The AMBRE Project: Origin and evolution of sulfur in the Milky Way
J. Perdigon, P. de Laverny, A. Recio-Blanco, E. Fernandez-Alvar, P. Santos-Peral, G. Kordopatis, M.A. Alvarez
AAstronomy & Astrophysics manuscript no. sulfur_article © ESO 2021February 4, 2021
The AMBRE Project: Origin and evolution of sulfurin the Milky Way (cid:63)
J. Perdigon (cid:63)(cid:63) , P. de Laverny , A. Recio-Blanco , E. Fernandez-Alvar , P. Santos-Peral , G. Kordopatis , and M.A.Álvarez Université Côte d’Azur, Observatoire de la Côte d’Azur, CNRS, Laboratoire Lagrange, Bd de l’Observatoire, CS 34229, 06304Nice cedex 4, France. CIGUS -CITIC- Department of Computer Science and Information Technologies, University of A Coruña, SpainReceived < date > / Accepted < date > ABSTRACT
Context.
Sulfur is a volatile chemical element that plays an important role in tracing the chemical evolution of the Milky Way andexternal galaxies. However, its nucleosynthesis origin and abundance variations in the Galaxy are still unclear because the number ofavailable stellar sulfur abundance measurements is currently rather small.
Aims.
The goal of the present article is to accurately and precisely study the sulfur content of large number of stars located in the solarneighbourhood.
Methods.
We use the parametrisation of thousands of high-resolution stellar spectra provided by the AMBRE Project, and combineit with the automated abundance determination GAUGUIN to derive local thermodynamic equilibrium (LTE) sulfur abundances for1855 slow-rotating FGK-type stars. This is the largest and most precise catalogue of sulfur abundances published to date. It covers ametallicity domain as high as ∼ / H] ∼ -2.0 dex. Results.
We find that the sulfur-to-iron abundances ratio is compatible with a plateau-like distribution in the metal-poor regime, andthen starts to decrease continuously at [M / H] ∼ -1.0 dex. This decrease continues towards negative values for supersolar metallicitystars as recently reported for magnesium and as predicted by Galactic chemical evolution models. Moreover, sulfur-rich stars havingmetallicities in the range [-1.0,-0.5] have very di ff erent kinematical and orbital properties with respect to more metal-rich and sulfur-poor ones. Two disc components, associated with the thin and thick discs, are thus seen independently in kinematics and sulfurabundances. The sulfur radial gradients in the Galactic discs have also been estimated. Finally, the enrichment in sulfur with respectto iron is nicely correlated with stellar ages: older metal-poor stars have higher [S / M] ratios than younger metal-rich ones.
Conclusions.
This work has confirmed that sulfur is an α -element that could be considered to explore the Galactic populationsproperties. For the first time, a chemo-dynamical study from the sulfur abundance point of view, as a stand-alone chemical element,is performed. Key words.
Galaxy: abundances – Galaxy: stellar content – Galaxy: evolution – Galaxy: discs - stars: abundances
1. Introduction
Sulfur is a chemical species of particular importance in the con-text of stellar nucleosynthesis and the chemical evolution ofgalaxies. It is a volatile element and, as a consequence, it is notblocked into the dust grains of the interstellar medium (ISM).It is therefore a good tracer of the chemical evolution of galax-ies, in particular at large redshifts (Savage & Sembach 1996).Moreover, from the stellar nucleosynthesis point of view, sul-fur is classified as an α -element (e.g. oxygen, magnesium, tita-nium, ...). It is indeed produced via α -capture in the inner lay-ers of massive stars (see e.g. Woosley & Weaver 1995; Nomotoet al. 2013). These chemical species are then released into theISM mostly through Type II supernovae on a relatively shorttimescale. It is therefore believed that the abundance of all these α -elements approximately follows the same behaviour duringthe Galactic chemical evolution. On the other hand, althoughalso partly produced in massive stars, iron is mostly produced (cid:63) Tables 5 and 6 are available in electronic form at the CDSvia anonymous ftp to cdsarc.u-strasbg.fr (130.79.128.5) or viahttp: // cdsweb.u-strasbg.fr / cgi-bin / qcat?J / A + A / (cid:63)(cid:63) Send o ff print requests to Patrick de Laverny and expelled into the ISM by Type Ia supernovae on a muchlonger timescale. Thus, we expect that the [ α / Fe] ratio (andthus [S / Fe]) remains almost constant with respect to iron con-tent in the metal-poor regime (i.e. as a plateau), correspondingto the epoch before the ignition of the first Type Ia supernovae.Then, this ratio is expected to decrease afterwards as soon asthe amount of released iron increases with time. Adopting suchproduction sites, Galactic chemical evolution models quite suc-cessfully reproduce the main observed behaviour of sulfur andother α -elements as a function of the metallicity (among otherrecent studies, see Prantzos et al. 2018; Grisoni et al. 2017; Pallaet al. 2020), confirming that the major production site of S isindeed Type II supernovae.On the observational side, a few studies have reported sul-fur abundance enhancements in the metal-poor regime. Francois(1988) was the first to suspect the α -like behaviour of sulfur for[Fe / H] < -1 dex with a plateau-like structure, and this was thenconfirmed by later studies (see e.g. Ryde & Lambert 2004; Nis-sen et al. 2004, 2007; Spite et al. 2011; Kacharov et al. 2015).These works invalidated the suggestion of Israelian & Rebolo(2001) who found a steady increase in [S / Fe] with decreasingmetallicities, leading to some suspected extreme sulfur-rich stars
Article number, page 1 of 14 a r X i v : . [ a s t r o - ph . GA ] F e b & A proofs: manuscript no. sulfur_article ([S / Fe] ≈ / H] < -2.0 dex). The plateau of sulfurin the metal-poor regime nonetheless appears with a large dis-persion mainly caused by the di ffi culty of analysing weak linesat these low metallicities. At higher but still subsolar metallic-ities (-1.0 dex < ∼ [Fe / H] < ∼ α -element, the decline of [S / Fe] with metallicity in the Galacticdisc is clearly observed and interpreted as the release of ironfrom Type Ia SNe at [Fe / H] < ∼ -1 dex (see e.g. Chen et al. 2002;Ca ff au et al. 2005; Ryde 2006; Matrozis et al. 2013). This there-fore supports the idea of a common nucleosynthetic origin forsulfur and other α -species.It is interesting to note that all these studies rely on rathersmall samples of stars (typically a few tens). However, in orderto be able to study possible di ff erent sulfur content in the variousGalactic populations, large statistics is necessary. This changedrecently thanks to a few studies. First, Luck (2015) reported S i abundances of ∼ / Fe]. Moreover, a strong tempera-ture dependency, probably caused by blending from unknownlines, caused blurring in the main picture. Then, Takeda et al.(2016) derived sulfur abundances for up to ∼
400 dwarfs andgiants, confirming the decrease in [S / Fe] with metallicity. Lateron, within the
Gaia -ESO Survey, Du ff au et al. (2017) managedto derive sulfur abundances for a sample of 1,301 Galactic stars,including stars in open and globular clusters, but with only halfa dozen stars below [M / H] ∼ -1.0 dex. Although the global be-haviour is again partially blurred by rather large measurementdispersions and temperature-dependent derived abundances, thisstudy seems to confirm the α -like behaviour of sulfur at subso-lar metallicities (-1 < ∼ [M / H] < ∼ α -element in the thin and thick discs,in rather good agreement with the literature models of Galac-tic sulfur evolution (Romano et al. 2010; Kobayashi et al. 2011;Prantzos et al. 2018).In the present study we profit from the spectra parametrisedwithin the AMBRE Project de Laverny et al. (2013) and pro-vided by the ESO archives of the HARPS, FEROS, and UVESspectrographs. We estimated precise and homogeneous sulfurabundances for a catalogue containing an unprecedented num-ber of stars, having metallicities varying from ∼ -2.0 dex to ∼ +
2. Derivation of the AMBRE sulfur abundances
This study has been carried out in the framework of the AM-BRE Project (de Laverny et al. 2013), whose first aim was toderive the main atmospheric parameters (e ff ective temperature T e ff , surface gravity log( g ), mean metallicity [M / H], and enrich-ment in α -elements with respect to iron [ α / Fe]) of ESO archivedspectra. This is performed thanks to the MATISSE algorithm(Recio-Blanco et al. 2006) trained with a specific grid of syn-thetic spectra (de Laverny et al. 2012). For the present analysis,we also use other AMBRE data products (for a detailed descrip- tion, see Worley et al. 2012) as the stellar radial velocity ( V Rad ),the signal-to-noise ratio (S / N), the full width at half maximum(FWHM) of the cross-correlation function (CCF) used to esti-mate V Rad (i.e. an estimate of the typical width of the lines, there-fore including the e ff ects of the rotational velocity), and a qualityflag of the stellar parametrisation (based on the computation of a χ between the observed and reconstructed spectra at the derivedstellar parameters).From these parametrised AMBRE spectra the sulfur abun-dances were then derived thanks to GAUGUIN, an optimisationmethod coupling a precomputed grid of synthetic spectra (seeSect. 2.2. for a description of this grid) and a Gauss-Newton al-gorithm. GAUGUIN was originally developed in the frameworkof the Gaia / RVS analysis within the
Gaia / DPAC for the estima-tion of the stellar atmospheric parameters: for the mathematicalbasis, see Bijaoui et al. (2010); and then, the first applications inBijaoui et al. (2012) and Recio-Blanco et al. (2016). A naturaland simple extension of GAUGUIN’s applicability to the deriva-tion of stellar chemical abundances was then initiated withinthe context of the
Gaia / RVS (DPAC / Apsis pipeline, Bailer-Joneset al. 2013), the AMBRE Project and the
Gaia -ESO Survey. Wefirst published a detailed description of the application of GAU-GUIN for the derivation of chemical abundances within the AM-BRE context in Guiglion et al. (2016). We also refer to Sect.2.2for its specific application to sulfur abundances.
For the selection of our analysed sulfur lines, we refer to theworks of Ca ff au et al. (2005), Du ff au et al. (2017), and Takedaet al. (2016) who studied multiplets 1, 6, and 8 (respectively at ≈ i multiplets in FGK star spectra, (iii) theexistence of almost blend-free spectral ranges, and (iv) the ex-pected weak non-local thermodynamic equilibrium (NLTE) ef-fects (see below), we selected the lines of multiplet 8 for thepresent analysis. It is known that these lines are almost unaf-fected by NLTE e ff ects since they are formed in deep atmo-spheric layers (Korotin 2009). For example, Takeda et al. (2016)and Korotin et al. (2017) have shown that NLTE departuresshould always be smaller than 0.1 dex for our sample stars.For the present analysis of the multiplet 8 lines the atomicdata of Wiese et al. (1969) were adopted and are reported inTable 1. We note that we have found a small systematic biasin the abundances derived from the 674.8 nm line with respectto the two others. This could reflect some possible small uncer-tainties in the atomic data adopted for the three components ofthis line (see Sect. 2.3.3). The proposed correction for the log g f ( + i lines completed by all the molecularand atomic linelists of Heiter et al. (2020), we computed a gridof synthetic spectra around the selected S i lines thanks to theTURBOSPECTRUM code (Plez 2012) and the MARCS modelatmospheres (Gustafsson et al. 2008) under the LTE, 1D, and hy-drostatic assumptions, adopting the Grevesse et al. (2007) solarchemical composition. For this specific sulfur grid, we followeda similar, but slightly updated, procedure as in de Laverny et al.(2012). The ranges of the atmospheric parameters are 4,000 ≤ T e ff ≤ + ≤ lo g (g) ≤ + g being in cm / s ), and -5.0 ≤ [M / H] ≤ + Article number, page 2 of 14. Perdigon et al.: The AMBRE Project: Origin and evolution of sulfur in the Milky Way
Table 1.
Adopted multiplet 8 sulfur line data (Wiese et al. 1969). line (nm) χ e (eV) log g f * * * Notes. ( * ) Probably underestimated, see text. above); up to 13 values of [ α / Fe] were considered for each valueof [M / H] depending on the availability of the MARCS models(with steps of 0.1 dex). Therefore, 25,961 MARCS model atmo-spheres with an [ α / Fe] enhancement consistent with that adoptedwhen computing the synthetic spectra were considered. Then,for each combination of these four atmospheric parameters, wecomputed spectra by varying sulfur abundances between -3.0 ≤ [ S / H ] ≤ + ff erentvalues of [S / H]). The adopted sulfur solar abundance is A S = T e ff , log( g ), and [M / H]following the prescription of the
Gaia -ESO Survey (version 2of the GES empirical relation based on microturbulence veloc-ity determinations from literature samples; Bergemann et al., inpreparation). The adopted micro-turbulence velocities vary from0.6 to 4.8 km / s, depending on the stellar parameters. We alsorecall that no stellar rotation is considered when computing thegrid spectra. Finally, the reference AMBRE sulfur grid consistsin about 675,000 spectra, covering the wavelength range from672 to 677 nm with a wavelength step of 0.0005 nm. Our data consists of a collection of about 100,000 ESO archivedspectra from the FEROS, HARPS, and UVES spectrographs, al-ready parametrised within the AMBRE Project (see Worley et al.(2012), De Pascale et al. (2014) and Worley et al. (2016), respec-tively). The number of selected spectra having a S / N higher than20 and a quality flag for their AMBRE parametrisation equal to 0or 1 (i.e. good or very good parametrisation; see the above AM-BRE papers), together with their atmospheric parameter rangesand their mean signal-to-noise ratio is summarised in Table 2.We note that two UVES setups cover the selected S i multiplet 8lines, and are thus identified as two separate spectrographs.In order to be analysed with our GAUGUIN pipeline in ahomogeneous way, the observed spectra were first corrected bytheir radial velocity. Then, the spectral resolutions of the HARPSand FEROS spectra were degraded to the UVES value ( R ∼ , . We adopted a sampling wavelength step of 0.005 nmfor the whole dataset in order to fulfil the Nyquist-Shannon cri-teria. The reference grid spectra was convolved and re-sampledaccordingly.Then, the prepared 99,271 spectra were ingested into ourchemical analysis pipeline. The spectra are first automaticallynormalised by comparing a synthetic and an observed spectrumover a ∼ ∼ i multiplet 8 in Table 1 are then analysed indepen-dently by comparing the observed and reference grid line profilesover domains covering 0.06nm, 0.05nm, and 0.08nm, centred atthe S i The AMBRE-sulfur catalogue that is presented and discussedin the next sections was built as follows. We first note that ourdataset contains, for some stars, a large number of repeated spec-tra (hereafter called ‘ repeats ’). We thus describe below how wederived sulfur abundances from the analysis of several repeats ofthe same star from which up to three distinct lines can be mea-sured.
Gaia
DR2 and adopted ID
The AMBRE spectra were collected with quite di ff erent instru-ments. The available spectra may therefore contain heteroge-neous names of targets and accuracy of coordinates. The firststep was thus to identify the corresponding observed stars, andin particular the identification of the spectra belonging to thesame star. For this purpose we made use of the Gaia
DR2 cata-logue (Gaia Collaboration et al. 2018) and adopted, when found,the
Gaia
DR2 ID. We refer to a forthcoming article for the de-tailed presentation of this cross-match between the
Gaia / DR2and the AMBRE catalogues. Briefly, this cross-match was per-formed using the stellar spectra coordinates and di ff erent checksbetween the derived AMBRE atmospheric parameters, T e ff es-timated from ground based photometry (2MASS and APASS,Skrutskie et al. 2006; Henden et al. 2018), and Gaia data (e.g. G-magnitude, B P − R P colours, T e ff , radial velocities, but for moredetails, see also Santos-Peral, 2021, submitted). We were ableto identify 5,076 distinct stars from the 99,271 AMBRE spec-tra. There is also a significant fraction ( ∼ Gaia
DR2 ID were found. Several of these spectraactually correspond to bright stars, absent from the
Gaia / DR2catalogue. In this case, we simply looked in Simbad for starshaving coordinates similar to the AMBRE coordinates within aradius of 10” and consistent parametrisation. We then adoptedfor them the corresponding name in the
Henry Draper catalogueas an ID, adding about 200 more stars into the initial sample.Finally, among the identified stars, a significant part has sev-eral repeats : about ∼
20% of the stars have more than ten re-peats, and more than ten stars have more than 1,000 associatedspectra. Such a large number of repeats allowed us to derive sul-fur abundances with very low internal uncertainties (see below).
For a given star, we decided to only keep the repeats with avery consistent set of stellar parameters among each other. Wetherefore rejected the spectra that depart too much from the me-dian values of all repeats. The threshold was arbitrarily chosento be 1 km / s, 100 K, 0.5, 0.10 dex, and 0.05 dex for Vrad, T e ff ,log( g ), [M / H], and [ α / Fe], respectively. A unique threshold canbe adopted for any spectrum since we recall that the AMBREparametrisation was performed at a constant spectral resolutionfor the three considered ESO spectrographs. We also note that
Article number, page 3 of 14 & A proofs: manuscript no. sulfur_article
Table 2.
Summary of the parameter ranges covered by our selected AMBRE spectra having a signal-to-noise ratio higher than 20.
Spectrograph Number of spectra T e ff (K) log( g ) ( g in cm / s ) [M / H] (dex) [ α / Fe] (dex) < S / N > σ (S / N)HARPS 88,178 [4015, 7620] [1.02, 4.95] [-3.43, 0.61] [-0.39, 0.59] 74 29FEROS 5,821 [4000, 7623] [1.00, 4.99] [-3.49, 0.95] [-0.38, 0.65] 98 52UVES / Red580 3,533 [3668, 7575] [1.10, 5.00] [-3.47, 0.72] [-0.33, 0.50] 192 89UVES / Red860 1,739 [3584, 7787] [0.00, 4.82] [-3.49, 0.75] [-0.39, 0.79] 157 51such a rejection procedure could help to reject possible spectro-scopic binaries for which their stellar parameters may seem tovary between di ff erent epochs of observation.On the other hand, the GAUGUIN pipeline was not able insome cases to derive a useful sulfur abundance. This directly re-sults from either a low spectrum quality (photon noise and / orcosmic rays) or from some atmospheric parameter limitations inthe synthetic grid. For example, in the present analysis the refer-ence grid is only valid for low-rotation stars. However, the AM-BRE parametrisation provides an indication of the line broaden-ing (and hence of the rotational velocity, among other broaden-ing mechanisms) thanks to the FWHM of the cross-correlationfunctions derived during the radial velocity measurement. There-fore, in order to reject stars whose rotational and / or macroturbu-lent velocities were possibly too high, we systematically rejectedall the spectra with a FWHM
CCF >
15 km / s for the three spectro-graphs, in agreement with previous estimates (see e.g. Guiglionet al. 2016). This value corresponds to rotational velocities typi-cally lower than ∼ / s (depending on the stellar types) atthe working spectral resolution.Moreover, for each analysed spectrum and detected sulfurlines, we also computed the lowest abundance ( upper limit )that could be estimated from their atmospheric parameters andsignal-to-noise ratio. We systematically rejected all the derivedabundances that were lower than or too close to twice this upperlimit .Finally, all these di ff erent criteria led to the rejection of about2 / We recall that our final working sample consists of 5,275 dis-tinct stars, of which the vast majority possess repeat spectra;each of them could have up to three measured lines. Our methodfor determining the sulfur abundance of a given star consistsof two steps: (i) determining separately the mean abundance ofthe three sulfur lines for the available repeats and (ii) estimat-ing the final sulfur abundance of each star by averaging up tothree available mean individual abundances from the previousstep. In both stages the same averaging method is employed,following the work of Adibekyan et al. (2015, hereafter A15).These authors investigated di ff erent methods of combining abun-dances extracted from di ff erent lines of a given element, andwe adopted their weighted mean ( WM ) procedure to estimateour sulfur abundances. We briefly describe below the adoptedmethodology, and refer to A15 for more details.For a set of N sulfur abundances [S / H] i the adopted W M is[S / H] = (cid:80) Ni = W i [ S / H ] i (cid:80) Ni = W i , (1)with W i being w ei g hts defined below. As shown by A15, this WM procedure has the advantage of successfully removing the e ff ect of outliers on the final abundance, without using any adhoc sigma-clipping procedure. The definition of the weights isbased on the distance (in terms of standard deviation, std ) be-tween the available abundances and their median ( med ): W i = dist i ; dist i = [ S / H ] i − med { [ S / H ] } std { [ S / H ] } . (2)In practice, if more than half of the repeats have the same abun-dance value, and if the abundance distribution is Gaussian, theassociated median of the absolute deviation (MAD), and hencethe standard deviation, could be equal to zero, resulting in an in-finite weight. To circumvent this e ff ect we decided, as suggestedby A15, to bin the dist i in boxes having a width equal to 0.5. Forexample, all the abundances with dist i ≤ . / . = W M abun-dance. The final sulfur abundance (step (ii) above) estimatedwhen more than one sulfur line was measured is also obtainedwith the same procedure. However, we note that if only twoS i lines are available, we provide their W M final abundanceonly if an abundance of the strongest S i / H] only if itcomes from the 675.7 nm component. Finally, each line abun-dance is associated with an error ∆ [ S / H ], proportional to the MAD amongst the repeats of the line. We adopted ∆ [ S / H ] = . × MAD ( { [ S / H ] i } ), i.e. the scaled MAD (corresponding toa 1 σ threshold for a Gaussian distribution).We note that the atmospheric parameters (including the S / N)associated with a given star have been averaged by adoptingthe same
W M procedure. The mean dispersions associated withthese means are equal to 6 K, 0.01, 0.006 dex, and 0.004 dex for T e ff , log( g ), [M / H], [ α / Fe], respectively. We also specify that thefinal sulfur abundance mean was performed after the 674.8 nmline was corrected for a systematic bias of -0.08 dex in [S / H]that could be associated with the non-calibrated line data (seeSect. 2.1). We validated our whole analysis procedure (includ-ing the line atomic data) by estimating the solar sulfur abun-dance derived thanks to a very high-S / N HARPS spectrum ofVesta and the solar FTS spectra of Wallace et al. (2011). For bothspectra degraded at R ∼ , / H] = -0.04 dexafter correcting the 674.8 nm line. Such a small bias was alsoseen in other reference stars (assuming [S / α ] = W M procedure, adopting the same weights as those adoptedfor the final abundance. If no repeats are available for somestars, the final errors were estimated from the dispersion betweenthe abundance of the accepted individual lines. Therefore, no dis-persions are reported for stars having only one available spectrain which only one line has been measured (the S i Article number, page 4 of 14. Perdigon et al.: The AMBRE Project: Origin and evolution of sulfur in the Milky Way
Several sources of uncertainties can a ff ect an abundance deter-mination. Moreover, looking at their di ff erent significances canhelp to flag the reported abundances and clean the sample forfuture scientific exploitation.Firstly, we estimated the sensitivity of the derived sulfurabundances due to possible uncertainties on the stellar atmo-spheric parameters. Table 3 presents the mean variations of [S / H]when considering typical AMBRE uncertainties ( external er-rors estimated by comparison with external catalogues) on T e ff ,log( g ), and [M / H] for the di ff erent stellar types contained in oursample. We note that typical uncertainties on the microturbu-lence velocity ( ± / s) and [ α / Fe] ( ± ff ect on [S / H]. It can be seen that the sulfur abundancesare mostly sensitive to the T e ff uncertainties, and this e ff ect isstronger for the coolest stars in which sulfur lines are weaker.The reported total uncertainties on the sulfur abundances havebeen estimated by summing quadratically the di ff erent contribu-tions.Then, since we were able to analyse several spectra of thesame stars ( repeats ) spanning a wide range of S / N, we checkedthe robustness of our automatic procedure for deriving sulfurabundances. The AMBRE large sample of repeat spectra al-lowed us to precisely quantify our typical internal uncertaintiesthat could result from several e ff ects (e.g. continuum normali-sation, spectra quality, radial velocity correction, di ff erences inthe atmospheric parameters). We list in Table 4 the internal er-rors for di ff erent stellar types. For a given line and a given stel-lar type it can be seen that the sulfur abundance scatter fromspectra to spectra is already very small at rather low S / N, andcan even be smaller for higher S / N. The robustness of our au-tomatic procedure is therefore confirmed. Moreover, we notethat our final sulfur abundances are computed by averaging the repeated measurements and up to three sulfur lines, when avail-able. Consequently, the relative uncertainties from star to starin the AMBRE-sulfur catalogue are therefore even smaller thanthe values given in Table 4. In any case, they can be neglectedwith respect to other error sources as errors caused by possibleatmospheric parameter uncertainties similar to those reported inTable 3.
3. AMBRE catalogue of sulfur abundances
The final AMBRE catalogue of mean sulfur abundances is re-ported in Table 5; a full version is available in electronic form.Having applied all the rejection criteria of Sect. 2.3.2 ( repeat spectra with departing stellar parameters, high-rotating stars, toolow S / N spectra and non-detected lines) and the above averagingprocedures, we finally provide the sulfur abundances with the as-sociated dispersion for 1,855 di ff erent stars. Among them, about10% are giant stars and the studied metallicity domain variesfrom ∼ -1.9 dex to ∼ + ff au et al. (2017). The AMBRE-sulfur catalogue is therefore the largest ever published. More-over, it covers a metallicity range larger than any other, particu-larly in the metal-poor regime since almost 30 stars have [M / H] < -1.0 dex.In Table 5 the stars are identified by their Gaia
DR2 ID, ex-cept for six that only have a HD name. This table contains themean atmospheric parameters ( T e ff , log( g ), [M / H], [ α / Fe]) to- gether with the mean S / N of the spectra kept when computingthe mean final sulfur abundances. We recall that these parame-ters were averaged by adopting exactly the same W M procedureas for the abundances (see previous section). For a given star wealso provide the number of analysed sulfur lines for each com-ponent of multiplet 8 ( N , N , N ) that have finally beenconsidered when computing its sulfur abundance. These num-bers range from unity (only one spectrum available) to severalthousands when a large number of repeat spectra were kept. Thelargest numbers of individual abundances derived for a given starare 5 194, 4 616, and 5 636 for the three sulfur components, re-spectively. We also report 1 165, 1 426, and 1 855 stars with atleast one measurement of the 674.3, 674.8, and 675.7 nm line,respectively. Moreover, 1,049 stars have the three sulfur linesmeasured in at least one of their spectra. Finally, as an indica-tor of the measurement uncertainty, we also report the dispersion( σ [ S / H ] ) between the di ff erent measurements obtained for a givenstar (equivalent to a line-to-line scatter between lines and / or re-peats). We note that this scatter can di ff er from the individualline error ∆ [ S / H ] described in the previous section. The meanof these scatters is equal to 0.06 dex, and a dispersion lowerthan 0.02 and 0.05 dex is found for 29% and 57% of the sam-ple stars, respectively. We can therefore safely conclude that thereported sulfur abundances are very precise and consistent witheach other. The derived sulfur abundances are shown in Fig. 1 asa function of the mean stellar metallicity. For positive metallic-ities it can be seen that the sulfur abundances behave, in a firstapproximation, almost as [M / H]. However, [S / H] becomes in-creasingly higher than [M / H] for negative metallicities down to[M / H] ∼ -1.0 dex, after which [S / H]-[M / H] stays almost constantdown to [M / H] ∼ -2.0 dex. This behaviour is very similar to thatof the α -elements, suggesting that sulfur belongs to this class ofchemical species (see discussion in Sect. 4.1).As a quality check of our derived sulfur abundances, we firstverified that no systematic trends are present between the AM-BRE [S / H] and the e ff ective temperature (over the range 4,500 –6,500 K) or the surface gravity (which varies between 1.5 and5.0 cm / s ). We also compared our complete sample of sulfurabundances (i.e. without selecting the best ones as done below)with those of CS20 estimated from the HARPS spectra of solar-type stars (see Fig. 2). This catalogue contains a very large num-ber of stars (223) in common with ours. This large number isexplained by the fact that both samples contain several HARPSspectra collected over similar epochs (most of them probablyidentical spectra). The median of the di ff erences between theliterature and the AMBRE values of [S / H] is insignificant (-0.027 dex) with a very small associated dispersion, the MADand the standard deviation being equal to 0.04 dex and 0.08 dexrespectively (see Fig. 2). We note that a large part of these smalldi ff erences can be explained by the di ff erent atmospheric pa-rameters adopted in the two studies. For example, the disper-sion associated with the di ff erences in T e ff , log( g ), and [M / H] are63 K, 0.14, and 0.05 dex, respectively. It can be seen, however,that the di ff erences in [S / H] seem to slightly increase towardscooler stars, revealing perhaps some di ff erences in the analysis(possibly caused by di ff erent considerations of some molecu-lar blends?). Moreover, CS20 seem to favour in their discussiontheir stars with e ff ective temperatures within ±
500 K around thesolar value, the range in which their errors are smaller. For thestars in common and having T e ff ≥ Estimated with the AMBRE pipeline during the parametrisation pro-cess, i.e. at a spectral resolution R ∼ , & A proofs: manuscript no. sulfur_article
Table 3.
Sensitivities of the sulfur abundances (in dex) caused by typical uncertainties on the stellar atmospheric parameters.
Cool giant Cool dwarf Solar-type Hot dwarf T e ff ∼ T e ff ∼ T e ff ∼ T e ff ∼ ∆ T e ff = ± ± ± ± ± ∆ log( g ) = ± ± ± ± ± ∆ [M / H] = ± ± ± ± ± ± ± ± ± Table 4.
Typical internal uncertainties (sulfur abundance scatters, index) for di ff erent stellar types and S / N bins. These numbers refer toa single line in one spectrum. Since the final reported abundances areobtained by averaging up to three S i lines and several repeat spectra(when available), the actual relative uncertainties from star to star aremuch smaller. Note: Only the S i Cool giant Cool dwarf Hot dwarfS / N ∼ ∼ ∼ ∼ ∼ ∼ i i i − . − . − . − . . . . [M/H] (dex) − . − . − . − . . . . [ S / H ] ( d e x ) T e ff ( K ) Fig. 1.
AMBRE-sulfur abundances [S / H] as a function of the mean stel-lar metallicity [M / H] for the whole sample. Error bars are the line-to-line dispersions listed in Table 5. The stars without error bars are thosewith only one measured sulfur line (the 675.7 nm transition), hence noline-to-line scatter can be estimated. to 0.03 dex and 0.06 dex, respectively (the median of the dif-ferences staying insignificant). As a consequence, we can safelyconclude that the agreement between the two independent analy-ses is very satisfactory, and this confirms the high accuracy of theAMBRE catalogue of automatically derived sulfur abundances.Finally, we provide in Table 6 new sulfur abundance of the
Gaia benchmark stars present in our sample (several of themhaving a very large number of available spectra). These FGK-type stars have carefully studied atmospheric parameters de-rived using di ff erent analysis techniques, and they are thus veryhelpful to calibrate and / or validate large spectroscopic surveys.Although these stars are already present in Table 5, their AM-BRE atmospheric parameters may di ff er slightly from the com-monly accepted values. We therefore recomputed their sulfurabundances from all their available spectra, assuming for each ofthem the accepted parameters summarised in Jofré et al. (2018)(where the adopted [ α / Fe] is the mean of the individual Mg, Si,Ca, and Ti abundances). We note that, for the Sun, we analyseda very high-S / N HARPS spectrum of Vesta. These new [S / H] T eff AMBRE (K) − . − . . . . ∆ [ S / H ] ( d e x ) MAD = 0.04 dex − . − . − . − . − . − . . . [ M / H ] A M B R E ( d e x ) Fig. 2.
Comparison between the AMBRE and Costa Silva et al.(2020) sulfur abundances. The ordinate axis refers to the di ff erence([S / H] AMBRE - [S / H] literature ). The blue horizontal line indicates the me-dian of these di ff erences (-0.027 dex), and the associated median abso-lute deviation is shown in the upper left corner. This MAD is equal to0.03 dex if we only consider stars warmer than 5400 K (see text). values in Table 6 di ff er only marginally (by a few hundredths ofdex) from those of Table 5 and this only results from the smalldi ff erences in the adopted atmospheric parameters. These newabundances should be favoured in any future calibration or vali-dation studies of sulfur abundances.
4. Behaviour and evolution of sulfur in the MilkyWay
In this section we analyse the catalogue presented above by se-lecting stars with the best measured sulfur abundances. For thispurpose we use the following: – For [M / H] ≥ -1.0 dex: We selected stars having a sulfur abun-dance dispersion σ [ S / H ] lower than 0.05 dex (i.e. those hav-ing at least two lines measured consistently). – In the same metallicity regime: We kept the other stars hav-ing only the 675.7 nm component, but measured in a high-quality spectrum (S / N > – For the metal-poor regime ([M / H] < -1.0 dex): Because fewerstars are available, we slightly relaxed these strict criteria byadding any stars having spectra with S / N higher than 50.The resulting subsample consists of 1,203 stars with very high-quality sulfur abundances (65% of the whole AMBRE-sulfurcatalogue, see top panel of Fig.3).We also built a golden sample by selecting stars with evenbetter sulfur abundances and satisfying the following criteria: – They should have at least three measurements of each indi-vidual S i line (i.e. with at least nine abundances available to Article number, page 6 of 14. Perdigon et al.: The AMBRE Project: Origin and evolution of sulfur in the Milky Way
Table 5.
AMBRE catalogue of LTE sulfur abundances.
Star ID a S / N T e ff (K) log( g ) [M / H] [ α / Fe] N N N [S / H] σ [ S / H ] Notes.
The full electronic table is available at the CDS. ( a ) Gaia
DR2 ID except HD-name for six very bright stars not present in the
Gaia second data release.
Table 6.
AMBRE LTE sulfur abundances of the
Gaia benchmarks stars adopting their recommended atmospheric parameters.
Star Gaia DR2 or HD S / N T e ff (K) log( g ) [M / H] [ α / Fe] N N N [S / H] σ [ S / H ] Solar-type stars α Cen B HD128621 45 5231 4.53 0.22 0.04 39 239 1013 0.25 0.25 τ Cet 2452378776434276992 89 5414 4.49 -0.49 0.23 615 1406 2505 -0.38 0.03Sun / Vesta — > α Cen A HD128620 80 5792 4.31 0.26 -0.02 11 12 17 0.17 0.0118 Sco 4345775217221821312 79 5810 4.44 0.03 0.02 1854 1863 2281 0.00 0.04HD 22879 3250489115708824064 124 5868 4.27 -0.86 0.33 12 17 20 -0.57 0.01 µ Ara 5945941905576552064 62 5902 4.30 0.35 0.00 626 1482 2849 0.14 0.02
FGK subgiants δ Eri 5164120762332790528 76 4954 3.76 0.06 0.04 0 18 275 0.12 0.07 β Hyi 4683897617108299136 102 5873 3.98 -0.04 -0.02 2601 1162 2762 -0.17 0.01 β Vir 3796442680947600768 89 6083 4.10 0.24 -0.13 236 204 249 0.12 0.01
F dwarfs
Procyon HD61421 128 6554 4.00 0.01 -0.04 2662 4844 5681 0.00 0.01HD 49933 3113219383954556416 105 6635 4.20 -0.41 0.04 300 164 470 -0.49 0.03
Cool giant (cid:15)
Vir 3736865265439463424 64 4983 2.77 0.15 -0.07 0 2 5 0.12 0.04
Notes.
The electronic table is available at the CDS. estimate their mean sulfur abundance) and a total dispersion σ [ S / H ] lower than 0.05 dex for their mean [S / H]. – We also included metal-poor stars ([M / H] < -1.0 dex) havingfewer measured lines, but spectra with a S / N higher than 150.This g olden sample contains 540 stars with extremely high-quality and precise sulfur abundances. They are shown in thethree bottom panels of Fig.3. The properties of these best starsare described in more detail below, but the same conclusionsare reached with the largest sample of 1,203 stars, although theslightly larger dispersion could slightly blur some of the fig-ures shown in the following. Finally, we discuss the sulfur abun-dances with respect to the mean metallicity ([S / M]) obtained bysubstracting the [S / H] estimated in the present work from themean metallicity [M / H] previously derived within the AMBREProject ([Fe / H] abundances estimated from individual iron linesbeing not available for the whole sample). α -element Because of its nucleosynthesis channel, sulfur belongs to thefamily of the α -elements, as do oxygen and magnesium, for ex-ample. For these chemical species Milky Way evolution modelspredict a specific variation of the abundance ratio of α -elementsto iron ([ α / Fe]) with respect to metallicity [M / H]. This is ob-servationally confirmed for most α -species, including sulfur to alesser extent (see the Introduction for references). However, thepresent AMBRE-sulfur catalogue o ff ers the possibility to drawa more global and homogeneous picture of the sulfur abundancevariations in the Milky Way than the previous studies, thanks toits large statistics, accurate measurements, and large metallicityrange covered.As already shown in Fig. 1 for the complete sample, andmore clearly shown in Fig. 3 for the high-precision and golden samples exhibiting our very best sulfur abundances, the abun-dance ratio of sulfur to mean metallicity ([S / M]) exhibits a vari-ation with respect to the mean metallicity that is very similar tothat of the mean [ α / Fe] ratio provided by the AMBRE Project
Article number, page 7 of 14 & A proofs: manuscript no. sulfur_article − . − . − . . . . . . . [ S / M ] ( d e x ) [M/H] (dex) − . − . − . . . . . . . [ S / M ] ( d e x )
540 stars [M/H] (dex) − . − . − . . . . . . . [ α / F e ] ( d e x ) − . − . − . − . − . − . − . − .
25 0 .
00 0 .
25 0 . [M/H] (dex) − . − . − . − . . . . . . [ S / α ] ( d e x ) MAD=0.05 dex
Fig. 3.
Ratio of sulfur abundance to mean metallicity [S / M] as a func-tion of the mean stellar metallicity [M / H] for the best derived abun-dances. Top two panels: Dispersion smaller than 0.05 dex and / or high-S / N spectra (top panel) and the same criteria plus a selection of starshaving at least three measurements of the three individual S i lines( golden sample, second panel from top). Bottom two panels: Behaviourof [ α / Fe] and [S / α ] vs. mean metallicity for the golden sample. (third panel of Fig. 3). The dispersion is smaller for [ α / Fe] be-cause it is derived from a much larger number of lines belong-ing to di ff erent chemical species than the maximum of three sul-fur lines studied in this work. This similar behaviour betweensulfur and α -elements is also clearly illustrated in the bottompanel of Fig. 3 where the mean [S / α ] value is equal to zerowith a very small dispersion over the whole metallicity range(MAD = / H] < ∼ -1.0 dex) can be studied thanks to the 27 stars analysed in thiswork (only those of the g olden sample are shown in Fig. 3).This number of stars is much larger than the numbers reportedby other large surveys of sulfur abundances (only two stars in − . − . − . − . − .
25 0 .
00 0 .
25 0 . [M/H] (dex) − . − . . . . [ S / M g ] ( d e x ) MAD = 0.03 dex
Fig. 4.
Ratio of sulfur to magnesium abundance [S / Mg] as a function ofthe mean stellar metallicity [M / H] for stars in common with Santos-Peral et al. (2020). The blue horizontal line indicates the median of[S / Mg] (-0.05 dex) over the whole metallicity domain and the associatedmedian absolute deviation is reported in the upper left corner. − . − . − . − . . . . . . [ S / M g ] ( d e x )
367 stars -0.20-0.100.000.100.200.300.40 [ S / M ] ( d e x ) − . − . − . − . − . − . − . − .
25 0 .
00 0 .
25 0 . [M/H] (dex) − . − . − . − . . . . . . [ S / E u ] ( d e x )
264 stars -0.20-0.100.000.100.200.300.40 [ S / M ] ( d e x ) Fig. 5.
Abundance ratios of sulfur to magnesium ([S / Mg], top panel)and to europium ([S / Eu], bottom panel) as a function of the mean stel-lar metallicity [M / H] colour-coded with [S / M], adopting Mg and Euabundances derived by Mikolaitis et al. (2017, top panel) and Guiglionet al. (2018, bottom panel).
CS20 and about half a dozen in Du ff au et al. 2017). It shouldbe noted that the mean [S / M] in the metal-poor regime is foundat + α -elements. The variation of[S / M] with metallicity for [M / H] < ∼ -1.0 dex is compatible witha flat behaviour, as confirmed by a locally weighted scatterplotsmoothing (LOWESS) fit of the data (Cleveland 1979). How-ever, a rather large dispersion (close to 0.15 dex) around thismean can be seen, probably caused by the di ffi culty in mea-suring the sulfur lines in this low-metallicity domain in spectrawith S / N that can be as low as 50 for the high-precision sam-
Article number, page 8 of 14. Perdigon et al.: The AMBRE Project: Origin and evolution of sulfur in the Milky Way ple. Nevertheless, our study therefore invalidates the suggestionof a steady increase in [S / M] with decreasing metallicities (seediscussion in the Introduction). However, the ratio [S / α ] seemsto become slightly positive (although with a large dispersion;see Fig. 3 bottom panel) for these low metallicities: the mean[S / α ] for [M / H] < ∼ -1.0 dex is close to 0.07 dex with a dispersionMAD = ff erentchemical species adopted to derive the α -abundances could dif-fer from one metallicity regime to another. As a consequence[ α / Fe] could more or less correlate with [Mg / Fe], for example,depending on [M / H]. We note, however, that the dispersion in[ α / Fe] looks larger than in [S / M] in this low-metallicity regime.Such a dispersion could therefore reveal the heterogeneity natureof the Galactic halo.We now focus on the supersolar metallicity regime([M / H] > ∼
0) where the values of [S / M] and [ α / Fe] shown in Fig.3 become negative. This decrease is extensively discussed inSantos-Peral et al. (2020) for magnesium and, independently,clearly seen again here for sulfur. This continuous decrease is notseen in the sulfur sample of CS20, and not always seen in otherGalactic studies of [ α / Fe] behaviours. We recall that Santos-Peral et al. (2020) show that the flattening reported by some pre-vious studies is an artefact created by an incorrect continuumnormalisation procedure in crowded-line spectra of very metal-rich stars. We also note that this decrease for sulfur and mag-nesium is in perfect agreement with Galactic evolution models.There is no reason why the production rate of α -elements (andsulfur) would suddenly increase around solar metallicity and / orwhy the iron production would be constant or smaller to producean almost constant [ α / Fe] ratio at high metallicity. For instance,Palla et al. (2020) predict an [ α / Fe] ∼ -0.2 dex at [M / H] ∼ + α -elements Mg and S show the same overallbehaviour with metallicity. The median [S / Mg] ratio over thewhole metallicity range is close to -0.05 dex with an extremelysmall dispersion (MAD = / H] < -0.25 dex and [M / H] > -0.1 dex(with similar small dispersions), respectively. We also note that,for these metallicity regimes, the median of [ α / Mg] is equal to-0.05 dex and 0.0 dex, respectively, leading to the slightly nega-tive mean [S / α ] ratio shown in Fig. 3 when [M / H] (cid:29) -1.0 dex.Such [S / Mg] ratios could be caused by di ff erent calibrationsadopted for the abundance derivations (although the proceduresadopted by these studies are very similar), but could also be realsince the production rates (yields) of these two elements couldbe slightly di ff erent. Our slightly negative [S / Mg] ratio in themetal-poor regime seems to di ff er from the positive [S / Mg] valuepredicted by Kobayashi et al. (2020, Fig.5) and Prantzos et al.(2018, Fig.13), whereas the agreement seems much better formetal-rich stars. Finally, the variation of [S / Mg] with metallicity(higher at higher [M / H]) could suggest that these two elementsdo not vary in perfect lockstep during the whole Galactic evolu-tion (but see Fig. 7, middle panel). We compare in the top panel of Fig.5 the sulfur-to-magnesium abundance ratio with respect to the metallicityadopting the magnesium abundances from the study of Miko-laitis et al. (2017). We considered the best derived magnesiumabundances of this work by rejecting those with too large errors,and found 367 stars in common. We also show in the bottompanel of Fig. 5 a similar plot showing the sulfur-to-europiumabundance ratio thanks to data derived by Guiglion et al. (2018).Here, we selected the best Eu abundances derived from at leasttwo lines and having a small dispersion (264 stars are found incommon with the AMBRE-sulfur sample). From these two plotsit can be seen that sulfur, magnesium, and europium are closelycorrelated and follow a similar behaviour: the abundance ratiosseem to be rather constant with the metallicity. More precisely,the top panel of Fig. 5 is very similar to Fig. 4, although the dis-persion is larger, probably resulting from the di ff erent method-oloy adopted by Mikolaitis et al. (2017) for deriving Mg abun-dances. Regarding the [S / Eu] ratio, it also seems to stay constantwith metallicity, although sulfur could be slightly underabundant(by about ∼ α nature of sulfur is clearly con-firmed by the AMBRE Project over a very large metallicity do-main. Thanks to Gaia / DR2 astrometry (Gaia Collaboration et al. 2018)and associated distances from Bailer-Jones et al. (2018), wecomputed the Galactic cartesian coordinates of our sample stars.As expected, our sample is predominantly composed of stars lo-cated in the solar vicinity: about 90% of them are found within200 pc of the Sun. Moreover, except for a few cases that arefound at a distance from the Galactic plane up to ∼ g alp y code Bovy (2015). We re-fer to Santos-Peral et al. (2021) for a detailed description of themethodology adopted for computing these orbits.The complex chemo-dynamical characteristics of the discstellar populations are still a matter of debate, and are constantlybeing updated thanks to progressively more complete samplesinside and outside the solar neighbourhood. It is today largelyadmitted that the Galactic disc presents a chemo-dynamical bi-modality, usualy described as the combination of two compo-nents: the thin and the thick disc. Since the discovery of thisbimodality in stellar density distributions (Yoshii 1982; Gilmore& Reid 1983) several studies have shown the kinematical (e.g.Bensby et al. 2003; Reddy et al. 2006; Kordopatis et al. 2017)and chemical (e.g. Adibekyan et al. 2012; Recio-Blanco et al.2014; Hayden et al. 2015) distinctions between the two compo-nents. The thick-disc presents a larger scale height and a shorterscale length, and it is kinematically hotter with respect to thethin-disc. It is also reported to be [ α / Fe] enhanced with respectto the thin disc at all metallicities. Finally, the stellar age distri-
Article number, page 9 of 14 & A proofs: manuscript no. sulfur_article − . . . . . V φ ( k m / s ) − . . . . . R a p o ( k p c ) [M/H] (dex) − . . . . . R p e r i ( k p c ) [M/H] (dex) − . . . . . E cc e n t r i c i t y − . − . − . − . − . − . − . − .
25 0 .
00 0 .
25 0 . [M/H] (dex) − . . . . . | Z m a x | ( k p c ) [ S / M ] ( d e x ) Fig. 6.
Ratio of sulfur to mean metallicity [S / M] as a function of themean metallicity [M / H] colour-coded (from top to bottom) by the rota-tional velocity ( V φ ), the pericentre and apocentre radii ( R peri and R apo ),the eccentricity, and the maximum height above the Galactic plane( Z max ). bution is older for the thick disc than for the thin disc. Di ff erentevolution models and simulations have tried to interpret the ob-served distributions. Although the Galactic disc evolution is stilla matter of debate, it is generally accepted that the thick discphase corresponds to the early disc component, and that about8 Gyr ago a discontinuity in the disc evolution allowed the for-mation of the thin disc. The study of the disc dichotomy usingsulfur abundances has been hindered until now by the lack ofgood statistics and the low precision of the abundance estimates.In this section we take advantage of our large sample of pre- cise sulfur abundances in the solar neighbourhood to analyse thechemo-dynamical correlations in the [S / M] versus [M / H] plane.We show in Fig.6 how the sulfur abundances of the best sam-ple defined earlier ( golden sample, see Fig.3) are related to someof their stellar orbital properties such as the pericentre and apoc-entre radii ( R peri and R apo ), the eccentricity ( e ), the rotational ve-locity ( V φ ), and the maximum height above the Galactic plane( Z max ).First, a gap can be suspected between the [S / M]-rich andthe [S / M]-poor populations, although its precise location cannotbe easily defined, as is often the case for the other α -elements.We note, however, that the chemical separation of the two disccomponents seems to depend on the studied α -element (see e.g.Mikolaitis et al. 2017). Nevertheless, the kinematical and dy-namical characterisation confirms that the observed [S / M] dis-persion at a given metallicity is not the result of the abundanceuncertainties, clearly illustrating the above-described disc bi-modality. This is particularly true for stars having metallicitieslower than -0.5 dex. As is the case for other α -species, the thickversus thin disc dichotomy is observed to be blurred at highermetallicities, and is today a matter of debate (Adibekyan et al.2012; Hayden et al. 2017).More specifically, it can be clearly seen in Fig. 6 that metal-poor sulfur-rich stars (MPSR, for metallicities found between[-1.0,-0.5]) have orbital properties that strongly di ff er from themore metal-rich sulfur-poor ones (MRSP): – First, regarding the rotational velocity, two disc componentsare present and they di ff er in V φ and [S / M]. The less en-riched sulfur stars (thin disc) have lower rotational velocitiesclose to the solar velocity. Only the most metal-poor starshave much lower rotational velocities. A gradient of V φ with[M / H] is also seen, as already mentioned for other α -species(see e.g. Recio-Blanco et al. 2014; Kordopatis et al. 2017). – Then, most of the MPSR are found in the inner Galactic re-gions contrarily to the MRSP that are predominantly locatedclose to the solar apocentrer and pericentrer radii. Radial gra-dients in both R peri and R apo can also be suspected (for a givenmetallicity bin), and they are correlated with the [S / M] en-richment. For instance, at a given metallicity (for [M / H] > ∼ -1.0 dex), stars having a smaller apocentre radius are the mostenriched in sulfur (for a similar discussion, see Hayden et al.2017). If thick-disc stars are defined as being more sulfur-rich (see Sect. 4.4), it can be easily seen that the thick disc ismore radially concentrated than the thin disc. – The eccentricity of the MPSR is higher than the that ofthe MRSP, and every star with a metallicity higher than ∼ -1.0 dex is on a quasi-circular orbit ( e < ∼ . / M] ratios are found oneccentric orbits, as is the case for halo stars, with a few ex-ceptions (also detected in Z max ) confirming the wide range ofstellar orbits in this metallicity range. – Finally, the bottom panel of Fig. 6 reveals a settling of thedisc stars for the MRSP, and the MPSR are mostly foundat several hundreds of parsecs above the Galactic plane, asalready described in Hayden et al. (2017).In summary, Fig. 6 shows that the two disc components can bedefined by studying both the kinematic and the sulfur content.This confirms the high quality of the sulfur abundances of theselected stars, and we show in Sect. 4.4 that the thin and thickdisc could also be defined based solely on their sulfur content,as could be done by adopting any other α -elements.Furthermore, Fig. 6 also reveals that all the metal-poor([M / H] < ∼ -1.0 dex) stars of the sample have much lower rota-tional velocities than the more metal-rich stars. These V φ values Article number, page 10 of 14. Perdigon et al.: The AMBRE Project: Origin and evolution of sulfur in the Milky Way − . . . . . [ S / M ] ( d e x ) -1.00-0.80-0.60-0.40-0.200.000.20 [ M / H ] ( d e x ) Stellar Age (Gyr) − . − . . . . [ S / M g ] ( d e x ) -1.00-0.80-0.60-0.40-0.200.000.20 [ M / H ] ( d e x ) Fig. 7.
Ratio of sulfur abundance to mean metallicity [S / M] (top panel)and to magnesium abundance [S / Mg] (bottom panel) as a function ofthe stellar ages for the main sequence turn-o ff and subgiant stars of thesample. The colour-coding corresponds to the mean stellar metallicity. lower than ∼
150 km / s are typical of halo star members. How-ever, it can also be seen in the bottom panel of Fig. 6 that alarge part of them are located very close to the Galactic plane.This could reveal that these halo stars are presently crossing theplane. Their large number could result from the complex selec-tion function of the AMBRE catalogue revealing a possible biastowards the identification of metal-poor stars in the solar neigh-bourhood. Santos-Peral et al. (2021, submitted) have estimated accurate andreliable ages for about 400 AMBRE stars using an isochrone fit-ting method, as in Kordopatis et al. (2016), and distances fromBailer-Jones et al. (2018, based on Gaia / DR2 parallaxes). Forthis purpose they selected only main sequence turn-o ff and sub-giant stars for which age estimates are more accurate since stellarages increase quickly when stars cross these regions of the HR-diagram. We refer to this article for a detailed description of theadopted methodology to derive the stellar ages adopted in thepresent study.Among our golden sample of stars with the best derived sul-fur abundances, about 10% are in common with Santos-Peralet al. (2021) and have ages with rather small uncertainties (themean of their age relative errors is equal to 17% with a disper-sion of 9%). The variation of [S / M] and [S / Mg] abundance ratiosas a function of the stellar ages and colour-coded with the meanmetallicity are shown in Fig. 7. The following can be seen: – The [S / M] ratio (top panel) of disc stars decreases towardsyounger and more metal-rich stars in a continuous way from ∼
10 Gyr to ∼ / M] val-ues for ages younger than ∼ / M] ver-sus age relation close to ∼ / Gyr for stars youngerthan ∼
10 Gyr (in very good agreement with the gradient re-ported by Santos-Peral, 2021, for magnesium). This contra-dicts some previous claims where most thin-disc stars seemto have almost constant [S / M] ratios with age; for instance,the slope in CS20 is two times smaller. However, the be-haviour shown in Fig. 7 is easy to interpret: As expectedby chemical evolution models, the [ α / Fe] content of stars inthe supersolar metallicity regime is believed to continue todecrease with time (the production of α -species by SNII isstrongly reduced with respect to that of iron-peak elementsin SNIa). More explicitly, younger metal-rich stars shouldalways have lower [ α / Fe] (and hence [S / M]) ratios, as pre-dicted by any chemical evolution model of the Galactic discs(for a recent reference, see Fig. 9 in Palla et al. 2020). – The stars older than ∼ / M] withage is much faster and steeper. Moreover, all the oldest starsin our sample, which are the most metal-poor, are enriched insulfur. Once again, this is in complete agreement with chem-ical model predictions. – It can also be seen in the bottom panel of Fig. 7 that sulfurand magnesium abundances stay close to each other what-ever the stellar age (but see the above discussion on this[S / Mg] ratio at di ff erent metallicities). This could reveal thatthere is no important variation in the yields of these twospecies with time, and could again confirm their commonorigin in SNII explosion.As a conclusion of the present subsection, it can be said thatsulfur can be considered a good chemical clock, like any other α -species, although some dispersion may be present (see e.g.Santos-Peral et al. 2021, submitted). A dichotomy in the [ α / Fe] abundances, associated with the thindisc–thick disc bimodality, has been found in the Galactic discstellar populations (see e.g. Adibekyan et al. 2011; Recio-Blancoet al. 2014). To date, such a chemical separation has been foundeither using an averaged [ α / Fe] index (see above references), butalso using di ff erent individual α -species such as magnesium (e.g.within the AMBRE context Mikolaitis et al. 2017; Santos-Peralet al. 2020) or other individual [ α / Fe] ratios (Mikolaitis et al.2014, among others). However, such a chemical dichotomy hasnever been established using only sulfur abundances. We notethat in the very recent study of CS20 the thin–thick disc separa-tion was first defined thanks to a global [ α / Fe] abundance ratiosand then applied to and / or checked against their sulfur content.We decided to follow an opposite approach by looking for a pos-sible definition of the Galactic components based purely on theirsulfur abundances. We indeed recall that, as already mentionedin Sect. 4.2, the dichotomy of the two disc components is clearlyseen in kinematics and is correlated to the sulfur content. Thisapproach is also favoured since the thin–thick disc separationis expected to slightly di ff er from one α -species to another, as Article number, page 11 of 14 & A proofs: manuscript no. sulfur_article − . − . − . . . . . . . [ S / M ] ( d e x ) ThinThickSRMRHalo [M/H] (dex) − . − . − . . . . . . . [ M g/ M ] ( d e x ) − . − . − . − . − . − . − . − .
25 0 .
00 0 .
25 0 . [M/H] (dex) − . − . − . . . . . . . [ S / M ] ( d e x ) Fig. 8.
T op : Ratio of sulfur abundance to mean metallicity [S / M] vs.mean metallicity [M / H] for stars with the best measured sulfur abun-dances (large number of analysed spectra, at least three measurementsof the three S i lines available and small dispersion). The green lineshows the adopted separation between the sulfur-rich and sulfur-poorstars. Middle : [Mg / M] ratio vs. [M / H] for the stars in common withSantos-Peral et al. (2020).
Bottom : Same as top panel, but for the com-plete sample of 540 stars with the best derived abundances (same sam-ple as in Fig. 3, second panel), and labelled according to their Galacticpopulation membership. In all these panels thin-disc, thick-disc, sulfur-rich, metal-rich, and halo stars are shown as blue filled circles, red tri-angles, orange squares, and green diamonds, respectively. already shown by Mikolaitis et al. (2014) or more recently byAmarsi et al. (2020). Such di ff erent behaviours for di ff erent α -elements could be real (slightly di ff erent nucleosynthesis chan-nels) or caused by (among other possibilities) di ff erent internaldispersions resulting from di ff erent residual systematics such asnon-LTE e ff ects and / or di ff erent sensibilities of the selected linesbetween dwarfs and giants (the proportion of dwarfs and giantsin either disc being di ff erent due to observational selection func-tions).Defining the sulfur-rich / sulfur-poor separation (associatedhereafter with the α -rich / α -poor or thick-disc / thin-disc Galac-tic populations, respectively) could be useful, among other in-terests, to compute the radial chemical gradients in both discs.For this purpose we first selected our best sulfur abundances by keeping only stars having a large number of analysed spectra.More specifically, we considered 200 stars having more than 20independent measurements of their 675.7 nm line, more than 50analysed lines in total including at least three measurements ofthe three S i lines, and, in addition, an internal dispersion amongall these measurements smaller than 0.05 dex. These best mea-sured stars are shown in the top panel of Fig. 8. We also defineda separation to disentangle the thin and thick discs (green line inFig. 8). It was drawn by examinating the [S / M] distributions in0.2 dex wide metallicity bins for [M / H] varying within [-1.0,0.0].Then, we looked for the position of the low-density regions sep-arating both disc sequences, and drew a straight line along theseminima. This green line was then extrapolated towards super-solar metallicites. Then, this separation was applied to both theentire and the golden samples. The resulting ∼
65% sulfur-poor,thin-disc stars are shown as blue filled circles below the greenline in the bottom panel of Fig. 8. Moreover, for the sulfur-richstars (found above the separation line), a gap in their numberdistribution might be present around [M / H] = -0.2 dex. Such agap was already suspected among α -rich stars around the samemetallicity by Adibekyan et al. (2011), who proposed to callthem high- α metal-rich stars (see also Gazzano et al. 2013). Al-though not clearly seen in sulfur and sometimes absent in otherstudies of α elements, depending on the analysed sample, wenevertheless decided to label such stars separately. This led to ∼
18% sulfur-rich metal-rich stars (hereafter SRMR, and shownas orange squares in Fig. 8) and ∼
15% thick-disc stars as redtriangles. Finally, the stars more metal-poor than -1.0 dex werelabelled as potential Galactic halo members (green diamonds).In confirmation of the above, several remarks can be maderegarding the bottom panel of Fig. 8. First, it can be again seenthat the sulfur abundances of the thin-disc stars and the SRMRstars continue to slowly decrease below [S / M] = / H]becomes positive. Then, most of the thin-disc stars have metal-licity higher than -0.5 dex, although an extension can be seendown to [M / H] ∼ -0.9 dex. The thick-disc stars follow the slope ofthe green line, and are preferentially located ∼ / H] ∼ -2.0 dex.Their halo membership is confirmed by their rather low rota-tional velocities, as shown in the top panel of Fig. 6. Their[S / M] ratio is compatible with a flat behaviour (like any other α -element) and they have a mean [S / M] ratio close to + / sulfur-poor stars are themost magnesium poor, (ii) most of the SRMR stars are alsomagnesium-rich / metal-rich, and (iii) thick-disc / sulfur-rich starstend also to have higher [Mg / M] ratios. There is only one halostar in this subsample, but its sulfur and magnesium enrichmentare consistent. Therefore, although the scatter in sulfur appearslarger than in magnesium, we can conclude that defining theGalactic populations from their [S / M] content would give resultsconsistent with those obtained from most commonly used ratios,for instance [Mg / M].
For the estimation of the Galactic gradient of sulfur for our ∼
400 stars belonging to the thin disc, we considered the guid-ing centre radius ( R g ), computed as the average of the pericen-tre and apocentre of the stellar orbits, as a proxy of the present Article number, page 12 of 14. Perdigon et al.: The AMBRE Project: Origin and evolution of sulfur in the Milky Way star Galactocentric distance. The gradient (and associated un-certainty) was estimated thanks to a Theil-Sen fit of the [S / M]versus R g points, the uncertainty being given by the lower andupper confidence levels of this fit. We find a small positive ra-dial gradient δ [S / M] /δ R g = + ± . dex / kpc in the thin discfor R g within 6 and 10 kpc. This gradient is slightly flatter thanthe [ α / Fe] gradient of 0.012 ± / kpc found between 5and 13 kpc using Gaia-ESO Survey abundances (Recio-Blancoet al. 2014, but no Gaia distances were available at that time)and that of Santos-Peral et al. (2020) estimated for magnesiumover iron for AMBRE stars ( + ± / kpc between 6 et11kpc). However, the Galactic (thin) disc gradient in sulfur hasrecently been estimated from 17 H ii regions with revisited Gaiadistances ranging between ∼ ∼
14 kpc(Arellano-Córdovaet al. 2020). This study reports a radial gradient in [S / H] equalto -0.035 ± . dex / kpc in good agreement within the error barswith our determination: δ [S / H] /δ R g = -0.05 ± . dex / kpc. Wealso note that these authors found a flat gradient for [S / O], con-firming independently the similar nature of these two chemicalspecies, and hence the α nature of sulfur.We also computed the radial gradient in the thick disc be-tween 4.5 and 15.5 kpc: δ [S / M] /δ R g = -0.014 ± . dex / kpc, inagreement within the error bars with the value reported by Recio-Blanco et al. (2014, -0.004 ± ∼
150 stars) would not change this sulfur-to-mean metal-licity radial gradient since we found -0.01 ±− . dex / kpc.
5. Summary
We have presented LTE sulfur abundances derived from the threemain components of the multiplet 8 system lines found around675 nm, which are known to be poorly a ff ected by NLTE ef-fects. This study analysed ∼ / H] abundances varying from -3.0 to + i line has been consideredto derive a mean sulfur abundance per star and an associated dis-persion, based on the line-to-line scatter and the available repeatspectra. The final catalogue contains abundances for 1,855 indi-vidual slow-rotating stars. About 90% of the sample consists ofFGK-dwarf stars and the remaining 10% are cool giants. Theirmean metallicity is between -2.0 to + Gaia benchmarkstars re-estimated by considering their recommended stellar pa-rameters.This AMBRE-sulfur catalogue allowed us to study the ori-gin and evolution of sulfur in the Milky Way. First, we haveshown that sulfur presents behaviours that are close to any other α -element. The mean [S / M] is equal to ∼ / H] < -1.0 dex and the distribution of [S / M] is compatible with aplateau-like behaviour in the low-metallicity regime. Then, amonotonic decline is found with increasing metallicity. More-over, this decline clearly continues for supersolar metallicity(without any slope change), as already independently found forAMBRE magnesium abundances (Santos-Peral et al. 2020). Allof this is also confirmed by the low [S / α ] ratios found over the whole studied metallicity range and by the similar behaviourof our sulfur abundances with metallicity compared to previousmagnesium and europium abundances, both elements being pre-dominantly produced by Type II supernovae (although still dis-cussed for europium).Then, thanks to Gaia DR2 astrometry, we have studied thecorrelation between the AMBRE-sulfur abundances and the stel-lar kinematic and orbital properties. A dichotomy in kinemat-ics and eccentricity is detected between sulfur-rich and sulfur-poor stars at metallicities lower than ∼ -0.5 dex. Two disc com-ponents that could be associated with the thin and the thick discs,with di ff erent sulfur content and kinematical properties are alsoidentified. We have then proposed that the thin-disc / thick-discdichotomy could be defined by solely considering sulfur abun-dances, as done in previous studies by considering any other α -species. Furthermore, a trend with small dispersion between stel-lar ages and sulfur content is found: [S / M] slowly increases withstellar age up to ∼
11 Gyr, whereas the metallicity decreases andthen a much steeper slope appears for older more metal-poorstars. Sulfur could thus be used as a chemical clock, althoughsome dispersion could appear when examining a larger sample.Finally, we have estimated the sulfur radial gradient in the thindisc and found a small positive gradient for δ [S / M] /δ R g consis-tent with previous studies, in particular the gradient derived fromthe sulfur content of H ii regions. The gradient in the thick disc isfound to be slightly smaller.This work therefore proposes that the Galactic chemical his-tory of sulfur is similar to that of a typical α -element, and thatthis chemical species could be adopted with confidence to studyGalactic stellar populations. Acknowledgements.
We are grateful to Robin Bonannini who starts apreliminary study of sulfur abundances with AMBRE data severalyears ago. This work has made use of data from the European SpaceAgency (ESA) mission
Gaia (https: // / gaia), pro-cessed by the Gaia
Data Processing and Analysis Consortium (DPAC,https: // / web / gaia / dpac / consortium). Funding for the DPAChas been provided by national institutions, in particular the institutionsparticipating in the Gaia
Multilateral Agreement. This research has also madeuse of the SIMBAD database, operated at CDS, Strasbourg, France. The authorsacknowledge financial support from the ANR 14-CE33-014-01. This work wasalso supported by the "Programme National de Physique Stellaire" (PNPS) ofCNRS / INSU co-funded by CEA and CNES. M.A.A. acknowledges supportfrom the CIGUS -CITIC, funded by Xunta de Galicia and the European Union(FEDER Galicia 2014-2020 Program) by grant ED431G 2019 /
01. Finally, mostof the calculations have been performed with the high-performance computingfacility SIGAMM, hosted by OCA.
References
Adibekyan, V., Figueira, P., Santos, N. C., et al. 2015, A&A, 583, A94Adibekyan, V. Z., Santos, N. C., Sousa, S. G., & Israelian, G. 2011, A&A, 535,L11Adibekyan, V. Z., Sousa, S. G., Santos, N. C., et al. 2012, A&A, 545, A32Amarsi, A. M., Lind, K., Osorio, Y., et al. 2020, arXiv e-prints,arXiv:2008.09582Arellano-Córdova, K. Z., Esteban, C., García-Rojas, J., & Méndez-Delgado, J. E.2020, MNRAS, 496, 1051Bailer-Jones, C. A. L., Andrae, R., Arcay, B., et al. 2013, A&A, 559, A74Bailer-Jones, C. A. L., Rybizki, J., Fouesneau, M., Mantelet, G., & Andrae, R.2018, AJ, 156, 58Bensby, T., Feltzing, S., & Lundström, I. 2003, A&A, 410, 527Biemont, E., Quinet, P., & Zeippen, C. J. 1993, A&AS, 102, 435Bijaoui, A., Recio-Blanco, A., de Laverny, P., & Ordenovic, C. 2010, in ADA 6- Sixth Conference on Astronomical Data Analysis, 9Bijaoui, A., Recio-Blanco, A., de Laverny, P., & Ordenovic, C. 2012, StatisticalMethodology, 9, 55Bovy, J. 2015, ApJS, 216, 29Ca ff au, E., Bonifacio, P., Faraggiana, R., et al. 2005, A&A, 441, 533Chen, Y. Q., Nissen, P. E., Zhao, G., & Asplund, M. 2002, A&A, 390, 225 Article number, page 13 of 14 & A proofs: manuscript no. sulfur_article
Cleveland, W. S. 1979, Journal of the American Statistical Association, 74, 829Costa Silva, A. R., Delgado Mena, E., & Tsantaki, M. 2020, A&A, 634, A136de Laverny, P., Recio-Blanco, A., Worley, C. C., et al. 2013, The Messenger,153, 18de Laverny, P., Recio-Blanco, A., Worley, C. C., & Plez, B. 2012, A&A, 544,A126De Pascale, M., Worley, C. C., de Laverny, P., et al. 2014, A&A, 570, A68Du ff au, S., Ca ff au, E., Sbordone, L., et al. 2017, A&A, 604, A128Francois, P. 1988, A&A, 195, 226Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al. 2018, A&A, 616, A1Gazzano, J. C., Kordopatis, G., Deleuil, M., et al. 2013, A&A, 550, A125Gilmore, G. & Reid, N. 1983, MNRAS, 202, 1025Grevesse, N., Asplund, M., & Sauval, A. J. 2007, Space Sci. Rev., 130, 105Grisoni, V., Spitoni, E., Matteucci, F., et al. 2017, MNRAS, 472, 3637Guiglion, G., de Laverny, P., Recio-Blanco, A., & Prantzos, N. 2018, A&A, 619,A143Guiglion, G., de Laverny, P., Recio-Blanco, A., et al. 2016, A&A, 595, A18Gustafsson, B., Edvardsson, B., Eriksson, K., et al. 2008, A&A, 486, 951Hayden, M. R., Bovy, J., Holtzman, J. A., et al. 2015, ApJ, 808, 132Hayden, M. R., Recio-Blanco, A., de Laverny, P., Mikolaitis, S., & Worley, C. C.2017, A&A, 608, L1Heiter, U., Lind, K., Bergemann, M., et al. 2020, arXiv e-prints,arXiv:2011.02049Henden, A. A., Levine, S., Terrell, D., et al. 2018, in American Astronomical So-ciety Meeting Abstracts, Vol. 232, American Astronomical Society MeetingAbstracts ff au, E., & Sbordone, L. 2015, A&A, 577, A18Kobayashi, C., Karakas, A. I., & Lugaro, M. 2020, arXiv e-prints,arXiv:2008.04660Kobayashi, C., Karakas, A. I., & Umeda, H. 2011, MNRAS, 414, 3231Kordopatis, G., Amorisco, N. C., Evans, N. W., Gilmore, G., & Koposov, S. E.2016, MNRAS, 457, 1299Kordopatis, G., Wyse, R. F. G., Chiappini, C., et al. 2017, MNRAS, 467, 469Korotin, S., Andrievsky, S., Ca ff au, E., & Bonifacio, P. 2017, in Astronomi-cal Society of the Pacific Conference Series, Vol. 510, Stars: From Collapseto Collapse, ed. Y. Y. Balega, D. O. Kudryavtsev, I. I. Romanyuk, & I. A.Yakunin, 141Korotin, S. A. 2009, Astronomy Reports, 53, 651Luck, R. E. 2015, AJ, 150, 88Matrozis, E., Ryde, N., & Dupree, A. K. 2013, A&A, 559, A115Mikolaitis, Š., de Laverny, P., Recio-Blanco, A., et al. 2017, A&A, 600, A22Mikolaitis, Š., Hill, V., Recio-Blanco, A., et al. 2014, A&A, 572, A33Nissen, P. E., Akerman, C., Asplund, M., et al. 2007, A&A, 469, 319Nissen, P. E., Chen, Y. Q., Asplund, M., & Pettini, M. 2004, A&A, 415, 993Nomoto, K., Kobayashi, C., & Tominaga, N. 2013, ARA&A, 51, 457Palla, M., Matteucci, F., Spitoni, E., Vincenzo, F., & Grisoni, V. 2020, arXive-prints, arXiv:2008.07484Plez, B. 2012, Turbospectrum: Code for spectral synthesis, Astrophysics SourceCode LibraryPrantzos, N., Abia, C., Limongi, M., Chie ffi , A., & Cristallo, S. 2018, MNRAS,476, 3432Recio-Blanco, A., Bijaoui, A., & de Laverny, P. 2006, MNRAS, 370, 141Recio-Blanco, A., de Laverny, P., Allende Prieto, C., et al. 2016, A&A, 585, A93Recio-Blanco, A., de Laverny, P., Kordopatis, G., et al. 2014, A&A, 567, A5Reddy, B. E., Lambert, D. L., & Allende Prieto, C. 2006, MNRAS, 367, 1329Romano, D., Karakas, A. I., Tosi, M., & Matteucci, F. 2010, A&A, 522, A32Ryde, N. 2006, A&A, 455, L13Ryde, N. & Lambert, D. L. 2004, A&A, 415, 559Santos-Peral, P., Recio-Blanco, A., de Laverny, P., Fernández-Alvar, E., & Or-denovic, C. 2020, A&A, 639, A140Savage, B. D. & Sembach, K. R. 1996, ARA&A, 34, 279Schönrich, R., Binney, J., & Dehnen, W. 2010, MNRAS, 403, 1829Scott, P., Grevesse, N., Asplund, M., et al. 2015, A&A, 573, A25Skrutskie, M. F., Cutri, R. M., Stiening, R., et al. 2006, AJ, 131, 1163Spite, M., Ca ff au, E., Andrievsky, S. M., et al. 2011, A&A, 528, A9Takeda, Y., Omiya, M., Harakawa, H., & Sato, B. 2016, PASJ, 68, 81Wallace, L., Hinkle, K. H., Livingston, W. C., & Davis, S. P. 2011, ApJS, 195, 6Wiese, W. L., Smith, M. W., & Miles, B. M. 1969, Atomic transition probabili-ties. Vol. 2: Sodium through Calcium. A critical data compilationWoosley, S. E. & Weaver, T. A. 1995, ApJS, 101, 181Worley, C. C., de Laverny, P., Recio-Blanco, A., Hill, V., & Bijaoui, A. 2016,A&A, 591, A81Worley, C. C., de Laverny, P., Recio-Blanco, A., et al. 2012, A&A, 542, A48Yoshii, Y. 1982, PASJ, 34, 365au, E., Andrievsky, S. M., et al. 2011, A&A, 528, A9Takeda, Y., Omiya, M., Harakawa, H., & Sato, B. 2016, PASJ, 68, 81Wallace, L., Hinkle, K. H., Livingston, W. C., & Davis, S. P. 2011, ApJS, 195, 6Wiese, W. L., Smith, M. W., & Miles, B. M. 1969, Atomic transition probabili-ties. Vol. 2: Sodium through Calcium. A critical data compilationWoosley, S. E. & Weaver, T. A. 1995, ApJS, 101, 181Worley, C. C., de Laverny, P., Recio-Blanco, A., Hill, V., & Bijaoui, A. 2016,A&A, 591, A81Worley, C. C., de Laverny, P., Recio-Blanco, A., et al. 2012, A&A, 542, A48Yoshii, Y. 1982, PASJ, 34, 365