Analysis of Galactic late-type O dwarfs: more constraints on the weak wind problem
W. L. F. Marcolino, J.-C. Bouret, F. Martins, D. J. Hillier, T. Lanz, C. Escolano
aa r X i v : . [ a s t r o - ph . S R ] F e b Astronomy&Astrophysicsmanuscript no. ms c (cid:13)
ESO 2018October 23, 2018
Analysis of Galactic late-type O dwarfs:more constraints on the weak wind problem ⋆,⋆⋆
W. L. F. Marcolino , J.-C. Bouret , F. Martins , D. J. Hillier , T. Lanz , C. Escolano LAM-UMR 6110, CNRS & Univ. de Provence, 38 rue Fr´ederic Joliot-Curie, F-13388 Marseille, France GRAAL-UMR 5024, CNRS & Univ. de Montpellier II, Place Bataillon, F-34095 Montpellier, France Department of Physics and Astronomy, University of Pittsburgh, Pittsburgh, PA 15260, USA Department of Astronomy, University of Maryland, College Park, MD 20742, USAReceived ; Accepted
ABSTRACT
Aims.
To investigate the stellar and wind properties of a sample of late-type O dwarfs in order to address the weak wind problem . Methods.
Far-UV to optical spectra of five Galactic O stars were analyzed: HD 216898 (O9IV / O8.5V), HD 326329 (O9V), HD 66788(O8V / O9V), ζ Oph (O9.5Vnn), and HD 216532 (O8.5V((n))). We used a grid of TLUSTY models to obtain e ff ective temperatures,gravities, rotational velocities, and to identify wind lines. The wind parameters of each object were obtained by using expandingatmosphere models from the CMFGEN code. Results.
We found that the spectra of our sample have mainly a photospheric origin. A weak wind signature is seen in C iv λλ ∼ − − − M ⊙ yr − ). A discrepancy of roughly 2 orders of magnitude is found between these mass-loss rates and the val-ues predicted by theory ( ˙ M Vink ), confirming a breakdown or a steepening of the modified wind momentum-luminosity relation at log L ⋆ / L ⊙ . .
2. We have estimated the carbon abundance for the stars of our sample and concluded that its uncertainty cannot causethe weak wind problem . Upper limits on ˙ M were established for all objects using lines of di ff erent ions, namely, P v λλ iii λ v λ iv λλ iv λ weak winds . Together with C iv λλ v λ M must be less than about -1.0 dex ˙ M Vink . Regarding the other transitions, theupper limits obtained still point to low rates: ˙ M must be less than about ( − . ± .
2) dex ˙ M Vink . We studied the behavior of the H α line with di ff erent mass-loss rates. For two stars, only models with very low ˙ M ’s provide the best fit to the UV and optical spectra. Weexplored ways to fit the observed spectra with the predicted mass-loss rates ( ˙ M Vink ). By using large amounts of X-rays, we verifiedthat few wind emissions takes place, as in weak winds . However, unrealistic X-rays luminosities had to be used (log L X / L Bol & − . Key words. stars: winds - stars: atmospheres - stars: massive - stars: fundamental parameters
1. Introduction:
Massive stars of spectral types O and B play an extremely im-portant role in astrophysics. They possess high e ff ective temper-atures ( T e f f > ffi cult for atmosphere and stellar evolutionmodels. These stars are known to be progenitors of fascinat-ing objects such as Red Supergiants (RSGs), Luminous BlueVariables (LBVs), Wolf-Rayet stars (W-Rs), and thus also ofsome of the most energetic phenomena in the Universe, i.e.,of type II supernovae and some γ -ray bursts (Massey 2003;Woosley & Bloom 2006). They also heavily a ff ect their host Send o ff print requests to : [email protected] ⋆ Based on observations made with the NASA-CNES-CSA FarUltraviolet Spectroscopic Explorer and by the NASA-ESA-SERCInternationalUltravioletExplorer, and retrieved from the MultimissionArchive at the Space Telescope Science Institute (MAST). Based onobservations collected with the ELODIE spectrograph on the 1.93-mtelescope (Observatoire de Haute-Provence, France). Based on observa-tions collected with the FEROS instrument on the ESO 2.2 m telescope,program 074.D-0300 and 075.D-0061. ⋆⋆ The Appendix is only available in electronic format. galaxies by transferring momentum, energy and enriched chem-ical elements to the interstellar medium (Abbott 1982; Freyer etal. 2003).Although they have been studied for decades, the properties,origin and evolution of OB stars still present several observa-tional and theoretical challenges. The dependency of their mass-loss rates ( ˙ M ) on the metallicity ( Z ) for example, as well as theire ff ective temperatures and wind structure (e.g. clumping) havebeen continuously debated in the literature during the last years(see for example Vink et al. 2001; Martins et al. 2002; Bouret etal. 2005; Puls et al. 2006; Crowther et al. 2006).Among several interesting issues currently under discussion(for a review see Puls 2008; Hillier 2008), one that have beenreceiving special attention is the so-called weak wind problem ,which is posed by late-type O dwarf stars. From a qualitativepoint of view, O stars with weak winds present mainly an absorp-tion spectrum, with the exception being a very few weak windlines. In fact, often only a weak C iv λλ iv λλ iv λ v λ v λ − M ⊙ yr − . Bouret et Marcolino et al.: Analysis of late-type O dwarfs al. (2003) were one of the first to suggest such low values afteranalyzing O dwarfs in the H II region NGC 346 in the SMC.The spectra of three objects of their sample could only be repro-duced by models using mass-loss rates of ∼ − to 10 − M ⊙ yr − . Similar results were found by Martins et al. (2004) for 4O dwarfs in the compact star formation region N81, also in theSMC. Later, weak winds were also found in some Galactic Odwarfs, demonstrating that they are not result of an environmen-tal (i.e. Z ) e ff ect (Martins et al. 2005). Despite these findings,as we will discuss later, very low ˙ M values are not of generalconsensus.There are interesting questions that O stars with very lowmass loss rates rise. From the stellar evolution point of view,it is well known that mass-loss is a fundamental ingredient inthe models. Very di ff erent values for this parameter can alterconsiderably the way the stars evolve. For example, mass-losscan change the rotational structure of a star (decreasing surfaceand internal velocities; Ω r ) due to removal of angular momen-tum combined with internal transport mechanisms (Meynet &Maeder 2000). Very low ˙ M ’s might thus imply that stars cankeep high rotational velocities and get closer to break-up veloci-ties. Low mass-loss rates can also change the way we understandthe evolution in the LBV and W-R phases. The total mass lostfrom the Main Sequence prior the LBV phase can be much lessthan currently thought. As a consequence, in order to be con-sistent with observed masses of hydrogen deficient W-R stars,other intense mass-loss mechanisms must occur (e.g. continuumdriven giant eruptions in the LBV phase) and / or evolutionarytime scales (e.g. of the WNL phase) must be changed (see Smith& Owocki 2006; van Marle et al. 2008). Given these considera-tions, to be sure that weak winds exist is now essential.From the radiative wind theory point of view, the low mass-loss rates obtained for late type O dwarfs also present challenges.While the last, state-of-art theoretical ˙ M predictions of Vink etal. (2000; 2001; ˙ M Vink ) present a good match for objects with log L / L ⊙ greater than about 5.2 (neglecting clumping), for late-type,less luminous objects a discrepancy of even a factor of 100 canbe found (Martins et al. 2005). Late O dwarfs data also suggest abreakdown of the modified wind momentum luminosity relation(WLR; Kudrizki & Puls 2000) or at least a change in its slope(see Fig. 41 of Martins et al. 2005). All these facts constitutewhat is now called the weak wind problem .A possible reason for the discrepancies aforementioned isthat mass-loss rates determinations based on UV lines could beincorrect, as it was suggested by Mokiem et al. (2007). The useof only one UV diagnostic line (i.e. C iv λλ ff ect of X-rays, and the wind ionisationstructure derived from the models are claimed to be consider-able sources of errors for the derivation of ˙ M . This means thatstronger winds could perhaps have been found if appropriate di-agnostic tools were used / available. According to Mokiem et al.(2007), three objects which are supposed to be in the weak windregime ( ζ Oph, CygOB2 M ’s are consistent with the-ory) if H α is used as the diagnostic.In order to clarify some of the issues described above andget more insight into the weak wind problem, we have ana- Another type of weak wind problem exists, which is the weakerwind signatures of some stars compared to others of the same spectraltype (e.g. θ Ori C; Walborn & Panek 1984). Throughout this paper,we consider ”weaker winds” compared to the predicted mass-loss ratesaccording to Vink et al. (2000) and to normal O stars of earlier spectraltypes. lyzed in detail far-UV, UV and optical high-resolution spectraof five Galactic late-type O dwarf stars. Their stellar and windphysical parameters were obtained using the codes TLUSTYand CMFGEN (Hubeny & Lanz 1995; Hillier & Miller 1998).With this work, the number of O8-9V objects studied by meansof state-of-art atmosphere models is increased considerably,since only six were previously analyzed (Repolust et al. 2004;Mokiem et al. 2005; Martins et al. 2005).The remainder of this paper is organized as follows. InSection 2 we describe the way we have selected our targets andthe observational material used. A description of the atmospherecodes and the adopted assumptions are given in Section 3. Theanalysis of each object of our sample is presented in Section 4.The derived stellar and wind parameters are presented later inSection 5 along with comparisons to previous results and the-oretical predictions. In Section 6 we discuss the carbon abun-dance and its relation with the mass-loss rate. In Section 7 wepresent estimates for the mass-loss rates of the programme starsusing other lines besides C iv λλ
2. Target Selection and Observations:
The Galactic stars claimed to have weak winds in the litera-ture belong mainly to the O8, O8.5, O9, and O9.5V spectraltypes (hereinafter we refer to them simply as O8-9V). In orderto carry out our investigation, we initially selected a sample ofseveral objects belonging to these classes by using the GalacticO Star Catalog (Ma`ız-Apell´aniz et al. 2004; hereinafter the GOSCatalog). Various stars were previously observed by our group(at optical wavelengths), and we have retrieved the spectra ofsome others from public available archives. After a first exam-ination, we neglected all those objects that presented a peculiarspectrum. This included for example stars known to be bina-ries (having contamination or known to show strong variabilityin the spectrum) or ON8-9V objects. Due to our aim to studywind lines, we have drastically decreased the number of objectsby considering only the ones having both UV and far-UV data.Furthermore, we have also neglected the stars in common withthe work of Martins et al. (2005), since they were analyzed withthe same atmosphere code utilized here (CMFGEN). In the end,we have chosen a subsample of five objects, listed in Table 1.
For the optical, we used data collected with the MPI 2.2m tele-scope, located at the European Southern Observatory (ESO), inLa Silla, Chile. The high resolution (R = ∼ = ∼ / N) ratio above 100 at about 5000Å.The data reduction, the order localization, background esti-mate / subtraction, and wavelength calibration, were performedusing the available pipeline (see Baranne et al. 1996). Each or- arcolino et al.: Analysis of late-type O dwarfs 3 Table 1.
Observational Data of the Programme Stars. far-UV UV OpticalStar Spectral Type a V Data Set b Date Data Set Date Instrument DateHD 216898 O9IV, O8.5V 8.04 F - A0510303 2000-08-03 SWP43934 1992-02-05 OHP / ELODIE 2004-08-29HD 326329 O9V 8.76 F - B0250501 2001-08-09 SWP48698 1993-09-21 ESO / FEROS 2005-06-25HD 66788 O8V, O9V 9.43 F - P1011801 2000-04-06 SWP49080 1993-11-03 ESO / FEROS 2004-12-22 ζ Oph O9.5Vnn 2.56 C - C002 1972-08-29 † SWP36162 1989-04-29 OHP / ELODIE 1998-06-11HD 216532 O8.5V((n)) 8.03 F - A0510202 2000-08-02 SWP34226 1988-09-11 OHP / ELODIE 2005-11-10 a Spectral types come from Lesh (1968), Schild et al. (1969), Garrison (1970), Walborn (1973), and MacConnell & Bidelman (1976). b Satellite used to obtain spectra: F = FUSE and C = Copernicus. † Orbits der was normalized by a polynomial fit to the continuum, speci-fied by carefully selected continuum windows. At last, we havemerged the successive orders to reconstruct the full spectrum.The spectrum of ζ Oph was obtained from the ELODIE archive(see Moultaka et al. 2004 for more details). We have retrievedsmall portions of its pipeline treated spectrum and then normal-ized them using polynomial fits. Di ff erent sources were used for the UV and far-UV data. Wehave retrieved spectra from the IUE and FUSE satellites usingthe Multimission Archive at STScI . Regarding the IUE data, thehigh resolution SWP mode was preferred ( ∼ . / LWR region were also used to help in the determi-nation of reddening parameters (R and E(B-V)). Regarding theFUSE data, we have relied only on the LiF2A channel, whichcomprises the ∼ / N) and clarity. Although small, theLiF2A region provided two very useful features for our study,namely, P v λλ iii λ ∼ ζ Oph, there is no FUSE dataavailable. Therefore, we have used far-UV data obtained withthe
Copernicus (“OAO-3”) satellite. The spectrum was retrievedfrom the Vizier Service and normalized “by eye”.
3. Models and Assumptions:
In order to obtain the stellar and wind parameters of the starsof our sample we have used the well known atmosphere codesTLUSTY (Hubeny & Lanz 1995) and CMFGEN (Hillier &Miller 1998). The TLUSTY code adopts a plane-parallel ge-ometry, assumes hydrostatic and radiative equilibrium, and anon-LTE treatment including line-blanketing is taken into ac-count. As such, it is suitable only to model lines that are notformed in a stellar wind, i.e., of photospheric origin. On the otherhand, the radiative transfer and statistical equilibrium equationsin CMFGEN are solved in a spherically symmetric outflow. Thee ff ect of line-blanketing is also taken into account, via a super-levels formalism (for more details see Hillier & Miller 1998).For the outflow we assume a β velocity law which connectssmoothly (at depth) with a hydrostatic structure provided byTLUSTY. MAST - http: // archive.stsci.edu / http: // webviz.u-strasbg.fr / viz-bin / VizieR
To start our analysis, we used a new grid of TLUSTY modelatmospheres based on the OSTAR2002 grid (Lanz & Hubeny2003). The new grid uses finer sampling steps in e ff ective tem-perature (1000 K ) and surface gravity (0.2 dex), updated modelatoms (for Ne and S ions, C iii , N ii , N iv , O ii , and O iii ), andadditional ions (Mg ii , Al ii , Al iii , Si ii , and Fe ii ; see Lanz &Hubeny 2007). In order to derive T e f f , we have used mainly He i λ ii λ i λ ii λ T e f f range from 1 to 2 kK . Ourderivation of log g was based on the fit to the H γ wings. For thisparameter, the uncertainty varies from 0.1 to 0.2 dex, depend-ing on the object. The rotational velocities ( vsini ) were adoptedfrom previous studies (e.g. Penny 1996; Howarth et al. 1997)and / or refined / estimated from the fitting process when neces-sary. After we have derived the basic photospheric properties(i.e. vsini , log g , and T e f f ) we have switched to CMFGEN tocontinue our investigation. As in previous studies, we noted agood agreement between TLUSTY and CMFGEN (e.g. Bouretet al. 2003). In general, conspicuous discrepancies were sortedout when we have increased the number of species or atomiclevels and / or super-levels in CMFGEN.Once e ff ective temperatures were obtained, we have com-puted the radii by adopting luminosities typical of the O8-O9dwarfs (see Table 4 of Martins et al. 2002), following the equa-tion R ⋆ = ( L ⋆ / πσ T e f f ) / . As we did not normalize the UVspectra, we have used the reddening parameters R and E(B-V)(following Cardelli et al. 1988) and the distance as free parame-ters to match the continuum in the IUE region. When a good fitwas achieved, the R, E(B-V), and the distance used were con-sidered representatives for the object in question. The values ob-tained were generally consistent with the ones estimated in theliterature (see Sect. 4). Whenever a large discrepancy was en-countered, we have revised our adopted L ⋆ and hence R ⋆ , keep-ing T e f f constant. A reasonable agreement is found between thefinal models and the observed absolute fluxes in the UV. In somecases however, local scalings of the theoretical continuum had tobe made.The two main wind parameters to be obtained are the mass-loss rate and the terminal velocity. Ideally, the determination ofthe mass-loss rates should be done by fitting di ff erent (sensi-tive) P-Cygni and emission lines in the optical (e.g. H α ) andin the UV (e.g. C iv λλ v λ iv λλ M isC iv λλ β = β ranging from Marcolino et al.: Analysis of late-type O dwarfs
Table 2.
Atomic Data Used in the CMFGEN Models.
Ion Basic Models Full Models i
30 30 30 30He i
69 69 69 69He ii
30 30 30 30C ii
22 22 338 104C iii
243 99 243 217C iv
64 64 64 64N iii
287 57 287 57N iv
70 44 70 44N v
49 41 49 41O ii
296 53 340 137O iii
115 79 115 79O iv
72 53 72 53Ne ii - - 242 42Mg ii - - 65 22Al iii - - 65 21Si iii - - 45 25Si iv
33 22 33 22P v
62 16 62 16S iv - - 142 51S v - - 216 33Fe iii - - 607 65Fe iv v iv λλ v ∞ ) is a very di ffi cult parameter to ob-tain for O8-O9V stars, since there are no clear, saturated P-Cygniprofiles. We have chosen to first use previous v ∞ estimates in theliterature and then change its value from the fits, if needed. Whensuch estimates were not available, we have started our modelsassuming v ∞ values between 1300-1700 km s − . We conserva-tively assume the uncertainty in this parameter to be about 500km s − . Regarding the mass-loss rate, we consider a conserva-tive uncertainty of ∼ ξ t of 15 km s − inthe co-moving computation of our models. The CMFGEN codeallows however a description of a depth dependent microturbu-lent velocity when passing to the observer’s frame, following theequation ξ ( r ) = ξ min + ( ξ max − ξ min ) v ( r ) / v ∞ . We have fixed ξ min at5 km s − (near the photosphere) and for ξ max we have assumeda value of 0 . v ∞ . As it was already reported in previous studies,we noted that slightly di ff erent values for these parameters donot present significant changes in the spectrum.We have used two sets of atomic data for CMFGEN in ourstudy. The first fits / models to the observed spectra, tests, and theexploration of higher mass-loss rates (Section 7) were done us-ing basic models , i.e., models with a reduced number of speciesand levels / super-levels. For our final model for each object, amore complete ( full ) set was used. In total, 4.5GB (2GB) ofmemory space (RAM) was required to compute each full ( ba-sic ) model. The details of the atomic data are shown in Table2. This approach could save a lot of computational time and re-sources, without compromising the results. The main di ff erencebetween these two sets of data is the quality of the optical fit.Some lines could only be reproduced after the inclusion of ad-ditional species / ions (e.g. Mg ii λ iii λ / or super-levels. In the ul-traviolet, only the 1850-2000Å region is a ff ected by the heavierdata set (being better fitted), due to the inclusion of Fe iii . In otherparts the changes are minor. A solar chemical abundance was adopted for all elements (following Grevesse & Sauval 1998).As it will be seen later in Section 4, it provides reasonable fits tothe observations. In this paper, only the amount of carbon is in-vestigated in detail (see Section 6), given its direct relation withthe weak wind problem through C iv λλ Currently, the e ff ect of wind density inhomogeneities (i.e.clumping) is implemented in CMFGEN through a depth depen-dent volume filling factor f = f ∞ + (1 − f ∞ ) e v ( r ) / v cl . The pa-rameter v cl regulates the velocity where clumping starts to beimportant. At velocities higher than v cl , f converges quickly to f ∞ . With this description, the model is homogeneous near thephotosphere. Because we have a lack of wind lines in the spectraof the stars studied here, we chose to not use clumping in ourmodels, i.e., f ∞ =
1. We stress that the use of clumping usuallyrequires a decrease in the mass-loss rate of a (previously homo-geneous final) model by a factor of 1 / p f ∞ , in order to fit (again)the observed spectra. Thus, the mass-loss in our final models canperhaps be overestimated by a factor of about three if f ∞ ∼ . ff ects of shocks due to wind instabilities. TheX-rays emissivity is basically controlled by a shock temperatureplus a filling factor parameter and it is distributed throughout theatmosphere (see Hillier & Miller 1998 for details). The e ff ectsof X-rays in the atmosphere of O dwarfs were discussed in de-tail by Martins et al. (2005). These authors convincingly showthat X-rays are not as important for early-type as they are forlate-type O stars (see also Macfarlane et al. 1994). For early-type objects (denser winds), the ionisation and wind lines suchas C iv λλ vi λλ v λ ff ected. In contrast, for late-type stars it is observed that theionisation structure and the C iv λλ v - vi , mass-lossrates about ten times stronger than in a model without X-rays aresometimes required to fit the observed C iv λλ × K and the fillingfactor is chosen to keep log L x / L Bol close to -7.0 (within ± . L x / L Bol of about -7.0 is the typical value ob-served (see e.g. Sana et al. 2006). For HD 326329 however, weused models with a higher ratio, in accord with the observations(log L X / L Bol ∼ − .
5; Sana et al. 2006). For ζ Oph, we use a log L X / L Bol ∼ − .
4. Spectral Analysis:
In this section we present the analysis of our sample. A brief in-troduction about each object is made, followed by the presenta-tion of our CMFGEN model fits to the UV and optical observedspectra. The discrepancies found are discussed and a comparisonto previous results in the literature is given when necessary.
HD 216898 belongs to the Cepheus OB3 association (Garmany& Stencel 1992). So far, no recent atmosphere models (i.e. uni- arcolino et al.: Analysis of late-type O dwarfs 5
Fig. 1.
Ultraviolet spectra of HD 216898 (green / light gray line) and our final model (black line; log ˙ M = − . Fig. 2.
Optical spectrum of HD 216898 (green / light gray line) and our final model (black / dashed line).fied, line-blanketed) were used to analyze this object. Its mass-loss rate was investigated two decades ago by Leitherer (1988),who could only establish an upper limit (log ˙ M < − .
97) fromthe H α profile.Our fits to the far-UV and UV spectra are presented in Figure1. The E(B-V) and distance used are 0.7 and 1.1kpc, respec- tively. Four intense transitions can be noted in the FUSE region:P v λλ iv λλ ,
28, and C iii λ iv λλ Marcolino et al.: Analysis of late-type O dwarfs
Fig. 3.
Ultraviolet spectra of HD 326329 (green / light gray line) and our final model (black line; log ˙ M = − . M = − .
55; see text for more details).C iv λλ iv lines ( an iron forest ). We note that a TLUSTY model withthe same photospheric parameters used in CMFGEN present asimilar fit to the far-UV and UV, with the exception of C iv λλ iv λλ M = − .
35. If we take the other physical parameters derived for thisobject ( T e f f , L ⋆ , and M ⋆ = gR ⋆ / G ) and use the recipe ofVink et al. (2000), we derive a predicted mass-loss rate of log˙ M Vink = − .
22. A discrepancy of more than two orders of mag-nitude is observed ! If we use mass-loss rates higher than theone in our final model, a deeper blue absorption (with respectto the line center) and an intense emission starts to be seen, i.e.,a more normal P-Cygni profile is developed, contrary to whatis observed. On the other hand, lower ˙ M ’s slowly approach thephotospheric prediction by TLUSTY, as expected.In Figure 2 we present our fit to the optical spectrum of HD216898. The five di ff erent spectral regions shown comprise thediagnostic lines used to derive the photospheric parameters (seeSection 3.1). The e ff ective temperature could be very well con-strained. The He i λ / He ii λ i λ / He ii λ T e f f = (34 ± kK .H γ has a reasonable fit but the model is somewhat stronger thanthe observed line. The same happens to H α . Nevertheless, theirwings are well matched with a log g of 4.0 ± i λ δ , H γ ,He i λ i λ ii λ iii λ iii λ ii ), He ii λ i λ α . These same transitions are easily identified in the otherobjects of our sample - and in other typical O8-9V stars - thathave vsini less than about 100 km s − . For rapid rotators such as ζ Oph and HD 216532, several profiles are broadened / blendedand in some cases we cannot distinguish individual transitions(see below). We highlight that the H α observed in HD 216898is symmetric. Indeed, all model predictions are symmetrical forthis line. However, as it is shown later, some objects present anasymmetric H α profile whose origin is not known. HD 326329 is located in the NGC 6231 cluster, at the nucleus ofthe Sco OB1 association. It belongs to a resolved triple system(O9V + O9V + B0V; GOS Catalog), but it is probably a single staritself since it does not show photometric or spectroscopic varia-tions (Garcia & Mermilliod 2001). So far, no atmosphere modelswere used to analyze its spectrum.In Figure 3 we present our final model to the far-UV andUV observed spectra. From the IUE continua, we have deriveda E(B-V) of 0.44 and a distance of 1.99 kpc , in agreement withthe study of the NGC 6231 cluster made by Baume et al. (1999).As in the case of HD 216898, a good agreement is found in theFUSE spectral region. The P v λλ iv λλ , iii λ iv λλ ff erent tests to fix this discrepancy (e.g. changesin the β velocity law; lower terminal velocities; increased tur-bulence), but they all turned out to be unsuccessful. We notethat in the study of Martins et al. (2005) no such problem wasfound. The O9 dwarfs analyzed in their study present a C iv λλ / broad feature. Interestingly, one morestar of our sample also present this characteristic, namely, HD216532 (see below).Despite the problem described above, we can still be con-fident about the mass-loss rate chose for HD 326329, log ˙ M = -9.22, within ± iv λλ M must be indeed low. When values for ˙ M higher than the one inour final model are used, a synthetic line with an intense (un-observed) emission and discrepant blue absorption appears (seeFigure 3; dotted line). On the other hand, when lower mass-lossrates are used, C iv λλ i and He ii lines are well reproduced witha T e f f = (31 ± kK . The fit to the H γ wings gives a surfacegravity (log g ) of 3 . ± .
1. Both T e f f and log g were previouslyestimated by Mathys et al. (2002) who found 31700 K and 4.6, arcolino et al.: Analysis of late-type O dwarfs 7 Fig. 4.
Optical spectrum of HD 326329 (green / light gray line) and our final model (black / dashed line). Fig. 5.
Ultraviolet spectra of HD 66788 (green / light gray line) and our final model (black line; log ˙ M = − . uvby β photometry. Although our derivedtemperature is compatible with their value, their higher log g is not supported by our analysis. In fact, a log g much higherthan 4.0 is not expected for O stars (see e.g. Vacca et al. 1996;Martins et al. 2002). Although a fairly good agreement is foundin several parts of the optical spectrum, we could not achieve a good fit to H α . This line is weaker than in the model and isalso asymmetric. Its blue wing has a “bump” which distorts theprofile. This is not seen in the other hydrogen lines. We have notused H α to derive any stellar or wind parameters of the objectsof our sample. However, we later discuss the e ff ects that di ff erentmass-loss rates have on this line (see Section 7.2). Marcolino et al.: Analysis of late-type O dwarfs
Fig. 6.
Optical spectrum of HD 66788 (green / light gray line) and our final model (black / dashed line). Figure 5 shows our fits to the ultraviolet spectrum of HD 66788.The C iv λλ M = -8.92. This value is much lower than the one pre-dicted, log ˙ M Vink = − .
95. The reddening and distance derivedare E(B-V) = = γ we derive a log g = . ± . i λ ii λ T e f f = (34 ± kK , as in HD 216898. Their rotational velocitiesare also practically the same, about 60 km s − . We note that theH α line in HD66788 is slightly asymmetric. Although the over-all agreement is reasonable, its blue wing has a weak absorptionnot predicted by the model. ζ Oph ζ Oph (also known as HD 149757) is a well studied runawaystar which presents several interesting characteristics. It has been known for a long time to present di ff erent kinds of spec-tral variability, such as discrete absorption components in theUV (DACs), emission line episodes, and line profile variations(LPVs) (see e.g. Howarth et al. 1984; Reid et al. 1993; Howarthet al. 1993; Jankov et al. 2000; Walker et al. 2005). This objectpresents also a very high rotational velocity: vsini values around400 or even 500 km s − were already reported in the literature(Walker et al. 1979; Repolust et al. 2004).A reason to include ζ Oph in our sample, is that it is consid-ered to have a ˙ M estimate based on H α which is in good agree-ment with the predictions of the radiative wind theory (Mokiemet al. 2005; 2007). Thus, it is interesting to check if a fit fromthe UV to the optical (including H α ) can be achieved withCMFGEN using a low ˙ M . We stress however, that the very highrotational velocity presented by ζ Oph can bring some problemsto the analysis. First, fast rotation can distort the shape and in-duce temperature and gravity variations through the stellar sur-face (see e.g. Fr´emat et al. 2005). Furthermore, it might implythat a stellar wind is not spherically symmetric. In such cases, 1Dunified atmosphere models should be considered as an approxi-mation. Another di ffi culty is that several features in the spectrumare broadened and sometimes blended. The analysis of diagnos-tic lines thus can present a greater di ffi culty than in the case ofslower rotators. arcolino et al.: Analysis of late-type O dwarfs 9 Fig. 7.
Ultraviolet spectrum of ζ Oph (green / light gray line) and our final model (black line; log ˙ M = − . M = − .
30; see text for more details).
Table 3.
Stellar and Wind Parameters.
Star HD 216898 HD326329 HD 66788 ζ Oph HD 216532Spec. type O9IV, O8.5V O9V O8-9V O9.5Vnn O8.5V((n))log g 4.0 ± ± ± ± ± ef f (kK) 34 ± ± ± ± ± − ) 60 80 55 400 190log L / L ⊙ ± ± ± ± ± ⋆ (R ⊙ ) 6 . + . − . . + . − . . + . − . . + . − . . + . − . M ⋆ ( M ⊙ ) 17 + − + − + − + − + − v ∞ (km s − ) 1700 ±
500 1700 ±
500 2200 ±
500 1500 ±
500 1500 ± M -9.35 ± ± ± ± ± M Vink -7.22 -7.38 -6.95 -6.89 -6.92log L x / L Bol -7.00 -6.69 -7.06 -7.31 -7.00log L / c -8.42 -8.41 -8.19 -8.29 -8.36E(B-V) 0.7 0.44 0.22 0.36 0.72R 3.5 3.1 2.8 2.9 3.5distance (pc) 1100 1990 4800 146 1150 Our final model and the IUE spectrum of ζ Oph are shownin Figure 7. A distance of 146pc and an E(B-V) = vsini involved, the pho-tospheric iron forest is considerably broadened. Despite weak,N v λ iii λ iv λλ iv λλ v λ M = − .
80, should be certain within a fac-tor of three (i.e. ∼ α , these authors derive log˙ M ( H α ) = − .
85. Our value is also much lower than the one pre-dicted according to Vink et al. (2000), log ˙ M Vink = − .
89. Wecome back to this question later in Section 7. Some other smallerdiscrepancies can be also seen in the Si iv λλ ∼ − ff orts, theycould not be sorted out. Due to the very high rotational velocity presented by ζ Oph, future studies using 2D atmosphere modelsmight be very useful to address some of these problems.In Figure 8 we present our model to the optical spectrum.As we mentioned earlier, a high rotational velocity convolutionis necessary to fit the broad features observed ( vsini ∼
400 kms − ). Despite the high rotation, the e ff ective temperature and sur-face gravity could be derived, but with a higher uncertainty thanin the other objects of our sample. From the He i and He ii lines,we have obtained a T e f f = (32 ± kK . From the H γ profile weestimate a low surface gravity compared to the other objects: log g = . ± .
2. It should be kept in mind that this value shouldbe regarded as an e ff ective gravity , i.e., the gravity attenuatedby rotation. A “centrifugal correction” can be applied in orderto obtain the true gravity, according to the equation log g true = log ( g e f f + vsini / R ⋆ ) (see details in Repolust et al. 2004). For ζ Oph, we find a log g true of ∼ .
8. This correction however, ne-glects any distortion in the stellar shape. A treatment allowingthe oblateness of the star was presented by Howarth & Smith(2001) for three fast rotators, including ζ Oph. These authorsderived a polar (equatorial) gravity of ∼ ∼ ⊙ (9.1R ⊙ ), and an e ff ective temperature of ∼ kK . The radius obtained from our T e f f and L ⋆ is 9.2R ⊙ .Therefore, our stellar parameters for ζ Oph are compatible tothe equatorial values found by Howarth & Smith (2001). They
Fig. 8.
Optical spectrum of ζ Oph (green / light gray line) and our final model (black / dashed line).also show agreement with other works in the literature (Repolustet al. 2004; Mokiem et al. 2005; Villamariz & Herrero 2005).Turning our attention to H α , we note that this line and afew others were observed to present emission episodes (in theform of double peaks) on di ff erent occasions (see e.g. Niemela& Mendez 1974; Ebbets 1981). However, in general, H α is ob-served to be a symmetric absorption, with an equivalent width( W λ ) of (2 . ± . ζ Oph’s quiescent spectrum (see Reid et al. 1993 andreferences therein). In our case, from the observed line we mea-sure a W λ of (3 . ± . ∼ quiescent -like,i.e., normal spectrum. However, the profile presents narrowerwings if compared to the one in our final model (see Figure 8,bottom). This problem can be also perceived in the model pre-sented by Mokiem et al. (2005) using the FASTWIND code (seetheir Fig. 12). A decrease in the vsini to 350 km s − improvesthe fit to the wings, but the line center gets slightly deeper thanobserved. Such lower vsini has also a negative e ff ect in otheroptical lines such as H δ and H γ . HD 216532 is a relatively fast rotator which belongs to theCepheus OB3 association. According to Howarth et al. (1997)its vsini is ∼
190 km s − , a value that we confirm from ourmodel fits. Leitherer (1988) have estimated its stellar parame-ters and derived an upper limit for the mass-loss rate from H α ,log ˙ M < − .
08. The present study is the first to analyze quanti-tatively its far-UV to optical spectra.Our fit to the far-UV and UV data is presented in Figure 9.For the E(B-V) and the distance, we inferred 0.72 and 1.15kpc,respectively. In the FUSE region, although the overall agree-ment is fair, some discrepancies can be noted. The observedP v λλ iv λλ ,
28 lines are deeper thanin the model. Although we can see that the synthetic contin-uum is somewhat higher than the observed, the problem remainsif we normalize the spectrum locally. Regarding the IUE spec-trum, as in HD 326329, we could not fit the C iv λλ iv λλ M = − .
22, should be certain within about ± arcolino et al.: Analysis of late-type O dwarfs 11 Fig. 9.
Ultraviolet spectra of HD 216532 (green / light gray line) and our final model (black line; log ˙ M = − . M = − .
78; see text for more details).slowly change the C iv λλ vsini of this object, ∼
190 km s − . With an e ff ective temperature of ∼ i λ ii λ γ profile, we have used a log g of 3 .
5. Analysis of the Results:
In this section we summarize the results of our spectral analysisand make a comparison to previous works and theoretical pre-dictions. The stellar and wind parameters obtained for each starof our sample are presented in Table 3. For comparison, we alsolist the theoretical mass-loss rates computed following the recipeof Vink et al. (2000), ˙ M Vink . Additional parameters such as thereddening and the distances are shown in the lower part of thetable. Regarding the X-rays luminosities, we recall that for HD216898, HD 66788, and HD 216532, we have fixed log L X / L Bol at the canonical value, i.e. at ∼ − .
0. For HD 326329 and ζ Oph, we chose L X / L Bol ratios close to the ones recently observed(Oskinova 2005; Oskinova et al. 2006; Sana et al. 2006).Overall, the stellar properties of our sample are quite homo-geneous. The e ff ective temperatures obtained range from ∼ kK , and the radii are between ∼ R ⊙ . Regarding the surfacegravities, the most di ff erent (lowest) values are presented by HD216532 and ζ Oph. In both stars, the log g measured should beinterpreted as e ff ective, i.e., they are attenuated by their fast ro-tation. These physical properties show a fair agreement with thelatest calibration of Galactic O star parameters, regarding theO8-9V spectral types (see Martins et al. 2005b).In order to better analyze the stellar and wind parameters de-rived, we also present our results in Figure 11, in two di ff erentways: the mass-loss rates are compared to the theoretical pre-dictions of Vink et al. (2000); and the wind parameters are usedto construct the modified wind momentum luminosity relation(WLR): ( ˙ Mv ∞ √ R ⋆ ) × L ⋆ . In each of these plots, we include theresults of Martins et al. (2005) regarding eleven O dwarfs, aswell as the data gathered by Mokiem et al. (2007) regarding Oand B dwarfs, giants, and supergiants (see their Table A1). In or- der to simplify the analysis and exclude metallicity e ff ects, onlyGalactic objects are considered.In the plots presented in Figure 11, we can perceive the samebasic result: the stars of our sample gather around the sameplaces occupied by the four O8-9V stars studied in Martins etal. (2005), namely, µ Col, AE Aur, HD 46202, and HD 93028.This means that: (i) we have also found a discrepancy of roughlytwo orders of magnitude between the measured and the predictedmass-loss rates for our programme stars; (ii) the modified windmomentum-luminosity relation indeed shows a breakdown or asteepening below log L ⋆ / L ⊙ ∼ . L ⋆ / L ⊙ > . M x log ˙ M Vink plot andbelow the fit to the data of Mokiem et al. (2007), they do notpresent a significant discrepancy if clumping is neglected .The results derived by us and Mokiem et al. (2005) regard-ing the star ζ Oph are connected in Figure 11. The large dif-ference observed is due to the di ff erent mass-loss rates derived.While we have log ˙ M = − .
80 from the C iv λλ M = − .
84 based on H α , us-ing the FASTWIND code. The other stellar and wind parametersdetermined in their study and ours are very similar. The radius, T e f f , and log L ⋆ / L ⊙ from this (their) paper are 9.2 (8.9) R ⊙ , 32.1(32) kK, and 4.86 (4.88), respectively. Regarding the wind ter-minal velocity, we (they) obtain 1500 (1550) km s − .As we mentioned above, the objects of our sample and thefour O8-9V stars analyzed in Martins et al. (2005) occupy aboutthe same locus in the panels shown in Figure 11. It is interestinghowever to analyze their results for other stars having also lowluminosities, i.e., with log L ⋆ / L ⊙ . .
2. Two objects in Martinset al. (2005) have a log L ⋆ / L ⊙ of about 5.2 ( ± . L ⋆ / L ⊙ ∼ .
8. They fall relatively far from the O8-9V groupin the plots in Figure 11. Contrary to the other four earlier Odwarfs, clumping was not used in their analysis. These three ex-amples are suggestive that in stars of the spectral types O6.5V, We recall that the theoretical models of Vink et al. (2000) do notinclude clumping.2 Marcolino et al.: Analysis of late-type O dwarfs
Fig. 10.
Optical spectrum of HD 216532 (green / light gray line) and our final model (black / dashed line).O7V, and O7.5V, a decrease of the wind strength starts to beseen, culminating in the O8-9V classes.Also at low luminosities are the following objects analyzedby Mokiem et al. (2005): Cyg OB2 τ Sco (B0.2V),10 Lac (O9V), and HD 217086 (O7Vn) (see Figure 11; filledsquares). They deserve special attention. In their case, theirmass-loss rates clearly present better agreement to the theoret-ical predictions and also allow them to follow relatively well theWLR. Interestingly, their ˙ M ’s were obtained from the H α line.Given these and our UV based results, the origin of the weakwind problem can be thought to reside in the di ff erent mass-lossrates diagnostics employed. In fact, Mokiem et al. (2007) havesuggested that the problem could be due to uncertainties in theUV method. We address and discuss these questions during therest of the paper.We emphasize that there are several other O8-9V stars withUV spectra similar to the ones found in our sample. Thus, itis likely that they possess similar wind properties, i.e., that thesame basic results we have obtained will be achieved for them ifwe use the same methods of analysis.
6. The Carbon Abundance
From the models, it can be easily seen that changes in the carbonabundance ( ǫ C ) a ff ect the C iv λλ ff erent combinations of ˙ M and ǫ C we can get satis-factory fits to this feature, as long as its optical depth is kept con-stant, i.e., at the observed value. This can be seen more directlyin terms of the Sobolev optical depth, which is proportional tothe mass-loss rate, the C iv ionization fraction ( q CIV ), and theabundance: τ CIV ∝ ˙ M q
CIV ǫ C (see for example Lamers et al.1999). In the context of the weak wind problem, this suggeststhat we could fit C iv λλ M and ǫ C . In orderto achieve this goal, we have built a small grid of models withthe following ǫ C /ǫ C ⊙ ratios: 0.2, 0.5, 1.0, 1.5, and 3 (mass frac-tions). Thereafter, we have analyzed the behavior of the follow-ing photospheric transitions: C iii λ λ iv λλ ǫ C , anaccurate determination of the carbon abundance is not feasibleat the moment. arcolino et al.: Analysis of late-type O dwarfs 13 Fig. 11.
Comparison of our results to previous ones in the literature and theoretical predictions. The objects of our sample arerepresented by star symbols; the O dwarfs studied by Martins et al. (2005) by open triangles; the O and B stars compiled byMokiem et al. (2007) that have log L ⋆ / L ⊙ . . M Vink ). The dashed line indicates a one-to-one relation.Right panel: modified wind momentum-luminosity relation (WLR). The solid line represents the best fit to the data in Mokiem et al.(2007). The dotted lines link the results obtained in this paper and in Mokiem et al. (2005) regarding the star ζ Oph. Low luminositydwarfs (log L ⋆ / L ⊙ . .
2) in this paper and in Martins et al. (2005) are explicitly indicated by their spectral types (see text for moredetails).In Figure 12 we present the fits obtained using models with ǫ C = . ǫ C ⊙ (20% solar), ǫ C = ǫ C ⊙ (solar), and ǫ C = ǫ C ⊙ (3 × solar). Although our grid is more complete, we illustrate onlyextreme ǫ C values (besides the solar one) to emphasize the dis-crepancies. From Figure 12, we can note that: (i) when an abun-dance of 0.2 ǫ C ⊙ is used, we cannot fit any of the lines shown; (ii)when a 3 ǫ C ⊙ value is used we cannot fit properly the C iii λ / broad wing appears in the synthetic line andthis is not observed; (iii) the 3 ǫ C ⊙ abundance is also not favoredfrom the fits to C iii λ ζ Oph, and HD216532. From (i) we can conclude that our sample must have a ǫ C > . ǫ C ⊙ . On the other hand, from (ii) and (iii) we are inclinedto conclude that ǫ C < ǫ C ⊙ . Regarding the C iv λλ ǫ C ⊙ upper limit is also suggested from the fits.From our analysis, we estimate the following abundancerange for the O8-9V stars in our sample: 0 . . ǫ C /ǫ C ⊙ . M Vink ). A simple test show that if an abundanceas low as 0.2 ǫ C ⊙ (not favored by our models) is used along with˙ M Vink , the model still present a very intense C iv λλ M , it cannot cause the weak wind problem. Theabundance range obtained does not leave enough room for themass-loss rates to be consistent with ˙ M Vink . A similar conclusionwas previously achieved by Martins et al. (2005), who quanti-fied the error on the mass-loss rate due to the uncertainties in theCNO abundances to be about 0.3dex.
7. Mass-loss Rates in Late-Type O Dwarfs: ˙ M Diagnostics:
In most cases, the C iv λλ M values.We have proceeded in the following manner: for each ob-ject of our sample, we have started with the ˙ M determined fromthe best fit to C iv λλ M Vink - which for each staris a function of T e f f , mass and luminosity. We observed that assoon as we have used values higher than in our final models (yetconsiderably lower than ˙ M Vink ), di ff erent wind profiles startedto appear which are not observed. We have thus used this find-ing to establish upper limits on ˙ M from di ff erent transitions, asdescribed below.First, in Figure 13, we illustrate the ˙ M diagnostics foundby following the procedure aforementioned. Besides C iv λλ v λλ iii λ v λ iv λλ iv λ M is low (i.e. ∼ − -10 − M ⊙ yr − ),N v λ M is increased theygradually have their profiles modified and some develop a veryintense P-Cygni profile when ˙ M Vink is used (see for instance Si iv λλ M upper ) for these ˙ M -sensitive lines wereobtained in di ff erent ways. For P v λλ iii λ iv λλ / or a blue absorption Fig. 12.
Determination of the carbon abundance in our programme stars. The observed spectra are indicated by solid line lines. Themodels shown have: 20% solar (0.2 ǫ C ⊙ ; dotted line), solar ( ǫ C ⊙ ; dashed line), and 3x the solar carbon abundance (3 ǫ C ⊙ ;dashed-dottedline).started to appear, deviating from the (purely photospheric) ob-served ones. For N v λ iv λ ff erent response.Instead of getting filled in its center, an asymmetric bluewardabsorption profile is formed. We have then determined ˙ M upper when a significant (blue) displacement from the observed fea- ture was reached. Such peculiar behavior indicates that this lineis formed in the very inner parts of the stellar wind. In the studydone by Hillier et al. (2003), a similar situation is found in thehomogeneous model for the O7 Iaf + star AV 83, specially re-garding S v λ iv λλ arcolino et al.: Analysis of late-type O dwarfs 15 Fig. 13. ˙ M -sensitive lines in the far-UV and UV regions. A vsini convolution of 200 km s − was applied in each spectrum. Otherphysical parameters (e.g. T e f f and v ∞ ) are held fixed at the values derived for HD 216532 (see Table 3). The models were normalizedand vertically displaced for clarity. Table 4.
Mass-loss Rates and Upper Limits Derived from Di ff erent Transitions (log units). Star ˙ M (C IV) ˙ M (C III) ˙ M (N V) ˙ M (N IV) ˙ M (P V) ˙ M (Si IV) ˙ M (Vink)HD 216898 -9.35 < -7.47 < -8.46 < -7.80 < -7.96 < -7.59 -7.22HD 326329 -9.22 < -7.73 < -8.22 < -7.49 < -7.73 < -7.96 -7.38HD 66788 -8.92 < -7.22 < -8.38 < -7.52 < -7.52 < -7.22 -6.95 ζ Oph -8.80 < -7.40 < -8.15 < -6.89 < -7.40 < -7.70 -6.89HD 216532 -9.22 < -7.45 < -8.21 < -7.57 < -7.45 < -7.70 -6.92 cated are the ones used in our best fits presented in Section 4. Thepredicted mass-loss rates are also shown for comparison. Belowwe present the models used to construct Table 4 and commenton each object. – HD 216898: In Figure A.1 we show the far-UV and UVobserved spectra of this object along with our final model,the models that determine the upper limits in ˙ M from eachline ( ˙ M upper ; not necessarily the same in each panel), and themodel with the Vink’s mass-loss rate. We can clearly notefrom Figure A.1 that: (i) the final model presents the best fitto the observed spectra; (ii) the model with ˙ M Vink presentsthe largest discrepancy; and (iii) the models with the upperlimits for the mass-loss rate do not present satisfactory fits.This is also valid for the other objects discussed below. Wenote that when we have increased ˙ M , the N v λ M Vink . – HD 326329: An analogous plot as shown for HD 216898is presented in Figure A.2 for this object. We observed thatthe N iv λ M in thiscase. Therefore, the upper limit for ˙ M derived from this lineis quite close to ˙ M Vink . For the other lines the situation wasmuch clearer, i.e., the upper limits could be easily obtained. – HD 66788:
In Figure A.3 we show the case of HD 66788.In general, the same line trends found in HD 326329 andHD 216898 are observed. We had no problems in thedetermination of ˙ M upper limits from each line. – ζ Oph:
The observed spectra and models are shown inFigure A.4. Regarding N iv λ M values could fit this feature. By increasing the mass-lossrate however, we noted that the absorption still went blue-shifted (as in the other objects). We conservatively assume a˙ M upper limit from this line equals to the Vink’s mass-loss.Another di ffi cult situation is presented by N v λ M Vink (i.e. with thestrongest mass-loss rate) has a synthetic P-Cygni emissionless intense than the one in the upper limit model. A slightlystronger emission is however found at ∼ ζ Oph (mainly the bluest one),
Fig. 14.
Mass-loss rates upper limits derived from di ff erenttransitions in the ultraviolet. The stars are: HD 216898 (cir-cled + solid line), HD 326329 (triangles + dashed line), HD 66788(squares + dotted line), ζ Oph (asterisk + dashed-dotted line), andHD 216532 (crosses + short-long dashed line). Arrows indicatethat the points represent upper limits.which is interpreted as being due to DACs (Howarth et al.1993; see their Figure 2). The situation for P v λλ iv λλ – HD 216532: In Figure A.5 is shown the models and observedspectra of HD 216532. We note that the C iii λ v λλ iv λλ M Vink . Regarding the P v λλ M can be derived withoutdi ffi culty as these lines turn to emissions when ˙ M is in-creased, and this is not what is observed.In Figure 14 we explore the results presented above in agraphical way. The mass-loss rates obtained are shown relativeto the predicted value for each star, i.e., relative to ˙ M Vink . ForC iv λλ iii λ iv λλ iv λ v λλ v λ we remind that the pointsrepresent only upper limits, i.e., the mass-loss rates must belower than what is displayed .Strikingly, we see from Figure 14 that in all cases the points,i.e., upper limits on the mass-loss rate, fall below the theoreticalline. The only exception is N iv λ ζ Oph,for which we have assigned a conservative upper limit value. Ifwe consider only the C iv λλ v λ M ’s must be less than about ten times the values predicted bytheory ! If we analyze all other transitions together, we see that Fig. 15.
Sensitivity of H α to di ff erent values of the mass-lossrate. The model with the lowest mass-loss has log ˙ M = − . M = − . − . ± .
2) dex. This meansthat the mass-loss rates are still very low: they must be less thanabout two to five times ˙ M Vink ! Taken together, these results bringfor the first time additional support to the reality of weak winds.If we focus only on one object, it is also interesting to notein Figure 14 that the points below the highest point (i.e. closestto Vink’s line) have the respective theoretical lines already indisagreement with the observed ones. In this sense, wheneverwe skip to models / points higher than others, Figure 14 can beinterpreted as errors being accumulated in the model fits.We remind that all the models presented in this paper do notinclude clumping. Therefore, some of the upper limits presentedabove could be in fact much lower if the wind of the objectsstudied here are not homogeneous. In this case, we could thenhave that: log ˙ M clumped ˙ M Vink = log ˙ M ˙ M Vink + log p f , where the first term in the right-hand side is in the verticalaxis shown in Figure 14. With a typical filling factor f = . α behavior: Although we have determined mass-loss rates from the far-UVand UV spectral regions, we have also analyzed the behavior of arcolino et al.: Analysis of late-type O dwarfs 17 H α to changes in ˙ M . This is an important point because as wehighlighted in Section 5, the use of H α by Mokiem et al. (2005)seems to indicate for a few stars that there are no weak winds,i.e., the ˙ M values obtained are compatible with ˙ M Vink (see alsothe discussion in Mokiem et al. 2007).Previously, the work of Martins et al. (2005) had alreadydemonstrated two things regarding this line: (i) in a low den-sity wind ( ˙ M ∼ − M ⊙ yr − ) the CMFGEN and FASTWINDpredictions agree; (ii) for ˙ M changes from ∼ − to 10 − M ⊙ yr − the resulting CMFGEN H α profiles are virtually identical.Here, we extend their analysis by presenting the changes in theH α profile for a large range in mass-loss: from ∼ − to 10 − M ⊙ yr − .In Figure 15 we plot models with the photospheric parame-ters fixed and di ff erent values for the mass-loss rate. First, wenote that only with a log ˙ M = − .
30 the H α line turned to emis-sion. For lower rates, the predictions are in absorption and there-fore the measurement of ˙ M in an observed spectrum must bedone based in the analysis of the wind filling of the photosphericprofile. In the example shown, models with a log ˙ M = − . M = − .
46, log ˙ M = − .
20 do not di ff er much. Despite afactor of one hundred in ˙ M , the changes in the profiles are toosmall to allow a secure discrimination of a best fit model to anobserved spectrum (see below). Overall, the models suggest thatour confidence in mass-loss rates derived from H α profiles islimited to values & − M ⊙ yr − . As it can be seen in Figure 15,only beyond this threshold noticeable changes start to be found.In Figure 16 we turn our attention to the stars of our sam-ple. The fits achieved to H α using our final models (i.e. with˙ M obtained from C iv λλ M Vink arepresented (see Table 3 for the corresponding values). First, if wefocus on HD 216898 and HD 66788, neither our final modelsnor the ones using ˙ M Vink match the observed H α . The modelsusing ˙ M Vink are weaker while our final models are stronger thanthe observed line center. However, the di ff erences compared tothe observations are very small: ∼ α , we find very hard to choosewith certainty between any of these models (and also the ones inbetween, i.e., with intermediate ˙ M ’s). As in Martins et al. (2005),we estimate that the position of the line core can have an errorup to 2%. However, if a nebular contamination is likely or thesignal-to-noise ratio of the spectra is not high enough, the uncer-tainty increases considerably. Such situations illustrate that it isnot straightforward to establish ˙ M values for O8-9V stars usingH α . For HD 326329 the situation is similar, but the ˙ M Vink modelpresents a better fit than our final model.Interestingly, for ζ Oph and HD 216532, our final modelsseem to be better than the ones using ˙ M Vink , and not only inthe line center. To use the predicted mass-loss rates for thesestars thus means that we can fit neither the UV nor the opti-cal lines! Regarding ζ Oph, Repolust et al. (2004) and Mokiemet al. (2005) have presented ˙ M measurements from H α basedon the FASTWIND code. The former authors derived an upperlimit of log ˙ M = − .
74 and the latter found a mass-loss rate oflog ˙ M = − .
85. In both studies, the β parameter used for thevelocity law was fixed at 0.8. Although the fit in Figure 16 uses β =
1, we have tested a model with the same predicted mass-loss (log ˙ M Vink = − .
89) and β = .
8. The synthetic H α linegets deeper, but it is still not enough to fit the observations. Wetherefore do not confirm the findings of Mokiem et al. (2005). In this example vsini =
80 km s − and the others parameters arefixed at the values derived for one of our stars, namely, HD 216532. We speculate that one reason for this can be the di ff erent dataset used and their reduction. Another alternative is the existenceof a discrepancy between the predicted H α profile by CMFGENand FASTWIND at ∼ − M ⊙ yr − . This is a possibility sincein the comparison made by Puls et al. (2005) we can note thatthese two codes can predict slightly di ff erent H α line intensities.In the case of an O8V or a O10V model for example (see theirFig. 17), FASTWIND tends to present a stronger H α absorptionthan CMFGEN. Such di ff erences can a ff ect mass-loss rate mea-surements for late O dwarfs obtained from this line.
8. Discussion:
It is clear from the previous section that when high mass-loss rates are used, i.e. above the values derived from C iv λλ Most impor-tantly, this is seen for lines of di ff erent ions, for each object of oursample . With these di ff erent diagnostics, the existence of weakwinds gains strong additional support. Given the importance ofthis result, we now discuss the validity of our models.What are the consequences if we still accept the mass-lossrates predicted by Vink et al. (2000) for late O dwarfs as thecorrect ones ? Given our findings, we must conclude that theexpanding atmosphere models used (from CMFGEN) lack newphysics or have incorrect assumptions or approximations (orboth). Even with high mass-loss rates (i.e. equal or close to˙ M Vink ) the models should provide a reasonable fit to the observedspectra, where only a few wind lines are seen. The current windstructure / ionisation then must be changed in a drastic way fora model to predict only a weak C iv λλ v λ M (about one hundred times higher than the ones in this paper oreven more!). One analogy that can be done is that the uppermostsynthetic spectrum shown in Figure 13 would have to turn to theone essentially photospheric (bottom). Although possible, thisscenario presents several di ffi culties.First and foremost, contrary to the recently reported case ofsome B supergiants, where a few key UV wind lines could not bereproduced by CMFGEN models (see Searle et al. 2008 and alsoCrowther et al. 2006) , we do find agreement with the observa-tions (see Figs. 1-10). Although in most spectra only one con-spicuous wind signature is observed, the lack of wind lines suchas P v λλ iii λ iv λλ iv λ In this case, a problem in the ionization structure of the models isclaimed. In fact, their wind structure seems to be distinct from the onesfound in O stars (Prinja et al. 2005). We note however, that X-rays werenot used in their analysis and might resolve some of the discrepancies.8 Marcolino et al.: Analysis of late-type O dwarfs
Fig. 16.
Fits to the observed H α lines in our sample using the mass-loss rates of our final models and the mass-loss rates predictedby theory ( ˙ M Vink ).observed spectra similar to ours, with perhaps more free inputparameters or assumptions (e.g. as in non-spherical winds).There is no clear answer as for the missing ingredient(s)in the current atmosphere models. For massive stars in gen-eral, it is true that although CMFGEN and other codes such asFASTWIND are considered state-of-art , di ff erent physical phe-nomena are not taken into account properly or are not taken intoaccount at all due to several technical challenges and lack of ob-servational constraints (e.g. non-stationarity; wind rotation; non-sphericity; realistic description of clumping; see Hillier 2008and references therein). It is possible that advances in some ofthese topics can cast a new light in the analysis of massive stars.However, the consequences regarding the weak wind problemare di ffi cult to foresee. In short, we believe that the inclusion of new physics or re-laxation of standard assumptions certainly deserves to be ad-dressed in future studies (with the appropriate observationalconstraints), but our atmosphere models for the O8-9V stars arethe best working hypothesis currently available.
Below we speculate about some possibilities worth to be in-vestigated, and thereafter present the hypothesis of highly (X-rays) ionized winds.
From the best fit models, we can examine how closely the mo-mentum absorbed by the wind matches that required to satisfythe momentum equation. In general, for the O8-9V stars, wefind that the wind absorbs too much momentum. This can beillustrated by noting that the mass-loss rate from a single satu-rated line is approximately L / c (Lucy & Solomon 1970). In all cases, this value exceeds our best fit mass-loss rates (see Table3). While none of the wind lines in the models are saturated,there are su ffi cient weak lines to yield a force similar, or larger,than the single line limit.It is unclear how this discrepancy can be removed.Possibilities include the existence of magnetic fields, and / or thepresence of a significant component of hot gas. The latter couldbe generated by shocks in the stellar wind. As noted by Drew etal. (1994) and Martins et al. (2005), it is possible that at the den-sities encountered in these winds the shocked gas never cools(see also Krtiˇcka & Kub´at 2009). Such gas would not be eas-ily detectable in the UV since it could be collisionally ionizedto such an extent that very few C IV and N V ions would ex-ist. Further evidence for a significant amount of hot gas comesfrom the observed X-ray fluxes; the required filling factors inthe models (while strongly mass-loss rate dependent) indicatethat the hot gas in the wind is not a trace component.Of interest in this regard are the distinct profiles presentedby C iv λλ ff er significantly among the stars (see forinstance the ones in HD 66788, HD 326329, and HD 216532).Also of concern is the general weakness of the P-Cygni emissioncomponents (with the possible exception of ζ Oph). Instead ofvery low mass-loss, such an absence might be related to compli-cated wind structures.
Although good fits could be achieved with very low ˙ M ’s, wehave still studied some alternatives to have agreement with theobservations using ˙ M Vink . It became soon clear that the most arcolino et al.: Analysis of late-type O dwarfs 19
Fig. 17. E ff ects of di ff erent log L X / L Bol ratios in the ultraviolet synthetic spectrum. Note the similarities between the middle andbottom spectra, despite the very di ff erent mass-loss rates.promising option was to explore high levels of X-rays emission.If most of the wind is ionized, i.e. is in the form of a hot plasma,we do not expect significant wind emission in the UV lines.In order to test the idea above, we explored models with highvalues for the L X / L Bol ratio. We have found that indeed the windemission is decreased in the desired manner. To illustrate this,we present in Figure 17 two models with the same (typicallyVink) mass-loss rate but with di ff erent values for log L X / L Bol :the canonical (-7.0) and a much higher ratio (-3.8). In addition,a third model with a very low mass-loss rate (log ˙ M = − .
3; asin a typical final model) is shown with a log L X / L Bol = − . L X / L Bol = − . M andlog L X / L Bol resembles the one with very low ˙ M and normal log L X / L Bol (see the spectra in the middle and bottom of Figure 17).In order to better quantify the result above, we have turnedour attention to the ionisation fractions of P v , N v , Si iv , andC iv . For the resonance lines P v λλ v λ iv λλ iv λλ M × q i - which means to find their cor-rect / observed optical depths (see for instance the case of P v λλ :˙ M × q i = Ψ i , (1)where Ψ i is a constant that depends on the observed spectrumand q i is the ionisation fraction of the ion i . This latter is usuallydefined as: q i = R . n i ( x ) dx R . n ( x ) dx , (2) It is important to note that if the ionisation fractions are reli-able / correct, we get ˙ M from the model fits. Otherwise, what is actuallyobtained is Ψ i . where x = v ( r ) / v ∞ and n i and n are the ion and elementnumber density, respectively. We can now compute what is theionisation fraction needed to fit the observations by using ˙ M Vink ,which we call q i , Vink . Since the constant Ψ i gives the appropriatefit to the observations, we can write: q i , Vink = Ψ i / ˙ M Vink . (3)After we found q i , Vink from the equation above, for each ion,we have computed several models with the mass-loss rate fixedat ˙ M Vink but with di ff erent values of log L X / L Bol . For each model,we have derived the ionisation fraction of each ion followingEqn. 2. We call these ionisation fractions q X , i.e., the ionisationfractions for a specific log L X / L Bol value. We have then verifiedwhich one of the several q X was equal to q i , Vink , in order to findfor which amount of X-rays we would have agreement with theobservations using ˙ M Vink .In Figure 18 we plot the ratio q i , Vink / q X for each ion versusseveral log L X / L Bol values. For simplicity, we illustrate the caseof only one object of our sample, HD 216898. The first thingto note is that the ions do not behave exactly in the same man-ner. This is not so surprising, since they have di ff erent electronicstructures and ionization potentials, and thus react di ff erently toX-rays. If we focus on C iv and P v , we can see that only whenvery high log L X / L Bol ratios are considered (around -3.0), q X getsclose to q i , Vink . Otherwise, their ratio is much lower than unity.In the case of Si iv , q i , Vink / q X remains very small regardless thevalue of log L X / L Bol . However, an increase is observed when weuse a log L X / L Bol near -3.0. For a log L X / L Bol ∼ − q i , Vink / q X ∼ × − . For log L X / L Bol ∼ −
3, this ratio ismuch higher, q i , Vink / q X ∼ . q X reach q i , Vink . The situation for N v is quite interesting: for both lowand high values of log L X / L Bol , the q X approaches q i , Vink . Thisis however, not hard to explain. For high values of log L X / L Bol ,we observed that practically all nitrogen is concentrated in N vi .Thus, the ionisation fraction q X obtained for N v is low as theionisation fraction of Vink, q Vink , computed from Eqn. 3. On theother hand, when very low log L X / L Bol ratios are considered, we
Fig. 18.
Ionisation fractions as a function of log L X / L Bol . Thevalues for q X were computed using Eqn. 2 and ˙ M Vink . The q i , Vink values were derived from Eqn. 3 (see text for more details).verified that most of the nitrogen is in N iii - iv . Thus, again, avery low q X is obtained for N v . In practice, this means that theobserved, weak N v λ M Vink with either very low or largeamounts of X-rays.Our conclusion from both Figures 17 and 18 is that theonly way to have high mass-loss rates ( ˙ M Vink ) and still agreewith the observed spectra is by using a log L X / L Bol & − . L X / L Bol values measured are normally about -7.0(see e.g. Sana et al. 2006) with a scatter of about ± .
0. It is en-couraging however, to explore in future studies di ff erent scenar-ios of X-rays emission in stellar winds.
9. Conclusions:
We have analyzed a sample of five late-type Galactic O dwarfsby using atmosphere models from the CMFGEN and TLUSTYcodes (HD 216898, HD 326329, HD 66788, ζ Oph, and HD216532). Model fits for the far-UV, UV and optical observedspectra were presented, from which stellar and wind parameterswere obtained. The mass-loss rates were obtained first using theC iv λλ ⊲ The stellar parameters obtained for our sample are quitehomogeneous (see Table 3). Surface gravities of about3.8 (after correcting for vsini ) to 4.0 and T e f f ’s of 30 to34 kK were obtained. These values show a good agreementwith the latest calibrations of Galactic O star parameters,regarding the spectral types of the programme stars. ⊲ By using the C iv λλ L ⋆ / L ⊙ . . ⊲ We have investigated the carbon abundance based on a setof UV (C iii λ . . ǫ C /ǫ C ⊙ . ⊲ We have explored di ff erent ways (besides using C iv λλ v λλ iii λ v λ iv λλ iv λ iv λλ v λ M upper limit rates. The results obtained show that the˙ M values must be less than about -1.0 dex compared to˙ M Vink . By considering the other lines, we still find very lowmass-loss rates. The results obtained show that ˙ M must be less than about ( − . ± .
2) dex compared to ˙ M Vink . Theybring additional support to the reality of weak winds. ⊲ Upper mass-loss rate limits derived from N iv λ iii λ iv λ iii λ ⊲ We have analyzed the H α line and we observed that itsprofile is insensitive to ˙ M changes when we are in the < − M ⊙ yr − regime. For some objects of our sample itis uncertain to choose between the H α fit presented by ourfinal models and the ones using ˙ M Vink . The interpretationof the results is hindered by uncertainties in the continuumnormalization of echelle spectra. Such situation gets evenmore complicated if a contamination by nebular emissionis likely. For the stars HD 216532 and ζ Oph, the fits tothe observed H α lines using ˙ M Vink were not satisfactory.This result shows that even when H α is used, lower thanpredicted mass-loss rates can be preferred. ⊲ We have investigated ways to have agreement with the ob-served spectra of the O8-9V stars using the mass-loss ratespredicted by theory ( ˙ M Vink ). The only mechanism plausiblethat we found is X-rays. By using high values for the log L X / L Bol ratio, we could observe that very few wind emis-sions takes place, as in models with very low mass-loss rates( ∼ − − − M ⊙ yr − ). However, the values needed tobe used (log L X / L Bol & − .
5) are not supported by the ob-servations, which usually measure log L X / L Bol values near-7.0. arcolino et al.: Analysis of late-type O dwarfs 21
Although our analysis was performed with state-of-art at-mosphere models, there are a couple of issues which still needto be addressed in future studies. For example, some stars ofour sample (HD 326329 and HD 216532) present a deep C iv λλ ζ Oph can besolved. Furthermore, a more realistic treatment or even alterna-tive descriptions of the X-rays emission in the atmosphere mod-els (e.g. in conformity with the magnetic confinement scenario;Ud-Doula & Owocki 2002) could bring valuable informations.First steps in this direction have been taken by Zsarg´o et al.(2009). As we have shown in Section 8.2, X-rays are an e ffi -cient mechanism to change the ionisation structure of the stellarwinds.It would be also very useful to analyze a sample of O starshaving log L ⋆ / L ⊙ near ∼ .
2. The disagreement between at-mosphere models and theoretical predictions for Galactic starsseems to start around this value. The candidate spectral typeswould be O6.5V, O7V, and O7.5V. Also, this same questionneeds to be better studied in metal-poor environments, i.e. in theLMC and SMC. We intend to investigate these issues in a futurepaper.From the point of view of the hydrodynamics, there are alsoimportant questions that could be addressed. In the work of Vinket al. (1999; 2000; 2001), a predicted mass-loss rate is computedbased on a set of ISA-WIND models (de Koter et al. 1997) plusthe use of a Monte-Carlo technique to compute the radiative ac-celeration ( g L ). Although the results obtained by their procedureare self-consistent, it should be kept in mind that the ISA-WINDatmosphere models do not include the e ff ect of X-rays. As thewind ionization can be changed significantly, new radiative ac-celerations can be found and perhaps new predicted mass-lossrates might be derived. Within this picture, the agreement withthe Vink et al. predictions for early O dwarfs may be understood,since their winds are not seriously a ff ected by X-rays.Our results bring new constraints to the weak wind problem.Galactic O8-9V stars seem to present very low mass-loss rates,as indicated by the C iv λλ Acknowledgements.
W. M. acknowledges the travel grant provided by IAU(Exchange of Astronomers Programme) and CNES for the postdoctoral fel-lowship. J-CB acknowledges financial support from the French NationalResearch Agency (ANR) through program number ANR-06-BLAN-0105. TLand DJH were supported by the NASA Astrophysics Data Program (grantNNG04GC81G). We wish to thank Ya¨el Naz´e for helping with the normalizationof the spectrum of HD 216898. We also wish to thank an anonymous referee foruseful comments which helped to improve the paper. This research has made useof the SIMBAD database, operated at CDS, Strasbourg, France.
References
Abbott, D. C., 1982, ApJ, 263, 723Baranne, A., et al., 1996, A&AS, 119, 373Baume, G., V´azquez, R. A., & Feinstein, A., 1999, A&AS, 137, 233Bouret, J. -C., Lanz, T., Hillier, D. J., Heap, S. R., Hubeny, I., Lennon, D. J.,Smith, L. J., Evans, C. J., 2003, ApJ, 595, 1182Bouret, J. -C., Lanz, T., Hillier, D. J., 2005, A&A, 438, 301Cardelli, J. A., Clayton, G. C., & Mathis, J. S., 1988, ApJ, 329, L33 Cohen, D. H., 2008, Massive Stars as Cosmic Engines, Proceedings of theInternational Astronomical Union, IAU Symposium, Volume 250, 17Crowther, P. A., Lennon, D. J., & Walborn, N. R., 2006, A&A, 446, 279de Koter, A., Heap, S. R., & Hubeny, I., 1997, ApJ, 477, 792Drew, J. E., Hoare, M. G., & Denby, M., 1994, MNRAS, 266, 917Ebbets, D., 1981, PASP, 93, 119Escolano, C., et al., 2008, in prep.Fr´emat, Y., Zorec, J., Hubert, A. -M., & Floquet, M., 2005, A&A, 440, 305Freyer, T., Hensler, G., & Yorke, H. W., 2003, ApJ, 594, 888Fullerton, A. W., Massa, D. L., & Prinja, R. K., 2006, ApJ, 637, 1025Garc´ıa, B., & Mermilliod, J. C., 2001, A&A, 368, 122Garmany, C. D., & Stencel, R. E., 1992, A&AS, 94, 211Garrison, R. F., 1970, AJ, 75, 1001Grevesse, N., & Sauval, A., 1998, Space Sci. Rev., 85, 161Hillier, D. J., 2008, Massive Stars as Cosmic Engines, Proceedings of theInternational Astronomical Union, IAU Symposium, Volume 250, 89Hillier, D. J., & Miller, D. L., 1998, ApJ, 496, 407Hillier, D. J., et al., 2003, ApJ, 588, 1039Howarth, I. D., Prinja, R. K., & Willis, A. J., 1984, MNRAS, 208, 525Howarth, I. D., et al., 1993, ApJ, 417, 338Howarth, I. D., Siebert, K. W., Hussain, G. A. J., & Prinja, R. K., 1997, MNRAS,284, 265Howarth, I. D., & Smith, K. C., 2001, MNRAS, 327, 353Hubeny, I., & Lanz, T., 1995, ApJ, 439, 875Jankov, S., Janot-Pacheco, E., & Leister, N. V., 2000, ApJ, 540, 535Kaltcheva, N. T., Hilditch, R. W., 2000, MNRAS, 312, 753Kaufer, A., Stahl, O., Tubbesing, S., Norregaard, P., Avila, G., Francois, P.,Pasquini, L., & Pizzella, A., 1999, Messenger, 95, 8Krtiˇcka, J., & Kub´at, J., 2009, astro-ph:0901.0223Kudrizki, R. -P., Puls, J., 2000, ARA&A, 38, 613Lamers, H. J. G. L. M., Haser, S., de Koter, A., & Leitherer, C., 1999, ApJ, 516,872Lanz, T., & Hubeny, I., 2003, ApJS, 146, 417Lanz, T., & Hubeny, I., 2007, ApJS, 169, 83Leitherer, C., 1988, ApJ, 326, 356Lesh, J. R., 1968, ApJS, 17, 371Lucy, L. B., & Solomon, P. M., 1970, ApJ, 159, 879MacConnell, D. J., & Bidelman, W. P., 1976, AJ, 81, 225Macfarlane, J. J., Cohen, D. H., & Wang, P., 1994, ApJ, 437, 351Ma´ız-Apell´aniz, J., Walborn, N. R., Galu´e, H. A., & Wei, L. H., 2004, ApJS,151, 103Martins, F., Schaerer, D., Hillier, D. J., 2002, A&A, 382, 999Martins, F., Schaerer, D., Hillier, D. J., & Heydari-Malayeri, M., 2004, A&A,420, 1087Martins, F., Schaerer, D., Hillier, D. J., Meynadier, F., Heydari-Malayeri, M., &Walborn, N. R., 2005, A&A, 441, 735Martins, F., Schaerer, D., Hillier, D. J., 2005b, A&A, 436, 1049Massey, P., 2003, ARA&A, 41, 15Mathys, G., Andrievsky, S. M., Barbuy, B., Cunha, K., & Korotin, S. A., 2002,A&A, 387, 890Meynet, G., & Maeder, A., 2000, A&A, 361, 101Mokiem, M. R., de Koter, A., Puls, J., Herrero, A., Najarro, F., & Villamariz, M.R., 2005, A&A, 441, 711Mokiem, M. R., de Koter, A., Vink, J. S., et al., 2007, A&A, 473, 603Moultaka, J., Ilovaisky, S. A., Prugniel, P., Soubiran, C., 2004, PASP, 116, 693Niemela, V. S., & M´endez, R. H., 1974, ApJ, 187, L23Penny, L. R., 1996, ApJ, 463, 737Prinja, R. K., Massa, D., & Searle, S. C., 2005, A&A, 430, L41Puls, J., 2008, Massive Stars as Cosmic Engines, Proceedings of the InternationalAstronomical Union, IAU Symposium, Volume 250, 25Puls, J., Urbaneja, M. A., Venero, R., Repolust, T., Springmann, U., Jokuthy, A.,& Mokiem, M. R., 2005, A&A, 435, 669Puls, J., Markova, N., Scuderi, S., Stanghellini, C., Taranova, O. G., Burnley, A.W., & Howarth, I. D., 2006, A&A, 454, 625Oskinova, L., 2005, MNRAS, 361, 679Oskinova, L., Feldmeier, A., & Hamann, W. -R., 2006, MNRAS, 372, 313Owocki, S. P.; Castor, J. I., & Rybicki, G. B., 1988, ApJ, 335, 914Reed, B. C., 1993, PASP, 105, 1465Reid, A. H. N., et al., 1993, ApJ, 417, 320Repolust, T., Puls, J., & Herrero, A., 2004, A&A, 415, 349Sana, H., Rauw, G., Naz´e, Y., Gosset, E., & Vreux, J.-M., 2006, MNRAS, 372,661Sana, H., Naz´e, Y., O’Donnell, B., Rauw, G., & Gosset, E., 2008, NewAstronomy, 13, 202Schild, R. E., Hiltner, W. A., & Sanduleak, N., 1969, ApJ, 156, 609Searle, S. C., Prinja, R. K., Massa, D., & Ryans, R., 2008, A&A, 481, 777Smith, N., & Owocki, S. P., 2006, ApJ, 645, L45Snow, T. P., & Jenkins, E. B., 1977, ApJSS, 33, 269
Snow, T. P., Lamers, H. J. G. L. M., Lindholm, D. M., & Odell, A. P., 1994,ApJS, 95, 163Ud-Doula, A., & Owocki, S. P., 2002, ApJ, 576, 413Vacca, W. D., Garmany, C. D., Shull, J. M., 1996, ApJ, 460, 914van Marle, A. J., Owocki, S. P., & Shaviv, N. J., 2008, MNRAS, 389, 1353Villamariz, M. R., & Herrero, A., 2005, A&A, 442, 263Vink, J. S., de Koter, A., Lamers, H. J. G. L. M., 1999, A&A, 350, 181Vink, J. S., de Koter, A., Lamers, H. J. G. L. M., 2000, A&A, 362, 295Vink, J. S., de Koter, A., Lamers, H. J. G. L. M., 2001, A&A, 369, 574Walborn, N. R., 1973, AJ, 78, 1067Walborn, N. R., & Panek, R. J., 1984, ApJ, 286, 718Walker, G. A. H., Yang, S., Fahlman, G. G., 1979, ApJ, 233, 199Walker, G. A. H., et al., 2005, ApJ, 623, L145Woosley, S. E., & Bloom, J. S., 2006, ARA&A, 44, 507Zsarg´o, J., et al., 2009, in prep. arcolino et al.: Analysis of late-type O dwarfs , Online Material p 1
Online Material arcolino et al.: Analysis of late-type O dwarfs , Online Material p 2
Appendix A: ˙ M upper limits: In this Section, we present the model fits used to establish upperlimits on the mass-loss rate from di ff erent UV lines. arcolino et al.: Analysis of late-type O dwarfs , Online Material p 3 F i g . A . . U pp e r li m it s f o r t h e m a ss - l o ss r a t e i n HD . arcolino et al.: Analysis of late-type O dwarfs , Online Material p 4 F i g . A . . U pp e r li m it s f o r t h e m a ss - l o ss r a t e i n HD . arcolino et al.: Analysis of late-type O dwarfs , Online Material p 5 F i g . A . . U pp e r li m it s f o r t h e m a ss - l o ss r a t e i n HD . arcolino et al.: Analysis of late-type O dwarfs , Online Material p 6 F i g . A . . U pp e r li m it s f o r t h e m a ss - l o ss r a t e i n ζ O ph . arcolino et al.: Analysis of late-type O dwarfs , Online Material p 7 F i g . A . . U pp e r li m it s f o r t h e m a ss - l o ss r a t e i n HD216532