2018A&A...619A.134H

The rapidly increasing number of stellar spectra obtained by existing and future large-scale spectroscopic surveys feeds a demand for fast and efficient tools for the spectroscopic determination of fundamental stellar parameters. Such tools should not only comprise customized solutions for one particular survey or instrument, but, in order to enable cross-survey comparability, they should also be capable of dealing with spectra from a variety of spectrographs, resolutions, and wavelength coverages. To meet these ambitious specifications, we developed ATHOS (A Tool for HOmogenizing Stellar parameters), a fundamentally new analysis tool that adopts easy-to-use, computationally inexpensive analytical relations tying flux ratios (FRs) of designated wavelength regions in optical spectra to the stellar parameters effective temperature (T_eff), iron abundance ([Fe/H]), and surface gravity (logg). Our T_eff estimator is based on FRs from nine pairs of wavelength ranges around the Balmer lines Hβ and Hα, while for [Fe/H] and logg we provide 31 and 11 FRs, respectively, which are spread between ∼4800Å and ∼6500Å; a region covered by most optical surveys. The analytical relations employing these FRs were trained on N=124 real spectra of a stellar benchmark sample that covers a large parameter space of T_eff~=4000-6500K (spectral types F to K), [Fe/H]~=-4.5 to 0.3dex, and logg~=1-5dex, which at the same time reflects ATHOS' range of applicability. We find accuracies of 97K for T_eff, 0.16dex for [Fe/H], and 0.26dex for logg, which are merely bounded by finite uncertainties in the training sample parameters. ATHOS' internal precisions can be better by up to 70%. We tested ATHOS on six independent large surveys spanning a wide range of resolutions (R=λ/Δλ~=2000-52 000), amongst which are the Gaia-ESO and the SDSS/SEGUE surveys. The exceptionally low execution time (<30ms per spectrum per CPU core) together with a comparison to the literature parameters showed that ATHOS can successfully achieve its main objectives, in other words fast stellar parametrization with cross-survey validity, high accuracy, and high precision. These are key to homogenize the output from future surveys, such as 4MOST or WEAVE.