IDENTIFYING CONTRIBUTIONS TO THE STELLAR HALO FROM ACCRETED, KICKED-OUT, AND IN SITU POPULATIONS

The formation history of the Milky Way is recorded in the present motions and chemical abundances of its stars. Ideally, to unravel the Milky Way's history, we would like a catalog containing spatial, kinematical, and chemical data for every star in the Galaxy. Large-scale photometric surveys, such as the Two Micron All Sky Survey (2MASS) and the Sloan Digital Sky Survey (SDSS), are bringing us closer to this goal: both have led to sweeping views of the structure of the Galaxy. Star counts from these surveys allow for detailed studies of the structure of each Galactic component (e.g., Skrutskie et al. 2001; Majewski et al. 2003; Bell et al. 2008; Jurić et al. 2008; Ivezić et al. 2008) and are rich sources for follow-up spectroscopic studies (e.g., Majewski et al. 2004; Yanny et al. 2009).

The present study looks at the spectroscopic properties of a sample of relatively nearby (d ≲ 9 kpc) M giant stars at mid-Galactic latitudes of 30° < |b| < 60° selected from the 2MASS catalog. M giants (1) are intrinsically bright stars and, hence, can be easily observed to large distances using even small telescopes; (2) can be readily identified on the basis of near-infrared, JHK photometry, and complete samples can be culled from full-sky catalogs (e.g., 2MASS); and (3) are limited to more metal-rich populations. These unique properties of M giants, when combined with the adopted latitude and magnitude selection criteria, make our survey particularly useful for exploring the structure of the thick disk and the metal-rich component of the nearby halo.

In this first paper describing our survey, we present the photometric and spectroscopic data for our current thick-disk-dominated sample of 1799 stars but focus on the detection and interpretation of those stars that do not kinematically conform to the typical behavior of the Milky Way thick disk population and are likely to be members of the stellar halo. This stellar halo sample is unique compared to other halo surveys of stars in that it covers an intermediate distance range (d < 9 kpc) and concentrates on a relatively rare halo tracer. For example, our survey volume is wider than that of the Hipparcos survey of the Sun's closest neighbors (d_{lim, Hip} ∼ 100 pc; Perryman et al. 1997), as well as kinematically selected solar neighborhood surveys (Eggen et al. 1962; Carney et al. 1996; Schuster et al. 2012), but is a much more local view of the stellar halo than the SDSS F–G type turnoff stars (visible to 20–40 kpc; Jurić et al. 2008; Bell et al. 2008) or the entire 2MASS M giant catalog (visible out to 100 kpc; Majewski et al. 2003). Our catalog is comparable in probed distances to the spectroscopic studies of the Century Survey Galactic Halo Project (Brown et al. 2008), RAVE (Steinmetz et al. 2006), and Carollo et al. (2007, 2010), although these studies employed different, more commonly used halo tracers: blue horizontal branch (BHB) stars; stars—both dwarfs and giants—with 9 < I < 12; and main-sequence turnoff (MSTO) stars or dwarfs from the SDSS DR7 calibration sample, respectively.

In this first analysis of our survey we explore the origin of the M giant population in the nearby stellar halo. There are three broad categories of formation scenarios typically postulated for halo stars.

• 1.  
  In-situ-halo stars form at comparable radii to their current locations within the dominant dark matter halo progenitor of the Galaxy. For example, Eggen et al. (1962) envisioned that the stellar halo could be formed from infalling gas, prior to the formation of the disk, during an early monolithic collapse phase for our Galaxy, as seen in the hydrodynamical simulations of Samland & Gerhard (2003).
• 2.  
  Kicked-out stars are formed initially more concentrated toward the center of the dominant dark matter halo, within either the bulge or disk, and are subsequently ejected to more eccentric orbits through minor or major mergers, as proposed by Purcell et al. (2010) and seen in hydrodynamical simulations by Zolotov et al. (2009) and McCarthy et al. (2012).¹¹ (Note that there were many earlier studies of this same process that focused on the thickening or destruction of disks rather than the formation of the stellar halo, e.g., Quinn et al. 1993; Walker et al. 1996.)
• 3.  
  Accreted stars form in separate dark matter halos that later merge with the Galaxy (as proposed by, e.g., Searle & Zinn 1978).

Note that in prior theoretical work, the first two categories were both simply termed in situ to indicate more generally stars formed in the dominant galaxy dark matter halo progenitor (as in, e.g., Abadi et al. 2006). Separate terms are introduced here for clarity.

Many prior studies have probed these formation mechanisms by looking at the distribution of stars in different dimensions of phase space. For example, the importance of an accreted population has traditionally been assessed by looking for residual groupings that are signatures of stars' original associations, whereas in-situ-halo or kicked-out populations are expected to be more smoothly distributed. All-sky photometric surveys have revealed rich substructure in the outer halo (e.g., Sgr tidal tails—Majewski et al. 2003; the Anticenter Tributaries—Grillmair 2006; the Orphan stream—Belokurov et al. 2007) that can best be explained by accretion events (e.g., Bullock et al. 2001; Bullock & Johnston 2005; Bell et al. 2008; Johnston et al. 2008; Cooper et al. 2010). In contrast, merger debris in the nearby halo fully phase-mixes on a much shorter timescale, leading to the expectation of negligible evidence for accretion identifiable as coherent spatial structures (e.g., Johnston et al. 2008; Sharma et al. 2010) and requiring additional dimensions of phase-space data to distinguish formation scenarios in this region. Adding the dimension of line-of-sight velocities helps: conservation of phase-space density during phase-mixing requires that as debris from accretion events becomes less dense with time, it should become colder in velocity (Liouville's Theorem; see, e.g., Helmi & White 1999), and coherent structures may be apparent even once stars are smoothly distributed in space. Indeed, substructure in velocities is apparent statistically over a large volume in BHB stars in the SDSS (and shown to be broadly consistent with model stellar halos built within a hierarchical cosmology; see Xue et al. 2011; Cooper et al. 2011), and individual clumps in velocity have also been detected in the halo using metal-poor MSTO stars in the SEGUE survey (Schlaufman et al. 2011), K giant stars (Majewski 2004), and the mixed populations in RAVE (Williams et al. 2011). Overall, group finding is more effective if even more dimensions of phase-space can be measured. For example, Majewski (1992) analyzed proper motions for a sample of 250 F–K dwarfs in the direction of the north Galactic pole that probes out to roughly 8 kpc; he measured a mean retrograde rotational velocity for the halo sample and detected a more coherent retrograde group of stars at a mean height above the Galactic plane of Z ∼ 4.6 kpc—findings suggestive of an accreted halo population. Subsequently, Majewski et al. (1994, 1996) obtained radial velocities (RVs) for a subsample of stars from the Majewski (1992) survey and found a significant amount of phase-space clumpiness in their halo sample. Once all six phase-space dimensions are known, conserved (or nearly conserved) quantities (such as energy, angular momentum, or orbital frequencies) can be calculated that can link stars from a common progenitor in the volume even if they are not at the same orbital phase, as has been done successfully for several nearby surveys (see Helmi & White 1999; Helmi & de Zeeuw 2000; Morrison et al. 2009; Gómez et al. 2010).

While the findings in the previous paragraph point to a significant fraction, and possibly the majority of the stellar halo, being accreted, some contributions from in-situ-halo and kicked-out stars are still possible. Structural, orbital, and/or metallicity trends in the stellar halo with radius—as well as transitions in those trends—have historically been taken as indicative of these different formation mechanisms and leading to "dual halo" models (e.g., in RR Lyraes, globular clusters, and BHB stars; see Hartwick 1987; Zinn 1993, 1996; Sommer-Larsen et al. 1997, respectively). Chiba & Beers (2000) analyzed a kinematically unbiased sample of 1203 stars with [Fe/H] < −0.6 in the inner halo (within 4 kpc of the Sun) and detected a gradient in the rotational velocity as a function of height above the Galactic plane for the more metal-rich stars in their sample—as seen for the in-situ-halo stars formed in the simulations of Samland & Gerhard (2003). Chiba & Beers (2000) also confirm the existence of the streams detected by Helmi et al. (1999)—further supporting some presence of an accreted population in this region. Carollo et al. (2007, 2010) studied the kinematics and metallicities of SDSS calibration stars for a larger volume (20 kpc) and found evidence for two populations: one of metal-rich stars on only mildly eccentric orbits (which they dubbed "inner halo") and a second of metal-poor stars on more eccentric orbits (which they dubbed "outer halo"). Similarly, Deason et al. (2011) analyzed BHB stars in SDSS (a sample that probes the outer halo to 40 kpc) and found a net retrograde rotation for metal-poor stars and a net prograde rotation for metal-rich stars. Such multi-component halos with distinct formation mechanisms for each component emerge naturally in the hydrodynamical simulations, with the inner halo (within 10–15 kpc) coming predominantly from stars (either in-situ-halo or kicked-out) formed within the main Galaxy dark matter halo progenitor and the outer halo (dominant beyond 15–20 kpc) from mergers and accretion of stars by the Galactic dark matter halo (Abadi et al. 2006; Zolotov et al. 2009; Font et al. 2011; McCarthy et al. 2012; Tissera et al. 2012). However, whether the distinct components observed in the stellar halo are indeed due to distinct formation mechanisms, and not merely a variety of accretion events, has yet to be proven.

In our own survey, the sample of stars is sufficiently far that distance and proper motion errors from extant data are too large to estimate their orbital properties accurately. However, chemical abundances can be derived from high-resolution spectra in general and thus provide an alternative avenue for exploring the origins of the stars. A star is branded at its birth by its chemical abundance patterns—a signature that is generally conserved throughout its lifetime (like orbital properties) and cannot be diluted by orbital phase-mixing. Moreover, stars deriving from a common origin should have similarities in their chemical abundance patterns, trends that are directly correlated to the details of their enrichment history. Hence, we can hope to "chemically tag" stars as members of the different populations via their abundance patterns (similar in spirit to Freeman & Bland-Hawthorn's 2002 proposal for the reconstruction of ancient star clusters in the stellar disk). The potential power of chemical tagging has already been demonstrated empirically in observations: in dwarf galaxies, for example, stars tend to have lower [α/Fe] at a given [Fe/H] than stars in the bulk of the Milky Way's stellar halo (Smecker-Hane & McWilliam 2002; Shetrone et al. 2003; Tolstoy et al. 2004; Geisler et al. 2005; Monaco et al. 2007; Chou et al. 2010a), and similar patterns have been seen in stars in stellar structures such as the Monoceros Ring (Chou et al. 2010b) and Triangulum–Andromeda Cloud (Chou et al. 2011), which lends support to the interpretation of such features as originating from disrupted dwarf galaxies. In a similar manner, a series of papers (Nissen & Schuster 2010, 2011; Schuster et al. 2012) have separated a sample of 94 nearby (within ∼335 pc of the Sun), metal-rich (−1.6 < [Fe/H] < −0.4) stars into "α-rich" and "α-poor" groups and shown systematic differences in the abundances, ages, and orbital properties of stars in these two groups that are suggestive of kicked-out and accreted origins, respectively.

Figure 1 shows two schematics to illustrate conceptually how this approach might be applied to our own sample. The lines in the left-hand panel show the expected temporal evolution, in the [α/Fe]–[Fe/H] plane, of the gaseous chemical abundance for systems with low/intermediate/high star formation efficiencies (SFEs—indicated by lines with increasingly dark shades of gray), which are assumed to correspond to stellar systems embedded in small/intermediate/large dark matter potential wells (see McWilliam 1997; Gilmore & Wyse 1998; Robertson et al. 2005). These systems could represent, for example, a Milky Way dwarf spheroidal (low SFE corresponding to low mass accreted systems), the LMC (intermediate SFE corresponding to intermediate mass accreted systems), and a Milky Way progenitor (high SFE and contributing to populations either formed in situ or kicked out from their original birthplaces). All of these systems are expected to have old stars with high [α/Fe] at low metallicity, which reflect the yields from explosive Type II SNe alone. Stars formed after the (delayed) onset of Type Ia SNe will become progressively more enriched by Fe and acquire lower [α/Fe] as Type Ia SNe produce α-elements much less efficiently. The transition point in Fe between the early and late stages reflects the metallicity that the gas has reached prior to the onset of Type Ia enrichment, which will be lower/higher for systems that have less/more rapidly converted their gas into stars.

Zoom In Zoom Out Reset image size

The right-hand panel of Figure 1 applies this intuition to show where populations with different origins might fall in this plane (these regions are defined more rigorously and empirically in Figure 7 and discussed in Section 5 using data from previous studies). The region outlined in blue should contain stars formed early (because of their low metallicities and high [α/Fe]) but that can now be in the halo via any of the three mechanisms (in-situ-halo, kicked-out, or accreted), and thus we do not anticipate being able to conclusively deduce their origins from this particular type of analysis. The region outlined in green is likely only to contain stars formed more recently (because of the low [α/Fe]) in small potential wells (because of the low Fe) and hence should be sensitive to a purely accreted population. The region outlined in orange is likely only to contain stars formed more recently (because of the low [α/Fe]) in deep potential wells (because of the high Fe) and hence should be insensitive to in-situ-halo populations (formed exclusively early on) and the majority of accreted populations (because they form in smaller potential wells—although note that there could be some contamination from high-mass, late accreted systems, as suggested in Figure 7). Stars that have formed recently in the Galactic disk and that were subsequently kicked-out might lie in this region (see Zolotov et al. 2010 for an illustration of this idea with hydrodynamical cosmological simulations). Overall, these expectations lead us to conclude that the fraction of our halo stars in the metal-poor and α-poor region is an indicator of the importance of late accretion, while the fraction with disk-like abundances but moving at large speed relative to the Sun is indicative of the contribution of a recently kicked-out population.

Thus, an additional advantage of focusing on abundance space as a probe of the relative proportions by origin of M giants in the stellar halo is that we only need to look for the expected systematic differences in the chemical composition of these populations (in-situ-halo, kicked-out, and accreted) overall rather than (for example) search for kinematical groupings from individual accretion events. Hence, we can explore formation scenarios with much smaller samples of stars than by looking at dynamics alone. Motivated by the promise of chemical abundances, which we could combine with the known RVs we already have for all program stars, we obtained high-resolution echelle spectra for 34 stars selected for follow-up based on their high, halo-like speeds relative to the disk. As a control sample we also observed five random thick disk red giants (chosen based on RVs similar to the bulk thick disk trend) and four M giant calibration stars from Smith & Lambert (1985) and Smith et al. (2000).

This paper is organized in the following way. In Section 2, we describe target selection, observations, and data reduction for the medium-resolution spectroscopy program of bright M giants. The RVs and the identification of RV outliers—stars that have high speeds relative to the bulk thick disk trend—are presented in Section 3. Details of the high-resolution follow-up spectroscopy program are given in Section 4. Our interpretation of the abundance data in terms of formation scenarios for the halo is given in Section 5. Lastly, we give a summary of the results and discuss them in the context of prior studies in Section 6. In a companion paper (Johnston et al. 2012), we develop a more complete understanding of the nature and implications of some possible dynamical groups in our survey by generating and analyzing synthetic observations of simulated stellar halos.

M giants can be distinguished from M dwarfs by their J − H and H − K colors (Bessell & Brett 1988; Carpenter 2001). M giants also dominate M dwarfs in catalogs of late-type stars to K ∼ 14. These facts were used by Majewski et al. (2003) to select M giants from 2MASS and map the streams of M giants from the Sagittarius (Sgr) dwarf spheroidal (dSph) galaxy and by Sharma et al. (2010) and Rocha-Pinto et al. (2003, 2004, 2006) to map and track the Triangulum–Andromeda star cloud, the Pisces overdensity, and the Monoceros/GASS/Argo feature. Our survey sample spans (J − K_S)₀ colors from 0.75 < (J − K_S)₀ < 1.24; this is similar to the range studied by Girard et al. (2006) in their study of the thick disk using red giants, although most (96%) of our red giants have (J − K_S)₀ > 0.85 (as in Majewski et al. 2003) to ensure a clean sample of M giants. The magnitude range of our entire sample is 4.3 < (K_S)₀ < 12.0, with a median magnitude of 7.4 (the majority, 1625 of 1799, have 5.0 < (K_S)₀ < 9.0). The magnitudes were dereddened using the maps from Schlegel et al. (1998).

A constraint in Galactic latitude of 30° < |b| < 60° was applied to our M giant catalog, with the lower limit in b applied to avoid excess contamination from the thin disk. Our nominal photometric sample contained approximately 12,000 stars, and in this paper we present spectroscopic observations for 1799, or roughly 15%, of these. The stars observed were selected randomly from the nominal sample and cover nearly all Galactic longitudes. The bulk of the observing was done at the Fan Mountain Observatory (FMO), located in Virginia, so there are gaps in coverage corresponding to the Southern Hemisphere. Of the 1799 stars, 149 were observed at Cerro Tololo Inter-American Observatory (CTIO) as part of a related program. Figure 2 shows in an Aitoff projection the spatial distribution of the 1799 program stars in Galactic coordinates. The nominal M giant catalog was matched to the UCAC2 catalog (Zacharias et al. 2004), and the gap in the northern Galactic hemisphere is due to the upper limit of δ = +52° in the UCAC2 catalog at the time of the survey inception. However, the very small amplitudes of the UCAC2 proper motions for the majority of the program stars typically result in unreasonably large relative errors (often larger than the derived space motions) in any derived kinematical parameters using them, so the proper motions are not actually utilized in the present work. Limitations from the use of the UCAC2 proper motions on 2MASS M giants are further explored in Majewski et al. (2006).

Zoom In Zoom Out Reset image size

Spectra were collected at the University of Virginia's FMO using the Fan Observatory Bench Optical Spectrograph (FOBOS; see Crane et al. 2005) on the 1 m astrometric reflector. FOBOS is a fiber-fed optical spectrograph that was designed for moderate-resolution (R ∼ 1500–3000) spectroscopy (Crane et al. 2005). FOBOS uses a grating with 1200 grooves mm⁻¹. The estimated resolution of the spectra is Δλ ∼ 4 Å. Our observing program began on UT 2005 February 25, and data presented here were taken in the years 2005–2008. The spectrograph is optimized for use over the region 4000–6700 Å but is limited by the camera, which has a SITe 2048 × 2048 CCD detector that at a linear dispersion of 1.0 Å pixel⁻¹ samples only ∼2000 Å of that range (selected by us to be 4000–6000 Å). On all but one night (when we were testing the efficiency of the setup) the detector setting used had a read noise level of 4.5 e⁻ and a gain of 6.1 e⁻ per ADU. For most stars, three spectra are taken and summed in two dimensions after the CCD frame preprocessing is completed. This combination of three images facilitates the elimination of cosmic rays. The minimum S/N to achieve the best possible RV precision (a few km s⁻¹) was found to be ∼20; this typically translated to total exposure times of 450–900 s for the M giants observed with FOBOS, which have magnitudes in the range 4.3 < (K_S)₀ < 9.7. Several RV standard stars from the Astronomical Almanac were also observed each night; these are used for cross-correlation templates when determining the RVs of the target stars. Standard stars were chosen to be of a similar spectral type as the program stars to minimize systematic offsets in the derived RVs. Several sets of bias frames were taken throughout each night. To remove pixel-to-pixel variations in the frames, a "milky flat" is created by illuminating an opal diffusing glass with a quartz–tungsten–halogen lamp. The object and comparison frames are flat-fielded (using the milky flat), trimmed, and bias subtracted using the task ccdproc in IRAF.¹² Comparison spectra were taken using neon, argon, and xenon lamps for calibrating wavelengths against the laboratory values for lines from these elements. The spectra are extracted from the two-dimensional images and converted to one-dimensional spectra and wavelength calibrated using the IRAF tasks apall and identify.

To obtain RVs, the extracted, flat-fielded, wavelength-calibrated spectra are cross-correlated with the standard star spectra using code developed by W. Kunkel (described in detail in Majewski et al. 2004). The cross-correlation code is run for each night of data, and each program star is cross-correlated to all of the standard stars (typically four to six) observed on that night. The RV reported for a program star is the average of the RVs from cross-correlation with multiple standards taken that night (with standard star spectra that produce poor cross-correlations against the others removed from the average iteratively).

The spectrum for a typical program star observed with FOBOS is shown in Figure 3. The dominant features in M giants in the spectral band we are studying are the Mg b triplet, at 5167, 5173, and 5184 Å, and the Na D doublet, at 5889 and 5896 Å. A number of strong Fe, Cr, and Ti lines/blends are also present. The redder stars in our sample (J − K_S > 1.1) show very strong titanium oxide (TiO) bands in their spectra; these M giants are cool enough (T_eff < 3560 K) that the TiO bonds in the star are not dissociated. Figure 4 shows the spectra for several stars observed with FOBOS covering a range of (J − K_S) colors; the strength of the TiO bands increases as the giants become redder and cooler.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Additional 149 M giants that fit our survey criteria were observed over UT 2004 October 8–11 at CTIO using the Cassegrain spectrograph on the 1.5 m telescope. The detector was a Loral 1K (1200 × 800 pixels) CCD with a read noise of 6.5 e⁻; the gain was set to 1.42 e⁻ per ADU. A grating with 831 grooves mm⁻¹ was used, with a resolution of Δλ ∼ 3.1 Å. Helium and argon lamps were used to take comparison frames for each target at the same telescope position to account for flexure variations. Ten quartz frames were taken each night and combined and normalized. The spectral range is 7650–8900 Å. The dominant feature in this region is the Ca ii triplet at 8498, 8542, and 8662 Å. The CTIO M giants have magnitudes in the range 6.3 < (K_S)₀ < 12.0. The reduction procedures for the CTIO program giants are similar to those used for the FMO reductions (the same cross-correlation code was used to determine the RVs).

For the FMO sample, the difference in the derived (from cross-correlation against each other) and published RVs for the standard stars is typically ±1–5 km s⁻¹, with no systematic offset in either direction. A total of 102 program stars were observed multiple times at FMO to test the stability of FOBOS and to gauge the S/N threshold for obtaining reliable RVs. The mean of the absolute value of the differences in the RVs for stars with multiple observations at FMO is 8.5 km s⁻¹. The RVs are fairly stable even at low S/N: stars with S/N below 20 have a mean absolute value in the difference of their RVs of 10.5 km s⁻¹. For the CTIO sample, the difference in the derived and published RVs for the standard stars is slightly higher, with variations ranging from ±1–10 km s⁻¹; as with the FMO standards, no systematic offset in either direction is seen. Repeat observations were also taken for 16 stars at CTIO. The mean of the absolute value of the differences in RVs for the CTIO repeat observations is 13.9 km s⁻¹. For a subsample of 34 stars, RVs were also determined from high-resolution echelle spectra (see Section 4.1). In Table 1, the RVs found from the high- and medium-resolution spectra for these 34 stars are reported (high-resolution/medium-resolution, denoted as v_{hel, h}/v_{hel, m}, respectively). The mean of the absolute value in the differences between the medium-resolution RVs and the high-resolution RVs is 6.2 km s⁻¹. Overall, considering the random errors in the medium-resolution RVs, the comparison of the medium/high-resolution values, and that the data set is dominated by FOBOS observations, we place the typical uncertainty level for the medium-resolution RVs at 5–10 km s⁻¹.

To test the chemical properties of the 34 selected stars, high-resolution echelle spectra were collected using the Astrophysical Research Consortium Echelle Spectrograph (ARCES) on the 3.5 m telescope at Apache Point Observatory on UT 2009 March 30 and UT 2009 April 2 and the CCD echelle spectrograph (ECHLR) on the 4 m Mayall telescope at the Kitt Peak National Observatory over UT 2010 February 26–March 2.

The ARCES uses a 2048 × 2048 pixel SITe CCD and has a resolution of R = 31, 500; the CCD has a gain of 3.8 e⁻ per ADU and a readout noise level of about 7 e⁻. Two sets of flat fields were taken to account for the strong gradient in response across the ARCES orders: one long set with a blue filter inserted and another short set with no filter (these are the red calibration frames). The blue and red frames were combined to create a master quartz flat. For wavelength calibration, thorium–argon (ThAr) lamp frames were taken.

Reduction of the ARCES data was carried out using various IRAF tasks in the echelle package. All images were overscan corrected and trimmed using ccdproc. The echelle orders were located and the trace defined for the spectra with apall. The task ecidentify was used to identify lines in a ThAr lamp spectrum. To minimize the effects of aliasing, the spectra were resampled along the dispersion axis using the magnify task. Scattered light was removed from the program star frames using the apscatter task, and the relevant echelle orders were then extracted. The program spectra were divided by the extracted master quartz flat, and the dispersion correction defined using the ThAr lamps was then applied to convert the pixel scale to a wavelength scale. As a final step in the reduction process, the spectra were continuum normalized.

ECHLR spectra were collected at KPNO using the 2048 × 2048 pixel T2KB CCD on the 4 m Mayall telescope. The gain setting for T2KB was 1.9 e⁻ per ADU with a read noise of approximately 4 e⁻. The ECHLR has a resolution of R = 35, 000. ThAr lamps were used to collect wavelength calibration frames, and the short-exposure quartz lamp was used for obtaining the flat-field images. The spectral reduction procedures for the ECHLR data are similar to those carried out for the ARCES data.

The photometric properties and the observational details of the stars observed at APO and KPNO are listed in Table 1. The IDs and photometry of the stars come from the 2MASS point-source catalog, where the IDs are the 2MASS R.A./decl. (J2000.0) coordinates. RV standards, taken from the Astronomical Almanac, were observed all nights at APO and KPNO. RVs from the echelle data were found using the IRAF fxcor task and are reported in Table 1 and are compared with the medium-resolution values. Cross-correlation between RV standards gives errors on the order of 0.5 km s⁻¹ for the high-resolution RVs. Based on the RV data in Table 1, there is apparently a systematic bias toward higher measured RVs for the medium-resolution spectra, such that v_{hel, h} − v_{hel, m} = −5.0 km s⁻¹. This offset may be due to variations in the centering of the star in the slit for the echelle data (FOBOS data are taken with fiber optics, which, due to radial and azimuthal scrambling, provide more uniform "slit functions" in the spectrograph).

4.2.1. Derivations

The APO and KPNO instrument setups give the best S/N per pixel for the region around 7400 Å. This particular region of the spectrum was selected due to its relative absence of molecular bands, which offers a good window for spectral analysis (Smith & Lambert 1990). The procedures used to derive the metallicities are similar to those used by Chou et al. (2007) in their study of M giants in the Sgr tidal tails. Equivalent widths (EWs) for 11 Fe i lines in the range 7443–7583 Å were used to determine the atmospheric parameters T_eff, log g, and [Fe/H] and the microturbulent velocity (ξ). The EWs were measured manually using the IRAF splot task.

The wavelengths of the 11 Fe i lines and their corresponding EW measured for each star are listed in Table 2. In some cases, an EW could not be measured due to a cosmic ray falling on the same pixel as an Fe line; these lines were removed from the list for that spectrum and are reported as "..." in Table 2. We adopt the same values for the excitation potentials (χ) and oscillator strengths (gf) for these lines as those used by Chou et al. (2007). Along with the EW measurements, stellar atmosphere models from the Kurucz ATLAS9 grids (Kurucz 1994) are used. Each stellar atmosphere model corresponds to a particular (T_eff, log g, [Fe/H]). We start with an initial guess for (T_eff, log g, [Fe/H], ξ)₀. The EWs for the sample lines and the model atmosphere are used as input to the MOOG local thermodynamic equilibrium (LTE) code (Sneden 1973). The starting value for T_eff is calculated from the Houdashelt et al. (2000) color–temperature relations for M giants, using the dereddened 2MASS J − K_S colors converted to the CTIO system and the relations of Carpenter (2001). An initial guess for log g is taken from the 10 Gyr Z_☉ isochrone from the Padova evolutionary track database (Marigo et al. 2008 ¹⁴). The surface gravity g of a star is related to its mass and size (luminosity). Once a star exhausts the hydrogen supply in its core, it will first brighten and move up the red giant branch (RGB); later, after core He exhaustion, the star eventually ascends the asymptotic giant branch (AGB). The difference in log g between a star on the RGB versus the AGB for cool giants, however, is quite small over a wide range of ages. Figure 6, which shows the T_eff–log g plane for 3 Gyr, 5 Gyr, and 10 Gyr solar metallicity (Z_☉ = 0.019) isochrones, demonstrates that the variation in log g between the RGB and AGB is on the order of 0.15 dex for stars in the range of effective temperatures for our sample, which is 3600–4000 K. The iterative scheme for determining abundances involves using the initial values of T_eff, log g, and [Fe/H] and iterating these values until the derived and model values of [Fe/H] converge. After each iteration, the correlation coefficient for A(Fe) as a function of the reduced EW (i.e., RW—the measured EW divided by the λ required for that atomic transition) is checked and ξ is adjusted to minimize the correlation coefficient before proceeding to the next iteration. An incorrect value of ξ leads to a physically unrealistic dependence of the elemental abundance on the EWs—as seen by a high correlation coefficient for the RW. The derived parameters for all program stars are listed in Table 3.

Zoom In Zoom Out Reset image size

Table 2. Atomic Lines and Equivalent Widths

ID	7443.02	7447.38	7461.52	7498.53	7507.26	7511.02	7531.14	7540.43	7547.91	7568.89	7583.79	7474.94	7489.57	7496.12
	(Å)	(Å)	(Å)	(Å)	(Å)	(Å)	(Å)	(Å)	(Å)	(Å)	(Å)	(Å)	(Å)	(Å)
	Fe i	Fe i	Fe i	Fe i	Fe i	Fe i	Fe i	Fe i	Fe i	Fe i	Fe i	Ti i	Ti i	Ti i
1521021+082320	53.5	34.5	87.7	...	...	162.4	100.1	56.9	20.0	94.2	147.0	26.7	56.5	42.1
1546044+061554	50.0	34.2	86.1	33.8	66.2	171.6	98.2	51.0	...	92.0	144.4	27.9	48.6	33.5
1535178+135331	54.3	52.1	104.5	51.7	...	168.5	110.3	58.3	32.0	99.0	156.6	26.7	104.7	...
1640358+060251	83.4	59.5	109.8	65.6	101.0	193.0	120.8	67.5	31.0	108.4	159.8	77.7	113.8	91.6
1509307+252912	40.1	31.2	80.4	28.1	52.6	...	94.2	39.2	17.8	...	135.4	17.2	37.5	...
1546468+270052	64.0	36.8	98.5	45.4	87.2	180.6	115.3	58.1	31.5	108.9	157.5	43.2	72.4	57.5
1530257+323409	63.0	47.1	103.8	52.6	70.6	171.1	101.3	60.0	24.2	94.6	147.0	89.0	115.0	102.5
1719596+301146	71.3	57.0	101.4	...	89.8	174.5	114.9	65.0	31.8	109.5	155.2	63.2	117.7	86.0
1731343+401543	88.0	60.4	130.1	71.4	108.5	201.0	130.8	76.0	35.9	123.2	175.6	95.7	138.2	107.8
0913092+450916	62.9	41.6	90.3	40.9	77.0	166.5	108.0	47.9	...	99.8	143.9	27.0	60.8	43.0
0807206+435418	...	60.0	131.2	77.0	113.3	207.9	137.5	79.0	45.4	130.1	186.3	99.9	136.2	110.8
0935556+414046	46.2	26.3	75.1	27.3	59.4	154.2	88.3	45.0	19.9	82.2	130.8	15.4	35.9	25.5
0903471+414944	60.0	43.5	91.8	50.1	77.0	161.1	102.0	47.3	20.0	96.4	140.8	54.2	83.8	70.3
0830244+283812	49.0	34.5	85.0	36.3	66.3	159.3	98.1	46.8	14.2	91.8	145.1	27.4	52.7	48.4
1011498+230504	55.1	34.0	94.5	40.0	73.0	168.0	114.0	...	21.0	93.1	153.9	30.7	63.0	49.0
0918573+145749	53.6	21.1	76.1	...	64.0	163.1	88.7	36.7	12.6	88.3	...	19.1	39.3	22.9
0938204+112950	60.5	37.4	94.6	39.1	78.5	165.4	107.3	54.4	21.6	105.4	...	37.1	76.5	65.4
1001593+114507	66.8	44.3	94.3	...	87.0	182.6	109.5	55.0	17.4	97.3	150.9	53.6	93.0	83.2
0932038+055521	67.5	45.0	94.1	43.8	74.7	162.9	102.3	59.7	33.8	98.7	149.0	77.5	105.7	97.8
0950147+082356	55.6	49.5	105.5	47.3	...	156.7	...	55.9	22.6	90.2	140.3	84.7	109.2	96.1
1101590+084043	41.9	...	71.8	...	60.0	143.0	100.7	44.2	18.7	80.6	147.2	...	43.8	33.0
1024571+004900	62.5	37.3	105.8	41.6	74.3	169.5	114.8	60.0	21.7	95.4	165.0	52.8	84.5	69.0
1101024−003004	42.2	25.1	81.8	24.8	61.3	158.7	92.3	...	...	80.4	139.1	16.5	41.0	30.4
1115376+000800	63.6	45.5	95.3	...	74.0	162.8	...	60.2	26.5	96.4	145.5	74.2	104.7	94.3
1054035−065124	59.1	35.4	97.0	40.8	74.0	158.0	93.8	53.7	17.5	89.1	147.6	58.2	95.4	81.4
1037414−121048	55.8	45.0	102.7	39.6	70.3	151.8	96.6	63.0	21.2	94.2	...	85.2	105.4	93.5
1136527+025949	44.5	26.3	81.3	27.4	66.2	162.8	95.9	35.8	15.6	85.6	146.6	20.6	45.5	35.7
1145072+013727	55.6	...	99.9	...	85.2	171.2	110.0	60.9	25.5	93.4	145.5	67.2	92.8	82.7
1105038−164000	48.0	20.9	86.2	21.7	76.2	175.3	106.6	37.9	...	87.3	150.1	30.4	62.1	45.5
1131243−055825	45.2	28.2	96.0	29.1	67.3	134.5	...	50.2	21.9	71.7	132.1	84.8	107.2	93.2
1206183−035045	49.6	36.9	94.3	42.5	72.4	158.5	101.6	56.5	19.6	99.7	152.4	60.0	91.7	80.0
1313176−164220	66.9	50.1	103.6	53.5	87.3	194.6	115.5	66.3	43.8	103.6	175.1	72.3	110.0	96.8
1314125−132352	66.7	43.0	93.0	52.1	84.1	173.5	106.4	56.5	26.7	97.2	147.2	58.7	95.3	75.1
1334567−055158	65.8	50.7	95.7	48.5	79.9	170.6	110.7	67.2	25.2	97.1	149.7	85.2	113.7	98.3
1355160−053350	50.0	32.0	73.1	25.1	57.0	146.5	89.3	...	...	80.6	149.4	53.8	69.0	64.4
1442035−162350	42.5	33.4	96.3	32.4	64.9	154.1	91.9	46.2	13.3	87.3	133.6	40.5	74.6	60.9
1404515−004157	56.2	33.2	112.4	45.1	70.6	171.2	113.5	63.2	20.6	104.3	167.4	84.1	113.7	91.7
1431302−051730	46.8	28.5	90.5	31.3	70.3	164.4	104.6	39.5	...	...	143.6	...	39.2	31.1
1518065−115536	55.7	44.5	98.3	41.9	75.6	161.5	103.4	56.7	...	100.0	136.8	55.8	84.8	69.4
Arcturus (APO)	72.0	45.9	91.7	...	96.1	193.4	116.9	56.5	29.4	108.1	150.8	38.1	77.7	57.6
Arcturus (KPNO)	65.5	45.5	94.6	31.0	89.3	185.0	115.8	48.0	31.6	106.6	148.4	35.9	72.9	54.1
α Tau (KPNO n1)	90.3	63.4	120.8	62.3	115.4	212.7	132.4	77.3	45.3	129.0	179.0	99.6	127.9	109.1
α Tau (KPNO n4)	95.7	65.8	120.0	73.5	113.9	211.0	132.2	78.5	45.8	121.6	181.0	90.4	124.4	104.0
δ Vir (APO)	83.0	62.7	119.9	73.8	103.1	193.3	...	76.4	40.1	112.2	164.8	99.3	126.8	121.3
δ Vir (KPNO)	76.1	54.3	117.4	68.0	95.6	188.5	...	69.7	44.4	112.0	164.2	99.9	131.5	120.6
ν Vir (APO)	73.4	50.7	101.7	59.2	85.7	171.8	110.7	60.9	29.3	98.7	155.4	76.3	109.3	93.4
ν Vir (KPNO)	66.9	51.2	97.0	57.2	82.8	175.3	107.4	60.3	25.4	93.0	149.9	76.4	103.9	92.5

Download table as:  ASCIITypeset image

Table 3. Derived Properties of the Program Stars

ID	Teff	log g	ξ	A(Fe)	[Fe/H]	e[Fe/H]	A(Ti)	[Ti/Fe]	e[Ti/Fe]	Distance
	(K)		(km s−1)							(kpc)
1521021+082320	3900	0.4	1.49	6.50	−0.95	0.07	3.85	−0.10	0.11	5.1
1546044+061554	3900	0.5	1.58	6.44	−1.01	0.11	3.76	−0.13	0.17	5.9
1535178+135331	3800	0.8	1.38	7.10	−0.35	0.14	4.14	−0.41	0.45	2.3
1640358+060251	3800	0.9	1.54	7.30	−0.15	0.16	4.61	−0.14	0.04	1.3
1509307+252912	3900	0.3	1.33	6.34	−1.11	0.13	3.56	−0.23	0.17	8.6
1546468+270052	3900	0.6	1.62	6.77	−0.68	0.09	4.08	−0.14	0.14	3.3
1530257+323409	3800	0.6	1.38	6.90	−0.55	0.15	4.70	0.35	0.12	2.2
1719596+301146	3900	1.2	1.38	7.30	−0.15	0.10	4.74	−0.01	0.12	1.8
1731343+401543	3800	1.1	1.72	7.44	−0.01	0.14	4.88	−0.01	0.04	0.9
0913092+450916	3950	0.7	1.48	6.74	−0.71	0.10	3.96	−0.23	0.08	4.9
0807206+435418	3900	1.3	1.72	7.52	0.07	0.13	5.01	0.04	0.07	1.3
0935556+414046	4000	0.5	1.33	6.37	−1.08	0.11	3.73	−0.09	0.13	2.6
0903471+414944	3900	0.6	1.42	6.71	−0.74	0.14	4.29	0.13	0.13	2.5
0830244+283812	3950	0.2	1.55	6.37	−1.08	0.10	3.95	0.13	0.18	3.2
1011498+230504	4000	0.7	1.63	6.57	−0.88	0.08	4.10	0.08	0.12	5.0
0918573+145749	4000	0.4	1.79	6.19	−1.26	0.14	3.75	0.11	0.16	8.6
0938204+112950	3900	0.5	1.56	6.63	−0.82	0.08	4.11	0.03	0.13	6.8
1001593+114507	4000	0.5	1.75	6.50	−0.95	0.13	4.47	0.52	0.15	6.6
0932038+055521	3700	0.9	1.20	7.33	−0.12	0.11	4.73	−0.05	0.13	1.6
0950147+082356	3600	0.5	1.35	7.12	−0.33	0.16	4.52	−0.05	0.10	1.6
1101590+084043	4000	0.4	1.47	6.31	−1.14	0.13	3.82	0.06	0.16	8.9
1024571+004900	3800	0.3	1.65	6.57	−0.88	0.09	4.05	0.03	0.13	4.7
1101024−003004	4000	0.3	1.59	6.21	−1.24	0.10	3.83	0.17	0.12	4.2
1115376+000800	3800	0.6	1.32	6.91	−0.54	0.08	4.55	0.19	0.11	6.2
1054035−065124	3800	0.0	1.42	6.46	−0.99	0.13	4.25	0.34	0.10	4.9
1037414−121048	3650	0.1	1.28	6.82	−0.63	0.12	4.41	0.14	0.13	3.0
1136527+025949	3950	0.4	1.69	6.27	−1.18	0.10	3.79	0.07	0.14	4.8
1145072+013727	3800	0.4	1.49	6.71	−0.74	0.08	4.27	0.11	0.16	4.5
1105038−164000	4000	0.3	2.05	6.12	−1.33	0.08	3.88	0.31	0.13	4.8
1131243−055825	3650	0.0	1.19	6.54	−0.91	0.13	4.40	0.41	0.10	3.5
1206183−035045	3700	0.1	1.47	6.57	−0.88	0.10	4.06	0.04	0.12	3.7
1313176−164220	3900	1.0	1.58	7.13	−0.32	0.15	4.66	0.08	0.10	5.4
1314125−132352	3950	0.9	1.40	6.92	−0.53	0.11	4.50	0.13	0.07	2.8
1334567−055158	3750	0.8	1.36	7.16	−0.29	0.09	4.69	0.08	0.08	2.8
1355160−053350	3800	0.1	1.47	6.28	−1.17	0.16	3.98	0.25	0.19	4.7
1442035−162350	3800	0.2	1.50	6.36	−1.09	0.12	3.93	0.12	0.11	4.8
1404515−004157	3700	0.1	1.76	6.56	−0.89	0.10	4.23	0.22	0.12	3.6
1431302−051730	4000	0.5	1.69	6.36	−1.09	0.08	3.75	−0.06	0.19	6.8
1518065−115536	3800	0.5	1.28	6.83	−0.62	0.10	4.19	−0.09	0.10	4.8
Arcturus (APO)	4250	1.4	1.73	6.88	−0.57	0.09	4.62	0.29	0.10
Arcturus (KPNO)	4250	1.4	1.76	6.82	−0.63	0.09	4.56	0.29	0.11
α Tau(KPNO n1)	3900	1.3	1.68	7.50	0.05	0.11	4.98	0.03	0.12
α Tau(KPNO n4)	3900	1.3	1.50	7.62	0.17	0.13	4.99	−0.08	0.06
δ Vir(KPNO)	3650	0.8	1.29	7.63	0.18	0.14	5.11	0.07	0.11
δ Vir(APO)	3700	0.8	1.35	7.62	0.17	0.14	5.05	0.03	0.15
ν Vir(KPNO)	3800	0.8	1.40	7.11	−0.34	0.15	4.58	0.02	0.12
ν Vir(APO)	3800	0.8	1.36	7.19	−0.26	0.14	4.64	0.03	0.07

Download table as:  ASCIITypeset image

In addition to [Fe/H], the ratio [Ti/Fe] was also derived. Once the atmospheric parameters for a star were determined using MOOG, these were fixed and used to measure A(Ti). Three Ti i lines were used to derive A(Ti): 7474.940, 7489.572, and 7496.120 Å. A solar gf value was derived for the Ti i line at 7474.94 Å and adopting A(Ti)_☉ = 4.90 from Asplund et al. (2005). The gf values for the other two Ti i lines (7489.57 and 7496.12 Å) were taken from Chou et al. (2007), and these are also solar gf values.

As a check of our methodologies, four calibration stars from the red giant spectral studies of Smith & Lambert (1985) and Smith et al. (2000) (collectively referred to as S&L) were observed with both the ARCES and ECHLR; a comparison between our [Fe/H] and [Ti/Fe] values for the calibration stars is given in Table 4. Typical uncertainties in the values derived for [Fe/H] in both studies are ∼ ±0.15 dex using this sample of Fe i lines. The mean difference in [Fe/H], and its associated standard deviation, between these two studies, in the sense of (this study − S&L), is −0.04 ± 0.16. This comparison indicates that the two studies are on the same abundance scale, with only a small mean offset and a scatter typical of the derived uncertainties in [Fe/H]. The values are in good agreement, and the discrepancies are due to differences in the stellar atmosphere models used.

Table 4. Standard Star Chemical Abundance Comparisons

Star	[Fe/H]	$[{\rm Fe/H}]_{\rm S\&L}$	[Ti/Fe]	[Ti/Fe]$_{\rm S\&L}$
Arcturus	−0.60 ± 0.09	−0.67 ± 0.11	0.29 ± 0.11	0.41 ± 0.12
α Tau	0.11 ± 0.10	0.07	−0.03 ± 0.08	−0.03
ν Vir	−0.30 ± 0.15	−0.02 ± 0.20	0.03 ± 0.20	0.06 ± 0.17
δ Vir	0.18 ± 0.17	0.13 ± 0.17	0.05 ± 0.13	0.07 ± 0.20

Download table as:  ASCIITypeset image

Although detailed non-LTE (NLTE) calculations have not been carried out here, limits to the simplifying assumption of LTE can be investigated using recent studies of NLTE line formation for iron by Bergemann et al. (2011) and for titanium by Bergemann (2011). As is the case for many NLTE calculations involving cool stars, the theoretical results depend on the choice of the value for the parameterized efficiency of neutral hydrogen collisions, S_H. In the case of Fe i/Fe ii, Bergemann et al. (2011) computed NLTE/LTE results using a small grid of model atmospheres and find that for Fe i in general, corrections to LTE-based abundances will become larger for increasing T_eff, decreasing surface gravity, or decreasing metallicity; the Fe i corrections increase most dramatically for decreasing [Fe/H] (see their Figure 3), especially for [Fe/H] ⩽ −2.0, and for warmer temperatures, T_eff ⩾ 5000 K (where Δ(NLTE – LTE) ⩾ +0.2 dex). The sample of M giants analyzed here is only moderately metal-poor ([Fe/H] ⩾ −1.3) and, when coupled to cooler effective temperatures, would be expected, based on the Bergemann et al. (2011) results, to require corrections to LTE Fe i abundances that would be ⩽+0.2 dex. Although NLTE corrections to LTE abundances in cool giants are not expected to be large, NLTE theoretical calculation of corrections that span the stellar parameter space analyzed here would be welcome.

The Ti i corrections to LTE from Bergemann (2011) are in the same sense as and qualitatively similar to those for Fe i discussed above. Given the stellar parameters and metallicities covered by the sample here, any corrections to [Ti/H] are not significant and, in particular, the critical values of [Ti/Fe] will have even smaller corrections from assuming LTE, with values ⩽0.10 dex; such uncertainties have no significant effect on conclusions drawn from the Fe and Ti abundances calculated here. As noted in Bergemann et al. (2011), their goal is to establish interactive routines to allow for estimating corrections to LTE-based abundances, so in the near future, it may be possible to provide more accurate corrections to LTE given particular stellar parameters and metallicities.

Observations of stars in clusters also support the theoretical calculations that suggest rather small departures from LTE for the Fe abundances in giants. Ramírez et al. (2001) analyzed stars in the mildly metal-poor globular cluster M71 ([Fe/H] ∼ −0.5) and found the same values of [Fe/H] (within ∼0.05 dex) for turnoff stars, subgiants, and giants (which span a range in T_eff from 6000 to 4500 K and log g from 4.1 to 1.5). These results support the smallish corrections to LTE Fe abundances suggested by the Bergemann et al. (2011) calculations.

4.2.2. Distances

In this study, we have analyzed the spectroscopic properties of a sample of M giants that are dominated by members of the Milky Way's thick disk and nearby halo components. From the RV distribution, we identified stars with RVs that lie outside those expected for typical thick disk stars at the same locations. These RV outliers are found to show some degree of spatial and kinematical coherence. We suppose that some of this coherence could be the signature of substructure (e.g., tidal tails) from accreted satellites.

To test our interpretation of the RV outliers, we looked at the chemical abundance patterns of 34 of these stars. We used the locations of the stars in the [Ti/Fe] versus [Fe/H] plane to attempt to assign them to one of the three potential populations of halo stars—those formed in situ, those that were accreted, and those that were kicked out of the disk. This cannot be done unambiguously in all cases because of expected overlaps in the chemical signatures of these populations. However, the [Ti/Fe] abundances for 17 of the RV outliers are systematically below the main halo trend and similar to the abundances seen in stars from Milky Way dwarf satellite galaxies. They are consistent with stars from MW dwarf satellite galaxies that have been accreted by the Milky Way (Shetrone et al. 2003; Tolstoy et al. 2004; Monaco et al. 2007) and inconsistent with expectations for abundance patterns from pure in-situ-halo or kicked-out growth. Another eight of the stars selected for high-resolution spectroscopic follow-up track the abundance trends set by the disk stars, even though they have halo-like kinematics. These are indicative of a population formed in the disk or bulge and subsequently kicked-out, although (1) some or all may have been accreted from a high-mass infalling object and (2) we estimate an upper limit to contamination by genuine disk stars even at these high RVs at a level that does not allow this association to be conclusive. The remainder of the high-resolution stars have high [Ti/Fe] at low [Fe/H] and could be part of any of the in-situ-halo, kicked-out, or accreted populations.

While it has had a long history (Hartwick 1987; Zinn 1993, 1996; Sommer-Larsen et al. 1997), the discussion of possible multiple mechanisms for the formation of the stellar halo has been revitalized in recent years by both the advent of large photometric data sets with follow-up low-resolution spectroscopic work (e.g., Carollo et al. 2007, 2010) and more modest but ever-expanding samples of nearby halo stars with high-resolution spectroscopy (e.g., Nissen & Schuster 2010, 2011; Schuster et al. 2012; Ishigaki et al. 2012). In several cases, the studies point out classes of halo populations with distinct properties that are argued to match expectations for the properties of populations with distinct formation histories seen in hydrodynamical simulations of galaxy formation: the models generally predict an inner halo (within ∼20–30 kpc) dominated by metal-rich stars formed within the main Galactic progenitor and an outer halo dominated by lower metallicity, accreted stars (Abadi et al. 2006; Zolotov et al. 2009; McCarthy et al. 2012). However, comparisons of the models to the data sets up to this point are necessarily inconclusive because the results of the simulations themselves are dependent on the (hard to model!) details of when, how, and where stars form in the main Galactic progenitor.

For example, Carollo et al. (2007, 2010) claim evidence for two populations in their sample of 10,123 nearby (within 4 kpc) SDSS calibration stars: one of metal-rich stars (with a metallicity distribution peaking around [Fe/H] = −1.6) on only mildly eccentric orbits, and a second of metal-poor stars on more eccentric orbits. These observations are very reminiscent of the hydrodynamical simulation results—indeed Carollo et al. (2010) interpret their metal-rich stars as an inner, in-situ-halo/kicked-out population and their metal-poor stars as an outer, accreted population—but a transition in the orbital and metallicity properties in these populations is not necessarily inconsistent with a purely accreted stellar halo. Moreover, the Carollo et al. (2007, 2010) interpretation is at odds with our finding of a large fraction of clearly accreted stars in our M giant sample of even higher metallicity stars ([Fe/H] > −1.2) than their metal-rich population. In fact, we might expect the M giants to be biased toward finding in-situ-halo/kicked-out stars because late-type giants are a metal-rich stellar population. On the other hand, the selection of RV outliers admits a bias in our sample toward the high-velocity tails of all populations—particularly accreted stars, which are expected typically to be on higher energy and higher eccentricity orbits than the other stars. This could explain why our sample is particularly sensitive to the accreted population. Further work is needed to definitively show which of these two biases should dominate in a sample such as ours.

Our results are more similar to those of Nissen & Schuster (2010, 2011) and Schuster et al. (2012), who find roughly equal-size populations distinct in age, abundances, and orbits when they divide their (metal-rich, [Fe/H] > −1.6) stellar halo sample between low-α and high-α stars in the [Mg/Fe]–[Fe/H] plane. They interpret the properties of the low-α population (in which the stars are found also to be younger and on more eccentric orbits) as being consistent with an accreted population—so that these authors find an accreted fraction for the stellar halo at these metallicities similar to our own estimates. It is interesting to note that we also find stars with significantly lower [α/Fe] (less than zero) than any of the stars in the Nissen & Schuster (2010) sample, possibly because our large survey volume (out to 9 kpc from the Sun) encompasses debris from more chemically extreme accreted progenitors not represented in their local sample (which is limited to within 335 pc of the Sun).

We explore the properties of relatively nearby, RV-selected halo M giant stars and conclude that the chemical properties of the stars in this sample show tentative evidence of distinct populations with distinct formation mechanisms. While close to 50% of the stars fall in the accreted region of chemical abundance space, a definitive assessment of the relative contributions from in-situ-halo/kicked-out stars is not possible, due to the sometimes ambiguous categorizations of stars based on their [α/Fe]–[Fe/H] abundance distributions alone. Nissen & Schuster (2011) have demonstrated that using abundance patterns along with more complete orbital information can play a key role in making this identification. Our own chemodynamical data point to the importance of mapping a significant volume of the halo to confirm that such local studies are representative of global properties. Follow-up work on the kinematics and more detailed chemical characterization of these stars would give more insight into their origin. Quantifying the size of the kicked-out population more generally could provide valuable constraints on the hydrodynamical models of galaxy formation.

A key finding is that our selection of M giants with unusual, halo-like RVs picks out a stellar halo population dominated by accreted stars. Hence, searching for accretion events in RV-outlying samples of M giants should be prolific and motivates further interest in the putative groupings of stars found in our RV survey. In a companion paper we explore what more the kinematical properties of these groupings could be telling us about their origins (Johnston et al. 2012).