Table Of ContentAstronomy&Astrophysicsmanuscriptno. esoagbsai_lang_ed (cid:13)cESO2017
January27,2017
The VLTI/MIDI(cid:63) view on the inner mass loss of evolved stars from
the Herschel MESS sample
C.Paladini1,2,D.Klotz2,S.Sacuto2,3,E.Lagadec4,M.Wittkowski5,A.Richichi6,J.Hron2,A.Jorissen1,M.A.T.
Groenewegen7,F.Kerschbaum2,T.Verhoelst8,G.Rau2,H.Olofsson9,R.Zhao-Geisler10,andA.Matter4
1 Institutd’Astronomieetd’Astrophysique,UniversitélibredeBruxelles,CP226,BoulevardduTriomphe,1050Brussels,Belgium;
e-mail:[email protected]
2 UniversityofVienna,DepartmentofAstrophysics,Türkenschanzstrasse17,1180Wien,Austria
3 University of Uppsala, Department of Physics and Astronomy, Division of Astronomy and Space Physics, Box 516, 75120,
Uppsala,Sweden
7 4 LaboratoireLagrange,UniversitéCôted’Azur,ObservatoiredelaCôted’Azur,CNRS,Blvddel’Observatoire,CS34229,06304
1 Nicecedex4,France
0 5 ESO,Karl-Schwarzschild-Str.2,85748GarchingbeiMünchen,Germany
2 6 NationalAstronomicalResearchInstituteofThailand, 191SiriphanichBldg., HuayKaewRd., Suthep, Muang, 50200Chiang
Mai,Thailand
n
7 KoninklijkeSterrenwachtvanBelgië,Ringlaan3,1180Brussel,Belgium
a
8 BelgianInstituteforSpaceAeronomy(BIRA-IASB),Ringlaan-3-AvenueCirculaire,B-1180Brussels,Belgium
J
9 OnsalaSpaceObservatory,Dept.ofEarthandSpaceSciences,ChalmersUniversityofTechnology,43992Onsala,Sweden
6 10 NationalTaiwanNormalUniversity,DepartmentofEarthSciences,88Sec. 4,Ting-ChouRd,WenshanDistrict,Taipei,11677,
2
Taiwan,ROC
]
R
Received;accepted
S
.
h ABSTRACT
p
- Context.Themass-lossprocessfromevolvedstarsisakeyingredientforourunderstandingofmanyfieldsofastrophysics,including
o
stellarevolutionandthechemicalenrichmentoftheinterstellarmedium(ISM)viastellaryields. Nevertheless,manyquestionsare
r
t stillunsolved,oneofwhichisthegeometryofthemass-lossprocess.
s Aims.TakingadvantageoftheresultsfromtheHerschelMasslossofEvolvedStarS(MESS)programme,weinitiatedacoordinated
a efforttocharacterisethegeometryofmasslossfromevolvedredgiantsatvariousspatialscales.
[
Methods.ForthispurposeweusedtheMID-infraredinterferometricInstrument(MIDI)toresolvetheinnerenvelopeof14asymptotic
2 giantbranchstars(AGBs)intheMESSsample.Inthiscontributionwepresentanoverviewoftheinterferometricdatacollectedwithin
v theframeofourLargeProgramme,andwealsoaddarchivedataforcompleteness.Westudiedthegeometryoftheinneratmosphere
7 bycomparingtheobservationswithpredictionsfromdifferentgeometricmodels.
0 Results.Asymmetriesaredetectedforthefollowingfivestars:RLeo,RTVir,π1Gruis,omiOri,andRCrt.AlltheobjectsareO-rich
4 orS-type,suggestingthatasymmetriesintheNbandaremorecommonamongstarswithsuchchemistry.Wespeculatethatthisfact
5 isrelatedtothecharacteristicsofthedustgrains. Exceptforonestar,nointerferometricvariabilityisdetected,i.e. thechangesin
0 sizeoftheshellsofnon-mirastarscorrespondtochangesofthevisibilityoflessthan10%.Theobservedspectralvariabilityconfirms
. previousfindingsfromtheliterature.ThedetectionofdustinoursamplefollowsthelocationoftheAGBsintheIRAScolour-colour
1
diagram: more dust is detected around oxygen-rich stars in region II and in the carbon stars in region VII. The SiC dust feature
0
doesnotappearinthevisibilityspectrumoftheUAntandSSct,whicharetwocarbonstarswithdetachedshells. Thisfindinghas
7
implicationsforthetheoryofSiCdustformation.
1
: Key words. Stars: late-type – Stars: AGB and post-AGB – Stars: mass loss – Circumstellar matter – Technique: high angular
v
resolution–Technique:interferometric
i
X
r
a1. Introduction evenathighredshift;Valianteetal.2009),anditcontributesto
buildingnewgenerationsofstarsandplanets.
Most of the material processed during the lifetime of low- to
intermediate-mass stars is returned to the interstellar medium The general picture that explains the mass-loss process as-
(ISM)duringtheasymptoticgiantbranch(AGB)stage.Thisma- sumes that stellar pulsation triggers shock waves in the atmo-
terial is crucial for the chemical evolution of galaxies (perhaps sphere. These shocks lift the gas above the stellar surface, cre-
ating dense cool layers where dust may form. Depending on
thechemistryofthestar,theorypredictsthattheradiationpres-
(cid:63) BasedonobservationscollectedattheEuropeanOrganisationfor
sure on dust or the scattering on micron-size dust grains drive
Astronomical Research in the Southern Hemisphere under ESO pro-
grammes 073.D-0711, 076.D-0620, 077.D-0294, 078.D-0122, 080.D- the stellar material away (Höfner & Dorfi 1997; Woitke 2006;
0801,081.D-0021,083.D-0234,086.D-0737,086.D-899,187.D-0924, Höfner2008). Animportantaspectofthemass-lossprocessthat
089.D-0562,090.D-410,091.C-0468,091.D-0344 is poorly understood is its geometry, i.e. the deviation of the
Articlenumber,page1of43
densitydistributionfromsphericalsymmetryondifferentspatial ing the thermal pulses known to occur during the AGB phase.
scales. The assumption about the geometry of the circumstel- Images provided by the Herschel/PACS instrument within the
larenvironmentaffectscalculationsofmass-lossratesandother frame of the Herschel Mass loss of Evolved StarS guaranteed
fundamentalparameters(Ohnakaetal.2008b). Inrecentyears, timekeyprogramme(MESS;Groenewegenetal.2011)showed
several observing campaigns were carried out with the purpose thatthemorphologyoftheouteratmosphere(R> 1000R )of
(cid:63)
ofinvestigatingthegeometryoftheenvelopeofAGBstars(ref- AGBs differs depending on various factors (for example inter-
erencesbelow). Observationssuggestthatthewindmechanism actionbetweenwindandtheinterstellarmedium,orwind-wind
may depend on the initial mass of the objects and on the evo- interaction;Coxetal.2012).
lution along the AGB (Habing & Olofsson 2003). Despite first Altogetheritisclearthatonehastoprobeallspatialscalesto
evidence in favour of overall spherical symmetry, some obser- understandthephysicsofthesecomplexoutflows. Whileprevi-
vations(Knappetal.1997,e.g. VHya)showverycomplicated ousstudieswiththeaimofdetectingasymmetriessufferedfrom
geometry at various spatial scales, and no consensus on its ori- a lack of (u,v) coverage (cfr. optical/infrared interferometry)
ginhasbeenreachedsofar(Habing&Olofsson2003). Under- and/orinstrumentswithsufficientsensitivities,theadventofnew
standing how the mass-loss shapes the envelope of AGB stars facilitieswithimprovedresolutionandsensitivitylikeHerschel,
iscrucialalsofortheprogeny. Althoughabinarycompanionis ALMA, and VLTI offers the unique chance to understand the
currentlythemostacceptedexplanation,othermechanismssuch mass-lossanddust-formationprocesses,andgenerallyspeaking,
as rotation velocity and magnetic fields might still play a role thelifecycleofdustandgasintheUniverse.
(De Marco 2009). Investigating the morphology of the atmo- In September 2010, we proposed a Large Programme (LP)
sphereofAGBstarsatdifferentspatialscalesandevolutionary totheEuropeanSouthernObservatory(ESO)tocomplementthe
stages(early-AGBandthermal-pulseAGB)helpstoclarifythe Herschelobservationswithobservationsusingthemid-infrared
pictureinthefollow-upstages. instruments MIDI on the Very Large Telescope Interferometer
ByscanningtheenvelopeofanAGBstarfromtheinsideto (VLTI) and VISIR on the Very Large Telescope (VLT). The
theoutsideonecandistinguishthefollowing: aims of the study are i) to establish whether asymmetries of
Innercircumstellarenvelope(CSE).Atmilliarcsecondscales the outer CSE originate in the dust-forming region or whether
(1-2.5stellarradii),closetothephotosphereofthestars,asym- they are only due to interaction with the ISM; ii) to evaluate at
metriesarefrequentlydetectedwithlunaroccultations(Richichi which height, the mass-loss process becomes manifestly non-
et al. 1995; Meyer et al. 1995) and optical interferometry spherical;andiii)tounderstandhowthegeometryoftheatmo-
(Ragland et al. 2006; Le Bouquin et al. 2009; Pluzhnik et al. spherechangesatthedifferentevolutionarystages(M-S-Cstars,
2009;Chiavassaetal.2010;Wittkowskietal.2011;Cruzalèbes andfromalmostdust-freetoverydustyobjects)withintheAGB
etal.2013a;vanBelleetal.2013;Mayeretal.2014;vanBelle sequence.
etal.2013;Cruzalèbesetal.2015). Theasymmetricstructures Inthispaper,wepresenttheprogrammeanddataoftheinter-
are often ascribed to convective patterns, but other interpreta- ferometric(MIDI)campaigninterpretedwithgeometricmodels.
tions are also invoked (mainly the effect of stellar rotation and All the ESO archive data available for the targets are incorpo-
binarity). Itis observedthatasymmetries inthebrightness dis- ratedintheanalysistogiveacompleteoverview. Thisfirstwork
tribution are more frequent for Miras (Cruzalèbes et al. 2015, willbefollowedbyaseriesofpapersincludingadetailedinter-
i.e. towards the end of the AGB life) and irregular variables pretationoftheMIDIdataintermsofmodelatmospheres. The
(Raglandetal.2006); asymmetriesaremorefrequentinC-rich VISIRobservingcampaignwasseverelyaffectedbybadweather
starsthanintheO-richstars. conditions;anewobservingproposalwasrecentlyaccepted.
Intermediate CSE. Between 2 and 10 stellar radii, asym-
The selection of the targets is described in Sect. 2.1, while
metries, and clumpiness are also observed for several objects
the strategy for observations is given in Sect. 2.2. The data re-
(Weigelt et al. 1998; Tuthill et al. 2000; Weigelt et al. 2002;
duction is reported in Sect. 2.3. We also used archive data that
Leão et al. 2006; Tatebe et al. 2006; Chandler et al. 2007; Pal-
are introduced in Sect.2.4. The programme used for the inter-
adini et al. 2012; Sacuto et al. 2013). In a very few cases the
pretationoftheinterferometricdataisdescribedinSect.3. Re-
asymmetries have a clear pattern resembling a spiral or a disc.
sultsarepresentedinSect.4. Thediscussion, conclusions, and
These cases are usually related to the presence of a hidden bi-
outlooks are given in Sects. 5 and 6, respectively. The detailed
nary companion (Mauron & Huggins 2006; Deroo et al. 2007;
description of every single target can be found in Appendix A,
Ohnaka et al. 2008a; Maercker et al. 2012; Mayer et al. 2013;
andthejournaloftheobservationsisprovidedinAppendixB.
Decin et al. 2014; Kervella et al. 2014; Ramstedt et al. 2014;
Lykou et al. 2015; Kervella et al. 2015). On the other hand,
many other authors detect time variability but no clear signa-
2. Targetselectionanddata
tures of asymmetries (Danchi et al. 1994; Ohnaka et al. 2005;
Wittkowski et al. 2007; Karovicova et al. 2011; Sacuto et al.
2.1. Targetselection
2011; Zhao-Geisler et al. 2011, 2012; Karovicova et al. 2013).
Toconfusethepictureevenmore,SiOmaserobservationsshow Asub-sampleoftheAGBstarsobservedwithintheHerschelkey
evidenceforclumpyisotropicmasslossintheatmosphereofO- programme MESS (Groenewegen et al. 2011; Cox et al. 2012)
richAGBstars,whileH OandOHmasers(atR>10R )prob- was selected on the basis of the following criteria: declination
2 (cid:63)
ingtheintermediate-outerpartofAGBstarsarelessconclusive accessible from the southern hemisphere where the ESO tele-
regardingthegeometry. scopes are located, brightness within the limits of the instru-
Outer CSE. Submillimiter observations of CO line profiles ments, and a range of chemistries and variability types. The
obtained towards M stars may deviate significantly from those IRAS two-colour diagram (van der Veen & Habing 1988) was
expectedfromasphericalenvelope(Knappetal.1998;Winters used as a reference with the purpose of sampling AGBs with
etal.2003;Klotzetal.2012a). ImagingintheCOradioemis- differentshellproperties(Fig.1). Everyregionincludestargets
sion lines of carbon stars revealed spherically symmetric thin withwell-definedinfraredcharacteristics: variability,andIRAS
detachedshells(Olofssonetal.2000),probablyoriginatingdur- LowResolutionSpectrometer(LRS)classification.
Articlenumber,page2of43
C.Paladinietal.:TheVLTI/MIDIviewontheinnermasslossofevolvedstars
sphericalobjects(suchasthedetached-shellobjectUAnt;Ker-
schbaum et al. 2010). For the non-spherical cases we selected
onebaselineorientedinthedirectionoftheasymmetryandone
baseline perpendicular to the latter. A third baseline, with ori-
entation in between the two, was selected to put constraints on
thepossibleelongation. Baselineswithrandomorientationwere
selected for the symmetric stars, as well as in the cases where
PACSimageswerenotyetavailable. Thus,weareabletocon-
strain any possible deviation from sphericity. We also tried to
samplethesamespatialfrequenciesbychoosingthesamebase-
linelengthsforallpoints. Thebaselinelengthalsohastobese-
lectedcarefullyforresolvingthedustformationzone. Giventhe
factthatnotallthetargetsofthesamplehadameasuredphoto-
sphericdiameter,thelatterwasestimatedthroughthe(V−K)re-
lationofvanBelleetal.(1999). FollowingtheresultsofDanchi
et al. (1994), we thus assumed that the diameter in the N band
Fig.1. TargetsofoursampleshownintheIRAStwo-colourdiagram
isapproximatelythreetimesthephotosphericdiameter,andthe
ofvanderVeen&Habing(1988).
baselinewasselectedaccordingly.
For planning the observations, we used ASPRO and AS-
ThemildlyvariableO-richAGBstarswithoutextendedcir- PRO2developedbytheJean-MarieMariottiCenterAsproser-
cumstellar shells are expected to populate region I. In region vice1.Theobservationswerecarriedoutbetween2011April23
II the objects are surrounded by young O-rich shells, while in and2012July01onthe1.8mAuxilliaryTelescopes(ATs). We
region IIIa the shells are more evolved and mass-loss rates are used the recommended observation sequence CAL-SCI-CAL.
higher. Most of the C-rich objects with relatively cold dust are Thefollowingselectioncriteriaforcalibratorstarswereapplied:
located in region VIa (so-called detached-shell objects), while brightness (difference between calibrator and science target of
theobjectslocatedinregionVIbhavehotoxygen-richdustclose ±1mag), position (RA and Dec as close as possible to the sci-
to the star and cold dust at larger distances. The variable stars encetarget),size(thecalibratorshouldbeassmallaspossible),
with evolved C-rich shells and infrared carbon stars are in re- andspectraltype(ifpossiblethecalibratorshouldhaveaspec-
gion VII. The panels not covered within the observing sample traltypeearlierthanM0). Thelistofcalibratorsandtheirmain
are populated by objects with optically thick envelopes having characteristicsarepresentedinTable2. ThejournaloftheMIDI
no visual counterparts, and by planetary nebulae. The list of observationsisavailableasonlinematerial(AppendixB).
targets is presented in Table 1 together with the location in the
IRAScolour-colourdiagram,themorphologicalclassidentified
byCoxetal.(2012)fromtheHerschel/PACSimages,andgen- 2.3. Observationsanddatareduction
eralcharacteristicssuchasspectraltype,variabilityclass,IRAS
12µmflux,period,distance,andmass-lossrate. The MIDI data were reduced using the data reduction pipeline
MIA+EWS2 (Jaffe2004a;Ratzka2005;Leinertetal.2004). A
detailed description of the data quality tests that were applied
2.2. Observingstrategy
during the data reduction can be found in Klotz et al. (2012b).
Inthisworkweconcentrateonpossibleasymmetriesdeveloping Datawerereducedwithallcalibratorsobservedinthesamenight
in the dust-forming region. Danchi et al. (1994) and more re- (if possible within ±2hours) and with the same baseline con-
centlyNorrisetal.(2012)observedthatdustisformingbetween figuration as the science target. The final calibrated visibilities
3 and 5 stellar radii, and even closer to the star in the case of are then the mean of all the visibilities, differential phases, and
Miravariables,asDraine(1981)alreadypredicted. Asaconse- fluxes reduced with suitable calibrators. The error is derived
quence,oneneedstohaveanangularresolutionof∼ 20masto from the standard deviation of that series. If the former error
observethelocusofdustformationforAGBswithin1kpc. For islowerthan±10%oronlyonecalibratorwasavailableduring
thisreason,theMIDIinstrument(Leinertetal.2003)installedat the night, a multiplicative error of ±10% was used (Chesneau
theESOVeryLargeTelescopeInterferometer(VLTI)onCerro 2007).
Paranal(Chile)untilMarch2015waschosen. TheMIDIinstru- The extraction of the differential phase was carried out fol-
ment is a two-beam combiner interferometer observing in the lowingthestandardprocedureofEWS(Jaffe2004b).Thediffer-
N band (8-13 µm). For every observation, the instrument de- ential phase was corrected for the changing index of refraction
livers one visibility spectrum (sometime shortened to visibility ofairbysubtractingalinearslope,sothatthemeanphaseover
in the text), one total flux spectrum, and one differential phase the N band is zero. We did not correct for higher order effects
spectrum(phasedifferencebetweendifferentspectralchannels; owingtowatervapourcontent(PWV).Instead,weemployeda
Sect. 4.2). Depending on the correlated magnitude of the tar- very simplistic approach, by calibrating the differential phases
get,eithertheHIGH-SENS(wherethecorrelatedandtotalflux using several calibrators taken over the night and deriving our
aremeasuredoneaftertheother)orSCI-PHOT(wherethecor- error estimate on the phase from the scatter of these multiple
related and total flux are measured simultaneously) mode was calibrations. The instrumental phase measured for the calibra-
chosen. AlltheLPdatahavespectralresolutionR=λ/∆λ∼30. torsduringthedifferentnightsisusuallystableandoftheorder
To optimise the observations for studying the geometry the of±5◦. Anyuncertaintyonthedifferentialphasecausedbythe
followingstrategywasused. Thechoiceforthebaselineconfig-
urationswassplitintwodifferentcategories,basedontheHer- 1 Availableathttp://www.jmmc.fr/aspro
scheldataavailable:configurationsfornon-spherical(suchasthe 2 http://www.strw.leidenuniv.nl/∼jaffe/ews/MIA+EWS-
bowshockinthecaseofTXPsc;Jorissenetal.2011), andfor Manual/index.html
Articlenumber,page3of43
Table1.Targetlist.
Target IRAScolour-colourregion Herschel/PACS(a) SpectralType Variability F Period Distance(b) Mass-lossrate
12
morphology [Jy] [d] [pc] [M yr−1]
(cid:12)
θAps II(O-richshells) Fermata M6.5III SRb 734.30 119(c) 113+7 0.4×10−7(d)
RCrt II(O-richshells) Eye M7III SRb 637.90 160(c) 261−+686 8×10−7(d)
RLeo I Fermata M8IIIe M 2161.00 310(e) 110−+5127 9.4×10−8(f)
TMic II(O-richshells) Fermata M7III SRb 493.80 347(c) 200−+1517 8×10−8(d)
RTVir IIIa Fermata M8III SR 462.20 375(g) 136−+3177 5×10−7(d)
−14
π1Gru II/IIIa/VII Irregular S5+G0V SRb 908.50 198(h) 153+23 4.6×10−7(i)
omiOri VIa/I Irregular S+WD SR 85.35 ... 200−+1383 <0.4×10−7(j)
−25
UAnt VIa Ring N:var Lb 167.50 ... 270+45 2×10−8(k)
RLep VII PointSource CIIe M 379.50 427(c) 470−+33401 1.6×10−6(k)
YPav VII PointSource CII SRb 72.38 233(c) 400−+112226 4×10−7(l)
TXPsc VIa Fermata CII Lb 162.90 ... 275−+7373 3.2×10−7(k)
SSct VIa Ring CII SR 65.31 148(c) 386−+21709 5.6×10−6(k)
AQSgr VII Fermata CII SR 56.64 199(c) 333−+7906 7.7×10−7(k)
XTrA VII Ring C Lb 201.00 ... 360−+6618 1.8×10−7(k)
−49
References.(a)Coxetal.(2012);(b)vanLeeuwen(2007);(c)Samusetal.(2009);(d)Olofssonetal.(2002a);(e)Whitelock&Feast(2000);
(f)Knappetal.(1998);(g)Imaietal.(1997);(h)Taburetal.(2009);(i)Jorissen&Knapp(1998);(j)Groenewegen&deJong(1998);(k)Bergeat
&Chevallier(2005);(l)Wintersetal.(2003).
Table2.Calibratorlist.
HD Spectraltypea F (a) θ(b) usedfor
12
[Jy] [mas]
18884 M1.5IIIa 234.7 12.28±0.05 omiOri,RLep
20720 M3/M4III 162.70 10.14±0.04 RLep
25025 M1IIIb 109.6 8.74±0.09 RLeo
29139 K5III 699.7 20.398±0.087 RLeo,UAnt
32887 K4III 56.82 5.90±0.06 RLep
39425 K1IIICN+1 28.0 3.752±0.017 omiOri
48915 A1V 143.1 6.08±0.03 RLeo,RLep,UAnt
50778 K4III 24.6 3.904±0.015 omiOri
61421 F5IV-V 79.1 5.25±0.21 RLeo
81797 K3II-III 157.6 9.142±0.045 θAps,RCrt,RTVir,RLeo
112142 M3III 47.0 5.90±0.7 RCrt,RTVir,AQSgr
120323 M4.5III 255.4 13.25±0.06 RCrt,RTVir,RLeo,UAnt,AQSgr
123139 K0III 56.9 5.33±0.057 RCrt,RTVir,RLeo,AQSgr
129456 K5III 21.4 3.37±0.014 TMic
133216 M3/M4III 200.7 11.154±0.046 RTVir
150798 K2II-III 144.0 8.76±0.12 θAps,RCrtRLeo,YPav,XTrA
151249 K5III 52.18 5.515±0.179 YPav
152786 K3III 82.1 8.02±3.23 θAps,RCrtRLeo,AQSgr
165135 K1III 23.4 3.47±0.015 TMic
167618 M3.5III 213.7 11.665±0.043 θAps,TMic
168454 K3III 62.17 5.874±0.026 SSct,AQSgr
169916 K0IV 31.2 3.995±0.019 TMic,YPav,SSct,AQSgr
177716 K1III 26.0 3.78±0.21 TMic
206778 K2Ib 103.9 7.59±0.046 θAps,RLeo,π1Gru
211416 K3III 59.3 5.92±0.28 AQSgr
224935 M3III 86.90 7.25±0.03 π1Gru
Notes.(a)http://simbad.u-strasbg.fr/simbad/;
(b) http://www.eso.org/observing/dfo/quality/MIDI/qc/calibrators_obs.html
lackofcorrectionforthedispersioneffectsduetoPWVisthere- fects, fluxes were only derived if the airmass difference to the
foretransferredtotheerrorsonourphases. science target was smaller than 0.2 and the calibrator was ob-
A word of caution must be issued concerning the MIDI served within ± 2 hours from the science target. Only calibra-
spectra. The water vapor content in the Earth atmosphere can tors of spectral type earlier than M0 were selected for the flux
changeconspicuouslyonatimescaleofhalfanhourwithoutany calibration. Henceforththepossibilitythatthesciencespectrais
changes in seeing or coherence time. So, it is possible that the contaminatedbypossibledustaroundthecalibratorisminimised
water vapour content and thus the transmission of the Earth at- (Chesneau2007). AddingtheLPtothearchivedata(Sect.2.4),
mospherechangesbetweenscienceandcalibrator. Whereasthe we collected a total of 201 visibility points; 60% of these data
calibration of the interferometric visibility is not affected, the wereofgoodqualityandwereusedinthiswork.
fluxes (i.e. the MIDI spectra) are affected. To limit such ef-
Articlenumber,page4of43
C.Paladinietal.:TheVLTI/MIDIviewontheinnermasslossofevolvedstars
2.4. Additionalobservationsandvariabilitycheck parametersgivenbyGEM-FINDarethe1σstatisticalerrorsde-
rivedfromthecovariancematrix(calculatedwithintheMPFIT3
ArchiveMIDIobservationswereavailableforTXPsc,AQSgr,
IDLroutinesimplementedinGEM-FIND).Itisknownthatthe
U Ant, T Mic, R Crt, R Leo, RTVir, π1 Gru, omi Ori, and
models in the Fourier space are not linear, therefore the errors
RLep. SomeofthesedatawereobservedinGRISMmode(spec-
are not Gaussian distributed. We tested the validity of the ap-
tral resolution R = 230). These high resolution archive obser-
proach with Monte Carlo simulations (Klotz et al. 2012b). We
vations were convolved to a spectral resolution R = 30 before
findthat1σerrorsfromtheMonteCarlosimulationsarecompa-
any comparison with the LP data. We also noticed that most
rabletothosederivedfromthecovariancematrix. Therefore,in
of these observations were carried out with different baselines
thefollowing,errorsarecomputedfromthecovariancematrix.
butatthesamepositionangle. Suchdatasetsallowustoprobe
For the study of the geometry of the circumstellar environ-
the atmosphere at different spatial scales, i.e. these data sets
ment, the following models were used to fit the data: circular
are optimal for tomography studies. Moreover these observa-
uniform disk (UD, representing an approximation of the stel-
tionscarryinformationaboutintra-cyleandcycle-to-cycleinter-
lar disk), circular Gaussian distribution (Gauss, approximation
ferometric and spectroscopic variability. When possible we as-
to an object with a molecular or dusty environment and limb
signedavariabilityphasecalculatedfromthevisuallightcurve
darkening), elliptical uniform disk (Ell. UD, such as in the UD
to every MIDI observation. For this purpose light curves were
casewithnon-centralsymmetricbrightnessdistribution),ellipti-
collected from the American Association of Variable Star Ob-
calGaussiandistribution(Ell.Gauss,suchasintheGausscase
servers(AAVSO),theAllSkyAutomatedSurvey(ASAS),and
with non-central symmetric brightness distribution). The latter
the Association Française des Observateurs d’Étoiles Variables
twomodelswereappliedonlyfortheobjectswithmorethantwo
(AFOEV). The phase is determined from the light curve using position angles available. In the case where a sufficient num-
thefollowingrelation: ber of observations sampling different spatial frequencies were
available,asphericaltwo-componentmodel(acircularUDplus
(cid:32) (cid:33)
φ= (t−T0) −int (t−T0) , circular Gaussian, where the two components typically repre-
P P sentthephotosphereandanopticallythindustand/ormolecular
component) was used additionally. In this latter case the diam-
wheretstandsforthedateoftheMIDIobservation(s)expressed eteroftheUDwasfixedtoavaluecorrespondingtotheθ(V−K)
inJuliandate,T isthephase-zeropointthatwasselectedasthe diameter(vanBelleetal.1999)tosimulatethecentralstar(orto
0
maximumlightclosestintimetothefirstMIDIobservation,and theobservedK−bandvalue,whenavailable).OnlytheGaussian
P is the period of variability already listed in Table 1. Visual envelopewasfitted. ThefitwithGEM-FINDwasperformedin
phases are assigned to the stars θ Aps, R Crt, R Leo, T Mic, twostages:firstonlytheLPdataarefitted,andafterwardstheLP
RTVir,RLep,YPav,SSct,andAQSgr. Thevaluesarelisted dataaremergedwiththearchivedataandanewfitisperformed.
inAppendixB,anderrorsareassumedtobeoftheorderof10% The reasoning behind this strategy is that the LP data are cho-
oftheperiod. sentosamplethesamespatialfrequenciesanddifferentposition
The analysis of spectroscopic variability is performed by angles, therefore they are more suitable for detecting possible
comparingMIDIspectraobtainedatdifferentvisualphase,and elongationsowingtoanon-centralsymmetricdistribution. The
also by comparing the MIDI spectra to available ISO and/or fit with all data (LP + archive) is performed for completeness,
IRAS spectra. The interferometric variability was studied by anditallowsustostudythestratificationofthestars.
comparing (when available) sets of visibilities at similar base-
line lengths and position angles, observed at different dates. If
nointerferometricvariabilityisdetected,onecanassumethatthe 4. Results
datacanbecombinedforthegeometricfit. The(u,v)-coverages
ThissectionsummarisesthegeneralfindingsoftheLP.Detailed
obtainedforallthedataoftheLP,includingthearchivedatais
discussionforthesingletargetsaregiveninAppendixA.
showninFig.2.
4.1. Visibilityversuswavelength
3. Geometricfitting
A visual inspection of the visibility spectrum reveals certain
A model-independent way to identify departure from spherical spectralfeaturescharacterisingthechemicalcompositionofthe
symmetryoftheCSEisbycomparingvisibilitiestakenatsimilar CSE. For this study, we inspect the visibilities in the range be-
baseline and different position angle. We performed this check tween0.1<∼V <∼0.9. InthecaseofV ≥0.9,itisnotpossibleto
where the data set allowed it. As a second approach we em- distinguish details of the spectral signature, or even distinguish
ploy the software GEM-FIND (GEometrical Model Fitting for thevisibilitiesfromthoseofapointsourceowingtothetypical
INterferometricData;Klotzetal.2012b)tointerpretourobser- errors. At V <∼ 0.1 the relation between visibility and spatial
vations. This software fits geometrical models to interferomet- frequency may not be univocal (as the visibility function may
ricvisibilitydata,wheredifferentsphericallysymmetric,centro- consist of several lobes). Fig. 3 shows the spectrally dispersed
symmetric,andasymmetricmodelsareavailable. Thedifferent visibilitycurvesforstarswithdifferentchemistry.
parameters of the models can be either wavelength dependent M-typeandS-typestars. Themostprominentmolecularfea-
(e.g.diameter, flux ratio of two components) or wavelength in- tureofoxygen-richstarsinthe N bandisSiOaround8µm. In
dependent(e.g.inclinationoraxisratioofadisk)asGEM-FIND some cases, this is followed by silicates and Al O dust. For
2 3
fits each wavelength point separately. This gives us the possi- stars not showing a pronounced silicate feature such as R Leo
bility to study the dependence of the model parameters on, for (upperleftpanelofFig.3),TMic,andtheS-typestarsomiOri
examplemolecularanddustfeatures. TheoutputofGEM-FIND and π1 Gru (upper right panel of Fig. 3), the visibility is rather
is a χ2 , the best-fitting parameters, and wavelength-dispersed
red
visibilitiesanddifferentialphases. Theerrorsofthebest-fitting 3 http://purl.com/net/mpfit
Articlenumber,page5of43
Fig.2. (u,v)coveragesobtainedforallthetargets.Thedifferentwavelengthsfrom8−13µmarecolour-coded(darktolight,respectively).
Articlenumber,page6of43
C.Paladinietal.:TheVLTI/MIDIviewontheinnermasslossofevolvedstars
Table3.Reducedχ2 fromGEM-FINDfitting.
red
Target Baseline Year N UD Gauss Ell.UD Ell.Gauss. UD+Gauss
OB
[m]
θAps 10–17 2011/12 6 0.78 0.69 0.41 0.38 ...
RCrt 10–16 2012 5 10.17 2.53 1.18 1.04 ...
10–64 2009/11/12 8 46.27 28.31 18.18 17.74 0.69
RLeo 11–16 2012 4 0.93 0.78 0.09 0.11 ...
11–64 2006/07 18 38.55 14.73 8.03 10.84 7.17
TMic 11–16 2011 4 0.28 0.27 0.09 0.11 ...
11–46 2004/11 7 1.16 0.96 0.49 0.46 0.24
RTVir 13–15 2012 2 1.31 1.13 ... ... ...
12–128 2008/09/11/12 14 49.10 25.13 22.80 20.51 3.04
π1Gru 10-15 2011 3 4.83 3.44 0.57 0.42 ...
10-62 2006/11 11 23.23 23.56 21.59 21.49 0.73
o1Ori 32–46 2011 7 1.30 1.30 0.79 0.78 ...
32-129 2005/11 14 2.04 2.01 1.90 1.85 1.43
UAnt 30 2012 1 ... ... ... ... ...
30-95 2008-2012 3 1.20 1.23 0.65 0.63 ...
RLep 34-40 2012 6 1.59 0.93 1.32 0.90 ...
34-79 2011/12 10 2.38 3.65 1.39 3.55 0.96
YPav 50-63 2011 4 0.77 0.78 0.44 0.44
TXPsc 60-140 2011 6 0.99 1.19 0.90 1.09 ...
11-140 2004-2011 15 1.28 1.32 1.29 1.32 1.34
SSct 40-45 2011 2 2.63 2.61 ... ... ...
XTrA 21-34 2011 5 0.95 0.89 0.31 0.30 ...
AQSgr 37-42 2011 2 2.39 2.47 ... ... ...
Notes. The range in baseline length is given as ’Baseline’. The number of observations used for the fitting is given as N . The first row
OB
correspondstothedataoftheLP.ThesecondrowcorrespondstofitsweretheLPandarchivedataweremerged. Ifonlydatapointswithsimilar
positionangleswereavailable, theellipticalmodelswerenotfitted. Theχ2 ofthemodelbestfittingthedataishighlightedinboldface. The
red
resultingbest-fittingparametersaregiveninTable4,whereasthedatesoftheobservationsaregiveninAppendixB.
Fig.3. Someexamplesofvisibilityspectraforgivenbaselines.ThecompletesamplecanbefoundinAppendixA.Molecularanddustfeatures
arehighlightedforstarswithdifferentchemistry: RLeoandRCrtareM-typestars, π1 GruisanS-typeobject, whileRLepandTXPscare
C-typestars.ThelowerrightpanelshowsanexampleofthevisibilityspectraofthecarbonstarUAnt(cf.Sect.4.1).Thetypicalerrorbarsareof
theorderof10%.
flat with a small bump at short wavelengths. The diameter in-
creases slightly at longer wavelengths. For stars showing the
dust and molecular features (R Crt shown in the central upper
Articlenumber,page7of43
panel of Fig. 3; RT Vir, and θ Aps) the visibility has a peak in 9µm. Thesedataareacquiredatdifferenttimes(seeSect.4.3),
the8-9µmregion,adecreaseofbetween9-11µm,andasubse- atthesamepositionangles,andverysimilarprojectedbaselines.
quentincrease. The lower panels show a much more complex behaviour with
C-typestars. Themoleculescontributingtothecarbonstars featuresat∼ 10and∼ 11.8µm.
opacity in the N band are mainly C H and HCN. Concerning The differential phase of RT Vir shown in Fig. 5 is charac-
2 2
thedust,evolvedcarbon-richobjectsshowSiCdustat11.3µm terisedbyajumpbetween8and9µm,followedbyamonotonic
and amorphous carbon dust (featureless). Examples of visibil- increase. For an interpretation of the RT Vir differential phase
ities of stars with SiC dust were shown by e.g. Ohnaka et al. withageometricmodel,werefertoSacutoetal.(2013).
(2007),Sacutoetal.(2011),Paladinietal.(2012),andRauetal. Allthenon-zerodifferentialphasesoccuratvisibilityspectra
(2015). The lower left panel of Fig. 3 shows the visibility of below10%.
theC-richMiraRLepwiththetypicaldropat11.3µmbecause
SiC. Y Pav, X Tra, and AQ Sgr show similar visibility curves.
The visibilities of carbon-rich stars without SiC have a typical
bowshape. Thevisibilitygetslower(i.e. thestarislarger)be-
tween8-9µmandafter12.5µmwherethemolecularopacityis
higher. Fig.3showsanexampleofthiskindofstar,i.e. TXPsc
(alsopresentedinKlotzetal.2013).
Other stars. The IRAS and MIDI spectra of U Ant show
thesignatureofSiCdustfeature,whilethereisnotraceofsuch
afeatureinthe(spatiallyresolved)interferometricobservations
(lowerrightpanelofFig.3). Thiscouldbetheresultofresolv-
ingoutpartofthetotalemission,onlyrevealingtheemissionof
the spatial scale to which the interferometer is sensitive (at the
employedbaselines).
ThecaseofSSctisslightlydifferent.ThereisnotraceofSiC
inthe(spatiallyresolved)interferometricobservationsnorinthe
MIDI spectrum. The dust feature is observed in the ISO spec-
trum recorded 14 years before the LP observations. The IRAS
spectrumofSSctobtained∼ 30yearsbeforeisverynoisybut
seems to agree with the MIDI spectrum (Fig. A.21). The lat- Fig.5. SameasFig.4forRTVir.
ter was derived by averaging data taken on two different days,
thereforeoneshouldbeabletoruleoutaproblemwiththecali-
bration. TheSSctobservationssuggestatemporalvariabilityin
thestratificationofSiC.
4.3. Spectroscopicandinterferometricvariability
To our knowledge it is the first time that such findings are
reportedforAGBstars. Section5providesadetaileddiscussion AsalreadymentionedinSect.2.4,weusedarchivespectroscopic
ofthesefindings. andinterferometricobservationstostudytheN-bandvariability.
The spectroscopic variability typically corresponds to the
variation in the colour (temperature) or a specific variation in
4.2. Differentialphase
certainlinestrength. Ontheotherhand,theinterferometricvari-
Exceptforthecaseswherethedifferentialphaseisequalto180◦, abilityisusuallyconnectedwithachangeinthemorphologyof
and/oritisaccompaniedbyanullvalueinthevisibility,anon- theobject,inparticularthespatialscaleoftheN-bandemission
zerodifferentialphasemeasuredbyMIDIimpliesanasymmetric region. Of course the real picture is more complex, and inter-
brightness distribution. The latter can be explained by two ef- ferometricvariabilitymightalsobeduetobrightnessvariation.
fects(Tristrametal.2014): first,theobjectiscomposedoftwo Whenthestellaratmosphereisspatiallyresolved,andthereare
sources (for example, the photosphere and resolved dust com- multiplecomponentsintheFOVoftheinterferometer(i.e. pho-
ponent)withdifferentspectraldistributionthroughthe N band; tosphereplusextendedmolecular/dustlayer,orphotosphereplus
and,second,theobjectiscomposedoftwoobjectswithaspatial clumpystructures),itispossibletoobservevariationinthevisi-
distribution more resolved at a certain wavelength than at an- bility(atthesamespatialfrequency)becauseofachangeinthe
otheracrosstheNband. Innatureweusuallyobserveamixture fluxratiobetweenthetwocomponents.
of these two effects, which are very difficult to distinguish, un- InFig.6wecomparedtheleveloftheMIDIspectrawithISO
lessonehasenoughinformationfordetailedmodellingortoat- and IRAS spectra (when the ISO observations were not avail-
temptanimagereconstruction. Non-zerodifferentialphasewas able). TheISOandIRASobservationsaretakenapproximately
observed in a few AGB stars. Usually this is interpreted as a 30 years apart from the MIDI observations. Such a compari-
typical signature of a disc (Kervella et al. 2014; Ohnaka et al. soncaninprincipleprovideinformationaroundlong-timevari-
2008a; Deroo et al. 2007) or the signature of a clump (Sacuto abilityduetodustformationand/ormass-lossvariation. Onthe
etal.2013;Paladinietal.2012). Sincemodellingveryfewdif- other hand, this can be caused by the FOV difference between
ferentialphasesgiveshighlynon-uniquesolutions,noattemptto the various telescopes. The FOV of the MIDI observations is
interpretthedifferentialphaseiscarriedouthere. ∼2.3(cid:48)(cid:48)×1.6(cid:48)(cid:48),anditissmallerthantheFOVofIRASandISO,
We report non-zero differential phase only for two objects: 45(cid:48)(cid:48)×45(cid:48)(cid:48),and33(cid:48)(cid:48)×20(cid:48)(cid:48),respectively). Onlyadetailedmod-
RLeoandRTVir(Figs.4and5). Themorphologyofthediffer- elling is able to distinguish between these effects. Such mod-
entialphasesofRLeocanbeclassifiedintwogroups,according elling is beyond our scope, and we simply report cases of sus-
totheprojectedbaselineusedfortheobservations.Thethreeup- pectedvariabilityandleavethemodellingtoafutureinvestiga-
perpanelsofFig.4showajumpofthedifferentialphasearound tion(Rauetal.2017,inpress).
Articlenumber,page8of43
C.Paladinietal.:TheVLTI/MIDIviewontheinnermasslossofevolvedstars
Fig.4. Non-zerodifferentialphasesmeasuredbyMIDIforRLeoorderedinincreasingprojectedbaseline(B ).
p
We observe that 3 stars out of 13 have a mid-infrared flux the interferometric variability for R Leo. The first set includes
very similar to the IRAS flux (U Ant, omi Ori, and R Crt). three observations taken with the short baseline configurations.
AQSgr, YPav, RTVir, andTMichaveafluxlevelbelowthat The visibility level is ∼ 0.6; the observations are taken at sim-
observed by ISO/IRAS. The shape of the spectrum is usually ilar visual phases, but one of them was observed six cycles be-
consistent, with an exception made for Y Pav where a calibra- fore. Novariabilityisobservedforthissetofdata(Fig.A.6,left
tion problem cannot be excluded. This hypothesis is also sup- panel).Thesecondsetofdataalsoincludesthreedatapoints,but
ported by the fact that in the Y Pav spectrum, one can still see thefirsttwowereaveragedbecausetheyweretakenwithintwo
thetelluricozonefeatureat9.7µm. consecutivedays(withverysimilarPAandprojectedbaseline).
ForRCrt,RLeo,RLep,andRTVir,wehadseveralMIDI TheobservationsareshownintherightpanelofFig.A.6andthe
spectraobservedatdifferentvisualphases. Byplottingtheflux differencebetweenthevisualphasesis0.13. Itisobviousthatin
a various wavelengths (8,10, and 12 µm) versus visual phase, this case we observe a variation in the visibility level from one
westudytheintra-cycleandcycle-to-cyclevariabilityofthestar visualphasetotheother.Howevertheseobservationsareassoci-
(Figs.A.3,A.5,A.9,andA.15). RCrtshowsnosignificantvari- atedwithadifferentialphasesignature(Fig.4,upperrow). The
ability (Fig. A.3). R Lep and R Leo show variations. RT Vir differential phase also changes from one set of observations to
is by far the star with the best coverage in phase. The varia- theother.
tion of the flux over the pulsation period resembles a sinusoid
(Fig.A.9). Thefluxvariationwithinthecyclecorrespondstoan
4.4. Thegeometricfittingresults
amplitudeofvariabilityof0.48magat8µm,and0.75magat10
and12µm. LargeProgrammedataonly. Asafirststepthegeometricmod-
Theinterferometricvariabilityofthevisibilityspectrumwas elsareonlyfittedtotheLPdata. AsdescribedinSect.2.2, the
studied only for θ Aps, R Leo, RT Vir, and R Lep. The intra- LPobservationssampleverysimilarspatialfrequencies(i.e. the
cycle observations of the carbon-rich mira R Lep are taken at same part of the star) at different position angles. In Table 3,
verysimilarvisualphases(0.01difference),thereforeitisnota foreachstar, wepresentinthefirstrowtheresultsofthefiton
surprise if no interferometric variability is detected (Fig. A.16, the large-programme data only. The model with the χ2 clos-
red
left panel). However, a cycle-to-cycle variation is observed in est to 1 is considered as the one best fitting the data, and it is
thelevelofthevisibilityspectrum(Fig.A.16,rightpanel). The highlightedinboldinTable3. Theellipticalmodelshavebeen
variationismorepronouncedinthemoleculardominatedregion testedfor10objectsoutof14becauseinsomecasesnotenough
between9and10µm. Wedonotobservevariationbetween11 datapointswereavailable. Oneoxygen-richstar(RCrt)outof
and 12 µm, where SiC is located. The visibility level is higher the 4 tested with elliptical models is asymmetric. Both S-type
beforethevisualphaseminimum(φV =1.43),correspondingto objects(π1 GruandomiOri)alsoshowindicationsofellipticity
asmallerdiameter. fromtheGEM-FINDfit. Outofthesampleof4carbon-richob-
Neither θ Aps and RT Vir show any evidence of variability jectsthatweretested, noneturnedouttobeasymmetric. Ithas
inthevisibilityspectrum.Twosetsofdataareavailabletocheck tobestressedoutthatanellipticalsolutiondoesnotnecessarily
Articlenumber,page9of43
Fig.6. IRASorISOspectra(blacklines)comparedwiththeMIDIspectroscopicobservations(shadedareas).
imply that the environment has a truly elliptical shape. It only symmetricsolutions;butthisresultcertainlydoesnotimplythat
means that the CSE is non-central symmetric. More complex the asymmetry detected by the dedicated LP data must be con-
geometriesthanellipsescannotbeexcluded. sidered spurious. Very likely the best-fitting model would be
LargeProgramme&archivedata.Ninestarsofoursamplehave an elliptical UD+Gaussian, but this model has too many free
archive data, and two of these stars have non-zero differential parameters to be tested versus our data set. By looking at the
LP+archive fitting results of Table 3, one notes the following:
phases. There are, therefore, a total of five targets showing ev-
sixstarsarebestfitbythecompositemodelUD+Gauss,i.e. six
idence of asymmetric environment. All these asymmetric stars
stars have an extended, optically thin, component. However, it
have O-rich chemistry and are located in the lower part of the
is possible that by adding more visibility points, other objects
IRAScolour-colourdiagram,asshowninpanela)ofFig.7.
alsoincreaseincomplexityandtheyarebestfittedwithacom-
Afterexcludinginterferometricvariability,wecombinedthe posite geometric model. This is very likely for X Tra, AQ Sgr,
data of the LP with those obtained from the archive. Besides θAps,andYPavwherethecoverageofthevisibilityspectrum
R Lep and TX Psc, all the other stars have archive data that does not extend below 0.6. Given the fact that no dust feature
sample mostly the same position angle but at different spatial is detected around S Sct, we expect that the visibility spectrum
scales. As already stated in Sect. 2.4, such data are optimal arereproducedonlywithonecomponent,evenbyaddingother
forstudyingthestratificationofthestar, buttheyareobviously visibility points. So far, in our sample only TX Psc has a visi-
less sensitive to asymmetries. Exception made to TX Psc, ev- bilitythatgoesdowntoV∼0.2withoutshowingdeparturefrom
ery time archive data are added to the fit, the composite (UD uniformdisc: nodustenvelopeisdetected. Thecarbonstarsin
plusGaussian)modelturnsouttobethebest-fittingmodel(Ta- regionVIa,followingtheloopforthecarbonstars,canbefitwith
ble3). Thisimpliesthatthestarshaveanextendedenvironment single component models and are supposed to be younger than
becauseofmolecularand/ordustopacities. Wenotethatthead- thoselocatedinregionVII.Theonlycarbonmiraofthesample
ditionalarchivedatawashawaytheellipticalsolutionforallthe isfitwiththecompositemodelinregionVII.Alltheobjectsin
threeobjectsmentionedatthebeginningofthissection. Thisis region VIa do not show SiC in the visibility. The geometry of
a consequence of the fact that the additional data always have the CSE increases in complexity from left to right, where this
thesamepositionangle. Becausetheseadditionaldataaremore
numerous than the LP data, they drive the fit solution towards
Articlenumber,page10of43