LAMOST OBSERVATIONS IN THE KEPLER FIELD: SPECTRAL CLASSIFICATION WITH THE MKCLASS CODE

Spectral classification can play important roles in large-scale spectroscopic surveys, such as the current LAMOST-Kepler survey, the Sloan Digital Sky Survey, and others. Those roles are closely tied to the two main goals of spectral classification. The first of those goals is to place a star within the broad population of stars—that is, to locate the star in the Hertzsprung–Russell diagram. This yields, via calibrations, first estimates for the physical parameters of the star, and thus classification is an important first step in the spectroscopic analysis of a star. The second goal of spectral classification is to identify astrophysically interesting stars. Spectral classification is a powerful tool specifically designed to accomplish that task with minimal effort. A catalog containing two-dimensional spectral types augmented with peculiarity codes constitutes a rich database that may be mined to find, for example, candidate solar twins, solar analogs, chemically peculiar A-type stars, barium dwarfs and giants, white dwarfs, or many other desired stellar categories.

The MKCLASS code (Gray & Corbally 2014, v1.07)⁷ , an expert system designed to classify blue–violet spectra on the MK Classification system, was employed to produce the spectral classifications reported in this paper. MKCLASS was designed to reproduce the steps skilled human classifiers employ in the classification process. Central to this process is the direct comparison of the program star with the MK standard stars which define the system. A unique feature of MKCLASS is its ability to recognize many of the common spectral peculiarities, such as chemically peculiar A-type stars (both Ap and Am stars), metal-weak stars, carbon anomalies (such as strong or weak CN and CH bands), stars with enhanced s-process abundances (the barium dwarfs and giants) and others. MKCLASS also has rudimentary code for recognizing non-MK spectral types, such as white dwarfs, carbon stars, Wolf Rayet stars, and various other emission-line stars. A description of the MK Classification System, and the various types of spectral peculiarities can be found in Gray & Corbally (2009).

MKCLASS is distributed with two standard libraries, one based on rectified blue–violet (3800–4600 Å) spectra with $R\sim 2200$, obtained with the Dark Sky Observatory (Appalachian State University) 0.8-m telescope using the GM spectrograph with a 1200 g mm⁻¹ grating. The second library is based on spectra obtained with the 600 g mm⁻¹ grating on the same instrument. Those spectra have a spectral range from 3800–5600 Å, a spectral resolution $R\sim 1100$, and are flux calibrated. Since the LAMOST spectra are flux calibrated, we chose to use the second library for this classification project. A list of MK standard stars that this library is based on may be found on the MKCLASS website. This required degrading the resolution of the LAMOST spectra to R = 1100, and truncating those spectra to the 3800–5600 Å spectral range. A preprocessor was written to accomplish those tasks utilizing the auxiliary programs distributed with MKCLASS.

For "normal" stars, MKCLASS outputs not only a spectral type on the MK system, but also a quality evaluation, of "excellent," "vgood," "good," "fair" and "poor." These ratings are based on ${\chi }^{2}$ differences between the program spectrum and the best matched interpolated MK standard. Very few classifications are given a rating of "excellent"; this requires almost exact correspondence between the program spectrum and the interpolated spectral standard. Spectral types with a "vgood" quality rating will have, typically, an uncertainty (one standard deviation) similar to that reported in Gray & Corbally (2014), that is ±0.6 spectral subtype in the temperature dimension where 1 spectral subtype is the difference between, for instance, F5 and F6. In the luminosity dimension, the uncertainty is about 0.5 luminosity class, where one luminosity class is the difference between, for instance, "V" and "IV." This uncertainty is somewhat dependent on spectral type. Luminosity classification is particularly difficult in the mid A-type stars, where the uncertainty might rise to ±1.0 luminosity class. For spectra with a "good" quality rating, the errors are typically twice those for the "vgood" category. Spectral types with "fair" are much more uncertain, and those with "poor" are unreliable. The last two quality categories can usually be traced to either low signal-to-noise ratio (S/N) spectra, or spectra with obvious defects. Bear in mind that classifications with certain spectral peculiarities, such as strong or weak CN or CH bands might be assigned a lower quality because of significant differences between the program spectrum and the best-matched standard.

Peculiar stars that are so peculiar that a match with a single MK standard is not possible are not given a quality rating. The most common example is an Am star, or "metallic-line" A-type star where the Ca ii K-line, the hydrogen lines, and the metallic-line spectrum are each given separate spectral types. The spectral type of such stars is written in the following way: kA2hF0mF2, where, in this example, A2 represents the K-line spectral type, F0 the hydrogen-line type, and F2 the metallic-line type.

Further comments on the reliability of the spectral types can be found in Section 4.

MKCLASS is a work in progress, and many improvements have been made to the code in the course of the LAMOST-Kepler project. These improvements have resulted in version 1.07 of this code, which may be obtained online at  http://www.appstate.edu/~grayro/mkclass/.

The main results of this paper are presented in the form of stellar classifications on the MK system in Table 1. That table consists of 81,171 entries giving MK classifications for the objects in the LAMOST-Kepler database that have plausibly classifiable spectra. The table is ordered in terms of the Kepler ID, except for those objects that do not have a Kepler designation; those objects are appended to the end of the table. The table consists of seven columns. The first column is the name of the original LAMOST fits file; many of these fits files are already in the public domain and may be downloaded from the LAMOST spectral archive (http://www.lamost.org). The unreleased spectra may be obtained upon request after becoming an external collaborator of the LAMOST-Kepler project.⁸ The second column is the Kepler Input Catalog designation expressed in the form KICNNNNNNNN. For those objects without a KIC designation, this column is blank. The third column is the stellar designation recorded in the header of the fits file. The fourth column is the S/N of the spectrum in the Sloan g-band (${\lambda }_{{\rm{eff}}}=4686$ Å), also taken from the fits header. The effective wavelength of this band coincides approximately with the center of the spectral range utilized for classification by MKCLASS with the libnor36 library (3800–5600 Å). The fifth column is the classification itself in the traditional format of an MK spectral type, with the temperature type first, the luminosity type second, followed by a list of detected peculiarities. For stellar types that are not on the MK system (such as white dwarfs), Table 2 lists the codes that MKCLASS uses for those types; not all of those codes can be found in Table 1. The sixth column of Table 1 is the quality, discussed in Section 3.2 above. The seventh column contains further notes on the quality (see below). Some example rows of Table 1 are reproduced in this paper. The remainder of that table (81,171 lines in total) may be obtained online.

While many of the LAMOST spectra are of excellent quality, there are some problems with the data set. Some of the spectra are too noisy to classify. We have written the preprocessor to reject those spectra with essentially no signal in the 3800–5600 Å range. Some of the spectra have gaps or are truncated, and since those gaps would interfere with the operation of MKCLASS, most of those spectra are also rejected by the preprocessor. Another problem with the LAMOST spectra that has been largely addressed with the latest LAMOST pipeline (v2.7.5) concerns sky, background, or scattered light subtraction (below referred to as "background subtraction"). In earlier pipelines, it was clear that this subtraction was not always properly executed, as the cores of strong lines (such as the Ca ii H & K lines) in some spectra, even those with high S/N, had negative fluxes. In other spectra too little background light was subtracted. In both cases the spectral type (and other types of analysis) could be badly affected.

This problem has been addressed, mostly successfully, with the latest pipeline, but some spectra still exhibit these problems. Spectra with consistently negative fluxes at the violet end, or negative fluxes in the cores of strong lines (such as the Ca ii K & H lines)—both indicative of excessive background subtraction—have been labelled as "unclassifiable" in Table 1. Spectra with some negative points (which may arise either from too much background subtraction, or from noise) are indicated with a "?" in the notes column. Spectra with too little background subtraction are very difficult to detect by machine, but as it turns out, the spectral types assigned to those spectra by MKCLASS are frequently given quality ratings of "fair" or "poor," and/or are assigned unlikely spectral types, such as hypergiants or extreme metal deficiency. Those types have not been removed from the Table 1, simply because it is too time consuming to scan visually through 81,000 spectra to accomplish that task. We estimate that the percentage of spectra that are badly affected by incorrect background subtraction amounts to about 3%, with another 10% somewhat affected. To be safe, all those spectral types with "?" in the Notes column, and/or those with quality ratings of "fair" or "poor" should be regarded with some suspicion. When quoting any of the spectral types in Table 1 in other studies, the spectral type should always be accompanied by the quality rating and the comments in the notes column.

The quality of the spectrum and the resulting spectral type may also be judged from the Sloan g-band S/N listed in the tables. The actual S/N of the spectra submitted to MKCLASS is slightly higher (by a factor of ∼1.6) than those listed because of the filtering procedure used to reduce the resolution of the LAMOST spectrum to that of the MKCLASS spectral library. Even taking that into consideration, many of the tabulated values of the S/N seem somewhat low compared to the actual appearance of the spectra. The S/N is correlated with the MKCLASS quality assignments. For spectral types rated as "excellent," the average S/N = 180; for "vgood," S/N = 73; "good," S/N = 42; "fair," S/N = 11, and "poor," S/N = 14. Spectral types rated "poor" have a slightly higher average S/N than those rated "fair" because of the greater proportion of faulty spectra (incorrect zeropoints, truncations, and gaps) as well as including spectra that are obviously overexposed, but which are assigned a very high S/N by the pipeline.

Yet another way of judging the quality and consistency of the MKCLASS spectral types is by comparing spectral types derived from two or more spectra taken of the same star. Some examples for spectral types from B–M are illustrated in Table 1. For spectra with reasonable S/N ($\gtrapprox 20$), the agreement is excellent and usually well within the errors expected for the different quality ratings. Some further examples are noted in Table 3.

Figure 2 shows histograms of spectral type frequency in Table 1 organized in panels according to luminosity type. While the stellar sample selected for the LAMOST-Kepler is not strictly a magnitude-limited sample (cf. De Cat et al. 2015), it does show the basic characteristics of such a sample. In a volume-limited sample the frequency of spectral types on the main sequence increases with decreasing luminosity, but in a magnitude-limited sample, as here, late K-type and M-type dwarfs appear with a low frequency because of their intrisically faint luminosity. Giants thus dominate the sample for spectral types later than K2. The most frequent stellar types in Table 1 are F and G dwarfs and subgiants, exactly the stellar types targeted by the Kepler mission.

Zoom In Zoom Out Reset image size

As noted in Section 4.1, MKCLASS has a rudimentary ability to identify stars with "non-MK" types, that is stars with spectral classifications that are not contained within the 2D framework of the MK system. These types are listed in Table 2, and the numbers in parentheses in that table give the frequencies with which they appear in Table 1. Note that the category "emission-line?," for which there are 609 entries in Table 1, is meant to be a grab-bag for all types of emission-line stars not accounted for in the other categories in Table 2, such as cataclysmic binaries, novae of various types, B[e] stars, etc. Unfortunately, it appears that many of the stars that fall into that category have spectra that are flawed with "spikes" arising from cosmic rays or, more likely, CCD read errors. In the early data sets from LAMOST (most notably, those taken during 2011), such flawed spectra were very common. It appears that in the newer data sets, which form the basis for this paper, such flaws now appear in less than 1% of the spectra.

Table 1 may be used for interesting statistical studies, provided those studies are not compromised by the selection criteria used in establishing the LAMOST-Kepler sample. A good example is determining the frequency of occurrence of Am stars (A-type metallic line stars—see Section 3.2). Abt (1981) studied the frequency of Am stars in the field and determined that Am stars constitute 32% of the dwarf and subgiant stars with spectral types between A4 and F1. In Table 1 we count a total of 3088 dwarfs and subgiants with spectral types between A4 and F1. That number includes 1067 Am stars with hydrogen-line types between those same limits. This gives a frequency of 34.6%, in excellent agreement with Abt.

The vast majority of Am stars have hydrogen-line and metallic-line types of F2 and earlier, although there is a small sample with hydrogen-line and metallic-line types as late as F5. These are known as the ρ Puppis stars (Gray & Garrison 1989). Some instances of those stars may be found in the database. Interestingly, there are a small number of stars that MKCLASS has classified using the Am notation (see Section 3.2) but which have hydrogen-line and/or metallic-line types beyond F5, some even in the G class. In some instances, these curious spectral types are derived from very noisy spectra. However, in other instances, it is clear that MKCLASS is using the Am notation to classify composite spectra—spectra that arise from the superposition of the spectra of two stars with very different spectral types. A common example (cf. Corbally 1987) is an A dwarf with a G giant.