Hackwell (1972) suggested that the spectra of many M-stars were not consistent with the view that the circumstellar dust is comprised solely of silicate dust, implying that other constituents should be sought. Treffers & Cohen (1974) made high spectral resolution observations of oxygen-rich stars and concurred with Woolf & Ney (1969) on the attribution of the circumstellar dust features to silicates, however they did not preclude the inclusion of other grain types.

There have been various attempts to classify the different oxygen-rich dust features seen in the IRAS LRS spectra of evolved stars (e.g. Little-Marenin & Little 1990; hereafter LML90, Sloan & Price 1995; hereafter SP95). LML90 have classified the variation in the spectral features from M-type AGB stars into six categories: featureless, broad, 3 component, sil++ (a "9.7 $\mu$ m'' feature with a strong feature on its long wavelength side centered at about 11.3 $\mu$ m), sil+ (a stronger "9.7 $\mu$ m'' feature with a weaker long wavelength feature) and sil (a strong "9.7 $\mu$ m'' silicate feature). They suggested that there is an evolutionary sequence in the spectral features, starting with a featureless continuum, then developing a broad feature, followed by a three component feature, a two component feature and finally increasingly strong silicate features. Following this work, SP95 devised another classification system and found that they could categorize the spectra into eight groups which formed a smooth progressive sequence from broad feature to "classic'' silicate emission. This classification system involved examining the fluxes at 10, 11 and 12 $\mu$ m (F₁₀, F₁₁ and F₁₂), and comparing the ratios of the fluxes, rather than a visual inspection of the feature shapes. Where their finding differed from those of LML90 was with respect to the 12.5 - 13.0 $\mu$ m feature. They found that this feature was apparent in approximately 40-50% of the spectra across all groups. Therefore, if the sequence is evolutionary, the 12.5 - 13.0 $\mu$ m feature is due to a dust-type whose formation is not related to dust evolution.

Sloan et al. (1996) continued this work on the IRAS LRS database and found that the 12.5 - 13.0 $\mu$ m feature was present primarily in the spectra of semiregular variables (SRs): 75 - 90% of SRb variables exhibit the feature, while only 20 - 25% of Miras exhibit this feature. Other than this strong tendency to be found in the spectra of SRb variables, the 12.5 - 13.0 $\mu$ m feature was not found to be related to any peculiarities that would distinguish the sources.

Onaka et al. (1989) had attributed the broad 12.5 $\mu$ m band seen in the spectra of some O-rich stars (see Hackwell 1972 and Vardya et al. 1986) to alumina (Al₂O₃). They also found that fits to nearly all their M star spectra could be improved by the inclusion of Al₂O₃ grains. They suggested that the only way to form the silicate grains was for them to grow on pre-existing grains of Al₂O₃which act as seed nuclei. However, for most polytypes of Al₂O₃, the infrared feature peaks at $\sim$ 11.5 - 12.0 $\mu$ m rather than 12.5 - 13.0 $\mu$ m. One polytype, $\alpha$ -Al₂O₃ (or corundum), has a peak in the right wavelength range, but also exhibits a feature at $\sim$ 21 $\mu$ m which is not seen in the LRS spectra. Therefore, Al₂O₃ seems unlikely to be the cause of the 12.5 - 13.0 $\mu$ m band seen in these spectra. This conclusion was supported by the work of Begemann et al. (1997), who suggested that the 12.5 - 13.0 $\mu$ m band is probably associated with silicates.

Justtanont et al. (1998) observed strong emission lines, attributed to CO₂, in the ISO-SWS spectra of stars which exhibit the 12.5 - 13.0 $\mu$ m feature. Their data showed that the strengths of the CO₂ emission lines and the 12.5 - 13.0 $\mu$ m feature are well correlated, implying that these features originate in the same region of the circumstellar envelope. Justtanont et al. (1998) asserted that the occurrence of correlated CO₂ and 12.5 - 13.0 $\mu$ m emission is indicative of a warm ( $\sim$ 650 - 1250 K) gas layer close to the star where the carriers of these features are formed. They also suggested that an increase in mass-loss would provide a mechanism to prevent formation and/or excitation of the CO₂.

Posch et al. (1999) used ISO-SWS data to characterize the observed 12.5 - 13 $\mu$ m feature and then compared this to the spectra of various appropriate minerals. They suggested that spinel (MgAl₂O₄) is the most likely candidate. The attribution of the 12.5 - 13.0 $\mu$ m feature to this mineral is discussed in more detail in Sect. 3.

Although much the focus of recent research has been on the nature of the 12.5 - 13.5 $\mu$ m emission feature, a number of other papers have focussed more on the general silicate features. Sloan & Price (1998; hereafter SP98) extended their previous LRS dataset (SP95) to include supergiants and S-stars. They found that the majority of supergiants fell into the "classic'' silicate groups defined for AGB stars. There was little evidence for supergiants with the broad feature. How their findings compare to the current work will be discussed later.

Sylvester et al. (1994, 1998) discovered the presence of 8.6- and 11.3 $\mu$ m "UIR'' emission bands (usually attributed to carbon-rich material, e.g. polycyclic aromatic hydrocarbons) superposed on the silicate emission features of a number of M supergiant stars, particularly amongst members of the h $\&\chi$ Persei double cluster. Such UIR-band emission was not found in the spectra of any AGB stars. Sylvester et al. attributed this stark contrast in emission properties to the presence of a chromospheric UV radiation field around the M supergiants, capable of dissociating the CO molecules which would otherwise have locked up most of the available carbon atoms in the circumstellar outflows. PAH molecules might then form from the liberated carbon atoms, and be fluorescently excited to emit in the IR by the ambient UV photons. Sylvester et al. only detected UIR-band emission from intermediate luminosity supergiants which had relatively broad silicate bands. The non-detectability of any UIR-band emission in their spectra of more luminous supergiants could be due to smothering of the UV photon field in the denser winds of these stars, or to lack of contrast of the bands against the much stronger silicate emission features of these stars.

Sylvester (1999) investigated the 10 $\mu$ m spectra of a number of oxygen-rich evolved stars whose IRAS LRS spectra had been classified as showing SiC 11 $\mu$ m band emission. Using UKIRT CGS3 spectra with higher S/N and spectral resolution, he found that two of these stars, both supergiants, instead showed 11.3 $\mu$ m UIR-band emission superposed on silicate features. The remainder showed standard 3-component O-rich dust features, with peaks at 10, 11 and 13 $\mu$ m, apart from one object with a self-absorbed silicate feature. There is therefore currently no evidence for SiC emission features in the spectra of any O-rich evolved stars.

1.2 Dust formation around oxygen-rich evolved stars

Scenarios for the formation of dust grains in circumstellar shells around oxygen-rich stars have been investigated by several groups over the last thirty years. It seems appropriate to discuss the types of dust grains that these models predict to appear in order to constrain our own attributions.

The first attempts to predict the types of grains that should be expected to form in these environments neglected the actual nucleation processes and concentrated on the chemistry (e.g. Gilman 1969 and references therein). The first investigation of dust formation processes in terms of nucleation and grain growth was by Salpeter (1974) who concluded that grain formation proceeded by nucleation of small refractory seed grains (i.e. oxides) onto which an "onion-layer'' mantle of the more abundant silicates could form. Since then the basic premise of grain nucleation and growth has remained the same (e.g. Sedlmayr 1989; Jeong et al. 1999), but the details change and the exact nature of the expected condensates is still being debated.

Pégourié & Papoular (1985) discussed the grain-types expected in circumstellar shells. Their model, based mainly on models of condensation in the solar nebula, implied that the precise nature of the grains formed is determined by the elemental composition and oxidation properties of the parent atmosphere, along with the density structure of the dust shell. According to their model: 1) the concentration of iron in silicates is always expected to be low (mole % Fe₂SiO₄(fayalite) $\sim$ 20%; and FeSiO₃ $\sim$ 10%); 2) Mg₂SiO₄(forsterite) forms before MgSiO₃ (enstatite) in a cooling atmosphere, but forsterite is converted into enstatite by gas-solid reaction with SiO. They did not expect the dust to be pure forsterite or even forsterite with some ( $\leq$ 20%) fayalite. Disequilibrium calculations showed that the dust shells of M-stars should also contain SiO₂, solid (metal) Fe, Ca₂SiO₄ and Al₂O₃.

Stencel et al. (1990) hypothesized the formation of "chaotic silicate''. In their scenario, a chaotic silicate forms from a supersaturated vapour containing metal atoms, SiO, AlO and OH in a hydrogen atmosphere. Inside the chaotic silicate, where both silicon and aluminium are less than fully oxidized, the higher reducing potential of Al would initially act to produce Al-O bonds at the expense of Si-O bonds. Thus, the stretching modes of solid, amorphous alumina would grow at the expense of the 9.7 $\mu$ m Si-O stretch associated with silicates. However, Si and O are approximately ten times more abundant than Al and therefore once the aluminium is completely oxidized, the Si and SiO components of the grain should begin to oxidize and thus increase the strength of the classic 9.7 $\mu$ m silicate band.

Tielens (1990) reviewed the thermodynamic and kinetic factors which go into determining which grain species form in O-rich circumstellar environments. This was taken to be the "classic'' condensation sequence and fits with the basic ideas of grain nucleation and growth presented here. This will be discussed in more detail in Sect. 4.

In a more recent paper, Kozasa & Sogawa (1997, 1998) suggested that seed nuclei are alumina (Al₂O₃) grains, so that the first dust grains that form would be "naked'' Al₂O₃ grains. These refractory grains can exist relatively close to the star. Further out, where the temperature allows formation of lower temperature condensates, these Al₂O₃ grains would act as seed nuclei and become coated with silicates. Such core-mantle grains would have a minimum size of $\sim$ 150 nm. This is very similar the Salpeter (1974) model. However, Kozasa & Sogawa also suggested that, even further from the star, homogeneous silicate nucleation can occur, so that there would be a population of very small (a few nanometres) pure silicate grains. They suggested that the classic 10 $\mu$ m silicate feature is due to these very small grains while the core-mantle Al₂O₃-silicate grains are responsible for the 12.5 - 13 $\mu$ m feature. We will return to this question in Sect. 3.

The utilisation of Al₂O₃ grains as seed nuclei was called into question by Jeong et al. (1999) who suggested that alumina is unlikely to be present in the gas phase and therefore could not be used to build the "critical cluster'' that constitutes a seed nucleus. However, Gail & Sedlmayr (1999) state that the relevant information needed to establish this is not available for alumina. Jeong et al. (1999) used TiO₂ as their nucleation seed material, after finding that it was thermodynamically and chemically the most favoured refractory species (see Gail & Sedlmayr 1998). Their models suggested that the first dust species to form, closest to the star, would be aluminium silicates (Al₆Si₂O₁₃), a result not found in other models. They also found that titanium-bearing dust should form. However, Ti is over thirty times less abundant than Al and 400 times less abundant than Mg, Si, and Fe, which probably precludes the formation of detectable quantities of Ti-rich dust. At somewhat lower temperatures, Jeong et al. expected Mg₂SiO₄ (forsterite), MgSiO₃ (enstatite) and SiO₂ to form. At even lower temperatures they predicted that iron oxides should form and eventually contribute $\sim$ 30% of the dust volume, contrary to the Lodders & Fegley (1999) statement that "stars do not rust''.

Gail & Sedlmayr (1999) predicted that M-star circumstellar shells should contain several different dust species. Their models, like those of Jeong et al. (1999), start with TiO₂ cluster formation; however, as mentioned above, Gail & Sedlmayr did not exclude the possibility of Al₂O₃ seed nuclei. They found the most abundant condensate to be olivine. The exact Fe/Mg ratio could vary and they even suggested that this ratio changes as the grain grows, so that there could be a gradient in the Fe/Mg ratio across the grains. They also found that metallic iron particles form (in agreement with Lodders & Fegley), as well as MgO (periclase) and a very small amount of SiO₂ material.

It is clear from the models discussed above that there is not a consensus as to the nature of the dust grains that form around O-rich stars. However, the models do provide insight into whether some dust species are at all plausible. The aim of the present paper is to compare and contrast the mid-infrared features found in the spectra of cool evolved O-rich stars. To this end, mid-infrared UKIRT CGS3 spectra have been obtained that have a significantly higher signal to noise ratio and a somewhat higher spectral resolution than those available in the IRAS LRS database, although the 7.5 - 13.5 $\mu$ m spectral coverage of these ground-based spectra is more restricted than the 7.5 - 22 $\mu$ m coverage of the IRAS LRS. We aimed in particular to determine whether any dust feature types are associated with particular types of star, and what can be learned from the differences in spectral features amongst the various stellar types.

$\begin{figure}\includegraphics[width=9cm]{h1934f1.eps}\end{figure}$

Figure 1: Mean spectra for each of the AGB star groups - progressing between the broadest feature (bottom right) and the narrowest silicate feature (top left)

1 Introduction

1.1 Previous observation of dust around O-rich stars

1.2 Dust formation around oxygen-rich evolved stars