AN ENHANCED SPECTROSCOPIC CENSUS OF THE ORION NEBULA CLUSTER

Located at less than one-half  kiloparsec from the Sun and at moderate galactic latitude (b ≈ −20), the greater Orion region is the nearest example of recent and ongoing massive star formation; see Bally (2008) for an overview. A subclustered OB association extends over many tens of parsecs and contains stars up to a few tens of megayears old. To the southeast, the Orion A and B molecular cloud complex harbors stellar nurseries each only a fraction of a parsec to a few parsecs in size, and less than one megayear old. The Orion composite is a revered touchstone for our developing understanding of star and planet formation processes on this range of spatial and temporal scales. The identification and detailed study of the young stellar population across the Orion complex is our best opportunity for probing in three dimensions a resolved star formation history.

Part of the so-called "integral shaped filament" of molecular gas (Bally et al. 1987), the northern part of the Orion A cloud has up to 100 mag of total visual extinction through its densest regions. The Orion Nebula, also known as M42 or NGC 1976, betrays a rich young stellar cluster, the Orion Nebula Cluster (ONC), the center of which contains the massive Trapezium stars.³ The current three-dimensional model of the Orion Nebula indicates a thin shell of ionized gas (Wen & O'Dell 1995) with the main ionizing source region θ¹ Ori C. Another ∼20 stars of type A0 and so or earlier are within the H ii region as is a sizable population (many thousand members) of later spectral type T Tauri stars. These lower mass stars are still pre-main-sequence and can be used to trace the star formation history as well as measure the initial mass function (IMF), which, in the case of young, unevolved, and co-located stars, is equivalent to the present day mass function.

In contrast to the sparsely distributed and older OB association, the ONC is relatively compact and more recently formed. Its projection on molecular cloud material significantly reduces contamination by background stars. However, the high and spatially variable extinction (e.g., Scandariato et al. 2011) renders fewer than 50% of the ∼3500 stars within this region with K < 14 mag also optically bright, I < 17.5 mag. Nevertheless, the bulk of the optically visible ONC members are extincted by 2 mag or less, with derived extinction values for fainter stars ranging up to 8 mag. The deepest optical surveys as well as most near-infrared surveys measure more heavily extincted members, but are more likely to suffer contamination at the faint end—both from intrinsically bright but highly reddened background giants seen through the cloud, and from intrinsically faint foreground and unreddened but late type field dwarfs.

Cluster membership can be determined kinematically in the ONC. Using astrometric techniques, Jones & Walker (1988) identified 891 probable members from 997 measured proper motions (measured among 1052 optically visible stars). Radial velocities (e.g., Tobin et al. 2009) have also been used to establish cluster membership. High precision three-dimensional space motion data from these and other ongoing investigations can be combined with precise cluster distance estimates (Jeffries 2007; Kraus et al. 2007; Hirota et al. 2007; Sandstrom et al. 2007; Menten et al. 2007) to study the cluster structure and dynamics and test theories for formation and evolution of dense clusters in molecular clouds. Additional observation and analysis work in this area is needed.

The ONC contains a broadly populated stellar mass function extending from a single massive (late-O type) star all the way down to low-mass stars (having late M types) and into the sub-stellar mass regime of brown dwarfs (L and likely T type), as well as planetary mass objects remaining to be spectrally typed and identified as such. With a mean stellar age for the optically visible stars of less than 2 Myr (Herbig & Terndrup 1986; Hillenbrand 1997; da Rio et al. 2010; Reggiani et al. 2011), the ONC is not only one of the largest but also one of the densest (∼10⁴ stars pc⁻³ in the inner cluster; McCaughrean & Stauffer 1994; Hillenbrand & Hartmann 1998) nearby star-forming regions. The ONC population was used to demonstrate definitively that it is possible to form both high and low mass stars as well as brown dwarfs in the same <0.1 pc region. From the lower mass stars, the estimated current star formation rate is ∼10⁻⁴ M_☉ yr⁻¹ (Hillenbrand 1997, hereafter H97).

The H97 study of the ONC employed optical spectroscopy and photometry to individually de-redden and locate over 900 stars on the H-R diagram, enabling estimation of their masses and ages via comparison to pre-main sequence evolutionary tracks. Work by Luhman et al. (2000), Slesnick et al. (2004), Riddick et al. (2007b), and Weights et al. (2009) increased the spectroscopically studied sample, employing optical and near-infrared spectral classification techniques. da Rio et al. (2010, 2012) estimated spectral types for M-type stars from narrow-band photometry, as discussed in more detail below. Additional studies such as those of Hillenbrand & Carpenter (2000), Muench et al. (2002), Lucas et al. (2005), and Andersen et al. (2011) used photometric techniques rather than spectroscopic ones, and arrived at IMF estimates that are useful in a statistical sense but do not determine masses and ages for individual stars. The general consensus from the above studies is that the ONC IMF rises from the highest masses to sub-solar masses, realizes a flattening in the 0.5–0.6 M_☉ regime, and peaks in the 0.2–0.3 M_☉ regime before turning over throughout the sub-stellar (brown dwarf) mass range. There is some discrepancy as to the exact form of the IMF below the peak and turnover in this cluster (and in many other young clusters). Increased spectroscopic samples as well as improved understanding of contraction models at young ages will be required for progress on the IMF in the low mass stellar and sub-stellar domain.

Recent reviews of the ONC region are those by Muench et al. (2008) on the stellar population and historical distance measurements, and by O'Dell et al. (2008) on the local ONC interstellar medium including circumstellar structures. The ONC stellar population has been the sample of choice for investigations of various stellar and circumstellar phenomena, in addition to the IMF studies described above, in part because it is the nearest and youngest example of an entire stellar/sub-stellar mass spectrum. Recent studies of circumstellar dust and gas include those of Megeath et al. (2012), Mann & Williams (2009), and Ricci et al. (2008); of disk accretion include Manara et al. (2012) and Sicilia-Aguilar et al. (2005); and of variability-disk connections include Morales-Calderón et al. (2012). Recent studies on stellar properties include topics such as rotation and disk-rotation connections (Rhode et al. 2001; Carpenter et al. 2001; Herbst et al. 2002; Herbst & Mundt 2005; Rebull et al. 2006; Cieza & Baliber 2007), lithium depletion (Palla et al. 2007), coronal activity/flaring (Feigelson et al. 2002; Getman et al. 2008; Prisinzano et al. 2008), stellar abundances (Cunha & Smith 2005), magnetic fields (Petit et al. 2008), and multiplicity (e.g., Simon et al. 1999; Reipurth et al. 2007; Kohler et al. 2006).

There is still great interest in improving our knowledge of stellar and circumstellar properties within the ONC. As described above, the H97 paper and its accompanying database have been used in subsequent years by many other investigators studying, e.g., the stellar/sub-stellar IMF; the age distribution and star formation history; stellar/sub-stellar angular momentum evolution, circumstellar disks, and disk evolution; accretion, magnetic activity, lithium, and other properties of young stars in the ONC. Recent examples of major surveys of the ONC with forefront facilities that reach unprecedented depth across the electromagnetic spectrum include those from X-ray (with Chandra; PI: E. Feigelson) to optical (with Hubble; PI: M. Robberto) to mid-infrared (with Spitzer; Megeath et al. 2012) and finally to millimeter (with the SMA and CARMA interferometers). Synoptic photometric monitoring studies by various groups at optical and infrared wavelengths (e.g., Morales-Calderón et al. 2011) continue.

Nearly all of these previous analyses and new surveys are limited in their analysis phases, however, due to lack of completeness of the ONC spectroscopic survey. Critically important is that due to the wide range in line-of-sight extinction values to individual stars, spectra are necessary for de-reddening and, thus, derivation of extinction estimates, and for subsequent placement on the H-R diagram from which stellar quantities such as stellar luminosity, stellar radius, and (further guided by theory) stellar age and mass, angular momentum, etc.

The optical color–magnitude diagram for the ONC suggests that the H97 optical photometric database is sensitive down to the hydrogen burning limit for stars less obscured than 2 mag of visual extinction and younger than 1 Myr (e.g., da Rio et al. 2010; H97). Infrared surveys extend well into the brown dwarf regime with the stellar–sub-stellar boundary identified at spectral type M6.5 given the age of the ONC. While several thousand young objects are known, only about 63% of the ∼1500 stars with I < 17 mag had spectral types following the H97 study (and only about 5% before). Considering both optical and infrared-only sources in the region, less than 25% of the total number of known sources have had their spectral types determined. While infrared spectroscopy will be required in order to obtain a complete spectroscopic census of the ONC stellar population, optical spectral types can be obtained for several hundred more objects than have been published thus far.

The present work aims to enhance the H97 optical spectroscopic census. By completing our optical study of the stellar population, it will be possible for other authors to revisit interests in many of the above goals.

Spectral type is a key characteristic of any star, and spectral typing is a typical first step in estimating the stellar effective temperature. This can lead, in combination with other data or assumptions, to derivation of other stellar quantities. The optical wavelength range is important for spectral typing because it contains a rich set of molecular and atomic absorption features that are temperature dependent. M type stars, which dominate the mass function, are characterized by metallic oxide TiO and VO bands. These molecular features diminish at hotter temperatures with mid-K and earlier stars classified using atomic line strengths and ratios involving, e.g., Ca i and Fe i. Reliable surface gravity signatures include the shapes of some of the molecular bands as well as the strength of the Na i doublet (λλ8183, 8194) for spectral types later than M2. For earlier type stars, most reliable surface gravity indicators lie blueward of our spectral coverage.

Relevant to the ONC, spectral types can be determined largely independent of reddening and nebular emission when these effects are modest, though in extreme cases they do bias or even hinder the analysis. For late type spectra, the ratios of the depths of metal oxide wavebands are not significantly affected by low reddening and the possibility of reddening can be taken into account when assigning types. Nebular emission, specifically from low Balmer (e.g., Hα) and high Paschen lines as well as the Paschen continuum can also affect the spectral typing process or even prevent it in extreme cases, especially for faint objects or if background subtraction (a combination of terrestrial sky and ONC nebular backgrounds) is ineffective. For late type stars the TiO and VO bands are generally strong enough such that the effects of minor nebular emission can be overcome. For earlier spectral type stars, however, strong nebular contamination is more problematic for spectral typing.

A final effect is that of non-nebular continuum and emission lines that are associated in young stars with disk-to-star accretion processes. Spectral typing can be influenced by non-photospheric contributions to emergent spectrum and in extreme cases the spectrum is accretion-dominated.

To classify the spectra we follow the same procedures as are described in Section 3.3 of H97. Specifically, the classification was performed using spectral indices to measure feature depths, in the spirit of O'Connell (1973), and followed up by visual examination of the data. For each feature, 30 Å wide bandpasses were used in creating spectral strength indices from defined on-feature and off-feature wavelength ranges; average fluxes over these bands were calculated. Riddick et al. (2007a) give a thorough description of the relevant molecular and atomic features and their behavior with temperature and surface gravity, as well as the challenges and pitfalls in classifying young stars associated with star formation regions. These effects include accounting for the observational influence of low surface gravity, accretion, large and variable reddening, and nebular contamination in the classification process.

The vast majority of the stars in the optically visible ONC sample are young late K and M type stars. For their classification metal oxide bands were used: four of TiO (λ6760, λ7100, λ7800, λ8465) and two of VO (λ7445, λ7865). These prominent bands are mainly temperature sensitive, making them ideal for the spectral classification. Figure 3 shows a sequence of K and M-type spectra from the WIYN/HYDRA and P200/Norris datasets. The TiO absorption bands used to initially classify the stars are indicated. Stars earlier than mid-K do not have strong TiO features and atomic line ratios were used, such as Ca i λ6162 versus Na i λ5892 for F2–K7 stars, Ca i λ6162 versus Mg i λ5175, upper Paschen lines, and other Ca i and O i lines for A0–F0 stars. Spectral typing for stars earlier than late-K is best done blueward of our spectral range.

Zoom In Zoom Out Reset image size

Classifying our red spectra quantitatively first involved two dimensional plots of spectral feature index pairs such as: TiO λ6760 versus TiO λ7100, TiO λ7100 versus TiO λ7800, and VO λ7445 versus VO λ7865. These diagrams were used to initially bin the late type ONC stars by finding the spectral type along the standard star sequence that is closest to the position of the target star. Second, cubic equations were used to fit the standard sequence for each TiO band. The standard stars included those at medium spectral resolution from Allen & Strom (1995) and Kirkpatrick et al. (1991) extending up to M6 (near the brown dwarf limit) and Kirkpatrick et al. (1999) beyond this.

For the majority of stars, different feature strength ratios predicted consistent spectral types. However, using line or band indices gives only an estimate for the spectral type. Hence, after assigning preliminary spectral types by the quantitative analysis techniques described above, each spectrum was examined visually against a grid of the standard star spectra in order to verify and in some cases slightly modify the quantitative spectral types. This step was undertaken separately by two of the authors.

Given the age estimate of the ONC, the stars were assumed to be dwarfs in the above quantitative classification methods, with surface gravity not explicitly taken into account for the initial classification. For spectral types later than about M5, however, there is a strong gravity dependence that must be considered in order to avoid classifying stars later in type than they really are. Since the TiO and VO bands acquire a distinct shape in proceeding from dwarf-like to giant-like surface gravities, particularly the red end of the TiO λ8465 band, this could be accounted for during the visual examination part of the process. Surface gravity effects are also seen in the strength of the Na i doublet (λλ8183, 8194 Å) at spectral types later than M2.

Indeed, being candidate pre-main sequence stars, our objects are expected to have surface gravities in between those of dwarf stars and giant stars, though closer to dwarfs. Thus following Slesnick et al. (2006a) who were themselves guided by the findings of Kirkpatrick et al. (1991), we investigated quantitatively the strength of the Na i λ8190 doublet in our spectra. This feature weakens towards lower values of surface gravity (see Figure 7 of Kirkpatrick et al. 2008 for a demonstration). We show in Figure 4 the same plot utilized by Slesnick et al. (2006b)—their Figure 4—of the gravity-sensitive Na i λ8190 index versus the temperature-sensitive TiO λ8465 and TiO λ7140 indices. ONC stars later than spectral type ∼M2 typically exhibit the intermediate Na i λ8190 line strengths expected from young 1–2 Myr old pre-main sequence stars. Based on this figure, we conclude that the vast majority of the stars M3 and later in our sample can be presumed ONC members given the consistency of their surface gravities with the assumed young age of the cluster. While a few of the M dwarfs could be interlopers along the line of sight with higher surface gravity, the ONC sample at I ≲ 17.5 mag appears to be largely uncontaminated by foreground dwarf stars.

Zoom In Zoom Out Reset image size

For those late type stars with medium to high signal-to-noise ratio and that are not severely affected by nebular contamination (which can contribute strong lines in either the on-feature or off-feature band regions) or strong accretion (which can render observed spectra too blue relative to the expected slope for the spectral type), the quantitative method is reliable and robust to visual examination. For stars with a significant amount of reddening, low signal-to-noise ratio, or both, experimentation has shown that the quantitative technique is in fact more trustworthy than ab initio examination by eye. Indeed, for most early-to-mid M type stars, the spectral types adopted were generally those assigned quantitatively. For the late M types, by-eye examination was more rigorous with detailed attention to surface gravity effects needed for accurate classification. At earlier types, some adjustments to the quantitative types were made for nearly all FGK stars. These were at the 1–3 sub-class level for late K stars but more for the earlier types for which our red region spectra include narrow features that are only weakly varying with temperature, and therefore challenging for our coarse quantitative technique. Methods employing equivalent width measurement would be more suitable for quantitative analysis than our wide bandpass methods at these earlier spectral types. As reported above, not all of the spectra were classifiable.

In order to assess the quality of the spectral typing, we can look at the sub-set of sources with more than one spectrum obtained and classified. As a reminder, we have 936 total spectra of 707 unique sources, of which 185 unique sources were observed more than once. Of these, only 89 sources have multiply derived spectral types; the remainder have only one or zero of the multiple spectra classifiable (due to either signal-to-noise or nebular contamination issues). Among the 89 stars, 78 have two classifiable spectra and 10 have three classifiable spectra. One source, H97 3132, had six spectra obtained of which five are classifiable. The range in the resulting, independently derived, spectral types has a median value of just one sub-class with standard deviation less than two sub-classes. Of the 89 repeated sources, 52 are typed within one sub-class and 71 within two sub-classes. The outliers with larger discrepancies tend to be G or K stars where accurate spectral typing is more difficult from red wavelength data. Notably, H97 3132 exhibited a range of only one sub-class among the types derived from the five different spectra, which were independently classified as M5.5, M6, M5, M5.5, and M5. It is thus clear that in the presence of good data and minimal complications from astrophysical effects, the methods are robust.

However, a number of systematics are present in the ONC region that can affect the spectral types. The main one is the accuracy of the sky/nebula subtraction. Some of our spectra have a blue spectral energy distribution but also late-type spectral features, which indicates that some nebular contamination remains. Although this slope/spectrum discrepancy can be seen in rapid accretors as well, the Orion Nebula itself predominantly causes the effect in our data sets. Several of the multiply observed stars with large spectral type discrepancies are cases in which one spectrum has much better sky subtraction than the other. For these stars it is easy to see that the later spectral type is the more appropriate one; however, other cases may exist where we have only one spectrum and have classified a star too early, e.g., G or K type, when it is really dominated by nebular continuum that if accurately subtracted would yield a later spectral type. As noted above, the presence of a strong accretion continuum can similarly make a spectrum appear blue as well as fill in the atomic absorption lines, resulting in a spectral classification that is biased earlier than the true type and we may indeed have some of these cases. Another complication to accurate spectral typing is reddening. While we believe our methods are able to account for modest amounts of reddening, it is possible that in extreme cases there are systematic effects that enter and cause larger than average spectral type errors. A final worry is the possible presence of cool spots, similar to those seen on older non-accreting but still active young stars. These could either produce or enhance late type features on the red side of our spectral range, leading to a bias towards a later spectral type assignment than truly characterizes the underlying star. In summary, one or more of the above astrophysical effects could be present and therefore there are possible spectral type biases in both directions.

Finally, we note that for comparison purposes we also ran our spectra through the distributed "SpTclass" software developed by Hernàndez et al. (2004). We found that the combination of spectral resolution and signal-to-noise effects, plus perhaps the lack of flux calibration in our dataset, rendered the SpTclass results suspect. Specifically, the results are somewhat degenerate for stars we classified by hand as F through K7 stars, with little useful output. Even for the standard stars in this spectral type range there was quite a lot of scatter and a bias in the F, G, and early K range for SpTclass to classify stars much later than their true types; in the K5–K8 range, however, there was good agreement. For the later type objects, M0–M9, we found a systematic offset both our object classifications and in tests with standard stars, of approximately 2–2.5 spectral subtypes, now in the sense that the SpTclass results were too early compared to the actual spectral type. We believe that this comparison demonstrates the need to classify stars relative to a grid of standards taken with identical or similar equipment. Furthermore, as our spectra are not flux-calibrated, having been taken with fibers rather than slits, there may be an additional explanation for the systematic effects reported above if the assumptions of the SpTclass software regarding continuum shape are being violated in our dataset. Regardless of any specific SpTclass biases, our experience suggests general caution when considering automated techniques for spectral classification, though they may be very useful when applied to data sets consistent with their development or in less challenging spectroscopy environments than the ONC.

The final spectral types for our ONC sample are presented in Table 2, where 555 are from the new WIYN/Hydra data and 99 are from the new Palomar/Norris data, with 35 stars classifiable in both spectral data sets. Our newly derived spectral types are on the same system as those in H97, having been assessed by the same author. Figure 5 shows the run of apparent I-band magnitude with spectral type. The scatter at any given magnitude can be attributed to cumulative differences in foreground extinction, accretion continuum excess, and/or stellar radius. For 254 of the 619 stars, the spectral types are the first ever reported. An additional 63 stars have spectral types in other literature more recent than the H97 study and compilation, as cited in Table 2.

Zoom In Zoom Out Reset image size

The strong continuum and line emission from the Orion Nebula combined with the difficulties of sky subtraction using multi-fiber instruments in high background fields such as the ONC, means that many of the emission lines traditionally used to diagnose accretion and outflow in T Tauri stars (e.g., Hα and other Balmer lines, [N ii], [S i] and other forbidden lines) cannot be used to identify such objects of interest among our sample stars. However, Ca ii triplet (λλ8498, 8542, and 8662) emission was identified by Hillenbrand et al. (1998) as being a robust accretion indicator in the ONC, even in the presence of strong nebular background since these high density Ca ii lines typically are not seen in nebular spectra at low spectral resolution. With the line strengths and profile widths that are observed in young stars, Ca ii triplet emission is attributed to accretion processes rather than to chromosphere activity that would produce weaker lines (Herbig & Soderblom 1980; Hamann & Persson 1992; Soderblom et al. 1993; Azevedo et al. 2006; Kwan & Fischer 2011). The physical conditions responsible for the emission appear to require high density, n_H ∼ 10¹² cm⁻³ and moderate to low temperature, T < 7500 K (Kwan & Fischer 2011).

Approximately 20% of our stars exhibit Ca ii triplet emission, consistent with the fraction reported in the earlier Hillenbrand et al. (1998) study. Figure 6 shows some example spectra of strong, medium, and weak emitters which occur mainly among the late K and early M spectral types while Figure 7 shows a histogram of the measured Ca ii λ8542 (pseudo) equivalent widths. We measured equivalent widths using splot in IRAF for each of the Ca ii triplet lines against the pseudo-continuum, as well as the nearby O i 8446 Å line. Notes were made concerning the strength of the upper Paschen lines and whether the Ca ii lines have significant H i contamination (plausibly from strong accretion, but more likely from incomplete nebular subtraction). Our identification of emitters has required that the Ca ii line series have either the 8542 Å line stronger than any 8600 Å pure H i, or 8498 Å stronger than 8542 Å. Because the local continuum is significantly depressed for M-type photospheres relative to earlier spectral types, there is an expected spectral type dependence of the measured equivalent widths for a given constant Ca ii line strength. We also note that many of the stronger emitters are variable among our several spectra, at a level up to 25%.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The range of equivalent width values in the ONC fully populates the range seen in other star-forming regions. A linear relationship between Ca ii triplet line flux and total accretion luminosity is now well-established in the literature (e.g., Muzerolle et al. 1998; Herczeg & Hillenbrand 2008; Dahm 2008). With knowledge of M_*/R_*, such measurements can lead to derivation of mass accretion rates. In the absence of stellar parameters, however, and with only equivalent widths rather then line fluxes (which would require either flux-calibrated spectra or contemporaneously obtained I-band photometry to mitigate errors due to variability) we cannot derive accretion rates here. We can comment, though, that the previously established correlations suggest that Ca ii λ8542 equivalent widths in the range 0.5–40 Å for low mass stars correspond to accretion luminosities in the range 10⁻⁵ to 0.1 L_☉ and accretion rates from <10⁻¹⁰ to 10^−6.5 M_☉ yr⁻¹.

We conclude that a wide range of accretion rates characterizes the young ONC population. Once stellar parameters are better established for our sample, these several hundred young stars can be further investigated for correlations between accretion properties and stellar mass and age.

As reported above, 619 stars are typed from our Norris and HYDRA data. While the observational focus of the new data acquisition was on those stars without previously determined spectral types, in practice we placed fibers on other ONC members rather than not assigning them at all. Thus the set of stars for which spectral types are derived in this paper overlaps significantly with those in the H97 paper. This enables comparison of results obtained with different equipment but using similar analysis methods.

In the vast majority of cases there is remarkable agreement between the previously and presently reported spectral types, as can be seen from examination of Table 2. This is despite the known observational challenges and the likely influence of spectrophotometric variability due to real changes in accretion/extinction parameters. Figure 8 illustrates the direct correlation between H97 and this work; the distribution exhibits a root-mean-squared scatter of 2.25 spectral sub-classes. However, some large discrepancies exist as well, notably when early-K versus mid-M spectral types have been derived for the same star. We advise in these cases that the later spectral type is likely the more accurate one since, as discussed earlier, there are plausible ways to make spectral types appear much earlier than the true type in some data.

Zoom In Zoom Out Reset image size

Histograms in Figure 9 show the distribution of stellar types from H97 compared to the distribution in the new spectral dataset. While the histograms are similar in terms of rising from earlier to later types with a peak around M5, there are notably fewer extremely late type stars seen here compared to what was reported in the H97 paper.

Zoom In Zoom Out Reset image size

da Rio et al. (2010) derive spectral types for 217 late type stars in the ONC based on narrow band photometry, employing a novel TiO photometric index defined as the measured magnitude in a 200 Å wide filter centered at 6200 Å relative to the magnitude at that same wavelength interpolated from a linear fit to V- and I-band photometry. da Rio et al. (2012) obtained a deeper photometric dataset and used a similar technique based on a longer wavelength 7700 Å narrow band filter, to derive spectral types and extinction values for 1280 objects. The TiO index (narrow band photometric) technique is in fact similar in spirit to the spectroscopic TiO band indices that we have employed here. The photometric spectral types from these two da Rio et al. studies range between M2 and M6, where the technique is valid, and have estimated errors of typically one to two sub-classes.

In Figure 10 we directly compare the da Rio et al. photometric spectral types with the spectroscopic spectral types quoted in Table 2; the correlation exhibits a root-mean-squared scatter of 1.75 spectral sub-classes. In total, a sample of 437 stars is available for comparison between the spectroscopic and photometric spectral types, with all 437 of those having 7700 Å TiO spectral types, and 315 having 6200 Å TiO spectral types as well, when both photometric methods could be applied. While the scatter seems large between the photometric and spectroscopic methods (filled symbols), it appears to be no worse at these late types than between the new spectroscopy and the old spectroscopy (open symbols) probing a slightly earlier spectral type range. It should be noted that, as discussed above, the error bars in this earlier regime are larger than those suggested at the later types from either the photometric methods (one to two sub-classes) or the spectroscopic methods (0.5 sub-classes based on repeated robust measurements of the same star but up to two sub-classes including systematic effects).

Zoom In Zoom Out Reset image size

da Rio et al. (2012) also devised procedures to de-redden their photometry and obtain both A_V and T_eff for individual late type stars. We have mimicked their analysis by performing synthetic photometry on our spectra based on the bandpasses specified by da Rio et al. (2012). Although the spectra we have are not flux calibrated, because the TiO index on-band and off-band filter profiles are so close in wavelength this should have only a minor effect. The resulting color–color diagram based on our spectra (not shown) is indeed similar in appearance to the one based on narrow band photometry that is presented in da Rio et al. (2012)—their Figures 5 and 11. Specifically, there is a narrow finger of stars having no indication of TiO depression but a range of reddening values, and then a spread along the TiO index or temperature-sensitive axis, as well as along a nearly orthogonal reddening vector.

We have further followed the da Rio et al. (2012) methodology for deriving both A_V and T_eff, finding for our sample extinction values in the 0–3 mag range and effective temperatures in the 2700–3900 K indicated. While illustrative of the consistency between different methods for deriving spectral types and then physical parameters from them, it should be noted that the scatter between the two TiO photometric methods (TiO 6200 Å in da Rio et al. 2010 versus TiO 7700 Å in da Rio et al. 2012) is itself 0.03 in log T_eff, corresponding to 1.5 spectral sub-classes in the M star regime, and comparable to the estimated errors in the photometric spectral typing method.

We conclude for mid- to late-M-type stars that spectral typing through the use of narrow-band photometric methods is no less accurate in the mean than spectral typing through spectroscopic methods. For early M stars, however, there is a systematic discrepancy with later types being assigned through the photometric method than appear reasonable from the spectroscopy. This could be due to nebular contamination in the spectra, the photometry, or both.

There is good correlation demonstrated overall in Figures 8 and 10, as well as general agreement over many decades in the spectral types derived by different authors for the same stars (see Table 2). However, there are some cases where large discrepancies are apparent.

We highlight the case of JW 20 as an illustration. The H97 spectral type (based on data from 1996 February) was M3.5. However, the spectrum presented here (based on data from 2006 January) is that of a K8 star, as illustrated in Figure 11. The star is only a small-amplitude optical variable (0.05 mag; Herbst et al. 2002) with all available literature and catalog near-infrared photometry consistent within 0.1 mag and the mid-infrared variability similarly small amplitude (Morales-Calderón et al. 2011). Nevertheless, between the two spectra taken 10 yr apart, the deep TiO and VO bands decreased in strength and Ca ii triplet absorption was revealed.

Zoom In Zoom Out Reset image size

We cannot explain this discrepancy in terms of signal-to-noise challenges nor can we rationalize an explanation through observational confusion since there are no other optical or near-infrared sources within 30''. Data for both spectra were taken with the WIYN/HYDRA fiber positioner and there are no indications among the other objects of problems with the pointing. Although the exact position of the sky fiber location was different, the sky subtraction method employed was the same, and we cannot see a way of producing these two different spectra if sky/nebula subtraction accuracy was the only variation.

Careful examination of Table 2 reveals several other such stars with large spectral type discrepancies. We are driven to the conclusion that in at least some cases, e.g., JW 20 and JW 798, astrophysical origin must be considered plausible. In other cases there may be explanations; as examples, for JW 625, H97 3030, and likely others, there is strong nebular contamination either on-source or in the available sky positions, while for H97 3067 a nearby brighter source could contaminate an on-source fiber spectrum during poor seeing conditions. These difficulties inherent to crowded, nebular star forming regions are non-astrophysical sources of discrepant spectral types that should be considered, in addition to low signal-to-noise, as possible classification biases.