Near-infrared spectra of clinopyroxenes: Effects of calcium content and crystal structure

Transcription

1 Meteoritics & Planetary Science 46, Nr 3, (2011) doi: /j x Near-infrared spectra of clinopyroxenes: Effects of calcium content and crystal structure Rachel L. KLIMA 1*, M. Darby DYAR 2, and Carlé M. PIETERS 3 1 Applied Physics Laboratory, Johns Hopkins University, Laurel, Maryland 20723, USA 2 Department of Astronomy, Mount Holyoke College, South Hadley, Massachusetts 01075, USA 3 Department of Geological Sciences, Brown University, Providence, Rhode Island 02912, USA * Corresponding author. rachel_klima@jhuapl.edu (Received 15 May 2010; revision accepted 18 November 2010) Abstract Pyroxenes are among the most common minerals in the solar system and are ideally suited for remote geochemical analysis because of the sensitivity of their distinctive spectra to mineral composition. Fe 2+ is responsible for the dominant pyroxene absorptions in the visible and near-infrared, but substitutions of other cations such as Ca 2+ change the crystal structure and site geometries and thus the crystal field splitting energies of the Fe cations. To define spectral systematics resulting from major pyroxene cations (Ca 2+,Mg 2+, and Fe 2+ ), we focus on a suite of pyroxenes synthesized with only Ca 2+,Mg 2+, and Fe 2+ in the two octahedral sites, specifically examining the effect of Ca 2+ on pyroxene absorption bands. The modified Gaussian model is used to deconvolve pyroxene spectra into component bands that can then be linked directly to crystal field absorptions. In orthopyroxenes and low-ca clinopyroxenes, Ca 2+ -content has a strong and predictable effect on the positions of the absorption bands. At a threshold of Wo 30, the crystal field environment stagnates and the bands cease to change significantly as more Ca 2+ is added. At Wo 50, when most of the sites are filled by Ca 2+, band positions do not change drastically, although the presence and strengths of the 1 and 2 lm bands are affected by even trace amounts of Fe 2+ in the site. It is thus apparent that next-nearest neighbors and the distortions they impose on the pyroxene lattice affect the electronic states around the Fe 2+ cations and control absorption band properties. INTRODUCTION Characterizing the specific pyroxene mineralogy of rock-bearing surfaces throughout the solar system allows us to address questions about the magmatic evolution and cooling history of those bodies. Pyroxene spectra are in most cases easily distinguished from other minerals, and band positions are well-known to vary in response to total iron and calcium content. Laboratory studies of natural pyroxenes have provided a background for near-infrared spectral analysis (e.g., Adams 1974; Hazen et al. 1978; Rossman 1980; Cloutis and Gaffey 1991; Burns 1993; Sunshine and Pieters 1993). However, our understanding of the crystal chemical constraints on the diversity of pyroxene spectra is inherently limited because most pyroxenes that have been measured and well-characterized in the literature are necessarily terrestrial and often contain exsolution, inversion, zoning, and or significant amounts of nonquadrilateral components. Understanding the relationship between the crystal structure of pyroxenes and their reflectance spectra is key to using near-infrared spectroscopy for remote geochemical analysis. Pyroxenes have two octahedral cation sites, and, each with different geometries. Cations exhibit a preference for either the or site based on their size and charge. The site is more distorted and larger than the site, and thus accommodates large cations. When the site is not completely filled by large cations such as Ca 2+, Fe 2+ prefers the site over the site both because Fe 2+ is slightly larger than Mg 2+ and because the distortion in the site results in a greater crystal field stabilization energy for Fe 2+ cations (Burns 1993). This 379 Ó The Meteoritical Society, 2011.

2 380 R. L. Klima et al. Diopside Hedenbergite Augite Pigeonite CaMg C2/c CaFe Pbca Mg P2 /c 1 Pbca Fe 2 2 Enstatite Ferrosilite Fig. 1. Configurations of the and polyhedra in different space groups within the pyroxene quadrilateral, viewed down the a-axis, with c horizontal and b vertical. Diopside, hedenbergite, and augite are all in the C2 c space group and have somewhat similar geometries in their and polyhedra. Enstatite and ferrosilite are very similar both at the endmembers and across the orthopyroxene composition range. Pigeonite is in the P2 1 c space group, but has coordination polyhedra that are more similar to the Pbca orthopyroxenes than the C2 c augites. preference is not absolute, and the amount of Fe 2+ partitioned into the and sites depends on the bulk composition of the pyroxene, temperature and pressure of formation, cooling rate, and subsequent metamorphism (e.g., Wang et al. 2005). The character and absorption band properties of near-infrared pyroxene spectra are a function of the Fe 2+ occupancy of the and sites as well as the specific geometry of those sites. The spectra thus represent a wealth of crystallographic and compositional information about the pyroxenes that must be carefully deconvolved. In this work, we build on our analyses of synthetic Ca-free orthopyroxenes (Klima et al. 2007) by investigating synthetic clinopyroxenes with compositions spanning most of the Ca-Fe-Mg pyroxene quadrilateral. The presence of Ca 2+ in the sites means that distortions are imposed on the next-nearest neighbor Fe 2+ (and Mg 2+ ) cations in the site. These distortions in turn change the crystal field splitting energies, and thus the absorption band positions, as observed in Ca-bearing clinopyroxenes (e.g., Adams 1974; Hazen et al. 1978; Cloutis and Gaffey 1991). Furthermore, Ca-bearing clinopyroxenes are not restricted to one space group, but rather generally take the form of P2 1 c pyroxene (e.g., pigeonite) at low Ca 2+ content, and C2 c (e.g., augite) at higher Ca 2+ content. We here explore the effects of structure and bulk composition on near-infrared spectra of the synthetic pyroxenes to refine the amount of crystal chemical and compositional information that can be derived from the spectra. BACKGROUND Clinopyroxene Structure and Composition Clinopyroxenes encompass a broad range of compositions within the pyroxene mineral group. We focus strictly on the Mg 2+ -Fe 2+ -Ca 2+ quadrilateral clinopyroxenes, without additional cations such as Al 3+,Cr 3+,Mn 2+,Na +, etc. All clinopyroxenes can be broadly grouped as monoclinic, but depending on the bulk composition, temperature, and pressure of formation, they may belong to either the P2 1 c or C2 c space groups (Fig. 1). Augite ((Ca,Na) (Mg,Fe, Al,Ti) (Si,Al) 2 O 6 ) and the Ca 2+ -saturated endmembers diopside (CaMgSi 2 O 6 ) and hedenbergite (CaFeSi 2 O 6 ) belong to the C2 c space group. Pigeonite ((Mg,Fe 2+,Ca)(Mg,Fe 2+ )Si 2 O 6 ), which ranges in composition from about Wo 5 Wo 15, exhibits C2 c symmetry at high temperatures (high pigeonite), but on cooling generally transforms to the P2 1 c space group (low pigeonite). In this work, we will use the term pigeonite to describe pyroxenes with a P2 1 c structure. The term augite will be used to refer to pyroxenes exhibiting a C2 c structure. Because of the differences

3 Near-infrared spectra of clinopyroxenes 381 Fig. 2. Graphical illustration of the difference in kinking between SiO tetrahedra in the P2 1 c (e.g., pigeonite) and C2 c (e.g., augite) space groups for pyroxenes; the linkages are outlined in thick black line for clarity. The chains are parallel to the z orientation in the pyroxenes. Variations in the angle of kinking, which is more extreme for the pigeonite, are related to distortions in the adjacent and polyhedra that may contain Fe 2+, Mg, or Ca. The c-axis lies along the length of the chains in all the structures. in structure, there is only full solid solution between augites and high pigeonites, and natural pyroxenes are often exsolved or inverted at some scale. All pyroxenes are composed of infinite chains of edge-sharing octahedra in the c direction, flanked by 6- or 8-coordinated sites. These are linked to parallel infinite chains of corner-sharing octahedra. The P2 1 c and C2 c space groups of interest in this study differ largely because of the size of cations in the site, which causes kinking and rotation of the Si-O chains that are linked to them (Fig. 2). In augites, the edges of the tetrahedra that share corners with other tetrahedra line up nearly straight, but in pigeonites, the angles between linked Si tetrahedral are kinked (Cameron and Papike 1980). The alternating chain structure of pigeonites results in site geometries for and that are somewhat similar to those observed in orthopyroxenes (Fig. 1). As crystal field splitting is directly a result of the ligand field surrounding the Fe 2+ cations (or the oxygen arrangement), pigeonites and orthopyroxenes should exhibit spectra that are more similar than augites and orthopyroxenes. Spectroscopy of Clinopyroxenes Spin-allowed crystal field transitions take place when transition metal cations reside in a ligand field (in the case of pyroxenes, this is an octahedral framework of closely packed oxygen atoms). For pyroxenes, the ligands surrounding the and cation sites produce an asymmetric electrostatic field, splitting the energy levels of d-orbital electrons of transition metals within these sites (Fig. 3). An incident photon may be absorbed at certain wavelengths, exciting an electron from a lower to a higher energy state (Burns 1993). If the resultant electronic transition is spin-allowed, it produces a prominent absorption feature such as the familiar 1 and 2 lm Fe 2+ pyroxene bands (Adams 1974; Cloutis and Gaffey 1991), which result from transitions made possible by splitting within the e g and t 2g orbitals around the and Fe cations. The relative intensities of such absorptions increase with the asymmetry of the octahedral site (Burns 1993). Thus, the asymmetric site in pyroxenes (cf. Fig. 2) results in the relatively strong 1 and 2 lm absorptions, while the more symmetric site produces much weaker absorptions near 1 and 1.2 lm. The energy of the crystal field splitting and the resulting band positions are a function of the ligand field, which changes in response to the structure and composition of the pyroxene. Previous studies of 1 and 2 lm crystal field band positions suggest that no distinct transitions are crossed, and that there is a smooth variation in pyroxene spectra across all structures and compositions (Adams 1974; Cloutis and Gaffey 1991). In clinopyroxenes, however, structural changes from the more orthopyroxene-like P2 1 c pigeonite structure to the C2 c augite structure should, in principle, result in changes in the ligand field that are manifested as spectral variations (Fig. 1). Pyroxenes that are near Ca 2+ -saturated exhibit drastically different spectra based on how much Fe 2+ is present in the site, and have been classified into two spectral groups (Adams 1975; Cloutis and Gaffey 1991). Type A pyroxenes exhibit two bands near 1 lm, and no 2 lm band, while type B pyroxenes exhibit spectra that are generally similar to Ca 2+ -undersaturated pyroxenes, with strong 1 and 2 lm bands. The type A pyroxenes, which exhibit only bands assigned to Fe 2+ in the site, are saturated with Ca 2+ or cations other than

4 382 R. L. Klima et al. ground under acetone, and mounted for collection of Mo ssbauer spectra. Compositional and fo 2 Analysis Fig. 3. Idealized 6-coordinated polyhedron around a hypothetical Fe atom, along with the crystal field splitting diagram. When the Fe atom is in octahedral coordination (as in pyroxene), the energies of the Fe 3d orbitals change as a function of proximity to electronic distributions of the surrounding oxygen anions. When the octahedron is not perfectly symmetrical, as shown in the pyroxene examples in Fig. 1, further splitting occurs among the e g and t 2g orbitals. Such splitting of the energy levels makes possible multiple electronic transitions that can give rise to absorption features such as the 1 and 2 lm bands seen in pyroxenes. in the site. In a comprehensive study of terrestrial high-ca 2+ (>Wo 45 ) pyroxenes, Schade et al. (2004) noted that most of the naturally occurring type A pyroxenes contain Mn or Fe 3+ in addition to the quadrilateral cations. Fe 2+ Pyroxene Synthesis METHODS The pyroxenes used in this study were synthesized by Don Lindsley and colleagues between 1972 and 2007 using the procedures detailed in Turnock et al. (1973). All samples were prepared from reagent grade chemicals and the oxygen fugacity was buffered at the iron-wu stite curve. Specific synthesis methods varied by composition, and were chosen to prevent nucleation of pyroxenoids and produce a single, homogeneous pyroxene. Pyroxene compositions were validated by Lindsley and colleagues using X-ray diffraction. The synthesis procedures produce pyroxenes that are fine powders, with individual grains approximately lm in diameter. Many of the powders formed clumps of grains, which were crushed manually and sieved to <45 and >45 lm grain size fractions for further analysis. The <45 lm size fraction was used for Vis NIR and Mo ssbauer spectroscopy, and, when available, the >45 lm size fraction was used to prepare grain mounts for electron microprobe analysis. Between 12 and 41 mg of each pyroxene was mixed with sugar, Pyroxene composition and homogeneity were investigated using the CAMECA SX-100 electron probe microanalyzer (EPMA) at Brown University. Electron backscatter was used to identify samples with unreacted starting oxides or excess mineral phases. No starting oxides were detected in any samples, although one sample contained trace amounts of fayalite. A number of mid-quadrilateral through high-ca 2+ samples contain a small amount of interstitial material that was found to be high in iron and sometimes calcium (Table 1). The bulk composition of this phase is indistinguishable from that of a pyroxene, but based on its texture, it is likely to be a quench glass. Minor (generally 1%) glass was reported for several of the samples previously described by Turnock et al. (1973). Although the glass is negligible in most cases, it has a modest effect on the spectra of most of the Ca 2+ -rich pyroxenes. The weak glass bands can be discerned in the Ca 2+ -rich pyroxenes because they contain less total Fe 2+ and their spectra therefore exhibit generally higher albedos and weaker pyroxene absorption bands than the other samples (no glass was detected for either hedenbergite sample). The compositions of the pyroxenes as measured by EPMA are presented on a pyroxene quadrilateral in Fig. 4 and listed in Table 1. The light gray region on the quadrilateral indicates a region in which pyroxenes were synthesized, but the resulting crystals contained two phases of pyroxene. These samples have been excluded from this work, but are described and modeled in Klima (2008). The corners of the quadrilateral are the pyroxene endmembers diopside (Di), hedenbergite (Hd), Enstatite (En), and Ferrosillite (Fs). Wollastonite (Ca 2 Si 2 O 6, or Wo) would lie at the top of the ternary plot, but compositions with Ca > 2 pfu per Si 2 O 6 are not true pyroxenes, but pyroxenoids. Some of the probed compositions are several percent less FeO-rich than the intended compositions, due to loss of FeO to the capsule or to a vapor during synthesis or to the interstitial glass. The samples that were furthest from their intended compositions, and therefore most likely to have their spectra affected by the glass contamination, are highlighted in gray in Table 1. Note that a number of the pyroxenes fall within the region of the pyroxene quadrilateral often referred to as the forbidden zone as pyroxenes of that composition are metastable with respect to silica, olivine, and augite at ambient atmospheric pressures and temperatures (Lindsley 1983). However, the pyroxenes in that region

5 Near-infrared spectra of clinopyroxenes 383 Table 1. Compositions and MGM band parameters for synthetic clinopyroxenes. Composition MGM parameters Synthesis Probed (1 lm) (2 lm) (1 lm) (1.2 lm) Relab ID Wo En Fs Wo En Fs C W I C W I C W I C W I DL-CMP ) ) )0.36 DL-CMP ) ) )0.67 DL-CMP ) ) )0.61 DL-CMP ) ) )0.41 DL-CMP ) ) )0.66 DL-CMP ) ) )0.86 DL-CMP ) ) )0.50 DL-CMP ) ) )0.53 DL-CMP ) ) )0.48 DL-CMP ) ) )0.18 DL-CMP ) ) )0.65 DL-CMP ) ) )0.30 DL-CMP ) ) )0.54 DL-CMP-067 a ) ) )0.21 DL-CMP ) ) )0.36 DL-CMP-073 a ) ) )0.14 DL-CMP-074 a ) ) )0.53 DL-CMP ) ) )0.28 DL-CMP ) ) )0.75 DL-CMP ) ) )0.37 DL-CMP ) ) )0.25 DL-CMP-076 a ) ) )0.22 DL-CMP ) ) )0.18 DL-CMP-079 a ) ) )0.35 DL-CMP ) ) ) )0.22 DL-CMP ) ) ) )0.40 DL-CMP ) ) ) )0.27 DL-CMP ) ) ) )0.52 DL-CMP ) ) ) )0.37 DL-CMP ) ) ) )0.80 DL-CMP ) ) ) )0.98 Notes: C = band center (nm); W = band full-width at half maximum (nm); I = band intensity (natural log reflectance). a Pyroxenes which differ in intended and actual iron content by more than Fs10. of the quadrilateral were synthesized at up to 22.5 kbar to be well within the pyroxene stability field during the experiments. Mo ssbauer spectra were analyzed to ensure that samples were free of Fe 3+ and to assess the proportion of Fe 2+ in the and sites. Iron-bearing glass, however, cannot be easily differentiated from pyroxene using Mo ssbauer spectra due to similarities in doublet parameters (Dyar 1985). Mo ssbauer spectra were collected at Mount Holyoke College using a WEB Co. model W100 spectrometer with a 45 mci 57 Co in Rh source. Spectra were calibrated against an a-fe foil of 6 lm thickness and 99% purity. Data were modeled using an in-house program from the University of Ghent, in Belgium. The Dist3e program models spectra using quadrupole splitting or hyperfine field distributions for which the subspectra are constituted by Lorentzian-shaped lines; it uses velocity approximations rather than solving the full Hamiltonian. Errors on isomer shift and quadrupole splitting are ±0.02 mm s )1, and errors on peak areas are ±<1% absolute. The spectra and raw data for all of the Mo ssbauer spectra can be found on the Mount Holyoke College Mars Mineral Spectroscopy Database ( Fits to these data will be reported in a work in preparation. None of the pyroxenes was found to contain Fe 3+ or additional iron-bearing crystalline phases (magnetite, wu stite, olivine, etc.). Spectral Methods Visible and near-infrared spectra were collected using the bidirectional reflectance spectrometer at the

6 384 R. L. Klima et al. En Di Synthetic pyroxenes No EPMA data available NASA Keck Reflectance Experiment Lab (RELAB) at Brown University. Spectra of the <45 lm grain size fraction were measured relative to a halon reference standard at 5 nm intervals over the wavelength range of lm and corrected for the small features of halon (Pieters 1983). An incidence angle of 30 and an emission angle of 0 were selected for bidirectional measurements. High resolution (0.5 nm interval) spectra were also collected in the visible between lm. Spin-allowed crystal field bands were modeled using the modified Gaussian model (MGM). The MGM deconvolves a spectrum into a continuum and a series of modified Gaussian curves that can be attributed to specific absorptions (Sunshine et al. 1990, 1999). It also allows individual pyroxene phases to be deconvolved from a composite spectrum (Sunshine and Pieters 1993). The MGM fits spectra using the natural log of reflectance instead of reflectance to better approximate absorbance. For this study, pyroxene spectra were modeled with a straight line continuum plus 2 3 broad bands in the visible to account for the Fe 2+ -O charge transfer band edge, and, in some cases, a band near 3 lm to model out water adsorbed on the sample. In theory, two spinallowed crystal field bands should occur for each of the pyroxene cation sites. However, the higher energy splitting of the and sites occurs at very similar wavelengths which cannot be resolved at these spectral resolutions for most pyroxene compositions. As the MGM provides a mathematical deconvolution of a spectrum, the model always improves when more Gaussian curves are allowed. However, this improvement does not mean that the fit is more accurate. It is important to reach a balance between using the minimum number of curves necessary to fit a spectrum and the theoretical origins of those bands. If Fig. 4. Pyroxene quadrilateral showing the compositions of pyroxenes used in this study. Note that the pyroxene 088 could not be microprobed and the plotted position is the synthesis composition. Pyroxenes synthesized within the light gray region were found to contain more than one crystalline composition and are thus excluded from this study. Orthopyroxenes along the En Fs tie line are described in detail in Klima et al. (2007). Hd Fs bands are going to be fit without restricting the model, it is particularly important to use the minimum number of absorptions needed to describe a spectrum. We have thus chosen to use one band in the 1 lm region to represent a combination + absorption, unless two are absolutely necessary (as is the case for the Wo pyroxenes). Additional spinallowed crystal field bands were added near 1.2 and 2 lm. In Ca 2+ -saturated pyroxenes containing high-iron glass, an additional broad absorption feature was needed near 1.8 lm to allow a proper fit of the 2 lm band. This wavelength is consistent with the longer wavelength absorption band of iron-bearing silicate glasses (Bell et al. 1976). An additional broad glass band should occur just beyond 1 lm, but because of extensive overlap with the 1.2 lm band, the model solutions did not clearly indicate that an additional band was necessary. We have omitted this band for the reasons discussed above. Initial analyses were conducted using only the pyroxenes that were close to the intended composition, as they are the least likely to be affected by the accessory glass. The pyroxenes with a larger proportion of glass (Table 1) were found to follow the trends observed for the more completely reacted samples, and have therefore been included in this study. RESULTS Near-Infrared Spectra ( lm at 0.5 lm Sampling) Quadrilateral Transects Constant Mg Number To highlight the variability in pyroxene spectra as a function of Ca 2+ and structure, four transects of roughly constant Fe 2+ -Mg 2+ ratio but varying Ca 2+ are shown in Fig. 5. The spectra of synthetic Ca-Free orthopyroxenes from Klima et al. (2007) that lie along each of these transects have also been included in each panel for comparison. The spectra in these figures have been offset by multiples of 0.1 to allow easy comparison of features. These transects represent pyroxenes with roughly constant Mg number (Mg 0, Mg 25, Mg 50, and Mg 80 ) with Mg number (Mg#) defined as n Mg (n Fe 2+ + n Mg ) 100. Shown in Fig. 5A are the spectra of pyroxenes in the Mg-free transect, ranging from hedenbergite through ferrosillite. The spectrum of pyroxene 083, a type A hedenbergite, is dominated by a composite absorption comprised of two bands near 1 and 1.2 lm. Progressing down the Mg-free limb of the pyroxene quadrilateral from pyroxene 082 through 087, the 1 and 2 lm bands strengthen, but their positions do not change much, suggesting that the geometry of the coordination polyhedra remains relatively constant.

7 Near-infrared spectra of clinopyroxenes 385 Fig. 5. Pyroxene spectra of transects cutting vertically through the pyroxene quadrilateral. Pyroxenes in each panel are of approximately constant Mg number, and grade from Ca-free orthopyroxene (dark blue spectra) through high Ca (orange and red spectra). A) Mg 0.B)Mg 25.C)Mg 50.D)Mg 80. This is in fact what is observed in single-crystal X-ray refinements (SREF) of the atom positions (Fig. 6). The site is completely filled with Fe 2+ throughout this transect because Ca 2+ cannot enter the site, so visual differences in the band strength of the 1.2 lm band are most likely due to albedo differences among the samples. For example, all the absorption bands in pyroxene 088 are weaker than in pyroxene 087. The absolute strengths of all absorption bands are depressed with very high Fe contents, although the relative strengths remain roughly constant until any band becomes optically saturated (this occurs more at lower Fe contents for the bands because of the distortion of the site). The 1 lm band occurs at slightly shorter wavelengths as the amount of Ca 2+ decreases along the full transect. The position of the 2 lm band in the spectra of pyroxenes 082 through 087 does not change significantly, despite the variability in Ca 2+. In contrast, the 2 lm bands of pyroxenes 088, reported to be a P2 1 c structure (Dowty and Lindsley 1973), and 061, an orthopyroxene, occur at shorter wavelengths as expected for their lower-ca 2+ contents because the shape of is different in the P2 1 c and Pbca space group pyroxenes. This change is also clearly seen in the SREF data in Fig. 6. The Mg 25 transect, shown in Fig. 5B, offers the most complete cut of single-phase pyroxenes through the quadrilateral. Starting at the most Ca 2+ -rich sample and moving down the transect, the strengths of the 1 and 2 lm absorption bands initially increase sharply until pyroxene 055 (Wo 26 ) is reached, after which the strengths of the 2 lm bands are approximately equal. As Mg 2+ is also present in this transect, the strengths of the 1.2 lm bands are dependent not only on the bulk composition of the pyroxenes but also on the partitioning of Fe 2+ and Mg 2+ between the and sites. The intensity of the 1.2 lm band is strongest for pyroxenes 066 and 070, in which a majority of sites are filled with Ca 2+, and decreases gradually from Wo 26 through Wo 0.

8 386 R. L. Klima et al. Fs 65 Wo 35 C2/c Fs 75 Wo 25 C2/c Fs 80 Wo 20 C2/c Fs Wo P2 /c As in the Mg 2+ -free system, the positions of the 1 lm bands move to shorter wavelengths with decreasing Ca 2+ content. However, the 2 lm band again occurs near 2.3 lm for all samples of Wo 20 or greater. The shifting of the 2 lm band is not visually obvious until the transition into low-ca clinopyroxenes and orthopyroxenes. The 2 lm band center for pyroxene 053 occurs at a similar wavelength to the P2 1 c pyroxene 088. Based on its composition, it is reasonable to conclude that pyroxene 053 may also exhibit a P2 1 c crystal structure. Shown in Fig. 5C is the Mg 50 transect. At Mg 50, the first evidence of the midquadrilateral miscibility gap becomes apparent. Although these pyroxenes were synthesized at temperatures and pressures designed to produce a single-phase pyroxene, it is extremely difficult to quench this metastable phase fully. Pyroxene 009, shown in light blue in Fig. 5, exhibits an abnormally broad 2 lm band, generally indicative of zoning or finescale exsolution. No exsolution or partial inversion to orthopyroxene was evident at the scale of the electron microprobe, although the sample was found to be compositionally zoned. A number of clinopyroxenes Fs Pbca Fig. 6. To-scale renderings of (left) and (right) polyhedra from single-crystal refinements of synthetic pyroxenes ranging down the Hd-Fs join. The Fs 100 data are from Sueno et al. (1976) and the other data are from Ohashi et al. (1975). These compare with the data shown in Fig. 5A over the same edge of the quadrilateral. were synthesized along the Mg 80 transect; however, all those with <Wo 25 were found to contain patches of high- and low-ca pyroxene when examined by EMPA. Thus, the Mg 80 transect, shown in Fig. 5D, is limited to only higher Ca pyroxenes. As observed in the more iron-rich pyroxenes, the positions of the 1 lm bands in the Mg 50 and Mg 80 transect move to shorter wavelengths with decreasing Ca 2+ content, and the 2 lm bands of all pyroxenes above Wo 20 occur near 2.3 lm. Quadrilateral Horizontal Transects Constant Ca We examined five transects of roughly constant Ca 2+ content to determine whether the effect of Fe 2+ can be distinguished from the effects of Ca 2+. Four of the transects are within the augite structural region of the quadrilateral, and one is within the pigeonite structural zone. Shown in Fig. 7A are near-infrared spectra of the pyroxenes with compositions near Wo 50. Despite these pyroxenes all being nominally Ca 2+ - saturated, only one sample (083) is a type A pyroxene. The remainder of the samples would be considered transitional between types A and B (or type AB). Though the small amount of Fe 2+ in the sites of these pyroxenes was essentially invisible to Mossbauer spectroscopy, it still has a pronounced effect on the reflectance spectra. In stark contrast to the spectra of lower-ca clinopyroxenes and orthopyroxenes, the absorption feature near 1 lm is clearly dominated by the absorption bands. In particular, the component of the 1 lm band, normally nested within and dwarfed by the 1 lm band, is clearly present in all spectra. The double bands are most clearly visible in the type A pyroxene 083. Though the spectrum suggests two bands of approximately equal intensity, MGM models consistently result in a stronger 1.2 lm band and a weaker 1 lm band (Klima et al. 2008). A similar result was found when Schade et al. (2004) performed MGM fits on their suite of natural high-ca pyroxenes. The bands are weak in comparison with the bands, yet relatively strong considering the very small amount of Fe 2+ present in the site. For example, in pyroxene 083, a slight band is present near 2 lm, and MGM models require a weak band in the 1 lm region as well (Klima et al. 2008). EPMA and Mo ssbauer measurements of pyroxene 083 predict that the site is completely saturated by Ca 2+ atoms, and should thus have no crystal field bands. Nevertheless, the presence of a band near 2.3 lm suggests that a trace of Fe must be present in at least some sites. This observation underscores the effect of the distortion of the site from regular octahedral symmetry, because the intensity of crystal field

9 Near-infrared spectra of clinopyroxenes 387 Fig. 7. Spectra of pyroxenes along horizontal transects through the pyroxene quadrilateral. Each panel contains pyroxenes of approximately constant Ca content. The lowest Fe pyroxenes are colored in red and the highest Fe pyroxenes are colored in blue. A) Ca-saturated pyroxenes (Wo 50 ). B) Augites with around Wo 40. C) Augites with around Wo 40. D) Augites with around Wo 30. E) Pigeonites with near Wo 10. The pyroxenes that were found to contain the highest proportion of interstitial glass (see Table 1) are shown as dashed lines. absorption bands is a function of both the iron content in a site and the distortion of the site. C2 c (augite) structure pyroxenes with <50% Ca exhibit more typical pyroxene spectral shapes. When Ca content is held approximately constant, the difference in 1 and 2 lm band position between pyroxenes of differing Fe content is minimal. As shown in Figs. 7B D, the shift of the 1 lm band in augites is only on the order of nm for any of the transects. The 2 lm band is also relatively unaffected by changes in Fe 2+ content, with the maximum difference between positions being roughly 40 nm. These observations are consistent with the long-held notion that there are few structural changes across the diopside-hendebergite solid solution (Rutstein and Yund 1969); note the similarities of the and polyhedra between diopside and hedenbergite in Fig. 1. In contrast, the positions of the 1 and 2 lm bands do shift significantly as a function of Fe 2+ within the pigeonite zone, as illustrated in Fig. 7E. DISCUSSION Spin-Allowed Crystal Field Bands in the Near-Infrared Absorption band parameters derived from MGM fits for each of the pyroxenes are listed in Table 1. Examples of the general fit types are shown in Fig. 8. In each panel, the bands attributed to each cation site are indicated. The measured spectrum is shown as a thick gray line and the modeled fit is shown as a thin black line. The residual as a function of wavelength is shown at the top of each panel. Shown in Fig. 8A is a typical low-ca clinopyroxene that falls within the pigeonite structural field. Only three major absorption bands were needed to model this spectrum, so the band near 1 lm

10 388 R. L. Klima et al. 0.5 Residual 0.5 Residual & & Natural Log Reflectance Wo En Fs A Wavelength (µm) Residual Wo En Fs B Wavelength (µm) Residual C Wo Fs (Type A/B) D Wo 50 Fs 50 (Type A) Wavelength (µm) Wavelength (µm) Fig. 8. Example MGM fits for the four basic types of pyroxenes in this study. The data are shown as a thick gray line, and the modeled fits are a thin black line superimposed. The residual error as a function of wavelength is shown at the top of each panel. A) Fit to a typical pigeonite sample. B) Fit to a subcalcic augite. C) Fit to a type AB hedenbergite. D) A type A hedenbergite. represents a combination of the strong and weaker absorptions. Augites with <Wo 45 were modeled similarly, as shown in Fig. 8B. In both cases, there is a residual error near 1 lm, likely due to the missing band. The symmetrical nature of the residual error indicates that the missing band is distorting the shape of the modified Gaussian curve because the two nested absorptions result in the composite band being saturated. Fits were attempted using an additional band near 1 lm, but the resulting deconvolutions were not physically realistic (i.e., bands became positive) or were directly dependent on the input parameters (i.e., the relative band intensities did not readjust as the model ran, but remained close to the values used as starting parameters). For pyroxenes with close to Wo 50, the additional band was critical to obtain a modeled fit. Examples are shown in Fig. 8 for a type AB (Fig. 8C) and type A (Fig. 8D) pyroxene. Spectral Variability as a Function of Ca Content Shown in Fig. 9 are a series of scatter plots of the MGM-derived 1 and 2 lm band positions along each of the pyroxene transects of constant Mg-Fe ratio. The 1 lm band moves regularly to longer wavelengths with increasing Ca 2+ content regardless of crystal structure. These trends are well described by linear fits (R 2 -values over 0.96). From Wo 0 through roughly Wo <15, the 2 lm band moves substantially to longer wavelengths with increasing Ca 2+ content. However, beyond Wo 20, the 2 lm band reaches roughly 2.3 lm and then remains near that value from roughly Wo 30 through Wo 50. This plateau in 2 lm band position occurs from near Wo through Wo 50 in all transects, regardless of iron-magnesium ratio. The decoupling of the 1 and 2 lm band position trends, which, according to crystal field theory should both be primarily or completely due to transitions of Fe 2+ in the site (Burns 1993), is unexpected, and will be discussed in more detail later. As the crystal field absorption band positions depend on the crystal field splitting, which in turn depends on the ligand field environment and crystal structure, the plateau in the position of the 2.3 lm band suggests that there is indeed a spectral transition

11 Near-infrared spectra of clinopyroxenes 389 Fig. 9. Variation of 1 and 2 lm band positions as a function of Ca content (vertical quadrilateral transects holding Mg number approximately constant). Orthopyroxenes are shown as open diamonds, pigeonites are shown as black diamonds, and augites are shown as gray diamonds. A trend line and R 2 -value is shown for the relationship of the 1 lm band position to the Ca content. The 2 lm band position does not vary linearly with Ca content. corresponding to the structural change from P2 1 c pigeonite to C2 c augite (Fig. 1). The almost-stationary position of the 2 band in C2 c clinopyroxenes implies that the ligand field around the sites does not change significantly even as more Ca 2+ is added to an augite (Fig. 6). Spectral Variability as a Function of Iron Content Shown in Fig. 10 is a scatter plot of the positions of the MGM-derived 1 and 2 lm bands as a function of Fe 2+ for the transects of approximately constant Ca content. For the highest Ca pyroxenes, there is no correlation between Fe 2+ content and the position of the 2 lm band, probably because Ca 2+ is controlling the distortion of the polyhedron and it is not changing across this transect. The 1 lm band moves to slightly longer wavelengths with increasing Fe 2+, but there is a large amount of scatter in band position. The general trends become slightly tighter for augites with lower-ca content. The relationship of Fe 2+ to the positions of the 1 and 2 lm bands become better described by linear fits as the amount of Ca decreases. The 1 lm band moves to slightly longer wavelengths, while the 2 lm band moves to slightly shorter wavelengths with the addition of more Fe 2+. For all augites, the difference in 1 lm band positions is roughly nm maximum, and that of the 2 lm band is around nm. Given the poor correlation and the small shift in position, it is doubtful that the Fs content of augites (especially higher Ca augites) could be reliably assessed from their absorption band positions. In contrast, the 1 and 2 lm bands of pigeonites trend consistently toward longer wavelengths as a function of Fe 2+ content. There is over 35 nm difference between the 1 lm band positions of a pigeonite of roughly Fs 40 and one of around Fs 90,and roughly 200 nm between the 2 lm band positions of the same pigeonites. This spectral behavior is more similar

12 390 R. L. Klima et al. Fig. 10. Variation of 1 and 2 lm band positions as a function of iron content (horizontal quadrilateral transects holding Ca approximately constant). Pigeonites are shown as black diamonds, and augites are shown as gray diamonds. A linear trend line is shown only for those relationships with an R 2 -value > 0.8. to that of orthopyroxenes than of other clinopyroxenes, probably due to the low total Ca in the structure. Overall Spectral Trends Shown in Fig. 11 are scatter plots of the 1 and 2 lm band positions as a function of Fe 2+ and Ca 2+ content for the full suite of clinopyroxenes. If all are considered together, irrespective of structure, the relationship between band position and Ca 2+ content is clearly much stronger than that with Fe 2+. All 1 lm bands move to longer wavelengths with increasing Ca 2+, and, as observed in the Mg transects, the 2 lm bands increase through roughly Wo 20 and then reach a plateau around 2.3 lm. When considered as a whole, the 1 and 2 lm bands appear to move to shorter wavelengths with increasing Fe 2+ content. However, this effect is overprinted (and obscured) by the variations caused by Ca 2+. Pyroxenes with Fs contents above 50% necessarily contain less Ca 2+ than those with lower Fs contents. When structural group is considered, the trends become quite different. Pigeonites have been colored black in Fig. 11, and augites are shown as gray. Although all 1 lm bands move to longer wavelengths with Ca 2+ content, the scatter in the pigeonites is greater, as illustrated in Fig. 11A. This is additional evidence that the Fe 2+ content has a stronger effect on pigeonites than augites. Indeed, in Fig. 11C, the pigeonites define a line that increases very regularly with Fe 2+ content. The two structural groups are even better separated when the 2 lm band is considered (Figs. 2B and 2D). Pigeonites move regularly to longer wavelengths with both increasing Fe 2+ and Ca 2+ content, followed by a gap from lm where no

13 Near-infrared spectra of clinopyroxenes µm band position 1 µm band position A C Wo Fs 2 µm band position 2 µm band position B Wo Fig. 11. Variations in the band positions of all clinopyroxenes as a function of Ca and Fe content. Pigeonites are shown as black diamonds, and augites are shown as gray diamonds. The region in 2 lm band position-space that corresponds to the structural transition from pigeonite to augite is shaded in gray. A) 1 lm band position as a function of Ca content. B) 2 lm band position as a function of Ca content. C) 1 lm band position as a function of Fe content. D) 2 lm band position as a function of Fe content. D Fs C2/c P2 1 /c C2/c P2 1 /c 2 lm bands are centered. There is no clear correlation between the 2 lm band positions for augites, which all plot above this gap, and Fe 2+ and Ca 2+ content. The relative changes in the 1 and 2 lm band positions can be clearly seen in a plot of 1 and 2 lm band centers, as used by Adams (1974) and Cloutis and Gaffey (1991). As illustrated in Fig. 12, the variation in 1 and 2 lm bands for orthopyroxenes and pigeonites follows a very regular trend. However, the augite data cluster in a cloud rather than along any type of regular line. If the opx-pigeonite trend is extended through the augite field, it defines a lower limit to the augite cluster. In previous studies, band 1 band 2 plots have been constructed using absolute band minima or band centers, without spectral deconvolution and usually without information on the crystal structure or space group (Adams 1974; Hazen et al. 1978; Cloutis and Gaffey 1991). For example, Hazen et al. (1978) show smoothly contoured quadrilaterals for the position of the 1 and 2 lm pyroxene absorption maxima versus composition. However, because of the mid-quadrilateral miscibility gap between pigeonites and augites, a large number of natural pyroxenes used in previous studies are extensively zoned or exsolved. If only single 1 and 2 lm band positions are measured for multiphase pyroxenes, it represents a combination of all phases and results in an apparently continuous trend in pyroxene band 1 band 2 positions that varies as a function of Fe 2+ Band 1: 1 µm band wavelength Natural OPX (Cloutis and Adams) Natural CPX (Cloutis and Adams) Synthetic Ca-free OPX Synthetic CPX (Pigeonites) Synthetic CPX (Augites) Band 2: 2 µm band wavelength Fig. 12. Band 1 and Band 2 diagram of synthetic pyroxenes compared with natural pyroxenes measured by Adams (1974) and Cloutis and Gaffey (1991). For orthopyroxenes and pigeonites, the synthetic pyroxenes define a much tighter trend in band 1 band 2 space than do the natural data. The augites maintain this relationship initially (at Wo 30 ), but begin to fan out at higher Ca content. (for low Ca 2+ -pyroxenes) and then Ca 2+ (Adams and McCord 1972; Adams 1974; Cloutis and Gaffey 1991). However, our synthetic samples with a better coverage of the Fe-rich compositions and a broader range of representation of space groups allow the finer details of changes with crystal structure to be discerned.

14 392 R. L. Klima et al. 1 µm band En 2 µm band En D i D i As seen in Fig. 12, when only single-phase pyroxenes are considered a band 1 band 2 plot separates pyroxenes into their structural groups (orthopyroxene, pigeonite, augite), suggesting that the previous studies were overly simplistic. Previous authors have used contoured quadrilaterals as a way to display the variability of 1 and 2 lm bands as a function of composition (Hazen et al. 1978; Cloutis and Gaffey 1991; Denevi et al. 2007). Shown in Fig. 13 are pyroxene quadrilaterals that have been contoured with the positions of the 1 and 2 lm bands. For reasons of the spectral transitions induced by structural changes, each structural group must be treated separately. As we do not have singlephase pyroxenes within the light gray region, we have avoided extrapolating our pigeonite trends to lower Mg contents. The slopes of the contours in the pigeonite fields are also somewhat qualitative, as most of our pigeonite data cluster around Wo 10. The 1 lm band is contoured using 10 nm intervals; however, the 95% confidence range is generally on the order of 10 nm for each of the modeled fits. No clear compositional trends could be contoured for the 2 lm band in augites, as the entire wavelength spread for augites is only about 40 nm. The quadrilaterals in Fig. 13 are qualitatively similar to those proposed by Cloutis and Gaffey (1991) and Denevi et al. (2007) for natural pyroxenes, if only Hd Hd Fig. 13. Pyroxene quadrilaterals showing the variation trends for the 1 and 2 lm bands. Note that the slopes within the pigeonite field (medium gray) are approximate and based on samples with a restricted range of Ca content (Wo 10 ). The augite field for the 2 lm band could not be explicitly contoured, as all samples exhibit 2 lm bands clustering around 2.3 lm. Fs Fs the regions where previous authors had data (or singlephase pyroxene data) are considered. If the natural pyroxenes measured by these authors were plotted on the diagrams constructed from the synthetic pyroxene data, they would also be within the error ranges of the contours. Unique to this data set are the large number of Fe 2+ -rich samples. In previous studies, contours had to be extrapolated through these compositions, so the gap in band parameter space between the pigeonites and augites was not clearly defined. A question that arises is: why is the 2 lm band seemingly insensitive to variations in Ca content for augites, yet the 1 lm band continues to shift regularly to longer wavelengths with increasing Ca 2+ content? The answer must lie in steric constraints on the crystal structures. The many single-crystal refinements for pyroxenes in the Ca-Mg-Fe system suggest some possible explanations. Consider again the Ca-Fe series from Fs 65 Wo 35 to Fs 85 Wo 15 on the right side of the quadrilateral, which was studied by Ohashi et al. (1975). The polyhedral volume and bond distances of the site do not change appreciably along this join except for a slight distortion of the octahedron at the Fs-rich end. So it is unlikely that changes in the octahedron are causing the wavelength shift of the 1 lm band. The site is solely responsible for the 2 lm absorption, and dominates the 1 lm absorption for all but the most Ca-saturated pyroxenes. Crystal field splitting is sensitive to the interactions between the 3d electronic states in the Fe cation and the surrounding oxygens. The two e g orbitals both have the majority of their electronic probability distribution along the x, y, and z orientations; repulsion between the oxygen anions and the electrons in the orbitals raises their energy. In a pyroxene site, distortion causes further splitting of the two e g orbitals: 3d x2 y2 has most of its electron probability in the plane of the x- and y-axes, while the 3d z2 orbital has most of its distribution along the z-axes. Each site is surrounded by four nonbridging oxygens (O1 and O2) that coordinate to the Fe atom only, and two to four bridging oxygens (O3) that bond to Si 4+ tetrahedra (Fig. 14). It is known from singlecrystal refinements that the Fe-O1 and Fe-O2 distances decrease as the smaller Fe 2+ substitutes for the larger Ca 2+ cation. At the same time, the Fe-O3 bond length increases (but only slightly) with Fe 2+ Ca 2+. It seems possible that the 1 lm band might be more affected by the Fe-O1 and Fe-O2 bond length changes, while the 2 lm absorption, resulting from a transition to an orbital with a fundamentally different electronic distribution, responds more to changes in the Fe-O3 bond distance. In fact, as shown in Fig. 14, the z direction is closest to the Fe-O2 bond, so it might be

15 Near-infrared spectra of clinopyroxenes 393 Fig. 14. coordination polyhedron from Fs 80 Wo 20 from Ohashi et al. (1975). more likely to have its 3d z2 orbital energy increased by such proximity. The x direction is not very close to any of the bonds, and the y orientation splits the angle between the O1 oxygens, so these would be expected to have less proximity to oxygen anions, and thus have lower energies, consistent with the energy of the 2 lm band (5000 cm )1 ). IMPLICATIONS FOR REMOTE SENSING One of the primary objectives in studying pyroxene spectra in-depth is to be able to determine what can and cannot be concluded when a pyroxene-rich spectrum is measured on a planet or asteroid. Currently, most global mapping of planetary surfaces only differentiates between the broad groupings of high and low Ca 2+ pyroxenes (Pieters 1993; Mustard and Sunshine 1995; Bibring et al. 2005; Mustard et al. 2005). Some asteroid studies, however, use a series of empirically derived equations to estimate specific pyroxene mineralogies for asteroids (Gaffey et al. 2002; Hardersen et al. 2004, 2006). Although previous studies of natural pyroxenes have led to the suggestion that the entire pyroxene mineral series exhibits a smooth trend of 1 and 2 lm band positions (Adams and McCord 1972; Adams 1974; Cloutis and Gaffey 1991), our investigation of pure, single-phase pyroxenes indicates that this is not the case. Mineral structure is one of the most important factors affecting the band parameters of pyroxenes. Structures in orthopyroxenes and pigeonites can be easily distinguished from augites based on their 1 and 2 lm band positions. However, to evaluate structure from near-infrared spectra, it is critical to ensure that the 1 and 2 lm bands belong to a single pyroxene, and do not represent an average of a mixture. This typically requires careful deconvolution of pyroxenes using a method such as the MGM. There is a strong desire in the planetary spectroscopy community to be able to measure pyroxene band parameters and easily relate them to a precise iron, magnesium, and calcium content. Gaffey et al. (2002) developed a set of seven equations for determining composition, as a single equation was not suitable for handling the entire pyroxene quadrilateral. Each of the equations has a stated Wo or Fs content for which it is valid, and they are applied iteratively until the equations converge on a consistent answer. If we treat the spectra of the synthetic pyroxenes as unknowns, applying their band centers to the equations in Gaffey et al. (2002), only the compositions of Ca-free orthopyroxenes with <Fs 30 are reliably predicted to be within the given error ranges for Fs and Wo content. As a whole, 20% of the equations converge to within ±Wo 5 of the true composition, and 13% converge to within ±Fs 5 of the actual composition. Part of this inaccuracy is because 12 of the synthetic pyroxenes (20% of the total) are outside of the chemical range for which equations are available (the Fe is too high). Even if only the pyroxenes within the designated ranges are considered, the success rate for predicting composition within ±Fs 5 is only 20% and is about 30% for ±Wo 5. These results suggest that simple parameterizations are inadequate to characterize the precise Mg-Fe-Ca composition of the pyroxene system. Based on our pyroxene analyses, compositional remote analysis cannot ignore structural constraints. We propose that when a pyroxene spectrum is observed, it should be analyzed using a decision-tree that reflects the constraints from this larger suite of well-characterized synthetic pyroxenes. If the 2 lm band occurs between 2.25 and 2.3 lm, the pyroxene is dominated by augite, and the 2 lm band is no longer useful for assessing the specific composition, but the 1 lm band can be used to determine the Ca content. If the 2 lm band occurs between 2.20 and 2.25 lm, it is likely that the target represents a mixture of pyroxene compositions, either intimately or mechanically mixed. If the 2 lm band occurs shortwards of 2.2 lm, the pyroxene is dominated by pigeonite and orthopyroxene and both positions of the 1 and 2 lm bands can be used to estimate the composition. Ultimately, a system incorporating additional constraints such as band width (which helps to determine whether the target is dominated by one or multiple pyroxenes) or relative strengths of the and bands (which addresses total iron content and cooling history) would provide much more robust compositional information about a planetary surface. CONCLUSIONS For clinopyroxenes, it had been previously recognized that the addition of the large Ca 2+ cation to the structure causes the 1 and 2 lm bands to shift to longer wavelengths. However, all other things being