The Infrared Telescope Facility (IRTF) Spectral Library: Cool Stars

Introduction

Spectral libraries play an important role in attempts to understand and classify individual stellar sources as well as to decompose the integrated spectrum of an aggregate system, such as a galaxy, into its various stellar populations. For example, the most widely used stellar classification process, as originally developed by Morgan et al. (1943), consists of comparing the spectrum of a star against a set of reference stellar spectra (for a review see Garrison, 1994). Infrared spectral libraries are particularly useful for studying the physics of cool stars (e.g. Joyce et al., 1998; Gautschy-Loidl et al., 2004), classifying and studying stars in nearby embedded young clusters (e.g., Greene & Meyer, 1995; Peterson et al., 2008) and optically obscured regions of the Galaxy (e.g., Figer et al., 1995; Frogel et al., 2001; Kurtev et al., 2007, for evolved, globular, and young clusters respectively), and studying the unresolved stellar populations of optically obscured extra-galactic regions using evolutionary population synthesis (EPS) (e.g., Lançon et al., 2007; Riffel et al., 2008). EPS techniques attempt to simulate observed galaxy spectra by combining individual stellar spectra from a library and thereby derive the chemical and evolutionary properties of the unresolved stellar populations (e.g., Fioc & Rocca-Volmerange, 1997; Leitherer et al., 1999; Bruzual & Charlot, 2003; Maraston, 2005; Bruzual, 2007; Ramos Almeida et al., 2009).

The niche of near-infrared (NIR $\sim$ 1-5 $\micron$) spectral classification is clear. While stars earlier than roughly M0 ($\sim$3800 K) are brighter at optical wavelengths, unobscured stars later than about M0 are brighter in the NIR and are thus better characterized at these wavelengths. Furthermore it makes sense to use infrared diagnostics only when optical extinction compromises optical diagnostics. The optimum infrared wavelengths for observation depend on the amount of extinction. For some objects, such as young stellar objects or evolved stars, the presence of circumstellar dust can result in significant excess continuum emission longward of 2 $\micron$. For this reason the $J$ and $H$ bands are perhaps best to characterize embedded young stars since they avoid the veiling due to warm dust in the $K$ band, while at the same time taking advantage of the reduced extinction relative to the optical (e.g. Meyer et al., 1998). On the other hand, heavily obscured objects without veiling are better characterized in the $K$ band, or even the $L^{\prime}$ band in extreme cases. Consequently, the ideal infrared spectral library should contain spectra covering a wide range of wavelengths to satisfy a variety of possible applications.

With the maturing of NIR spectrographs and detector arrays, it has become possible to generate increasingly sophisticated NIR libraries of stellar spectra. Ivanov et al. (2004) presented a compilation of NIR spectral libraries available at that time. In Table 1 we revise and update this list. (The list does not include spectral libraries covering mostly L and T dwarfs. For our purposes a ‘library’ is assumed to contain more than ten objects.) All of these libraries have shortcomings since none of them contains a large sample of stars, with a range of metallicities, covering all spectral types and luminosity classes, with spectra spanning a large wavelength range. In light of this, we have undertaken a project to construct an improved spectral library using the facility NIR spectrograph, SpeX, at the 3.0 m NASA Infrared Telescope Facility (IRTF) on Mauna Kea, Hawaii. The result of this work is the IRTF Spectral Library, which we are presenting in a series of papers. In the first paper of this series, Cushing et al. (2005) presented the spectra of M, L, and T dwarfs. The current paper presents 210 spectra of F, G, K, and M stars with luminosity classes between I and V (with mostly near-solar metallicities), and includes some asymptotic giant branch (AGB) stars, carbon stars, and S stars. The spectra of all of these stars, including the 13 L dwarfs and 2 T dwarfs from Cushing et al. (2005), and the gas giant planets (spectra summed along the central meridian), are available in digital form on the IRTF websitehttp://irtfweb.ifa.hawaii.edu/~spex/IRTF_Spectral_Library. Additional papers on hot stars are currently in preparation.

There are several important features of the IRTF Spectral Library. The wide wavelength range of $\sim$0.8-5 $\micron$ (with a larger subset at 0.8-2.4 $\micron$) is covered in only two cross-dispersed instrument settings. For each setting, several spectral orders are simultaneously recorded during a single exposure. In addition, most of the spectral orders in each setting have significant wavelength overlap with the adjacent spectral orders. These instrumental aspects minimize potential calibration problems posed by stitching together multiple non-overlapping wavelength ranges observed at different times (e.g., sequentially), a situation typically encountered with observations obtained with non-cross-dispersed (single-order) spectrographs. The signal-to-noise (S/N) is better than $\sim$100 across most of this range (except for the regions of poor atmospheric transmission and for $\lambda>4~{}\micron$) and the resolving powers of $R\equiv\lambda/\Delta\lambda\approx$ 2000 at 0.8-2.4 $\micron$, and $R\approx$ 2500 at 2.4-5 $\micron$, enable the accurate measurement of spectral type and luminosity class using established equivalent width and line ratio criteria (see §3.1). In contrast to some other NIR spectral libraries, the continuum shape is preserved during data reduction (for details see §2.3) which is particularly useful for characterizing cool stars with strong molecular absorption bands that have been observed at low-resolution $R\sim$100. Preserving the continuum shape also allows for absolute flux calibration by scaling the spectra to published Two Micron All Sky Survey photometry (2MASS, Skrutskie et al., 2006) and for the computation of synthetic colors (e.g. $Y-J$, $J-H$, $H-K$, and $K-L^{\prime}$).

OBSERVATIONS AND DATA REDUCTION

As described by Morgan & Keenan (1973), the MK spectral classification system “is a phenomenology of spectral lines, blends, and bands, based on a general progression of color index (abscissa) and luminosity (ordinate). It is defined by an array of standard stars located on a two-dimensional spectral type versus luminosity-class diagram. These standard reference points do not depend on specific line intensities or ratios of intensities; they have come to be defined by the totality of lines, blends, and bands in the ordinary photographic region” (emphasis added). In the MK system, the classification gives the spectral sub-type and luminosity class (e.g. K0 III); this is the observational analogue to the projection on the luminosity-temperature plane (H-R diagram) for stars of a particular composition. Abundance adds a third dimension to the two-dimensional MK diagram and is represented by additional symbols determined by the relative intensities of lines or bands that reveal compositional differences from the Sun. For example, as a means of distinguishing a solar metallicity Population I giant K0 III star from one with a lower metal/hydrogen abundance, the classification of the latter becomes K0 III CN-1, or K0 III CN-2 (e.g. Morgan & Keenan, 1973). With better quality spectra increased precision in spectral classification is possible. For example, giants can often be subdivided into luminosity subclasses IIIa, IIIab, and IIIb. A fundamental characteristic of the MK system is that a finite array of discrete cells (spectral types) represents a continuum i.e. spectra of stars of a given spectral subtype (e.g. K5 V) are not all identical. The precision attainable with MK classification has been estimated to be $\pm$0.6 spectral subtypes for B and A dwarfs by Jaschek & Jaschek (1973), and $\pm$0.65 spectral subtypes for G and K dwarfs by Gliese (1971). This precision depends upon observational dispersion (heterogeneous group of observers and instruments) and cosmic dispersion (e.g. chemical composition effects, and rotation effects).

We attempted to construct a sample of stars with undisputed spectral types, traceable to the original developers of the MK classification system. The original MK standard stars (Johnson & Morgan, 1953; Morgan & Keenan, 1973) are generally too bright for us to observe. To that end, for the majority of the sample, we chose stars with classifications given by Morgan & Abt (1973), Morgan & Keenan (1973), Morgan et al. (1978), Keenan & McNeil (1989), Keenan & Newsom (2000)http://www.astronomy.ohio-state.edu/MKCool, which in several cases were supplemented by stars taken from compilations of MK standard stars by Garcia (1989) and Jaschek (1978). Whenever references gave conflicting classifications, we chose the most recent revision. For F stars, we supplemented the lists generated from the aforementioned references with stars whose spectral types are given by Gray & Garrison (1989), Gray et al. (2001), and Abt & Morrell (1995). Additionally, for M stars, we included objects with classifications given by Kirkpatrick et al. (1991), Henry et al. (1994), and Kirkpatrick et al. (1997), again deferring to the latest revised classifications whenever conflicting or multiple spectral types were found in the various sources. In a few instances stars with less certain classifications were observed in order to fill gaps in our coverage of spectral types due to observing limitations.

Despite their known variability in spectral type, and relative rarity, we included asymptotic giant branch (AGB) stars in our sample because their high luminosity makes them important in EPS studies of galaxies. Population synthesis and star counts in clusters indicate that AGB stars contribute more than 50% of the $K$-band light of stellar populations at 0.1 to 1 Gyr after an instantaneous burst of star formation (Lançon et al., 1999; Lançon, 1999). The AGB phase is also important in providing feedback in the chemical evolution of galaxies. AGB stars are intermediate mass stars ($\sim$0.8-8$M_{\odot}$) which ascend the asymptotic giant branch in the HR diagram when helium and hydrogen ignite in shells surrounding their cores (this phase lasts about $2\times 10^{5}$ yr). Shell burning in young AGB stars is stable but becomes increasingly unstable as the stars become more luminous which leads to thermal pulsations. These stars are known as thermally pulsating AGB (TPAGB) stars. TPAGB stars are recognizable by a variety of observational criteria by which they are variously named: characteristic spectra (late-M, S, and C stars), pulsating variability (Mira variables, long-period variables), mass loss and maser emission (OH/IR stars). In our sample TPAGB stars are identified by their variability types (L: irregular, SR: semiregular, and M: Mira) given in the General Catalog of Variable Stars (GCVS, Kholopov et al., 1998)http://www.sai.msu.su/groups/cluster/gcvs/gcvs. About 40 TPAGB stars are included in our sample.

Mass loss eventually removes the hydrogen-rich stellar envelope, effectively terminating the TPAGB phase. The central star subsequently evolves to higher temperatures while the circumstellar envelope expands and cools, exposing the star. Ionizing wind and radiation from the star quickly form a planetary nebula (PN). The transition from TPAGB to PN is known as the post-AGB (PAGB) or protoplanetary nebula phase and lasts a few thousand years. Although PAGB stars were not targeted in our sample, several supergiants that were observed have some of the characteristics of PAGB stars (see §4.3). (For a comprehensive review of AGB stars see Habing & Olofsson, 2003).

In order to obtain high S/N out into the thermal infrared ($\sim$2.3-5 µm), we selected relatively bright stars. Consequently, they tend to be local and therefore of mostly solar composition. Figure 1a shows the distribution of metallicities for stars in our sample with spectroscopic measurements of [Fe/H] (Cayrel de Strobel et al., 1997). The distribution is typical for stars in the solar neighborhood (Nordström et al., 2004).

The object name, spectral classification and associated reference, GCVS variable type, $V$, $B-V$, and 2MASS ($J$, $H$, and $K_{S}$) magnitudes for each star in the sample are given in Table The Infrared Telescope Facility (IRTF) Spectral Library: Cool Stars. All objects in the sample have declinations $-30>\delta>+70$ degrees, a range set by the latitude of IRTF, and an airmass $<$ 2 for good telluric correction. The stars have $K$-band magnitudes of $\sim$11 $>K>~{}$0; the faint limit was set by the desire to obtain high S/N spectra in less than about 30 minutes of integration time, and the bright limit corresponds to detector saturation in the minimum exposure time of 0.1 s (although several brighter targets were observed using ad hoc methods). The total number of stars in the sample is 210. Table 3 gives the composition of the sample by spectral type and luminosity class. Due to practical limitations of observing time there is no multi-epoch coverage of variable stars or large numbers of stars with non-solar metallicity. However, we anticipate future observing campaigns with SpeX at IRTF will add to the sample.

2. Observations

The observations were carried out over a period of eight years using SpeX at the IRTF. A detailed description of SpeX is given by Rayner et al. (2003). Briefly, SpeX is a 0.8$-$5.4 $\mu$m, medium-resolution, cross-dispersed spectrograph equipped with a 1024 $\times$ 1024 Aladdin 3 InSb array. The entire 0.8 to 5.4 $\mu$m wavelength range can be covered with two cross-dispersed modes, the short-wavelength cross-dispersed mode (SXD), and the long-wavelength cross-dispersed mode (LXD). The SXD mode provides simultaneous coverage of the 0.8$-$2.42 $\mu$m wavelength range, except for a 0.06 $\mu$m gap between the $H$ and $K$ bands, while the LXD1.9, LXD2.1, and LXD2.3 modes cover the 1.9$-$4.2, 2.20$-$5.0 and 2.38$-$5.4 $\mu$m wavelength ranges, respectively. For nearly all stars, the 0$\farcs$3 (2 pixel) slit was used for both the SXD and LXD modes, providing resolving powers of 2000 and 2500, respectively. Measurements of arc lines obtained with the internal calibration unit indicate that the FWHM is 2 pixels at all wavelengths for the 0$\farcs$3 slit. (The resolving power R varies by $\sim$20% across a spectral order since it depends on the changing grating diffraction angle.) The length of the slit for these modes is 15$\farcs$0 and the spatial scale is 0$\farcs$15/pixel. The spectrograph also includes a high-throughput low-resolution $R\sim$ 200 prism mode and a single-order 60$\arcsec$ long-slit $R\sim$ 2000 mode. An autonomous infrared slit viewer employing a 512 $\times$ 512 Aladdin 2 InSb array is used for object acquisition, guiding, and imaging photometry. The slit viewer covers a 60$\arcsec$ $\times$ 60$\arcsec$ field-of-view at a spatial scale of 0$\farcs$12/pixel. An internal K-mirror image rotator enables the field to be rotated on the slit. Calibration observations are obtained using the internal calibration unit consisting of flat field and arc lamps, integrating sphere, and illumination optics which reproduce the beam from the telescope. A log of the observations including the object name, spectral type, UT date of observation, spectroscopic mode, resolving power, exposure time, associated telluric standard star, and sky conditions, is presented in Table 4.

To facilitate subtraction of the additive components of the total signal (electronic bias level, dark current, sky and background emission) during the reduction process, the observations were obtained in a series of exposures in which the target was nodded along the slit between two positions separated by 7$\farcs$5, and a sequence of nodded pairs was taken to build up S/N. A minimum of three pairs was taken (six spectra) to allow noisy pixels (due mainly to cosmic ray hits) to be rejected by a sigma clipping algorithm. Guiding was done on spill-over from the science target in the slit using the infrared slit-viewing camera. In the SXD mode, where atmospheric dispersion is significant compared to the the slit width of 0$\farcs$3 (see Figure 2), the image rotator was set to the parallactic angle prior to each observation. As discussed in §2.3, observing at the parallactic angle minimized spectral slope variations. This is not as important in the LXD mode where atmospheric dispersion is an order a magnitude smaller.

An A0 V star was observed before or after each science object to correct for absorption due to the Earth’s atmosphere (see Figure 3) and to flux calibrate the science object spectra. The airmass difference between the object and “telluric standard” was almost always less than 0.1 and usually less than 0.05. However, in a few cases where there was a paucity of nearby A0 V stars, the airmass difference was as large as 0.15. Standard stars were also chosen to be located within 10 degrees of the science object whenever possible, to minimize the effects of any differential flexure in the instrument between observations of the object and standard. This limit on the angular distance provides a good compromise between the requirements to match airmass, minimize flexure, and find suitably bright standard stars. On those few occasions when it was necessary to observe a telluric standard more than 10 degrees away from the object due to a lack of A0 V stars in certain parts of the sky, we found that telluric CO2 (predominantly at 2.01 $\mu$m) features were sometimes not adequately removed by the standard star despite a good airmass match and good correction of telluric H2O. (See, for example, Figure 54, where the F7 III and F8 III stars were corrected with telluric standard stars at separations and airmass differences of 14 degrees and 0.09, and 21 degrees and 0.05, respectively.) We attribute this to the possibility that telluric H2O and CO2 are not well mixed and to patchy CO2 distribution. Finally, a set of internal flat field exposures and argon arc lamp exposures were taken after each object/standard pair for flat fielding and wavelength calibration purposes.

3. Data Reduction

We reduced the data using Spextool (Cushing et al., 2004), the facility IDL-based data reduction package for SpeX. The initial image processing consisted of correcting each science frame for non-linearity, subtracting the pairs of images taken at the two different slit positions, and dividing the pair-subtracted images by a normalized flat field. In each frame, the spectra in the individual orders were then optimally extracted (e.g., Horne, 1986) and wavelength calibrated. (All wavelengths are given in vacuum.) The extracted spectra in each order from the set of frames for a given object were then combined using the median. This resulted in a single spectrum in each order for a given object.

The spectra in the individual orders were then corrected for telluric absorption and flux calibrated using the extracted A0 V spectra and the technique described in Vacca et al. (2003). In addition to correcting for the absorption due to the atmosphere, this process also removes the signature of the instrumental throughput and restores the intrinsic (i.e., above the atmosphere) spectrum of each science object in each spectral order. Briefly, this technique scales a theoretical model spectrum of Vega to the observed visual magnitude of the observed standard star, convolves it to the observed resolution, and adjusts the H I line strengths to match the observed strengths of the standard star. The ratio of the adjusted model spectrum to the observed spectrum of the A0V star gives the telluric correction spectrum (which also includes correction for the instrument throughput) in each order. The science object spectrum is then divided by the telluric spectrum. The resulting flux calibration is accurate to about 10 percent. The sharp and deep telluric absorption features are marginally sampled with a 2 pixel-wide slit so when the object and telluric correction spectra are ratioed, residuals remain at the wavelengths of these features due to a small amount of instrumental flexure between the object and standard star positions. In order to minimize these systematic errors, the telluric correction spectrum is first shifted relative to the object spectrum until the noise in these regions is minimized. Typically these shifts are $\sim$0.1$-$0.2 pixels for telescope movements of 10 degrees.

In principle, the flux density levels of the telluric-corrected spectra in two adjacent orders should match exactly in the wavelength region where they overlap; in practice we find offsets of usually less than one percent, although occasionally as large as three percent. The level mismatch was removed by scaling one spectrum to the level of the other. The scale factor was determined from a section of the overlap region where both spectra were judged to have sufficient S/N to allow an accurate determination. The telluric-corrected and scaled spectra in the individual orders were then merged together to form a single, continuous spectrum for the science object. Regions of strong telluric absorption were then removed. The precise wavelength intervals removed depended on the transparency of the atmosphere at the time of observation but always included the $\sim$2.5-2.8 $\micron$ and $\sim$4.2-4.6 $\micron$ regions. For each object, the SXD and LXD spectra were combined in a manner similar to that used to combine the individual orders. A scale factor was determined from the overlapping wavelength region and then used to adjust the SXD and LXD spectra to a common level.

The next step in the reduction process was to absolutely flux calibrate the spectra using the 2MASS photometry listed in Table The Infrared Telescope Facility (IRTF) Spectral Library: Cool Stars. For each spectrum, we computed correction factors based on the $JHK_{S}$ photometry given by,

The spectrum of each object is then shifted to zero radial velocity. To do this, we first selected two stars, HD 219623 (F8 V) and HD 201092 (K7 V), as representative of F-G, and K-M stars, respectively, and measured their apparent radial velocities using the observed wavelengths of strong, isolated atomic lines. For HD 219623 we measured the positions of the Pa $\epsilon$ (0.9548590 $\mu$m), Pa $\gamma$ (1.0052128 $\mu$m), Pa $\beta$ (1.282159 $\mu$m), Mg I (1.4881683 $\mu$m), Mg I (1.7113304 $\mu$m), and Br $\gamma$ (2.166120 $\mu$m) lines; for HD 201092 we used the Si I (1.0588042 $\mu$m), Mg I (1.1831408 $\mu$m), Mg I (1.4881683 $\mu$m), and Mg I (1.7113304 $\mu$m) lines. The observed radial velocities for these two stars were determined by averaging the radial velocity measurements obtained from these lines. The standard error on the mean radial velocity was $\sim$3 km s-1 for both stars. We then shifted the spectra of these two stars in wavelength to correspond to zero radial velocity. To determine the radial velocities of the remaining stars in the library, we cross-correlated the 1.05$-$1.10 $\mu$m spectra of the F-G stars (which contain isolated lines of Si, C, Mg, and Fe) against that of HD 219623 and the 2.285$-$2.33 $\mu$m spectra of the K and M stars (which contain the $\Delta\nu$=$+$2 CO overtone bands) against that of HD 201092. The peak of each cross-correlation function was fitted with a second order polynomial to determine the radial velocity. Based on our implementation of the cross-correlation technique described by Tonry & Davis (1979) as well as a comparison of the radial velocities derived from different wavelength regions in each spectrum, we estimate the uncertainty in our radial velocity values to be generally less than 30 km s-1. The spectrum of each object was then shifted to zero radial velocity using the radial velocity derived from the cross correlation. It should be noted that, in order to preserve the accuracy of our data, we did not re-sample or re-bin the final spectra to a common wavelength scale after shifting them, and therefore each spectrum has a unique wavelength array. In most cases, the velocity shifts are small: the average shift was found to be 4.5 $\pm$ 37 km/s with a maximum of about 131 km/s. Both values are smaller than our velocity resolution (150 km/s per resolution element).

Although the stars in our sample are generally bright and nearby, some stars show evidence of interstellar reddening. For those applications requiring true spectral energy distributions (e.g. EPS studies) reddening needs to be corrected. Consequently, as a final step in the data reduction we corrected the spectra for reddening. We determined the $E(B-V)\equiv(B-V)-(B-V)_{0}$ color excess from the observed ($B-V$) color and and an intrinsic $(B-V)_{0}$ color appropriate for its spectral type. The observed colors were taken from the Mermilliod (2006) catalog and are on the Johnson photometric system. For the few stars that were not included in this catalog, we adopted the (B-V) color given by Kharchenko (2001), Leggett (1992), and Beauchamp et al. (1994). We adopted the calibration of intrinsic colors as a function of spectral type by Fitzgerald (1970). For the M dwarfs, we found that the Fitzgerald (1970) values gave unreasonably large color excess values for stars that are very nearby (less than 20 pc). Furthermore, while other calibrations of intrinsic colors (e.g., Schmidt-Kaler, 1982) agree well with that of Fitzgerald (1970) at earlier spectral types, they differ markedly for M stars, with the later calibrations becoming progressively redder. For these reasons, we adopted the intrinsic colors given by Leggett (1992) for the M dwarfs, which generally yield very small (or even negative) values for $E(B-V)$ for these stars. To derive intrinsic colors for stars with intermediate spectral types and luminosity classes not tabulated in the combined FitzGerald/Leggett calibration we performed a two-dimensional surface interpolation over the intrinsic color values as a function of spectral subtype and luminosity class. The distribution of E(B-V) values for the stars in our sample is shown in Figure 1b. As expected for the bright (and generally nearby) stars in our sample, the distribution is peaked near 0. Fitting a Gaussian to the values of $E(B-V)<0$ indicates that the uncertainty on the color excesses is about $\sigma$=0.036 mag. All color excesses determined to be less than zero were set to 0 in the dereddening process. Furthermore, based on our findings for the standard deviation of the distribution, we chose not to correct any spectra with $E(B-V)<0.108$. The 66 dereddened stars along with their ($B-V$), ($B-V)_{0}$, $E(B-V)$, and $A_{V}$ values are given in Table 5.

where $f_{\lambda}(\lambda)$ is the absolutely flux calibrated spectrum and $A_{\lambda}$ is the extinction law as a function of wavelength. We adopted the NIR law given by (Fitzpatrick & Massa, 2007) with $R_{V}$=3.0. For this law,

We note that these dereddened spectra should be used with care because the dereddening process assumes an intrinsic stellar color and a mean Galactic extinction law that may not be accurate or appropriate in all cases. For example, we find $E(B-V)=0.06$ (or $A_{V}$=0.18) for the M6 dwarf Gl 406 which resides at a distance of only 2.4 pc (van Altena et al., 1995). No significant extinction is expected at this distance. In addition, the M giant stars and PAGB and TPAGB stars may have local extinction due to dust formation in their cool outer atmospheres which is difficult to separate for any interstellar extinction that may be present. Variable stars are also problematic due to changes in the intrinsic color. Therefore unless otherwise noted, we will continue to use the uncorrected spectra in the remainder of the analysis, including the figures. Nevertheless, the dereddened spectra are available on the IRTF website (see footnote 2).

where the symbols have the same meaning as in Equation 1. The results for the 53 stars with relatively good (5%) 2MASS photometry (from Table The Infrared Telescope Facility (IRTF) Spectral Library: Cool Stars) are plotted in Figure 5. The average differences are $<\Delta_{J-H}>=0.00\pm 0.04$ (RMS), $<\Delta_{H-K_{S}}>=0.02\pm 0.04$, and $<\Delta_{J-K_{S}}>=0.01\pm 0.05$. Although the colors derived from the 2MASS photometry are not as precise as those measured from the spectra via synthetic photometry (better than 1%), the photometric residuals on the sample of 53 cool stars indicate that our measurements of spectral slope are accurate to within a few percent for F, G, K, and M spectral types. Similar (but larger) effects have been observed by Goto et al. (2003) while using adaptive optics with medium resolution spectroscopy. Their simulations demonstrate that spectral slope variations result from changes in the fraction of light from the object transmitted by a finite width slit as a function of wavelength. Therefore, temporal changes in seeing, guiding, and differential atmospheric refraction, are probably the cause of the small variations we observe. Although SpeX does not have an atmospheric dispersion corrector, we minimized the effects of differential refraction (and therefore the wavelength dependent light loss through the slit) by moving the internal image rotator to observe at the parallactic angle (see Figure 6 in Rayner et al., 2004) and by combining multiple spectra together.

As a further test of the accuracy of the spectral slopes, we have computed synthetic $Z-Y$, $Y-J$, $J-H$, $H-K$, and $K-L^{\prime}$ colors of eight A0 V stars observed with the same instrumental setup. The synthetic color for any two bandpasses $X$ and $Y$ is given by,

The mean colors of the 8 A0 V stars are $<<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mi>Z</mi><mo>−</mo><mi>Y</mi></mrow><annotation encoding="application/x-tex">Z-Y</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.7667em;vertical-align:-0.0833em;"></span><span class="mord mathnormal" style="margin-right:0.0715em;">Z</span><span class="mspace" style="margin-right:0.2222em;"></span><span class="mbin">−</span><span class="mspace" style="margin-right:0.2222em;"></span></span><span class="base"><span class="strut" style="height:0.6833em;"></span><span class="mord mathnormal" style="margin-right:0.2222em;">Y</span></span></span></span></span>>$$=-0.015\pm 0.007$ (RMS), $<<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mi>Y</mi><mo>−</mo><mi>J</mi></mrow><annotation encoding="application/x-tex">Y-J</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.7667em;vertical-align:-0.0833em;"></span><span class="mord mathnormal" style="margin-right:0.2222em;">Y</span><span class="mspace" style="margin-right:0.2222em;"></span><span class="mbin">−</span><span class="mspace" style="margin-right:0.2222em;"></span></span><span class="base"><span class="strut" style="height:0.6833em;"></span><span class="mord mathnormal" style="margin-right:0.0962em;">J</span></span></span></span></span>>$$=0.007\pm 0.006$ (RMS), $<<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mi>J</mi><mo>−</mo><mi>H</mi></mrow><annotation encoding="application/x-tex">J-H</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.7667em;vertical-align:-0.0833em;"></span><span class="mord mathnormal" style="margin-right:0.0962em;">J</span><span class="mspace" style="margin-right:0.2222em;"></span><span class="mbin">−</span><span class="mspace" style="margin-right:0.2222em;"></span></span><span class="base"><span class="strut" style="height:0.6833em;"></span><span class="mord mathnormal" style="margin-right:0.0813em;">H</span></span></span></span></span>>$$=0.019\pm 0.005$ (RMS), $<<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mi>H</mi><mo>−</mo><mi>K</mi></mrow><annotation encoding="application/x-tex">H-K</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.7667em;vertical-align:-0.0833em;"></span><span class="mord mathnormal" style="margin-right:0.0813em;">H</span><span class="mspace" style="margin-right:0.2222em;"></span><span class="mbin">−</span><span class="mspace" style="margin-right:0.2222em;"></span></span><span class="base"><span class="strut" style="height:0.6833em;"></span><span class="mord mathnormal" style="margin-right:0.0715em;">K</span></span></span></span></span>>$$=-0.005\pm 0.004$, $<<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mi>K</mi><mo>−</mo><msup><mi>L</mi><mo mathvariant="normal" lspace="0em" rspace="0em">′</mo></msup></mrow><annotation encoding="application/x-tex">K-L^{\prime}</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.7667em;vertical-align:-0.0833em;"></span><span class="mord mathnormal" style="margin-right:0.0715em;">K</span><span class="mspace" style="margin-right:0.2222em;"></span><span class="mbin">−</span><span class="mspace" style="margin-right:0.2222em;"></span></span><span class="base"><span class="strut" style="height:0.8019em;"></span><span class="mord"><span class="mord mathnormal">L</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.8019em;"><span style="top:-3.113em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mtight">′</span></span></span></span></span></span></span></span></span></span></span></span></span>>$$=-0.001\pm 0.010$. Given that we expect the mean colors to be zero, the mean colors of the A0V stars represent the precision in our ability to measure spectral slope. Taken together, the 2MASS and A0V star photometry indicates that we can measure spectral slope to within a few percent. Finally, slope variations introduced due to uncertainties in the spectral types of the A0 V standard stars are small since an uncertainty of a 0.5 subtype at A0 is equivalent to ${\delta}_{J-H}=0.003$, ${\delta}_{H-K}=0.005$, and ${\delta}_{J-L^{\prime}}=0.010$, (Cox, 2000).

DATA AND ANALYSIS

Digital versions of the spectra are available at the IRTF website, which contains a full description of the data products. The data are available in text or Spextool FITS format and files can be downloaded individually or bundled together in a tar file. The files contain wavelength, flux, and error. The errors include the photon Poisson noise and read noise (Vacca et al., 2004) for both the object and associated telluric standard, which are then propagated through each step of the reduction process. The file headers contain more information (including object name, epoch, spectral type, observing modes, 2MASS $JHK$ magnitudes, measured radial velocity, flux, and wavelength units). As an example, the flux and S/N spectra of HD 63302 (K1 1a-Iab) and HD 10696 (G3 V), are shown in Figure 6. We estimate that the actual S/N in our fully reduced stellar spectra is limited to less than 1000 by systematic errors in the flat field measurement even though the formal S/N can be greater than 1000. Also, systematic errors in telluric correction (e.g. if the airmasses of an object and standard star are not ideally matched), systematic errors in the slope, as well as any errors in the spectral type of the A0 V telluric standard are not accounted for in the formal S/N estimate (see Figure 6).

Representative spectra, covering 0.8 to 5.0 $\mu$m, of dwarfs, giants, and supergiants in our sample are shown in Figures 7 to 9. Given the large wavelength range and change in flux wavelength and spectral type, it is difficult to display the spectra at a scale that allows close examination of all of the interesting spectral features in one plot. For display purposes we therefore use $\lambda f_{\lambda}$ (normalized at the given wavelength) as a function of $\log\lambda$. The plots show the general trend of spectral features with MK spectral type. Feature identifications and changes with spectral type are described in more detail in §3.2, §3.3, and §3.4.

It should be noted that MK spectral classification is based on comparing an $optical$ spectrum to a set of stars (anchor points) that define certain spectral types. Therefore it is not necessarily the case that the trends in NIR spectral features will follow the optical types. Also, the exact numbers for the effective temperature, surface gravity, and composition (i.e. metallicity) are model dependent (i.e. not directly observable) and will change as models improve. Although there are no formal MK classification criteria for the NIR, much work has been conducted on NIR spectral classification. The pioneering study on cool stars was done by Kleinmann & Hall (1986) who identified a number of temperature and luminosity sensitive atomic (Na I, Ca I, Br $\gamma$) and molecular (CO, H2O) features in the $K$ band. Subsequent studies developed a variety of spectral type versus equivalent width (EW) indices for cool stars. Notable examples amongst these are the studies by Origlia et al. (1993), Dallier et al. (1996), Meyer et al. (1998), Ramírez et al. (1997), Förster Schreiber (2000), Gorlova et al. (2003), Davies et al. (2007), Ivanov et al. (2004), and Mármol-Queraltó et al. (2008). We do not propose to add to these and other NIR classification schemes but present the IRTF Spectral Library as a resource for further investigation. Examples of EWs of several prominent features in data as a function of spectral type and luminosity are, however, presented in §4.1.

As an example of the mismatches in the trends of the NIR spectral features with MK type, Figure 10 shows a spectral sequence of K giant stars from 0.8 to 2.45 $\mu$m based on their MK types. From the trend in overall spectral shape and depth of the first overtone vibration-rotation bands of CO in the $K$-band ($\sim$2.29-2.5 $\mu$m), two of the stars appear out of sequence (they are bluer and with shallower CO absorption than expected) and should appear earlier in the sequence by about one spectral subtype. When the stars are corrected for reddening (see Figure 11 and Table 5) the sequence of spectral shapes behave as expected but the CO band depths still appear to be out of sequence. A possible explanation is that these stars are slightly metal poor (see for example Mármol-Queraltó et al., 2008). However, this explanation is inconsistent with the assigned MK spectral types, one star being of approximately solar metallicity (HD 137759, K2 III), and the other star being slightly metal rich (HD 114960, K3.5 IIIb CN0.5 CH0.5) although neither star has a formal [Fe/H] measurement. Another possibility is the uncertainty in assigning the MK spectral type as discussed in §2.1. Given these uncertainties, spectral classification using the IRTF Spectral Library is probably best done by comparing a given stellar spectrum of an unknown type to an ensemble of spectra for a sequence of MK types, rather than trying to find the closest individual spectral match. The effect of reddening should also be considered.

Plots of the dwarf sequence F3, F5, F9, G2, G8, K1, K7, M3, and M7, for the $I$, $Y$, $J$, $H$, $K$, and $L$ bands, respectively, are shown in Figures 12 to 17. Luminosity effects at spectral types F, G, K, and M, across the 0.8-5 $\mu$m range, are shown in Figures 18 to 21. Plots showing luminosity effects at spectral types F5 and G6 for the $I$, $Y$, $J$, $H$, $K$, and $L$ bands, respectively, are shown in Figures 22 to 27; likewise luminosity effects at spectral types K5 and M5 are shown in Figures 28 to 33. A 0.8-5 $\mu$m sequence of M, S, and C giants is plotted in Figure 34. All these spectra are discussed in more detail in the subsequent sections.

2. Atomic Line Identifications

The S/N of the spectra is high enough that almost all of the absorption features seen in the spectra are real and not noise. However given that the resolving power of the spectra is $R\sim$2000, the carriers of only the strongest and most isolated lines can be unambiguously identified. We therefore use the line identifications presented in the high resolution ($R\sim$100000) spectral atlases of the solar photosphere (Wallace et al., 1993; Wallace & Livingston, 2003) and Arcturus (K1.5 III, Hinkle et al., 1995, 2000). We selected only those absorption lines in the atlases with depths less than 0.8 (where the continuum has been normalized to unity) because weaker lines are only marginally detectable in our spectra.

Tables 6 and 7 list all atomic metal lines with depths less than 0.8 in the solar and Arcturus atlases, respectively. The wavelengths of the lines are in vacuum and were taken primarily from the atlases themselves. The revised solar atlas (Wallace & Livingston, 2003) does not present the wavelengths of the lines so we obtained them from an earlier version of the atlas (Livingston & Wallace, 1991). A few of the lines in the revised atlas either lack identifications in the original atlas or have since been identified as being carried by two lines. We determined the wavelengths of these lines using the National Institute of Standards and Technology Atomic Spectra Databasehttp://physics.nist.gov/PhysRefData/ASD/index.html, the Atomic Line Listhttp://www.pa.uky.edu/~peter/atomic/, and the Fe I line list compiled by Nave et al. (1994).

3. Molecular Band Identifications

Molecular absorption bands are prominent in the spectra of late-type stars. Identifications for many of these features can be found in e.g., Spinrad & Wing (1969), Brett (1990), Brett (1989), Lançon & Wood (2000). Table The Infrared Telescope Facility (IRTF) Spectral Library: Cool Stars lists the vacuum wavelengths and identifications for each of the band heads identified in the spectra. Below we describe some of the absorption bands in detail.

Absorption bands of TiO that arise from the $\gamma$ ($A$ ${}^{3}\Phi$ $-$ $X$ ${}^{3}\Delta$), $\epsilon$ ($E$ ${}^{3}\Pi$ $-$ $X$ ${}^{3}\Delta$), $\delta$ ($b$ ${}^{1}\Pi$ $-$ $a$ ${}^{1}\Delta$), and $\phi$ ($b$ ${}^{1}\Pi$ $-$ $d$ ${}^{1}\Sigma$) systems are conspicuous in the spectra of late-type stars over the 0.8 to 1.5 $\mu$m wavelength range (see Figures 7, 8, 9, 12, 28, 29, 30). The $\gamma$ and $\epsilon$ systems involve triplet electronic states and thus exhibit triple-headed bands while the $\delta$ and $\phi$ systems involve singlet electronic states and thus exhibit only single-headed bands.

The $\Delta\nu=-2$ bands of the $\gamma$ system (0.82$-$0.86 $\mu$m) are often identified in both low and high resolution spectra of late-type stars (e.g., Kirkpatrick et al., 1991; Tinney & Reid, 1998; Reiners et al., 2007). In particular, four TiO band heads ($R_{2}$, $Q_{2}$, $R_{1}$, $Q_{1}$) are identified between 0.849 and 0.860 $\mu$m (see Figures 12 and 36). However as can be seen in both our data and the high resolution spectra of late-type M and early-type L dwarfs, (Tinney & Reid, 1998; Reiners et al., 2007), two additional band heads of similar depths exist at 0.8508 and 0.8582 $\mu$m. Together, the six band heads appear to form two sets of triplet band heads. The 1$-$1 and 2$-$2 bands of the $\epsilon$ system exhibit band heads at these wavelengths and therefore it is likely that these features arise from the $\epsilon$ system alone or are a combination of the $\gamma$ and $\epsilon$ systems. We also note that three of the $\Delta\nu=-2$ band head classifications listed in Gatterer et al. (1957) were marked as uncertain so we have confirmed these using the theoretical line list of Schwenke (1998).

It has been known for some time that absorption bands of TiO are present in the spectra of M giant stars near 0.93 $\mu$m (Spinrad & Newburn, 1965). However, to our knowledge, these bands were not detected in the spectra of M dwarfs until the work of Cushing & Vacca (2006). These band heads arise from the $\Delta\nu=-1$ band of the $\epsilon$ system. Although telluric and intrinsic H2O absorption make identifications in this region difficult we nevertheless are able to identify the triplet band heads of the 0$-$1 (0.9211, 0.9221, 0.9233 $\mu$m), 1$-$2 (0.9279, 0.9289, 0.9300 $\mu$m), and 2$-$3 (0.9345, 0.9356, 0.9368 $\mu$m) transitions (see Figures 12 and 28). Finally, we note that while the wavelengths of the 2$-$3 and 3$-$4 band heads measured by Linton & Broida (1977) agree with our observations, they do not match the positions of the band heads predicted by Schwenke (1998).

3.2 VO

Absorption bands that arise from the $B$ ${}^{4}\Pi<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mo>−</mo></mrow><annotation encoding="application/x-tex">-</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.6667em;vertical-align:-0.0833em;"></span><span class="mord">−</span></span></span></span></span>X$ ${}^{4}\Sigma^{+}$ This system is identified as the “C” system in Gatterer et al. (1957), Solf (1978), and Tinney & Reid (1998). and $A$ ${}^{4}\Pi$ $-<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mi>X</mi></mrow><annotation encoding="application/x-tex">X</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.6833em;"></span><span class="mord mathnormal" style="margin-right:0.0785em;">X</span></span></span></span></span>{}^{4}\Sigma^{-}$ systems of VO are present in the spectra of M stars over the 0.8 to 1.3 $\mu$m wavelength range. We identify only the 0$-$0 ($\sim$1.06 $\mu$m) and 0$-$1 ($\sim$ 1.18 $\mu$m) bands of the $A$ ${}^{4}\Pi$ $-<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mi>X</mi></mrow><annotation encoding="application/x-tex">X</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.6833em;"></span><span class="mord mathnormal" style="margin-right:0.0785em;">X</span></span></span></span></span>{}^{4}\Sigma^{-}$ system (see Figures 8, 13, 29, 39). Although there are certainly absorption features arising from the 0$-$1 band of $B$ ${}^{4}\Pi<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mo>−</mo></mrow><annotation encoding="application/x-tex">-</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.6667em;vertical-align:-0.0833em;"></span><span class="mord">−</span></span></span></span></span>X$ ${}^{4}\Sigma^{+}$ system centered near $\sim$0.85 $\mu$m in the spectra of late-type stars (Keenan & Schroeder, 1952; Tinney & Reid, 1998), they are simply too weak to be detected in our data.

3.3 ZrO

A series of four ZrO band heads arising from the 0$-$0 band of the system $b$ ${}^{3}\Pi<span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mo>−</mo></mrow><annotation encoding="application/x-tex">-</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.6667em;vertical-align:-0.0833em;"></span><span class="mord">−</span></span></span></span></span>a$ ${}^{3}\Delta$ of ZrO (Phillips et al., 1979) are found in the spectra of M giant stars and carbon stars (see Figures 9, 21, 28, 34). Three additional band heads are also present at 0.932060, 0.934417, and 0.935860 $\mu$m but their corresponding transitions are unknown. Only three of the band heads ($R_{1(c)}$, $R_{1(d)}$, and 0.932060 $\mu$m) can be conclusively identified in the spectra given the complexity of the spectra at these wavelengths, but we include the other band heads for completeness. Additional ZrO band heads (Hammer & Davis, 1981; Joyce et al., 1998) in the 1.0$-$1.2 $\mu$m wavelength range are not seen in our spectra.