F336W−F814W ∼ 2.5), forming a vertical sequence of objects on the right side of the plot. This results from the fact that most massive (>10⁵ M_☉) clusters in the Year 1 sample are old globular clusters.

Zoom In Zoom Out Reset image size

Figures 12 and 13 show that the sample of clusters assembled from the PHAT data set provide a top-to-bottom assessment of the M31 cluster population. Few data sets have the ability to sample objects across a variety of stages in cluster evolution over such a large, uninterrupted mass range. This diversity makes the PHAT cluster sample a valuable tool to better understand the formation and dissolution of star clusters.

7.1. Luminosity Functions

Luminosity functions are a basic, model-independent measure of a stellar cluster population. Up to this point, observations of spiral galaxies have provided largely consistent results, where luminosity functions are well fit by a power law ($N \propto L^{-\alpha _{L}}$) with indices of about −2 or steeper (Larsen 2002; Gieles 2010; Chandar et al. 2010b). In addition, there is growing evidence of steeper slopes at the brightest cluster luminosities (M_V < −9; Whitmore et al. 1999; Gieles et al. 2006; Haas et al. 2008), suggesting the possibility of a Schechter-like cluster mass function with an exponential truncation at the high-mass end (Larsen 2009).

These previous studies have generally focused on the bright end of the luminosity function, fitting clusters with luminosities greater than ∼2 × 10⁴ L_☉ (equivalent to M_V ≲ −6). In contrast, the faint end of the luminosity function has relatively poor constraints due to detection and catalog completeness difficulties encountered by previous extragalactic cluster studies. However, due to our ability to identify low-luminosity clusters in PHAT survey data, we can use the Year 1 cluster sample to probe the shape of the luminosity function down to limits only previously accessible in the Magellanic Clouds. These new constraints are interesting because of the sensitivity of the faint-end luminosity function to cluster disruption (Larsen 2009).

To assess the shape of the PHAT Year 1 luminosity functions in each of the six filter passbands, we consider two different samples of objects for fitting: all 601 clusters, and a subsample of 534 disk clusters that excludes objects from the bulge-dominated Brick 1. The cluster population within Brick 1 is dominated by old, massive globular clusters associated with the galaxy's bulge component, in addition to the significantly brighter completeness limits in this region. The disk sample represents a set of predominately younger objects that are likely more homogeneous in terms of formation and destruction history, simplifying the interpretation of its resulting luminosity functions. We narrow down the samples further to exclude clusters with uncertain photometry (σ > 0.5 mag), performing the quality cut individually for each passband. The fraction of objects in each filter that meets this quality requirement is listed in Table 4. Finally, we convert aperture-corrected total magnitudes for the selected clusters to absolute magnitudes, correcting for the M31 distance modulus and Galactic foreground reddening. We note that no correction for dust attenuation within M31 has been applied; we will derive these correction factors on an object-by-object basis as part of a future age and mass fitting analysis (M. Fouesneau et al. 2012, in preparation).

To characterize the power-law slope of the luminosity function, we perform a linear fit to constrain the slope of log dN/dM (the logarithm of the number of clusters per magnitude) as a function of absolute magnitude. For each passband, we bin the clusters using variable-sized magnitude bins to ensure fair weighting of the data, following the suggestion of D'Agostino & Stephens (1986; and more recently in Maíz Apellániz & Úbeda 2005; Haas et al. 2008) to group the data such that each bin represents an equal number of clusters, N. We choose N = 15, but find the results are insensitive to the particular number chosen. We fit to data points brighter than absolute magnitude completeness limits: −4.0 for F275W, F336W, and F475W; −5.0 for F814W; and  − 6.0 for F110W and F160W. These limits are conservative; they are equivalent to >80% completeness for all bricks, as found in Section 3.2. We convert the resulting magnitude-based slopes to equivalent luminosity function slopes (α_L) and report these as our primary results. The luminosity functions and their associated power-law fits for the complete and disk-only cluster samples are presented in Figures 14 and 15; the values of the fitted slopes are listed in Table 5. We separate the results into two groups to isolate the F275W, F110W, and F160W passbands, which carry the potential for larger systematic uncertainties due to the smaller fraction (∼50%–75%) of clusters with well-determined photometry available at these wavelengths. As such, we will focus our subsequent discussion on results from the F336W, F475W, and F814W passbands in Figure 14.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 5. Luminosity Function Fits

Passband	αL(All Clusters)	αL(Disk Clusters)
F275W	−1.72 ± 0.07	−1.76 ± 0.08
F336W	−1.73 ± 0.06	−1.90 ± 0.08
F475W	−1.79 ± 0.06	−2.06 ± 0.09
F814W	−1.74 ± 0.05	−2.18 ± 0.09
F110W	−1.68 ± 0.06	−2.20 ± 0.11
F160W	−1.56 ± 0.05	−1.82 ± 0.08

Download table as:  ASCIITypeset image

Considering the complete sample of Year 1 clusters, we find luminosity function power-law slopes in the F336W, F475W, and F814W passbands that are all flatter than −2. While these measurements agree with the general trend of flatter slopes measured at fainter luminosities (e.g., in NGC 45 and M51; Mora et al. 2007; Haas et al. 2008), we suggest these flat slopes result from the inclusion of a relatively large number of luminous, evolved globular clusters associated with Brick 1 and included in the complete cluster sample. We find that although clusters in Brick 1 make up ∼15% of the complete sample by number, ∼50% of sample members with m_F475W < 19 (M_ap ≲ −5.5) are Brick 1 clusters with spectroscopically derived ages of >10 Gyr (Caldwell et al. 2009, 2011). Globular clusters are known to follow a Gaussian-like luminosity function shape with a peak magnitude around M_V of −7 and −8, depending on factors such as metallicity and age (e.g., Harris 1991; Barmby et al. 2001). The dissimilar luminosity function shape of the globular cluster subpopulation significantly influences the slope of the overall complete Year 1 luminosity function.

To determine the luminosity function shape that is characteristic of the dominant, young cluster population found in M31, we examine the fitting results for the disk-only sample of clusters, presented in the bottom row of Figure 14. We find uniformly steeper slopes (α_L < −1.9) for the F336W, F475W, and F814W disk-only luminosity functions when compared to the complete Year 1 sample results. In addition to the closer agreement between these slopes and the canonical −2 power law, we also observe a significant trend where the luminosity functions at bluer wavelengths are flatter than at redder wavelengths. This behavior can be explained by wavelength and age-dependent cluster mass-to-light ratios (Gieles 2010).

To visualize the sensitivity of the faint-end luminosity function slope to differences in cluster dissolution, we compare our F475W luminosity functions to shapes predicted by four canonical dissolution models in Figure 16. The analytical luminosity functions were calculated by Larsen (2009) under the assumption of a constant cluster formation history, an underlying Schechter mass function (with cutoff of 2 × 10⁵ M_☉), and continuous, non-stochastic sampling of the stellar initial mass function.²² In the case of mass-dependent cluster dissolution, Larsen (2009) finds that the slope of the luminosity function should flatten to > − 2 at M_V > −8 or −9, while scenarios with mass-independent dissolution (and the case of zero dissolution) result in slopes that always remain steeper than −2.

Zoom In Zoom Out Reset image size

The complete Year 1 cluster sample appears to be better described by a mass-dependent disruption model, while the steeper disk cluster luminosity function appears to agree with the shape of the mass-independent or the no disruption case. However, due to the degenerate effects of underlying cluster age and mass distribution differences and variations in cluster disruption, such simple interpretation is not possible. We opt to defer further analysis and interpretation until we obtain robust age, mass, and extinction determinations for individual clusters. This crucial information will allow for the separation of the degenerate effects that influence the shape of the luminosity function and better constrain the characteristics of the underlying cluster population.

7.2. Cluster Sizes

Cluster size is a fundamental parameter that provides constraints on the dynamical state of stellar systems. While many interesting correlations involving cluster sizes require the determination of other cluster parameters (e.g., mass–radius and age–radius relations), from sizes alone we can compare the overall properties of the cluster sample to those in other galaxies and look for interesting sub-populations of clusters within the sample.

We plot R_eff as a function of F475W magnitude in Figure 17. In this parameter space, the sample separates well into two groups: Brick 1 clusters that are predominately old (∼10 Gyr), massive globular clusters associated with the galaxy bulge as discussed in Section 7.1, and clusters from the remaining bricks that are associated with Andromeda's star-forming disk. Overall, the Brick 1 clusters are more compact and luminous than the disk clusters, and show a trend in which R_eff decreases with decreasing F475W luminosity. Along with the cluster data, we plot a line of constant surface brightness anchored at F475W of 21.2 and an R_eff of 3 pc, representing the 50% catalog completeness limit as determined in Section 3.2. The distribution of clusters with respect to this line suggests that, to first order, catalog completeness roughly follows a constant surface brightness cutoff.

Zoom In Zoom Out Reset image size

The disk cluster sample includes objects with a broad range of ages and suffers from significant completeness truncations, making its magnitude–radius distribution difficult to interpret. On the other hand, the Brick 1 clusters are similarly aged and are generally much brighter than the completeness limit. This allows us to conclude that the moderate trend in which fainter Brick 1 clusters are more compact is real. We do not currently know the physical origin of this trend, but plan to follow up using ages and metallicities from Caldwell et al. (2011) and masses from Strader et al. (2011).

The left panel of Figure 18 presents the principal result of our size analysis, which shows that the overall size distribution of the Year 1 PHAT cluster sample can be described approximately as a log-normal distribution, where the median R_eff value is 1.68 pc (044). This median R_eff is similar to, but smaller than, values found in other galaxies. In M83, Bastian et al. (2011) find an overall median value of ∼2.4 pc for a sample of mostly young clusters, while McLaughlin & van der Marel (2005) find a median value of ∼3.3 pc for a sample of Milky Way globular clusters. Additionally, Barmby et al. (2007) find a median value of ∼2.5 pc for a sample of previously known M31 globular clusters.

Zoom In Zoom Out Reset image size

We can explain the differences in median cluster sizes by considering how the composition of the Year 1 PHAT cluster sample differs from the M83, Milky Way, and existing M31 samples. One possible reason for the median mismatch with the Milky Way and M31 globular cluster samples is that the Year 1 PHAT sample is a heterogeneous mix of mainly young clusters, while the globular samples are uniformly much older. However, an age-based explanation is ruled out because the old, globular-dominated Brick 1 subsample has a similar, but still smaller, median R_eff of 1.69 pc. The R_eff distribution for the Brick 1 subsample is shown in the right panel of Figure 18.

An explanation for the median differences becomes clear when we consider the galactocentric radii of the existing globular cluster samples. A correlation between R_eff and galactocentric radius has been shown to exist in many galaxies such that inner clusters are more compact (Barmby et al. 2007). When we limit the Milky Way and previous M31 cluster samples to include only objects with galactocentric radii less than 3 kpc, which roughly matches the radial extend of the Brick 1 sample, the median R_eff for these two samples drops to ∼2.0 pc. These lower medians are in better agreement with the Brick 1 subsample, easing the tension between the values derived for the two globular cluster samples and for the PHAT clusters.

To explain the median R_eff difference between the young M83 cluster sample and the Year 1 PHAT sample, instead of age and galactic position, we consider how completeness limits influence the shape of the R_eff distribution. In terms of numbers, the Year 1 cluster sample is dominated by compact, low-luminosity objects that lie just above our detection threshold. At these faint magnitudes, surface brightness limits prevent us from detecting clusters across the full range of possible R_eff values. As a result, the overall cluster sample is biased toward compact objects, driving the median toward smaller values. However, if we only consider clusters with F475W <20.5, the biasing effects of surface brightness limits are greatly reduced. The R_eff distribution for this luminous subsample is shown in the right panel of Figure 18. Without the faint, compact portion of the sample, the median cluster size rises to 2.32 pc, which agrees well with the observed value obtained for clusters in M83. In conclusion, cluster sizes for the Year 1 sample are consistent with results derived by previous works in similar spiral galaxies once sample completeness and composition are taken into account.

7.3. Objects of Interest

7.3.1. Red Supergiant Clusters

We identified a subsample of objects with unusually red F475W−F814W colors in Figure 12. Within the selection box, we identify ∼15 clusters that show extreme integrated colors. Upon inspection, we find that many of these clusters appear to host luminous red supergiant (RSG) stars, which explain their anomalous integrated colors. Two examples of this class of clusters, PC57 and PC1127, are shown in Figure 19.

Zoom In Zoom Out Reset image size

RSG stars are massive (8 to 25 M_☉) stars that have left the main sequence and are traversing the top of the CMD during the helium burning phase of stellar evolution. Individual luminous RSG stars have the ability to bias the integrated colors of low-mass (<10⁴ M_☉) clusters because stochastic sampling of the stellar mass function in low-mass systems can cause the small number of evolved stars to vary by factors of a few (low-mass clusters typically host between 0 and 4 evolved stars). This variation creates wild fluctuations in the total luminosity and integrated color of the cluster, which depend on the particular number of evolved stars progressing through this relatively short evolutionary phase.

The existence of these color outliers confirms that stellar population modeling of low-mass clusters must account for stochastic sampling of the stellar initial mass function to obtain accurate age and mass determinations (see, e.g., Fouesneau & Lançon 2010). Models that assume a fully sampled stellar initial mass function cannot reproduce objects with integrated colors as red as those shown by RSG clusters. Our future cluster analysis will benefit from the use of cluster models that account for stochastic fluctuations in integrated light caused by the RSGs.

More importantly, our ability to resolve individual stars in clusters provides a number of important benefits with respect to the RSG clusters. Resolved star photometry enables us to use CMD fitting techniques to obtain cluster parameter determinations, allowing us to avoid the use of biased integrated measurements all together. In addition, once we obtain age, mass, and attenuation information for the clusters, we can use this information to tag individual RSGs and use these constraints to improve calibration of late stage stellar evolution for massive stars at high metallicity.

7.3.2. Massive Clusters

Globular clusters, with old characteristic ages (∼10 Gyr) and large masses (>10⁵ M_☉), have long been a target of study in M31. Although the Year 1 cluster sample is numerically dominated by low-mass clusters, it contains many massive clusters as well. While we identified few new globular clusters as part of the Year 1 search, our high spatial resolution imaging has enabled us to confirm or eliminate a large number of globular cluster candidates from existing catalogs. In addition to the 63 out of 71 highly ranked clusters we confirm from the RBC, we affirm 12 and reject 19 possible candidates. The PHAT survey also enables detailed analysis of these objects by means of resolved star photometry for cluster members. We plan to place better constraints on red giant branch (RGB) and horizontal branch morphology as a function of metallicity through these detailed studies.

In addition to the old globular clusters, we have also newly identified a number of intermediate-mass (∼10⁴ M_☉), intermediate-age (∼1–3 Gyr) clusters. These objects are interesting targets for study because of the relative rarity in the Milky Way (Friel 1995), where most similarly aged Galactic clusters are less massive. While a small number of these intermediate-mass, intermediate-age clusters are already known to exist in M31 (Caldwell et al. 2009, 2011), increasing the sample size for this class of object should help to better understand their origin and evolution. Investigation of these objects will complement the study of younger (<1 Gyr) massive clusters by Fusi Pecci et al. (2005), Caldwell et al. (2009), and Perina et al. (2010).

The most remarkable massive cluster in the Year 1 catalog is the cluster SK142C (PC1156). This object is the sample's most discrepant color outlier, appearing at the redward extreme of all cluster photometric measurements, with an F475W−F814W color of ∼3.5. This cluster appears to be an intermediate-mass, intermediate-age, highly reddened cluster. Along with cutout images showing the cluster's heavily reddened appearance, we plot the cluster's resolved star CMD in Figure 20. The CMD shows an extended, highly reddened RGB, indicative of an intermediate cluster age. We perform a visual comparison to solar metallicity Padova isochrones (Girardi et al. 2010) and find that the CMD is consistent with an age of ∼1–3 Gyr and A_V of 2.25–2.75 mag, assuming large uncertainties associated with visual "chi-by-eye" isochrone fitting. Using this combination of age and reddening, we use the cluster's F160W luminosity to estimate a cluster mass of ∼1.5 × 10⁵ M_☉. Due to its massive nature and its somewhat intermediate age, this cluster is a rare and interesting object akin to newly discovered objects in the Milky Way (Figer et al. 2006; Davies et al. 2011, 2012).

Zoom In Zoom Out Reset image size

In this paper, we presented the first installment of the PHAT stellar cluster catalog. We introduced the Year 1 cluster sample, consisting of 601 clusters identified through a visual search of high spatial resolution HST imaging. The PHAT cluster sample significantly increases the number of clusters known in M31; the catalog presented here represents more than a factor of four increase in the number of known clusters within the current survey area. We presented a basic assessment of the cluster sample, including positional, size, and photometric information.

We have shown that the PHAT cluster sample hosts a large range in ages and masses. This wide-ranging sample provides the opportunity for a top-to-bottom study of stellar cluster evolution processes. The PHAT sample includes low-luminosity objects missed in studies of distant external galaxies, while covering an uninterrupted range of cluster masses, unlike Milky Way cluster samples.

When the survey is complete, the PHAT cluster catalog will be among the largest and most comprehensive surveys of star clusters in any galaxy. This work presents results derived from the first 25% of the survey data; we estimate that the final sample will include ∼2500 clusters. Over the duration of the survey, we plan to periodically publish updates to the catalog to include new clusters and to revise object classifications as we gather additional information about the sample. Age and mass determination analysis for the Year 1 sample is currently underway (M. Fouesneau et al. 2012, in preparation), which will provide the means to explore a host of different topics, including the cluster mass function and cluster dissolution. The Year 1 cluster sample will also be the basis for analysis of structural parameters, resolved star content, and integrated spectroscopy, but we also look forward to future studies enabled by the complete, four-year PHAT cluster catalog.

The authors acknowledge the collective efforts of the entire PHAT team in this project. Also, the authors thank the anonymous referee for a prompt and useful report. This research made extensive use of NASA's Astrophysics Data System Bibliographic Services. Support for this work was provided by NASA through grant number HST-GO-12055 from the Space Telescope Science Institute, which is operated by AURA, Inc., under NASA contract NAS5-26555. D.A.G. acknowledges financial support from the German Aerospace Center (DLR) and the German Research Foundation (DFG) through grants 50 OR 0908 and GO 1659/3-1, respectively.

Facility: HST (ACS, WFC3) - Hubble Space Telescope satellite

As introduced in Section 3.1, we present a catalog of objects visually identified as background galaxies in Table 6. Along with positional information, we present a rough assessment of the galaxy size.

We performed a comparison of object classifications from existing cluster catalogs to those in the PHAT Year 1 cluster catalog in Section 5. Here we provide comments on individual catalog comparisons, including detailed classification statistics and object-by-object classification revisions. We discuss individual classifications statistics for the RBC (Galleti et al. 2004), Kim et al. (2007), Caldwell et al. (2009, 2011), Peacock et al. (2010), and three of the four HKC studies (Krienke & Hodge 2007; Hodge et al. 2009, 2010; no objects from Krienke & Hodge 2008 lie within the Year 1 PHAT footprint). Classification results broken down by object type are presented in Table 7.

The Revised Bologna Catalog. Our search for RBC objects that lie within the Year 1 PHAT imaging footprint produces a catalog of 187 objects. These objects include 71 clusters, 36 cluster candidates, and 80 non-cluster objects. After reanalyzing these classifications using the by-eye ranking results and following up on other objects that were not selected as PHAT candidates, our revised object classifications show excellent agreement with the original RBC results. Individual object revisions are provided in Table 8, but the bulk of the changes are due to our ability to determine the nature of the RBC's candidate objects (RBC Flag = 2). Of the 36 candidates, we confirm 12 objects as clusters in the PHAT catalog, reject 23 candidates (including two originally controversial cases, one duplicate entry, and one object not found in PHAT imaging), and transfer the final candidate to the PHAT possible cluster sample. In terms of catalog accuracy, only five "confirmed" clusters (RBC Flag = 1) were confidently revised to non-cluster objects, four of which were Kim et al. (2007) objects (see additional discussion below). Overall, the high level of consistency between the two catalogs gives us great confidence in galaxy-wide accuracy of the RBC cluster sample.

Kim et al. (2007). The catalog content of Kim et al. (2007) is also accounted for in the results of the RBC cross-comparison, but the resulting cluster confirmation and rejection statistics from this work is worth individual mention. As pointed out in Caldwell et al. (2009) and Peacock et al. (2010), the Kim et al. (2007) catalog has a high level of contamination, namely from misclassified stellar sources. Considering all three cluster quality categories from Kim et al. (2007) together, we find 38 objects from this catalog that fall within the Year 1 PHAT footprint. Of these, only 37% of those objects appear in the PHAT catalog as confirmed or possible clusters. Even when taking the quality rankings into consideration, we find that (50%, 27%, 38%) of the (A, B, C) objects appear in the catalogs from our present work.

Caldwell et al. (2009, 2011). These two papers revised classifications from v3.4 of the RBC (many of which were adopted in v4.0) using low-resolution spectra, ground-based imaging from the Local Group Galaxy Survey (LGGS; Massey et al. 2006), and HST-based images. As this catalog has been incorporated into the latest version of the RBC, the discussion of individual object classifications is included in Table 8. When comparing our current PHAT classification with those from Caldwell et al. (2009, 2011), we find good agreement between the two works. We revise classifications for 6 clusters out of 183 that fall within the Year 1 footprint.

Peacock et al. (2010). The clusters considered by Peacock et al. (2010) are included in the RBC comparison discussed above, and classifications for each of these objects is already provided in Table 8. Many of the reclassifications performed in Peacock et al. (2010), which was based on v3.5 of the RBC, have proven to be accurate (including the stellar classification of numerous Kim et al. 2007 candidates). Overall, we find good consistency between the Peacock et al. (2010) catalog and our PHAT classifications.

Hodge–Krienke Catalogs. Similar data and methodology (including common authors in the case of P.H.) lead to strong consistency between the HKC and the work presented here. We find 54 clusters from the HKC that match entries in the PHAT cluster catalog and 8 additional objects that match possible clusters, out of a total of 75 objects that lie within the boundaries of the PHAT Year 1 footprint. Unlike in the case of ground-based catalogs, the 13 HKC objects not recovered as part of the PHAT cluster search mostly fall into the category of asterisms. In other words, these objects were interpreted by the PHAT search team as uncertain or unlikely cluster candidates, and omitted from the final catalog presented in this work. This subset of objects shows that for the least luminous, least massive clusters, defining the difference between an object that is potentially a bound stellar system and simply a chance collection of stars seen in projection becomes a difficult and subjective task. Table 9 presents proposed revisions to the HKC. In closing, we note that we do not compare to cluster identifications presented in Williams & Hodge (2001) because classifications in this work were superseded by the HKC.

We provide a comparison between PHAT cluster photometry and the results of Peacock et al. (2010), Fan et al. (2010), Hodge et al. (2009), and Hodge et al. (2010). We compare photometry for these selected studies because they provide magnitudes derived in similar photometric passbands as part of a uniform analysis. We choose not to compare to photometry provided by the RBC due to the heterogeneous nature of their unified photometric measurements (see Galleti et al. 2004 for details). We note that both Peacock et al. (2010) and Fan et al. (2010) find good agreement between photometric results and that of Barmby et al. (2000), upon which the photometry of the RBC is based.

The optical aperture photometry of Peacock et al. (2010) was derived from Sloan Digital Sky Survey (SDSS) ugriz imaging. We compare photometric results for 77 clusters found in both data sets. We calculate transformations to convert from AB-based g- and i-band photometry to Vega-based F475W and F814W magnitudes using transformations from Girardi et al. (2008), assuming a single-age, 1 Gyr solar metallicity population that represents a median value for the age-dependent transformation:

$$\begin{equation} {\rm F475W} = g_{{\rm AB}} + (0.06 \pm 0.02), \end{equation} \tag{ C1 }$$

$$\begin{equation} {\rm F814W} = i_{{\rm AB}} - (0.54 \pm 0.10). \end{equation} \tag{ C2 }$$

The resulting photometric comparison is presented in the top row of Figure 21. We find that the PHAT magnitudes are fainter than those of Peacock et al. (2010) by ∼0.25 mag. This offset can be explained by the difference in aperture sizes, as seen in Figure 22. The Peacock et al. (2010) apertures have radii that are twice as large on average, and up to four times as large as those used in this work. In the case of the brightest comparison cluster, B127-G185 (PC1425), the size of the aperture (10'') was large enough to include the nearby cluster NB89 (PC1426), explaining the ∼0.5 mag difference for this object. In a comparison of F475W−F814W versus (g − i) optical colors, the agreement is quite good for brighter clusters (F475W < 17) with increasing scatter and a redward bias of Peacock et al. (2010) colors for fainter clusters. These differences are likely caused by field contamination within the large Peacock et al. (2010) apertures and by blending in their low-resolution (1''–2'' seeing) ground-based imaging. In summary, we find systematic differences between the photometry derived in this work and that of Peacock et al. (2010), but the overall agreement is adequate and remaining differences can be readily explained by the effects of image resolution and aperture size differences.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In another recent compilation of ground-based aperture photometry, Fan et al. (2010) derived Johnson–Kron–Cousins UBVRI magnitudes from LGGS imaging (Massey et al. 2006). The Vega-based PHAT F336W and F814W magnitudes are nearly equivalent to the U and I magnitudes from Fan et al. (2010), allowing us to directly compare photometric results for 68 clusters common to both samples. We present the results in the second row of Figure 21, showing good agreement between the two data sets. While Fan et al. (2010) magnitudes appear slightly brighter for faint clusters, the overall agreement is improved with respect to the Peacock et al. (2010) results. One explanation for the improvement between ground-based and HST-based results could trace back to the superior image quality of the LGGS data when compared to the SDSS imaging of Peacock et al. (2010). The average seeing for the LGGS data is ∼1'', while varying between 1''–2'' for the SDSS imaging. This results in the ability for Fan et al. (2010) to reduce their aperture sizes, such that they are only a factor of ∼1.5 bigger than those used by PHAT.

The photometry of Hodge et al. (2009) and Hodge et al. (2010) allows us to compare two sets of HST-based photometric results. Out of the four M31 cluster catalog papers that compose the HKC, we choose two where the tabulated photometry is provided in native, Vega-based HST filter magnitudes. Although the data used in these studies are similar to those used in this work, there are fundamental differences in the photometry techniques employed. As opposed to our growth curve method of aperture size determination, the HKC adopts an isophotal aperture determination, such that apertures extend out to a chosen surface brightness limit. Further, the methods of sky background determination differ. Our sky levels are determined using local sky background estimates measured in annular rings around the photometric aperture, while the HKC uses samplings of the sky background taken across the HST image. Due to these differences in methodology and the lack of any correction to account for total cluster light, we expect the comparison to show some scatter and a systematic offset such that the PHAT photometry is brighter.

For the Hodge et al. (2009) comparison, we use 18 objects common to both data sets and present the results in the third row of Figure 21. We compare our F475W photometry to magnitudes measured in a similar F450W passband, using a transformation derived in the same way as Equations (C1) and (C2):

$$\begin{equation} {\rm F475W} = {\rm F450W} - (0.11 \pm 0.10). \end{equation} \tag{ C3 }$$

There are several significant (Δ > 0.5 mag) outliers in the F475W magnitude comparison in which the photometry from this work is brighter by up to ∼1 mag. To check the validity of our measurements, we performed photometry for the three most discrepant outliers on the WFPC2 images used by Hodge et al. (2009), employing photometry procedures used in this work. We found resulting magnitudes that agree within ∼0.1 mag of our original PHAT photometry values. This result indicates that the magnitude offsets are likely due to differences in photometric technique, specifically related to sky background determination.

The bottom row of Figure 21 presents the comparison of Hodge et al. (2010) photometry for 13 common objects. We find better overall agreement with this data set when compared to the Hodge et al. (2009) results. We compare F814W magnitudes (no filter transformations required) and F336W−F814W colors, and find a scatter of ∼0.2 mag between the two data sets. The comparison of photometry between our current PHAT study and the work of Hodge et al. (2009, 2010) shows good overall consistency. Systematic biases and a small number of significant outliers likely stem from differences in measurement technique.