DOOp, an automated wrapper for DAOSPEC
Tristan Cantat-Gaudin, Paolo Donati, Elena Pancino, Angela Bragaglia, Antonella Vallenari, Eileen D. Friel, Rosanna Sordo, Heather R. Jacobson, Laura Magrini
aa r X i v : . [ a s t r o - ph . I M ] D ec Astronomy & Astrophysicsmanuscript no. DOOp_AA_final c (cid:13)
ESO 2018August 3, 2018
DOOp, an automated wrapper for DAOSPEC
Tristan Cantat-Gaudin , , Paolo Donati , , Elena Pancino , Angela Bragaglia , , Antonella Vallenari , Eileen D. Friel ,Rosanna Sordo , Heather R. Jacobson , and Laura Magrini Dipartimento di Fisica e Astronomia, Università di Padova, vicolo Osservatorio 3, 35122 Padova, Italy INAF-Osservatorio Astronomico di Padova, vicolo Osservatorio 5, 35122 Padova, Italy Dipartimento di Fisica e Astronomia, Università di Bologna, Viale Berti Pichat, 6 / INAF-Osservatorio Astronomico di Bologna, via Ranzani 1, 40127 Bologna, Italy ASI Science Data Center, I-00044 Frascati, Italy Department of Astronomy, Indiana University, Bloomington, IN 47405, USA Massachusetts Institute of Technology and Kavli Institute for Astrophysics and Space Research, 77 Massachusetts Avenue, Cam-bridge, MA 02139, USA INAF-Osservatorio Astrofisico di Arcetri, Largo Enrico Fermi 5, 50125 Firenze, ItalyReceived date / Accepted date
ABSTRACT
Context.
Large spectroscopic surveys such as the Gaia-ESO Survey produce huge quantities of data. Automatic tools are necessaryin order to e ffi ciently handle this material. The measurement of equivalent widths in stellar spectra is traditionally done by hand orwith semi-automatic procedures that are time-consuming and not very robust with respect to the repeatability of the results. Aims.
The program DAOSPEC is a tool that provides consistent measurements of equivalent widths in stellar spectra while requiringa minimum of user intervention. However, it is not optimised to deal with large batches of spectra, as some parameters still need tobe modified and checked by the user. Exploiting the versatility and portability of BASH, we have built a pipeline called DAOSPECOption Optimiser (DOOp) automating the procedure of equivalent widths measurement with DAOSPEC.
Methods.
DOOp is organised in di ff erent modules that run one after the other to perform specific tasks, taking care of the optimisationof the parameters needed to provide the final equivalent widths, and providing log files to ensure better control over the procedure. Results.
In this paper, making use of synthetic and observed spectra, we compare the performance of DOOp with other methods,including DAOSPEC used manually. The measurements made by DOOp are identical to the ones produced by DAOSPEC when usedmanually, while requiring less user intervention, which is especially convenient when dealing with a large quantity of spectra. LikeDAOSPEC, DOOp shows its best performance on high-resolution spectra (R >
20 000) and high signal-to-noise ratio (S / N > Key words. techniques: spectroscopic
1. Introduction
We are presently living in an era of large astronomical surveysthat are delivering an unprecedented amount of information, likethe current APOGEE (Apache Point Observatory Galactic Evo-lution Experiment, Allende Prieto et al. 2008) and RAVE (Ra-dial Velocity Experiment, Zwitter 2008) and those to come in thenear future such as GALAH (Galactic Archaeology with HER-MES, Barden et al. 2010), 4MOST (4-metre Multi-Object Spec-troscopic Telescope, de Jong 2011) or MOONS (Multi-ObjectOptical and Near-infrared Spectrograph, Cirasuolo et al. 2011).The Gaia ESO Survey (hereafter GES, see Gilmore et al. 2012)is a public spectroscopic survey that started at the end of 2011,carried out on FLAMES at Very Large Telescope, targeting morethan 10 stars over the course of five years, systematically cov-ering all major components of the Milky Way, from ancient halostars to star forming regions, providing the first homogeneousoverview of the distributions of kinematics and detailed elemen-tal abundances. The data analysis of the GES is a complex taskcarried out by di ff erent groups. When dealing with a huge quan-tity of astronomical data, it is essential to have tools that eco- nomically process large amounts of information and produce re-peatable results.Some automatic spectrum analysis procedures rely onthe minimisation of the χ di ff erence between the observedspectrum and a set of synthetic ones (for instance SME,Valentin & Piskunov 1996), or the projection of the spec-trum on a vector basis constructed from theoretical spectra(Recio-Blanco et al. 2006, MATISSE). Other procedures arebased on the classical method consisting in measuring equiva-lent widths (EWs) that can be analysed with codes like MOOG(Sneden et al. 2012) or WIDTH9 (Kurucz 2005). Equiva-lent widths analysis is widely used, and codes like FAMA(Magrini et al. 2013) or GALA (Mucciarelli et al. 2013) weredeveloped recently.We present here a tool developed to measure the EWs ofa large number of spectra in a fully automatic way. Thistool, called DAOSPEC Option Optimiser pipeline (DOOp), usesDAOSPEC (Stetson & Pancino 2008, hereafter SP08) to mea-sure the EWs and optimises its key parameters in order to makethe measurements as robust and repeatable as possible. The aimof this paper is to describe the main characteristics of DOOp and Article number, page 1 of 9 o show the results that can be achieved when combining DOOpwith an analysis code such as FAMA, or any other abundanceanalysis program based on EW measurements.DOOp, along with a user guide, is avail-able to the community via its webpage: http://web.oapd.inaf.it/GaiaESO/DOOp
2. DAOSPEC in a nutshell
The DAOSPEC code is fully described in SP08, and we willsimply note its main characteristics and underline a few pointsthat are important for the best use of DOOp: DAOSPEC is anautomated Fortran program to measure EWs of absorption linesin high-resolution (typically, higher than 20 000) and high signalto noise (S / N higher than 30) spectra of stellar atmospheres. Themeasured lines are matched with a user-provided line list.The code employs a fixed full width at half maximum (orscaled with wavelength for echelle spectra) to facilitate de-blending, and estimates the continuum with Legendre polyno-mials after all the fitted lines are removed from the spectrum.These two characteristics are not present in codes like SPEC-TRE (Fitzpatrick & Sneden 1987), ARES (Sousa et al. 2007),or EWDET (Ramírez et al. 2001), that all leave the full widthat half maximum as a free parameter for each line, as is com-monly done when measuring EWs with the IRAF task splot .This makes DAOSPEC especially useful on crowded spectra.One important feature of DAOSPEC (see Fig. 4 in SP08) isthat the continuum on which the EW fits are based is not the true continuum of the spectrum (i.e., the continuous star emis-sion after all the lines are excluded), but an e ff ective continuum ,which is the true continuum depressed by a statistical estimateof the contaminating lines (the unresolved or undetected ones,producing a sort of line blanketing). This greatly improves theestimate of the unblended EW of each line in crowded spectra, asdemonstrated in Sect. 3.2.1 of SP08, but it is often perceived asbeing too low by those who are used to employing traditionalinteractive methods for the continuum fitting procedure. Thediscrepancy between the true continuum and the e ff ective con-tinuum increases with line crowding (i.e., spectrum metallicity,especially for giants) and with decreasing S / N or resolution ofthe spectra.Full width at half maximum (FWHM) and continuum place-ment are strictly correlated: if the continuum level of a spectrumis altered, so is the FWHM of each line. This is why the threemost important parameters for a successful use of DAOSPECare: (1) the FWHM estimate; (2) the continuum placement; and(3) the residual core flux parameter, which is the flux at thecore of saturated lines, expressed in percent of the local con-tinuum level; therefore, DOOp is designed to provide the bestfine-tuning of these three parameters.Another characteristic of DAOSPEC, which is uncommonfor EW measurement programs (both interactive and auto-mated), is that it provides a number of quality estimates of eachEW measurement. These are the formal fitting errors in the sin-gle line, the quality parameter Q of each single line (a compari-son of the local residuals around each line with the average resid-uals of the whole spectrum), and the average residuals, expressedas a percentage, over the whole spectrum. The codes mentionedbefore do not, to our knowledge, provide an uncertainty on thefit of individual lines (except for EWDET), and none of themperforms a global estimate of the quality of fit. These featuresof DAOSPEC are used by DOOp to estimate the e ff ects of a sys-tematically incorrect continuum placement, for example. The DAOSPEC code relies on statistical evaluation to con-sistently estimate the FWHM of the lines and place the e ff ectivecontinuum across the whole spectral range, which means that itperforms better on wider ranges than on smaller ones. Whendealing with spectra from an echelle spectrograph that deliversindividual orders, it is safer to use a merged spectrum rather thanmeasuring each order separately. This applies of course also toDOOp.
3. DOO pipeline
The DOOp code is an algorithm which optimises the parametersof DAOSPEC in order to get the best measurements of EWs.The fine tuning of the parameters is obtained through a fullyautomatic and iterative procedure and is tailored to the intrinsiccharacteristic of the spectrum that is going to be analysed. Thisprocedure is performed by di ff erent scripts written in BASH and IRAF built around DAOSPEC.The DAOSPEC parameters on which DOOp focuses are thefollowing: • short wavelength limit (SH) • long wavelength limit (LO) • minimum radial velocity (MI) • maximum radial velocity (MA) • residual core flux (RE) • FWHM (FW)An exhaustive description of these parameters can be foundin SP08 or in the DAOSPEC manual publicly available . Herewe describe briefly their meaning. The SH and LO parametersspecify the spectral range over which DAOSPEC will measureequivalent widths. The MI and MA parameters set the veloc-ity range in which DAOSPEC is allowed to estimate the radialvelocity (RV) of the star. Imposing a restricted range of possi-ble RV through MI and MA reduces the risk of mismatching thelines (for instance in spectra with very few or very broad lines)and helps to find the right value. It also reduces the computationtime. The RE parameter tells the program the residual flux at thecore of the deepest line in the spectrum. The FW sets the esti-mate of the resolution of the spectrum in units of pixel. All theother DAOSPEC parameters are set to default values but some ofthem must be specified in the input file of DOOp (see Sect. 3.2).Among these, the most important is the order of the poly-nomial (OR) used to fit the continuum. This parameter is not optimised by DOOp, and is kept fixed to the value provided bythe user. The OR parameter should be chosen with care and inSect. 3.1 we discuss its importance and its impact on the EWmeasurement. The choice of continuum fitting when analysing a stellar spec-trum is always of great importance. A good model of the con-tinuum must follow the main large-scale features in a spectrum,and a general rule of thumb is to use a polynomial of an order BASH is a Unix shell, a free software in common with all the oper-ating systems based on UNIX and Linux. IRAF is distributed by the National Optical Astronomical Observa-tory which is operated by the Association of Universities for Researchesin Astronomy, under cooperative agreement with the National ScienceFoundation. or Article number, page 2 of 9ristan Cantat-Gaudin et al.: DOOp, an automated wrapper for DAOSPEC similar to the number of waves seen in the spectrum. For a largewavelength range, one must use a polynomial of higher order.Di ff erent choices of continuum can result in di ff erences in EWsof up to 2 mÅ for some lines, as will be illustrated in Sect. 5.3when comparing measurements with literature values obtainedwith DAOSPEC. The work flow of the pipeline is summarised in Fig. 1. Thepipeline reads two input files. One is needed to set the basic op-tions of the algorithm of the whole pipeline, such as the namesof the output files and the convergence parameters for Module 2(described later in this section). The other input file contains thelist of spectra to analyse and a set of six parameters for each ofthem. The first five parameters are the DAOSPEC parametersOR (order of the polynomial for continuum fitting), FW (firstguess of the FWHM), FI ( = = • Module 1: this module provides the SH, LO, RE, MI, andMA parameters, as well as the first EW estimates. The SH,LO, and RE values are determined calling IRAF twice withtwo di ff erent IRAF scripts. The first looks for the startingand ending wavelength of the spectrum avoiding glitches andbad values that may be found at the spectrum borders, settingthe two parameters as the first / last wavelength for which thespectrum has a non-negative value. The second looks forthe strongest lines in the spectrum (H α , H β , Mg b triplet,CaII triplet) to set properly the RE parameter. If negativevalues are found, or if none of these strong lines are de-tected, a default value is imposed. The MI and MA param-eters are set accordingly to the input RV: if 0, then a widerange is imposed, otherwise a smaller range is used. Theseranges can be defined by the user ( ±
500 km / s and ±
10 km / sare typically reasonable values). At the end of Module 1 afirst EW measurement is performed on the spectra for whichDAOSPEC did not encounter computational problems. Thismeasurement is not the best one, as not all the parameters areoptimised. • Module 2: this module provides the best FWHM parame-ter for DAOSPEC. For each analysed spectrum, an initialvalue of FWHM is needed. If the user wishes to optimisethe FWHM (by setting the parameter FI = ffi cult in some spectra (for instance because of di ff erent signal-to-noise ratios). In such a case it can be su ffi cient to reach con-vergence for only 50 or 70% of them and use the medianvalue of the FWHM to measure the others (which is done byModule 3). In the most general case of a batch containingspectra of potentially di ff erent FWHMs (because the spectrawere obtained with di ff erent instruments, under di ff erent skyconditions or because of rotating stars) it is recommendedto require 100% of the spectra to reach convergence of theFWHM, so that this parameter is estimated in an independantway for each spectrum. If after 30 iterations the FWHM hasnot converged for some spectra, Module 3 will take care ofthem. • Module 3: this module determines the FWHM that will beused for the final EW measurements. It computes the medianFWHM of all the spectra for which it converged. Dependingon the choice of the user, it will use this median value forthe spectra that did not converge, or for all the spectra in thebatch including those that converged . This second optionmay be more suitable if all the spectra are known a priori tohave the same FWHM. Of course, the spectra for which theuser had required to use a fixed FWHM (by setting FI = the final EWmeasurements of DOOp . • Module 4 is designed to perform two di ff erent tasks. Oneis to provide the fit and normalised spectra in FITS format,because DAOSPEC only provides the continuum and resid-ual FITS files. The other task is to provide EW measure-ments obtained by over- and underestimating the continuumlevel. The amount by which the continuum will be shiftedis proportional to the dispersion of the residuals in the fi-nal DAOSPEC fit. The results coming out of these experi-ments can be used to quantify the error in the EW measure-ment due to the placement of the continuum. For UVESspectra (R =
47 000), the dispersion of the residuals of thefit ranges typically from 3% for a S / N of 30, to 1% or lessfor a S / N above 100. Altering the continuum placement bythese amounts can lead to di ff erences of 10 and 3 mÅ re-spectively, although this is certainly a conservative estimateand can depend on other factors such as the metallicity ofthe star. A description of how the EWs are changed whenaltering the continuum placement is done by SP08 in theirSect. 3.4.3 and Fig. 2. Module 4 uses IRAF script to obtainthe FITS files, while DAOSPEC is called to perform the EWmeasurements on the spectra with artificially imposed con-tinuum levels. This module works with the output producedby Module 3. • Utilities: together with the main algorithm of the pipeline,small scripts are provided to perform standard operations onthe output files obtained by Module 3. They are presentlyused to produce the input files for the abundance analysisprograms GALA and FAMA and to print out a summary ofthe analysis, including a log of the possible errors. With aprovided script, the user can easily visualise and comparethe spectra and their corresponding fit (see Fig. 2).One advantage of the pipeline is that it can be easily cus-tomised. For example, if only the RV measurement is needed,one can set the pipeline to use only Module 1. If, instead, onewants to test the e ff ect of changing the continuum order whilekeeping all the other parameters fixed, it is also possible. Fur-thermore, if the user prefers to pre-normalise their spectra us-ing a personal routine, indicating a continuum order of -1 tellsDAOSPEC not to perform any normalisation. Article number, page 3 of 9 ig. 1.
The tasks performed by DOOp are organised in several modules. The figure shows the dependencies of the modules and their mainresults.
Examples of the resulting fits obtained with DOOp on a spec-trum of Arcturus can be seen in Figs. 2 and 3.
DOOp exploits the versatility of the BASH shell, which usu-ally comes together with a set of smaller stable and powerfulprogramsl to perform operations on files and data. It uses theBASH language and a small set of these programs to handle allthe logical operations needed to accomplish the optimisation ofthe DAOSPEC parameters. The external programs used by thepipeline are DAOSPEC, IRAF, and Python. The latter is calledto display the interactively zoomable plots that allow the qualityof the fits to be controlled.On one hand, a script code is more fragile than monolithiccodes because it does not pass any compilation checks and more-over the use of di ff erent programs interacting together can be lessportable. On the other hand, script languages (Perl, Python, andJavaScript, to name a few well-known examples) are now widelyused to make small programs for which the computational speedis not a key requirement. They are particularly intuitive, easy,and fast to use and their popularity often ensures the portabilityof the code.From the beginning, DOOp was meant to be used by di ff er-ent groups in di ff erent locations and the need for portability wasa top priority. It has been tested on di ff erent operating systemswith di ff erent versions of BASH and IRAF, hence we can guar-antee a full compatibility with at least the software we could test.It has been used on 32-bit and 64-bit Linux kernels (Ubuntu andCent OS distributions) and with Mac OS. The BASH versions of the machines used range from the old 3.2.48 (2007) to the newer4.1.2 (2009). We used two di ff erent IRAF versions: 2.14 and2.16. The DOOp pipeline provides a robust method of measuring EWsin stellar spectra by limiting as much as possible human inter-vention in the process and facilitating the control of the results.This makes it especially convenient when dealing with batchesof hundreds or thousands of spectra produced during observa-tional campaigns such as the GES, and when di ff erent groupswork with the same data.For example, the estimate of the RE parameter usually re-quires manual measurements for each spectrum, while DOOpperforms automatic measurements. It also automatically sets theSH and LO parameters for each spectrum, with great gain oftime, and reduces the velocity range in which DAOSPEC looksfor the radial velocity when the RV parameter is at least approx-imately known.Ensuring that DAOSPEC converges to the best value for theFWHM in the spectrum can be a very lengthy procedure if donemanually, even for a single spectrum. We show in Sect. 4.1 howthe choice of initial FWHM can a ff ect the convergence and thefinal EW measurements. The DOOp code takes care of this con-vergence process automatically.The DAOSPEC code works by fitting gaussians to the ab-sorption lines but does not return any file containing the globalfitted spectrum. Instead, it returns a fits file containing the con-tinuum that was fitted, and a second file containing the residuals Article number, page 4 of 9ristan Cantat-Gaudin et al.: DOOp, an automated wrapper for DAOSPEC (cid:0) [ (cid:1) A]01000200030004000500060007000 f l u x fwhm=11.164, 21 FeI and 1 FeII lines.observed spectrumcontinuum6000 6200 6400 6600 6800 (cid:2) [ (cid:3) A]0.00.20.40.60.81.0 n o r m a li z e d f l u x N O D A O S P E C O U T P U T A V A I L A B L E normalizedDaospec fit Arcturus
Fig. 2.
UVES-POP spectrum of Arcturus seen through the graphi-cal interface of DOOp.
Top: original spectrum and fitted continuum.The flux is given in arbitrary units, as the instrument response was notcorrected for. Information is displayed on the FWHM of the lines andthe number of lines that were identified from the line list.
Bottom: nor-malised spectrum and fit. (cid:4) [ (cid:5) A]0.40.50.60.70.80.91.0 n o r m a li z e d f l u x N O D A O S P E C O U T P U T A V A I L A B L E normalizedDaospec fit Arcturus
Fig. 3.
Detail of the normalised spectrum of Arcturus from Fig. 2 andcorresponding fit. of the fit. The DOOp code automatically combines these fileswith the original spectrum to create a file containing the fittedspectrum. It also contains tools for direct plotting of the re-sults (using the popular Python library Matplotlib). Each moduleprints out a log file containing a summary of its action on eachspectrum, thus allowing an e ffi cient control over the whole pro-cedure. When DAOSPEC is run, the format of the output filesmatches the line list provided by the user. In its current ver-sion, DOOp can automatically convert the output files producedby DAOSPEC to the format needed by FAMA and GALA if theuser provides a line list in the right format. Given the structureof DOOp, organised in independent modules, adapting the codeto work with a custom format of line list or produce a specificformat of output can be easily done. In any case, the output filesof DAOSPEC are kept by DOOp, and one can use them exactlyas they would when running DAOSPEC manually.
4. Tests on synthetic spectra
We applied DOOp to the synthetic spectra used by SP08 to ex-plore the boundaries in S / N, resolution, and pixel sampling underwhich DAOSPEC performs optimally. We did not run the testson the spectra of resolution 5 000 and 10 000, as SP08 show thatthese resolutions are too low for DAOSPEC to perform well (andtoo low for a reliable EW analysis).
We have run tests on the synthetic spectra used in SP08 to com-pare the EWs recovered after single run of DAOSPEC (whichmeans that no condition on the convergence of the FWHM isimposed) and a run of DOOp (which automatically imposes theconvergence of the FWHM), on spectra of various resolutionsand signal-to-noise ratios. The true FWHM of these syntheticspectra is 5 px. The top-right and bottom-right panels of Fig. 4show the average di ff erence between the true EWs and the EWsmeasured with DAOSPEC after one run only, starting from dif-ferent initial values of the FWHM. The measurements are clearlydependent on the initial FWHM. The top-left and bottom-leftpanels of the same figure show the di ff erence between the trueand measured EWs, after DOOp has imposed the convergenceof the FWHM. It is clear that the convergence process carriedout by DOOp makes the measurements independent of the in-put FWHM. This experiment was conducted on spectra at vari-ous resolutions and signal-to-noise ratios, showing that DOOp ismore stable in all cases.Ensuring the convergence of the FWHM can also have con-sequences on the line detection. Starting from a too large valuewill cause DAOSPEC to ignore some features that it should actu-ally be fitting. Running DAOSPEC again using the newly foundFWHM as a new initial value for the convergence helps to find allthe lines. Figure 5 shows two fits obtained with an input FWHMof 10 px, on a spectrum of resolution R =
20 000 and S / N = We have compared the measurements of DOOp, with those ofSP08, who optimised manually all the parameters (implyingmultiple runs of DAOSPEC). The spectra used for these testsare the same as in Sect. 4.1, with an additional four spectra ofdi ff erent pixel sampling (true FWHM of 1, 2, 3 and 10 px). Theresults of DOOp show excellent agreement with those obtainedby SP08. The average di ff erences and the dispersion betweenboth sets of measurements are reported in Table 1. The fit uncer-tainty is the average uncertainty on the fit of the single lines, thatDAOSPEC computes from the least-square fitting procedure (see Article number, page 5 of 9 (cid:1) ∆ E W (cid:0) [ m ◦ A ] DOOp
SNR = 100R 20000R 40000R 60000 2 3 4 5 6 7 8 9 10i put FWHM [px]−10−8−6−4−202 (cid:1) ∆ E W (cid:0) [ m ◦ A ] DAOSPEC (o e ru )
SNR = 100R 20000R 40000R 600002 3 4 5 6 7 8 9 10i put FWHM [px]−8−6−4−202 (cid:1) ∆ E W (cid:0) [ m ◦ A ] DOOp
R = 35000SNR 10SNR 30SNR 50SNR 100 2 3 4 5 6 7 8 9 10i put FWHM [px]−8−6−4−202 (cid:1) ∆ E W (cid:0) [ m ◦ A ] DAOSPEC (o e ru )
R = 35000SNR 10SNR 30SNR 50SNR 100
Fig. 4. Di ff erence between our measurements and the true EW for synthetic spectra with a true FWHM of 5 px, at various resolutions and S / N. Top-left: average di ff erence between the EWs measured with DOOp and the true EWs, for various input values of the FWHM, in synthetic spectraat three di ff erent resolutions. Top-right: same as top-left panel, but for measurements obtained without imposing the convergence of the FWHM.
Bottom-left: same as top-left panel, for spectra of four di ff erent S / N. Bottom-right: same as bottom-left, but without imposing the convergennceof the FWHM.
Table 1.
Comparison between the EWs measured by DOOp and byDAOSPEC on synthetic spectra.R S / N FWHM h ∆ EW i r.m.s. fit uncertainty[px] [mÅ] [mÅ] [mÅ]35 000 100 1 -0.7 0.2 5.135 000 100 2 -0.3 0.1 6.335 000 100 3 -0.4 0.1 4.635 000 100 10 0.3 0.2 2.620 000 100 5 -0.3 0.4 5.640 000 100 5 -0.3 0.1 3.460 000 100 5 0.0 0.3 1.835 000 10 5 -0.6 0.1 4.835 000 30 5 -0.6 0.3 3.735 000 50 5 -0.6 0.1 3.635 000 100 5 -0.4 0.3 3.4 Notes.
The di ff erence is given in the sense DOOp-SP08. SP08 for details). The di ff erence and uncertainty decrease withpixel sampling, resolution and signal-to-noise, as is expected.Four of these cases are illustrated in Fig. 6.
5. Comparing the measurements of DOOp withliterature
We have checked the measurements obtained with DOOp againstalready published measurements obtained with other methods,to ensure that our method is able to reproduce the same results,in particular those of DAOSPEC used manually, on real stellarspectra. The stars used in these comparisons are giant, metal-rich stars (-0.5 < [Fe / H] < We have measured EWs in a spectrum of Arcturus downloadedfrom the UVES-POP archive (Bagnulo et al. 2003) and de-graded from a resolution of 80 000 to 47 000, which is theresolution of the UVES-FLAMES spectra.. These EWs werecompared with measurements by Friel et al. (2003), hereafterF03, on the high resolution spectrum of Arcturus published byHinkle et al. (2000). They have normalised the spectrum us-ing the IRAF CONTINUUM task and measured the EWs us-ing the IRAF task SPLOT. The comparison (Fig. 7) shows good Article number, page 6 of 9ristan Cantat-Gaudin et al.: DOOp, an automated wrapper for DAOSPEC ◦ A]0.50.60.70.80.91.0 n o r m a li z e d f l u x non-convergedconverged Fig. 5.
The dotted line is a synthetic spectrum of resolution R =
20 000.The true value of the FWHM for this spectrum is 5 px, but for this testwe have used 10 px as a starting point. The blue line is the fit obtainedwith DAOSPEC run only once. The cyan line is the fit obtained afterconvergence of the FWHM. Refining the value of the FWHM not onlyallows for a better fit of the line profiles, but also avoids DAOSPECmissing some lines. −4−2024 ∆ E W [ m ◦ A ] ∆EW=0.3 ±0.2 m ◦ AR=35000, SNR=100, FWHM=10px−4−2024 ∆ E W [ m ◦ A ] ∆EW=-0.6 ±0.3 m ◦ AR=35000, SNR=30, FWHM=5px−4−2024 ∆ E W [ m ◦ A ] ∆EW=-0.4 ±0.3 m ◦ AR=35000, SNR=100, FWHM=5px20 40 60 80 100EW [m ◦ A]−4−2024 ∆ E W [ m ◦ A ] ∆EW=0.0 ±0.3 m ◦ AR=60000, SNR=30, FWHM=5pxSynthet c spectra - DAOSPEC vs DOOp
Fig. 6. Di ff erence in EW (in the sense DOOp-SP08) for lines in fourdi ff erent synthetic spectra. The results for all the synthetic spectra usedin these tests are given in Table 1. agreement. The DOOp measurements have a general o ff set of-2.1 mÅ, with an r.m.s. dispersion of 3.8 mÅ. An o ff set of thismagnitude is consistent with the way DAOSPEC sets the contin-uum level (as discussed in Sect. 2 above). We have measured EWs in two stars of NGC 2477 and sixstars of Be 29, in UVES spectra previously measured and pub-
F03 [m ◦ A]−30−20−100102030 ∆ E W [ m ◦ A ] (cid:1)∆EW(cid:0)= -2.1 m ◦ Ar.m.s. = 3.8 m ◦ A 5000 5500 6000 6500λ [ ◦ A]−30−20−100102030 ∆ E W [ m ◦ A ] Arcturus - SPLOT vs DOOp
Fig. 7. Di ff erence in EW (in the sense DOOp-F03) plotted againstEW (left panel) and wavelength (right panel) for Arcturus. The averagedi ff erence is -2.1 mÅ, with an r.m.s. dispersion of 3.8 mÅ. B08 [m ◦ A] 30 20 100102030 ∆ E W [ m ◦ A ] (cid:1)∆EW(cid:0)= -2.9 m ◦ Ar.m.s. = 4.3 m ◦ ANGC2477 - 364490 20 40 60 80 100 120 140EW
B08 [m ◦ A]−30−20−100102030 ∆ E W [ m ◦ A ] (cid:1)∆EW(cid:0)= -0.2 m ◦ Ar.m.s. = 4.8 m ◦ ANGC2477 - 36363 5000 5500 6000 6500λ [ ◦ A]−30−20−100102030 ∆ E W [ m ◦ A ] NGC2477 - 364495000 5500 6000 6500λ [ ◦ A]−30−20−100102030 ∆ E W [ m ◦ A ] NGC2477 - 36363NGC2477 - SPECTRE vs DOOp
Fig. 8. Di ff erence in EW (in the sense DOOp-B08) plotted againstEW (left) and wavelength (right) for two stars of NGC2477. lished by Sestito et al. (2008) and Bragaglia et al. (2008) (here-after B08), respectively. These stars are red giants of nearly so-lar metallicity. The authors have normalised the spectra usingthe IRAF task CONTINUUM, and fitted Gaussian profiles us-ing the code SPECTRE. We measure in general smaller EWsby 1 to 5 mÅ. Again, this can be explained by the behaviourof DAOSPEC in terms of continuum placement. The results fortwo stars of NGC 2477 are illustrated in Fig. 8. The identifiersof the stars are taken from the ESO Imaging Survey catalogue,as reported in B08. The o ff set between the measurements ofB08 and those of DOOp is inside the range of di ff erences ex-pected when comparing di ff erent methods (see for instance thecomparisons of SP08 between DAOSPEC and other tools). Theslightly di ff erent o ff set observed between the blue and red partof the spectral range comes from the fact that UVES spectra aresplit between a blue and a red arm, and both ranges were mea-sured independently (by B08 and by us), resulting in a slightlydi ff erent continuum adjustment. As a final test we have compared our EWs with measurements byPancino et al. (2010), hereafter P10, using DAOSPEC on three
Article number, page 7 of 9
20 40 60 80 100 120 140EW
P10 [m ◦ A]−30−20−100102030 ∆ E W [ m ◦ A ] (cid:1)∆EW(cid:0)= 0.1 m ◦ Ar.m.s. = 0.2 m ◦ ACr110 - 21080 20 40 60 80 100 120 140EW
P10 [m ◦ A]−30−20−100102030 ∆ E W [ m ◦ A ] (cid:1)∆EW(cid:0)= 0.2 m ◦ Ar.m.s. = 1.3 m ◦ ACr110 - 2129 5000 5500 6000 6500 7000λ [ ◦ A]−30−20−100102030 ∆ E W [ m ◦ A ] Cr110 - 21085000 5500 6000 6500 7000λ [ ◦ A]−30−20−100102030 ∆ E W [ m ◦ A ] Cr110 - 2129Cr110 - DAOSPEC vs DOOp
Fig. 9. Di ff erence in EW (in the sense DOOp-P10) plotted against EW(left) and wavelength (right) for two stars of Cr110. stars of Cr 110 and two stars of NGC 2420. These stars arered giants of solar metallicity. The spectra have a resolution of30 000 and a S / N of 70. Figure 9 shows the perfect agreementbetween the two sets of measurements for two stars of Cr 110(identifiers from Dawson & Ianna 1998). Such a good agree-ment is expected if the FWHM and residual core flux were care-fully set when using DAOSPEC manually, which can be timeconsuming. The DOOp code does not produce better resultsthan those expected from the most careful use of DAOSPEC,but making the procedure automatic reduces the sources of er-rors and makes it humanly possible to deal with large numbersof spectra.However, the critical issue of setting the continuum remains.The refinements of choosing a continuum order remain arbitrary,and two users fitting a continuum of slightly di ff erent order onthe same spectrum may find slight di ff erences in the measure-ments of EWs (rms of about 2 mÅ). The star 2129 of Fig. 9 is anexample of such a case, where small di ff erences varying acrossthe spectral range can be observed.
6. Analysis of benchmark stars
To assess the e ff ect of the di ff erent EW measurements on the de-termination of the stellar parameters, the stars used in Sect. 5are not ideal test cases because the authors use di ff erent linelists, and using only the lines in common between their lists andours does not provide su ffi cient statistics (for instance, we needenough FeII lines in common to derive a reliable gravity).As a final validation for the whole procedure, we mea-sured the EWs for four well-studied stars, Arcturus, Procyon,HD 23249, and the Sun. The first three are bright stars (V = -0.04,0.37, and 3.51, respectively) for which a rich literature is avail-able in the PASTEL database . Their spectra were taken fromthe UVES-POP archive. These spectra were obtained at the VLTwith the UVES instrument at a resolution R ∼
80 000 that we de-graded to 47 000 (which is the nominal resolution of the GESUVES spectra). The list of spectral lines we have used is theGES line list (Heiter et al. in prep.). The available UVES-POP http://vizier.u-strasbg.fr/viz-bin/VizieR?-source=B/pastel Table 2.
Atmospheric parameters of the four benchmark stars.star T e ff ∆ T e ff log g ∆ log g [Fe / H] ∆ [Fe / H][K] [K] LiteratureArcturus 4302 120 1.68 0.31 -0.53 0.12HD 23249 5025 255 3.84 0.17 0.07 0.15Procyon 6583 162 4.06 0.15 -0.01 0.16Sun 5777 ... 4.44 ... 0 ...This studyArcturus 4352 31 1.78 0.12 -0.45 0.13HD 23249 5108 73 3.82 0.15 0.07 0.15Procyon 6647 35 3.83 0.08 0.02 0.06Sun 5755 40 4.30 0.20 0.02 0.09
Notes.
For the literature values the numbers are the average valuesfound in PASTEL, and their associated errors are their standard de-viations. For this study, the parameters and their errors are given byFAMA. l o g ( g ) ProcyonSunHD23249Arcturus
Fig. 10. T e ff and log g for our four benchmarks. The filled symbols areour results, while the empty symbols are the various values found in theliterature. data cover the optical range, from 3040 to 10 400 Å, of whichwe have used the ranges 4760 − − , degraded from a resolution R ∼
120 000 to 47 000. Afterrunning DOOp, we passed the output files to FAMA to obtain theatmospheric parameters of these stars.The e ff ective temperatures, surface gravities and metallici-ties we obtain for these four stars are in good agreement withthe values available in literature, as shown in Fig. 10 and 11, andsummarised in Table 2.
7. Conclusion
The current and future large-scale spectroscopic surveys requireautomatic procedures for batch-processing large numbers of stel-lar spectra. Based on DAOSPEC, DOOp provides a robust and http://archive.eso.org/wdb/wdb/eso/repro/form Article number, page 8 of 9ristan Cantat-Gaudin et al.: DOOp, an automated wrapper for DAOSPEC [ F e / H ] ProcyonSunHD23249Arcturus
Fig. 11.
As in previous figure, but for T e ff and [Fe / H]. convenient way of measuring EWs and produces results of thesame quality as DAOSPEC used manually, while requiring lessuser intervention, thus making the results more reproducible andthe process faster. DOOp is able to optimise the key parametersof DAOSPEC, but not the order of the polynomial used for thecontinuum fitting, which still has to be chosen by the user. Weshow that di ff erent choices of continuum order can lead to smalldi ff erences in the EWs, up to ± Acknowledgements
This work was partially supported by the Gaia Research forEuropean Astronomy Training (GREAT-ITN) Marie Curie net-work, funded through the European Union Seventh Frame-work Programme (FP7 / References