INTERSTELLAR SILICATE DUST IN THE z = 0.89 ABSORBER TOWARD PKS 1830-211: CRYSTALLINE SILICATES AT HIGH REDSHIFT?

Our observations, summarized in Table 1, were obtained using the IRS on the Spitzer Space Telescope (Werner et al. 2004). These data consist of 69 individual spectral exposures, each observed in two nod positions, totaling 198 minutes (3.3 hr) of integration on-target, centered at (α, δ)_J2000 = (18^h33^m3989, − 21^d03^m404). The two nods, obtained using the IRS staring mode, facilitate both background subtraction and the identification of spectral features induced by bad pixels. Given the spectral breadth of the silicate absorption features in previously studied quasar absorption systems (Kulkarni et al. 2007b, 2011), we chose to utilize the low-resolution mode (R ∼ 60–130), rather than the high-resolution mode (R ∼ 600), in order to reach a higher sensitivity (Houck et al. 2004). In addition to the LL1/2 and SL1/2 spectra, bonus (third) order SL and LL spectra were obtained, which facilitate joining the first- and second-order spectra. Our complete data set spans a spectral region from 5.13 to 39.90 μm; this broad wavelength coverage is crucial for the accurate determination of the underlying quasar continuum which we require to measure the strength of the broad silicate features.

The IRS slit cannot individually distinguish the two lensed PKS 1830-211 quasar images; thus, we cannot probe two distinct lines of sight through the absorber. The lower-wavelength data (SL) are obtained with a (36–37) × 57'' slit and a 18 pixel⁻¹ scale Si:As CCD, while the longer-wavelength data (LL) are obtained with a (105–107) × 168'' slit and a 51 pixel⁻¹ scale Si:Sb CCD (Houck et al. 2004). Given this resolution, we are unable to spatially differentiate the two lensed images, which are separated by (Δα, Δδ) = (− 0642, −0728) and embedded in a fainter Einstein ring (Winn et al. 2002), and instead we address, in this study, the composite spectrum which constrains the average dust properties within the quasar absorber. The NE lensed quasar image is significantly brighter at all wavelengths (Lehár 2000; Courbin et al. 2002; Winn et al. 2002), and we assume that our continuum is dominated by this line of sight. The weaker (SW) component is more heavily dust-obscured, since it passes through a region of enhanced star formation in the western spiral arm of the host (Winn et al. 2002), and will contribute more significantly to our measured silicate dust absorption, as it does in detections of molecular absorption (see, e.g., Lehár 2000; Frye et al. 1997). Furthermore, we note that several neighboring objects are spatially inseparable from our quasar spectrum. These sources include (see illustrations in Courbin et al. 2002 and Winn et al. 2002) an M4 dwarf star, a foreground galaxy at z ∼ 0.19, and possibly a second star and a second galaxy close to the z = 0.886 absorber host. As explained in Appendix B.2, none of these other sources is expected to contribute significantly to our observed spectrum.

Our analysis utilizes the basic calibrated data (BCD) reduced spectra, and associated uncertainties and bad pixel masks, produced by version 18.7.0 of the Spitzer automatic pipeline. These spectra have been bias and dark subtracted, and a characteristic flat field has been applied. Additionally, the data have been corrected for systematic effects such as droop and dark-current drift, and cosmic-ray hits have been identified and flagged in the corresponding masks.

We have further processed the Spitzer BCD spectra by utilizing IRSCLEAN (v1.9-2.0) to identify and remove rogue pixels and cosmic rays, and to reduce the background noise in the frames. In addition to replacing bad/rogue pixels identified by the masks and iterative algorithms, we have implemented a slightly non-traditional application of the IRSCLEAN software in which we also manually identified for replacement all pixels in the spectra backgrounds with visibly high/low signal in order to smooth the background. We did not apply IRSCLEAN-replacements, even for noted bad pixels, within the quasar spectrum or its environs, in order to prevent any spurious spectral signatures from being added to our data by the algorithm. We then manually removed bad/rogue pixels along or near the quasar spectrum using the epix task in IRAF, and replacing points with the median of the surrounding pixels. For pixels along the quasar spectrum, the unweighted average of bordering points along the spectral axis was used instead of the median. We were generally conservative with identifying and replacing these pixels, and only targeted strongly discrepant points, e.g., those pixels with saturated flux or isolated pixels with substantially higher/lower values than neighboring pixels. Pixels producing all other spectral features were retained as plausibly physical in origin, and those which were produced by transitory events, such as cosmic rays, or pixel imperfections were effectively removed when we combined the data from the two nod positions.

We combined the cleaned individual BCD exposures in each nod position using the IDL script coad, which automatically propagates the uncertainties and mask files, and then subtracted the backgrounds using the IRAF task imarith, propagating the uncertainties in quadrature. For the SL data we subtracted one nod position from the other, in order to remove the background contribution. For the LL data, however, there exists an anomalous scattered light feature (see Appendix B.1). This spatially broad and intense feature prevented us from subtracting the two nods to remove the background. Therefore, we combined with coad all of the off-target LL1 frames (obtained while the quasar was in the LL2 order) and vice versa to create off-target background frames, which were then subtracted from the spectra in each nod position. We applied the IRAF task fixpix to smooth over two weak foreground stellar spectra present in the off-target LL1 spectra utilized for background subtraction.

We extracted the one-dimensional (1D) spectra for each order using the Spitzer IRS Custom Extractor (SPICE) algorithm, set for a point-source extraction with optimal weighting, using the default extraction aperture in each order, with no subsequent corrections for fringing. Our final 1D spectra were extracted with a varying spatial width over the dispersion axis, as per the standard SPICE templates, sampling a broader spatial extent at the longest wavelengths within each order. In Appendix B.1 we describe how the results of the SPICE extraction were checked using different treatments of the background and aperture selection. A visual inspection of the two-dimensional (2D) spectra prior to extraction showed no apparent fringing in the SL2, SL1, or LL2 spectra, and only weak fringing in LL1. The IRSFRINGE manual stipulates that fringes are not spectrally resolved in the SL module, while in the LL module fringes resulting from one of the filters are present in the raw un-calibrated data, with an amplitude of 3%–10%, depending on wavelength. These are largely corrected through the applied pipeline flat fielding. As a test, we applied IRSFRINGE (v. 1.1) to our extracted LL data and found that the algorithm did not automatically detect any significant fringes, and when forced to correct for any measurable fringes, produced only a negligible statistical and visual improvement in the data quality, with small spurious spectral features resulting in a few cases. Therefore, we have applied no fringe corrections to our extracted spectra.

In order to improve the S/N for our spectra, we have combined the 1D extracted spectra from each of the two nod positions in an unweighted average. The flux density uncertainties have been propagated in quadrature, and reflect both the measurement uncertainty estimated through the BCD pipeline reductions and the scatter of the two nods relative to their mean. These uncertainties are dominated by variations between the two nod positions. Although the spectra extracted from the two nod positions are generally consistent, there exist regions with small discrepancies where, e.g., one of the nods exhibits a "convex" feature, while the other exhibits a "concave" feature. These small variations are at most 10% relative to the mean spectrum in the SL2, SL1, and LL2 data. In the LL1 data, a discrepancy on the order of ±15%–20% is observed over an extended spectral region from ∼22 to 26 μm; this feature does not affect our measurements of the 10 and 18 μm features in the z = 0.886 absorber. After consultation with the Spitzer Science Center staff, we conclude that the origin of this feature is likely scattered light, as discussed in Appendix B.1.

We have manually trimmed the nod-combined spectra to eliminate lower quality data points, and subsequently we normalized and joined the four SL–LL spectra to construct a single, continuous spectrum for the PKS 1830-211 system. These trimming and joining procedures are detailed in Appendix A.1. Our combined spectrum has then been shifted to the rest frame of the quasar absorber host, assuming an absorber redshift of 0.886. The final combined spectrum is illustrated in the left panel of Figure 1, with the individual SL and LL orders differentiated by color. There, and in all other figures, we depict the absorber rest-frame wavelength, unless otherwise noted.

Zoom In Zoom Out Reset image size

In order to search for and study the 10 and 18 μm silicate features, we have normalized our spectrum by a fit to the quasar continuum. As detailed in Appendix A.2, we explored a range of approximately 20 different normalization fits, and identified a third-order Chebyshev polynomial as producing the best fit to the underlying continuum. This quality assessment was based on a combination of a visual inspection of the continuum fits and resultant continuum-normalized spectra, and an examination of the relative contributions by the continua to the reduced chi-squared values for the template fits used in characterizing the optical depth and shape of the silicate absorption profile (for details, see Section 3). Our final normalization is illustrated with a black line in the left panel of Figure 1, with the resultant normalized spectrum of PKS 1830-211 depicted in the right panel. As is evident in this figure, there is a broad absorption feature present at ∼10 μm, as well as a shallower absorption feature near 18 μm. These features are associated with silicate dust in the z = 0.886 foreground absorber, as discussed in the remainder of the paper. Additionally, there are several weaker absorption features in the 5–8 μm region, and some emission features both longward and shortward of the 10 μm silicate absorption feature; the possible origins of these features are discussed in Appendix A.3, as they are not the primary focus of this analysis.

In order to understand the physical origin of the 10 μm silicate feature and the chemical composition of the dusty material, we have fitted a series of template optical depth profiles (see Tables 2 and 3) to our observed spectrum. Our fitting assumes simple radiative transfer through the cloud, such that I/I₀ = exp [ − τ], where τ ≡ a_nτ_norm; τ_norm is the optical depth profile for the template, normalized to have a maximum peak depth of 1.0 over the full spectral extent of profile specified in Tables 2 and 3. The peak optical depth normalization factor (a_n) for each template profile is selected to produce the minimum reduced chi-squared (χ²_r). In order to obtain the associated 1σ error bars, we have (cubic-spline) interpolated the two values in the χ²_r versus τ curve for which χ²_r − χ_{r, min}² = 1. The primary fitting has been performed over the 10 μm spectral region, i.e., 8.6–12.5 μm. This fitting region has been selected to focus on contributions to the χ²_r from portions of the spectrum near the absorption feature. For those objects in which the template profile additionally covers the 18 μm silicate feature, we have performed a second fit, termed the "full" fit, extending from 8.0 to 19.45 μm. This extended fitting region includes contributions to the χ²_r from not only the 10 and 18 μm features, but also from the region of the continuum between these two features, which as discussed in Appendix A may also contain some emission from non-silicate sources. In those template profiles for which only a more limited wavelength range is covered, we have reduced this expanded fitting range appropriately, as detailed in Table 4.

To constrain the composition of the material producing the 10 μm silicate absorption in the z = 0.886 absorber, we fit more than 100 unique optical depth profiles, derived from astrophysically observed and laboratory sources in the literature, as detailed in Tables 2 and 3. These templates probe a wide range of laboratory profiles for natural and synthetic terrestrial minerals, and environments including the solar system (comets), circumstellar material, the Galactic diffuse ISM, dense molecular clouds, and extragalactic objects (e.g., ultraluminous infrared galaxies; ULIRGs). For brevity, we include a subset of 18 of the most representative profiles in the bulk of this paper, as detailed in Table 2 (observational templates) and Table 3 (laboratory templates). However, in order to more fully explore the possible minerals comprising the detected silicate dust, we additionally examine an expanded subset of the templates, including 44 laboratory and 18 observational profiles, including comets, in Appendix C. The full range of profiles are illustrated in Figure 2 (observed) and Figure 3 (laboratory). This complete set of templates spans a range of chemical compositions and degrees of silicate crystallinity, ranging from the purely amorphous (diffuse ISM (Kemper et al. 2004), amorphous laboratory templates) to 10%–15% crystallinity (asymptotic giant branch (AGB) star; Sylvester et al. 1999; Kemper et al. 2001; Molster & Kemper 2005; ULIRG; Spoon et al. 2006; Kemper et al. 2011), to purely crystalline laboratory silicates.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The optical depth fits obtained using the 18 templates are presented in Table 4. We list the peak optical depth for fitting both solely over the 10 μm spectral region (8.6–12.5 μm) and over the 8.0–19.45 μm fitting range. In Figures 4 and 5, we overplot the fitted profiles on top of our PKS 1830-211 spectrum, using a template profile sampling of Δλ = 0.01 μm for uniformity. Considering our complete subset of 18 profiles, we find 0.11 ⩽ τ₁₀ ⩽ 0.13 for the observationally based profiles, and 0.10 ⩽ τ₁₀ ⩽ 0.28 for the laboratory-based profiles. Most of these are visibly poor fits, as discussed in the following sections. Our best-fit estimate of the peak optical depth is τ₁₀ = 0.27 ± 0.05 for the crystalline olivine hortonolite template. This best fit was determined from both a visual inspection of the fits and based on the χ²_r, and is at the highest end of the optical depth range derived using our template profiles.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In order to ensure sampling uniformity for all of the computed fits, we have (cubic-spline) interpolated the values in each of the template profiles at the wavelengths at which our PKS 1830-211 spectrum has measured data points. By sampling every template at the wavelengths of our PKS 1830-211 data points, we ensure consistency when comparing the derived fits. This is essential since some of the laboratory profiles have been finely sampled with significantly submicron measurements, while the observationally derived profiles are more coarsely sampled. A consequence of sampling at our observed spectral dispersion scale is that we may potentially miss extremely narrow lines, but studies examining unresolved atomic/molecular transitions in similar IRS spectra, e.g., star-forming galaxies (Smith et al. 2007), suggest that this is not a significant problem.

Although the template profiles are sampled at a range of resolutions, since observationally based templates cannot achieve the high resolutions of laboratory measurements, we do not believe that this is an issue for our analysis; the locations of the peaks play a more significant role than their shape. In order to test the effects of resolution variations, we have convolved three of the most-structured, crystalline olivine template profiles with two different Gaussians representative of the Spitzer IRS instrumental resolution; one of which represents the lowest achieved resolution typical of LL1, and one of which is more representative of an average spectral resolution across the wavelength range covering the 10 μm feature. We find that for the more average spectral resolution there is a negligible change in the derived optical depth, with Δτ = 0.01. For the worst-case scenario, the optical depth is at most lower by Δτ = 0.04–0.05, which is consistent within our 1σ stated uncertainties. Since the instrumental resolution varies with wavelength over each IRS spectral module, it is difficult to correct for the instrumental profile. Given that our tests indicate no significant impact on the derived optical depths, no significant improvements in the χ²_r for our best-fitting profiles, and no visual improvement in the quality of the fit, we have made no corrections for the resolution variations between our template profiles and the PKS 1830-211 spectrum.

While the laboratory-based profiles are arguably more precisely measured, and do not include the same observational uncertainties as astrophysical sources, we also include templates derived from astrophysical sources in this analysis for the following reasons: (1) they may contain compounds which are not commonly found terrestrially or synthesized in a laboratory; (2) they may trace minerals in temperature and density environ-ments more physically similar to our quasar absorber than can be achieved in ground-based laboratories; and (3) in order to establish the similarity or dissimilarity of higher-redshift extragalactic dust with that locally, it is essential to directly compare Galactic and other extragalactic dust sources with our sample.

3.2.1. Fits with Galactic Source Templates

3.2.2. Fits with Extragalactic Source Templates

3.3.1. Fits with Amorphous Silicate Templates

3.3.2. Fits with Crystalline Olivine Templates

Given the relatively poor fits for both the amorphous silicates and for the observational sources, and the fact that the more crystalline-rich sources provide better fits than those which are crystalline-poor, we next considered crystalline olivine silicates. We examined fits for four such olivine templates (Mg_2xFe_{2 − 2x}SiO₄) spanning a range of Mg/Fe ratios: synthetic forsterite (x = 1.0), natural olivine (x = 0.94), natural hortonolite (x = 0.55), and synthetic fayalite (x = 0.0). As illustrated in Figure 6, these minerals are able to explain the bulk of our profile shape and substructure, including the offset in the peak wavelength, the breadth of the feature, and the majority of the substructure peaks, without invoking secondary absorption and emission mechanisms such as would be required for amorphous silicates. The best fit is produced by hortonolite, both over the 10 μm fitting region, and when considering the expanded fitting region covering both the 10 μm and 18 μm absorption features, as illustrated in Figure 5. As is evident in Figure 6, hortonolite well reproduces the general shape of the profile, but cannot fully account for the leftmost peak (∼9 μm). Additionally, the rightmost peak (∼11 μm) is offset to slightly longer wavelengths relative to the PKS 1830-211 absorber spectrum. By comparison, forsterite, which produces the second best fit, is able to match the shape and location of the rightmost (∼11 μm) substructure peak, but finds the middle peak (∼10 μm) and the intermediate ridge between these two features offset to slightly lower wavelengths. This strongly suggests that a crystalline olivine with an intermediate Mg/Fe ratio could simultaneously fit both peaks. While natural olivine is such an intermediate profile, it provides a poorer fit than either forsterite or hortonolite, fitting neither peak perfectly, suggesting an unexplored ratio between 0.55 and 1.0 may provide the best match. In order to determine the exact Mg/Fe ratio, longer-wavelength data would be helpful as the similar profile shapes for the four olivine compositions are significantly different longward of 20 μm, as discussed by Jäger et al. (1998).

Zoom In Zoom Out Reset image size

3.3.3. Fits with Other Laboratory Templates

3.3.4. Evidence for Multiple Minerals in Combination

As our template fits have revealed that hortonolite produces the best fit, but that it cannot explain the ∼9 μm absorption substructure, we finally considered whether a combination of minerals could fully explain the absorption feature observed in the PKS 1830-211 spectrum. To test this we considered a two-mineral mixture within the dusty quasar absorption system, combining hortonolite and a second mineral, such that I/I₀ = exp [ − (a_nτ_{1, norm} + b_nτ_{2, norm})]. The optimal values for a_n and b_n are obtained simultaneously, with the uncertainties conservatively determined by ascertaining parameter values which produce χ²_r − χ_{r, min}² = 1 within a given parameter, while holding the other parameter fixed. The resultant bi-variate fits are presented in Table 5 and depicted in Figure 7.

Zoom In Zoom Out Reset image size

Table 5. Two-mineral Fitting of the 10 μm Feature

Profile1+Profile2	τ1	τ2	χ2r	Fit-λrange	τ2/τ1
Hortonolite + second silicate
Horton+silica	0.25 ± 0.05	0.02+0.03−0.02a	3.37	8.60–12.50	0.09
Horton+avg.serp.	0.26 ± 0.05	0.01+0.03−0.01a	3.72	8.60–12.50	0.02
Horton+enstatite	0.24 ± 0.05	0.02+0.03−0.02a	3.44	8.60–12.50	0.10
Horton+SiCb	0.27 ± 0.05	0.01+0.06−0.01a	3.74	8.60–12.50	⩽0.02
Amorphous olivine + second silicate
Horton+am.olivineb	0.26 ± 0.05	0.01+0.03−0.01a	3.75	8.60–12.50	⩽0.02
Am.olivine+SiC	0.07 ± 0.02	0.19 ± 0.06	6.74	8.60–12.50	2.49

Notes. Fits resulting from bi-variate profile combinations, i.e., exp [ − (a_nτ_{1, norm} + b_nτ_{2, norm})]. In the first column, we list the two combined minerals, followed by the peak optical depth for the first and second minerals, the reduced chi-squared of the fit, the wavelength range over which the fit was performed, and the ratio of the derived optical depths for mineral 2 and mineral 1; the nomenclature is as detailed in Table 4. All wavelengths are in μm. Hortonolite, which produced the lowest χ²_r in Table 4 is used for the first profile in every combination but the last. The best fit is the hortonolite + silica (χ²_r = 3.37) although it is only a marginal improvement over the pure hortonolite fit (χ²_r = 3.73). ^a1σ limit formally extends below 0.0; lower limit is considered to be effectively 0. ^bτ₂ value is at the lowest boundary explored in the calculation, τ₂ = 0.005, and is consistent with zero. We thus discount this fit.

Download table as:  ASCIITypeset image

We find that the combination of hortonolite with average silica, average serpentine, or enstatite all reduce the χ²_r for the fit over the 10 μm feature, and none of these profiles has significant features, for which would we need to account, in the 12.5–17 μm region. We have isolated the two best combinations in Figure 8, in which it is evident that the strongest contribution is from hortonolite. The largest difference between these two fits is in the region between the 9 μm and 10 μm features where the enstatite combination exhibits an additional, very small, inflection. None of these fits quite matches the height of the region intermediate between the 9 μm and 10 μm features, though, and while overall the fit is improved by the addition of the second mineral, there is only a weak contribution from this secondary material. For instance, in the combination of enstatite with hortonolite, enstatite is only contributing 10% of the column density relative to hortonolite. By adding these minerals we slightly reduce the hortonolite optical depth from 0.27 ± 0.05 to 0.24 at minimum (for enstatite). The optical depth of the second material is consistent with zero in every combination. We also considered the combination of SiC with hortonolite, but find that this is not an improvement.

Zoom In Zoom Out Reset image size

Alternatively, we note that it is possible that the ∼9 μm absorption feature is produced by ammonia (NH₃) absorption, which is seen in some ice-rich Galactic environments, such as W33A, a dust-embedded young stellar object (YSO; Gibb et al. 2000, 2004). As discussed in Gibb et al. (2004), NH₃ has an umbrella mode transition at 9.3 μm which, when NH₃ is a small component of the ice, shifts to 9.0 μm. We compared the modeled H₂O:NH₃ (100:9) ice at 50 K for W33A from Gibb et al. (2000) with our spectrum and find that both the overall shape and location of that absorption feature is well matched by our 9 μm absorption. Furthermore, ortho-NH₃ has previously been detected in the PKS 1830-211 z = 0.886 quasar absorption system by Menten et al. (2008), with a total column density of ∼10¹⁴ cm⁻².

In addition to considering combinations of hortonolite with a second material, we also considered amorphous olivine in combination with either hortonolite or with SiC. We find that in the latter fit, SiC dominates the fit, and that while an improvement over pure amorphous olivine, is still a relatively poor fit in comparison with crystalline olivine, since the combined profile is unable to explain the shorter-wavelength substructure features. The hortonolite and amorphous olivine fit is dominated completely by hortonolite, with little to no contribution from amorphous olivine. The amorphous olivine contribution produces a τ ⩽ 0.005, the lower limit considered in the combination, leading us to conclude that the contribution from crystalline silicates (1−N_am/N_cry) is ⩾95%, following Spoon et al. (2006).

Crystalline silicates have been observed in a multitude of Galactic and extragalactic sources, but they generally contribute <15% of the silicate mass. Bowey & Adamson (2002) have examined whether the broad 10 μm silicate feature could instead represent a superposition of numerous crystalline silicate features, but Molster & Kemper (2005) deem this scenario unlikely given that the corresponding expected crystalline silicate resonances at longer wavelengths have not been observed. We discuss in the following subsections a comparison of the PKS 1830-211 quasar absorber crystallinity with other Galactic and extragalactic sources, and with other quasar absorption systems, and finally we discuss the physical environments for crystalline silicates.

4.1.1. Comparison with Other Astrophysical Sources

4.1.2. Comparison with Other Quasar Absorbers

4.1.3. Physical Conditions

While we believe that crystalline silicates provide the simplest explanation for the observed substructure within the 10 μm absorption feature, we now explore a variety of scenarios to assess whether the observed features could be reproduced without crystalline silicates. The first alternative explanation which we explore is the possibility that the observed structure is the result of a combination of a broad absorption feature, produced by an amorphous silicate of olivine composition, and a series of other additional atomic and molecular absorption/emission features within the quasar absorber material. This possibility is worth exploring because (1) the 10 μm amorphous silicate feature is present in many Galactic and extragalactic sources; (2) this absorption system is exceedingly rich in molecular material (Wiklind & Combes 1996, 1998; Muller et al. 2011); (3) narrow absorption and emission features have been observed in ULIRGs (Spoon et al. 2006); and (4) previous detections of crystalline silicates in astrophysical sources, e.g., ULIRGs, have been much weaker than in our profile.

The first test of this scenario which we have implemented is to manually mask potential absorption and emission features within the broader 10 μm absorption feature. We have then attempted to fit the remaining points with a series of four amorphous olivine profiles which were selected as providing the best range of olivine templates (see Table 10 for template details). We have implemented these fits applying two different masking combinations: mask A which attributes the 9.3–10.05 μm and 10.35–10.85 μm regions to unidentified, superposed emission features, and mask B, which includes both of these purported emission features, as well as an absorption feature at 10.85–11.3 μm. (We have also considered two other masks which exclude the 8.65–9.25 μm spectral region, in which the enstatite, silica, or serpentine material, or ammonia, absorption, may be present, instead of the 9.3–10.05 μm region; but these two masks do not fit the data as well as mask A or mask B, and so we do not illustrate these fits.) We have adopted the same fitting procedure as utilized in Section 3.1. These fits are illustrated in Figure 9 and parameterized in Table 6. We present the χ²_r values in this table in order to facilitate comparisons between the different templates and masking schemes, but we note that the bulk of the structure in the 10 μm region is masked, and as a result relatively few points in the masked profiles are being considered in the χ²_r. For the less aggressive mask A, the amorphous olivine silicate fits remain too narrow/shallow relative to the PKS 1830-211 profile and peak at shorter wavelengths.

Zoom In Zoom Out Reset image size

As a second test we have employed the same masking of the PKS 1830-211 absorption spectrum, and we have additionally allowed the amorphous olivine templates to be arbitrarily shifted in wavelength space to produce the best fit. We permit this wavelength shifting because (1) as discussed in Appendix C, variations in silicate grain morphology can produce marked offsets in the peak of the optical depth profile; and (2) recent work by R. Nikutta (2011, private communication) suggests that sizable wavelength shifts in the 10 μm peak can be produced around quasars with a clumpy dust torus model which combines simultaneous silicate detections in absorption and emission, in conjunction with the rising quasar continuum. Although we are not looking at dust within a quasar torus, the work of Nikutta raises the possibility that a combination of absorption and emission by clumpy dust could contribute to the observed wavelength offsets. The results of this fitting are shown in Figure 10 and are tabulated in Table 6. We find that using mask B with shifting of +0.1 to +0.5 μm, we can achieve relatively low χ²_r values. However, as previously noted, this is based on substantial masking of the absorption feature, and there are few points which are being directly fitted within the deepest part of the optical depth profile; the bulk of the fitting leverage is being applied by the wings. Given the reduced number of sample points being fit, we do not consider the masked fits to be as reliable as those in our primary analysis.

Zoom In Zoom Out Reset image size

Furthermore, the masking scenario is contingent upon all of the identified emission and absorption features being physically explicable by molecular and atomic transitions. In order to determine whether there are plausible atomic and molecular lines which could produce these features, we have considered prominent lines which are seen in other extragalactic and Galactic sources (see Figure 11). As is evident from this figure, none of the commonly observed features are broad enough to explain the significant substructure within our absorption feature. We note that H₂ S(3) emission, which is commonly seen in ULIRGs, may be present in our z = 0.886 absorber, producing the small hump in the larger substructure peak. This emission, however, cannot explain the entire substructure peak, as the line is not only relatively narrow in comparison, based on the line widths in star-forming galaxies (Smith et al. 2007) and ULIRGs (Spoon et al. 2006), but it is also not centered on the broad substructure peak seen in our data. We also do not see prominent H₂ S(2) or H₂ S(1) emission at longer wavelengths, although these features could be lost in the noise. The emission feature from 10.35 to 10.85 μm could be associated with the [S iv] 10.51 μm feature, which is again seen in some ULIRGs (Spoon et al. 2006) and star-forming galaxies (Smith et al. 2007), but our feature appears broader, offset to longer wavelengths, and more prominent than the unresolved line seen in these other systems. The best candidate to explain the absorption feature from 10.85 to 11.3 μm remains crystalline silicates, based on the work of Spoon et al. (2006); however, we note that if we invoke a combination of amorphous silicates with molecular/atomic transitions to explain the bulk of the substructure in the profile, we require a much smaller contribution from crystalline silicates than claimed elsewhere in this paper.

Zoom In Zoom Out Reset image size

In general, these superposed emission features in conjunction with broad amorphous absorption appear inadequate to explain the detected substructure. Also, there is no strong evidence that all of these particular features can be physically produced and detected. Emission lines of H₂ and [S iv] are detected in ULIRGs, but these objects are undergoing extreme star formation and can be detected in emission. By contrast, the PKS 1830-211 quasar absorber host is a regular spiral which is barely detected in Hubble Space Telescope (HST) images, with no noted pronounced star formation. It seems unlikely that strong emission lines of these species would be produced in such an environment, and certainly not in the required line strengths/breadths to explain our substructure peaks.

Lastly, we address the possibility that the substructure we attribute to crystalline silicates is, in fact, a by-product of PAH emission overlaid on a broad silicate absorption feature. Given the location of the expected PAH emission features in the rest frame of the z = 0.886 absorber, illustrated in Figure 15, we deem this unlikely. While the 10.68 μm PAH emission feature in the rest frame of the absorber is consistent in location with the peak between the 10 μm and 11 μm absorption subfeatures, this would still require a shift in the amorphous olivine template and an additional absorption mechanism to explain the ∼9 μm feature. The subpeak between 10 and 11 μm (in the z = 0.886 absorber rest frame) aligns with the 5.7 μm PAH feature in the quasar rest frame, but this feature is quite weak in most quasars (Hao et al. 2007). We deem it more likely that our substructure originates from crystalline silicates than from a wavelength-shifted amorphous olivine profile in combination with superposed PAH emission, although we concede that this subpeak may be enhanced by some PAH contributions.

An alternative scenario to explain the structure in the PKS 1830-211 absorber spectrum is that we are detecting superposed amorphous silicate features from two absorption sources at different redshifts which serendipitously combine to produce the appearance of multi-peaked structure. This is plausible in that the NE quasar sight line is passing within 10 kpc of the center of a z = 0.19 foreground galaxy (Lovell et al. 1996), and so our spectrum is also probing the material in the outskirts of this z = 0.19 foreground galaxy, which may contain silicate dust. If we perform a bi-variate fit in which we consider an amorphous olivine template at the rest frame of our z = 0.886 quasar absorption host, combined with an amorphous olivine template at the rest frame of the z = 0.19 galaxy, as in Appendix B.2, we do not find this to be viable, however. At the wavelength at which we would expect 10 μm silicate absorption in the z = 0.19 foreground galaxy, we instead detect an emission-type feature. We, therefore, conclude that the structure in the 10 μm region of the PKS 1830-211 absorber spectrum is not the superposition of 10 μm silicate absorption from the absorber host galaxy and 18 μm silicate absorption from the z = 0.19 foreground galaxy. We also note that while such a superposition might account for two of the substructure peaks, it would not account for three, and would require a third absorption system to explain the ∼9 μm absorption feature, or once again invoke additional mineral/molecular absorption in this region. Furthermore, while in principle two absorption features could combine to produce the observed "W" shape formed by the 10–11 μm substructures, individual amorphous olivine silicate profiles are too intrinsically broad to produce a "W" of the required shape.

We have also repeated the bi-variate fitting both allowing for a foreground galaxy at any arbitrary redshift, in combination with the z = 0.886 absorber, and allowing for two systems at arbitrary redshifts out to z = 1, with possibly no absorption from the z = 0.886 system. These scenarios cover the possibility in which a dark object which is rich in silicate material was missed in previous studies, as well as the possibility that the z = 0.19 galaxy might in fact be at a higher redshift closer to the z = 0.886 galaxy, as suggested by Lehár (2000), or that the z = 0.886 galaxy might in fact be multiple galaxies, as proposed by Courbin et al. (2002). However, we do not find any such "hidden" object responsible for producing superposed silicate absorption features in our spectrum.

Finally, we have examined the possibility that we are detecting a combination of crystalline silicate absorption at z = 0.19 with amorphous olivine absorption at z = 0.886. While we do see structure near 6–7 μm, which we have attributed in Appendix A.3 to H₂O and other molecular absorption, we note that these features occur in the z = 0.19 rest-frame 10 μm spectral region. We thus tested whether this structure might be attributable to crystalline silicates at z = 0.19. However, when we fit the hortonolite spectrum to the z = 0.19 rest-frame absorption spectrum, we find that the hortonolite features are too narrowly spaced to explain the ∼6–7 μm features, and that the structure at longer wavelengths is not well reproduced by 18 μm crystalline silicates in the rest frame of the z = 0.19 galaxy. Thus, we discount this scenario.

Lastly, we consider the possibility that the PKS 1830-211 spectrum could be produced by a broad amorphous olivine feature associated with the z = 0.886 absorber in combination with superposed foreground or coeval emission. For instance, if the presumed z = 0.19 galaxy is instead a massive galaxy located at z = 0.886, as proposed by Lehár (2000), or the z = 0.886 host galaxy is instead comprised of two neighboring galaxies, as proposed by Courbin et al. (2002), then we might expect a blend of silicate absorption with either some limited emission or absorption reflecting a differing chemical composition, from the second galaxy. This scenario has largely been excluded by the bi-variate fitting tests in Section 3.3.4, in which we did not find compelling evidence arguing for two distinct chemical compositions, although we cannot exclude a weak contribution from a second source. Alternatively, we could be detecting broad silicate absorption along the SW quasar line of sight probing the molecule-rich material, while simultaneously detecting emission from, e.g., crystalline silicates, along the other quasar line of sight. However, the broad amorphous silicate absorption would still need to be offset in wavelength to fit our spectrum, as would the crystalline silicate emission template; and no single offset or explored crystalline silicate could simultaneously reproduce all of the observed peaks/substructure. Furthermore, as in the case of invoking atomic and molecular emission lines, it is not clear that such emission should be detectable from the z = 0.886 host galaxy, or any companion galaxies.

Alternatively, some of the foreground emission could be produced by a transient object within our own Galaxy or solar system, perhaps connected to the scattered light feature discussed in Appendix B.1. The flux from a foreground object could be probing a longer rest-frame wavelength region than the z = 0.886 absorber, where crystalline silicate resonances would be more abundant. For instance, Bouwman et al. (2010) find abundant crystalline silicates around a young, low-mass M4 star with a proto-planetary disk, and as discussed in Appendix B.2, there is an M4 dwarf within our slit. However, we do not believe that the flux from this particular star contributes significantly to our PKS 1830-211 absorption spectrum, as detailed in Appendix B.2, nor is it believed to be a particularly young stellar system, and so any crystalline silicates produced during its youth are expected to have dissipated. Furthermore, the likelihood of a transient source providing significant crystalline structure seems unlikely, and as discussed in Appendix B.1, we do not expect contamination from the scattered light feature to be strong enough to produce structure at the level we observe in the PKS 1830-211 spectrum.

As addressed in Section 2.2, we have joined the four individual nod-combined SL and LL spectra in order to form a single spectrum for PKS 1830-211, following a manual trimming of the data to eliminate lower quality data points, i.e., those points which appear visually discrepant or noisy. As discussed further in Appendix A.4, the exclusion of these data points does not significantly impact our derived fits and conclusions. The majority of the removed points constitute isolated data points where a bad pixel was not removed through our conservative spectral cleaning described in Section 2.1. The remaining points occur in regions where the two nods showed discrepant spectral features, indicative of bad pixel contamination in one position, or occur beyond 38 μm where the noise notably increases. Lastly, we have eliminated points at the edges of each spectral order which do not connect smoothly with the adjoining orders. As in Section 2.1, we have been as conservative as possible when making these cuts, so as not to eliminate spectral features which may be potentially physical in origin. We illustrate the spectrum which would result when excluding the bulk of this clipping in the left, bottom two panels of Figure 12 (clip v3 and v4). As is evident in this figure, there are several additional data points with large error bars, which are absent in our fiducial clipping (v1), but the overall shape of the spectrum is unaltered.

Zoom In Zoom Out Reset image size

Following this manual clipping of the data, we have normalized and stitched together the data in the SL2, SL1, LL2, and LL1 spectral orders, to form a single SL spectrum and a single LL spectrum for the PKS 1830-211 system. Within the SL and LL orders we have normalized the first- and second-order data to the bonus (third) order. The individual data points from the bonus orders have been excluded from our final spectrum, as they provide no additional information since they align with the data in the retained spectral orders and exhibit similar features within the overlapping spectral regions; furthermore, their inclusion would lead to uneven sampling over the spectral regions spanned by the bonus order relative to the full spectrum. We note that the joining of the LL1 and LL2 spectral orders occurs within the 10 μm feature of our z = 0.886 absorber, which is the focus of our analysis, and so could potentially impact the shape of this feature. However, as the orders align well, have overlapping data points at the edges of the orders, and match the bonus order spectrum, we do not believe that the observed substructure illustrated in Figure 1, and attributed in Section 4 to physical mechanisms, is affected by this joining.

We have generated the final PKS 1830-211 spectrum by scaling down the LL spectrum and joining it with the SL spectrum. An examination of the well-separated longer (radio) and shorter (X-ray) wavelength data for PKS 1830-211 in the literature does not convincingly argue for either scaling up the SL data or scaling down the LL data. Since the SL data generally have smaller associated formal uncertainties, we have given that order preference in setting the absolute flux normalization. This has resulted in final normalization-scaling factors of 0.97, 0.99, 0.91, and 0.87 relative to the original extracted data for the SL2, SL1, LL2, and LL1 data, respectively (including the shifts of the first- and second-order data to match the corresponding bonus order data), with a slight overlap in SL and LL data points at the location of the joining. These scaling factors are consistent with scalings employed in other analyses (e.g., Spoon et al. 2006; Smith et al. 2007). As our analysis is based on a quasar-continuum-normalized spectrum, any offsets in the absolute flux normalization will not impact our conclusions. We have also considered a slightly modified version of the joining in which we employed additional clipping at the long-wavelength edge of the SL spectrum, and then shifted the LL spectrum down by a larger factor. This alternative joining is illustrated in the clip v2 (and clip v4 for the less aggressively clipped data) panels of Figure 12 and does not significantly impact our conclusions.

We have, lastly, experimented with binning the data by a factor of three, using an unweighted average. While, as illustrated in the right panel of Figure 12, this can reduce the uncertainty associated with each data point, and produce a cleaner looking spectrum, it can also obscure the finer structure in the absorption feature. Given that we find indications of significant substructure in our analysis, we have opted against binning. As discussed in Appendix A.4, however, had we chosen binning it would not have altered our conclusions that crystalline olivine provides the best match to the 10 μm data feature.

In order to fit the silicate template profiles in Section 3.1 to the PKS 1830-211 absorber silicate features (in the rest frame of the z = 0.886 absorber), we have normalized the observed spectrum by a fit to the underlying continuum. We have first examined the data at both longer (radio) and shorter (X-ray) wavelengths to ascertain whether a single quasar continuum could be fit through an extended frequency range. However, given the discrepancies in the literature, the known temporal variability across a wide spectral range of the quasar, and uncertainty in the shape of the quasar spectral energy distribution in the IR region (Donnarumma et al. 2011), as well as the large wavelength separation between our data and these supplementary data, we decided to select a continuum which fits in detail solely our own IR spectral data.

We then explored a range of more than 20 fits to the quasar continuum. In the analysis of Kulkarni et al. (2007b, 2011), the underlying quasar continuum, in the quasar absorption systems, was well represented by a polynomial or power law. Thus, in this analysis we considered not only a linear polynomial and a power-law normalization, but also third- to fifth-order Chebyshev polynomials, third-order Legendre polynomials, and single component cubic-spline fits. The power-law and linear polynomial fits were determined using a least-squares algorithm, while the higher order polynomial fits were obtained using the IRAF task continuum with unweighted fitting in linear space, and no averaging of the sample points. A subset of the most representative of these fits is tabulated in Table 7 and illustrated in Figure 13. Unless specifically noted, each of these fits was obtained using masking to exclude the four prominent absorption features in the spectrum: 5.5–6.2 μm, 6.5–7.1 μm, 8.5–12.2 μm, and 14.8–18.3 μm in the z = 0.886 absorber rest frame. We also forced the IRAF-fitted curves to pass through points longward of 18.5 μm in order to constrain the curvature of the continuum normalization; the implications of this constraint are addressed below. Based on the relative contributions of the continua to the observed and laboratory fits over solely the 10 μm feature and over the combined 10 and 18 μm features, as discussed in Section 3.1, and a visual inspection of the resultant normalizations, we have determined that the third-order Chebyshev polynomial provides the optimal fit. As addressed in Appendix A.4, however, adopting an alternative normalization would not have altered our conclusions in this paper.

Zoom In Zoom Out Reset image size

Lastly, we have experimented with forcing the third-order Chebyshev polynomial to pass through/near certain critical parts of the spectrum. Since the continuum fitting is unweighted, this prevents anomalous, high uncertainty points, like those at the longest wavelengths, from skewing the fit. In Figure 14, we illustrate five alternative assignments of these regions. The fiducial fitting solely forces the curve to pass through points at the longest wavelengths (λ ≳18.5 μm), termed "A" in the figure. We have also examined forcing the spectrum through points between the 10 and 18 μm features (termed "B" in the figure), points just shortward of the 10 μm absorption feature (termed "C"), points just shortward of the 6 μm H₂O absorption feature (termed "D"), and points in the substructure peak near 9.6 μm (termed "E"). These variations are illustrated by the shaded gray regions in Figure 14, and listed in Table 7, with the resultant continuum variations shown by the red lines in the left side of Figure 14. These variations can impact the relative depth of our absorption feature, as addressed in Appendix A.4, and illustrated in the right side of Figure 14, but they do not alter our conclusion that the absorption feature in the PKS 1830-211 spectrum is best represented by crystalline olivine.

Zoom In Zoom Out Reset image size

An examination of our final normalized spectrum, shown in the right side of Figure 1, illustrates that in addition to the prominent 10 μm and 18 μm absorption features, which are the focus of this analysis, there are several additional emission and absorption features, which we briefly address here. First, there are two absorption features located near 6 μm and 7 μm, respectively. The first of these we attribute to water ice (H₂O). Galactic observations of, e.g., ice-rich-embedded YSOs have also found this feature, and a direct comparison of these Galactic profiles with our absorption feature finds good agreement in both the breadth and shape of the H₂O feature (Gibb et al. 2000; Chiar et al. 2011). The slightly longer-wavelength feature is also seen prominently in some Galactic environments, alongside the H₂O feature, e.g., in the W33A spectrum (Gibb et al. 2000). Although its origin is not completely understood in the literature, it is likely attributable to other molecular transitions associated with the ices. Given that the PKS 1830-211 z = 0.886 absorber is well documented to be rich in molecules, we find that these associations are plausible.

We also note the presence of a visible emission-type feature just shortward of the 10 μm absorption feature, near 8 μm, which we contend may be evidence of PAH emission. As illustrated in Figure 15, the location of the feature is consistent with the 7.7 μm complex and 8.3–8.8 μm PAH emission features in the rest frame of the z = 0.886 absorber. Following the classifications of Peeters et al. (2002) and van Diedenhoven et al. (2004), the shape of the feature is similar to a blend of perhaps Type A and Type C PAH emission, perhaps originating in a range of stellar sources. We note that if this feature does originate in PAH emission, it is unusual that we do not see evidence of the usually prominent 11.3 μm complex emission, although the relative strengths of the PAH features can vary, for example, due to variations in the ionization levels (see, e.g., Tielens 2008; Smith et al. 2007). These PAH ratio variations are illustrated in Figure 16 comparing the PAH features for two star-forming galaxies, NGC 5866 and NGC 3198, rescaled and overlaid on top of our PKS 1830-211 spectrum. Furthermore, it is possible that our absorption feature could be hiding some weak PAH emission, i.e., that the rightmost (∼11 μm) substructure peak may indeed be deeper or broader than our spectrum indicates. This would not alter our assessment that the primary constituent of the silicate dust is crystalline olivine, but might alter our ascribed species of crystalline olivine or the requisite secondary materials (Section 3). Finally, we note that in the Q0852+3435 quasar absorption spectrum examined by Kulkarni et al. (2011), there is also evidence of not only the two absorption features near 6 and 7 μm, but also the emission features shortward and longward of the 10 μm absorption; thus, these emission features may be physically connected to an enriched environment.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Finally, we address the "emission" feature longward of our 10 μm absorption. The physical origin of this feature is less clear. It may originate from one or a combination of mechanisms including weaker PAH emission features in the rest frame of the absorber, emission from the 7.7 μm PAH complex emission in the frame of the quasar host galaxy, H₂ S(2) emission in the frame of the z = 0.886 absorber (which would be consistent with evidence for the H₂ S(3) line at shorter wavelengths), or it may be related to the scattered light feature addressed in Appendix B.1, which provides the greatest contamination in this region, as is evidenced by the larger associated error bars. Given this range of plausible mechanisms, and the added complication of the contaminating scattered light, we do not attribute this feature to an artifact of the quasar-continuum normalization. We note, however, that treating this emission as an artifact, and eliminating it through an altered continuum shape, is encompassed by our tests explored in Figure 14 and Table 7, and we emphasize that these normalizations would not impact our qualitative conclusions in this paper.

Lastly, in order to ensure that our data processing choices regarding clipping, binning, and joining (Figure 12), the quasar-continuum polynomial (Figure 13), and the continuum shape (Figure 14) have not impacted our conclusions, we have repeated our template fitting described in Section 3.1 using the alternative normalizations in Table 7. We find that, as detailed below, while the continuum variations can have some impact on the optical depth, even the most extreme plausible normalizations do not alter our conclusions that the observed structure is best represented by crystalline olivine.

We find that neither the possible clipping, joining, and binning options, nor the variations in the shape of the adopted polynomial normalization, significantly impact our peak optical depth measurements. All explored clipping, joining, and binning options increased the derived optical depth by values within 1σ, and over the 10 μm region hortonolite remains the best fit. The linear and power-law normalizations produce optical depth measurements which are consistent over the 10 μm range, but we discount the fitting over the full fitting range because of their visibly poor fits at the longest and shortest wavelengths. We also discount the fifth-order Chebyshev, as we believe that it has too much curvature and is overfitting the continuum, and thus removing some of the residual substructure. The remaining continuum normalization fits are all consistent within 1σ of the fiducial fit, and also consistently find that hortonolite produces the best fit.

Finally, we examine the impact of the different shape constraints on the third-order polynomial. We find that the more aggressive forcing, which eliminates the emission peaks preceding and following the 10 μm absorption feature, can raise the optical depth to at most 0.42 ± 0.05 using the hortonolite template, or to a maximum of 0.37 using the crystalline olivine template composition which provides the best fit. The Δτ = 0.10 is nearly consistent within the 1σ uncertainties, relative to the fiducial fit. We also note that imposing these constraints we consistently find that crystalline olivine provides the best fit, although the precise species of olivine varies. We prefer our fiducial normalization to these alternatives, because it is able to better fit the lowest-wavelength data in our spectrum, which is underfit in these more aggressively forced polynomials. If we employ less aggressive forcing, as explored in test "f1" (see Figure 14), the 10 μm optical depth (τ₁₀) decreases slightly to 0.26, while the fit including longer wavelengths (τ_full) increases to 0.34; it is at this long-wavelength end that the continuum normalization differences are the most significant. We believe that our adopted polynomial, however, provides a more reasonable fit to the overall continuum and is less influenced by noisy points at the longest wavelengths. Lastly, using a polynomial shape which passes through the ∼9.6 μm substructure peak, we find that the optical depth is decreased to τ₁₀ = 0.20 and τ_full = 0.24, both of which are consistent within 1σ of our fiducial fits, and again, argue for crystalline olivines as the origin of the substructure.

First, we have identified an anomalous emission feature in our 2D LL spectra which we associate with scattered light. The relatively broad feature is adjacent to our quasar spectrum at approximately (α, δ)_J2000 = (18:33:36, −21:03:41.3) in each of our LL1, LL2, and LL3 spectra. The centroid of this feature is approximately 8 pixels away from the quasar spectrum at the wavelength relevant to our 10 μm absorption feature. This prominent anomalous feature exhibits a spatially narrower distribution at low LL wavelengths, which broadens progressively as the wavelength increases; at the longest LL1 wavelengths it appears to be bifurcated with the light generally concentrated in two spatially distinct prongs, and a diffuse contribution in-between. Thus, while at the shortest LL wavelengths we do not believe that this feature substantially contributes to our spectrum, at the longest wavelengths it visibly crosses through our narrow PKS 1830-211 spectrum.

We have determined that the origin of this feature is most likely scattered light, from a bright off-field object. A visual inspection of both ground-based and HST visible and near-IR data reveals no significant objects at the spatial location of the feature. Furthermore, we have utilized SPICE to extract the spectrum of this feature, and compared it with a catalog of known stellar spectra observed with the Spitzer IRS (Ardila et al. 2010), and with quasar spectra; the feature does not match any of these spectral shapes. The Spitzer Science Center help desk has confirmed that this feature is in all probability a very rare scattered light feature, of unidentified origin, although the most likely candidates are either an IR bright source near the galactic plane or a transitory object from our own solar system. Neither we, nor the help desk, could find any bright optical or IR sources in the immediate vicinity of our target. We have also ruled out detector features, such as latents or bad dark corrections, as the origin of this feature.

As this scattered light feature is spatially broad, and visibly crosses through the quasar spectrum at the longest wavelengths, it may be the origin of the observed discrepancies between the two LL1 nod positions in the 22–26 μm spectral region. This location corresponds to the 11.7–13.8 μm region in the rest frame of the z = 0.886 absorber. As discussed in Section 2.2, a discrepancy on the order of 15%–20%, in which one nod position exhibits a broad emission-type feature while the other exhibits a broad absorption-type feature, is present over this spectral range. In order to identify the origin of this discrepancy, which is larger than those observed in the SL and LL2 spectral orders, we have first examined the position of the slit-centering to determine whether perhaps an offset in centering is causing a foreground/background object to fall within the slit for one of the nod positions, but not for the other. However, the header information indicates that there are variations of at most (Δα,Δδ) = (0018, 0009) and (0020, 0014) within the two nod positions, and no significant offsets (within 1 pixel) in-between the nod positions; IRS exhibits pointing uncertainties on the order of 1'' (which is smaller than the pixel scale). There is no evidence to suggest that we have a slight pointing mismatch, which could result in the inclusion of extra flux from a foreground/background object in only some of the exposures, nods, or orders.

We have also explored whether the difference between the two nod positions could stem from the background subtraction. As we were unable to subtract the background by differencing the two nod positions because of the anomalous scattered light feature, we instead constructed a background generated from the combined off-source LL2 images. It is plausible that a background gradient, independent from the scattered light feature, might be contaminating one nod background or region of the chip but not the other. In order to check this, we utilized SPICE to extract a background spectrum from multiple locations across our LL2 background image, as well as across a region of each of the LL1 spectra not covered by the quasar or the scattered light feature. We do not detect any substantial variations between the intrinsic backgrounds in the two nod positions, nor do we detect any significant spatial variations in the subtracted background, indicative of either a gradient or of a region of slightly bad pixels underlying one of the nod positions. We have also confirmed that the feature is not a result of any treatment of the background by the extraction algorithm, by replacing the pixels within the scattered light feature of the 2D spectrum with representative background pixels and repeating the extractions; the extracted structural features remain unchanged. An additional test using the IRAF task apall confirms that treatment of the background in our extraction algorithm is not the origin of the derived structure.

Additionally, we have explored whether the difference in the 22–26 μm spectral region could stem from a localized event along one of the spectra, such as an isolated cosmic-ray burst or transitory event. In order to test this, we extracted the combined 20 exposure spectrum, for each nod, as well as four subset spectra with 5 exposures each. We find that all of the spectral features in each nod position, including the discrepancy, are virtually identical between these five extractions. Therefore, it is unlikely that an isolated transitory event produced the discrepancy.

Thus, the most likely explanation for the discrepancy in the 22–26 μm region is contamination from the scattered light feature, which peaks in intensity over this 4 μm region, and more strongly affects one of the nod positions than the other. An inspection of the 2D spectrum shows that the scattered light feature is both the most narrow and has the highest surface brightness at the short wavelength end of LL1, where it is well separated (∼8 pixels) from the quasar spectrum. As the wavelength increases, the breadth of this feature also increases. However, the potential impact of this increased contamination at the longest wavelengths is mitigated by the decrease in surface brightness of the feature with increasing wavelength. In order to determine whether a narrower extraction window might minimize this contribution, we have compared a SPICE extraction obtained using the default point-source extraction template, which is of variable spatial width becoming broader at longer wavelengths, with two manual template extractions, one of which is 3 pixels wide, and one of which is 5 pixels wide. Modifying the extraction template impacts the absolute calibration of the extracted spectrum, but as we are only utilizing the manual extractions to compare features within the spectrum, this is not an issue. While we do find that the continuum level is slightly different between the extractions, as expected, we find no significant differences in the shape of the PKS 1830-211 spectrum or in any of the spectral features, along either nod position. Thus, employing a narrower extraction template would not change our analysis. We also have no method by which to separate any possible contamination occurring directly over the location of the quasar spectrum from the quasar spectrum itself, as this scattered light feature does not exhibit any standard spectral type. We thus rely on the combination of the two nod positions to effectively cancel out the contamination pattern within this 22–26 μm region, with resultant larger uncertainties associated with the data points.

The second contaminant of our quasar absorption system spectrum comes from several foreground objects which are also present within the slit. However, none of these objects seems to be bright enough to affect our analysis, as we describe below. As discussed briefly in Section 2.1, in addition to the two images of the lensed quasar and the z = 0.886 quasar absorber host galaxy within the slit, a foreground star and a lower-redshift foreground galaxy are also encompassed by the IRS slit in all spectral orders. The brighter of the two quasar sight lines, termed the NE or A component in the literature, is located at 18^h33^m39931 and −21^d03^m3975 (J2000), while the dimmer, and more dust-obscured quasar sight line to the SW is located at (Δα, Δδ) = (− 064, −073) from the NE component (Subrahmanyan et al. 1990; Jin et al. 1999; Winn et al. 2002). The host galaxy for the absorber, which has been identified as an Sb/Sc spiral, is separated by (Δα, Δδ) = (− 033, −049) from the NE quasar component (Winn et al. 2002). The absorber subtends a physical scale of 7.6 kpc/'', assuming H_o = 73 km s⁻¹ Mpc⁻¹ and a standard ΛCDM cosmology. Our requested central slit position of 18^h33^m3989, − 21^d03^m404 places it close to the center of this quasar absorption system.

The foreground star which is also present within the slit, termed S1 in the literature, dominates the combined light distribution at optical wavelengths, but has a minimal impact at the mid-IR wavelengths under consideration in this analysis. This star, which is located at (Δα, Δδ) = (009, 053) relative to the NE quasar component, has been identified as a likely M4 dwarf star by Courbin et al. (2002) and Djorgovski (1992). Star S1 dominates the total light at visible wavelengths, contributing between 54% and 74% of the total light at F555W (V) and 68%–83% of the total light at F814W (I), based on the magnitudes for the star and other objects within the slit provided by Winn et al. (2002) and Courbin et al. (2002); differences in the total contribution by S1 reflect variations in separating the light contributed by different objects by these authors. In the near-infrared, however, S1 contributes a much smaller percentage of the total light: 32%–44% at F160W (H) and 17%–23% at F205W (K) (Lehár 2000; Courbin et al. 2002). Extrapolating these rapidly decreasing relative contributions to the spectral region covered by the IRS, we would expect ∼1% of the total flux to be contributed by S1 at the shortest (5 μm) wavelengths covered, and ≲0.01% at the longest wavelengths covered. These findings of a negligible contribution to the total flux are reconfirmed by rescaling the IRS spectra for three brighter early M dwarf stars (HD180617, GJ687, and GJ849) provided in the compilation of IRS stellar spectra by Ardila et al. (2010); we find ≈0.5% contribution to the total flux near 5 μm, and 0.001% contribution to the total flux near 35 μm. In the region where the z = 0.886 quasar absorber 10 μm silicate feature is located, we find only a 0.01% contribution to the spectrum by these rescaled, similar stars from the Spectral Atlas. The small contribution of the M4 dwarf to our combined spectrum is a result of the rapidly decreasing strength of the stellar emission at mid-IR wavelengths, with the increasing strength of the underlying quasar-continuum flux density. Furthermore, we have verified through an inspection of the HD180617 stellar spectrum that we do not expect any significant structural features to be produced in our spectrum from an M4 dwarf star. We thus exclude the possibility that the spectral features we see in the PKS 1830-211 spectrum are produced by this foreground star.

The second potential foreground contaminant to our quasar spectrum is the z = 0.19 (Lovell et al. 1996; Wiklind & Combes 1998; Courbin et al. 2002; Winn et al. 2002) galaxy, termed G2 in the literature, which also lies within the slit, and is separated by (Δα, Δδ) = (024, −249) from the NE quasar component. We do not believe that this galaxy will produce a significant contribution for three reasons. First, it is substantially (Δm_F205W = 3.35 mag) fainter than the NE quasar sight line in the near-IR and is less reddened than the quasar (Lehár 2000). Hence, any mid-IR emission produced by the galaxy should be negligible. Second, any spectral features contributed by the galaxy will be well separated spectrally from those of the 0.886 quasar absorber host galaxy. Third, the central regions of the foreground galaxy are not being directly illuminated by either of the quasar sight lines, and so material from the foreground galaxy should not be producing significant absorption in the quasar spectrum. However, the closer, NE quasar line of sight passes within 10 kpc of the center of this foreground galaxy (assuming a standard ΛCDM cosmology with H_o = 73 km s⁻¹ Mpc⁻¹) and so may be probing the outer regions of this z = 0.19 foreground galaxy. Thus, in order to confirm that none of the prominent features which we observe coincide with silicate absorption in a z = 0.19 galaxy, we have shifted the intermediate amorphous olivine profile (AmOliv) into the rest frame of the z = 0.19 galaxy and repeated our fitting procedure. As illustrated in Figure 17, we find that while the purported 10 μm feature could be explained by 18 μm silicate absorption in the z = 0.19 foreground galaxy, there is no corresponding 10 μm absorption feature for this foreground galaxy. At the wavelength at which we would expect to detect 10 μm silicate absorption in the z = 0.19 foreground galaxy, there is a weak emission-type feature detected instead. We, therefore, conclude that the structure we see in the PKS 1830-211 spectrum does not originate from a superposition of 10 μm absorption from the quasar absorber and 18 μm absorption from the z = 0.19 foreground galaxy.

Zoom In Zoom Out Reset image size

The deblending of the objects in this field, based on their HST WFPC2 and NICMOS images, remains uncertain, and it is possible that there are further additional objects contained within the slit. Courbin et al. (2002), Meylan et al. (2005), and Lehár (2000) have identified a second foreground star, termed star P, located at (Δα, Δδ) = (− 03, −05) relative to the NE quasar component; Winn et al. (2002) have instead argued that this object is the bulge of the z = 0.886 galaxy, and the current photometry is unable to distinguish between the two scenarios. Given that this object is 0.02–0.05 times fainter than star S1 at F205W, its emission would not have a significant impact on the derived profile should it indeed be a foreground star. Furthermore, Courbin et al. (2002) argue that the object considered to be a single z = 0.886 galaxy by Winn et al. (2002), may in fact be comprised of two galaxies near z = 0.89, one of which exhibits significant spiral features. The impact of two companion galaxies is addressed briefly in Section 4, in which we have searched for simultaneous absorption, or simultaneous emission and absorption, at arbitrary redshifts in order to explain the observed spectral features. However, this explanation does not adequately account for the spectral structure which we observe, which indicates that if two galaxies are contributing to the observed silicate absorption profile, they either have a similar chemical composition and redshift or one of the objects is dominating the absorption profile.

We first examine an expanded set of observationally based templates for astrophysical sources ranging from comets to star-forming galaxies. Although these profiles are inherently subject to more sources of uncertainty than the laboratory profiles, which are the product of controlled experiments, they can provide valuable insight into the astrophysical environment which might be lacking should we solely use laboratory templates. We list our expanded set of observational templates, separated by environment, in Table 8, with the derived fits listed in Table 9 and illustrated in Figure 18. It is apparent both visually and from examining the χ²_r of the fits, that none of these objects matches the profile in our system as well as crystalline olivine. This is significant because it suggests that our system is dissimilar from the Galactic ISM and from the ISM in several star-forming galaxies.

Zoom In Zoom Out Reset image size

The profiles which best match our system are the two comets and the ULIRG 060301-7934 (when excluding the absent 11.3 μm PAH feature). What unites these systems is that they are all crystalline-rich. While we do not have a precise estimate of the crystallinity for the Bradfield and Levy comets, comets, in general, are found to be up to 90% crystalline. The ULIRG 060301-7934 has a 13% crystallinity and provides a better fit than either ULIRG 18443+7433 or ULIRG 00397-1312 (Spoon et al. 2006), both of which exhibit slightly lower levels of crystallinity. These three ULIRGS were selected as the best matches from those in the Spoon et al. (2006) study; the ULIRGs with 6.5%–9% crystallinity generally provided even poorer fits.

The Galactic and stellar environments generally provide relatively poor fits. The best fit is from the AGB star, and since AGB outflows have been found to be crystalline-rich (10%–15%), this is consistent with our need for silicate crystallinity. The diffuse ISM sources, which were adequate for reasonably fitting the silicate absorption features in the Kulkarni et al. (2007b, 2011) quasar absorbers, are a poor fit here, being shallower and offset to lower wavelengths relative to our feature. The circumstellar environments, dense molecular clouds, and stars embedded within molecular clouds all suffer from this same general failing. Some of the molecular clouds, such as ρ Oph, characterized as a disturbed environment (Bowey et al. 2003), and the visually similar (not depicted) Serpens profile, produce significantly inferior fits when compared with, e.g., profiles for sources probing the Taurus molecular cloud. Interestingly, profiles for sources embedded within the dense molecular clouds tend to produce slightly better fits than those illuminating the dark molecular clouds, which may stem from the fact that young sources surrounded by circumstellar material can be rich in crystalline silicates (e.g., see review in Molster & Kemper 2005).

We also explore, in the context of the amorphous olivine silicate profiles, the range in derived optical depths which can occur as a result of variations in the shapes and sizes of the dust particle grains (Table 10), using profiles from a variety of literature sources, as noted in that table. The reviews by Henning et al. (2005) and Molster & Kemper (2005) emphasize that variations in grain size, shape, and metallicity can have an impact on the shape and peak wavelength of the silicate profiles. This is well illustrated in Figures 6 and 7 of Chiar & Tielens (2006) for variations in the dust particles' shapes, including solid versus porous particles, and solid glass versus a continuous distribution of ellipsoids. These differences are most pronounced in the 18 μm region.

We first explore the impact of the variations in grain shape by fitting three separate amorphous olivine profiles to our PKS 1830-211 absorption features: one which is typical of solid spheres, one which is typical of a porous continuous distribution of ellipsoids, and one which is intermediate between these two extremes. Figure 19 illustrates the visible differences in the fits which can result from these variations. As detailed in Table 11, the best-fit results from the profile representing the continuous distribution of ellipsoids with a porous structure, although none of the profiles adequately fits our substructure, as discussed elsewhere, and all have relatively poor χ²_r values. Furthermore, as detailed in Table 11, while these variations impact the quality of the fit, they have relatively little effect on the derived optical depth which only varies from 0.11 ⩽ τ₁₀ ⩽ 0.13.

Zoom In Zoom Out Reset image size

We next consider the impact of variations in grain size, considering three similar particle distributions with particle sizes of 2 μm, 1.5 μm, and those in the Rayleigh limit, all taken from Dorschner et al. (1995). We find that these size variations can have an even more significant impact both visually (see Figure 19) and in terms of the χ²_r (Table 11), compared with the explored particle shape variations. The best fit over the 10 μm range is produced by the 2.0 μm particles, while over the full fitting range the 1.5 μm particles provide a better fit. However, despite the range in relative quality of the fits, all poor in comparison with our crystalline olivine fits, the optical depths are consistently in the 0.11–0.13 range.

Given our results in Section 3 that the PKS 1830-211 z = 0.886 absorber silicate profile is best represented by crystalline olivine, we explore in this section the effects of varying the chemical composition (i.e., the Mg/Fe ratio), particle shape, and material temperature on the derived fits. To implement these tests, we examined a total of 19 different varieties of crystalline olivines, as detailed in Table 10. The derived fits using these profiles are illustrated in Figure 20, where it is evident that both particle shape and size variations can significantly alter the fit; temperature has a reduced effect over this region of the spectrum, but would play a larger role at longer wavelengths.

Zoom In Zoom Out Reset image size

We first consider the variations in composition by exploring six distinct olivine compositions, as well as two different hortonolite profiles, since there can be variations between different laboratory samples. These profiles include both naturally occurring (terrestrially found) compounds, as well as more synthetic blends, as described in the references given in Table 10. As discussed in Section 3.3.2, we see that as the Mg/Fe ratio is varied, the leftmost (∼11 μm) and central (∼10 μm) substructure features are increasingly well or poorly fit, and despite exploring seven olivine varieties, we do not find any that perfectly fit our data. However, the wide range of fits which we do find leads us to believe that a ratio of Mg/Fe exists that will fully, simultaneously fit these two features. The best fits are obtained using the two hortonolite profiles, an intermediate Mg/Fe ratio, with the profile from Jäger et al. (1998) providing slightly superior results. The next best fit is provided by forsterite, the Fe-free end of the olivine compositional spectrum. As with the amorphous olivine, we find similar fits if we fit over solely the 10 μm range, or if we expand fitting to include the full range. The best fit produces an optical depth of τ₁₀ = 0.27 ± 0.05, and the full set of profiles ranges from 0.23 ⩽ τ₁₀ ⩽ 0.28, well within the 1σ limits. Since these templates span the full range of Mg/Fe ratios, we anticipate that the true Mg/Fe ratio will likewise produce an optical depth within this range.

We next explore the effects of variations in particle shape using four chemically consistent compounds (Mg_1.9Fe_0.1SiO₄) close to forsterite, from Fabian et al. (2001). These variations, which include considering a naturally occurring particle distribution, as well as spherical particles, and two variations of a continuous distribution of ellipsoidal (CDE) particles produce even more visually and statistically significant variations in the quality of the fit than the chemical variations discussed above. The best fit is produced from the powdered olivine, and it is apparent that the spherical particle distribution is not a good fit. However, again despite significant variations in fit quality, the derived optical depth measurements are self-consistent within 1σ across both shape variations and fitting regions.

Lastly, we consider the effects of variations in temperature. As addressed in detail by Koike et al. (2006), variations in temperature can induce visible shifts in the profile features, particularly at long wavelengths. To investigate whether these can impact our fits, we have considered both fayalite (the pure Fe olivine) and forsterite (the pure Mg olivine) over a range of temperatures from 10 to 300 K. The variations for forsterite are minimal in terms of optical depth, although Δχ²_{r, 10} = 0.9. For fayalite, Δτ_{10, full} = 0.02, which is consistent within 1σ, and the Δχ²_{r, 10} = 0.24, which is somewhat smaller. From these tests, we conclude that particle shape, size, and composition may significantly impact the quality (χ²_r) of our derived fits, albeit not the optical depth, while in this regime temperature variations are largely insignificant.

Lastly, we explore an expanded set of laboratory-derived non-olivine silicate profiles including three amorphous silicate minerals, five crystalline pyroxenes, four phyllosilicates, three crystalline silicate blends, and four types of SiC (Table 10, with included references for template origins). These substances span a wide range of chemical compositions, some of which have been proposed as potentially significant contributors to Galactic dust sources, such as the phyllosilicates (Dorschner et al. 1978), and others of which are known to be prominent in specific environments, such as SiC which is found in meteorites and carbon stars. This large variety of compounds is illustrated in Figure 21. However, we find that none of the materials fits the silicate absorption feature as well as virtually all of the crystalline olivine silicates, although as discussed in Section 3, some of the materials such as silica, synthetic enstatite, serpentine, and SiC can fit a subset of the structural features. In conclusion, while in principle a combination of several of these minerals might be constructed to produce a plausible fit to the substructure, and certainly several of them are candidates to explain the ∼9 μm feature in combination with crystalline olivine, they do not provide a compelling alternative, by themselves, to crystalline olivine when explaining the origin of the detected absorption feature.

Zoom In Zoom Out Reset image size