APOGEE Spectral Libraries

This page describes the synthetic spectral libraries that are used by ASPCAP to determine stellar parameters and abundances.

Context and Overview

As described on the ASPCAP page, the APOGEE/ASPCAP analysis functions by comparing observed spectra to a library of synthetic spectra, and associating the parameters of the best-matching synthetic spectrum to the object. The synthetic libraries are created by using a spectral synthesis code, which calculates a spectrum given input about the structure of a stellar atmosphere (a model atmosphere) and input about the atomic and molecular absorption lines (a linelist)

For DR17, new spectral grids have been constructed using the Synspec (reference) spectral synthesis code that incorporate NLTE level populations for Na, Mg, K, and Ca (Osorio et al. 2020). This differs from previous data releases that have used the Turbospectrum LTE synthesis code. However, while the results using the Synspec grids are the primary DR17 results, a set of supplemental analyses are provided in separate files that allow interested users to investigate effects of NLTE/LTE, different synthesis codes, plane-parallel versus spherical geometry, etc.

The spectral synthesis samples the spectrum at higher resolution than the observed spectra. To make a comparison, the synthetic grids are convolved with a line-spread-function (LSF) to match the observed spectrum. Since the LSF varies across the APOGEE detectors, spectral grids are calculated for four different LSFs for each of APOGEE-N and APOGEE-S, and an appropriate grid is chosen based on the mean fiber number of the combined spectra that are run through ASPCAP. For more details, see Holtzman et al. (2018).

Model atmospheres

The model atmosphere grid was created specially for APOGEE by the MARCS (reference) group. The grid provides a rectangular coverage of parameter space in five dimensions: Teff, log g, [M/H], [$\alpha$/M], and [C/M]. The rectangular nature of the grid implies that there are many regions in which we do not expect to find stars, and in some of these regions, the calculations may be suspect because of extreme combinations of parameters. However, a rectangular grid is needed for the ASPCAP methodology, in particular, the FERRE code.

Orphan Osorio et al. (2020) link.

There are also some locations in the atmosphere grid where the structure calculations do not converge, leading to "holes" in the atmosphere grid. Spectra for these "holes" are filled by interpolation from surrounding synthetic spectra using the method of radial basis functions (RBF), i.e., they are filled at the synthesis level, not at the atmosphere level. For more information, see Jonsson et al (2020).

The model atmosphere grid is available at ....

Line list

A custom line list has been created for APOGEE as described in Smith et al. (2021) and Shetrone et al. (2015) It contains hundreds of thousands of lines in the APOGEE spectral window. Data for atomic lines was compiled from several sources, but some of the atomic data, e.g., the log gf and damping parameters, were adjusted based on comparison of synthetic spectra to very high resolution spectra of the Sun and Arcturus, although the adjustments were limited to be within the laboratory uncertainties of the input values. For molecular lines, data were adopted directly from several literature sources.

Different synthesis codes use the linelist in different formats. The APOGEE linelists are available at ...

Spectral grids

The spectral grids cover ranges of Teff, log g, [M/H], [$\alpha$/M], [C/M], [N/M], microturbulent velocity, and, for dwarf grids, rotational velocity. To be able to process all of the APOGEE stars, the grids need to be kept small enough to fit into available computer memory. This is achieved by two methods: 1) the coverage of stellar parameter space is split into five regions based on Teff and log g, and grids are constructed for each region, and 2) the spectra in each grid are compressed using principal component analysis (PCA) which reduces the size by almost an order of magnitude with limited loss of accuracy.

The spectral grids are calculated at a sampling of 0.05 Angstroms.

For DR17, five grids are calculated: two giant grids (GK and M), and three dwarf grids (F, GK, and M). The coverage and spacing of the models in these grids is described in Jonsson et al (2020) and is summarized below:

Library Table

Libraries
Class Dimensions Teff log g log vmicro log vmacro log vrot [M/H] [C/M] [N/M] [α/M]
GK giant 7 3500 to 6000 0 to 5 -0.301 to 0.903 f([M/H]) 0. -2.5 to 0.5 -1.5 to 1 -0.5 to 2 -0.75 to 1
    step: 250 step: 0.5 step: 0.301 ... ... step: 0.5 step: 0.25 step: 0.5 step: 0.25
GK dwarf 8 3500 to 6000 0 to 5 -0.301 to 0.903 0. 0.176 to 1.982 -2.5 to 0.5 -0.5 to 1.5 -0.5 to 0.5 -0.75 to 1
    step: 250 step: 0.5 step: 0.301 ... step: 0.301 step: 0.25 step: 0.25 step: 0.5 step: 0.25
M giant 7 2500 to 4000 -0.5 to 5 -0.301 to 0.903 f([M/H]) ... -2.5 to 0.5 -1.5 to 1 -0.5 to 2 -0.75 to 1
    step: 100 step: 0.5 step: 0.301 ... ... step: 0.5 step: 0.5 step: 0.5 step: 0.5
M dwarf 8 2500 to 4000 -0.5 to 5 -0.301 to 0.903 0. 0.176 to 1.982 -2.5 to 0.5 -0.5 to 0.5 -0.5 to 1.5 -0.75 to 1
    step: 100 step: 0.5 step: 0.301 ... step: 0.301 step: 0.5 step: 0.5 step: 0.5 step: 0.5
F 8 5500 to 8000 1 to 5 -0.301 to 0.903 0. 0.176 to 1.982 -2.5 to 0.5 -0.5 to 0.5 -0.5 to 1.5 -0.75 to 1
    step:250 step: 0.5 step: 0.301 ... step: 0.301 step: 0.25 step: 0.25 step: 0.5 step: 0.25

Synspec NLTE

The primary grids used for the DR17 analysis were made using the Synspec spectral synthesis code (reference) , version s54h. NLTE population levels for Na, Mg, K, and Ca were supplied from the calculations of Osorio et al. (2020).

Turbospectrum 20 (LTE)

In previous data releases, grids were constructed using the Turbospectrum (reference) code. For DR17, a separate set of Turbospectrum grids were constructed to allow some comparison arising from the use of different spectral synthesis codes. The DR17 Turbospectrum grids differ from those used in previous data releases in that they are made using a newer version of Turbospectrum, Turbospectrum 20, which includes better handling of Stark broadening.

Turbospectrum calculations are fully LTE. However, Turbospectrum, unlike Synspec, is able to do the radiative transfer calculations in spherical geometry, which can become important for the most luminous (and largest) giant stars.

Synspec LTE

To facilitate the comparison of different codes, another set of grids were constructed using Synspec, but without the NLTE level populations, i.e. a full LTE calculation that can be directly compared with Synspec NLTE grid and results.

Turbospectrum 20 plane-parallel

Finally, to investigate the effects of using spherical geometry, a set of Turbospectrum grids was created using plane-parallel geometry for all models. These differ from the set of Turbospectrum grids described above only for the giant star grids.

Line spread functions

The line spread function (LSF) for the APOGEE spectrograph varies as a function of wavelength and location along the pseudo-slit. A parameterization of the LSF for each fiber is made as part of the data reduction using Gauss-Hermite polynomials, based on observations of night sky lines (which is a bit tricky, because many of these are doublets).

To account for some of the LSF variation, we calculate grids for four separate regions along the pseudo-slit within which the variation is relatively small, using an average LSF within each region; see Holtzman et al (2018) for additional discussion.

The spectra that are processed by ASPCAP are combined from multiple visits, in which the star may have been observed in different fibers. Usually, however, the variations in plugging do not move the stars too far along the pseudo-slit. To keep things manageable, we calculate a mean fiber for each combined spectrum, and choose the LSF grid based on this.

The LSF convolution with the spectral synthesis is done at subpixel sampling. The subsampled LSFs used for each grid are available at ....

Data files

Data files ...