Using APOGEE stellar parameters

Introduction / summary

Automated determination of stellar parameters for tens of thousands of stars from IR data is a challenging problem, and one that we certainly have not fully solved. As a result, there are some uncertainties and potential issues in the current stellar parameters that are being released as part of DR10. All potential users of these parameters should carefully consider the information presented on this page, and note that a significant fraction of the data for the sample being released has some cautionary note attached to it! Certainly, these parameters are subject to change in future data releases as we improve our understanding and techniques!

To understand the caveats presented here, users should understand the basics of how the ASPCAP pipeline works, and what the fundamental parameters affecting the IR spectra are, as described in the ASPCAP pipeline description.

Executive summary: The ASPCAP fits for DR10 provide Teff, log g, [M/H], [α/M], [C/M], and [N/M]; for definitions of these parameters, see the ASPCAP documentation. We believe that useful results for Teff, log g, and [M/H] are recovered for giants, albeit with some small calibration offsets applied (both raw and calibrated values are provided). [α/M] also seems fairly good, but there are some additional uncertainties for Teff<~4200 K. Results for [C/M] and [N/M] are considerably less reliable, and, since the spectra of cool stars are dominated by absorption lines due to OH, CO, and CN, which are sensitive to the abundances of those elements, uncertainties in all parameters are larger for Teff<~4200 K. For warmer stars (T>5500 K) and dwarfs (log g > 3.8), parameters are less well determined, but they are well enough determined to identify these as warm stars and dwarfs!

We release two sets of ASPCAP stellar parameter results: validated/calibrated data and raw ASPCAP results. The validated/calibrated data are released for Teff, log g, [M/H], and [α/M] for giants, while the raw ASPCAP results present the data for these parameters without calibrations applied, as well as for [C/M] and [N/M]. The raw data are likely to have systematic offsets and considerable uncertainty, but we release it because the calibrated data are derived from the raw data with all of its uncertainties. The calibration relations used are documented on the ASPCAP docmentation page and are taken from Meszaros et al (2013), where the data used to derive them are presented.

No calibrated parameters are released for stars with log g>3.8. No calibrated results for [C/M] or [N/M] are released for any stars.

ASPCAP flags

While DR10 presents stellar parameters and some abundances from the ASPCAP pipeline, users must be aware of several issues before using these indiscriminately. To facilitate this, every star has associated bitmasks, ASPCAPFLAG and PARAMFLAG, that flag known and potential issues with some or all of the stellar parameters. Users need to be sure to consult the bitmasks and use the derived parameters accordingly. ASPCAPFLAG is used to flag certain parameters as not to be used (BAD) or to be used with caution (WARN), and is also used to flag stars that are likely to be unreliable. If the BAD or WARN flag is set for any parameter, PARAMFLAG is an array of bitmasks that gives information, for each parameter, about the reason why that parameter was flagged.

Overall, quantities that are marked as BAD should not be used. Quantities that are marked as WARN have potential significant uncertainties, but are likely to be ballpark correct, or at least contain some relative information. For example, surface gravities at high gravities (dwarfs) are known to be significantly inaccurate, but they are good enough to identify that these stars are indeed dwarfs; all of these are flagged as WARN. In general, users are cautioned not to use data with WARN bits set for detailed quantitative chemical analysis indiscriminately. However, we also note that a large fraction of the DR10 data has such bits set, and that the stars with WARN bits set fall preferentially in certain regions of parameters space. In this respect, the elimination of stars with WARN bits from a given sample may subject it to important biases. Obviously, we are working hard to improve our understanding and the performance of the pipeline. Data released in DR10 represent our current best efforts, and they will certainly be improved in subsequent data releases.

To facilitate use, we provide summary STAR_WARN and STAR_BAD bits in addition to individual bits for different parameters and different conditions (as discussed further below). The STAR_BAD bit is set if any critical individual BAD bits are set (specifically, if TEFF, LOGG, CHI2, COLORTE, ROTATION, SN BAD bits are set, or any parameter is near grid edge). The STAR_WARN bit is set if the WARN bits are set for Teff (TEFF) or log g (LOGG) (but not for [M/H], [α/M], [C/M], or [N/M]) or for WARNings related to CHI2, ROTATION, SN, or COLORTE. Because of the definition, note that it is possible for a star not to have STAR_WARN fit, but still to have ALPHAFE_WARN or METALS_WARN set!

In addition to the ASPCAP flag bits, several star level flag bits as derived from the spectra may also be relevant, as encoded in the summary star data and in the STARFLAG bitmask. In particular, users may want to beware of stars with significant RV variation (as indicated by VSCATTER), as these are likely indicative of binarity (although not all cases of binarity are expected to lead to stellar parameter issues, e.g., if the luminosity ratio is large), and be more cautious about stars that are flagged as having potential issues from persistence in the detectors (PERSIST_HIGH, etc. bits in the STARFLAG mask).

One might hope that uncertainties in the stellar parameters are well represented by the parameter uncertainties produced by the pipeline. Unfortunately, the expected random errors are very small, as the true uncertainties almost certainly arise from systematics. We provide some estimate of the uncertainties from such effects, but overall, it is challenging to provide realistic uncertainties for all of the parameters. See more discussion of this in the uncertainties section below.

Details of different bits in the ASPCAP flags are discussed below:

Calibration of raw ASPCAP results

Results for effective temperature, surface gravity, and overall metal abundance have been calibrated by comparing APOGEE results for stars in stellar clusters spanning a wide range of metallicities, to data from the literature on those same stars. These comparisons show that there are likely some small, systematic errors in the raw ASPCAP parameters. The cluster stars have been used to derive corrections to the raw parameters, and most users probably want to use these calibrated parameters. However, the calibration relations have only been derived in certain regions of parameter space (effective temperature and surface gravity), so one needs to be more cautious about using parameters for stars that fall outside these regions. In such cases, no values are provided for calibrated parameters (apart from a mistake in populating Teff, see caveats page, and the {PARAM}_BAD or {PARAM}_WARN bit (where {PARAM} is one of TEFF, LOGG, METALS, ALPHAFE, CFE, or NFE) in ASPCAPFLAG is set along with the CALRANGE_BAD or CALRANGE_WARN bit in PARAMFLAG.

Specifically,

the Teff calibration is valid within the range 3500 < Teff < 5500 K and for log g < 3.8 (but see caveats)
the log g calibration is valid only within the range log g <3.8, 3500 < Teff < 5300, and -2.5 < [M/H] < 0.5.
the [M/H] calibration is valid only for log g <3.8 and Teff < 5500

In general, we have applied calibration relations for the red giants in the sample, with log g < 3.8 and Teff < 5500. For most other stars, calibrations have not been applied because we have fewer (or no) observations of calibrating objects outside the giant regime. As a result, for DR10, the stellar parameters for hotter stars and for stars of higher surface gravity cannot be trusted too much; stars with Teff > 5500 and/or log g > 3.8 are almost certainly warm stars and/or stars with higher surface gravity, but, beyond that, the values returned by the ASPCAP pipeline may not be highly accurate.

For all stars we supply both the original fit parameters as returned by the ASPCAP pipeline, as well as the parameters after the calibration relations have been applied. In the allStar and aspcapStar files, the fit parameters are provided in the FPARAM array, while the calibrated parameters are provided into the PARAM array. In the CAS database, the fit parameters are preceded by FIT_ (e.g. FIT_TEFF, FIT_LOGG, etc.), while the calibrated parameters are stored in TEFF, LOGG, etc.

Poor matches to synthetic spectra (CHI2)

Reduced CHI2 as a function of Teff for the DR10 sample

The synthetic spectra may not always be good matches to real spectra because of possible issues with the line list and/or model atmospheres adopted, or because of a peculiar chemical composition (e.g., carbon stars, S-type stars, etc.). It is likely that the quality of the results will be better in some regions of parameter space than in others. In particular, results at the coolest temperatures may be less reliable, judging from the quality of the matches to the synthetic spectra. The distribution of CHI2 with temperature for the entire sample is given in the figure to the right.

For DR10, we set CHI2_WARN if CHI2>10 and CHI2_BAD if CHI2>30. Note that this flags a significant fraction of stars, including most of the stars at cooler temperatures, leading to a strong bias about which stars appear in DR10 without this bit set.

Signal-to-noise (SN)

DR10 includes all data taken in the first year of survey operations. However, since most stars are observed multiple times, data for some stars in DR10 will not yet have been observed at the target signal-to-noise. Since the quality of ASPCAP results depends on S/N, there may be issues with the ASPCAP results for these stars. SN_BAD and SN_WARN bits in ASPCAPFLAG flag such stars.

Color-temperature relation (COLORTE)

Rotation/broad lines (ROTATION)

The technique used by ASPCAP -- matching against a template library of sythetic spectra -- is only valid to the degree to which stars are represented in the library. In particular, the library spectra are all for non-rotating stars, and any stars with significant rotation will not be well matched. While most giants are expected to be slow rotators, the same cannot be said for main sequence stars. An estimate of rotation is made during the radial velocity determination by comparing the width of the cross-correlation peak when cross-correlating the spectrum with best-matching template to the width of the autocorrelation of the best-matching template. When the observed peak is significantly broader than the autocorrelation peak, a bit is set, and this is used to trigger the ROTATION_BAD or ROTATION_WARN bit in ASPCAPFLAG.

Degeneracies between parameters

While we characterize the spectra using seven parameters (effective temperature, surface gravity, microturbulence, overall metal abundance, alpha-element abundance, carbon abundance, and nitrogen abundance), there are likely to be significant degeneracies between some parameters in some regions of parameter space.

Teff flags (TEFF)

Accuracy of the ASPCAP effective temperatures have been judged by comparing to temperatures obtained from photometric temperature indicators, and in some case, with effective temperatures from the literature, based on analysis of high resolution spectra of cluster stars included in the APOGEE sample. Results are not fully consistent: comparison with spectroscopic temperatures show scatter, but little mean offset, while comparison with photometric temperatures suggest a small offset at cooler temperatures with ASPCAP underestimating the temperatures, and ASPCAP overestimating temperatures at warmer temperatures (>5000K). Based on these, we have applied a small temperature correction to the ASPCAP results.

log g flags (LOGG)

Surface gravities have been checked/calibrated using observations of Kepler asteroseismic stars, for which independent surface gravities can be derived, and from cluster stars, for which surface gravities can be determined, for a given effective temperature, from theoretical isochrones for the age and chemical composition of the clusters (to the level to which these isochrones are accurate). These comparisons suggest that the ASPCAP gravities are systematically high, and provide calibration relations that we use to correct the ASPCAP gravities.

Unfortunately, the asteroseismic gravities are only available for relatively metal-rich stars. At the metal-poor end, observations in globular clusters suggest that the ASPCAP gravities are systematically even higher. This comparison is perhaps less secure than that from the asteroseismic stars, because the cluster comparison is with gravities that are derived from isochrones and the observed photometric sequences; as a result, the expected gravities depend on accurate temperature estimates as well as the assumption that the isochrones accurately represent real stars.

Our gravity calibration is derived from a combination of the stars with asteroseismic gravity and clusters at low metallicity. It is a significant correction, so there is some concern about gravities, although the formal scatter of the calibrators after correction is about 0.17 in log g.

For dwarfs, one expects stars to lie along sequences predicted from stellar models, but we find that the ASPCAP gravities are systematically low in this regime. Since we have less data to calibrate the issues here, we do not provide a correction for dwarfs, and warn that, while dwarfs are correctly identified as such by the ASPCAP results, the derived surface gravities have some significant issues.

metallicity ([M/H]) flags (METALS)

Metallicities ([M/H]) have been validated through observations of globular and open clusters. These suggest that the raw ASPCAP metallicities have some small offsets that are characterized as a function of metallicity, and these offsets have been applied to giants.

[α/M] flags (ALPHAFE)

Alpha-element enhancements ([α/M]) have been validated through observations of globular and open clusters. These suggest that the raw ASPCAP [α/M] seem generally reasonable. However, at the lowest and highest metallicities, there seems to be some degeneracy between [α/M] and [M/H], such that clusters at these metallicities have a larger spread in both quantities, with a correlation between the two. As a result, the ALPHAFE_WARN flag is set for very low and high metallicities.

IN addition, at the cooler temperatures, T < 4200, the results for [α/M] are correlated with the results for [C/M] and [N/M]. Since there is some uncertainty about the quality of these, all such stars are flagged with ALPHAFE_WARN.

[C/M] flags (CFE)

Comparison between ASPCAP [C/M] with literature data show large scatter and a significant zero point uncertainty. These data at this stage should be taken with extreme caution and we discourage their use for detailed chemical evolution analysis.

[N/M] flags (NFE)

Comparison between ASPCAP [N/M] with literature data show large scatter and a significant zero point uncertainty. These data at this stage should be taken with extreme caution and we discourage their use for detailed chemical evolution analysis.

ASPCAP uncertainties/errors

ASPCAP calculates internal parameter uncertainties based on the fit to the spectrum. However, there are also external uncertainties based on systematic issues and the quality of the models. The ASPCAP calibration documentation discusses these uncertainties in more detail.