6 Discussion and conclusions

We have confirmed through detailed polarimetric imaging the existence of remarkable circumstellar envelopes around the carbon stars R Scl and U Ant. They have been previously reported in CO radio lines (in the form of detached gas shells; Olofsson et al. 1996), in the case of U Ant in thermal dust emission (in the form of detached dust shells; Izumiura et al. 1997), and in scattered stellar light through direct imaging (Paper I). We have also established that part of the light scattered in these shells is significantly polarised, $\approx$$30{-}50\%$. The polarisation data made it possible to establish the spatial structure of the scattering medium, which takes the form of detached, geometrically thin shells. We analysed the polarised scattered light under the assumption that it is entirely due to dust scattering. The results for U Ant strongly support this, and makes it likely that the same applies to R Scl.

The polarimetry imaging of the light scattered around R Scl and the modelling of dust scattering are consistent with the existence of a thin (2 $\hbox{$^{\prime\prime}$ }$, or $1.1\times10^{16}$ cm) shell with a radius of 20 $\hbox{$^{\prime\prime}$ }$ ( $1.1\times10^{17}$ cm). The modelling further suggests that as much as 70% of the observed total scattered flux can be due to scattering by dust grains. The required dust mass to achieve this is estimated to be a few $\times10^{-6}~M_{\odot}$. Using this dust shell model we are also able to explain the measured thermal dust emission at far-IR wavelengths. However, there appears to be room also for other scattering agents, which produce no polarised emission. We suggest that this scattering is due to K and Na atoms, which have strong resonance lines inside the used filters. If this interpretation is correct, we believe that the most likely explanation to the absence of polarisation in the line scattered radiation is optical depth effects. New data on scattered light in filters not covering these resonance lines and spectroscopic information will shed light on this issue. We note here that in Paper I we concluded that the observed circumstellar scattered flux towards R Scl was too strong for only line scattering (the maximum scattered to stellar flux ratio is given by the ratio of the line width and the filter width). Hence, also this result is consistent with both dust and line scattering, but it depends on the uncertain calibration of the data.

The CO radio line observations towards this star give a different picture, but they are not easily interpreted since the emission may emanate from both a present mass loss envelope and a detached shell (the CO observational results are presented in Olofsson et al. 1996). The angular resolution of the observations do not allow a separation of these two components. We note here that such present mass loss envelopes exist in all the other objects with detached CO shells. We have used the numerical code presented by Schöier & Olofsson (2001) to model the CO line emission. Input data appropriate for a C-rich CSEs were also taken from this paper. It is possible to explain the observed line intensities towards the star with only a present mass loss envelope if the mass loss rate is as high as $1.1\times10^{-5}~M_{\odot}$ yr^-1. However, the observed CO(3-2) radial intensity distribution is much more extended than the model one, and the observed double-peaked CO(3-2) line profile cannot be reproduced. In addition, the magnitude of the mass loss rate is much higher than those provided by other estimates (e.g., Gustafsson et al. 1997), and it is incompatible with the star being an optically bright carbon star. Thus, the CO line emission must have a significant contribution from a detached shell. The inability to separate any present mass loss rate envelope emission from that of the detached shell makes it difficult to estimate the size of the detached CO shell. Olofsson et al. (1996) made a crude estimate assuming that all of the CO radio line emission comes from the detached shell. The result was a shell radius of 9 $\hbox{$^{\prime\prime}$ }$ and an upper limit of 12 $\hbox{$^{\prime\prime}$ }$ to the shell width. The shell radius must be regarded as a lower limit. The observations of Gustafsson et al. (1997), described below, suggests that the CO shell lies outside about 10 $\hbox{$^{\prime\prime}$ }$. Further modelling of the CO line emission has shown that a dust envelope of the size and mass estimated from the scattering data would under normal circumstances (e.g., with a normal dust-to-gas mass ratio the estimated dust mass and shell size corresponds to a gas mass loss rate in the range $(0.5{-}1.5)\times10^{-5}~M_{\odot}$ yr^-1) be detected in CO radio line emission (although this depends somewhat on the uncertain kinetic temperature), provided that the CO molecules are not photodissociated (and we know there exist much larger shells than that of R Scl in which the CO has survived, e.g., those of TT Cyg and S Sct). However, it would produce highly double-peaked line profiles in both the CO(2-1) and CO(3-2) lines, and there is no apparent evidence of this. Thus, with some considerable uncertainty, we conclude that the dust shell at 20 $\hbox{$^{\prime\prime}$ }$ contains very little CO gas, and by inference very little gas.

Additional data are available. Detections of emission in radio lines of HCN, CN, and CS are reported for this source (Bujarrabal & Cernicharo 1994; Olofsson et al. 1996). These molecules require high densities and temperatures in order to be excited, and have not been detected in any of the sources with large detached CO shells (Olofsson et al. 1996). It is possible that their emission originates close to the star in a present mass loss wind. Gustafsson et al. (1997) performed spectroscopic observations towards R Scl. They clearly detected KI resonance line scattered light of circumstellar origin, which declines as the third power of the angular distance from the star out to a distance of $\approx$10 $\hbox{$^{\prime\prime}$ }$ (beyond this the emission is lost in the noise). This is consistent with a steady mass loss rate wind extending to about 10 $\hbox{$^{\prime\prime}$ }$. Using these data they estimate a mass loss rate of about 10 $^{-6}~M_{\odot}$ yr^-1. Such an envelope produces much weaker CO radio lines than observed, e.g., only about 25% of the CO(1-0) line intensity is accounted for (using appropriate input data for a C-rich CSE), i.e., the presence of a detached shell is inferred.

Therefore, the data suggest the existence of three different circumstellar components, an inner one due to the present mass loss epoch of the star, a middle one producing the bulk of the CO radio line emission, and an outer one mainly composed of dust. We summarize this in Fig. 9.

Figure 9: Shells detected around R Scl and U Ant through different observational probes. The arrows indicate the direction of uncertainty in the shells inner and outer radii.

The polarimetric imaging of U Ant confirms the existence of multiple shells which scatter the stellar light, see Fig. 9 for a summary. However, only the outermost component, weak in total intensity, is detected in polarised flux, shell4. Modelling of the polarised intensities shows that dust scattering is able to explain all of the emission from this outer component, but less than 30% of the total scattered light observed towards U Ant can be accounted for in this way. Either a very different grain composition, without polarising properties, or e.g. resonance line scattering by K and Na atoms provides the bulk of the observed scattered light. The former appears unlikely since the dust component in shell4 accounts also for the measured IRAS fluxes. A simple analysis in Paper I indicates that the latter is certainly possible. In particular, the different morphological appearances in the two filters can be attributed to different optical depths in the KI and Na D resonance lines, and the AARP in the F77 filter suggests (at least partially) optically thin, isotropic scattering (i.e., scattering without a strong forward efficiency as expected from dust). In favour of this interpretation is also that only the shell4 component was (marginally) detected in a filter which contains no resonance lines (Paper I).

The CO radio line data of U Ant reveal a detached CO gas shell which coincide spatially with shell3 (Olofsson et al. 1996). The angular resolution of the CO observations is rather poor ($\approx$15 $\hbox{$^{\prime\prime}$ }$), and the dynamic range is limited. Therefore, we cannot exclude that there exists also CO emission from, at least, shell2 and shell4, but it must be considerably weaker than that of shell3. The detailed spatial structure of the CO gas can only be resolved by interferometer observations. Hence, our conclusion is that in the case of U Ant there is good evidence that the dust and the gas has separated.

In both stars we found evidence, stronger in the case of U Ant, of a shell of dust which has separated from the rest of the circumstellar medium. The reason for such a separation is not clear. In the case of U Ant it may be explained by a gas-grain drift scenario. The drift velocity is estimated to be about 3 km s^-1, a value which is reasonable for a mass loss rate which is estimated to have risen to a value of about $10^{-5}~M_\odot$ yr^-1 during the formation of the detached CO shell (Schöier & Olofsson 2001). In the case of R Scl the separation between the gas and dust shell is uncertain. Another possibility is hydrodynamical effects (Steffen & Schönberner 2000; Simis et al. 2001). These would tend to produce differences between the gaseous and the dusty media. Effects of this kind may also lie behind the multiple-shell structure seen towards U Ant.

Clearly, observations made in other filters and using other techniques are needed in order to make progress in this study. Direct and polarimetry imaging performed in high-quality filters containing no resonance lines will supply important information. Spectroscopic observations of the KI and Na D lines will help in disentangling the contributions from dust and gas scattering (see e.g., Mauron & Querci 1990; Plez & Lambert 1994; Gustafsson et al. 1997). In high resolution, where the stellar light is diluted, such observations will also make it possible to observe regions close to the star. R Scl and U Ant are both obvious targets once radio interferometers become available for southern sky objects.

S Sct and TT Cyg, two other bright carbon stars with detected CO gas shells, are good targets for a study similar to that performed here for R Scl and U Ant. However, these stars are in the vicinity of the galactic plane. The presence of numerous field stars and the more extended circumstellar medium around these objects make it more difficult to image the scattered stellar light (though it is feasible, and results will appear in a forthcoming paper). In these cases, it is possible to obtain optical absorption spectra towards background stars lying behind the shells. This novel technique has been recently used by Kendall et al. (2002) to probe the CSE of IRC+10216. They report successful detections of circumstellar KI and Na D resonance lines.

Acknowledgements

François Ménard is thanked for allowing us to use his original Monte Carlo scattering code. Financial support from the Swedish Science Research Council is gratefully acknowledged. DGD is grateful for a NOT/IAC graduate study stipend.