RUSTY OLD STARS: A SOURCE OF THE MISSING INTERSTELLAR IRON?

Our sample consists of 35 highly evolved, (strongly) pulsating stars in globular clusters (Sloan et al. 2010, Paper I), observed using the Spitzer Space Telescope Infrared Spectrograph (IRS). These long-period variable stars lie near or above the red giant branch tips of 19 Galactic globular clusters, which range in metallicity from [Fe/H] = –1.62 to –0.10. Observed with Spitzer's low-resolution modules, the spectra cover λ = 5.2–38 μm with a resolving power of R ~ 60–120.

One carbon-rich star (Lyngå 7 V1) is discounted, leaving a sample of 34 stars. Of these, silicate features at 10 and 20 μm are seen in 24 objects, implying oxygen-rich chemistry. More pertinently, the remaining 10 stars either have very weak silicate features or appear to be "naked".

JHKL-band light curves are also available (Paper I) for 25 of our 34 objects, which allows us to construct the photometric fluxes of these objects at the pulsation phase of their IRS spectrum.

Figure 1 shows the L−[8] color for these objects, where the 8 μm photometry was derived by convolving the IRS spectrum with the Spitzer IRAC 8 μm filter. A truly naked photosphere should have L − [8] ≈ 0, once interstellar reddening has been taken into account⁷ (cf. Boyer et al. 2008). All 25 stars, however, have a positive color. This implies a flux excess at 8 μm which cannot be caused by commonly assumed dust species (silicates, alumina) under normal conditions of temperature and grain size. This excess flux at long wavelengths is confirmed by comparison to marcs photospheric model spectra (Gustafsson et al. 1975, 2008; McDonald et al. 2009) as shown below.

Zoom In Zoom Out Reset image size

From our sample, four example stars have been selected for detailed modeling: NGC 5927 V3, a highly evolved asymptotic giant branch (AGB) star with a strong, narrow, 10 μm silicate feature; NGC 6352 V5, a less luminous giant star with unusual dust features; and NGC 362 V2 and NGC 5139 V42, "naked" RGB-tip or AGB stars showing optical TiO bands (indicating they are oxygen-rich) and infrared dust excesses, but no silicate features (Smith et al. 1999; McDonald et al. 2009; Boyer et al. 2009).

For a dust-free comparison, we use a stellar atmosphere model created using the marcs code (Gustafsson et al. 1975, 2008) at 3500 K, [Fe/H] = –1.0, [α/Fe] = +0.3 (see McDonald et al. 2009 for details of these models). Being very similar objects near the giant branch tip, we expect the effective surface temperature below the dust-producing zone to be 3750 ±~250 K. We use the near-infrared flux to fit the photospheric continuum. A low-temperature, metal-rich model therefore represents an intentional bias toward a model with a more-luminous infrared spectrum compared to its near-infrared flux. This reduces the calculated excess to rule out disparity between the modeled and real photospheric spectra.

Figure 2 shows the observed spectra, alongside fits to the stellar wind contribution using dusty (Nenkova et al. 1999). Using the marcs model atmosphere as an input, we fit a stellar wind model using metallic iron grains (Ordal et al. 1988). For each model, we assume a radiatively driven wind; and a standard Mathis–Rumpl–Nordsieck grain size distribution (Mathis et al. 1977), given by n(a) = a^−q, q = 3.5 for a = 5–250 nm.

Zoom In Zoom Out Reset image size

We correct for interstellar reddening using the absorption profiles from McClure (2009). We further assume that A_V = 3.2E(B − V) and that A_V/A_K = 7.75. E(B − V) values for NGC 362, 5927, and 6352 are from Harris (1996), E(B − V) = 0.10 is assumed for ω Cen, based on Harris (1996) and McDonald et al. (2009). The correction applied is thus quite small and negligibly different from that applied in Paper I.

NGC 362 V2 and ω Cen V42 are best fit with a dust shell truncated to an outer radius (r_outer) of ~100 times the inner radius (r_inner= the dust formation radius). (Normally, fits are extended to r_outer/r_inner 1000.) This suppresses the flux at λ>30 μm. A similar effect can also be achieved by using a grain distribution with fewer large grains: both effects may be in play (we only model a truncated shell here). The strength, and indeed existence of, this suppression depends crucially on the subtraction of the infrared background (caused by Galactic interstellar medium and other cluster objects). While every effort has been made to make sure this has been done correctly, we cannot be conclusively sure that such a suppression and thus truncation exist.

In NGC 5927 V3 and NGC 6352 V5, we also observe other dust species, which emit at λ>8 μm. Identified species include silicates (magnesium-rich olivines and pyroxenes) at 10 and 20 μm and amorphous alumina (Al₂O₃) at 11–15 μm (optical constants from Draine & Lee 1984; Begemann et al. 1997). Related crystalline minerals with a smaller degree of amorphization are likely responsible for the sharp 9.6 and 13.1 μm features seen in NGC 6352 V5 (Sloan et al. 2003). Note that the fit to NGC 5927 V3 produces a fairly good match to the 10 μm feature, but at a wavelength which is too red. This can be caused by a number of factors, including grain size and porosity (Voshchinnikov & Henning 2008). The 20 μm feature has an unknown carrier: a promising candidate is magnesio-wüstite (Mg_xFe_1−xO; see, e.g., Posch et al. 2002; Lebzelter et al. 2006; Paper I). This provides a good fit to the feature in NGC 6352 V5. The comparatively red peak wavelength seen in NGC 6352 V5 further suggests that x ≈ 0; i.e., that it is near-pure FeO (Henning et al. 1995). We return to this feature later.

We could not fit a single-temperature dust model to these two stars, implying that the dust species are not in thermal contact with each other. Iron and alumina were fit at temperatures of ~1000 K. In NGC 6352 V5, silicates and FeO were fit at ~450 K; in NGC 5927 V3, the strong 10 μm peak suggests mainly glassy silicates, which we fit at ~400 K.

Table 1 lists our fitted parameters, including dust temperatures at the outer and inner edges of the dust envelope (T_outer, T_inner) and the size of the envelope used (r_outer/r_inner, r_inner ~ 2 R_*). We do not include mass-loss rates, as explained at the end of this section.

Table 1. Parameters of dusty Models Presented in Figure 2.

Parameter	NGC 362 V2	ω Cen V42	NGC 5927 V3		NGC 6352 V5
Assumed luminosity (L☉)	1826	1862	2000		2000
Assumed [Fe/H]	−1.16	−1.62	−0.37		−0.70
Assumed E(B − V)	0.05	0.10	0.45		0.21
References	1, 3	2	4		4
Wind component			$\overbrace{\qquad \quad \qquad }$		$\overbrace{\qquad \quad \quad \qquad }$
			A	B	A	B
Composition	100%	100%	100%	100%	96% Fe	70% Silicate
	Iron	Iron	Iron	Silicate	4% Al2O3	30% FeO
Tinner dust envelope edge (K)	800	1000	1100	400	1000	450
Touter dust envelope edge (K)	194	173	⋅⋅⋅	⋅⋅⋅	⋅⋅⋅	⋅⋅⋅
router/rinner	50	120	1000	1000	100	100
Optical depth at 0.55 μm	0.28	0.70	0.55	0.45	1.05	0.12

References. (1) Boyer et al. 2009; (2) McDonald et al. 2009; (3) E(B − V) from Harris 1996; (4) luminosity estimated, metallicity from Harris 1996.

Download table as:  ASCIITypeset image

Featureless infrared excess can be produced by a variety of circumstellar material. This can take the form of free–free emission, emission from shells of molecular gas (a "molsphere"; e.g., Tsuji 2000), extremely large silicate grains, or amorphous carbon grains. Figure 3 shows modeled fits for these possible emission sources, modeled in an identical way to the iron fits listed above.

Zoom In Zoom Out Reset image size

Ionization in a chromosphere or shock fronts propagating through the outer atmosphere can give rise to free–free emission, as observed by radio observations toward Galactic Miras (Reid & Menten 1997). The optical depth of free–free emission in a region of thickness r can be calculated as

$$\begin{equation} \tau _\nu = \int _r 3.28 \times 10^{-7} \left(\frac{T_{\rm e}}{10^4 {\rm K}}\right)^{-1.35} \!\! \nu _{\rm GHz}^{-2.1} \left(\frac{n_{\rm e}}{\rm cm^{-3}}\right)^2 \frac{\delta r}{\rm pc}, \end{equation} \tag{ 1 }$$

where the electron density, n_e, is recoverable from the electron temperature, T_e, via the Saha equation. At low frequencies, where τ_ν 1, the emission scales as F_ν ∝ ν⁻² for an isothermal, isobaric region; and F_ν ∝ ν^−0.6 for an isothermal wind expanding with constant velocity. For ω Cen V42, we find our spectrum would require τ_ν = 1 at 3–4 μm, and that F_ν ∝ ν^−1.6 for τ_ν 1. This more-closely approximates an isobaric density profile, which would be difficult to achieve in an expanding wind. Very similar conditions would have to exist in all 34 of our stars, as the shape of their spectra are very similar.

An isobaric chromosphere producing a spectrum with this optical depth cannot exist except in the following case: the wind must be mostly ionized (6000 K) and either the chromosphere extends to an unphysical radius (70 R_☉) and/or the mass-loss rate becomes unphysical (10^−5.5 M_☉ yr⁻¹) for a thermally expanding wind. Departing from an isobaric state makes conditions even more prohibitive. While chromospheres may be present in these stars, this infrared excess is not seen in chromospherically active stars that are less evolved (McDonald & van Loon 2007; Boyer et al. 2009). In V42, circumstellar ionization appears dominated by pulsation shocks. Reid & Menten (1997) show that, for Galactic Miras, the temperature and density increases provided by shocks in the outer atmosphere are insufficient to produce an ionized layer that would be optically thick in the infrared. We thus rule out free–free emission as a dominant contributor to our spectra.

The molsphere creates a dense "forest" of molecular lines. Water lines are the most important, and cause emission throughout the spectrum, but mainly in the 6–8 μm and >20 μm regions. This emission is seen in some of our example stars (Paper I). However, the water bands in our spectra are very weak compared to stars with naked molspheres (e.g., β Peg— Tsuji et al. 1997). Gas-phase model emission spectra (Figure 3), created from the line list of Partridge & Schwenke (1997) using SpectraFactory⁸ (Cami et al. 2010), fail to reproduce the flatness of the spectrum for any reasonable combination of temperature and column density. Figure 3 shows the highest column density we consider feasible, and even this contains ~10⁻⁷ M_☉ of water or ~0.01% of the star's oxygen. Furthermore, most molsphere host stars still require an extra opacity source (such as iron grains) providing flux at 6–8 μm (Verhoelst et al. 2009). Infrared molecular emission should therefore not make a significant contribution to our spectra.

Of the commonly considered dust species only metallic iron and amC produce a featureless infrared continuum that matches the observed "naked" spectra. Other dust species (e.g., graphite) produce either strong spectral features or inflections which do not fit observed spectra. Spectral features can be damped by unusual conditions, notably exceptionally large grains (Höfner 2008): these cannot simultaneously reproduce both the excess at λ < 8 μm and λ>20 μm at the observed levels (Figure 3) and are not expected in metal-poor stars where grain growth may be hampered by a lack of constituent elements.

Observationally, amC can also provide an acceptable fit to our example stars over most of the spectrum. The remaining discrepancies may be fixable by fine-tuning the wind parameters (grain size, density distributions, temperatures, background subtraction, etc.). Detection of amC as dominant dust species in an oxygen-rich environment would be unexpected. In giant stars, carbon and oxygen are bound into CO near the stellar surface, locking away the least abundant element. The remaining majority element, either carbon or oxygen, dominates the dust chemistry. CO may be dissociated in atmospheric shock waves or by chromospheric UV irradiation (Beck et al. 1992). This can lead to the production of small amounts of other carbon-rich molecules: mainly CO₂ (observed in our stars: see Paper I), but also HCN and CS (Duari et al. 1999). Small quantities of carbon-rich dust may then form (Höfner & Andersen 2007), though oxygen-rich dust production is still expected to be dominant. In our sample, however, the featureless component often appears to be the dominant species, and no evidence for oxygen-rich dust is found.

Having ruled out the alternatives, we conclude that the source of the dust excess must be iron grains, and that these must dominate the condensed material in some stars. If we naïvely assume identical populations of spherical grains, the fraction of warm silicates in the "naked" stars being 1% of the iron content (Figure 3; the same would be true at a 3% level for an amC component). The silicate fraction may increase if iron grains are made smaller and oblate, thus presenting a higher cross section per unit mass: extreme needle-like grains are required if silicates are to become the dominant dust species by mass. As we do not know the grain shape, it becomes difficult to estimate a mass-loss rate for an iron wind. Spherical grains produce unfeasibly low expansion velocities of order ~m s⁻¹ for a pure iron wind; expansion velocities for NGC 5927 V3 and NGC 6352 V5 were not calculated as the model assumes a zero initial velocity for both temperature components, which will not be the case if they form at different radii. The mass-loss rates and expansion velocities can also be significantly modified by altering, e.g., the grain size or density (single crystal versus conglomerate), but reasonable changes still fail to produce feasible values, except for extremely elongated grains.

This implies that iron alone may not be able to drive a dusty wind in these stars. Other factors may also accelerate the wind. These could include small amounts of amorphous carbon (Höfner & Andersen 2007); energy deposition either from shocks created by stellar pulsations (Bowen 1988), or magnetic reconnection (Pijpers & Hearn 1989); increased opacity from circumstellar molecular gas (Elitzur et al. 1989); stellar pulsation, either by direct acceleration or dissipation of shocks; or line-driving of H i or Ca ii atoms, or H₂O molecules (e.g., Elitzur et al. 1989; Bowen 1989). The driving mechanism is thus likely to be complex, with radiation pressure on metallic iron being only one of several accelerants which drive matter from the star. Obtaining an accurate radial profile of the wind's outflow velocity will be crucial in determining which factors dominate this acceleration.

Efficient production of metallic iron by oxygen-rich AGB stars would explain several iron depletions. Gas-phase depletion of iron extends to the interstellar medium (Savage & Bohlin 1979). Iron grains contribute to infrared emission from this medium (Chlewicki & Laureijs 1988) and iron is seen to be destroyed in shocks (Lebouteiller et al. 2008). Iron is also depleted by >90% in the gas phase of planetary nebulae (Delgado Inglada et al. 2009). Direct evidence for iron condensation in AGB winds comes from depletion patterns in binary post-AGB stars. In these stars, the wind is captured in a circumbinary disk, from where gas re-accretes onto the star. These stars show large depletions of refractory elements, with iron being among the most depleted (Maas et al. 2005) thus indicate a history of iron-rich dust production.

The 20 μm feature, seen in NGC 6352 V5, is also observed in NGC 5927 V1 and Terzan 5 V6. It has a width (FWHM) of ~3μm and a line-to-continuum ratio up to 50%. This feature has previously been seen in low mass-loss stars; it correlates with the presence of an 11 μm shoulder, and 13.1 and 28 μm peaks (Little-Marenin & Little 1988; Sloan et al. 2003). The related peaks are also present in our sample, and we confirm that the 20 μm feature disappears for more evolved (i.e., more luminous) stars. The 20 μm and associated features are barely detectable in the naked stars, strongest in stars with weak silicate emission and disappear when silicates dominate (Figure 4). This suggests they are produced either during the formation or destruction of silicate grains. The identification of the feature with Mg_xFe_1−xO is unconfirmed mainly because it does not predict the associated features, but the proposed identifications for those (corundum, pyroxenes; Sloan et al. 2003) do not provide good fits at 20 μm, making it plausible that a range of minerals form under these conditions. We suggest that these features may be caused by the chemical modification of silicate grains: silicate grains which adsorb iron into the lattice have higher opacity, and can be heated until the iron-bearing component dissociates from the grain (Woitke 2006). This is expected to occur at any radius from the star at which dust formation is efficient (Gail & Sedlmayr 1999). In this model, metallic iron is produced first, but at higher metallicity and mass-loss rates, the iron becomes partly incorporated into other grain types including iron-oxide and iron-containing silicates.

Zoom In Zoom Out Reset image size

Studies of inclusions within meteoric primitive solar system material have traced dust grains to specific origins in AGB stars, supernovae and novae, through their isotopic ratios (Zinner 2003; Messenger et al. 2005). Recently, the first pre-solar iron-oxide grain has been identified (Floss et al. 2008). The iron and oxygen isotopic ratios in it are indicative of an origin in an AGB star. This supports the hypothesis that AGB stars are a source of interstellar iron dust, and have contributed to the solar system's iron content.

Figure 4 shows the variation of dust types (see Paper I) with bolometric magnitude and metallicity. Silicate-rich winds are only seen at high luminosity or high metallicity, with iron dominating the outflow in lower-luminosity, lower-metallicity stars. This indicates that the dust condensation sequence depends on metallicity and luminosity. Gail & Sedlmayr (1999) show that, under equilibrium conditions, iron is only expected to condense before silicates in regions with high gas pressure. If this applies to our stars, it implies high gas pressure at low metallicity and/or low luminosity, as may be expected if the outflow velocities where iron condenses (at the outside edge of the molsphere) are very low. Under non-equilibrium conditions, grain growth chemistry is strongly controlled by kinetics (Tachibana et al. 2009), which may favor the formation of iron.

In Gail & Sedlmayr (1999), silicates form before iron at lower densities and temperatures. The restriction of silicates toward higher metallicity and luminosity (Figure 4) suggests that their circumstellar envelopes are more extended, favoring silicates. Silicate production is expected to increase with both stellar metallicity and luminosity (van Loon 2000).