MaStar Stellar Parameters
Accurate stellar parameters are essential for applications of the stellar library. Although some stars have stellar parameters derived by other surveys, such as APOGEE, LAMOST, and SEGUE, the sources are heterogeneous and they could have systematics relative to each other. In addition, some stars were selected based on photometry and do not have existing stellar parameters. Therefore, we want to derive stellar parameters uniformly for the whole sample using the MaStar spectra, which have high signal-to-noise ratio and excellent flux calibration. We provide here a value-added catalog that contains multiple sets of stellar parameter measurements derived based on MaStar spectra using different methods, and a master set of stellar parameters made by taking the median among those based on different methods. There are two versions of this catalog: v1 is the original version released with DR17, v2 contains improved parameter measurements for three of the four sets of parameters. We recommend using the latter. The catalog is described in full detail by Yan et al. (in prep) along with the papers on the individual efforts: Hill et al. (2022), Imig et al. (2022), Chen et al. (in prep), Lazarz et al. (in press, for v2), and Hill et al. (in press, for v2). Please cite all these papers when using this catalog.
Algorithm Details
The methods are:
- JI: Derived using a neural network which models flux as a function of parameters and is trained on a combination of empirical MaStar spectra with ASPCAP parameters and the model spectra produced by Allende Prieto et al. (2018). This method also derives the microturbulence parameter. (Imig et al. 2022). This set does not change between v1 and v2 of the catalog.
- DL: Full-spectrum fitting with a Markov Chain Monte Carlo (MCMC) sampler using interpolated BOSZ model spectra with continuum shape information included in the chi-square calculation. Extinction is fitted as a by-product. No photometry priors are used. In Version 1, the continuum-based component of the chi-square calculation is scaled to have approximately 1/10th as much influence as that of the continuum-normalized component in the fitting. Version 2 is updated to give the continuum-based component and the continuum-normalized component equal influence in the fitting, with improved uncertainty calculation. In addition, the extinction measurements in v2 are allowed to go slightly negative to account for flux calibration systematics. The quality flags for extinction are also updated. (Lazarz et al. in press).
- LH: Full-spectrum, single-template, pPXF fitting with an MCMC sampler, using a flat prior for Teff and log g based on the Gaia color-magnitude diagram. The continuum is modeled with a multiplicative polynomial and observations are fit with interpolated BOSZ and MARCS model spectra. In v2, we recalculate Teff, log g, [Fe/H], and [alpha/Fe] using improved diagnostics to flag unreliable parameters. For example, for [alpha/Fe] we now use the chi square of the model fit, the MCMC autocorrelation time and the error in our measurement to indicate the reliability of parameter recovery. Furthermore, in v2 we have corrected an issue in our pipeline that caused almost discrete values of [Fe/H] (see Hill et al. in press).
- YC: Full-spectrum fitting using both the BOSZ and MARCS model spectra without interpolation, with the result produced by a Bayesian average and a flat prior based on the Gaia color-magnitude diagram. The continuum is modeled with a multiplicative polynomial. Compared to v1, v2 has significantly improved [alpha/Fe] measurements, which are computed after first constraining the three main stellar parameters. Then in an expanded model grid that includes varying [alpha/Fe], a similar algorithm is used to estimate the [alpha/Fe] together with Teff, logg, and [Fe/H].
The master set of parameters is derived by taking the median of the parameters of the four methods when they are available and are considered valid. Each set of parameter measurements has an associated validity flag to indicate whether the parameters of that method is valid for each star or each visit. For the median set, we provide a few columns (INPUT_GROUPS and INPUT_GROUPS_NAME) to indicate results from which methods are included in the median calculation. The median for [alpha/Fe] is treated differently from the other three properties. Additional quality filtering is done on [alpha/Fe] beyond that indicated by the validity flag associated with each method. The columns, INPUT_ALPHA_GROUPS and INPUT_ALPHA_GROUPS_NAME, indicate which sets of [alpha/Fe] parameters were included in the median [alpha/Fe] calculation.
The parameter derivations are done for each visit spectrum individually and only for those good quality visits. For those stars with multiple good visits, multiple measurements are generated which could be used to assess the internal uncertainty. We provide two tables (files) in this VAC. One has one entry per visit giving the per-visit measurements. The other has one entry per star which gives the median measurements among the many good visits of the star. The master set for the per-star table is done by first taking median among multiple visits and then taking median among multiple methods.
The metallicity measurement is critical for assigning spectra to different metallicity bins when building stellar population models. We applied a calibration on 3 of the 4 sets of [Fe/H] measurements using the overlap sample with APOGEE and the parameters derived by ASPCAP. The median of the calibrated [Fe/H] is provided in this VAC along with the median of the original [Fe/H] measurements. For the parameters from each method, we only provide the original [Fe/H] but not the calibrated ones. The calibration formula are all quadratic in the following form with the coefficients given in the tables below.
[Fe/H]cal = a[Fe/H]2+b[Fe/H]+c
| v1 | |||
|---|---|---|---|
| Method Abbrv. | a | b | c |
| JI | 0 | 1 | 0 |
| DL | -0.1161 | 0.7702 | 0.0755 |
| LH | -0.2317 | 0.7882 | 0.1337 |
| YC | -0.2418 | 0.9382 | 0.1260 |
| v2 | |||
| Method Abbrv. | a | b | c |
| JI | 0 | 1 | 0 |
| DL | -0.1183 | 0.7000 | 0.0694 |
| LH | -0.1798 | 0.9514 | 0.1634 |
| YC | -0.2543 | 0.8007 | 0.1329 |
The VAC also provides the total metallicity ([Z/H]) in addition to the iron-to-hydrogen ratio ([Fe/H]). This is derived by combining [Fe/H] and [alpha/Fe] measurements to compute the total number of atoms heavier than Helium, relative to the number of Hydrogen atoms. We assume the abundance pattern among the alpha-elements and that among the other elements respectively follow those in the solar abundance computed by Asplund (2005).
Data Access
The per-star file, mastar-goodstars-v3_1_1-v1_7_7-params-(version).fits can be downloaded at this link with the data model here. We recommend using v2. The table is also available on the SkyServer under the DR17 context with the name "mastar-goodstars-param". The table for v1 is available under the DR17 context while the table for v2 will be available under the DR18 context (expected December 2022). It can be queried using either the SQL Search tool or the CasJobs data access tool.
Caveats
The stellar parameter estimates presented here are our best effort. If you find quality issues, please let us know by emailing Renbin Yan at rbyan (at) cuhk.edu.hk .
