Astronomical Science

Molecular and Dusty Layers of Asymptotic Giant Branch Stars Studied with the VLT Interferometer

Markus Wittkowski1 Low to intermediate mass stars, including and their effects on the mass-loss mech­ Iva Karovicova1 our Sun, evolve to red giant stars and anism are also not well understood. David A. Boboltz2 subsequently to asymptotic giant branch Because of the extension of the atmos­ Eric Fossat3 stars after the hydrogen and helium phere, temperatures are cool enough Michael Ireland4, 5 ­supplies in the core have been exhausted so that molecules form, leading to a sce­ Keiichi Ohnaka6 by nuclear fusion. An AGB star is in the nario where molecular layers lie above Michael Scholz7, 8 final stage of stellar evolution that is the continuum-forming photosphere. Francois van Wyk9 driven by nuclear fusion, where a degen­ Dust is formed and believed to be accel­ Patricia Whitelock9, 10 erate carbon–oxygen core is surrounded erated within this extended atmospheric Peter R. Wood11 by hydrogen- and helium-burning layers, region with a certain condensation Albert A. Zijlstra12 a huge convective envelope and a very sequence, where dust species of higher extended and diluted stellar atmosphere. condensation temperatures form closer Mass loss becomes increasingly impor­ to the stellar surface and dust species 1 ESO tant during AGB evolution, both for stellar with lower condensation temperatures 2 United States Naval Observatory, evolution, and for the return of material form at larger distances. There is also a Washington, DC, USA to the interstellar medium. It reduces the notion that different dust species may 3 Laboratoire d’Université convective stellar envelope until the star prevail at different stages along the AGB, d’Astrophysique de Nice, France begins to shrink and evolves at constant along with increasing stellar luminosity 4 Dept. of Physics and Astronomy, luminosity toward the hotter post-AGB and increasing mass-loss rate. ­Macquarie University, Sydney, Australia and planetary nebula (PN) phases, and is 5 Australian Astronomical Observatory, thus the most important driver for the Optical interferometry at near-IR and mid- Epping, Australia ­further stellar evolution (e.g., Habing & IR wavelengths has proved to be a pow­ 6 Max-Planck Institut für Radioastrono­ ­Olofsson, 2003). Mass loss from AGB erful tool to study the extended atmos­ mie, Bonn, Germany stars is also one of the most important phere and dust condensation zones of 7 Zentrum für Astronomie, University of sources for the chemical enrichment of AGB stars, because of its ability to spa­ Heidelberg, Germany the interstellar medium and of galaxies. tially resolve these regions. Indeed, red 8 School of Physics, University of giants and AGB stars have historically ­Sydney, Australia Depending on whether or not carbon has been prime targets for optical interfer­ 9 South African Astronomical Observa­ been dredged up from the core into the ometry, because of their brightness and tory, Cape Town, South Africa atmosphere, AGB stars appear in obser­ the match of their angular size to the typi­ 10 Astronomy Dept., University of Cape vations to have an oxygen-rich or a car­ cal spatial resolution of optical interfer­ Town, Rondebusch, South Africa bon-rich chemistry. A canonical model of ometers. However, despite the long his­ 11 Australian National University, the mass-loss process has been devel­ tory of such measurements, they still ­Canberra, Australia oped for the case of carbon-rich chemis­ continue to provide important new con­ 12 Jodrell Bank Centre for Astrophysics, try, where atmospheric carbon dust has straints on the open questions discussed University of Manchester, United a sufficiently large opacity to be radia­ above. In particular, the near-infrared ­Kingdom tively accelerated and driven out of the (NIR) instrument AMBER and the mid- gravitational potential of the star and infrared (MIR) instrument MIDI of the VLT where it drags the gas along. For the Interferometer (VLTI) have been shown Mass loss from asymptotic giant branch case of oxygen-rich chemistry, the details to be well suited for new measurements (AGB) stars is the most important driver of this process are not understood, and of AGB stars because of their unprec­e­ for the evolution of low to intermediate are currently a matter of vigorous debate. dented ability to provide spectro-­ mass stars towards planetary nebulae. Questions remain also for the carbon-rich interferometric observations with spectral It is also one of the most important case, such as regarding the formation resolutions of 35–12 000 (AMBER) and sources of chemical enrichment of the of the recently detected oxygen-bearing 30–230 (MIDI). Here, we report on very interstellar medium. The mass-loss pro- molecule water in the inner atmospheres recent results (­Karovicova et al., 2011; cess originates in the extended atmos- of carbon-rich AGB stars by the Herschel Wittkowski et al., 2011) from our ongoing phere, whose structure is affected by space mission (Decin et al., 2010). programme to characterise molecular stellar pulsations, and where molecular and dusty layers of AGB stars using the and dusty layers are formed. Optical AGB stars experience stellar pulsations, VLTI instruments. interferometry resolves the extended from semi-regular variable stars (SRVs) atmospheres of AGB stars and thereby on the early AGB to large-amplitude long- enables us to obtain measurements period variable stars on the more evolved Recent VLTI observations of the intensity profile across this region. AGB, including Mira variables and more We present an overview of recent results dust-enshrouded stars at the tip of the VLTI observations of AGB stars using from our spectro-interferometric obser- AGB. Pulsations and the induced shock the instruments MIDI and AMBER started vations of AGB stars using the near- fronts are expected to play a crucial role with the measurements described by and mid-infrared instruments AMBER for the structure of the extended atmos­ Ohnaka et al. (2005) and Wittkowski et and MIDI of the VLT Interferometer. pheres, but the details of these processes al. (2008), respectively. These studies

24 The Messenger 145 – September 2011 demonstrated that the spectro-interfero­ Stellar cycle/phase 01234 metric capabilities of MIDI and AMBER 6 are well suited to characterise the dusty and molecular layers of AGB stars. AAVSO

We obtained MIDI data of the Mira varia­ 8 AFOEV ble AGB stars S Ori, GX Mon, RR Aql, R Cnc, and X Hya between 2004 and 2009. Some of these observations were 10 aimed at performing a mid-infrared moni­ toring of AGB stars in order to investi- gate intracycle and cycle-to-cycle varia­ g bility, and others were designed to com­ ma 12 plement AMBER observations to provide V a more complete picture of the dusty and molecular layers at the same time. Mid- 14 infrared interferometric monitoring of the sources listed above has been performed by Iva Karovicova during her PhD thesis, MIDI 16 together with theoretical simulations of VLBA the expected variability. For example, the observations of RR Aql described in Karovicova et al. (2011) include 52 obser­ 18 3000 3500 4000 4500 vations obtained at 13 epochs between Julian Date – 2450000 April 2004 and July 2007 covering three pulsation cycles (Figure 1). Figure 1. (Top) Visual light curve of RR Aql with our Figure 2. (Bottom) Uniform disc diameter as a func­ epochs of MIDI observations indicated in green. tion of wavelength obtained from VLTI/AMBER Epochs of Very Long Baseline Array (VLBA) observa­ observations of the Mira variable AGB star R Cnc. Using the AMBER instrument, we se­­ tions are shown in blue. From Karovicova et al. The red line denotes the best-fit prediction from the cured data of the Mira variables R Cnc, (2011). recent dynamical atmosphere model series CODEX. X Hya, W Vel, RW Vel, and RR Aql From Wittkowski et al. (2011). between 2008 and 2010 using the medium reso­lution mode with a spectral resolution of 1500. First results from 22 R Cnc these campaigns have been presented VLTI/AMBER 1.6 by Wittkowski et al. (2011). ) 20 E0-G0-H0 as er

Some of these observations were coor­ (m er 18 1.4

dinated with concurrent VLBA observa­ diamet tions of the SiO, H2O, and OH maser UD diamet emission (observation epochs shown in 16

1.2 ed

Figure 1), which provide complementary disk information on the kinematics and geom­ m 14 etry of the molecular layers in which the or

1.0 Normaliz maser emissions originate. Unif 12

H2O CO 0.8 10 12C16O 12C16O 12C16O 12C16O 12C16O 13C16O Structure of molecular layers 2–0 3–1 4–2 5–3 6–4 5–3 2000 2100 2200 2300 2400 The medium resolution (R ~ 1500) near- Wavelength (nm) infrared AMBER observations of the Mira variables of our sample confirmed a char­ so that the source appears larger and the study and to quantification of this effect acteristic wavelength-dependent shape of visibility smaller, and lower at other wave­ because of its wide wavelength coverage the visibility function that is consistent with lengths so that the source appears smaller with high spectral resolution. In our earlier low resolution AMBER observa­ and the visibility larger. Our observations measurements, the corresponding wave­ tions of S Ori and that can be understood confirm this effect, which has previously length-dependent uniform disc (UD) within the molecular layer scenario. In this been detected by interferometric obser­ diameters show a minimum near the scenario, the opacity of molecular layers, vations using a few narrowband filters. near-continuum bandpass at 2.25 μm. at NIR wavelengths, most importantly H2O Hereby, the AMBER instrument has shown They then increase by up to 30% toward and CO, is larger at certain wavelengths to be particularly well suited to further the H2O band at 2.0 μm and by up to

The Messenger 145 – September 2011 25 Astronomical Science Wittkowski M. et al., Molecular and Dusty Layers of Asymptotic Giant Branch Stars

70% at the CO bandheads between 200 2.29 μm and 2.48 μm. Figure 2 shows the wavelength-dependent UD diameter for R Cnc the example of the source R Cnc. VLTI/AMBER 150 E0-G0-H0

Recently, new dynamical model atmos­ ) pheres for Mira variables were developed eg (d

that are based on self-excited pulsation e models and opacity-sampling radiation 100 phas treatment (CODEX model series; Ireland et al., 2008; 2011), and which became available at the time of our data analysis. These models predict intensity profiles Closure that are in excellent agreement with our 50 H O CO visibility data and thus the uniform disc 2

12C16O 12C16O 12C16O 12C16O 12C16O 13C16O diameters, as also indicated in Figure 2. 2–0 3–1 4–2 5–3 6–4 5–3 In order to estimate the effective temper­ 0 atures of the sources at the time of our 2000 2100 2200 2300 2400 observations, we coordinated simultane­ Wavelength (nm) ous near-infrared photometry at the South African Astronomical Observatory Figure 3. Closure phase as a function of wavelength of the star, and thus to the observed (SAAO). The effective temperature values for the example of the Mira variable AGB star R Cnc inhomogeneities. An inhomogeneous or obtained with VLTI/AMBER. From Wittkowski et al. obtained are consistent with those of the (2011). clumpy molecular environment in the model atmospheres, as are the distances extended atmosphere may have impor­ obtained from the radii and bolometric tant implications for the non-LTE chemis­ magnitudes. Altogether, the agreement of atmosphere, where inhomogeneities are try in this region and may help to under­ our observations with the CODEX model present at distances where molecules are stand the formation of certain molecules atmosphere series increases the confi­ formed and where molecular emission such as water in carbon-rich AGB stars dence in the model approach, and thus in originates. (Decin et al., 2010). It may also be the the scenario where the stratification of source of observed clumpy molecular the extended atmosphere is largely deter­ With the limited amount of visibility data features in the circumstellar environment mined by the pulsation that originates in that we currently have, we cannot recon­ at larger distances. the stellar interior and in which the molec­ struct the full morphology of the molecu­ ular layers lie above the continuum-form­ lar layers. It would thus be important to ing photosphere. conduct an interferometric imaging cam­ Variability of dusty layers paign of AGB stars at high spectral reso­ lution. However, since the visibility data The observed MIDI visibility curves of Inhomogeneities in molecular layers are overall well consistent with spherical RR Aql show significant wavelength models, we can already estimate that the dependence with a steep decrease from The AMBER closure phase functions of deviation from point symmetry originates 8–~ 9.5 μm and a slow increase in the our targets exhibit significant wavelength- from substructure at a relatively low flux 9.5–13 μm range. The shape of the dependent non-zero values at all wave­ level for an overall spherical intensity dis­ 8–13 μm flux spectrum with a maximum lengths (Figure 3 shows the example tribution. For example, the values of R near 9.8 μm is known to be a characteris­ of R Cnc). Non-zero values of the closure Cnc in the water vapour band shown in tic silicate emission feature. Our rich sam­ phase are indicative of deviations from Figure 3 could be explained by an unre­ ple of MIDI data of RR Aql cover pulsation point symmetry. Our data confirm non- solved (up to 3 milliarcseconds [mas]) phases between 0.45 and 0.85, i.e. mini­ zero closure phase data of Mira variables spot contributing 3% of the total flux at a mum to pre-maximum phases, for a total that have been previously observed by separation of about 4 mas (for compari­ of three cycles. For many different pulsa­ other interferometers. Again, the AMBER son, the photospheric diameter is esti­ tion cycles and phases, we obtained data data provide additional information, thanks mated to 12 mas). at similar projected baseline lengths and to their spectro-interferometric nature, position angles. This gives us a unique that helps to constrain where within the There are several physical mechanisms opportunity to perform a meaningful direct atmosphere the asymmetric structure that may lead to asymmetric stellar sur­ comparison of interferometric data ob­ originates. In fact, we observe a complex face structures. For our targets, we favour tained at different pulsation phases and wavelength-dependent closure phase a scenario where the pulsation in the stel­ cycles. An example of such a direct visibil­ signal that correlates with the features of lar interior induces chaotic motion in the ity comparison is shown in Figure 4. the molecular layers, most importantly outermost mass zones of the extended

H2O and CO. This indicates a complex atmosphere that leads to different exten­ We concluded that our data do not show non-spherical stratification of the extended sions of the mass zone on different sides any evidence of intracycle or cycle-to-

26 The Messenger 145 – September 2011 cycle visibility variations within the probed 1.0 range of pulsation phases (~ 0.45–0.85) + Bp: 28.60 P. A.: 69.67 phase: 2.45 and within our visibility accuracies of ×+ Bp: 29.05 P. A.: 70.06 phase: 2.61 about 5–20%. This implies either that the 0.8 × Bp: 30.62 P. A.: 72.04 phase: 2.74 mid-infrared sizes of the molecular and dusty layers of RR Aql do not significantly vary within our phase coverage or that e the conducted observations are not sen­ 0.6 sitive enough to detect such variations. mplitud

The MIDI photometry exhibits a 1–2σ sig­ a nature of intracycle and cycle-to-cycle 0.4 flux variations which are most pronounced toward the silicate emission feature at Visibility 9.8 μm. Additional observations with a dedicated mid-infrared spectrograph, 0.2 such as the instrument VISIR at the VLT, are recommended to confirm the flux ­variations. 0.0 8 9 10 11 12 13 We performed a number of model simu­ Wavelength (µm) lations in order to investigate the visibility and photometry variations that are theo­ Figure 4. MIDI visibility amplitudes of RR Aql for retically expected in the 8–13 μm wave­ ­different pulsation phases (indicated by different ­coloured points) within the same cycle, chosen to length range for the typical parameters of investigate intracycle visibility variations. From RR Aql. We used a radiative transfer Karovicova et al. (2011). model of the dust shell where the central source is described by the dust-free dynamic model atmosphere series, as 1.2 they were used to model the near-IR AMBER data. We varied model parame­ 1.0 ters such as the phase of the central atmosphere model and the opacity and inner radius of the dust shell to investi­ 0.8 gate the expected photometry and visibil­ ity variations during a pulsation cycle. An example of such a simulation is shown 0.6 mplitude in Figure 5. a ty 0.4

These model simulations show that isibili ­visibility variations are indeed not V expected for the parameters and obser­ 0.2 vational settings of RR Aql at wave- lengths of 8–13 μm within the uncertain­ 0.0 ties of our observations. Variations in the flux spectra may in some cases just –2.0 be detectable. Thus, our observational 8910 11 12 13 result of a constant visibility curve and Wavelength (µm) only slightly varying flux spectra at wave­ lengths of 8–13 μm are consistent with, Figure 5. An example of expected intracycle visibility and not contradicting, theoretical expec­ variation based on a simulation using two atmos­ phere models with a phase difference of 0.2 and a tations of a pulsating atmosphere. We dust shell that is located closer to the star with a used our simulations as well to determine higher optical depth toward stellar minimum com­ the best baseline configurations for the pared to stellar maximum. The solid green and red parameters of certain targets in order lines denote the two global models of this simulation, the dash–dotted lines the unattenuated stellar con­ to optimise future interferometric observ­ tribution, the dotted lines the attenuated stellar con­ ing campaigns that aim at characterising tribution and the dashed lines the dust shell contri­ the small expected visibility variations. bution. From Karovicova et al. (2011).

The Messenger 145 – September 2011 27 Astronomical Science Wittkowski M. et al., Molecular and Dusty Layers of Asymptotic Giant Branch Stars

Characteristics of dusty layers 1.0

We characterised RR Aql’s dust shell by using radiative transfer models of 0.8 shells of Al2O3 and/or silicate grains with independent inner radii and opacities, 0.6

­following earlier such attempts in the liter­ mplitude ature. We obtained best-fit results for a RR Aql with a silicate shell alone. The 0.4 addition of an Al2O3 shell did not result in any improvement in the model fits. The Visibility best-fit model for our average pulsation 0.2 phase of ΦV = 0.64 ± 0.15 includes a sili­ cate dust shell with an optical depth of 0.0 τV = 2.8 ± 0.8 and an inner radius of 8910 11 12 13 Wavelength (µm) Rin = 4.1 ± 0.7 RPhot (where RPhot is the photospheric radius) and uses a central intensity profile corresponding to an Figure 6. Radiative transfer model of RR Aql (solid confirm earlier suggestions of a sequence atmosphere model with an effective tem­ lines) compared to the visibility data of one of our where the dust content of stars with low epochs including two observations. The solid lines perature of 2550 K. Figure 6 shows a indicate our best-fit model, while the contributions of mass-loss rates is dominated by Al2O3 comparison of our radiative transfer model the attenuated stellar and dust components alone grains, while the dust content of stars to the RR Aql visibility data of one of are denoted by the dotted and dashed lines, respec­ with high mass-loss rates predominantly our epochs. We conclude that a radiative tively. From Karovicova et al. (2011). exhibit substantial amounts of silicates. transfer model of the circumstellar dust shell that uses dynamical model atmos­ radius of 2–2.5 stellar radii and/or a sili­ pheres representing the central stellar cate dust shell with a typical inner radius References source can reproduce well the spectral of 4–5 stellar radii. This result is con­ Decin, L. et al. 2010, Nature, 467, 64 shape of both the visibility and the pho­ sistent with the scenario of a dust con­ Habing, H. J & Olofsson, H. 2003, Astronomy and tometry data (not shown). densation sequence where Al2O3 grains astrophysics library, New York, (Berlin: Springer) form closer to the stellar surface and Ireland, M. et al. 2008, MNRAS, 391, 1994 In addition to RR Aql, we determined ­silicate grains at larger radii, which is as Ireland, M. et al. 2011, MNRAS, in press (arXiv 1107.3619) best-fit dust shell parameters for other well consistent with the condensation Karovicova, I. et al. 2011, A&A, 532, A134 sources of our sample (S Ori, GX Mon, temperatures. Our best-fit values of the Ohnaka, K. et al. 2005, A&A, 429, 1057 Wittkowski, M. et al. 2008, A&A, 479, L21 and R Cnc). The mid-IR visibility and optical depths of the Al2O3 and silicate spectra of each star can be described by dust shells together with the mass loss Wittkowski, M. et al. 2011, A&A, 532, L7 an Al2O3 dust shell with a typical inner rates adopted from the literature may

A composite of images from VISIR and NACO of the red supergiant star Betelgeuse. The image is 5.63 by 5.63 arcseconds in extent and shows the exten­ sion of the nebula around the star to be some 26 milli-parsec (550 AU). The supergiant star itself is resolved by interferometry at 44 milli-arcseconds (3.4 AU). The larger image is taken with VISIR in ther­ mal-infrared filters sensitive to dust emission and the compact image (inside the occulting spot) is taken with NACO in J-, H- and K-band filters. Some of the non-spherically symmetric eruptions close to the star in the NACO image can be traced to much larger distances in the VISIR image. More details can be found in Kervella et al., A&A, 531, 117, 2011 and in Release eso1121.

28 The Messenger 145 – September 2011