The Apache Point Observatory Galactic Evolution Experiment (APOGEE)

Modern astrophysics has taken two general observational approaches to understand the evolution of galaxies. On the one hand, increasingly larger aperture telescopes, on the ground and in space, give access to the high-redshift universe and offer "low-resolution" snapshots of ever earlier phases of galaxy evolution. On the other hand, increasingly efficient, multiplexing photometric and spectroscopic instrumentation, often on smaller, workhorse telescopes, has made possible enormous, definitive surveys of nearby galaxies, yielding a "high-resolution (HR)" view of the present state of these systems. These data can be tested against "end state" predictions for the growth of large structures in the universe to provide critical constraints on cosmological models—so-called "near-field cosmology". These two observational approaches—overviews of global properties at high redshift versus more detailed information at low redshift—provide complementary information that must be accommodated by evolutionary theories.

The highest-granularity information about galaxy evolution is provided by stars in our own Milky Way, whose present spatial distributions, ages, chemistry, and kinematics contain fossilized clues to its formation. Guided by detailed models for the chemical and dynamical evolution of stellar populations, critical telltale signatures and correlations within the above observables provide constraints on the model predictions for physical quantities that cannot be observed directly, such as the history of star formation, the early stellar initial mass function (IMF), and the merger history of Galactic subsystems. This "Galactic archaeology" remains the principal basis by which models for the formation and chemodynamical evolution of the Milky Way and analogous systems are formulated and refined. The vast literature on Milky Way stellar populations as tools for understanding Galactic evolution has been reviewed in the past by, e.g., Gilmore et al. (1989), Majewski (1993), and Freeman & Bland-Hawthorn (2002), and more recently by Ivezić et al. (2012), Rix & Bovy (2013), and Bland-Hawthorn & Gerhard (2016).

These efforts are of course greatly aided by access to expansive, carefully designed, homogeneous, and precise databases of properties for stellar samples that span large regions of the Galaxy and include all of the principal stellar populations. Modern archetypes of such databases are large photometric surveys like the Two Micron All-Sky Survey (2MASS; Skrutskie et al. 2006) and the Sloan Digital Sky Survey (SDSS; York et al. 2000). Over the past decade, these photometric catalogs have been widely exploited for insights into the nature of the Milky Way and probing the complexities of Galactic structure—e.g., halo substructure (e.g., Majewski et al. 2003; Rocha-Pinto et al. 2004; Belokurov et al. 2006; Grillmair 2009), satellite galaxies (e.g., Willman et al. 2005; Belokurov et al. 2007), the warp of the disk (e.g., López-Corredoira et al. 2002; Reylé et al. 2009), and the still unresolved, composite anatomy of the bulge (e.g., Robin et al. 2012), which includes the recently found X-shaped feature (e.g., McWilliam & Zoccali 2010; Nataf et al. 2010) and one or more central bars (e.g., Hammersley et al. 2000; Alard 2001; Cabrera-Lavers et al. 2007). Follow-on, low- and medium-resolution (MR) spectroscopic programs provide additional dynamical discrimination of, and context for, these structures as well as general information on their chemical makeup (e.g., mean metallicities and, in some cases, an additional dimension of chemistry, such as [α/Fe]); these broad brushstrokes represent an important step in characterizing stellar populations and constraining galactic evolution models.

Meanwhile, HR stellar spectroscopy has become an increasingly indispensable tool for providing the necessary detail to discriminate galaxy evolution models. Accurate multi-element chemical abundances provide insight into the stellar IMFs, and histories of star formation and chemical enrichment of stellar populations, which, in turn, fuel ever more sophisticated galactic dynamical and chemodynamical models (e.g., Chiappini et al. 2001, 2003; Sellwood & Binney 2002; Abadi et al. 2003; Bournaud et al. 2009; Schönrich & Binney 2009; Minchev & Famaey 2010; Bird et al. 2013; Minchev et al. 2013, 2014; Kubryk et al. 2015). Coupled with orbital information derived from precise radial velocities, these data probe the role of dynamical phenomena such as large-scale dissipative collapses, mergers, gas flows, bars, spiral arms, dynamical heating, and radial migration.

Conventional echelle spectroscopy programs to deliver HR spectroscopic data useful for Galactic archaeology demand substantial resources, often on the world's largest telescopes. Consequently, while heroic efforts have been devoted to surveying stars in a wide variety of environments—including, e.g., dwarf spheroidals, globular clusters, the Magellanic Clouds, tidal streams, and the Galactic bulge—until very recently the solar neighborhood was the only region for which multiple hundreds or thousands of observations had been assembled for "Galactic field stars" (e.g., Edvardsson et al. 1993; Bensby et al. 2003; Fuhrmann 2004; Venn et al. 2004; Nissen & Schuster 2010; Soubiran et al. 2010; Adibekyan et al. 2012, 2013; Bensby et al. 2014). These studies traditionally relied on kinematically selected samples to harvest from the nearby stars of accessible apparent brightnesses a broad spread of stellar ages and population classes. For stellar populations not represented in the solar neighborhood, like the Galactic bulge, and for in situ studies of field stars outside of the solar neighborhood, HR observations are only now generating samples with hundreds of stars. In the inner Galaxy where foreground dust obscuration is a formidable challenge, many previous samples were concentrated to a handful of low extinction sight lines, such as Baade's Window. Unfortunately, the aggregate of these piecemeal collections of spectroscopic data, heterogeneously assembled, can give a biased and incomplete view of the Milky Way.

Truly comprehensive evolutionary models for the Milky Way must be informed and constrained by statistically reliable, complete, or at least unbiased Galactic archaeology studies, which require the construction of large, truly systematic, and homogeneous chemokinematical surveys covering expansive volumes of the Milky Way and sampling all stellar populations, including, in particular, those dust-obscured inner regions where the bulk of the Galactic stellar mass is concentrated. A number of ambitious "Galactic archaeology" spectroscopic surveys that aim to fill this need (1) have been previously undertaken, such as RAVE (Steinmetz et al. 2006), SEGUE-1 (Yanny et al. 2009), SEGUE-2 (Rockosi et al. 2009), and ARGOS (Freeman et al. 2013), (2) are currently underway, such as LAMOST (Cui et al. 2012), Gaia/ESO (Gilmore et al. 2012), GALAH (Zucker et al. 2012), and Gaia (Perryman et al. 2001), (3) or are envisaged, e.g., those associated with the WEAVE (Dalton et al. 2014), 4MOST (de Jong et al. 2014), and MOONS (Cirasuolo et al. 2014) instruments. Although each of these surveys focuses on large samples of ≳100,000 stars, all of the past and ongoing endeavors are based on optical observations and are therefore strongly hampered by interstellar obscuration in the Galactic plane (Figure 1, bottom); this makes it challenging to sample significant numbers of stars within the very dusty regions of the Milky Way that are both central to constraining formation models and encompass most of the Galactic stellar mass (and some projects, like the RAVE, SEGUE, and GALAH surveys, specifically avoid low Galactic latitudes). Therefore, with optical wavelength surveys, it is challenging to assemble a systematic census having comparable or sufficient representation of all Galactic stellar populations and across wide expanses of the Galactic disk and bulge.

Zoom In Zoom Out Reset image size

While other surveys, such as BRAVA (Rich et al. 2007), ARGOS (Freeman et al. 2013), and Gonzalez et al. (2011) aim to fill at least part of this void by specifically focusing on the Galactic bulge, they utilize target selection criteria that differ from those of surveys of other parts of the Milky Way, which makes it difficult to generate a holistic picture of stellar populations and their potential connections. Moreover, apart from GALAH and the Gaia/ESO survey, these other studies are limited to MR spectroscopy (R < 10,000; Figure 1), and so they are unable to provide reliably the kind of detailed elemental abundance information that is now a key input to the models, while at the same time the moderate velocity precisions can limit their sensitivity to more subtle, second-order dynamical effects (e.g., perturbations by spiral arms and the bar, dynamical resonances, velocity-coherent moving groups and streams).

In contrast to previous and ongoing surveys, the Apache Point Observatory Galactic Evolution Experiment (APOGEE) in Sloan Digital Sky Survey III (SDSS-III) was designed to tackle the fundamental problem of galaxy formation through the first large-scale, systematic, precision chemical and kinematical study specifically optimized to include the exploration of the "dust-hidden" populations in the Milky Way. As planned, APOGEE aimed to build a database of HR (R ∼ 22,500), near-infrared (NIR; 1.6 μm H-band) spectra for over 10⁵ stars—predominantly red giant branch (RGB) and other luminous post-main-sequence stars—across the Milky Way, but with particular emphasis on obtaining significant representation from heavily dust-obscured parts of the Galactic disk and bulge. Operationally, this plan, now successfully executed, exploits several key advantages:

• a.  
  NIR observations profit from a selective extinction many times lower—for the H-band, a factor of 6 in magnitudes (i.e., a factor of 250 times in flux) than at visual wavelengths.
• b.  
  HR spectra provide the chemical abundance and radial velocity precision needed for constraining Galactic evolutionary models and, in the H-band, sample lines of numerous elements up to and including the iron peak and for which non-LTE departures are typically small.
• c.  
  Collectively, RGB stars, and asymptotic giant branch (AGB) and red supergiant (RSG) stars are good tracers of the disk, bulge, and halo; together, they sample populations of all ages and metallicities, and are luminous in the NIR. Meanwhile, the high sky density of these stellar types—combined with the large, 3° diameter Sloan 2.5 m telescope field of view (FOV) and high throughput, multifiber plugplate handling system—allows the simultaneous observation of several hundred targets at a time, and thousands per night.

Together, these advantages translate into a Milky Way survey trade-space "sweet spot" that permits efficient, high resolution, NIR spectroscopic study of large numbers of stars that are not easily accessible to optical programs, and enables a consistent database of stellar spectra to be assembled across all Galactic stellar populations, from the inner bulge to the more distant Galactic halo. Thus, with APOGEE, it is possible for the first time to explore and compare with great statistical significance the chemokinematical character of all Milky Way stellar subsystems using the same set of chemical elements and line transitions represented in data of uniformly high quality that has been gathered, processed, and analyzed identically.

The principal scientific goals of APOGEE, which together provide a broad but integrated approach to furthering our understanding of galaxy evolution, are:

• a.  
  To measure high-precision abundances for multiple elements in $\sim {10}^{5}$ stars across the Galaxy, and derive distributions of these chemical properties to constrain Galactic chemical evolution (GCE) models. Among the target elements are the preferred GCE tracers and most common metals—i.e., carbon, nitrogen, and oxygen—as well as other metals with particular sensitivity to the star formation history (SFH) and the IMF of stellar populations.
• b.  
  To derive high-precision kinematical data useful for constraining dynamical models for the disk, bulge, bar, and halo, and for discriminating substructures within these components.
• c.  
  To access the often ignored, dust-obscured inner regions of the Galaxy, and for the observed stars in these regions derive the same data as are available for other, more accessible stellar populations, which will also be included in the survey; furthermore, by collecting large survey samples, provide statistically reliable measures of chemistry and kinematics in dozens of Galactic zones (R, θ, Z) at the level currently available in the solar neighborhood.
• d.  
  To contribute to explorations of the early Galaxy by inferring the properties of the earliest stars, thought to reside or to have resided within a few kiloparsecs of the Galactic center (Tumlinson 2010). This can be achieved either by detecting them directly if they survive to the present day, or (more likely) by measuring their nucleosynthetic products in the most metal-poor stars that do survive.
• e.  
  To achieve a dramatic (more than two orders of magnitude) leap in the total number of available high resolution, high signal-to-noise ratio (S/N) infrared stellar spectra, which will enable substantial advances in stellar astrophysics, GCE modeling, and dynamical modeling of the Milky Way.

Among the more specific issues that APOGEE addresses are:

• a.  
  Completing the first systematic determination of the 3D chemical abundance distribution—for numerous elements—across the Galactic disk, determining the Galactic rotation curve, and examining correlations between abundances and stellar kinematics at all disk radii.
• b.  
  Determining the distribution functions of chemical abundances for a variety of elements in the bulge, bar(s), and inner disk, and probing correlations between abundances and kinematics there, with the goal of investigating the physical mechanisms that connect these structures and determining the origin(s) of the bulge.
• c.  
  Establishing the nature of the Galactic bar(s) and spiral arms and their influence on the disk through detailed assessment of abundances and velocities of stars in and around them.
• d.  
  Assessing the properties that discriminate the thin and thick disks to clarify the nature and origin of the latter.
• e.  
  Drawing a comprehensive picture of the chemical evolution of the Galaxy via the placement of strong constraints on the IMF and star formation rates of the bulge, disk, and halo as a function of position and time.
• f.  
  Searching for and probing the chemistry and dynamics of low-latitude substructures in both the disk and halo, whether from dynamical resonances or the accretion of satellites.
• g.  
  Investigating the kinematics and chemistry of the Galactic halo and its substructure, and using them to assess the relative contribution of accreted versus stars formed in situ.
• h.  
  By reference to other available optical, near-IR, mid-IR and radio data, exploring the interstellar medium, mapping the Galactic dust distribution, and constraining variations in the interstellar extinction law.
• i.  
  By combining spectroscopic data with the detailed information on stellar structure provided by asteroseismology surveys, deriving accurate ages for Galactic field stars, which provide key timestamps for the exploration of all manner of Galactic evolutionary phenomena.
• j.  
  Through the marriage of accurate stellar parameters and detailed chemical compositions from APOGEE with accurate asteroseismological data, providing fundamental constraints on models of the structure of stellar interiors, opening doors to progress in important areas of stellar physics.

The goal of the present paper is to give a broad overview of the APOGEE survey, with particular focus on the scientific motivations and technical rationale that led to the instrument and survey design choices (Section 2). The "birds-eye" descriptions of the APOGEE project given here are at a level intended to give the potential user of APOGEE data sufficient background to understand the basic structure of the instrument (Section 3) and survey (Section 4), how the data were collected (Section 5) and processed (Section 6), and what the data look like and how they may be accessed (Section 8). We also summarize how the survey met its intended goals (Section 7), further illustrated via several science demonstrations (Section 7.4). The latter also introduce some of the variety of applications to which APOGEE data may be directed. Based on the success of the APOGEE project, a new collaboration has been formed to expand upon this initial survey via the APOGEE-2 project; these and related future efforts are discussed briefly in Section 9.

This paper is a primer for those interested in a general understanding of the overall structure of the APOGEE survey. For more details on particular aspects of the survey, users are encouraged to consult a series of focused technical papers that address various specific elements of the survey, the software and algorithms used to produce the publicly released databases, and post-survey assessments of the calibration and accuracy of the data (Table 1). These papers and other survey documentation are described further in Section 8.4. Online information describing the data release file formats and available online tools for data visualization and download may be found at http://www.sdss.org.

Recent technology development has made high resolution NIR spectroscopy a new and vigorous area of astrophysical investigation, particularly in the area of stellar atmospheres analysis. The value and promise of high resolution NIR spectroscopy for exploring stellar abundances is attested by the growing number of papers on the subject over the past decade using instruments suitable for the purpose on the world's largest telescopes—e.g., CRIRES on the VLT, NIRSPEC at Keck, IRCS at Subaru, and, formerly, Phoenix at Gemini-South (e.g., Rich & Origlia 2005; Cunha & Smith 2006; Cunha et al. 2007; Ryde et al. 2010; Tsuji & Nakajima 2014). While the flow of high resolution NIR data has recently seen a dramatic upturn, the study of stellar photospheres on the basis of NIR spectroscopy has a long tradition (e.g., see the early review by Merrill & Ridgway 1979). The current state of the art in interpreting these data is proving highly successful, competitive with, and complementary to, traditional analyses in the optical (see references below).

To probe the largest distances in the Galaxy most easily, one should focus on the intrinsically brightest population tracers. A particular advantage realized by working in the NIR is that the intrinsically brightest common stars found in differently aged populations—RGB, AGB, and RSG stars (collectively referred to as "giants" throughout this paper)—all have cool atmospheres and are even brighter in the infrared than at optical wavelengths. Moreover, selecting for red stars in dereddened color–magnitude diagrams (CMDs) made from a magnitude-limited survey like 2MASS guarantees a virtually giant-dominated sample. Fortunately, the analysis of giant star atmospheres is an area that has received particular attention in high resolution NIR spectroscopy, given that these stars are the most accessible in star clusters, resolved galaxies (like the Magellanic Clouds), and fields toward the Galactic Center, like Baade's Window. The earlier papers by Smith & Lambert (1985, 1986, 1990) focusing on the CNO abundances in red giant stars were among the first efforts to explore chemical abundances from high-resolution spectra in the infrared. More recently, the analysis of high-resolution spectra in the H-band for stars in the Magellanic Clouds as well as the Galactic bulge and center (Smith et al. 2002; Rich & Origlia 2005; Cunha & Smith 2006; Cunha et al. 2007; Ryde et al. 2010) have helped to demonstrate the feasibility of determining precise chemical abundances in the H-band and have helped to lay the foundation for the APOGEE Survey.

Choice of the specific NIR wavelength range to be used for APOGEE involved optimizing a trade-off between competing desires:

• a.  
  Penetration of Interstellar Dust: the longer the infrared wavelength observed, the smaller is the sensitivity of the light to the extinguishing effects of interstellar dust, and the greater is the ability of the survey to penetrate highly obscured regions of the inner Galaxy.
• b.  
  Thermal Background: at longer wavelengths, the contribution of the thermal background increases and becomes significant in the K-band and beyond.
• c.  
  Airglow: the intensity of airglow emission (particularly from OH) varies across the NIR, with the weakest lines in the J-band and the strongest in the H-band.
• d.  
  Telluric Absorption: the ranges of the ground-based NIR bands are defined by major telluric absorption bands, most especially from CO₂ and H₂O; however, bands of various strengths from these molecules, as well as from CH₄, O₂ and O₃, are found all across the NIR.
• e.  
  Available Line Transitions: some key atomic elements, like Fe, C, N, and O (the latter expressed in absorption lines due to diatomic molecules such as CO, OH, and CN) are represented by spectral features all over the NIR, whereas other interesting elements, like K, F, Al, and Sc, have only a few lines.

Weighing the various aspects of this trade-space led to the selection of the H-band for APOGEE, with relatively strong weighting given to the first two considerations above: while the K-band is less sensitive to dust extinction than the H-band (${A}_{K}/{A}_{V}\sim 1/9$ compared to ${A}_{H}/{A}_{V}\sim 1/6;$ e.g., Cardelli et al. 1989), the H-band still confers a powerful degree of insensitivity to dust, whereas, in the meantime, S/N considerations motivate avoiding the large K-band backgrounds. Moreover, a K-band instrument requires much greater consideration for mitigating contamination from local sources of thermal background than does an instrument working in the H-band.⁵³

Unfortunately, although the above thermal background issues favor it, the H-band does include by far the strongest lines of the OH airglow spectrum. On the other hand, in principle, with high enough resolution, the impact of those airglow lines could be confined to a small fraction of the total spectrum, whereas in the K-band the thermal background would affect all pixels. In the ultimately selected APOGEE spectral range, the airglow spectrum includes about a dozen strong lines and a few dozen weaker lines (e.g., Figure 2); coincidentally, these lines span the entire APOGEE spectral region, which makes them potentially useful for wavelength calibration.

Zoom In Zoom Out Reset image size

The shape of the telluric absorption spectrum strongly drove the primary part of the H-band worth considering for APOGEE. The H-band itself was defined as the atmospheric transmission window between the strong and broad water absorption bands at ∼1.4 μm and ∼1.9 μm. By far, the lowest absorption in this region is in the range of approximately 1.5–1.75 μm, although this region is punctuated by the 30013 ← 00001 and 30012 ← 00001 bands⁵⁴ of the CO₂ molecule (Miller & Brown 2004), which cover roughly the $\lambda \lambda 1.568\mbox{--}1.586$ and $\lambda \lambda 1.598\mbox{--}1.617$ μm spectral intervals, respectively (Figure 2). An initial, two-detector design of APOGEE sought to avoid most of these bands, but eventually they were almost fully included in the near-contiguous wavelength coverage of the final, three-detector APOGEE instrument (Section 2.3).

In principle, different NIR windows offer some variance in available elements, but for many important elements (C, N, O—the most abundant metals in the universe—and the fiducial element Fe), there is ample representation in all three of the NIR bands (J, H, and K). Inspection of the Hinkle et al. (1995) infrared atlas reveals the J-band to have lines for almost the same set of elements as the H-band, but the H-band lines tend to be stronger in the spectra of giant stars than their J-band counterparts, as attested by the inspection of medium resolution NIR spectra from the IRTF library (see, e.g., Rayner et al. 2009, in particular their Figures 10 and 11). And while a number of α-elements are represented in either the H- or K-bands, other atoms with few transitions are represented in only one or the other (e.g., the H-band offers the important odd-Z elements Al and K). While these trade-offs—typically between elements tracking similar nucleosynthetic families—were not strong drivers in the decision process leading to the choice of the broadband NIR bandpass in which to operate (i.e., J versus H versus K), they did play a larger role in fine tuning the precise limits of the wavelength coverage (see below). Fortunately, the H-band, preferred for other reasons given above, was determined to offer an appealingly wide range of chemical elements that could be sampled, covering a range of nucleosynthetic pathways.

A detailed inspection of the infrared spectrum of Arcturus by Hinkle et al. (1995, Figure 2) was used to define the specific limits of the APOGEE spectral range. Initially, a survey of potentially accessible elements (atomic and in molecular combinations) in the H-band was made, and showed potentially useful representation from the following elements: C, N, O, Na, Mg, Al, Si, S, K, Ca, Ti, V, Cr, Mn, Fe, Co, and Ni (element by element maps are shown in Figure 34 in Appendix A). This is a useful subset of atomic species with which to probe most types of nucleosyntheses. Moreover, many of these elements are now accessible to integrated spectroscopy of extragalactic systems, which makes it possible to place the Milky Way in context with other galaxies having a range of masses and morphological types. Unfortunately, conspicuously absent from this initial assessment are any significant lines from neutron-capture elements, a general problem across the NIR.⁵⁵

The above panoply of H-band-accessible elements offers a number of potentially interesting insights into various aspects of GCE (see, e.g., Matteucci 2001 and the recent review of nucleosynthesis and chemical evolution by Nomoto et al. 2013):

• a.  
  Carbon, Nitrogen: important elements produced in significant amounts in intermediate-mass stars (Ventura et al. 2013), and thus sensitive to the ∼100 Myr timescales of star formation and chemical evolution. Carbon is synthesized in both massive stars ($M\geqslant 10{M}_{\odot }$) and lower-mass AGB stars ($M\sim 1-4{M}_{\odot }$), in roughly equal amounts (Nomoto et al. 2013). Because AGB stars produce no Fe, [C/Fe] can present an interesting behavior as a function of time in systems with ongoing star formation and chemical enrichment—initially increasing due to the contribution by core-collapse Type II supernovae (SN II) and AGB stars, then declining as a result of the onset of enrichment by Type Ia supernovae (SN Ia). Moreover, because oxygen is produced in large amounts by SN II, the C/O ratio tracks the relative contributions of low- to intermediate-mass stars versus massive stars in a given stellar population. Nitrogen is produced efficiently in intermediate-mass AGB stars (Karakas 2010), and there are suggestions in the literature (Chiappini 2013 and references therein) for an important contribution by massive stars as well. Analysis of integrated spectra of M31 globular clusters (Schiavon et al. 2013) and early-type galaxies (Schiavon 2007; Conroy et al. 2014) suggests that secondary enrichment was important in these systems. Although N can exhibit complicated behavior as a result of chemical evolution, it provides information on the relative importance of intermediate-mass stars to chemical evolution. Stellar evolution effects introduce an important caveat in the use of C and N abundances for chemical evolution studies, as dredge-up and/or deep mixing (e.g., Karakas & Lattanzio 2014) displace the abundances of these elements from their original main-sequence levels. This additional complexity can be dealt with in at least two ways. Following a data-based approach, one can take advantage of the sheer size of the APOGEE sample to focus on subsamples within the same evolutionary stage, thus minimizing, or even entirely cancelling out, stellar evolution effects (e.g., Schiavon et al. 2017). Conversely, a model-based method resorts to theoretical predictions of the dependence of internal mixing on stellar mass, metallicity, and evolutionary stage, generating estimates of relative ages of stellar populations (e.g., Masseron & Gilmore 2015). As a proof of concept, Martig et al. (2016) used the APOKASC sample as a training set to show that it is possible to disentangle, to first order, stellar evolution from chemical evolution effects on the abundances of carbon and nitrogen, making possible the determination of ages for over 50,000 APOGEE field giants (see also Ness et al. 2016a). The ages thus obtained correlate with distance from the Galactic midplane and [α/Fe] more or less in expected ways, though there are some surprises (see Figure 28 below).
• b.  
  Oxygen: the quintessential SN II yield from hydrostatic He burning in massive stars and the most abundant element in the universe, after hydrogen and helium. The timescale for the release of oxygen by SN II is much shorter than that of iron by SN Ia (e.g., Tinsley 1979). Therefore, one can argue that [O/H] is a more suitable and sensible chronometer and independent variable than [Fe/H] as a surrogate for "metallicity" in investigations of temporal abundance ratio variations benchmarked by overall enrichment level. That iron is more commonly used to indicate stellar metallicity is at least partly historically rooted in the relative ease with which [Fe/H] can be estimated from analysis of HR blue/optical spectra of solar-type stars. However, because the H-band includes many OH and CO lines that can be easily measured (and modeled) in the spectra of cool giants, APOGEE can provide reliable and precise [O/H] abundances for large stellar samples to lend better insights into crucial observables such as, e.g., the age–metallicity relation in different Galactic subcomponents. Moreover, stellar oxygen abundances can be more directly compared with gas-phase metallicities, which are predominantly based on measurements of oxygen lines (e.g., Kewley & Ellison 2008). The [O/Fe] ratio has been extensively used as an indicator of the relative contribution of SN II and SN Ia to chemical enrichment, which makes it sensitive to the timescale and/or efficiency for star formation as well as the shape of the high-mass end of the IMF (e.g., Tinsley 1979, 1980; Wheeler et al. 1989; McWilliam 1997).
• c.  
  Magnesium: another important α-element, Mg is an excellent complement to O. Its main isotope, ²⁴Mg, is produced in massive stars during carbon burning. Therefore, magnesium can also constrain enrichment by SN II, having become commonly used in part because it is relatively easier to measure than oxygen in optical spectra, with early abundances being based on MR spectra (Wallerstein 1962; Tomkin et al. 1985; Laird 1986). When combined with oxygen, magnesium can both probe the importance of Wolf-Rayet winds in chemical evolution and provide insights into the slope of the stellar IMF (e.g., Fulbright et al. 2007; Stasińska et al. 2012; Nomoto et al. 2013, and references therein). Magnesium is also important as the main element constraining the [α/Fe] ratio from integrated-light studies of extragalactic stellar systems (e.g., Worthey et al. 1992; Schiavon 2007). Thus, Mg measurements may provide a key bridge between Galactic and extragalactic chemical composition studies and facilitate the placement of the Milky Way within the broader context of galaxy evolution. In early-type galaxies (Worthey et al. 1992) and, to a lesser extent, in the bulges of spirals (Proctor & Sansom 2002), magnesium is found to be enhanced relative to iron, which is commonly interpreted as due to a short timescale for star formation in those systems.
• d.  
  Sodium, Aluminum: odd-Z elements. Sodium is produced during carbon burning and returned to the ISM via SN II. Aluminum, in turn, is expected to be produced mostly during neon burning, with only a small contribution from carbon burning. The SN II yields for these elements are moderately dependent on metallicity (Nomoto et al. 2013). Both Na and Al also participate in H burning in intermediate-mass stars (e.g., Karakas 2010), so these elements can also monitor the impact of intermediate-mass stars on chemical evolution. Interestingly, studies of chemical evolution in the Galactic thin and thick disk and halo reveal different trends for the abundances of these elements as a function of [Fe/H] (e.g., Bensby et al. 2014).
• e.  
  Silicon, Sulfur: these α-elements are produced mostly in SN II (with small amounts in SN Ia). Silicon, as ²⁸Si, is the most abundant product of oxygen burning, with the dominant sulfur isotope, ³²S, also synthesized in oxygen burning (e.g., François et al. 2004; Nomoto et al. 2013). The abundances of these elements, in principle, provide constraints on the stellar IMF by comparison to the abundances of lighter α-elements O and Mg (e.g., McWilliam 1997).
• f.  
  Potassium: another odd-Z element whose chemical evolution is poorly understood. Shimansky et al. (2003) suggest that the evolution of K comes from hydrostatic oxygen burning and we expect an increase in [K/Fe] with [Fe/H].
• g.  
  Calcium, Titanium: two more elements with strong ties to SN II yields, but which may also have some fraction produced in SN Ia (e.g., François et al. 2004; Nomoto et al. 2013). In Galactic populations, these elements display trends similar to those of O, Mg, Si, and S, but there has been debate in the literature as to whether they behave like SN Ia products in early-type galaxies (e.g., Milone et al. 2000; Saglia et al. 2002; Cenarro et al. 2004; Schiavon 2010; Conroy et al. 2014).
• h.  
  Vanadium: produced in both explosive oxygen-burning and silicon burning, ⁵¹V is synthesized through radioactive parents, ⁵¹Cr and ⁵¹Mn, and is made in both SN II and SN Ia (Nomoto et al. 2013). Reddy et al. (2006) find [V/Fe] to be approximately solar in the thin disk and slightly enhanced in the thick disk (by about 0.1 dex).
• i.  
  Manganese: while most iron-peak elements follow iron, Mn does not, with [Mn/Fe] decreasing with decreasing [Fe/H]. Manganese is produced mainly from radioactive decay of ⁵⁵Co in both core-collapse and Type Ia supernovae (Nomoto et al. 2013); the dominant source of Mn has not been definitively identified.
• j.  
  Chromium, Iron, Cobalt, Nickel: these elements represent the Fe-peak in APOGEE spectra and are produced in varying amounts in both SN Ia and SN II.

The mere presence of a line transition, of course, is not sufficient for it to provide scientifically useful abundance measurements. As a means to assess the identified lines, extensive tests were made of model RGB spectra of different metallicities ([Fe/H] = −2, −1, 0) at a number of potential spectrograph resolutions to determine their suitability for 0.1 dex precision measurements (see Section 2.3). Given the results of these tests, and to inform the final selection of the specific spectral coverage, these elements were ranked in a prioritization scheme that considered not only the nucleosynthetic family to which the element belonged and their value to mapping GCE, but the strength and number of the available transitions:

• a.  
  Top priority (i.e., "must-have" elements): C, N, O, Mg, Al, Si, Ca, Fe, Ni.
• b.  
  Medium priority (i.e., valuable elements worth trying to include in APOGEE, but that should not necessarily drive requirements for the survey): Na, S, Ti, Mn, K.
• c.  
  Lower priority (i.e., "if at all possible" elements—interesting elements but not deemed essential for success): V, Cr, Co.

A census of the H-band shows that the reddest third (approximately 1.7–1.8 μm) is the most deficient in interesting spectral lines whereas the middle third (approximately 1.6–1.7 μm) has the highest density. Moreover, the 1.7–1.8 μm subwindow has significantly worse telluric absorption (Figure 34). This ultimately drove the primary APOGEE wavelength of interest to roughly 1.5–1.7 μm. The precise wavelength limits were set by the specific line transitions desired, after detailed assessment of resolution and S/N considerations.

The ultimately adopted wavelength setting includes sufficient lines for abundance work on all of the top and medium priority elements listed above. However, a subsequent assessment of the available lines for the low priority elements suggested that abundances for Cr and Co would be very difficult to obtain reliably, given the excitation potential, $\mathrm{log}{gf}$, and strength in the Arcturus spectrum of these lines. Therefore, abundances of Co and Cr were not attempted in the first round of elemental abundance determinations leading up to DR12. The additional element Cu, on the other hand, was not considered as a viable APOGEE product when the survey was initially conceived, but later Cu was successfully explored in FTS spectra of standard stars in the APOGEE region by Smith et al. (2013). Over time, as a better understanding of available line transitions in the APOGEE spectral range is achieved and as the performance of the APOGEE Stellar Parameter and Chemical Abundances Pipeline (ASPCAP) continually improves, the sensitivity of APOGEE to all three of these elements, as well as others such as P and Ge, can be reevaluated and reliable abundances will possibly be made available in future data releases.

As with most spectrographs, the precise specifications of the APOGEE spectrograph were the product of balancing the competing benefits of HR, high S/N, and a broad wavelength range. To model these factors, we calculated a series of synthetic H-band spectra for RGB stars (${T}_{\mathrm{eff}}=4000$ K, $\mathrm{log}g=1$) with [Fe/H] = −2, −1, 0, at a number of values for resolving power between R = 15,000 and 30,000. For each case, we computed two spectra, one with solar-scaled composition, and a second in which the abundance of a particular element, X, was modified by Δ[X/Fe] = 0.1. These calculations were used to derive an estimate of the ${\rm{S}}/{\rm{N}}/\mathrm{pixel}$ required to measure abundance variations of the order of 0.1 dex at each resolution, as described in Appendix B. The results are summarized in Figure 3.⁵⁶

Zoom In Zoom Out Reset image size

These calculations give rise to a number of general considerations:

• a.  
  Clearly the highest S/N is required at the lowest metallicities and resolutions, with metallicity being the stronger driver. For instance, measuring the Mg abundance to 0.1 dex at [Fe/H] = −2.0 would require ${\rm{S}}/{\rm{N}}\gt 50$ at R = 15,000 and ${\rm{S}}/{\rm{N}}\gt 25$ at R = 30,000. At the other extreme, measuring K to 0.1 dex requires ${\rm{S}}/{\rm{N}}\gt 700$ at R = 15,000 and ${\rm{S}}/{\rm{N}}\gt 400$ at R = 30,000 for the same metal-poor star (outside the range shown for this element in Figure 3).
• b.  
  The Galactic thin disk is dominated by stars with $[\mathrm{Fe}/{\rm{H}}]\gt -1$, for which the number of elemental abundances that can be determined with 0.1 dex precision is maximum for a given S/N. For example, at R = 21,000 and ${\rm{S}}/{\rm{N}}=100$, we are able to measure all of the listed elements except Na, S, and V for thin disk stars.
• c.  
  For more metal-poor stars, the challenging elements (at the top of Figure 3 and Tables 3 and 4) are measurable with less demanding precision. It might also be possible to do at least a statistical analysis of abundance patterns in metal-poor stars with the minimum nominal S/N by combining spectra for multiple stars of similar chemistry or position in phase space.
• d.  
  Obviously, for a constant exposure time, we can achieve higher S/N by probing stars of brighter magnitudes and thereby recover more of the challenging lines.

Even more specifically, this analysis led to the following considerations:

• a.  
  Na is challenging for all but the most metal-rich stars (even ignoring that the available Na lines are affected by non-negligible blending by molecular lines), but we have Al as a substitute. Therefore, Na was not used as a requirement driver.
• b.  
  V is similar in chemical character to Al, and behaves similarly to the α-elements Ca and Ti (Reddy et al. 2006). Therefore, loss of this element for some stars was not considered a substantial setback.
• c.  
  S is perhaps the most valuable element with weak lines in the potential APOGEE line list. The S i lines at 15422 Å and 15469 Å are the two cleanest lines, whereas the strongest line at 15478 Å is blended with a strong Fe i feature. In some ways, Si can play the same role in terms of constraining the high-mass end of the IMF, though the combination of S and Si is better. While it was expected that S could be measured for bright stars, it was accepted that S should not be a requirement driver at the nominal survey magnitude limit.
• d.  
  Given the above logic that we would not use Na, V, or S to drive the survey specifications, it seemed reasonable to adopt the measurement of the stellar K abundance for $[\mathrm{Fe}/{\rm{H}}]\gt -1$ stars as a requirements driver.
• e.  
  For metal-poor stars ([Fe/H] ≲ −1), it was considered desirable to have, at minimum, O, C, Fe, Mg, Si, Al, Ca, and Ni, making the measurement of Ni in all stars a requirements driver.
• f.  
  Overall improved resolution lowers the S/N requirements, but the gains from R = 15,000 to R = 21,000 are modest, according to the calculations. However, the above estimates were assumed to be somewhat optimistic, given that telluric lines, sky emission, and blends of stellar lines were not considered. Telluric and sky lines will be better removed at higher resolution. All elements studied have at least some lines that are free of telluric or sky interference for most stellar RVs, and fairly isolated at solar metallicity and intermediate temperatures (${T}_{\mathrm{eff}}\simeq 4000$ K). However, at cooler temperatures and similar metallicities, molecular lines due to CN, CO, and/or OH affect virtually all wavelengths in the H-band.

Taking into consideration these calculations and the wavelengths of the transitions of the target elements (all those listed above, except Na, V, and S), we obtained the following constraints on wavelength coverage: The blue limit of the APOGEE range was set to capture the single available K i line at 15160 Å as well as the best Mn i lines at 15157–15263 Å, for reasons discussed above. Meanwhile, the red limit was set by the goal to make sure to include at least one of the three Al i lines at 16720–16770 Å.⁵⁷ The specified wavelength range also needed to account for potential heliocentric velocity variations in Galactic stars, and a contingency of ±700 km s⁻¹ was adopted.

Initially, it was thought that the goals for the APOGEE science might be met with a baseline, single grating instrument sampling two disjoint H-band windows, but a desire to sample multiple lines for each element for redundancy, as well as the greater than linear gains of increased spectral resolution, drove to a three-detector design with nearly continuous coverage from the K i to Al i lines. Nevertheless, even with three detectors, the desired minimal spectral resolution leaves the short wavelength end slightly undersampled. To address this problem, it was decided that the three detector spectrograph would include a mechanism by which the focal plane arrays can be dithered precisely by half-pixel steps. By taking exposures in dithered pairs, the spectral resolution can be recovered as properly (Nyquist) sampled through interpolation of the paired exposures during post-processing.

To summarize this critical information in a single, readily accessible location, we list in Table 2 the final choices of resolution, S/N, and wavelength limits mandated by the considerations above, along with other hardware specifications discussed in Section 3.

Table 2.  Summary of APOGEE Instrument Characteristics

Property	Performance
On-sky field of view (typical declinations)	30 diameter circle
On-sky field of view (high airmass)	15 diameter circle
Total number of spectrograph fibers	300
Fiber center-to-center collision limit on plugplate	70 arcsec
Fiber scale on sky (diameter)	2.0 arcsec
Detectors	2.5 μm cutoff, 20482 pixel, Teledyne H2RG Imaging Sensors
Detector pixel size	18 μm
Detector wavelength regions	1.514–1.581, 1.585–1.644, 1.647–1.696 μm
Littrow ghost position	1.6056–1.6067 μm
Littrow ghost intensity	0.150% of full fiber intensity
Dispersion (at 1.54, 1.61, 1.66 μm)	0.326, 0.282, 0.235 Å pixel−1
Point Spread Function (spatial FWHM) (at 1.54, 1.61, 1.66 μm)	2.16, 2.14, 2.24 pixels
Line Spread Function (resolution element) FWHM (1.54, 1.61 1.66 μm)	2.01 , 2.44 , 3.14 pixels
Median native (λ/FWHM) resolution (at 1.54, 1.61, 1.66 μm)	23,500, 23,400, 22,600
Predicteda throughput (1.54, 1.61 1.66 μm)	13, 15, 10%
Measuredb throughput (1.61 μm)	20 ± 2%
S/N for H = 12.2 K0III star in an 8 × 500 s visit (1.61 μm)	105
Specific fiber numbers most affected by excessive persistence	1–100

Notes.

^aCalculated as the product of the wavelength-dependent transmittance or reflectivity for all components of the as-built telescope+instrument design. ^bBased on measured flux for stars of known H magnitude. Error bars reflect uncertainties regarding extinction by Earth's atmosphere and (seeing-induced) fiber losses.

Download table as:  ASCIITypeset image

A final issue that had no bearing on the instrument design but did bear on the allocation of survey resources is that of unidentified lines. At the start of the survey, approximately 6% of all lines deeper than 5% of the continuum within the APOGEE wavelength interval were still not identified with a transition from a given excitation and ionization state of a known chemical element. This number went up to 20% when all lines deeper than 1% of the continuum were considered. To improve this situation, the APOGEE team initiated a collaboration with a team of laboratory astrophysicists. For details, we refer the reader to Appendix E.

In the initial formulation of the APOGEE experiment, a stellar radial velocity precision of 1 km s⁻¹ was established as a requirement to be met by the combination of instrument capabilities and data reduction and analysis software. For many problems in large-scale Galactic dynamics—e.g., measuring the disk rotation curve or the velocity dispersions of stellar populations, sorting stars into populations, looking for kinematical substructures—a velocity precision at the level of 1 km s⁻¹ per star is not only suitable, but substantially better than has been available in these kinds of investigations heretofore. However, the combination of HR and a very stable instrument platform made possible achieving kinematical precision beyond these initial survey specifications. In fact, the APOGEE instrument and the existing radial velocity software routinely deliver radial velocities at a precision of ∼0.07 km s⁻¹ for S/N > 20, while the survey provides external calibration sufficient to ensure accuracies at the level of ∼0.35 km s⁻¹ (Nidever et al. 2015; Section 7.3)—a level of performance that allows more subtle dynamical effects to be measured. For example, the detection of pattern speeds of—or kinematical substructure in the disk due to perturbations and resonances from—spiral arms, the bar, or other (e.g., dark matter) substructure (e.g., Dehnen 1998; Famaey et al. 2005; Junqueira et al. 2015), the detection of stellar binary companions (e.g., Terrien et al. 2014), the assessment of stellar membership in star clusters (e.g., Terrien et al. 2014; Carlberg et al. 2015) or extended stellar kinematic groups (i.e., "moving groups" or "superclusters") in the disk (e.g., Eggen 1958, 1998; Montes et al. 2001; Malo et al. 2013), and the accurate measurement of stellar velocity dispersions in star clusters or satellite galaxies (Majewski et al. 2013) are all made possible with radial velocity measurements of the rms precision and external accuracies routinely achieved by APOGEE for main survey program stars. Nevertheless, it has been shown that even greater precision and accuracy may be obtained from APOGEE spectra, which increases the sensitivity to even lower mass stellar companions (Deshpande et al. 2013) and greatly benefits the exploration of the intricate dynamics of young star clusters (Cottaar et al. 2014; Foster et al. 2015).

Until recently, the largest detailed chemical abundance studies were typically focused on stars in the solar neighborhood, and included samples of order 10³ stars (Bensby et al. 2003; Venn et al. 2004). A primary goal of APOGEE is to obtain similar-sized samples of several thousand stars in many dozens of Galactic zones across the Galaxy, and this led to the basic technical requirement to obtain data on 100,000 stars distributed across all major Galactic populations. For example, a typical prediction from GCE models that we aim to test is gradients in mean abundance for critical elements (Fe, C, N, O, Al) in disk populations, with differences in the models seen at the level of a few 0.01 dex at each radial or vertical point in the Milky Way. Discriminating the present models demands an accuracy in mean abundances of ∼0.01 dex per Galactic zone, or more than 100 stars with 0.1 dex accurate abundances in that zone assuming $\sqrt{N}$ statistics. Similar precisions are needed to determine, within each zone, the variation of [X/Fe] with [Fe/H] or [O/H] (which are important discriminants of the IMF and SFH), and therefore require 100 stars with 0.1 dex accurate abundances in each metallicity bin. Thus, deriving not only mean abundances but accurate and useful multidimensional abundance distribution functions (such as [α/Fe] and [Fe/H]) in each zone requires orders of magnitude more stars per zone. Such accounting (e.g., [dozens of Galactic zones][∼20 metallicity bins][100 stars/bin]) leads to samples of $\sim {10}^{5}$ stars. Fortunately, such numbers were estimated to be achievable if a three-year observing campaign were feasible within the duration of SDSS-III (which had a well-defined end of mountain operations in the summer of 2014; Section 2.7).

While a $\sim {10}^{5}$ sample of stars with R ≈ 22,500 spectra is orders of magnitude larger than had been previously available for Galactic archaeology, implicit to making this a true milestone is that the stars be distributed systematically and widely across the Galaxy to include (a) fields that cover a substantial part of the Galactic bulge including the Galactic Center, (b) fields that span a substantial fraction of the Galactic disk from the Galactic Center to and beyond the longitude of the Galactic Anticenter, (c) high-latitude fields to sample the halo, and (d) fields that probe a variety of specific targets of interest, such as star clusters (valuable as both science targets and calibration standards) and known Galactic substructures (e.g., the bar, disk warp/flare, tidal streams). In addition, a small fraction of the survey time/fibers would be available for potential ancillary science projects (Section 4.3), though these would drive neither the science requirements nor instrument design, nor significantly impact the net throughput of the main survey.

For APOGEE's primary target—evolved stars—the survey seeks to reach across the Galactic disk in moderate extinction, to the Galactic Center in fairly heavy extinction, and to the outer halo in low extinction. With some variation due to metallicity, the tip of the red giant branch (TRGB) lies at ${M}_{H}\sim -5.5$. AGB stars extend still brighter, whereas red clump stars have ${M}_{H}\sim -1.5$. To achieve the goal of readily and abundantly sampling all Galactic populations in situ, it was required that for "typical" survey fields and exposure times APOGEE routinely reach a depth of H = 12.2, which translates to probing the TRGB to 35 kpc for no extinction and to $\gt 8.5\,\mathrm{kpc}$ (i.e., to at least the distance of the Galactic center) through ∼3 magnitudes of H-band extinction (${A}_{V}\sim 18$). Thus, H = 12.2 was adopted as the "baseline" magnitude limit for "normal" APOGEE fields.

Special consideration was required for bulge fields, for which the considerable zenith distance even at culmination translates to short observing windows and more extreme differential refraction at APO. To enable greater bulge spatial coverage, a baseline magnitude limit of H = 11.1 was implemented for these fields to reduce the integration time by a factor of three. Nevertheless, Galactic center distances are reachable for TRGB stars when ${A}_{H}\lesssim 2$.

However, in other fields, longer integrations were anticipated to enable APOGEE to probe red clump stars in low extinction fields to >8.5 kpc or TRGB stars to >50 kpc, or TRGB stars to the Galactic Center in fields with ${A}_{H}\sim 4$ (${A}_{V}\sim 25$). Such longer fields were not only desired for APOGEE, but they were a necessary part of the observing plan because, initially, APOGEE shared bright time observations with the Multi-Object APO Radial Velocity Exoplanet Large-area Survey (MARVELS) project (Ge et al. 2008; Zhao et al. 2009). The baseline MARVELS program observed fields for $\geqslant 24$ epochs at about 1 hr per visit; thus, at least some APOGEE fibers on these same cartridges could sample fainter stars by accumulating integrations of up to 24 hr. At first, MARVELS "controlled" half of the bright time,⁵⁸ so that about half of the APOGEE time was in these "long fields." Subsequent termination of the MARVELS observing campaign in the second year of APOGEE observations enabled some reconfiguration of our observing plan (Section 4.1.2).

For throughput and target selection requirements, the APOGEE team assumed that it would be able to observe during 95% of the available bright time (i.e., after accounting for a $\sim 50 \%$ loss for weather plus one SDSS-III-wide engineering night per month) for the final three years of SDSS-III. This high fraction would be achieved by carrying out simultaneous MARVELS/APOGEE observing with the two instruments sharing the focal plane. The ability to carry out such simultaneous observations was thus a technical requirement for achieving the desired survey size and depth.

It was determined from the onset that APOGEE would feature a 300 fiber spectrometer, which is the number of spectra that can be maximally packed along the spatial direction on a 2048 pixel-wide detector (allowing ∼7 pixels per spectrum, assumed to be sufficient to span both the point-spread function (PSF) of each spectrum and leave dark gaps between). Initially, it was assumed that 50 fibers would be needed for simultaneous observations of sky and telluric calibration stars.⁵⁹

As discussed above, the requirement of detailed and precise chemical composition determinations drives requirements of S/N ≈ 100 per pixel at a resolution R ≈ 22,500. The requirement to observe $\sim {10}^{5}$ stars then implied that, after adopting conservative estimates for all variables, the instrument would have to achieve this S/N at the typical observation depth H = 12.2 in 3 hr of total integration time: ${N}_{\mathrm{stars}}\,\approx$

• a.  
  (3 year observing campaign) ×
• b.  
  (11 months per year⁶⁰ ) ×
• c.  
  (30 nights per month) ×
• d.  
  (11 hr per night) ×
• e.  
  (40% bright time) ×
• f.  
  (95% of bright time to APOGEE) ×
• g.  
  (50% clear weather) ×
• h.  
  (250 targets per field) /
• i.  
  (3 + 1.5 hr per field⁶¹ ) $=\,1.15\times {10}^{5}.$

More detailed analyses that, for example, included lost nights for engineering time, various weather models, and more sophisticated observing plans all yielded estimates within 15%–20% of this conservative estimate.

Because the majority of APOGEE targets are RGB stars, a substantial fraction are expected to be single-lined binaries. The amplitudes of radial velocity variations in binary stars can reach as much as 10–20 km s⁻¹; thus, it is useful for such binary systems to be identified and flagged so that they can, when necessary, be removed from APOGEE kinematical samples to minimize the deterioration of the precision of derived bulk dynamical quantities for stellar populations, such as, e.g., velocity dispersion measurements.

Identification of the radial velocity variability associated with single line binaries can be achieved by splitting the total integrations for each star into visits optimized in cadence to identify the binaries with problematic barycentric velocities. Given the expected instrument throughput, it was understood early on that to reach the distances of interest for studying a large fraction of the Galaxy (and in particular, crossing the full extent of even just the near side of the disk for the nominal Galactic plane pointing), detector integrations of multiple-hour net length would be needed. However, because differential refraction limits the range of hour angles over which any given drilled fiber plugplate⁶² can be employed without incurring significant light loss, it is necessary to break long exposures into multiple observing stints—either using plugplates drilled for different hour angles (potentially observed on the same night) or using the same plate observed on multiple nights. It was most efficient and natural to adopt the latter solution and exploit the multivisit strategy for the added purpose of binary star identification.

For effective identification of binaries, more velocity samples over a longer time baseline is always preferable. This desire must be balanced against that of survey efficiency, which pushes in the direction of breaking long exposures into the fewest possible number of visits, to limit the fraction of time surrendered to overheads of plugplate cartridge (Section 3.1) changing and field acquisition. While mountain observing staff showed that this overhead can be as low as 12–15 minutes per plugplate cartridge change, 15–20 minutes is a more realistic "typical" situation. Under these circumstances, field visits of less than 30–45 minutes accrue substantial overhead. Moreover, frequent visits of such short duration place substantial physical demands on the observers. In any case, there were only eight available "bright time" Sloan plugplate cartridges available on which to put APOGEE fibers, so that no more than eight APOGEE plugplates were observable on a given night. Thus, given the tradeoffs between observing efficiency and differential refraction limits as well as the eight-cartridge limit, it was decided that the baseline APOGEE visit would include about an hour of integration plus overhead (see Section 5.1) and that the "nominal" survey field integration of ∼3 hr length (see Section 2.7) would be divided into no less than three visits.

With this basic multivisit plan in place, one last requirement imposed is the adopted cadence for the visits. To understand the potential effects of binary stars on measured APOGEE dynamical quantities, and to assess the best way to distribute three visits over time to maximize the ability to detect "problem" binaries, a series of Monte Carlo simulations of stellar populations having nominal binary fractions and mass, period, and orbital eccentricity properties was undertaken. The details of these models, wherein the parent sample of typical APOGEE targets had their radial velocities sampled over varying time intervals and net spans, are given in Appendix C.

These simulations showed that the majority of binary systems (∼74 %) are not expected to adversely affect the kinematical measurements, where "adversely affected" had been defined as a measured velocity of the primary star that is $\gt 2$ km s⁻¹ different from the true systemic motion of the binary system. Given the above visit strategy, the most effective way of identifying the remaining 26% of binaries is by calculating the radial velocity difference between every combination of paired measurements and flagging stars showing a maximum velocity difference above a certain threshold (we adopted for our modeling 4 km s⁻¹). These simulations indicated that, for a set of at least three radial velocity measurements of 0.5 km s⁻¹ precision, a temporal baseline spanning at least one month was sufficient to make evident at least a third of the remaining (26%) binaries most likely to have a significant impact on the APOGEE survey velocity distributions. While longer baselines improve detectability, that improvement is only marginally better for baselines lengthened to a full season of typical object visibility (Figure 35). On the other hand, a strategic decision was made to try to finish observations of fields within an observing season whenever practical, not only so that analysis of survey results could begin soon after the survey began, but to lower the risk of leaving many fields with incomplete observations at the end of survey, should weather or other considerations confer a net deficit of observing hours to complete the field plan. Thus, a requirement of at least a 25 day span for the visits to a single plugplate was adopted as a rule, with a minimum span between epochs of 3 days.

In their CORAVEL study, Famaey et al. (2005) find 13.7% of their local K giant sample to be in binaries and that with their strategy (two radial velocity measurements per star spanning two to three years) and 0.3 km s⁻¹ velocity accuracy, "binaries are detected with an efficiency better than 50% (Udry et al. 1997)." Famaey et al. actually find an even lower binary fraction of only 5.7% for their M giant sample. These numbers suggest that one might expect 27.4% and 11.4% binary fractions among K and M giants, respectively. If two-thirds of 26% of these (i.e., 9%) slip through the APOGEE ability to detect them, then perhaps only a few percent of APOGEE targets would remain kinematically "problematic," with measured velocities deviant from their systemic motion by more than 2 km s⁻¹. Even this fraction is likely an upper limit because (a) one month is the minimum temporal baseline, whereas, at survey end, the median baseline for all multivisit fields is almost two months (see Figure 18(b)); (b) a significant fraction of the primary APOGEE sample, ∼32,600 stars or 30%, had more than three visits, by design or circumstance (see Figure 18(a)); and (c) the per-visit velocity precision is substantially better than 0.5 km s⁻¹ (at 0.07 km s⁻¹; see Section 10.3 of Nidever et al. 2015). An initial assessment of the lower limit of the actual detected binary fraction within APOGEE is of order 4% (7% if substellar mass companions are included), based on the results of Troup et al. (2016).

The APOGEE instrument leverages the wide-field (3° diameter FOV) capability of the Sloan 2.5 m telescope (Gunn et al. 2006) at Apache Point Observatory, New Mexico (USA), and the highly efficient and proven survey infrastructure that has led to the very successful SDSS-I and SDSS-II suites of experiments using optical spectrographs (Smee et al. 2013). For the optical spectrographs, which are mounted to the telescope back end, short-length (1.8 m) fiber optic bundles run directly from the telescope focal plane to the pseudo-slits of the spectrographs. In contrast, because of the sheer size of the APOGEE spectrograph, it is housed in a separate building adjacent to the 2.5 m Sloan Telescope and fed light via a single, approximately 40 m "long fiber" run from the telescope. This set of 300 "long fibers" (called the "fiber link") is permanently attached to the APOGEE instrument with one end of each fiber hermetically sealed inside the cold, evacuated cryostat and terminating at the spectrograph "pseudo-slit". The warm end of the fiber link terminates at the base of the telescope.

At the telescope, APOGEE employs the same plugplate system designed for use in the SDSS-I and SDSS-II surveys (Owen et al. 1998; Siegmund et al. 1998), and, indeed, makes use of eight plugplate cartridges from those previous surveys that were converted to "bright time" operations through the incorporation of distributed and mingled "anchor blocks" of fibers linked to the MARVELS and APOGEE instruments. As with other Sloan spectrographic surveys, aluminum plugplates with precision-drilled holes matching the positions of APOGEE targets in a specific sky field are interchanged and manually plugged with the cartridge "short fibers" each day in preparation for the ensuing night-time observing, The APOGEE fibers are step index, multi-mode, low-OH (i.e., "dry") fibers with a 120 μm diameter silica core that subtend 2 arcsec of sky at the telescope focus. The sets of "short fibers" installed in the fiber plugplate cartridges terminate in pluggable, stainless steel ferrules that impose a fiber-to-fiber proximity limit (the so-called "fiber collision limit") of 71 arcsec. Each APOGEE anchor block of six fibers covers a linear extent across the plugplate cartridge equivalent to a roughly circular sky patrol area of about a 10 (22 cm) radius. However, the distribution of these anchor blocks is non-uniform, forming a ring around the central part of the field. This arrangement allows either a uniform plugging across the entire plate or a higher central concentration with all 300 fibers in a relatively narrow FOV. The latter application is for those plugplates that are drilled for high airmass (low declination) fields, where all targets are selected within a restricted FOV (potentially as small as a 1° diameter circle; see Section 5.2).

One primary difference in the plugging process for APOGEE plates compared to previous SDSS projects is that APOGEE fiber plugging imposes one level of fiber management by separating fibers into three H-magnitude-defined groups that are plugged into holes corresponding to the faintest, mid-brightness, and brightest thirds of targets on each plate. This fiber management is accomplished by color-coding the target holes on each plate by their brightness rank as red, green, or blue, and filling these holes with fibers having matching colored sheathing. No other requirement is imposed on which science fiber is plugged into which hole, which means that a particular star may be observed by completely different short and long fibers on different visits. Each anchor block has two fibers of each color, so that the "bright," "medium," and "faint" fibers are evenly distributed across the field. At the spectrograph end, these different fibers are sorted along the pseudo-slit into a repeating pattern of faint−medium−bright−bright−medium−faint to ensure that bright spectra are never placed next to faint spectra in the spectrograph focal plane (this pattern of alternating spectrum brightness may be seen in Figure 14 below). This scheme minimizes the degree of contamination of any given spectrum by overlapping wings of the PSF from the adjacent spectrum of a brighter star.

During survey operations, the short-length fibers in each of the eight plugplate cartridges are mated to the long-length fibers approximately hourly (after each cartridge/plugplate change) via a custom-built "gang connector" that simultaneously mates each of the 300 short fibers with its corresponding long fiber to within a few μm accuracy. Because of the frequency of this mating operation, the need for efficiency of operations, and the sometimes dusty conditions at the site, no index-matching gel is used in this fiber coupling operation; as a result, there are some light losses at the connector, but they are small enough (5%) to justify the simplification of operation.

An additional modification is required for the APOGEE fiber mapping system. In the case of the optical Sloan spectrographs, fiber mapping is undertaken after each plugplate is plugged by running a laser directly up the pseudo-slit (which is integrated as part of the cartridge) and recording which plugged fibers light up on the plugged plate in succession. In the case of APOGEE, however, the true instrument pseudo-slit is inaccessible, as it lies within the cryostat. Therefore, a warm copy of that pseudo-slit is mated to the gang connector for this operation.

3.2.1. Technological Innovations

3.2.2. Basic Instrument Layout

The basic optical design of the spectrograph leverages the successful optical design of the multifiber optical SDSS spectrographs (Smee et al. 2013), but modified as needed for high spectral resolution at infrared wavelengths. The basic optical layout of the APOGEE instrument is illustrated in Figure 4 and was built in a highly modularized fashion with distinct subcomponents:

Cryostat: past the gang connector, the long fibers route to the spectrograph and enter the cryostat via vacuum feed-throughs at the cryostat vacuum wall (keeping the fibers intact avoids the introduction of another optical junction and thus minimizes throughput losses and focal ratio degradation). The 1.4 × 2.3 × 1.3 m cryostat is a specially designed, stainless steel, liquid nitrogen-cooled vessel⁶³ that encloses the optical bench and instrument subcomponents, themselves surrounded by an aluminum cold radiation shield and thermal blankets. The entire ∼1.8 metric ton (2 U.S. tons) system sits on four pneumatic isolation stands to minimize vibration.

Optical Bench: the spectrograph optics are mounted to a three-inch thick, cold aluminum optical bench suspended from the vacuum shell. A 97 liter LN₂ tank suspended from the bottom of the cold plate maintains a temperature of about 78 K in the vicinity of the camera and gives the instrument a 5.5 day hold time, although an automatic filling system tops off the dewar every morning after observing is over. Although overall the bench-mounted, fiber-fed spectrograph confers a distinct advantage in maintaining a vibration-free, temperature-stable system with a constant gravity vector that ultimately makes the APOGEE instrument deliver spectra that are extremely stable and repeatable, the small nightly variation in LN₂ liquid level was later found to induce slight variations in the mechanical flexure on the cold plate that can be observed as small, ∼0.1 pixel shifts in the spatial position (and even smaller, ∼0.01 pixel shifts in the spectral position) of the spectra on the detector over the course of the night; fortunately, these slowly varying changes can be mapped and accounted for by regularly observed flat-field exposures.

Pseudo-slit and Collimator: the final 2 m of the long-length fiber link trains are contained within the cold volume and terminate on a curved pseudo-slit. A 350 μm fiber-to-fiber spacing at the pseudo-slit maps to 6.6 pixel spacings between spectra on the detectors. An "uncorrected Schmidt" camera, used in reverse, collimates the light of each of the fibers. In keeping with the Schmidt design, the fiber tips are carefully positioned to lie on, or close to, a curved surface with radius of curvature approximately one-half that of the collimator and to emit light in a direction orthogonal to that surface, so that they axially point back as close as possible to the center of curvature of the pseudo-slit. Moreover, the pseudo-slit and spherical collimator mirror have a common center of curvature near the system pupil, which is at the approximate position of the spectrograph grating. The fiber ends are also aligned on a curved lateral surface to ensure that at the detector the fiber ensemble gives straight slit images; this lateral curve enables each fiber to deliver the same rest wavelength range on the detectors, but creates a different location for the "Littrow Ghost" on each fiber (Figure 5; Section 3.4). As the only active means to effect small changes in instrument focus, the collimator is controlled by three-axis (tip−tilt−piston) movement. This capability is also useful for implementing dithering in the spatial dimension, an operational mode periodically activated for the creation of spatially-smoothed instrument flat fields. The resulting optical design is on-axis so that the pseudo-slit is an obscuration in the collimated beam.

Zoom In Zoom Out Reset image size

Cold Shutter and Flat Field Illumination: a "swinging gate" cold shutter with a light trap (not shown in Figure 4) covers the pseudo-slit to prevent excessive light leaking into the cold volume when the spectrograph is not taking observations. This minimizes the accumulation of unwanted charge that could contribute to detector "persistence" (see Section 3.4). This mechanism also contains a set of infrared light-emitting diodes that can provide a diffuse illumination onto the detectors useful for creating flat-field exposures.

Fold Mirrors: two fold mirrors are used for efficient packaging of the optical train within the cryostat. The second mirror flat is also a dichroic, which passes light longward of the APOGEE spectral range into a trap behind the mirror; this assists in the mitigation of stray, thermal light (see below).

VPH Grating: the disperser is a three-segment mosaic VPH grating, the first ever fabricated by KOSI. Because the required grating size exceeds that which can be recorded in a single VPH exposure, the APOGEE VPH is made by recording, in close temporal succession, three adjacent segments in gelatin on a common fused-silica substrate. While prototype mosaic VPH's have been fabricated in the past, none have been deployed in an astronomical instrument. The VPH has 1009.3 grooves/mm, operates in first order with a 54° angle of incidence, and delivers a peak efficiency of 90% near the center of the APOGEE spectral range (40% at the edges).

Camera: the wavelength-dispersed beams are focused by a six-element refractive camera designed and fabricated by NEOS. The APOGEE camera is very large for an astronomical spectrograph, weighing 250 lbs and with lenses ranging from 237 to 394 mm in diameter. The optical materials that can be practically considered for such large elements are severely limited. Fortunately, the narrow wavelength range of APOGEE means that only two materials—monocrystalline silicon (for four of the lenses) and fused silica (for the other two)—are necessary (and those two materials are also, coincidentally, very robust to thermal shock). When combined with (minimal) absorption through the fused silica and the performance of the anti-reflection coatings, the throughput for all six lenses is 93% across the wavelength coverage.

Detector Array: the spectra are recorded on three Teledyne Imaging Sensors H2RG, 2.5 μm cutoff, 2048 × 2048 pixel detector arrays. These are mounted in a detector mosaic opto-mechanical package similar to that used for the NIRCam instrument for the JWST, with the arrays tilted to approximate the field curvature of the optical system within a tolerance of 15 μm through precise shimming of piston, tip, and tilt. These detectors are operated in sample-up-the-ramp mode, with readouts every 10.7 s; thus, "images" of the spectra are in data cubes for each of the three arrays. As mentioned above, the arrays lie on a movable stage which is used for "dithering" translation of the entire assembly in the dispersion direction; in practice, images are taken in pairs with half-pixel shifts, which, in the data processing, can be used to recover the full sampling of the spectral line-spread profile (Section 6.3).

Baffling and Other Stray Light Mitigation: mitigation of stray light is an important consideration for achieving the required S/N because the APOGEE wavelength range is small compared to the wavelength sensitivity range of the detectors; of particular concern is thermal light, because of the use of 2.5 μm cutoff arrays. Zeroth-order light transmitted through the grating is intercepted with a blackened panel behind the VPH. Of greater concern are first-order wavelengths outside of the nominal APOGEE wavelength range. Light shortward of 1.0 μm is absorbed by the four silicon elements in the camera. Thermal light is mitigated in several ways: (1) the fibers are cooled over 2 m of travel within the cryostat before reaching the pseudo-slit, (2) the "long-pass" dichroic on the front face of the second fold mirror and a broadband anti-reflection coating on the backside creates a light dump that intercepts some 95% of the residual thermal light ($\lambda \gt 2$ μm) before if gets to the grating, and (3) the silicon lenses in the camera have anti-reflection coatings that, together, permit transmission of only 9% (3%) or the thermal light to the detectors at 2.3 μm (2.5 μm).

Calibration Box: unlike the optical SDSS spectrographs, which take wavelength calibration images by illuminating the closed telescope covering petals, APOGEE employs a separate, off-telescope calibration module that enables access to calibration lamps at any time. When not attached to a bright time plugplate cartridge, the APOGEE gang connector can be connected to separate fiber runs that terminate at an integrating sphere that can illuminate the fibers with nominal $f/5$ light (to mimic the telescope) with either a ThArNe hollow-cathode lamp, a UNe hollow-cathode lamp, or a tungsten halogen light source. During commissioning and testing, the sphere also at times was illuminated with a precision-controlled blackbody source. Two possible fiber links to the integrating sphere are available: (a) a "DensePak" bundle with a full set of 300 fibers, and (b) a "SparsePak" bundle that sends light to every sixth fiber in the spectrograph focal plane. The latter is particularly useful for evaluating the wings of the PSF and the effects of scattered light.

Instrument Control: at the observer level, operation of the APOGEE instrument takes place through scripted sequences in the Sloan Telescope User Interface (STUI; Section 5.1). For manipulation of the spectrograph detectors, the STUI interfaces with a Digital Signal Processor (DSP) based controller that provides both clocking and bias/power supply voltages to the three arrays. All three share a common clock and most of their bias lines, with just a few power supply voltages unique to the individual arrays. This ensures common timing for the three arrays as they are read out. The readout scheme utilizes "sampling up the ramp" (SUTR), where the arrays are clocked and read out continuously and non-destructively with a period of 10.7 s. The data are formatted as a single output image containing the data for all three arrays, and including the three H2RG reference outputs. Because the array clocking is DSP based, the interval between reads is very stable, which allows for an accurate curve of growth analysis of the developing signal in each pixel (Section 6.1). This is facilitated by the rearrangement of the 2D, three-array data frames into time series data cubes for each array during post-processing of the data for each observation.

A white paper describing the potential of high throughput, multifiber, NIR spectroscopy on the SDSS 2.5 m telescope was presented to the Astrophysical Research Corporation (ARC) Futures Committee by Skrutskie & Wilson (2015) in 2005 August. The APOGEE project, refining the concept to a focus on HR spectroscopy of Milky Way stars, was proposed as an SDSS-III⁶⁴ project in 2006 August and was officially approved by the ARC Board as one of the four SDSS-III projects in 2006 November. The APOGEE instrument Conceptual Design Review was held in 2008 April, with the goal of having the spectrograph collecting data on the Sloan Telescope by 2011 Q2. The instrument Preliminary Design Review took place in 2009 May, with approval to start fabrication given at a Critical Design Review (CDR) held in 2009 August. Despite the technical challenges enumerated in Section 3.2.1, the primary APOGEE hardware construction phase spanned only 16 months from CDR to obtaining spectra of bright stars in 2011 February.⁶⁵ The instrument was delivered to APO in 2011 April and on-site first light occurred on the evening following deployment of the fiber train, on 2011 May 6—consistent with the original instrument development schedule.

Soon after, commissioning observations revealed that while instrument performance met, or exceeded, the requirements, it was also suffering from significant astigmatism preventing simultaneous focus in the spatial and spectral directions. In addition, placement of the detectors, particularly the array recording the reddest wavelengths, was non-optimal. While the source of the astigmatism has yet to be identified confidently, it was quickly mitigated by introducing a slight cylindrical bend on the first fold mirror by inducing a calculated axial force along the center of the mirror backside. On the other hand, the realignment of the focal plane arrays, which required shipping the entire detector assembly package to the University of Arizona, was not performed until APO "summer shutdown" in 2011 July. Thus, APOGEE observations taken from 2011 May–July suffer a reduced resolution of R ∼ 15,000 across the "red" detector array. Observations collected during this phase of operations are commonly referred to as "pre-shutdown" or "commissioning" data; although released publicly, application of APOGEE data reduction and analysis pipelines to those data does not lead to data products within the science quality requirements and thus are generally not of survey quality. Though most of these initial observations were eventually repeated (see Figure 10), they are of some value, for example, for the study of time series phenomena.

Official APOGEE survey data collection commenced after summer shutdown, 2011 August. The instrument parameters given in Table 2 pertain to this in-focus configuration of the APOGEE spectrograph, which was maintained throughout the remainder of SDSS-III operations (which concluded 2014 July). Throughout this period, APOGEE observations were smoothly carried out by the SDSS observers with minimal daily oversight by the APOGEE team and no loss of time due to instrument problems.

The overall instrument performance is obtained from a variety of test data taken on site. Table 2 summarizes the instrument characteristics. Much greater detail on the instrument performance can be found in J. C. Wilson et al. (2017, in preparation), Nidever et al. (2015), and Holtzman et al. (2015).

Wavelength Ranges: while the APOGEE spectrograph was designed to meet technical specifications that included the specific wavelength limits set by the 15160 Å potassium line and the 16720–16770 Å aluminum lines (Section 2.3), the spectral range recorded by the detectors extends almost to 1.7 μm (Figure 5), although these "extra" wavelengths were not designed to meet the resolution, throughput, and other technical specifications and did not drive design considerations. Of course, because of the physical limitations of butting detectors together, the spectral coverage is interrupted by interchip gaps. The exact wavelength region falling onto the array ensemble can be controlled by micro-positioning of the dithering stage, but the default position of the instrument delivers the wavelength regions on each chip as shown in Figure 5 and given in Table 2.

PSF, LSF, and Resolution: image quality can be judged from the delivered line-spread function (LSF) and PSF across the arrays. The PSF has an FWHM of typically 2.16, 2.14, and 2.24 pixels at the fiducial centers (1.54, 1.61, and 1.66 μm)⁶⁶ of each of the three array wavelength spans (Figure 5). The wings of the fiber PSFs reach far enough from the peak that there is a small amount of overlap between the PSFs of adjacent fibers on the detector focal plane. When the magnitude difference between stars on adjacent fibers is large, contamination of the spectrum of the fainter one by the wings of the brighter spectrum can become important. The amount of contamination varies across the three arrays, but analysis of commissioning data showed that between ∼0.1% and 0.2% of the total power of the PSF is located within 3 pixels of the central pixel of the adjacent PSF. It is for this reason that the fiber management scheme described in Section 3.1 was implemented. The LSF also varies across the arrays, both as a function of wavelength and fiber, as discussed by Nidever et al. (2015, see their Figures 14–16). In particular, it is slightly undersampled in the blue part of the spectrum, which is what motivated our use of the detector array dithering mechanism during observations (Sections 2.3 and 3.2.1). The resulting resolution in the properly sampled spectra varies by ∼25%, peak to peak, being higher at shorter wavelengths. Typical values at 1.55, 1.61, and 1.66 μm are R = 23,500, 23,400, and 22,600 (for details, see Nidever et al. 2015).

Instrument Throughput: observations of stars with well-known 2MASS magnitudes make possible empirical estimates of the throughput of the APOGEE instrumental apparatus. The end-to-end (i.e., from primary mirror to detector) measured throughput has a peak of 20% ± 2% at $\lambda \sim 1.61$ μm. This number is somewhat higher than expected from predictions based on the product of the component-by-component (measured or manufacturer-supplied) wavelength-dependent throughputs (see Table 2). However, the throughput varies as a function of wavelength, with flux dropping by about 50% from the middle of the spectral range to the red end (J. C. Wilson et al. 2017, in preparation). The main factor contributing to the dependence on wavelength is the VPH grating efficiency, with the telescope correctors contributing to a lesser extent (see Table 5 of J. C. Wilson et al. 2017, in preparation). Of course, the behavior of the S/N as a function of wavelength also depends on stellar color. These throughput measurements have obvious implications for the S/N achieved under survey conditions; those are discussed in Section 7.2. However, a key to the ${\chi }^{2}$-minimization procedure adopted by the APOGEE software (specifically, ASPCAP—see Section 6.5) is that it takes into account the pixel-to-pixel variation in S/N.

Array Persistence: as with most Teledyne infrared detector arrays, those installed in the APOGEE instrument have a small degree of image persistence, which results in the carryover of latent charge from exposure to exposure. Even after particularly strong "stimulus exposure," the residual charge impacting subsequent exposures is minor, so that normal persistence does not typically affect most APOGEE data (see Section 4.8 in Nidever et al. 2015). However, roughly one-third (in the spatial direction; see Figure 5) of the detector used for the bluest wavelengths is affected by excessive and long-lasting "superpersistence," which appears to behave like normal persistence, but with significantly greater accumulated charge and a very long time constant (see Section 5 of Nidever et al. 2015). Thus, intensely exposed pixels on one image can yield inordinately "hot" pixels in subsequent exposures. A small portion of the "green" array⁶⁷ (a thin "frame" around the edges) is also affected. The seriousness of this phenomenon has had a strong influence on our observing procedures—e.g., the timing and strength of calibration exposures and the imposition of a bright limit to targeted sources—with the goal of limiting unnecessary overexposure whenever possible. This problem also motivated the installation of the cold shutter (Section 3.2.2) to prevent stray light from entering the instrument when not in use. While there is hope that the effect of the superpersistence on the data may be correctable in software, it is a complex hysteresis problem that we currently have not fully resolved and no mitigation is currently implemented or planned for up to, and including, the future Data Release 14. Holtzman et al. (2015) explore the effects of the superpersistence (see their Section 5.7) and find that while the derivation of stellar parameters does not seem to be impacted significantly, the derivation of certain elemental abundances (N, Mg, S, K, Ca, Ti, Mn, and Ni) may be affected for stars for which some of their visit spectra land in the affected array regions. To dramatically reduce the impact of superpersistence on post-SDSS-III observations with the APOGEE spectrograph (Section 9.2), the shortest wavelength detector was replaced in the summer of 2014.

Ghosts: Despite mitigation efforts, there remain at least two in-band sources of stray light in the form of ghosts: (1) a Littrow ghost of each fiber (created by light reflected off the detector surface, recollimated by the camera, recombined by, and reflected from, the grating, and reimaged by the camera onto the detector; Burgh et al. 2007) forms on the detector at 0.4% the intensity of all of the recorded spectral light in each fiber near the spectrograph Littrow position at 1.604 μm. Because the pseudo-slit is actually curved, the Littrow ghost centers this excess light at a slightly different wavelength for each fiber, ranging from 1.6056 to 1.6067 μm (∼35 pixels; Figure 4). This spectral region was chosen through an optimization procedure aimed at minimizing the impact of the loss of absorption lines due to ghost overlap on the quality and diversity of the final APOGEE abundances. Optimal ghost positions were identified for which only very few interesting lines are lost, and for which in all cases there are other lines for the same element making up for the missing ones. The final position was selected so as to minimize any additional grating tilts that could lead to a substantial change in spectral resolution. The resulting spectral interval happens to coincide with the natural position of the Littrow ghost for the nominal APOGEE grating with no fringe tilt. The FHWM of the ghost in the wavelength dimension is about 9 Å (∼32 pixels). (2) Fiber tip ghosts occur from light that reflects off the detector face, transits through the entire instrument in reverse, reflects off the fiber face (or metallic parts adjacent to the fiber), and returns through the instrument a third time, back to the detector. While ghost intensity varies with wavelength and fiber position, stray light analysis of the optical design predicts that the ghost images should have spot size rms radii approximately 2–5 times larger and intensity $\leqslant 1.5 \%$ of primary images at the detector. Moreover, ghost images should arrive within 1 pixel of the primary image positions. No indication of the presence of such ghosts have been found in APOGEE frames, which suggests that they are indeed arriving very near the primary image.

Examples of the appearance of stellar spectra as obtained by the APOGEE spectrograph are shown in Figures 6–9. Figure 6 shows stars ranging from spectral type O to M; the primary APOGEE science targets are of type G and K, whereas most of the early spectral types were observed as telluric standards and some M types were selected by the random sampling of the parent distribution (Section 4.2). Across the temperature range of the primary survey target types (G-K stars), it is still possible to discern line strength variations, as shown in Figure 7. A primary driver of the APOGEE project is the exploration of chemical abundance variations among its late-type stellar sample; Figure 8 demonstrates the appearance of RGB stars of similar temperature but a 2.2 dex metallicity spread. To show greater detail and a broader array of chemical species, Figure 9 highlights the blue array spectra for two giant stars separated by about 2.2 dex in [Fe/H].

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

4.1.1. Field Selection Principles

4.1.2. Field Selection Evolution

Initial Survey Design—Co-Observing and Deep Fields: for its expansion into bright time observing, the SDSS-III collaboration planned to capitalize on the existence of two new fiber-fed instruments that could operate simultaneously from shared plugplates. Because the MARVELS project required many visits (≥24) to the relatively bright stars in each of its target fields, whereas APOGEE had always been planned to have at least some deep field probes, the original SDSS-III scheme was for 75% of the bright time to be in co-observing mode, with the remaining 25% of bright time given to APOGEE to observe fields of no interest to MARVELS.

This 75% shared survey time was distributed in a series of 24- and 12-visit fields across fields predominantly sampling the disk or halo (the 12-visit fields being those that MARVELS began observing before APOGEE started). The disk fields were chosen in a regular "picket fence" Galactic longitude distribution with the multiple visits at each picket cycling through different plate designs that allow for more total visits for fainter stars. The adopted distributions of the field Galactic latitude and target magnitude were based on modeling stellar population distributions using both the TRILEGAL (Girardi et al. 2005) and Besançon (Robin et al. 2003) Galaxy models and meant to ensure ample representation of stars from the Intermediate Population II, "thick disk". Further information about this modeling is given in Appendix D; an example of the results is given in Figures 11 and 12.

The co-observed halo pointings focused on fields containing globular clusters, which serve as both science and calibration targets. Typical globular cluster stars having HR spectroscopy in the literature are faint enough to require deep APOGEE observations. The multiple globular cluster visits also make it possible to increase the number of cluster stars to be sampled, given the limitations posed by fiber collisions (Table 2). Additional high-latitude deep pointings were placed in fields traversed by halo substructures, such as the Sagittarius stream (with field placement guided, e.g., by the results of Majewski et al. 2003) or the Virgo Overdensity (e.g., Vivas et al. 2001; Newberg et al. 2007; Jurić et al. 2008).

The remaining 25% of bright time was distributed to various classes of "APOGEE-only" fields: (1) across the bulge, a critical Galactic region for our primary science goals (Section 1.3), (2) at low declinations not viable for MARVELS work, (3) filling in the relatively large gaps between the disk deep field "pickets", and (4) toward open clusters useful for calibration at high metallicity. The limited accessibility for fields in classes (1) and (2), but a desire for statistically significant stellar samples and good spatial coverage in these regions, resulted in most of them being reduced to having single 1 hr visits, which then mandated brighter magnitude limits there (H = 11.1, see Section 4.2.3) to achieve our target S/N. Moreover, these fields had to be reduced to only a 1°–2° diameter because of the severe differential refraction experienced over the course of a 1 hr visit at high airmass.

This observing plan dictated the survey selection function for initial observations (much of 2011).

Reconfiguration at MARVELS Descope: just prior to commencement of formal APOGEE survey operations, a decision was made to curtail the MARVELS program over the course of the following year. A select number of MARVELS fields that had already obtained at least 12 pre-APOGEE epochs of MARVELS observation were chosen for completion, but reduced to either 6 or 12 hr APOGEE visits. Thus, in the end, only a handful of the original 24 hr fields were preserved (primarily the "deep disk midplane spokes" at l = 30°, 60°, and 90° and a few globular cluster fields: see Figures 10 and 11).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

However, the sudden, substantial increase in "APOGEE-only" bright time observing allowed a number of new pointings to be added to the baseline APOGEE field plan:

• a.  
  more bulge pointings, including fields useful for cross-calibration to the BRAVA (Rich et al. 2007) and ARGOS (Freeman et al. 2013) surveys;
• b.  
  additional calibration of open and globular clusters;
• c.  
  numerous 3 hr fields to give finer angular sampling at (b = 0°, ±4°, ±8°, and ±12°) between the preserved 12/24 hr pickets at l = 30°, 60°, 90°, 120°, 150°, 180°, and 210°;
• d.  
  rings of high-latitude fields at b = +30°, ±45°, +60°, and +75°.

The greater flexibility afforded by the increased control over field placement also aided in the implementation of (a larger set of) Ancillary Science programs.

Incorporation of the Kepler Field: The success of ESA's CoRoT mission (Auvergne et al. 2009) and NASA's Kepler mission (Borucki et al. 2010), and, in particular, the asteroseismology programs for each (Michel et al. 2008; Chaplin et al. 2010; Gilliland et al. 2010)—through which non-radial oscillations were detected and characterized for a substantial sample of RGB stars and subgiants (Mosser et al. 2010; Hekker et al. 2011)—presented a special opportunity for the APOGEE program. The asteroseismic frequencies are sensitive probes of stellar masses and radii (Chaplin & Miglio 2013). Apart from providing invaluable independent measurement of stellar gravities for testing and calibrating the APOGEE stellar parameters pipeline (Section 6.5), when combined with precision abundance measurements of the quality that APOGEE could provide, asteroseismically measured stellar masses can provide reliable age estimates, at the level of 15% (Gai et al. 2011). The opportunity to obtain such reliable age data for a large number of field stars is unprecedented and provides pivotal temporal benchmarks for a survey of GCE. Moreover, the APOGEE instrument presents the only practical means to obtain HR spectroscopic assays for a large fraction of this Kepler sample, because, serendipitously, the FOV of the SDSS plugplates is nicely matched to the size of a Kepler tile.

After formally establishing collaborative agreements—the APOGEE-Kepler Asteroseismic Science Collaboration (APOKASC) and the CoRoT-APOGEE Collaboration (COROGEE)—plans were put into place to reorganize substantially the APOGEE targeting to include a large fraction of the KASC giant/subgiant sample and CoRoT giants in their Galactic center/anticenter fields. While deep paintings of the two Kepler tiles containing the star clusters NGC 6791 and NGC 6819 had been in the original APOGEE plan, the remaining 19 Kepler tiles were now included with two 1 hr visits having unique sets of targets each.⁶⁸ This resulted in observations of (a) some 8000 APOKASC giant stars, along with (b) about 600 subgiant and dwarf stars, whose ages could be determined using gyrochronology (e.g., van Saders & Pinsonneault 2013), as well as (c) targets for other ancillary science programs (e.g., eclipsing binaries). Further details on the Kepler field targeting can be found in Zasowski et al. (2013) and Pinsonneault et al. (2014). APOGEE also targeted 121 giant star candidates in one plate designed for the CoRoT LRa01 ("anticenter") field and 363 giant candidates on three plates designed for the LRc01 ("center") field.

Survey Year-three Modifications: By the third observing year, overall clearer than average winters put the survey of the anticenter disk ahead of schedule. This enabled expansion of the Galactic anticenter grid with broader and denser coverage—an advantage for exploration of the disk warp, disk flare, and low-latitude substructure like the Monoceros and TriAnd structures—as well as the expansion of several Ancillary Science programs at these LSTs. On the other hand, our somewhat lagging spring and summer field schedule was greatly aided in the final year by the addition of both twilight and dark time observing campaigns (Section 5.3), which were key to completing fields in and around the Kepler field.

4.1.3. Final Field Plan

APOGEE targeting consists of (1) the "main sample" or "normal science targets", (2) "special targets", which include (among others) calibration stars with measured stellar parameters and abundances from other spectroscopic studies, star cluster members, and targets submitted by one of APOGEE's Ancillary Science programs, and (3) a sample of early-type stars observed as telluric absorption monitors for each exposure. A complete and detailed discussion on how each of these targets is selected, and how they are identified within the publicly released databases, is given by Zasowski et al. (2013). We only give a broad overview here, with an emphasis on motivations for the overall procedures followed.

4.2.1. "Minimum Criteria" Philosophy

4.2.2. Source Catalogs and Supplemental Data Used

4.2.3. Main Survey Targets

4.2.4. Calibration Fibers

Several motivations led to the inclusion of an ancillary science program in the APOGEE survey plan: (1) the APOGEE spectrograph, its mating to the very large FOV Sloan 2.5 m Telescope, and the extremely effective multifiber optical interface between the two represent a unique, state-of-the-art capability applicable to a broad range of groundbreaking Galactic science applications that may not fall within the purview of the primary APOGEE mission. (2) Not all interesting and relevant Galactic science could be included in the primary survey program, but could be addressed in a limited way through an ancillary science program. (3) Some science programs that might be worth pursuing as main survey science require some verification and testing in pilot programs. (4) Leaving some amount of survey time in reserve allows the opportunity to respond to new developments in the field or to incorporate originally unanticipated science of great value.

Given these motivations, 5% of the total fiber-hours⁷⁰ of the APOGEE survey were made available for a formal APOGEE Ancillary Science Program. The main criteria for selecting such proposals were that the ancillary observations result in novel and compelling scientific contributions and that they not impact negatively the primary objectives of the APOGEE survey. Especially compelling were proposed programs that could enhance the productivity and impact of the primary APOGEE survey. Several of the most meritorious proposals in the APOGEE calls for ancillary science that had as primary goals the improved calibration of the APOGEE database. A few approved ancillary science programs served as the basis for a major redefinition of APOGEE targeting to include significant attention to Kepler mission targets (Section 4.1.2).

Three calls for proposals to the APOGEE Ancillary Science Program were solicited: 2010 September, 2012 March, and 2013 March. Two flavors of ancillary science targeting were implemented: (a) sets of individual fibers placed on specific targets in already-existing APOGEE survey pointings, and (b) use of up to all ∼230 APOGEE "science" fibers in a new pointing not already within the general APOGEE survey plan. The selected Ancillary Science programs are described in Appendix C of Zasowski et al. (2013). Note that, while all collected APOGEE spectra are automatically processed through the data reduction and analysis pipelines, for some of the programs focused on targets significantly different from those in the main survey, there is no guarantee that the automatically generated data products are optimal, or even reliable. All special processing and analysis of Ancillary Science Program data are the responsibility of the principal investigators of each selected project.

Once prioritized lists of selected targets (science, telluric calibrators, sky positions) have been generated for each plate design, they are fed to standardized SDSS plate design software. This software takes the input targets' celestial coordinates and generates the final linear (x, y) plugplate drill pattern for the plate design. The software accounts for potential fiber collisions between all fibers from both APOGEE and MARVELS, as well as collisions between science fibers and acquisition or guide fiber bundles. The algorithms also take into account the field curvature of the Sloan 2.5 m Telescope (to which the plugplates are bent during observing) and the differential refraction expected for the nominal hour angle at which each plate of a given declination might be observed. In some cases, due to the uncertainty in scheduling, multiple plates might be generated from the same plate design input files, differing only in the potential hour angle of observation.

In addition to establishing the precise coordinates for each star based on refraction considerations, the plate design code also sorts the intended targets into three magnitude bins of 100 stars each for the "red/green/blue" fiber management scheme described in Section 3.1. Figure 14 illustrates how this fiber management scheme creates a repeating pattern of variable spectrum brightness as a function of fiber pseudo-slit position as projected onto the spectrograph focal plane.

The fiber plugplates are drilled on a six-axis, computerized (CNC) milling machine at the University of Washington, and then shipped to APO. At APO, the plates are manually marked to identify which holes correspond to stars designated to red, green, or blue-sheathed fibers by way of an overhead projection onto the aluminum plate of the fiber plugging color scheme (Figure 13). Note that the red/green/blue = bright/medium/faint division of stars in each plate design is not directly correlated to any designated cohort divisions, except as the sorting by magnitudes of stars in the cohorts places them into an appropriate fiber color by default.

As with all SDSS observing, APOGEE observing was typically conducted with the use of a package of standard operating scripts that orchestrate nightly activities through the observatory STUI (Section 3.2.2).

APOGEE science observing was based on standardized "visits" (Section 2.8) to a scheduled set of fields using corresponding plugplates designed and drilled for specific hour angles (Section 4.4), and plugged with fibers in advance. Each standard visit consisted of eight 500 s exposures taken at two array dither positions ("A" and "B"; Section 3.2.1) in two ABBA sequences. A 500 s exposure consists of a sequence of 47 detector readouts, performed in intervals of 10.7 s, which generates a SUTR data cube (see Section 3.2.2). This ∼67 minute exposure sequence plus two dark exposures taken during the change of the plugplate cartridges yields a typical visit length of 75 minutes. A plugplate was typically revisited on multiple nights to build up the required S/N according to the cadence rules described in Section 2.8.

To operate usefully in less than ideal weather conditions and to take full advantage of extra pockets of observing time, guidelines had to be established for maximizing the usefulness of "non-standard visits". Therefore, a minimum data quality to count a visit as successful was set at at least one AB dither pair with each 500 s exposure having ${\rm{S}}/{\rm{N}}\geqslant 10$, the minimum needed to derive the stellar radial velocity at the required survey precision.⁷¹ To aid in the assessment of exposure quality, the observers had access to "quick look" reductions (simplified versions of the data reduction pipeline; Sections 6.2–6.3) of the data in near real time that produced plots of accumulated S/N as a function of magnitude. Over the course of a night, the available APOGEE time would be divided into standard field visits, with any additional observing time allocated to either gathering extra S/N on a particular plate or creating a "short visit" with a new plate, at the discretion of the observing staff, to maximize observing efficiency. In particular, by reference to the "quick look" reductions, the last visit with any particular plugplate could be shortened or lengthened as appropriate to ensure that the target total S/N for that plugplate was obtained.

Because telescope guiding is done at optical wavelengths, APOGEE plates were observed with the guiding software making refraction corrections to keep 1.6 μm light in the fibers. In the case of fields observed jointly with MARVELS, the guiding wavelength was set to a compromise wavelength of 1.1 μm.

The stability of the APOGEE instrument relaxes the requirements on the amount of calibration needed on a nightly basis. At the beginning and end of each night with potential APOGEE observations, the gang connector is connected to the calibration box to collect a standard calibration sequence that includes long dark frames as well as exposures of the tungsten halogen, ThArNe, and UNe lamps at both dither positions (Section 3.2.2). At the end of the night, we also take a set of internal flat fields. In addition, once each night 4× ABBA exposures are taken with all fibers on sky; the resulting airglow spectra are used for monitoring the LSF and PSF of the instrument.

A full observing night can generate ∼100 GB of data, which are then compressed and transferred from the mountain to the Science Archive Server (SAS; see Section 8.3), where they are stored in disk. These raw data consist of large data cubes containing all the 47 readouts making up every single 500 s exposure. The subsequent processing and reduction of these data are described in Section 6.1.

From 2011 Q2 to 2014 Q2, APOGEE (and, initially, MARVELS in parallel) operated during all bright time (lunar phase $\lt 39 \%$), as well as all "gray" time (lunar phase 39%–56%) for LSTs when the North Galactic Cap was not visible. APOGEE observations pushed the Sloan Telescope to several new observing regimes and limits—e.g., with respect to lunar phase, airmass, twilight, cadencing, and sharing of the focal plane by two different instruments (Figure 15). With a number of observing constraints different than those required by the optical programs, integrating the APOGEE program into SDSS operations added new layers of complexity to telescope scheduling and plugplate cartridge organization, especially on nights shared between all three operating surveys (BOSS, MARVELS, and APOGEE). Within the APOGEE portions of nights, internal scheduling software was developed to organize the nightly observing for efficiency, and to account for APOGEE observing constraints, as well as those for MARVELS during joint operations. These APOGEE scheduling constraints included:

• a.  
  Moon avoidance: Observations were not allowed within 15° of the moon (30° for MARVELS shared observations). However, because the ecliptic passes directly through the Galactic bulge, this limit was loosened to 10° for bulge observations; without this adjustment, the amount of potential bulge observations would have been reduced by 50%.
• b.  
  Airmass limits: The central regions of the Galaxy, containing highly prized APOGEE targets, transit at very high airmass at APO. Compared to the optical, NIR observations benefit from reduced differential refraction and atmospheric extinction, which made it possible to undertake the desired extreme airmass observations. One important limitation, however, is the still significant differential atmospheric refraction at low elevation, which forced the adoption of more limited drilled areas (1°–2°) on the plugplates. Despite the smaller angular coverage, it was easy to fill all the science fibers in these fields, due to the high stellar density of the central regions of the Galaxy. Those fortunate advantages made it possible for APOGEE to probe the Galactic bulge, the Galactic center, and even farther south (to $\delta =-32^\circ$). Hardware limits set the maximum APOGEE airmass (X) to X < 3.2, but a limit of X < 1.7 was necessary for MARVELS co-observed plates. APOGEE utilized the standard $X\gt 1.01$ limit of the Altitude-Azimuth mounted Sloan Telescope.
• c.  
  Hour angle: The APOGEE windows of opportunity were set so that plates had to be observed with no more than 0.5 arcsec of differential refraction across the plate. However, the reduced H-band differential refraction also allowed greater APOGEE flexibility in observing plugplates farther from their nominally drilled hour angles than is possible for optical observations.
• d.  
  Plate cadence: As discussed in Section 2.8, the nominal survey plates had to be observed over at least three visits each meeting the minimum S/N per visit requirement (Section 5.1) with separations of at least 3 days between the two closest observations and at least 25 days between the first and last observation.

Survey plate scheduling was done by a module originally designed to optimize cadence observations for the MARVELS survey that was later adapted to account for both cadence and S/N constraints of the APOGEE survey. Beyond accounting for the above constraints, the scheduling software invoked several additional rules to optimize efficiency.

For example, bulge plates and other plates with limited observability windows were given highest priority. Other plates were given relative priorities that accounted for their individual cadence histories and net accumulated S/N. Special attention was needed for scheduling of "non-standard" visits to take advantage of occasional extra pockets of observing time. For example, standard visits for the eight bright time cartridges were insufficient to fill the available time on long winter nights; in this case, longer than standard visits could be applied to, e.g., (a) halo plates that have been designed with fainter than main survey stars (see Section 4.2.3), (b) plates that, due to poor weather or prematurely ended previous visits, were behind on S/N accumulation despite satisfying cadence constraints, or (c) plates that could, conversely, be "pre-loaded" with extra S/N allowing useful, but shorter than standard, visits on other (e.g., shorter) nights. In the interest of steady progress on the completion of fields, another, albeit more loosely followed, scheduling strategy was that the full set of observations for nominal, three-visit plugplates, if at all possible, not stretch beyond one observing season.

On long nights, when the full eight fiber plugplate setups could be observed, APOGEE was able to record spectra for 1840 target stars, along with 280 hot telluric star calibrators and 280 sky fibers (Section 4.2.4).

5.3.1. Twilight Observing

5.3.2. Year 3 and Dark Time Campaign

5.3.3. Bright Standard Star Calibration

The APOGEE program consists of two distinct observing campaigns—"commissioning" (2011 May–July) and "survey" (2011 August–2014 July)—divided by the change in spectrograph optical configuration during the shutdown in summer 2011 (see Section 3.3). "Commissioning" observations consisted primarily of one-visit and three-visit fields to test instrument performance, calibration, and limitations. The "survey" observations were conducted over the originally intended three-year APOGEE campaign from 2011 August to 2014 July and produced survey-quality data during 520 days spanning over 1900 individual field visits. The entire three-year survey campaign was conducted uninterrupted, with the instrument continuously sealed and cold in the same optical state to provide an extremely uniform data set.

At the end of every observing night, APOGEE data are compressed and transferred to the SAS (Section 5.1), and all data reduction is done subsequently off the mountain. In the following, we briefly describe the steps leading to the generation of a final APOGEE combined spectrum.

In this first processing stage, each data cube is corrected for standard detector systematic effects and converted into a 2D array. Every individual readout is corrected for bias variations in the detectors and electronics. Bias measurements are performed on a combination of pixels generated by the readout electronics and a set of reference pixels around the edge of each detector. Next, a dark frame resulting from the combination of multiple individual exposures is subtracted from each individual readout. The 2D arrays are then generated through linear fits to the time series of SUTR readouts for each pixel, and the best-fitting slope is multiplied by the exposure time to generate the final pixel counts. The process allows for detection, correction, and flagging of pixels affected by cosmic rays. Finally, 2D arrays are corrected for pixel-to-pixel sensitivity variations through division by a normalized flat-field frame. The output of this reduction step for one visit is eight calibrated 2D arrays, four for each dither position.

As a next step, spectral extraction and wavelength calibration are performed on each 2D array. Spectra are extracted through modeling of the spatial PSF of all 300 fibers as a function of wavelength in a way that accounts for the overlapping of the PSFs between adjacent spectra. The model is fit to a high S/N flat-field frame obtained immediately after each science exposure.

Wavelength calibration is the next stage of the reduction, and as usual, is based on arc lamp exposures. Because each fiber occupies a different position in the pseudo-slit, fiber-to-fiber wavelength scale variations exist, so that individual calibrations for each fiber are necessary. The APOGEE spectrograph is stable enough that a single polynomial relation is adopted for each fiber, with zeropoint corrections applied on the basis of measurements of central wavelengths of airglow lines. In conformity with previous SDSS standards, APOGEE adopts vacuum wavelengths. For details of the adopted conversion between vacuum and air, see Nidever et al. (2015).

The overall wavelength scale suffers drifts linearly over time, due to a slowly varying flexure in the instrument optical bench as the liquid nitrogen tank depletes over time (Section 3.2.2). Every time the tank is refilled, the scale undergoes a large "reset" shift, which brings it back to the original scale. These shifts are measured using a set of bright airglow lines and the wavelength scale is corrected accordingly. The accuracy of the resulting wavelength solution at any given pixel of an APOGEE spectrum is of order 0.1 pixel or 0.03–0.04 $\mathring{\rm A}$ (Nidever et al. 2015). The outputs of this reduction stage for one visit are eight wavelength-calibrated 1D dither spectra, four for each dither position.

In the next reduction stage, dither pairs are combined into well-sampled 1D spectra, sky subtraction is performed, and the signature of telluric absorption is removed.

The shift between the spectra in each dither pair is determined to high accuracy through cross-correlation of the two spectra. Before combination, each dither spectrum is subject to sky subtraction. The H-band sky background is characterized by strong OH emission lines and a faint continuum, which is stronger in the presence of clouds and moonlight. The contribution of sky background to the spectrum of any science fiber is determined through interpolation of the spectra of the four closest fibers from among the 35 sky fibers distributed across the APOGEE FOV (Section 4.2.4). Because of fiber-to-fiber LSF differences, the subtraction of sky lines is not perfect, and can result in the presence of significant residuals in pixels situated at or near the positions of very strong lines that render these pixels useless for science. While future improvements in the reduction pipeline may ameliorate the situation, the S/N in those pixels is nevertheless unavoidably deteriorated due to high Poisson noise.

Telluric line absorption in the APOGEE spectral region due to the rovibrational transitions of the H₂O, CO₂, and CH₄ molecules is removed through the fitting of telluric absorption models to observations of the 35 telluric standards distributed across the field (Section 4.2.4). For each telluric standard, synthetic telluric spectra based on model atmospheres by Clough et al. (2005) are fitted to the full family of absorption lines from each molecule separately to generate scaling factors to the model spectrum of each molecule at the position of each telluric calibration fiber. Polynomial surfaces are then fitted to describe the spatial variation of the scaling factors, and the correct scaled model is determined for each science fiber through interpolation within those surfaces. For each science fiber, models are then convolved with fiber-specific LSFs and divided into the science spectrum. Nidever et al. (2015) provide a quantitative assessment of the performance of the telluric line correction. They find that the average error in telluric correction is of order ∼1%–2% of the stellar continuum (see Nidever et al. 2015 for more details).

Although the above telluric correction method works well, it has shortcomings related to errors in the wavelength solution, and uncertainties in both the telluric absorption model and the adopted LSFs. Because a large fraction of APOGEE pixels are affected by telluric absorption (albeit at a very minor level for the majority), improvements in telluric correction are a high priority for future pipeline improvements.

Each sky-subtracted, telluric-corrected pair of dither spectra are then combined into a single better-sampled spectrum, using the shifts determined as described above. Each of these resulting spectra are then coadded to generate a single visit spectrum.

Flux calibration consists of two steps. First, an approximate relative flux calibration is applied to dither spectra to remove the spectral signature of instrumental response; this response function was determined through observation of the blackbody spectrum from a calibration source (Section 3.2.2). Later on, after dither spectra are combined to generate visit spectra, the latter are scaled to match the object's cataloged H-band magnitude. Because the spectra are later reshaped through polynomial fits to the pseudo-continuum prior to performance of stellar parameter and abundance analysis, flux calibration is not a critical aspect of data processing.

Radial velocities (RVs) are one of APOGEE's key data products. There are two main steps related to the RV determination within APOGEE. One step determines relative RVs between different visits, and the other converts these measurements into an absolute scale.

In both steps, RVs are determined via a cross-correlation between the object spectrum and a particular template. Observed and template spectra are initially both in a log-linear wavelength scale, so that a Doppler correction can be performed by shifting all pixels by the same value. Before cross-correlation, bad pixels are flagged and the pseudo-continuum is normalized through a low-order polynomial fit to the spectrum of each detector separately. A Gaussian fit is performed to the cross-correlation distribution and the position of the peak and its error are converted into a velocity shift and uncertainty.

Visit RVs are determined through an iterative process via cross-correlation with the combined spectrum. Initial relative RVs, obtained from cross-correlation with the highest S/N visit spectrum, are used to bring all visit spectra to a common velocity scale, and making possible the production of an initial combined spectrum. The process is then iterated by adopting the most recently created combined spectrum as a template.

Absolute RVs are obtained through cross-correlation of the combined spectra with synthetic spectra from an "RV mini-grid," which is a subset of the APOGEE spectral grid (Section 6.5.1) and consists of 538 spectra over a wide range of stellar parameters and chemical compositions. With steps of 200 K in ${T}_{\mathrm{eff}}$ (increasing toward 1000 K in the very hot end), 1.0 dex in $\mathrm{log}g$, and 1.0 dex in [M/H], this spectral grid is much coarser than that adopted in the determination of stellar parameters (Section 6.5). Each combined spectrum is cross-correlated with all spectra of the mini-grid and the absolute RV resulting from the analysis is that inferred from the best-matching synthetic spectrum, which in turn is selected through ${\chi }^{2}$ minimization. The numbers resulting from this cross-correlation are further adjusted by the barycentric correction to produce heliocentric RVs (see Nidever et al. 2015 for details). The impact of template mismatch in the derived RVs was gauged by comparing DR12 values with those of Troup et al. (2016), who derived accurate RVs for a sample of 382 stars with companion candidates adopting as templates the best-matching ASPCAP synthetic spectrum, rather than the RV mini-grid. The average difference between the two sets of RVs is −0.05 km s⁻¹, with an rms of 0.94 km s⁻¹, which is entirely negligible for the vast majority of science applications of APOGEE spectra.

Elemental abundances are another primary data product of the APOGEE survey. Stellar parameters—effective temperature (${T}_{\mathrm{eff}}$), surface gravity ($\mathrm{log}g$), metallicity ([M/H]), and microturbulence (${\xi }_{t}$)—are also necessary stepping stones toward elemental abundances and spectroscopic parallaxes. An understanding of the possible systematic effects on the derived elemental abundances and distances inferred from APOGEE spectra requires a good grasp of the procedures followed for the derivation of atmospheric parameters and metallicities. In this section, a brief description of those procedures is provided, but the reader is referred to García Pérez et al. (2016) for details.

The ASPCAP routine implements a two-step process: (1) the determination of stellar parameters from a fit of the entire APOGEE spectrum to model spectra, and (2) the adoption of these parameters as inputs for a fit to relatively narrow spectral windows containing features associated with each particular element to derive its abundance. In the following subsections, we describe each of the main ASPCAP processing steps.

6.5.1. Grid of Synthetic Spectra

Stellar parameters are obtained through determination of the best-fitting synthetic spectrum from across an extensive grid spanning six stellar atmospheric parameter dimensions (${T}_{\mathrm{eff}}$, $\mathrm{log}g$, [M/H], [α/M], [C/M], and [N/M]) by ${\chi }^{2}$ minimization (Section 6.5.3). The accuracy of the results is fundamentally dependent on the fidelity with which spectra from the synthetic grid reproduce real stellar spectra. We briefly describe the main ingredients entering the calculation of this spectral grid and refer the reader to Zamora et al. (2015) for further details.

Synthetic spectra were calculated using the Advanced Spectrum Synthesis 3D Tool (ASS$\varepsilon$T) code (Koesterke 2009), adopting 1D model atmospheres calculated in local thermodynamic equilibrium (LTE) by Mészáros et al. (2012) and a line list customized for the analysis of APOGEE spectra (Shetrone et al. 2015; Appendix E). The adopted model atmospheres were calculated using the ATLAS9 code (Kurucz 1993), adopting newly computed opacity distribution functions as described by Mészáros et al. (2012) and the solar abundance pattern of Asplund et al. (2005), as well as variations in the abundances of carbon and α elements. Spectra were calculated over the following ranges of stellar parameters:

• a.  
  $3500\leqslant {T}_{\mathrm{eff}}\leqslant 15000\,K$
• b.  
  $0\leqslant \mathrm{log}g\leqslant 5$
• c.  
  $-2.5\leqslant [{\rm{M}}/{\rm{H}}]\leqslant +0.5$
• d.  
  $-1\leqslant [{\rm{C}}/{\rm{M}}]\leqslant +1$
• e.  
  $-1\leqslant [{\rm{N}}/{\rm{M}}]\leqslant +1$
• f.  
  $-1\leqslant \alpha /M]\leqslant +1,$

where all α elements are assumed to vary in lockstep. The separation between the grid nodes in ${T}_{\mathrm{eff}}$, $\mathrm{log}g$, and chemical composition parameters is not constant. For instance, steps of $250\,K$ in ${T}_{\mathrm{eff}}$ were adopted for stars cooler than $6000\,K$ because spectral features vary more steeply with ${T}_{\mathrm{eff}}$ in cool stars, whereas coarser steps of $1000\,K$ were adopted for warmer stars. By the same token, $\mathrm{log}g$ steps of 0.5 dex were adopted for ${T}_{\mathrm{eff}}\lt 6000\,K$ and 1 dex for warmer stars. Steps of 0.5 dex in [M/H] were chosen everywhere except for stars with ${T}_{\mathrm{eff}}\gt 8000\,K$, for which steps of 1 dex were employed. For both [C/M] and [α/M], steps of +0.25 dex were adopted, whereas in the [N/M] direction the grid is coarser, with steps of 0.5 dex instead. For warm stars with ${T}_{\mathrm{eff}}\gt 6000\,K$, no CNO variations are present, as these stars are too warm for any relevant molecular formation. For more details, see Table 1 of García Pérez et al. (2016).

The chemical compositions adopted matched those used in the generation of the model photospheres, except for the case of nitrogen, whose variation was not seen to affect the photospheric structure in an important way.

The line list resulted from an initial implementation of the Kurucz line list, improved by introduction of both theoretical and laboratory transition probabilities (gf values) following an exhaustive critical search of the existing literature, and further supplemented by laboratory values of key transitions obtained by our collaborators (e.g., Wood et al. 2014) by request (see Appendix E). Further refinement of gf values and damping constants was achieved through spectral synthesis of the solar and Arcturus spectra (see Shetrone et al. 2015 for details), where departures from laboratory values were capped at no more than twice the nominal uncertainties.

The synthetic spectra are smoothed to the APOGEE resolution (R = 22,500) by convolution with a single, empirically determined, average APOGEE LSF (Nidever et al. 2015; Holtzman et al. 2015) and sampled into a logarithmic scale to match the sampling of the APOGEE data ($\sim {10}^{4}$ wavelengths). Synthetic spectra are further normalized through fitting of a polynomial to the upper envelope of the spectrum, for comparison with observed spectra treated in the same way (see below).

Efficient computation would require storage of the entire spectral grid in memory, which is currently not practical. Therefore, fluxes are compressed using Principal Component Analysis (PCA) and it is the PCA-compressed grid that is compared with the observed spectra for atmospheric parameter determination. To expedite calculations further, the grid is split into two distinct subgrids, with ${T}_{\mathrm{eff}}$ spanning ranges approximating those of GK (3500–6000 K) and F (5500–8000 K) spectral types (see Zamora et al. 2015).

Each synthetic spectrum is characterized by seven parameters, namely ${T}_{\mathrm{eff}}$, $\mathrm{log}g$, [M/H], [α/M], [C/M], [N/M], and ${\xi }_{t}$ (microturbulent velocity). With multiple nodes in each parameter, the final seven-dimensional spectral subgrids consist of about 1.7 million (GK stars) and 1.4 million (F stars) spectra covering the entire range of expected atmospheric parameters and chemical compositions.

Abundances of individual elements are defined as follows:

$$\begin{eqnarray}&&[{\rm{X}}/{\rm{H}}]={\mathrm{log}}_{10}({n}_{{\rm{X}}}/{n}_{{\rm{H}}})-\,{\mathrm{log}}_{10}{({n}_{{\rm{X}}}/{n}_{{\rm{H}}})}_{\odot },\end{eqnarray} \tag{ 1 }$$

where n_X and n_H are the number, per unit volume of the stellar photosphere, of atoms of element X and hydrogen, respectively. The metallicity [M/H] is defined as an overall scaling of metal abundances for a solar abundance pattern, while [X/M] is the deviation of element X from that pattern:

$$\begin{eqnarray}&&[{\rm{X}}/{\rm{M}}]=[{\rm{X}}/{\rm{H}}]\,-\,[{\rm{M}}/{\rm{H}}].\end{eqnarray} \tag{ 2 }$$

Because the search for the best-fitting spectrum within a 7D space is considerably slow at present, the library dimensionality has been reduced to 6 (thereby reducing the overall size of the libraries by a factor of 5) by constraining microturbulent velocities through the adoption of a relation with surface gravity (see details in Holtzman et al. 2015 and García Pérez et al. 2016). For ${T}_{\mathrm{eff}}\gt 8000\,K$, where molecular lines are entirely absent, the grid is described by only three parameters, ${T}_{\mathrm{eff}}$, $\mathrm{log}g$, and [M/H].

6.5.2. Pre-processing of Observed Spectra

6.5.3. Stellar Atmospheric Parameter and Abundance Determinations

Upon concluding three years of operation, the APOGEE survey obtained over half a million spectra of 163,278 stars. Of these, 12,140 were obtained during commissioning but were never reobserved. Due to the issues discussed in Section 3.3, these commissioning data do not meet the survey science requirements. Therefore, the total number of survey quality targets delivered in DR12 is 151,138. Of these, 14,692 are telluric standards, which leaves a net total of 136,446 survey quality science targets. Thus, APOGEE exceeded by more than 35% the original technical requirement on sample size. Such a substantial increase over the required performance was achieved due to the combination of factors described in Sections 4.1.2 and 5.3.

Determining the sample breakdown according to Galactic component is not simple, because it requires disentangling overlapping stellar populations in every field, which is particularly challenging at low latitudes. Nevertheless, we provide below a simple breakdown of the sampled survey stars by rough field type, with the caveat that of course not all targets in a field belong to the Galactic component defining these types, which, for simplicity, we categorize broadly by Galactic coordinates:

• a.  
  13,473 stars in bulge fields ($| b| \lt 16^\circ$, $-10^\circ \lt l\lt 11^\circ$);
• b.  
  54,988 stars in halo fields ($| b| \gt 16^\circ$);
• c.  
  82,677 stars in disk fields ($| b| \lt 16^\circ$, $11^\circ \lt l\lt 350^\circ$).

The halo numbers are inflated by inclusion of fields targeting nearby stars at relatively large Galactic latitudes (some examples are the Kepler fields and Ancillary Science fields focused on nearby clusters or associations). Accounting for those, the total number of halo field targets drops to fewer than 40,000.

Along these lines, some specialized target classes of particular interest include:

• a.  
  12,443 stars in Kepler/CoRoT fields;
• b.  
  2035 stars in Sagittarius dSph core fields;
• c.  
  3782 stars in fields in the direction of other known halo substructure, including streams associated with the Sagittarius dSph;
• d.  
  7291 stars in fields placed on suspected halo overdensities from the Grid Giant Star Survey (Bizyaev et al. 2006; Majewski et al. 2012);
• e.  
  8112 stars in star cluster fields (fields specifically targeting open or globular clusters, but not counting disk fields in which open clusters were serendipitously observed);
• f.  
  12,115 objects in Ancillary Science fields;
• g.  
  880 bright stars observed with the NMSU 1 m telescope + the APOGEE spectrograph.

Figures 16 and 17 show the computed⁷⁴ Galactic spatial distributions for the main APOGEE survey targets and demonstrate that the targeting plan as implemented achieved its general goals for Milky Way coverage (compare to Figures 11 and 12). A full description of the data is given in Holtzman et al. (2015).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

As mentioned in previous sections, a unique aspect of the APOGEE survey is that the majority of the stars are visited multiple times. Figure 18(a) shows the distribution of numbers of stars having any particular number of visits over the course of the survey. As can be seen, the vast majority of stars have three visits, but a significant tail of stars have been visited many times more. For example, of order 1000 stars have more than two dozen visits. In many cases, these stars were observed for over a year, and some for as long as three years (Figure 18(b)). Such time series data enable a variety of scientific applications, including the search for stellar companions (Section 7.4.2) and stars with spectral variability (Section 7.4.1).

Zoom In Zoom Out Reset image size

Achievement of survey requirements in depth would only be possible if the instrument throughput met the original specifications of S/N = 100 pixel⁻¹, at the specified resolution and sampling, at H = 12.2 in 3 hr of integration time. Those numbers translate into an expected ${\rm{S}}/{\rm{N}}=105$ for a combination of three visits with eight exposures of 500 s duration each. Overall survey performance is, of course, also impacted by observing conditions (transparency, seeing) and procedures adopted to track and ensure accumulated signal over multiple visits. Delivered performance in this area is demonstrated by Figure 19, which shows the distribution of S/N measured in three-visit spectra of all stars (main science and ancillary targets, amounting to ∼1600 stars) in a narrow range of (undereddened) magnitudes (12.15 < H < 12.20) at the three-visit plate limit of H = 12.2.⁷⁵ The distribution is nearly Gaussian (despite the long tail toward very low S/N), with a peak at S/N ∼ 93. The FWHM spread in S/N is about 30 or $\sigma \sim 13$, which, for $\overline{{\rm{S}}/{\rm{N}}}=93$ translates into a ${}_{-0.32}^{+0.28}$ mag (1σ) dispersion in flux. A small fraction of this spread can be ascribed to photometric uncertainties in the 2MASS catalog (${\sigma }_{H}\sim 0.02$ mag at H = 12.2),⁷⁶ but most of it is likely due to variable seeing and extinction due to clouds, since conditions during data collection were often non-photometric. The latter reflects the realities of accommodating variable weather conditions within a semi-rigid observing plan structured on gathering set numbers of paired 500 s standard exposures for each plugplate (nominally a dozen pairs for each three-visit plugplate). Assuming that the whole of the scatter is due to weather and seeing variations, one finds that for the best observing conditions, represented by those observations more than 1σ (2σ) above the median of the distribution, the S/N achieved was about 106 (118), indicating that the spectrograph delivered, and probably exceeded, the throughput needed to meet the S/N requirement established at the survey outset.

Zoom In Zoom Out Reset image size

More important than achieving a nominal S/N requirement is the ability to achieve the desired uncertainties on higher level survey data products, and in this regard APOGEE has generally succeeded. A good estimate of RV precision can be obtained from analysis of repeat measurements of single stars. In DR12, the RV residuals for stars with multiple visits and S/N $\geqslant$ 20 peak at ∼70 m s⁻¹ (Nidever et al. 2015). Uncertainties in the cross-correlation method tend to be smaller for spectra with lots of sharp absorption lines. Therefore, RVs are more accurate for cooler and/or metal-rich giants than for metal-poor and/or hotter/high surface gravity stars, whose spectra are characterized by fewer and/or broader absorption lines, respectively. In any case, the precision achieved surpasses the stated requirement (Section 2.4). The absolute RV zeropoint of APOGEE is estimated to be ∼350 ± 30 m s⁻¹, from comparison with data from other sources for stars in common (Nidever et al. 2015).

Even though no technical requirements were specified for the stellar parameters ${T}_{\mathrm{eff}}$ and $\mathrm{log}g$, they drive the accuracy of derived chemical abundances and spectroscopic parallaxes. Figure 20 shows the final DR12 values (Holtzman et al. 2015) for APOGEE stars in a "spectroscopist's HR diagram." The final stellar parameters adopted and shown in this plot result from calibration of the raw parameters delivered by ASPCAP against photometry and abundance data from the literature, and asteroseismological surface gravities from APOKASC (Pinsonneault et al. 2014), as described by Holtzman et al. (2015). There is very good agreement between stellar parameters from APOGEE spectra and independent theoretical predictions, shown by the isochrones. Of course, some scatter of the data about the isochrones is expected given the dispersion of the sample in age and [α/Fe] at any metallicity. Moreover, some unresolved issues remain. The most important example happens for red clump stars, which should be distinctly separated from the red giant branch. Stellar evolution models, which are confirmed independently by accurate gravities from asteroseismology for a subsample of APOGEE stars in the Kepler field (e.g., see Figure 18 of Pinsonneault et al. 2014), predict a narrow range of surface gravities for red clump stars. Such a feature is not seen in APOGEE data, which instead show a correlation between ${T}_{\mathrm{eff}}$ and $\mathrm{log}g$. Another unresolved issue is the slightly too warm APOGEE temperatures at the cool end. These are areas where refinements in ASPCAP will bring improvements in the future. For a further discussion of these issues, see Holtzman et al. (2015) and García Pérez et al. (2016).

Zoom In Zoom Out Reset image size

The precision and accuracy of the derived elemental abundances are reviewed in detail by Holtzman et al. (2015). To achieve the goals in probing GCE specified in Section 1.3, APOGEE had a specification for a precision in measured chemical abundances of 0.1 dex, and an accuracy of 0.2 dex. The precision of the actually derived APOGEE elemental abundances was established by measuring the dispersion of the measurements at fixed ${T}_{\mathrm{eff}}$ in clusters stars, where the abundances are expected to be constant. Precisions were found to be better than 0.1 dex for all top and medium priority elements, thus formally meeting the original requirement. For many elements, precisions are better than 0.05 dex and for only one low priority element (vanadium) is the precision worse than 0.1 dex.

Comparison of APOGEE [Fe/H] with mean literature abundances of cluster stars based on HR optical spectroscopy shows that the 0.2 dex accuracy requirement is met everywhere, except perhaps for [Fe/H] lower than $\sim -1.5$. In view of the presence of clear systematic differences between "raw" derived APOGEE values via ASPCAP and literature values (where APOGEE values are higher by up to 0.2 dex), a calibration to convert APOGEE [Fe/H] into the literature scale was derived (Holtzman et al. 2015). For some of the other elements, abundance ratios—i.e., [X/Fe], computed by adopting the non-calibrated [Fe/H]—compare well with the literature, in spite of some scatter, which is likely due to the heterogeneity and uncertainties in the literature data. For a few elements, such as Ca and Ti, trends with ${T}_{\mathrm{eff}}$ are found. In the absence of a statistically significant homogeneous sample of optical data from the literature, it is difficult at this stage to draw firm conclusions regarding the APOGEE accuracy for some of the elemental abundances. Nevertheless, while the original requirements are met for most of the elements, more work is required, on one hand, to refine the elemental abundances delivered by ASPCAP and, on the other, to more firmly establish the systematic differences that will inevitably be present between the APOGEE abundance scales and those of other large observational programs (typically performed at optical wavelengths).

The achieved performance and capabilities of APOGEE are perhaps most eloquently demonstrated by examples of how the collected data may be used in a wide variety of science applications. In this section, we briefly discuss a few examples of published and in preparation research based on APOGEE data, with an eye toward illustrating the breadth of potential APOGEE science, both within the confines of Galactic astronomy and beyond. The order of presentation of these science examples roughly tracks the degree of processing of the APOGEE data, as described in Section 6—i.e., from direct analyses of the spectral character of sources, to analyses of derived stellar velocities, to explorations of bulk metallicity and then more detailed chemical abundance patter data, to even higher level results made possible by inclusion of derived stellar ages, and concluding with analyses incorporating the former information for the analysis of star clusters and the interstellar medium.

7.4.1. Time Series Spectral Data

The time series data that are available from multi-epoch observations of the majority of APOGEE targets (e.g., Figure 18) lend themselves to a number of useful applications. These include, e.g., the identification of binary stars that partly motivated this survey strategy (Section 2.8). For instance, Figure 21 shows an example of a double-lined spectroscopic binary made evident through the multi-epoch spectroscopy). As another example of the interesting phenomena that a detailed exploration of APOGEE's time series spectral data might uncover, Figure 22 shows the strong variation in a Brackett series line for one of the numerous Be stars discovered by APOGEE (Chojnowski et al. 2015) within the sample of hot stars observed for telluric absorption correction.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

7.4.2. Time Series Precision Velocity Data

The high precision of the APOGEE radial velocity measurements, made possible by the tightly controlled environmental conditions (gravity vector, temperature, vacuum) of the APOGEE spectrograph combined with the availability of multi-epoch visits, makes the survey sensitive to not only stellar, but also substellar mass companions. The opportunity for large-scale statistical explorations of brown dwarfs and even hot Jupiters is illustrated by Figure 23, which shows one of a number of exoplanet candidates identified within the APOGEE database. In this case, the star, HD 114762, has a known $11{M}_{\mathrm{Jup}}$ companion in an 83.9 day period (Latham et al. 1989). Before recognizing that this was a previously known exoplanet, the APOGEE team, through a project to automatically fit multi-epoch APOGEE velocity data, determined this source to have a $14{M}_{\mathrm{Jup}}$ companion in a 77.9 day period. This result, based on only 11 APOGEE spectra, is not far from that already determined by Latham et al. The rms of the best fit to this relatively bright star system yields a residual of only 33 m s⁻¹; further refinements in the velocity measurement pipeline may bring similar rms precisions to fainter stars and perhaps enable further improvements in velocity precision for brighter stars.

Zoom In Zoom Out Reset image size

7.4.3. Radial Velocities Across the Galaxy

While not all stars in the APOGEE database will have more than several radial velocity measurements (see Figure 18), almost all of them will have mean radial velocities measured with an accuracy greater than is typically found in previous large spectroscopic surveys of Milky Way stars. Combined with its systematic probe of all stellar populations, APOGEE's radial velocity database makes possible new large-scale explorations of Galactic dynamics (e.g., see Figure 24). Because APOGEE was designed to include extensive and deep probes of the low-latitude Milky Way, including the highly extinguished regions of the disk and bulge, previously poorly studied regions of the Galaxy have now been opened up to extensive kinematical canvassing. As a result, a number of Galactic dynamical phenomena have already been discovered or reexamined with unprecedented statistical significance, including the discovery of a kinematically extreme population in the bulge (Nidever et al. 2012), likely associated with the Galactic bar (Aumer & Schönrich 2015; Molloy et al. 2015); a comprehensive analysis of the dynamics of the bulge (Ness et al. 2016b); a new derivation of the circular velocity of the Sun afforded by the more global APOGEE view of disk kinematics (Figure 24; Bovy et al. 2012, 2015); and finally, an assessment of the power spectrum of non-axisymmetric velocity perturbations of the disk, of which most can be attributed to the action of the central bar (Bovy et al. 2015).

Zoom In Zoom Out Reset image size

7.4.4. Metallicities Across the Galaxy

A primary goal of the APOGEE survey is to probe the large-scale distribution of stellar chemical compositions across the Galaxy. Typical results are demonstrated by the [M/H] metallicity gradients shown in Figure 25, where the distribution of APOGEE targets projected onto the XY plane is displayed. Accurate measurements of gradients across a range of distances, both Galactocentric and away from the Galactic plane, provide strong constraints on models for the formation of the thick and thin Galactic disks (for further details on the application of APOGEE to these problems, see Bovy et al. 2014; Anders et al. 2014; Hayden et al. 2014, 2015; Nidever et al. 2014).

Zoom In Zoom Out Reset image size

7.4.5. Multi-element Abundance Variations

The power of APOGEE to deliver similar information for multiple chemical elements and with great statistical power is illustrated by Figure 26, a representation of the metallicity gradient in the Milky Way's disk using APOGEE spectra directly. We stack the pseudo-normalized spectra of RC stars in the DR12 APOGEE-RC catalog (Bovy et al. 2014) as a function of Galactocentric radius in bins of width $0.1\,\mathrm{kpc}$ and normalize by the stacked spectrum at the solar circle ($R=8\,\mathrm{kpc}$). Because the RC occupies a small region in $(\mathrm{log}g$, ${T}_{\mathrm{eff}})$, the ${T}_{\mathrm{eff}}$ and $\mathrm{log}g$ values for RC stars are all very similar, so that spectrum-to-spectrum differences in absorption line strengths are mostly due to star-to-star elemental abundance variations. This figure clearly demonstrates that metal absorption increases toward the center of the disk and decreases toward the outskirts for every element with transitions in the APOGEE spectral region (except for vanadium, which has such a weak line that absorption disappears almost completely around the solar circle). A quantitative measurement of the overall radial [M/H] gradient of the RC sample is $-0.09\pm 0.01\,\mathrm{dex}\,{\mathrm{kpc}}^{-1}$ (Bovy et al. 2014).⁷⁷

Zoom In Zoom Out Reset image size

7.4.6. Analysis of Abundance Patterns

A central rationale for APOGEE's sensitivity to the abundances of multiple chemical elements is to access the information contained in variations of chemical abundance patterns, particularly for patterns containing elements synthesized on different evolutionary timescales within populations. The power of APOGEE to probe GCE in this way is demonstrated by Figure 27, which shows the spatial distribution of [Mg/Fe] with [Fe/H] in different spatial zones of the Galactic disk. Figure 27 is a more elemental-specific example of the more general [α/Fe] distributions explored in Anders et al. (2014), Nidever et al. (2014), and Hayden et al. (2015), and shows that APOGEE not only confirms the existence of a bimodal distribution of [α/Fe] at fixed [Fe/H], but that this bimodality exists over a large extent of the disk, including for previously uncharted Galactocentric distances for $| Z| \lt 0.5$ kpc (see Hayden et al. 2015 for an even more extensive coverage of the disk). These results pose important constraints on disk formation models, in particular calling for the presence of two stellar populations with different chemical enrichment histories in the outer Galactic disk. That APOGEE data can provide such data across a large multidimensional space of elements probing different nucleosynthetic pathways is further demonstrated by the results shown in Figure 30 below.

Zoom In Zoom Out Reset image size

7.4.7. Stellar Ages

Collaboration between the APOGEE and the Kepler and CoRoT asteroseismology teams has greatly increased the power of APOGEE data to shed light on Galaxy formation scenarios. The combination of APOGEE chemical compositions with asteroseismological data makes possible the determination of accurate masses, and thus ages, for a large sample of APOGEE field giants (Pinsonneault et al. 2014). Demonstrations of the power of analyzing combined detailed chemical compositions and ages from these APOGEE subsamples are starting to emerge (e.g., Epstein et al. 2014; Chiappini et al. 2015; Martig et al. 2015); one example is demonstrated in Figure 28, which shows the unexpected discovery (Martig et al. 2015) of a small sample of stars with high [α/Fe] and relatively low ages (∼3–5 Gyr). The mere existence of intermediate-age stars with such high [α/Fe] constitutes a serious challenge to existing chemical evolution models and is, as yet, not fully explained.

Zoom In Zoom Out Reset image size

7.4.8. Open Clusters

The APOGEE database has made the targeting of star clusters a priority, not only because they provide reliable abundance calibration references, but because they are extremely useful stellar population components of great interest in their own right. Open clusters, in particular, provide a powerful, independent means by which to explore the chemistry, kinematics, and extinction of the Galactic disk because cluster distances can be precisely gauged via isochrone fitting, made all the more accurate when the variable of metallicity can be removed using spectroscopic metallicities. An important output of such isochrone fits are cluster ages, which provide critical and confident timestamps for benchmarking evolutionary studies across the disk. Fortunately, apart from a set of key open clusters that drove specific plugplate pointings, the vast majority of the large APOGEE sample of over 150 open clusters was a natural product of its extensive canvassing of the Galactic plane, and only required attention to the allocation of fibers on already overlapping plugplates from the APOGEE grid to known or suspected members of these clusters. As part of the Open Cluster Chemical Analysis and Mapping (OCCAM) survey (Frinchaboy et al. 2013), these data have already been used to study disk metallicity gradients and abundance patterns as a function of both space and time; an example of the opportunities opened up with this kind of analysis is demonstrated in Figure 29.

Zoom In Zoom Out Reset image size

7.4.9. Globular Clusters

Unlike the open clusters, where spectra of large samples of cluster members are easily generated as part of APOGEE's systematic coverage of the Galactic disk, considerable investment of effort, planning, and survey time was needed to ensure ample attention to globular clusters. Many globular clusters are distant, which makes even their giant stars relatively faint and means that the stars are densely packed on the sky relative to the fiber-to-fiber collision radius (see Section 2). These two facts typically necessitated many visits to specialized fields (see Figure 10), with multiple plugplate designs to accumulate long integrations on fainter targets and to overcome the collision radius limitations to generate reasonable sample sizes. Nevertheless, the globular cluster visits were deemed high priority both because the member stars are extremely useful calibration targets and because the chemistry of globular clusters is itself interesting and relevant to understanding the evolution of Milky Way stellar populations.

The study of globular cluster formation is undergoing a revolution since the discovery of the commonality of multiple populations in these systems (e.g., Piotto et al. 2007; Milone et al. 2008; Gratton et al. 2012), which has been established predominantly on the basis of observations collected from the Southern Hemisphere. APOGEE is already making important contributions to the study of globular cluster evolution through the determination of accurate multi-element abundances of relatively large samples of northern cluster members. This is illustrated in Figure 30, which, for the majority of the 10 clusters shown (eight that were never previously studied in this way), gives evidence for strong internal Al abundance variations—in some cases anti-correlated with Mg—a feature commonly interpreted as the signature of the presence of multiple stellar populations. Interestingly, APOGEE has shown that some globular clusters (M3, M53) display apparently discrete Al−Mg distributions (i.e., Al-rich and Al-poor groups), while others (e.g., M13, M5) may show a more continuous distribution in the Al abundances (Mészáros et al. 2015). APOGEE data have also been used to test previous claims of the presence of multiple populations in the well-known, unusually large, old but super-metal-rich open cluster, NGC 6791 (Cunha et al. 2015).

Zoom In Zoom Out Reset image size

A distinct advantage afforded by these APOGEE studies of globular clusters is that the inferred chemistry of the member stars may be directly compared to those of the numerous field stars surveyed in precisely the same way. An example of the power of such comparisons, with interesting implications for our understanding of both globular cluster and galaxy bulge formation, is the discovery of remnants of globular cluster destruction in the Galactic bulge—long predicted by theoretical arguments (Tremaine et al. 1975)—through the identification of chemical signatures typical of globular cluster stars in an APOGEE sample of field bulge stars (Schiavon et al. 2017).

7.4.10. Mapping Interstellar Extinction

The combination of APOGEE spectroscopy with available photometric databases, such as those provided by 2MASS or the Spitzer IRAC surveys, increases the scope of approachable astrophysical problems. For example, comparison of spectroscopically derived atmospheric properties with measured broadband colors can give key, star-by-star measurements of the distribution of interstellar reddening, and therefore a powerful means by which to explore the three-dimensional distribution of dust. For example, Figure 31 shows the distribution of derived K_s-band extinction across the Galactic plane, with the extinctions derived by comparing observed colors to the intrinsic colors expected from theoretical isochrones matched to the ASPCAP-derived values of T_eff, $\mathrm{log}g$, and [M/H] (see Schultheis et al. 2014), and distances derived from comparing apparent and associated isochrone absolute magnitudes. Apart from showing global trends in dust properties, studies such as this can help calibrate or highlight inadequacies in extinction maps derived by other means.

Zoom In Zoom Out Reset image size

7.4.11. Mapping Diffuse Interstellar Bands

The accuracy with which the APOGEE spectral library matches the intrinsic spectra of typical giant stars makes possible further explorations of interesting astrophysics that is unrelated to the stars themselves. For instance, disagreements between spectral synthesis and observations may reveal the presence of absorbers/emitters along the line of sight. A compelling example of this is shown in Figure 32, where the difference between an observed spectrum and its best-matching synthetic template reveals the presence of a relatively strong diffuse interstellar band (DIB; Herbig 1995). Using the fact that this same DIB feature is seen in the vast majority of APOGEE spectra along disk and bulge sight lines, Zasowski et al. (2015) for the first time thoroughly mapped the spatial distribution and velocity field of a DIB carrier over many kiloparsecs of the Galactic disk (Figure 33). Comparison of these properties to those of other components of the ISM (dust, atomic and molecular gas) will yield critical evidence toward the identification of the carriers of these mysterious spectroscopic signatures—a longstanding puzzle in Galactic astrophysics.

Zoom In Zoom Out Reset image size

There are two main sets of APOGEE data products. The first consists of wavelength-calibrated, normalized, 1D rest-frame spectra, and the radial velocities resulting from analysis of visit spectra. These data products are the outputs of the data reduction and radial velocity pipelines, described in Sections 6.1–6.4. The second set of data products contains the stellar parameters (${T}_{\mathrm{eff}}$, $\mathrm{log}g$, [M/H]) and the abundances of up to 15 elements, namely, C, N, O, Na, Mg, Al, Si, S, K, Ca, Ti, V, Fe, Mn, and Ni. The latter set results from application of ASPCAP (Section 6.5) to the spectra. To ensure that the high-level data can be reproduced by the community, ingredients that are crucial for the stellar parameter and elemental abundance determinations—such as the adopted model atmospheres (Mészáros et al. 2012), line list (Shetrone et al. 2015), and grids of synthetic spectra (Zamora et al. 2015) —are are also publicly available.

Zoom In Zoom Out Reset image size

The data products from the three-year APOGEE survey have been included in several SDSS-III data releases (DRs). In Data Release 10 (DR10, announced 2013 July), calibrated APOGEE spectra, radial velocities, stellar parameters, and a limited set of elemental abundances were made available for 57,454 stars (Ahn et al. 2014). DR12 (made available in 2015 January) includes the data described in Section 8.1 for all ∼136,000 primary science targets in the SDSS-III/APOGEE sample (Alam et al. 2015; Holtzman et al. 2015). In addition, radial velocities for the ∼17,000 hot stars⁷⁸ used as telluric standards, as well as spectra and derived parameters for approximately 900 bright stars observed using the APOGEE feed from the NMSU 1 m telescope (Section 5.3.3), are also included in DR12. In 2016, a further data release, DR13 (SDSS Collaboration et al. 2016), was made that included new reductions of these same APOGEE spectra but with improved telluric subtraction, upgraded pipelines, and grids of synthetic spectra (including rotational broadening for dwarf star spectra and extension to cooler stars, to ${T}_{\mathrm{eff}}=3500$ K) and updated abundance determinations, including the addition of several atomic species (C I, P, Ti II, Co, Cu, Ge, and Rb) previously not analyzed or newly found to have lines in the APOGEE sprectra. Further improvements in the pipelines and data will be found in DR14, to be released in the summer of 2017.

APOGEE data can be accessed in a number of different ways through the http://www.sdss.org Web site. The SDSS-III Catalog Archive Server (CAS) contains the high-level APOGEE data products, namely, radial velocities, radial velocity dispersion, ${T}_{\mathrm{eff}}$, $\mathrm{log}g$, [M/H], and elemental abundances, as well as information relevant to target selection, including, e.g., coordinates, 2MASS, WISE, Spitzer/IRAC, and Washington+DDO51 photometry, proper motions, where available, and assumed interstellar extinction. These data can be accessed via low-level SQL scripts, which allow users to select subsamples according to suitable criteria in the CasJobs link. Lower level data products, including the visit and combined spectra of sources searchable via standard, basic, or advanced online forms, can be accessed with the SAS. The SAS also provides access to the directory tree containing the full data set (for expert users), as well as a summary FITS file containing all of the catalog information listed above. The SkyServer offers another way of interfacing with the data, through a "quick look" tool that makes possible non-interactive visualization of images and spectra of sources that are searchable by position and ID, via an online form. Finally, both the SAS and SkyServer link to a Web tool that allows for a quick assessment of data and model quality, through interactive viewing of the (final) observed spectra overplotted with the best-fitting synthetic spectrum.

The primary source of in-depth information for users of APOGEE data are the technical papers listed in Table 1. Additional, often less detailed, information can be obtained in the Web pages associated with each data release. Users of APOGEE are strongly urged to peruse those Web pages or technical papers for a complete understanding of the data quality, limitations, and caveats that are specific to each data release (Ahn et al. 2014; Holtzman et al. 2015). Users of high-level data products should pay particular attention to the documentation pertaining to ASPCAP (Mészáros et al. 2012, 2013; Smith et al. 2013; Shetrone et al. 2015; Zamora et al. 2015; García Pérez et al. 2016), where the limitations and known systematics in the APOGEE stellar parameters and elemental abundances are described. On the other hand, users of APOGEE spectra will be interested in information about possible systematics induced by instrument features and reduction procedures (Nidever et al. 2015; J. C. Wilson et al. 2017, in preparation). A good grasp of the possible biases introduced by APOGEE's target selection procedure and field placement strategy (Zasowski et al. 2013) is encouraged for those exploring global properties, trends, and correlations for stars across the Galaxy, and for those comparing these empirical results to chemical and/or chemodynamical models of stellar populations.

As part of APOGEE-2 (Section 9.2), we continue to develop our data reduction and analysis capability. In addition, it is expected that our understanding of the atomic and molecular physics related to the relevant line transitions will continue to evolve. Therefore, to maintain consistency across the integrated APOGEE-1 and APOGEE-2 efforts, future APOGEE data releases will be expected to include updated analyses of the APOGEE-1 data.

In 2012, the APOGEE team proposed to continue with an APOGEE-2 program in Sloan Digital Sky Survey IV. Beyond an immediate continuation of operations on the Sloan Telescope, a key feature of this proposal is the extension of the APOGEE-2 program into the Southern Hemisphere, in a collaboration with the Observatories of the Carnegie Institution for Science (OCIS) to use the du Pont 2.5 m telescope for this purpose. The du Pont is very much a forerunner of the Sloan Telescope, having been designed with a wide FOV (Babcock 1977) and with a heritage in fiber plugplate spectroscopy (Shectman et al. 1996) that influenced the subsequent design of the SDSS system. APOGEE-2 was approved as an SDSS-IV program in 2012, and APOGEE-2 began observing at APO in 2014 September. The current instrument at APO is the same as that described in Section 3 with one important modification in that the blue detector with "superpersistence" has been replaced with a cosmetically cleaner array free of this problem.

While initially it was unclear whether APOGEE-2 would operate with two spectrographs working on the Sloan and du Pont Telescopes in parallel or whether the existing spectrograph would be moved to the du Pont after an initial two-year campaign on the Sloan Telescope, by 2014 November complete funding was secured to build a second APOGEE spectrograph to make parallel operations possible. With the active participation of OCIS, Chilean universities, and other SDSS-IV collaborators, the southern infrastructure development, including the creation of a scaled-down version of the Sloan fiber plugplate system (cartridges, plugging station, mapping system, cartridge transport and loading system), is underway, as is the construction of an infrastructure "mock-up and training facility" in a special laboratory at the University of La Serena, Chile. The second APOGEE spectrograph, which is planned as a near duplicate of the original with only minor alterations (e.g., to include seismic mitigation and to incorporate various other "lessons learned"), is currently planned to begin operations on the du Pont Telescope in mid-2015.

When it is necessary to distinguish it from the APOGEE-2 survey, the original "APOGEE" survey will in the future be referred to as "APOGEE-1." Henceforth, the term "APOGEE" will generally be intended to refer to the combined APOGEE-1 and APOGEE-2 surveys, which together will provide a fully comprehensive, all-sky view of the Milky Way.