K2O-rich trapped melt in olivine in the Nakhla meteorite: Implications for petrogenesis of nakhlites and evolution of the Martian mantle

Authors


Abstract

We used new analytical and theoretical methods to determine the major and minor element compositions of the primary trapped liquid (PTLs) represented by melt inclusions in olivine and augite in the Martian clinopyroxenite, Nakhla, for comparison with previously proposed compositions for the Nakhla (or nakhlite) parent magma. We particularly focused on obtaining accurate K2O contents, and on testing whether high K2O contents and K2O/Na2O ratios obtained in previous studies of melt inclusions in olivine in Nakhla could have been due to unrepresentative sampling, systematic errors arising from electron microprobe techniques, late alteration of the inclusions, and/or boundary layer effects. Based on analyses of 35 melt inclusions in olivine cores, the PTL in olivine, PTLoliv, contained (by wt) approximately 47% SiO2, 6.3% Al2O3, 9.6% CaO, 1.8% K2O, and 0.9% Na2O, with K2O/Na2O = 2.0. We infer that the high K2O content of PTLoliv is not due to boundary layer effects and represents a real property of the melt from which the host olivine crystallized. This melt was cosaturated with olivine and augite. Its mg# is model-dependent and is constrained only to be ≥19 (equilibrium Fo = 40). Based on analyses of 91 melt inclusions in augite cores, the PTL in augite, PTLaug, contained (by wt) 53–54% SiO2, 7–8% Al2O3, 0.8–1.1% K2O, and 1.1–1.4% Na2O, with K2O/Na2O = 0.7–0.8. This K2O content and K2O/Na2O ratio are significantly higher than inferred in studies of melt inclusions in augite in Nakhla by experimental rehomogenization. PTLaug was saturated only with augite, and in equilibrium with augite cores of mg# 62. PTLaug represents the Nakhla parent magma, and does not evolve to PTLoliv by fractional crystallization. We therefore conclude that olivine cores in Nakhla (and, by extension, other nakhlites) are xenocrystic. We propose that PTLoliv and PTLaug were generated from the same source region. PTLoliv was generated first and emplaced to form olivine-rich cumulate rocks. Shortly thereafter, PTLaug was generated and ascended through these olivine-rich cumulates, incorporating fragments of wallrock that became the xenocrystic olivine cores in Nakhla. The Nakhla (nakhlite) mantle source region was pyroxenitic with some olivine, and could have become enriched in K relative to Na via metasomatism. A high degree of melting of this source produced the silica-poor, alkali-rich magma PTLoliv. Further ascension and decompression of the source led to generation of the silica-rich, relatively alkali-poor magma PTLaug. Potassium-rich magmas like those involved in the formation of the nakhlites represent an important part of the diversity of Martian igneous rocks.

Introduction

The nakhlites constitute a group of eight closely related augite-rich meteorites, widely interpreted to be igneous cumulates that crystallized from basaltic magma on Mars (seven samples reviewed by Treiman 2005; an eighth described by Jambon et al. 2010). They consist of euhedral to subhedral crystals of augite (approximately 51–81 vol%) and olivine (approximately 3–16 vol%), embedded in approximately 8–40 vol% fine-grained mesostasis. Both augite and olivine grains have homogeneous cores (Wo 38–40, mg# 60–66, and Fo 35–45), with zoned, more FeO-rich rims. The nakhlites have similar bulk chemical compositions, and the variations in modal abundances, mineral compositions, and textures among them (S1) are consistent with variation mainly in cooling rate (Mikouchi and Miyamoto 2002; Mikouchi et al. 2003, 2006, 2012). Based on these variations, it has been suggested that the nakhlites solidified over a range of depths in a single shallow intrusion or a series of lava flows that were emplaced nearly simultaneously (Mikouchi and Miyamoto 2002; Mikouchi et al. 2003, 2006, 2012; Bridges and Warren 2006; Hammer 2009). A single parent magma, or a series of closely related parent magmas, for all nakhlites is also consistent with their common crystallization age of approximately 1.3 Gyr (Nyquist et al. 2001) and inferred highly depleted mantle source region (Harper et al. 1995; Borg et al. 2003; Jones 2003). In addition, emplacement in a single lithologic unit on Mars is consistent with their similar cosmic ray exposure ages of approximately 10–11 Ma, suggesting that they were ejected from Mars by a single impact event (Eugster et al. 2002, 2006; Huber et al. 2012).

However, the chemical composition of the nakhlite parent magma(s) has proven elusive. Because nakhlites are cumulate rocks, their parent magma compositions are not preserved in their bulk rock compositions and must be determined by indirect methods. Many nakhlite parent magma (NPM) compositions have been proposed (S2), the majority of which are based solely on Nakhla. These compositions were obtained using a variety of methods, including (1) mass balance, (2) mineral/melt partitioning, (3) crystallization experiments, (4) melt inclusions in olivine, and (4) melt inclusions in augite. In S2, we describe each of these methods, discuss its advantages and disadvantages, and evaluate the NPM compositions that have been based on it.

Although most of the proposed NPMs (S2) are similar in having high CaO contents and high Fe/Mg ratio compared with common terrestrial basalts, they differ significantly in SiO2 and Al2O3 contents, and even more so in alkali element abundances. Their K2O contents range from 0.3 to 2.8 wt%, Na2O contents from 0.8 to 3.2 wt%, and K2O/Na2O ratios from 0.2 to 2.3. Notably, all proposed NPMs that have been derived from melt inclusions in olivine have much higher K2O contents and K2O/Na2O ratios than those derived by any of the other methods (S2). High K2O content in the NPM would have significant implications for the depth and composition of the magma's source region (Jones 1989; Longhi and Pan 1989; Righter et al. 2008), and thus for the structure and evolution of the Martian mantle. The importance of a possibly K2O-rich NPM has been underscored by the discovery of K2O-rich rocks on the Martian surface (Schmidt et al. 2013; Stolper et al. 2013).

Given the uncertainty in the composition of the NPM in general, and the unresolved puzzle of its K2O content in particular, we undertook a new study to determine the composition of the primary trapped liquid (PTL) represented by melt inclusions in olivine in Nakhla. This work is independent of previous efforts (Harvey and McSween 1992; Treiman 1993), and utilizes new analytical and theoretical methods in the study of melt inclusions (see S2). In particular, we focus on obtaining accurate K2O contents, and on testing whether the high K2O contents obtained in previous studies (e.g., Treiman 1993) could have been due to unrepresentative sampling, systematic errors arising from electron microprobe techniques, late alteration of the inclusions, and/or boundary layer effects. We then use similar methods to determine the composition of the PTL represented by melt inclusions in augite in Nakhla (this is the first study to do so using analytical methods, rather than experimental rehomogenization). Based on the two trapped melt compositions obtained from this work, we propose a new scenario for the formation of Nakhla and, by extension, all nakhlites. Finally, we discuss the implications of our results for the composition and evolution of the nakhlite mantle source region.

Samples and Analytical Methods

We obtained 9 polished thin sections of Nakhla, in which we observed 37 polymineralic inclusions in 20 host olivine crystals (Table 1; S4). All inclusions were characterized petrographically with backscattered electron imaging (BEI) and energy-dispersive spectral (EDS) analysis, using the Zeiss EVO50 XVP scanning electron microscope (SEM) in the Department of Geosciences at the University of Massachusetts, Amherst (U. Mass). Compositions of individual phases in the inclusions and bulk compositions by grid analysis (described below) were obtained by electron microprobe analysis (EMPA), using the following instruments: Cameca SX-Ultrachron and Cameca SX-50 (U. Mass), Cameca SX-100 and Cameca SX-50 (Lunar and Planetary Laboratory, University of Arizona), Cameca SX-100 (American Museum of Natural History), and Cameca SX-100 (Johnson Space Center, Houston).

Table 1. Thin sections of Nakhla studied in this work.
Section no.SourceaNo. of polymineralic inclusions in olivineNo. of host grains
  1. a

    USNM = U.S. National Museum (Smithsonian); IOM = Institute of Meteoritics; FMNH = Field Museum of Natural History; AMNH = American Museum of Natural History.

NK4USNM22
NK11USNM21
NK12USNM112
NK13USNM11
NK19USNM83
NK20USNM11
NK1137IOM44
NK3887-1AMNH53
NKFMNHFMNH33
Totals 3720

Following the convention of Treiman (1993), melt inclusions are designated by section number or name, crystal letter, and melt inclusion number: e.g., NK12A1 refers to the first numbered inclusion (usually the largest) in olivine crystal A in section NK12. Inclusions that were previously studied by Treiman (1993) have the same designation as in that work.

X-Ray Mapping of Host Olivine Crystals

Prior to any EMPA, we obtained elemental X-ray maps of two olivine crystals (NK19A and NK12A) that contained melt inclusions. Maps were collected on the Cameca SX-Ultrachron, using an optimized CeB6 cathode at 15 kV, 600 nA (regulated), with 100 ms pixel dwell time and 2 μm step size (600 × 600 pixels). Elements, acquired lines, and monochromators were Al Kα (TAP), Ca Kα (LPET), P Kα (VLPET), S Kα (VLPET), and Cr Kα (LLIF).

Grid Analysis

Bulk compositions of individual inclusions (present bulk compositions of the exposed sections of the inclusions, pbc) were obtained using grid analysis as an alternative to spread beam (rastered or defocused beam) analysis (see S2 for a discussion of methodologies in the study of melt inclusions). In this technique, a rectangular grid (or multiple grids) of points is set up, covering the area of the inclusion. Then, an automated routine is run to analyze each of these points using standard, fully matrix-corrected, focused (or 2 μm sized) beam wavelength-dispersive analyses. The resulting set of analyses can then be processed by simple averaging (after identifying and eliminating analyses of surrounding olivine), to obtain a bulk composition for the inclusion (e.g., S5-a and S5-b). Compared with spread beam analysis, grid analysis has the advantage that it greatly reduces the number of mixed phase analyses and thus minimizes errors introduced by the unequal density effect (Warren 1997; Lindstrom 1999). It also allows for much better separation of the inclusion from the surrounding host, thus eliminating the need to estimate the degree of overlap with the host in the spread beam (e.g., Treiman 1993). In addition, the obtained set of analyses can be processed to yield other useful information, such as modal abundances, individual phase analyses, and phase or elemental distribution maps (S5-c and S5-d).

The step size of the grid should ideally be determined by the smallest grain size within the inclusion, subject to a minimum spacing of approximately 2 μm (limited by nominal diameter of focused electron beam), but was actually determined in this study by constraints of funding (smaller step size ⇒ more analyses ⇒ more time ⇒ more money). Typical step sizes ranged from 2 to 9 μm, and total time per grid analysis (i.e., per inclusion) ranged from approximately 18 to 70 h. In all cases, step sizes were adequate to ensure accurate representation of the major phases in the inclusions (pyroxenes, glasses, feldspars). The larger step sizes may have resulted in less accurate determination of TiO2 and P2O5, oxides concentrated in the small, minor phase Fe-Ti oxides and phosphates. Conditions of the analyses and sample preparation procedures were designed to minimize beam damage (in glasses and other fragile phases) that might result in erroneously low alkali concentrations, following a series of tests described below.

Electron Microprobe Analysis (EMPA)

Immediately following the X-ray mapping, we obtained bulk compositions (11 elements) by grid analysis of two melt inclusions, one in each of NK19A and NK12A (inclusions NK19A1 and NK12A1). Analyses were carried out at JSC, using the CAMECA SX-100. Conditions were 15 keV, 20 nA beam current, and 10–40 s count times on peaks, with backgrounds measured in every analysis. Natural and synthetic minerals, glasses, and metals were used as standards. Step sizes were 10 μm (NK19A1) and 13 μm (NK12A1).

Results obtained from these grid analyses showed bulk K2O and Na2O contents that were much lower (e.g., in NK19A1 by factors of approximately four and six, respectively) than those obtained by Treiman (1993) for the same inclusions. Subsequent SEI (secondary electron imaging) and BEI observation of these inclusions revealed significant damage to glasses, feldspars, and “iddingsite,” in the form of parallel lines (or stripes) of pits with approximately 2 μm spacing (S6). Based on the 2 μm spacing, it is clear that this damage was caused by the X-ray mapping, rather than by the grid analyses themselves. To determine whether this damage was responsible for the low Na2O and K2O contents obtained in the grid analyses, we analyzed individual phases in the inclusions, and obtained Na2O and K2O contents for the glasses that were significantly lower than those given in Treiman (1993) for the same inclusions (e.g., 0.1–0.6% Na2O and 0.7–1.6% K2O, compared to 1.2% and 4.9%, respectively, in NK19A1). We then polished down both samples until no visible evidence of the damage (stripes) remained. This required significantly more aggressive polishing than a typical treatment to remove beam-damaged C-coat; i.e., at least 10–15 min on 3 μm grit, followed by a series of finer grits. Reanalysis of the glasses after polishing showed much higher alkali contents (e.g., average approximately 6% K2O and approximately 2.8% Na2O in glass in NK19A1). These tests showed that X-ray mapping for trace elements in the host crystals (under typical high current conditions) severely compromises the compositional information that can be obtained from melt inclusions unless they are adequately repolished after the mapping. Thus, the results from the first two grid analyses were discarded, and we did not carry out further P-mapping of host crystals in this study.

The majority of the individual phase and grid analyses (13 elements, including Cl and F) presented in this paper were obtained at U. Mass using the CAMECA SX-50 electron microprobe, operating at 15 keV and 10 nA. Counting times ranged from 5 to 20 s on peaks, with a nominally 2 μm diameter beam. Na and K were always analyzed first, with shorter counting times than other elements. Na and F were analyzed using a PCO monochromator, which is optimized for light element analysis. Natural and synthetic minerals and glasses were used as standards.

We found that these conditions minimized alkali loss during the analyses (thus optimizing the accuracy of Na2O and K2O concentrations), but still left beam damage, with detectable loss of Na and, to a lesser extent, also K (probably because the analyses continued for longer than it took to measure Na and K) in glasses. In one extreme case, we witnessed a drop in measured Na2O from approximately 8.6 to 1 wt%, and in K2O from approximately 3 to 2 wt% upon reanalysis of glass in an inclusion that had been previously analyzed in a grid. Similar effects were observed regardless of whether grid analysis preceded individual phase analysis or vice versa. The extent of repolishing that was required to remove this damage was, again, found to be greater than that typically needed to remove C-coat (although much less than that needed to remove the damage from the X-ray mapping). Thus, we concluded that reliable analyses of glasses (and possibly other fragile phases as well), whether individual phase analyses or part of grid analyses, can only be obtained on freshly (re)polished samples (including samples newly borrowed from museums if their history is unknown). All data reported here were obtained under such conditions. Because this procedure was time consuming, we did not obtain individual phase analyses from all inclusions (although in many cases, good individual phase analyses were obtained as part of the grid analysis).

Some grid analyses were obtained at The University of Arizona (CAMECA SX-50 and SX-100) and at the American Museum of Natural History (CAMECA SX-100). Analytical conditions were similar to those used at U. Mass.

Results

Inclusions in Olivine: General

As described previously (Treiman 2005 and references therein), Nakhla contains approximately 7–15% olivine, occurring as subhedral grains or, more commonly, anhedral masses intergranular to and partly poikilitically enclosing augite grains (Fig. 1). These masses (or grains) range up to several mm in size, generally larger than the augite grains. They show slight normal zoning, with Fo ranging from approximately 42 to 30.

Figure 1.

Backscattered electron images (BEI) of Nakhla, showing olivine (light gray) and augite (darker gray). a) Section NK19. Box around olivine crystal NK19A, with large melt inclusion near center (see Fig. 3). b) Section NK 1137. Irregularly shaped masses of olivine poikilitically and partially poikilitically enclosing augite grains.

The olivines contain both monomineralic and polymineralic inclusions. The monomineralic inclusions are quite common (Fig. 1) and consist of rounded grains of augite having the same composition as cumulus augite cores (i.e., they are poikilitically enclosed augite grains). The polymineralic inclusions, most of which we interpret to be melt inclusions, are much less abundant. Typically, many olivines in a section have none, while a few olivines have 1–2. Two exceptional olivine grains are NK19A with 5 and NK12A with 9 polymineralic inclusions (Figs. 2a and 3a). In addition, all olivines contain thin lamellae and tiny (≤ μm-sized) blebs consisting of symplectic augite-spinel intergrowths (Fig. 3c), which have been interpreted as products of exsolution (Mikouchi et al. 2000).

Figure 2.

Olivine grain NK12A, containing nine melt inclusions. a) Backscattered electron images with melt inclusions labeled. b) Phosphorus K-α X-ray map of NK12A, shown at approximately same scale and orientation as in (a). c) Drawing from Donaldson (1976) of swallowtail morphology of olivine, with crystallographic directions. Based on the phosphorus X-ray map (b), NK12A appears to have a relict core of this morphology. d) Chromium K-α X-ray map of same area as in (b).

Figure 3.

Olivine grain NK19A, containing five melt inclusions. a) Backscattered electron image with melt inclusions labeled. b) Phosphorus K-α X-ray map of NK19A, shown at approximately same scale and orientation as in (a). c) Higher magnification BEI of part of NK19A, showing crystallographically oriented exsolution lamellae consisting of symplectic intergrowths of magnetite + pyroxene. d) Drawing of inferred internal growth zones in NK19A. Based on phosphorus X-ray map (b) and crystallographic orientation inferred from the symplectic exolutions (c), the grain appears to have a relict core with hopper (Donaldson 1976) skeletal morphology. e) Chromium Kα X-ray map of same area as in (b). f) Drawing from Donaldson (1976) of hopper skeletal morphology of olivine, with crystallographic directions.

X-ray Mapping of Olivines NK12A and NK19A

Figures 2 and 3 show P-Kα and Cr-Kα X-ray maps of olivines NK12A and NK19A, both of which contain many polymineralic inclusions. In BEI (Fig. 2), olivine grain NK12A appears to consist mainly of one subhedral crystal (a tabular polyhedron; Donaldson 1976), with an outermost rim that partially encloses surrounding augite grains poikilitically. It appears to be connected to a smaller sub/anhedral olivine grain at one corner (upper left of Figs. 2a and 2b). Fe/Mg zonation is present, but subtle (not visible in BEI), with Fo approximately 42 near the center of the grain and Fo approximately 39 near an edge (not quite the outermost rim, which is more ferroan). The P X-ray map reveals many internal details similar to those described by Milman-Barris et al. (2008) for P zonation in natural and experimental olivines. The grain appears to have a relict P-rich core with swallowtail morphology (Donaldson 1976), and fine-scale concentric oscillatory zoning of P-rich and P-poor olivine in the subsequently infilled area between the wings. Superimposed on these features are P-poor zones surrounding all nine of the polymineralic inclusions in the grain. All of the inclusions appear to be located in the infilled areas, rather than in the relict core. Based on the concentric zonation, the three largest inclusions appear to belong to the same generation, and all the smaller inclusions to have been trapped at later times. The P-poor zones surrounding the inclusions are sometimes very large relative to the size of the inclusion, and are typically asymmetric with respect to the inclusion (as observed also by Milman-Barris et al. 2008), suggesting that they have migrated. The Cr X-ray map does not show the relict core and oscillatory zoning seen in the P map. Most of the internal detail seen in the Cr map is due to the abundant spinel-pyroxene exsolutions (Mikouchi et al. 2000) in the olivine, and the tiny chromites rimming the inclusions. Most of the inclusions are surrounded by Cr-poor (i.e., exsolution-free) zones similar in size and morphology to the P-poor zones.

In BEI (Fig. 3), olivine NK19A appears to be a single subhedral crystal (tabular polyhedron; Donaldson 1976), with outermost zones partially enclosing surrounding augite grains poikilitically. It is homogeneous at Fo approximately 35.5, except for its outermost rim, which is more ferroan. Spinel-pyroxene exsolutions, known to be parallel to the (100) plane (Mikouchi et al. 2000), are abundant in the grain, and can be used to constrain its crystallographic orientation (Fig. 3). Again, the P X-ray map reveals many internal details similar to those described by Milman-Barris et al. (2008). The grain appears to have a relict P-rich core with hopper skeletal morphology (Donaldson 1976), based on its apparent shape and inferred crystallographic orientation. The large melt inclusion NK19A1 occupies the central embayment in the core; the smaller inclusions A2–A5 (plus several others that were too small to study) are symmetrically arrayed on each side of the central embayment along a [111] diagonal. The olivine overgrown on the P-rich core is concentrically zoned around it, with several oscillations of P-poor and P-rich zones. The outermost zone is interrupted in several places by embayments of surrounding augite grains. The melt inclusions are all surrounded by P-poor halos. Again, the Cr map does not show the relict core and zoning seen in the P map, but shows clearly the orientation of the spinel-pyroxene exsolutions and the tiny chromite grains rimming the inclusions.

Polymineralic Inclusions in Olivine: Petrography

Petrographic characteristics of each of the 37 polymineralic inclusions in olivine examined here are summarized in S4 (along with BEI of each). The inclusions range from approximately 25 to 450 μm in apparent maximum dimension (Fig. 4) and have rounded, commonly elliptical, shapes (Figs. 5 and 6; S4). All inclusions less than approximately 170 μm in size have a relatively simple texture and phase assemblage, in which augite occurs as a rind around the inclusion and in the interior as skeletal/dendritic crystals with interstitial glass (Fig. 5). All inclusions > ~150 μm in size have more complex textures and phase assemblages (Figs. 6 and 7), including large euhedra or massive crystals of orthopyroxene (in addition to skeletal/dendritic augite), and either plagioclase feldspar or K-feldspar or both. The K-feldspar occurs principally as an interstitial phase texturally equivalent to the glass. The plagioclase occurs as large, irregularly shaped masses, or as an interstitial phase. One inclusion contains an approximately 100 μm long patch of scapolite (Fig. 6a), a phase that has not previously been reported in a Martian meteorite. Minor phases occurring in inclusions of any size are chromite, Fe-Ti oxides, phosphates, Fe-sulfide, and rare silica blebs. Chromite occurs as minute wisps decorating the augite rims, and, in a few cases, as larger grains within the inclusions (Fig. 6a).

Figure 4.

Histogram of number of inclusions having “simple” phase assemblages and textures versus “complex” phase assemblages and textures (as defined in text), as a function of apparent inclusion size, for the 37 polymineralic inclusions in Nakhla studied in this work.

Figure 5.

Backscattered electron images of melt inclusions in olivine in Nakhla with “simple” phase assemblages consisting of augite (rims and dendritic/skeletal crystals) and glass (gl), with minor Fe-Ti oxide, Fe-sulfide, phosphates, and silica blebs.

Figure 6.

Melt inclusions in olivine in Nakhla with “complex” phase assemblages. a) NK12A1, the largest melt inclusion observed in this study (see Fig. 2a). Superimposed backscattered electron image (BEI) and K k-α X-ray map. b) NK20A1. BEI combined with K k-α, and Na k-α X-ray maps. c) NK19A1 (see Fig. 3a). BEI. opx = orthopyroxene; idd = iddingsite.

Figure 7.

a) Backscattered electron images (BEI) of melt inclusion NK38871A1 in olivine in Nakhla, containing large grains of orthopyroxene and sodic plagioclase, in addition to area of dendritic/skeletal augite with interstitial K-feldspar and glass. b) BEI combined with K k-α and Na k-α X-ray maps of area outlined by box in (a). c) BEI of melt inclusion NK13A1 in olivine in Nakhla, containing orthopyroxene, augite, albitic plagioclase, and K-feldspar. Feldspars are significantly altered to iddingsite. d) BEI combined with K k-α and Na k-α X-ray maps of area outlined by box in (c). e–f) BEI of unusual inclusions NK12A7 and NK12A8 in olivine in Nakhla. Note absence of augite rind or skeletal/dendritic crystals, and absence of glass. These may be xenoliths rather than melt inclusions. plag = plagioclase; opx = orthopyroxene; aug = augite; idd = iddingsite; ol = olivine; K-spar = K-feldspar; sulf = Fe-sulfide (pyrrhotite).

Many (but not all) inclusions contain some “iddingsite,” a commonly observed alteration product in Martian meteorites (Treiman 1993; Anand et al. 2005; Stopar et al. 2007; Changela and Bridges 2010; Hallis and Taylor 2011). In these inclusions, it appears to replace primarily glass and/or feldspar (as has been previously observed in nakhlites; Treiman et al. 1993). In a few cases, it appears to replace a mafic phase (ferroan augite, orthopyroxene, or fayalite?) that occurred as overgrowths on dendritic augite crystals. It sometimes appears to have been intruded into the inclusions from veins in the surrounding olivine (Fig. S4-v). In places (e.g., Fig. 7f), it contains numerous tiny needles (too small to analyze) of a phase that is brighter than the iddingsite in BEI and could be lawrencite (based on EDS analyses showing higher Cl and Fe contents than in needle-free iddingsite). Estimates of iddingsite abundance were used to assess degree of alteration for each inclusion, using a qualitative scale from none to major (S4). Two of the largest inclusions, NK12A1 (Fig. 6a) and NK19A1 (Fig. 6c), were point-counted manually from collages of combined BEI and elemental X-ray maps. Results are given in Table 2.

Table 2. Modal abundances of melt inclusions NK12A1 and NK19A1 in olivine in Nakhla.
 NK12A1NK19A1
Area%wt%bArea%wt%b
  1. a

    Phase could not be identified and/or cracks and pits.

  2. b

    Converted from area% using the following densities (in g/cc): augite 3.2; glass 2.5; K-feldspar 2.6; iddingsite 2.8; scapolite 2.6; Fe-Ti oxides 4.3; chromite 4.8.

Augite5964.94957
Glass108.73733.8
K-feldspar2018  
Scapolite10.9  
Fe-Ti oxides2322.1
Chromite0.50.8  
Iddingsite43.767.2
Othera3 6 

Two large inclusions in olivine grain NK12A (Fig. 2a), NK12A7 and NK12A8, have textures and phase assemblages that differ significantly from those of all other polymineralic inclusions (Figs. 7e and 7f). Neither of these inclusions contains dendritic/skeletal crystals or glass, but rather consist of coarse crystals or formless masses of their constituent phases. They also lack the rinds of pyroxene decorated with tiny chromite grains that are found in all other inclusions. NK12A7 (Fig. 7e; Fig. S4-k) consists principally of a 320 μm long lath of orthopyroxene, plus two distinct compositions of plagioclase. One side of the orthopyroxene lath is graphically intergrown with albitic plagioclase. Minor phases include phosphate, Fe-Ti oxides, Fe-sulfide, and olivine. There is a moderate amount of iddingsite, which appears to be principally replacing the albitic plagioclase. NK12A8 (Fig. 7f; Fig. S4-l) consists mainly of massive augite and calcic plagioclase, with minor Fe-sulfide, Ca-phosphate, hercynitic spinel, and a myrmekite-like intergrowth of olivine and K-feldspar. It also has a large amount of iddingsite, which, in one area, appears to be replacing K-feldspar.

Polymineralic Inclusions in Olivine: Phase Compositions

Glasses in 15 polymineralic inclusions have average compositions of approximately 59–70% SiO2, 15–18% Al2O3, 3–7% K2O, and 2–7% Na2O, with K2O/Na2O ratios approximately 0.5–2.5 (Table 3). Analyses of plagioclase in NK20A1 and NK38871A1 show near-stoichiometric compositions of Ab65Or3.6 and Ab59Or2.7, respectively (Table 4). Analyses of K-feldspar in NK12A1, NK20A1, and NK38871A1 (Figs. 6 and 7) give average compositions of ~Or87Ab13, Or63Ab32, and Or52Ab35, respectively, but show greater deviations from stoichiometry than the plagioclase (Table 4). These areas may be mixtures of K-feldspar and glass, based on their nonhomogeneous appearance in BEI and X-ray maps (e.g., Fig. 7b), and the presence of significant FeO and P2O5 in the analyses (Table 4). The scapolite in NK12A1 (Fig. 6a) contains approximately 3.9% Cl, and has the structural formula approximately (Na2.8Ca0.7K0.3)3.9(Si8.3Al3.7)12ClO24 (Table 4). Augites in the inclusions are highly zoned and, except in NK12A8 (below), were not analyzed in this work. Analyses of augite in a few of the inclusions studied here are given by Treiman (1993). The orthopyroxenes in NK38871A1, NK20A1, and NK11A1 have compositions of Wo approximately 2–5 and mg# 44–50. Analyses of iddingsite in the inclusions show variable compositions, with approximately 30–50% SiO2, 24–35% FeO, and approximately 0.5–1.9% K2O (Table 5). There is no correlation between the composition of iddingsite and the phase that is apparently being replaced.

Table 3. Analyses of glass in melt inclusions in olivine in Nakhla.
 NK11A1 glassNK11A2 glassNK12A1 glassNK19A1 glassNK19A2 glassNK19A4 glassNK19D1 glass
Avg (2)StdevAvg (2)Stdev(1)Avg (9)StdevAvg (2)StdevAvg (2)Stdev(1)
SiO268.60.770.91.269.572.51.358.60.467.51.763.4
TiO20.160.010.100.060.200.170.030.320.020.010.000.08
Al2O316.30.415.70.615.815.00.315.21.419.01.218.1
Cr2O3bdl bdl bdlbdl 0.130.05bdl 0.05
FeO0.900.161.400.211.16bdl64.80.91.360.042.86
MgObdl 0.210.280.44bdl 5.61.30.100.010.84
MnObdl bdl 0.04bdl 0.180.01bdl 0.06
CaO2.50.33.52.31.31.31.18.71.01.91.84.3
K2O5.80.11.90.05.36.00.32.90.23.90.83.9
Na2O3.60.32.50.82.72.80.62.92.02.93.33.9
P2O51.270.351.600.450.040.310.551.410.230.750.901.67
Fbdl bdl bdlbdl bdl 0.140.000.25
Cl0.24 0.35 0.160.120.040.330.030.460.020.07
O=F,Cl−0.05 −0.08 −0.04−0.03 −0.07 −0.16 −0.11
Total99.3 98.1 96.599.0 101.1 97.9 99.6
 NK19E1 glass NK19E2 glass NK1137A2 glass NK1137A4 glass NK1137A1 glass NK1137A3 glass NK38871B1 glass NK38871C1 glass
(1)(1)(1)(1)Avg (2)StdevAvg (3)StdevAvg (5)Stdev(1)
  1. bdl = below detection limit.

SiO272.372.068.468.170.41.069.71.667.64.469.3
TiO20.080.080.070.150.080.040.050.020.110.050.05
Al2O317.616.417.514.716.70.218.60.416.00.915.9
Cr2O3bdl bdlbdlbdl bdl1bdl bdl
FeO0.911.061.381.061.400.181.390.553.606.211.21
MgObdlbdl0.700.080.350.430.540.540.640.900.16
MnObdl bdlbdl0.060.06bdl bdl bdl
CaO0.71.31.32.92.71.61.31.01.90.72.9
K2O4.65.72.93.22.60.12.90.26.50.72.7
Na2O4.83.77.25.36.10.67.12.22.70.13.5
P2O50.030.370.321.511.270.530.090.031.070.602.22
Fbdlbdl0.20bdlbdl bdl bdl bdl
Cl0.310.210.440.440.33 0.29 bdl 0.32
O=F,Cl−0.07−0.05−0.18−0.10−0.07 −0.07   −0.07
Total101.3100.999.797.0101.6 101.7 100.1 98.3
Table 4. Analyses of scapolite and feldspars in melt inclusions in olivine in Nakhla.Thumbnail image of
Table 5. Analyses of iddingsite in melt inclusions in olivine in Nakhla.
 NK11A1 iddingsiteNK19A1 iddingsiteNK19D1 iddingsiteNK20A1 iddingsite
(1)Avg (6)Avg (4)(1)
  1. bdl = below detection limit.

SiO240.441.734.550.6
TiO20.020.030.120.02
Al2O35.54.87.72.1
Cr2O3bdlbdlbdlbdl
FeO32.430.834.824.8
MgO6.17.04.57.1
MnO0.200.440.340.29
CaO0.540.783.091.56
K2O0.91.91.00.5
Na2O0.130.540.520.01
P2O50.110.151.960.05
Fbdlbdl0.60bdl
Cl0.400.360.630.41
O=F,Cl−0.09−0.08−0.39−0.09
Total86.788.589.587.3

Compositions of phases in NK12A7 and NK12A8 are given in Table 6. The orthopyroxene in NK12A7 differs from that in other polymineralic inclusions in having very low Wo (<1). The plagioclase intergrown with this orthopyroxene is albitic, approximately Ab74Or9. The other (massive) plagioclase in the inclusion is calcic, approximately An66Or1. The augite in NK12A8 is Wo approximately 35, mg# 56, which is similar to the rims on Nakhla cumulus augite (Treiman 2005); it has distinctly lower TiO2 and Al2O3 contents than typical augite in melt inclusions in olivine in Nakhla (Treiman 1993). The main plagioclase in this inclusion has a ternary composition, An59Ab20Or21. The K-feldspar intergrown with olivine (Fo 41) is Or94Ab5. The hercynite is (Fe0.77Mg0.25)2.02(Al1.91Cr0.02Ti0.09)2.01O4.

Table 6. Analyses of phases in inclusions NK12A7 and NK12A8 in olivine in Nakhla.Thumbnail image of

Polymineralic Inclusions in Olivine: Present Bulk Compositions

Present bulk compositions (pbc; see S3 for melt inclusion terminology) of 21 melt inclusions obtained by grid analysis are given in Table 7. The two most heavily altered inclusions, NKFMNHC1 and NK19A3, have significantly lower SiO2 and higher FeO than any of the other inclusions. Excluding these two heavily altered inclusions, pbc from grid analyses have approximately 50–60% SiO2, 5–10% Al2O3, 10–19% FeO, 5–10% MgO, 11–15% CaO, 0.2–3% K2O, 0.6–3% Na2O, and 0.03–0.3% Cl. Bulk fluorine was below the detection limit in all inclusions (the only phases observed to contain detectable F were phosphates).

Table 7. Present bulk compositions of 21 melt inclusions in Nakhla determined by grid analysis.
InclusionNK12A1NK20A1NK38871A1NK19A1NK11A1NK12A3NK13A1NK38871B1NK4A1NKFMNHB1NK12F1NK12A2
SiO251.452.750.954.050.454.953.553.749.557.151.756.9
TiO21.10.70.91.00.91.10.80.91.110.91.20.9
Al2O37.98.37.18.97.77.79.58.98.0310.38.19.9
Cr2O30.180.040.030.040.030.200.030.070.120.070.110.14
FeO16.115.017.115.417.713.515.812.318.6110.214.410.1
MgO6.25.79.44.67.05.35.65.86.964.66.96.2
MnO0.330.330.410.280.340.250.360.240.420.210.340.20
CaO12.412.410.911.511.914.310.513.511.5110.913.412.1
K2O3.22.30.82.21.51.40.252.21.72.51.61.1
Na2O0.721.71.61.31.30.692.961.160.992.141.291.38
P2O50.500.730.690.690.900.540.531.000.180.790.820.88
Cl0.110.100.150.140.140.110.110.200.090.220.030.16
O=Cl−0.02−0.02−0.03−0.03−0.03−0.03−0.03−0.05−0.02−0.05−0.01−0.04
Total100100100100100100100100100100100100
K2O/Na2O4.41.40.51.71.12.00.11.91.71.21.30.8
Molar Fe/Mg1.41.51.01.91.41.41.61.21.51.21.20.9
Size (μm)45030024021019017016013013011010090
Host Fo42.141.340.435.440.342.530.037.938.334.336.042.7
Major phases

augite

K-spar

scapolite

glass

augite

opx

Na-plag

K-spar/glass

augite

opx

Na-plag

K-spar/glass

augite

glass

augite

opx glass

augite

glass

augite

opx

Na-plag

K-spar?

augite

glass

augite

glass

augite

glass

augite

glass

augite

glass

Minor phases

chromite

Fe-Ti ox

phos

chromite Fe-Ti ox

phos sulfide

chromite

Fe-Ti ox

phos

chromite

Fe-Ti ox

phos

chromite

Fe-Ti ox

phos

sulfide

chromite

Fe-Ti ox

phos

silica

chromite

Fe-Ti ox

phos

chromite

Fe-Ti ox

phos

chromite

Fe-Ti ox

phos

sulfide

chromite

Fe-Ti ox

phos

chromite

Fe-Ti ox

phos

chromite

Fe-Ti ox

phos

IddingsiteMinorMinorMinorMinor+Minor+MinorModerateMinorModerateNoneMinorMinor
Phase analyses?YesYesYesYesYes  Yes    
Where analyzedU. MassU. AZU. AZU. MassU. AZU. MassAMNHU. MassAMNHU. AZU. AZU. Mass
Grid step size9 μm8 μm9 μm8 μm6 μm4 μm5 μm6 μm4 μm6 μm5 μm3 μm
InclusionNK19E1NK4B1NK19E2NK12A5NK11A2NKFMNHC1NK12A4NK19A3NK19A2Complex Avg (5)aSimple Avg (12)aWeighted Avg (17)b
  1. a

    All averages exclude the four inclusions with major or moderate amounts of iddingsite. Complex and simple refer to textures and assemblages as defined in text.

  2. b

    Average of all inclusions weighted by area (following Treiman 1993). ox=oxide; phos=phosphate

SiO252.454.249.659.452.848.855.344.850.051.954.052.3
TiO22.31.60.91.31.21.00.80.80.60.941.11.0
Al2O38.49.97.29.68.18.98.55.15.48.08.58.1
Cr2O30.240.200.310.110.240.110.130.130.110.070.160.11
FeO14.011.518.49.512.318.111.922.816.716.312.915.4
MgO7.15.78.35.97.36.77.39.79.86.66.76.3
MnO0.360.290.370.180.180.320.270.520.430.340.280.32
CaO11.812.812.411.615.112.313.514.014.911.813.012.2
K2O1.31.11.20.710.640.910.690.230.482.01.22.3
Na2O1.161.280.860.791.171.600.560.681.121.31.11.1
P2O50.741.170.490.840.880.810.741.070.530.700.780.65
Cl0.150.280.110.160.140.310.160.170.080.130.150.12
O=Cl−0.03−0.06−0.02−0.04−0.03−0.07−0.04−0.04−0.02   
Total100100100100100100100100100100.0100.0100.0
K2O/Na2O1.10.81.40.90.50.61.20.30.41.51.12.0
Molar Fe/Mg1.11.11.20.90.91.50.91.31.01.41.11.4
Size (μm)757565655352523330   
Host Fo38.139.536.741.641.732.539.035.835.4   
Major phases

augite

glass

augite

glass

augite

glass

augite

glass

augite

glass

augite

glass

plag

augite

glass

augite

glass

augite

glass

   
Minor phases

chromite

Fe-Ti ox

phos

sulfide

chromite

Fe-Ti ox

phos

sulfide

chromite

Fe-Ti ox

phos

sulfide

chromite

Fe-Ti ox

phos

chromite

Fe-Ti ox

phos

sulfide

chromite

Fe-Ti ox

phos

sulfide

chromite

Fe-Ti ox

phos

chromite

Fe-Ti ox

phos

sulfide

chromite

Fe-Ti ox

phos

   
IddingsiteNoneMinorMinorNoneNoneMajorNoneMajorNone   
Phase analyses?Yes Yes Yes   Yes   
Where analyzedU. MassAMNHU. MassU. MassU. AZU. AZU. MassU. MassU. Mass   
Grid step size4 μm3 μm2 μm3 μm3 μm3 μm2 μm2 μm2 μm   

The grid analysis of NK12A1, which contains both K-feldspar and K-rich glass (Fig. 6a; S5), shows the highest bulk K2O content and K2O/Na2O ratio, with 3.2% K2O and 0.7% Na2O. These values are in reasonable agreement with values of 3.0% K2O and 0.6% Na2O estimated by combining modal abundances (assuming that area% = volume% and converted to wt% using estimated densities) of K-feldspar, glass, and scapolite (Table 2) with average compositions of these phases (Tables 3–5). The grid analysis of NK19A1 (Fig. 6c) shows bulk K2O and Na2O of 2.2 and 1.3%, respectively. These values are in reasonable agreement with values of 2.1% K2O and 1.0% Na2O estimated by combining modal abundance of glass (Table 2) with the average composition of this phase in the inclusion (Table 3).

Six of the inclusions for which we obtained grid analyses were also studied by Treiman (1993), so we can compare the new pbc with those obtained from his rastered beam analyses. A source of uncertainty in the latter is the degree of overlap of the beam with surrounding olivine. Treiman (1993) estimated that the average rastered beam analysis for eight analyzed inclusions contained 28% olivine, and so subtracted this amount of olivine from the averaged composition. He acknowledged this to be an approximation, because the degree of overlap probably varied from inclusion to inclusion. To permit direct comparison of the new olivine-free compositions with those of Treiman (1993), for each of the six inclusions, we calculated the amount of olivine that had to be subtracted from the rastered beam composition of Treiman (1993) to produce an SiO2 content equal to that obtained in our grid analysis. Calculated values (as detailed in S7) ranged from approximately 12 to 29% olivine. Table 8 compares the resulting olivine-free rastered beam results with the new results from grid analysis. There is no reason to expect exact agreement between these two sets of data (even if there were no systematic differences resulting from the differences in method), because repolishing of the samples in the interim between the two studies may well have exposed sections of the inclusions with slightly different compositions. Focusing on the alkalis, which are the main interest here, Table 8 shows that, for all six inclusions, K2O contents from the grid analyses are lower by approximately 9–50% relative. For four of the inclusions, Na2O contents from the grid analyses are lower by approximately 10–135% relative. The lower alkali contents in the grid analyses might be explained by exposure of less glass and/or feldspar in the inclusions, judging by Al2O3 contents (which are approximately 10–17% lower in the grid analyses) and SiO2/(FeO + MgO) ratios (which are approximately 10–20% lower in the grid analyses. But, this does not appear to explain the entire difference. Thus, it appears that alkalis are systematically higher in the rastered beam analyses, which could be a result of the unequal density effect in mixed phase analyses (see S3).

Table 8. Comparison between present bulk compositions obtained by grid analysis and those obtained by rastered beam analysis (Treiman 1993) for six individual polymineralic inclusions in olivine, and area-weighted averages of inclusions in olivine in Nakhla from this work and the Treiman (1993) data
 (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)
 NK11A1NK12A1NK12A2NK12A3NK19A1NK20A1Average
 T93aThis workT93aThis workT93aThis workT93aThis workT93aThis workT93aThis workThis workT93a
 Rastered beamGrid analysisRastered beamGrid analysisRastered beamGrid analysisRastered beamGrid analysisRastered beamGrid analysisRastered beamGrid analysisWeighted avg (17)Weighted avg (8)
  1. a

    Compositions from Table 2 of Treiman (1993), minus amount of olivine calculated to yield same SiO2 content as corresponding olivine-free grid analysis from this work (see S7).

SiO250.450.451.451.456.956.954.954.954.054.052.752.752.352.5
TiO21.20.91.21.11.20.91.01.11.01.00.70.71.01.0
Al2O39.07.79.27.911.29.98.77.79.98.99.78.38.19.5
Cr2O30.10.030.00.180.20.140.10.200.00.040.00.040.110.05
FeO15.417.713.416.17.410.113.113.514.915.413.515.015.413.5
MgO6.27.05.66.24.86.24.85.33.24.64.65.76.34.9
MnO0.20.340.20.330.10.200.20.250.30.280.40.330.320.26
CaO12.911.913.412.412.312.113.714.311.611.512.712.412.213.0
K2O1.81.54.13.21.41.12.11.42.42.22.62.32.33.1
Na2O1.61.30.70.73.31.40.60.71.81.31.91.71.11.4
P2O50.90.900.70.500.90.880.60.540.70.690.90.730.650.78
Total99.899.899.899.999.899.899.999.999.799.999.799.9100.0100.0

Among the present bulk compositions obtained in this work for 21 melt inclusions, there appears to be a positive correlation between K2O content and inclusion size (Fig. 8a). In addition, among the nine of these inclusions for which individual phase analyses were obtained (Tables 3–5), there is a positive correlation between K2O content of glass and K2O content of the bulk composition (Fig. 8d), which suggests that the correlation between bulk K2O content and inclusion size is not due to unrepresentative sampling. Nevertheless, the quality of the correlation (r2 = 0.44) is not high. There is a suggestion of a similar (but even lower quality) correlation between bulk Na2O content and inclusion size, but the largest inclusion, NK12A1 (which has the highest K2O content), falls distinctly off this trend, with very low Na2O (Fig. 8b). There are no apparent correlations of K2O content or inclusion size with any other elemental abundance or ratio (e.g., SiO2/[FeO + MgO]), nor with degree of alteration (Fig. 8a) or composition of host olivine (Fig. 8c). Although there is no clear indication as to how alteration (i.e., presence of iddingsite) affects bulk inclusion compositions, we eliminate the four inclusions that show moderate to major amounts of iddingsite from all further discussion.

Figure 8.

Present bulk compositions of 21 melt inclusions in olivine in Nakhla from grid analyses in this study. a) Diameter of inclusion versus wt% K2O in bulk composition. b) Diameter of inclusion versus wt% Na2O in bulk composition. c) Composition of olivine host versus wt% K2O in inclusion. d) wt% K2O in present bulk composition versus wt% K2O in constituent glass of the same inclusion.

Average pbc for all (5) “large, complex” inclusions versus all (12) “small, simple” inclusions are shown in Table 7. These compositions are similar, with the most notable differences being lower SiO2/(FeO+MgO) ratio and higher alkali contents in the former. Table 7 also gives the average of all 17 (low-alteration) inclusions, weighted by approximate exposed area of the inclusion. This result can be compared (Table 7) with the similarly weighted inclusion average of Treiman (1993), after correcting the latter for 21% overlap with olivine (see S7). The comparison shows similar compositions, with the main differences being that the new average is slightly more mafic (lower SiO2/[FeO + MgO] ratio and lower Al2O3, K2O and Na2O contents). The new weighted average K2O content of 2.3 wt% compares with 3.1% in the Treiman (1993) composition. However, K2O/Na2O ratios are similar (2.0 versus 2.2). These observations suggest that the main difference between the two compositions is sampling of a slightly higher proportion of mafic phases (relative to glass and/or feldspar) in the new composition.

Inclusions in Augite

We surveyed an area of approximately 0.5 cm2 of section NK12, consisting of approximately 90% augite. As described previously (Treiman 2005 and references therein), augite crystals average approximately 1 × 0.2 mm in size, with the larger grains being euhedral to subhedral prisms. They have relatively homogeneous, magnesian cores (mg# approximately 62.5, Wo 38.5) with irregular, 10–30 μm wide FeO-rich rims (mg# approximately 47–27; Treiman 1990). Melt inclusions in augite are much more abundant than those in olivine, and also much smaller. Most augite grains have at least one inclusion. We examined a total of 91 melt inclusions approximately 1–15 μm in size, of which 24 were located in augite cores and 67 were located in the more FeO-rich rims (Fig. 9). The inclusions in the rims are commonly located very close the boundary with the core (Figs. 9a–c), suggesting that they were trapped in response to a change in crystallization conditions (Roedder 1984). In addition, submicron-sized glassy inclusions are abundant in some cores. The melt inclusions located in augite grain cores consist dominantly of glass and Fe-Al-rich augite, in apparently various proportions. The pyroxene typically forms a halo around the glass and commonly shows negative crystal shapes against surrounding core augite (Figs. 9d–f). In addition, some inclusions contain subhedral crystals of pigeonite (Figs. 9d–f), and many contain grains of Fe-Ti oxide, pyrrhotite, and/or hercynite (Fe-Al-rich spinel). In most cases, the opaques occur as tiny (submicron-sized) grains, but we observed several inclusions with much larger opaques (5–10 μm sized) that apparently occupy a large percent (25–50%) of the volume of the inclusions (Fig. 9f). None of the melt inclusions in augite cores that we studied are penetrated by fractures or surrounded by radial fractures, although the augite within the inclusions often contains short cracks that end at the boundary with the surrounding core augite. The melt inclusions located in augite grain rims consist of glass and halo pyroxene, the latter being difficult to distinguish from the FeO-rich augite rims pyroxene in BEI. Five of the inclusions originally thought to be melt inclusions in rims turned out to be inclusions of albitic plagioclase with K2O ≤ 0.5 wt%.

Figure 9.

Backscattered electron images of melt inclusions in augite in Nakhla. aug = augite; mi = melt inclusion. a–c) Melt inclusions are more common in augite rims than in augite cores, and tend to be located very near the boundary between rims and cores. d–f) Melt inclusions in augite cores, consisting dominantly of Fe-Al-rich wall augite and glass, with minor pigeonite, pyrrhotite, Fe-Ti oxide, and hercynitic spinel. Inclusion in f) has exceptionally large grain of pyrrhotite that is probably a cognate grain.

Glass, halo augite, and core augite compositions were obtained from 10 melt inclusions in cores and 29 in rims (the majority of observed inclusions that were large enough for EMPA). Glasses in inclusions in cores have K2O contents ranging from approximately 1.2 to 5.5 wt%, with an average of 3.0 wt% (Table 9; Fig. 10). They have K2O/Na2O ratios ranging from 0.4 to 1.7, with an average value of 0.9 (Table 9). Glasses in inclusions in rims have K2O contents ranging from 0.6 to 4.6 wt%, with an average of 1.6 wt% (Table 9; Fig. 10). They have K2O/Na2O ratios ranging from 0.6 to 4.7, with an average value of 1.6 (Table 9). Considering the degree of internal variation, neither K2O nor Na2O contents are statistically different between the two sets of glasses (Table 9). There appears to be a correlation between SiO2 and Al2O3 contents among all glasses combined (Fig. 10b). There is no correlation of SiO2 with K2O (Fig. 10b) or with Na2O within either of the two sets or among the combined glasses.

Table 9. Compositions of glasses and pyroxenes in melt inclusions in augite in Nakhla.
 Glass in melt inclusions in augite coresGlass in melt inclusions in augite rimsPyx in melt inclusions in augite coresPyx in melt inclusions in augite rimsAugite
Avg (12)StdevAvg (28)StdevAvg (8)StdevAvg (23)StdevCores
SiO272.65.770.15.249.90.949.80.9052.1
TiO20.060.040.070.080.460.120.420.090.21
Al2O315.82.414.71.43.01.81.90.90.79
Cr2O30.030.040.030.030.210.100.240.090.41
FeO1.51.52.22.320.01.919.91.414.3
MgO0.500.90.740.889.60.69.40.813.2
MnO0.010.030.030.040.560.080.540.060.42
CaO2.12.82.31.816.50.617.50.818.3
Na2O3.40.83.01.00.290.160.270.210.16
K2O3.01.64.11.6bdl bdl bdl
P2O50.390.870.551.3bdl bdl bdl
Fbdl bdl bdl bdl bdl
Cl0.330.14bdl bdl bdl bdl
Total99.8 98.2 100.61.2100.20.8100.0
K2O/Na2O0.90.41.61.0     
mg#    46.23.045.63.262.3
Wo    36.31.638.01.538.3
Figure 10.

Glasses in melt inclusions in augite in Nakhla. a) Histogram of K2O contents of glasses in inclusions in augite cores and inclusions in augite rims. b) SiO2 content versus Al2O3 and K2O content of glasses in inclusions in augite cores and inclusions in augite rims.

Pyroxene halos in the inclusions in augite cores, as well as the pyroxene rims of the augite grains, have higher FeO (mg# approximately 40–50) and Al2O3 (up to approximately 4 wt%) contents than augite cores (Table 9). Analyses of augite cores (Table 9) are consistent with those in Treiman (2005).

Discussion

What Are NK12A7 and NK12A8?

As described above, inclusions NK12A7 and NK12A8 have textures, phase assemblages, and phase compositions that differ significantly from those of all other polymineralic inclusions (Fig. 7). The absence of skeletal/dendritic crystals, the characteristic rind assemblage (augite with tiny chromites), and glass all suggest that these are not melt inclusions. Dendritic/skeletal crystals are a common feature in crystallized melt inclusions (Frezzotti 2001), because heterogeneous nucleation and consequent overcrystallization of the host phase onto inclusion walls typically lead to undercooling and hence supersaturation of daughter minerals. Nucleation of daughter minerals on the wall phase, forming complete or partial rinds, is another common feature of crystallization within small closed systems. Thus, the coarse-grained textures and absence of concentric structure in these inclusions suggest that they did not crystallize from trapped liquids in situ. The phase assemblages and compositions of these inclusions (particularly myrmekite-like and graphic intergrowths, presence of hercynite, and Or94 feldspar) suggest highly evolved assemblages, such as might be expected in nakhlite mesostasis. Thus, it is possible that these objects are not inclusions at all, but rather areas of mesostasis protruding into the olivine crystal from the third dimension and misleadingly appearing as inclusions in the random section that is exposed. However, their feldspar compositions do not match any of those reported for Nakhla mesostasis (Treiman 1993). In addition, the appearance of these inclusions in the P X-ray map of their host olivine is similar to that of the comparably sized inclusion NK12A1 and, in fact, all the other inclusions in this olivine, in that they are surrounded by asymmetric, P-poor halos (Fig. 2b). These observations strongly suggest that NK12A7 and NK12A8 are indeed inclusions, and were trapped in the olivine at approximately the same time as the melt represented by NK12A1. One possibility is that NK12A7 and NK12A8 are xenoliths.

Melt Inclusions in Olivine

With the exception of NK12A7 and NK12A8, all the polymineralic inclusions in olivine have textures and phase assemblages that suggest that they represent trapped liquids (see Treiman 1993 for additional arguments to this effect), and it is a reasonable hypothesis that these liquids were samples of the melt from which their olivine hosts grew. The question, then, is whether all inclusions represent the same PTL—i.e., do they all represent the same magma and were they all trapped at approximately the same stage of evolution of that magma?

One of the major difficulties in attempting to retrieve the composition of the PTL from a set of highly crystallized melt inclusions in section is that it is not possible to distinguish real differences in inclusion size and/or composition from apparent differences due to unrepresentative sectioning (see S3). The only reasonable approach to this problem is to analyze as many inclusions as possible. In this regard, Nakhla is a rare and fortunate case among meteorites because its total mass is large (approximately 10 kg) and many sections are available. The number of melt inclusions analyzed in this study (35) is far more than are often available in studies of Martian meteorite melt inclusions.

The 35 melt inclusions we analyzed show a large range in apparent size (25–450 μm diameter) that is unlikely to be due entirely to unrepresentative sectioning. For example, if all the melt inclusions in NK12A were actually as large as NK12A1 in the third dimension, they would occupy nearly all the volume of the olivine grain. Furthermore, there appear to be significant differences in the texture and phase assemblages of the large “complex” inclusions compared with the smaller “simple” inclusions. Thus, it is conceivable that these two groups constitute distinct populations, representing two different (although probably related) PTLs. Based on the apparent correlations of inclusion size with K2O and Na2O content (Figs. 8a and 8b), the “small simple” population (average K2O = 1.2 wt%) could represent the earliest melt, while the “large complex” population (average K2O = 2.0 wt%) represents a later stage of magmatic evolution of the same melt. However, magmatic evolution leading to higher alkali contents should also lead to higher SiO2/(FeO + MgO) ratio, which is the opposite of what is observed in the two averages (Table 7), so it is not clear that average bulk compositions of the two groups support this interpretation.

Additional clues as to which inclusions were trapped earliest may come from the P X-ray maps. As discussed above, the P X-ray map of olivine grain NK19A (Fig. 3b) shows a P-rich core with skeletal hopper morphology. Four small melt inclusions are located in the P-rich core, and the central open cavity (re-entrant) is now occupied by the large melt inclusion NK19A1. This would imply that the small melt inclusions were trapped earlier than the large one, a sequence that is intriguingly the opposite of what one would infer by looking only at a BEI of this grain. The P X-ray map of olivine grain NK12A (Fig. 2b) is more difficult to interpret. Again, as discussed above, this P map shows a P-rich skeletal core, in this case with swallowtail morphology. All eight of the inclusions in this grain are now located in olivine that would have crystallized as later infilling of the swallowtail core's wings. Based on the concentric zoning of this olivine, one would conclude that the inner, larger inclusions were trapped before the small outer ones. On the other hand, it is possible that some or all of the inclusions migrated within the crystal after they were trapped (Sonzogni et al. 2011). Milman-Barris et al. (2008) interpreted the large size and asymmetric shapes of P-poor zones surrounding inclusions in olivine to be a result of a dissolution-replacement process during inclusion migration in a thermal gradient. If this is the case, then some of the smaller inclusions may originally have been located in the swallowtail wings, and thus been trapped earlier than large ones located in the infilled material. On the whole, evidence from the P X-ray maps does not provide a clear answer as to which (if either) population of inclusions was trapped earliest.

The alternative interpretation is that all the observed melt inclusions represent a single PTL, i.e., one magma trapped as multiple, variously sized inclusions at approximately the same stage of evolution. In this case, the difference in texture and phase assemblage between the large and small inclusions could be explained by differences in disequilibrium crystallization sequence in large versus small closed volumes (e.g., nucleation effects), rather than by differences in bulk composition. We cannot explain the apparent correlation of inclusion size with K2O content (Fig. 8a), but the correlation is weak and possibly spurious. Given these many uncertainties, we conclude that there is no compelling evidence that one of the subpopulations is earlier than the other and will use the weighted average of all 17 inclusions (Tables 7 and 8) as the best estimate of the composition of the earliest melt trapped in Nakhla olivine.

Reconstruction of the Composition of the Primary Trapped Liquid in Olivine

The composition of the PTL can be reconstructed from the weighted average pbc (Tables 7 and 8) by re-addition of wall olivine that crystallized from the trapped melt, followed by adjustment of the Fe/Mg ratio of the resulting composition to account for post-entrapment Fe/Mg re-equilibration of the inclusions with their hosts. The mechanism and effects of Fe/Mg re-equilibration of melt inclusions with olivine hosts are well understood (Danyushevsky et al. 2000; Goodrich et al. 2009; S3), and melt inclusions in Nakhla show typical signs of such re-equilibration: wall olivine is homogeneous, with Fe/Mg ratio identical to that of host olivine; daughter pyroxenes are likewise homogeneous, with Fe/Mg ratio very similar to that of the olivine (Treiman 1993); residual glasses have very low concentrations of FeO and MgO (Danyushevsky et al. 2000); and the average bulk inclusion composition has an Fe/Mg ratio (Table 7) that is far too low to be in equilibrium with the olivine. The consequence of re-equilibration for retrieving the PTL is that it becomes impossible to determine the amount of wall olivine that crystallized in the inclusions without an independent estimate of either the FeO content or the olivine component of the PTL (Sobolev and Danyushevsky 1994; Danyushevsky et al. 2000, 2002a, 2002b). Neither of these values can be determined from an equilibrated cumulate. In such cases, we can obtain a minimum estimate of the olivine component of the PTL from the requirement that the PTL be saturated with olivine (i.e., cosaturated with olivine + pyroxene). There is no way to know how much olivine crystallized from the PTL before it reached olivine-pyroxene cosaturation, but this limitation is not restricted to melt inclusion studies and pertains to any technique for determining the parent magma of an equilibrated cumulate (Longhi and Pan 1989). An additional factor in the case of Nakhla is that the olivine hosts themselves may have re-equilibrated with the external magma. As recognized by previous workers (Treiman 1986, 1993; Longhi and Pan 1989), Nakhla's olivine (core Fo of approximately 40) is far too FeO-rich to be in equilibrium with the cores of its augite (mg# approximately 62–63). This has been attributed to re-equilibration of the olivine with late melt, with the inference that it was originally Fo approximately 53, in Fe/Mg equilibrium with the augite cores (Longhi and Pan 1989).

In reconstructing the composition of the PTL from the weighted average pbc, we used the MELTS program (Ghiorso and Sack 1995) to determine the minimum amount of wall olivine that had to be added to achieve olivine-pyroxene cosaturation. Calculations were run at oxygen fugacities (fO2) of approximately QFM to QFM-1, and proceeded by trial and error, adding olivine to the PTL and checking the liquidus phases until olivine saturation was achieved. Adding the amount of olivine determined in this manner (approximately 20 mole%), and adjusting Fe/Mg ratio for equilibrium with olivine of Fo 53 (DFe/Mg[ol/liq] = 0.35; Filiberto and Dasgupta 2011), yields the PTL composition given in column 1 of Table 10. Carrying out the same exercise for the recalculated (Table 8) weighted average bulk inclusion composition based on the data of Treiman (1993) yields the PTL composition given in column 3 of Table 10 (designated T93, to distinguish it from the original NK93 of Treiman 1993). The two PTLs are very similar in most respects. Focusing on the alkalis, the PTL obtained in this work has lower K2O (1.8 versus 2.3 wt%) and, to a lesser extent, Na2O (0.9 versus 1.0 wt%) than that based on Treiman (1993), resulting in a slightly lower K2O/Na2O ratio (2.0 versus 2.2 wt%). Nevertheless, the new K2O content is approximately 4.5–7 × higher, and the new K2O/Na2O ratio is approximately 5–6 × higher, than those of Nakhla parent magma compositions obtained by other methods (Table 10). This cannot be due to alteration, because we eliminated highly and even moderately altered inclusions from consideration. And it is unlikely to be due to unrepresentative sectioning, because almost all 35 melt inclusions that we studied have high K2O content. We conclude that the melt trapped in olivine in Nakhla (PTLoliv) really was unusually K2O-rich, by Martian standards.

Table 10. Reconstructed composition of trapped melts in olivine (PTLoliv) and augite (PTLaug) in Nakhla, compared with proposed nakhlite parent magma compositions.Thumbnail image of

Is the High K2O Content of the PTL in Olivine Due to Boundary Layer Effects?

Although high K2O content and high K2O/Na2O ratio appear to be real characteristics of the melt trapped in olivine in Nakhla, it is still possible that they are due to boundary layer effects and therefore are not characteristics of the external body of magma from which the olivine was crystallizing. Melt inclusions formed during rapid crystal growth (e.g., skeletal, hopper, dendritic shapes) are likely to trap large volumes of the boundary layer, while those formed during slower crystal growth will not (Faure and Schiano 2005). The P X-ray maps of olivine grains NK12A and NK19A (Figs. 2 and 3) can be used to indicate crystal growth style during trapping of the melt inclusions in these grains, and therefore assess the possibility of boundary layer effects. For example, the skeletal hopper morphology of the relict P-rich core in NK19A (Fig. 3) is a strong indicator of rapid crystal growth (Donaldson 1976), with the large, K2O-rich melt inclusion NK19A1 representing an embayment of melt that was subsequently closed by overgrowth. This type of melt inclusion might exhibit strong boundary layer effects (Faure and Schiano 2005). However, studies of interdiffusion in mixing of silicate melts suggest that alkalis diffuse orders of magnitude faster than other cations, such as Al, Ca, and Ti (Baker 1992). If this is applicable to melt inclusions, then boundary layer effects should result in the Nakhla inclusions being depleted in both K2O and Na2O, and enriched in CaO and Al2O3 relative to the far-field liquid, which implies that boundary layer effects cannot explain the high K2O content of the Nakhla inclusions. On the other hand, Milman-Barris et al. (2008) observed boundary layers adjacent to experimentally rapidly grown (hopper, dendritic) olivines that were enriched by up to approximately 30% in K2O and/or Na2O. This suggests that further work (including experimental) may be needed to quantify and understand the processes responsible for boundary layer effects during trapping of melt inclusions. Nevertheless, a back-of-the envelope calculation demonstrates that even boundary layer enrichments of K2O by 30% would be grossly inadequate to account for the difference between the bulk K2O content of PTLoliv and those of NPMs inferred by other methods. Approximating the inclusion NK19A1 as a sphere of radius 120 μm, with a K2O content of 0.3 wt%, a boundary layer that was enriched in K2O by 30% would need to be thicker than the radius of the inclusion to produce an apparent bulk K2O content of 1.8% in the trapped melt. Moreover, K2O-enriched boundary layers would have a larger effect on the bulk compositions of small inclusions compared with large ones, so that smaller inclusions should show higher K2O. The data (Fig. 8a) do not show such a trend; if they show any trend at all, it is the opposite (also see Sonzogni and Treiman 2013). We conclude that boundary layer effects are not responsible for the high K2O content of PTLoliv. High K2O content was therefore a property of the magma from which the olivine cores crystallized. In the following sections, we consider whether this magma was the Nakhla parent magma.

Comparison of the PTL in Olivine with Proposed Nakhla Parent Magmas

If Nakhla crystallized as a closed system, the bulk composition of the rock minus that of the cumulus minerals should give the composition of the intercumulus melt with which the cumulus minerals were in equilibrium (see S2). This melt can then be equated with the NPM. It follows that a simple mass balance calculation can be used to test proposed NPMs. Results of such a calculation, compared with the most viable previously proposed NPMs (S2), are shown in Fig. 11 on a plot of MgO versus K2O. This plot shows that PTLoliv (or, obviously, T93) could not be obtained by subtracting augite, olivine, or any combination of the two minerals from the bulk composition of Nakhla, because their K2O contents are too high. The same is true of NK3 (Harvey and McSween 1992), which was also derived from melt inclusions in olivine using a different method (S2).

Figure 11.

Plot of MgO versus K2O content of bulk nakhlites, proposed Nakhla parent magma compositions, and primary trapped melts represented by melt inclusions in olivine (PTLoliv) and augite (PTLaug) cores from this work. Trends from Nakhla bulk composition indicate subtraction of pure augite and olivine of core compositions. Nakhlite bulk compositions cited in Treiman (2005). Nakhla parent magma compositions from following sources: T93 (Treiman 1993; revised in this work); NK3 (Harvey and McSween 1992); GH6 (Varela et al. 2001); NA03 (Stockstill et al. 2005); NPM-05 (Sautter et al. 2012); B′ (Treiman 1986, 1993); N′ (Longhi and Pan 1989; Treiman 1993).

In contrast, the proposed NPM compositions B′, N′, GH6, NA03, and NPM-05 all have significantly lower K2O contents, which would be consistent with subtraction of some combination of olivine and augite from the bulk rock composition. In the case of B′, this observation is trivial, because B′ was calculated by exactly this procedure (Treiman 1986, 1993; see S3). The same is true of N′, because its K2O content was based on the bulk rock K2O/Na2O ratio (Longhi and Pan 1989).

More important is that the three proposed NPM compositions derived from melt inclusions in augite, GH6 (Varela et al. 2001), NA03 (Stockstill et al. 2005), and NPM-05 (Sautter et al. 2012), have K2O contents 3–6 × lower than what we infer from melt inclusions in olivine. In principle, the melt inclusions in augite should offer the least model-dependent means for deriving the composition of the Nakhla parent magma, because augite is the only undisputed cumulus phase in the rock and because melt inclusions in augite are less prone to post-entrapment re-equilibration of Fe/Mg with their hosts (Danyushevsky et al. 2000, 2002b; Gaetani and Watson 2000). All three studies of melt inclusions in augite that have been conducted so far have used experimental rehomogenization to retrieve the composition of the PTL (Varela et al. 2001; Stockstill et al. 2005; Sautter et al. 2012). In the following section, we use our observations of melt inclusions in augite to reconstruct a PTL composition using analytical methods instead, for comparison with the results of experimental rehomogenization.

Melt Inclusions in Augite and Reconstruction of the PTL in Augite

Our observations of melt inclusions in augite suggest that they constitute two populations—inclusions in grain cores and inclusions in grain rims. The inclusions in cores appear to be identical to type 1a inclusions identified by Sautter et al. (2012). The homogeneity of the augite cores suggests that all melt inclusions in cores were trapped at approximately the same early stage of magmatic evolution. The absence of SiO2-K2O (Fig. 10b) or SiO2-Na2O trends among the glasses supports this interpretation. In contrast, the observed SiO2-Al2O3 trend (Fig. 10b) could suggest trapping over a period of magmatic evolution. However, this trend is essentially an augite-control trend, and could be due to different degrees of crystallization of augite within the inclusions rather than external magmatic evolution (assuming that augite was the only liquidus phase). The observation that the two populations of glasses (those in cores and those in rims) completely overlap one another in both SiO2 and Al2O3 content, and that glasses in cores span the entire range (Fig. 10b), further suggests that the trend is not due to external magmatic evolution. Thus, we interpret the inclusions in cores to represent a single trapped melt that provides the best sample of the parent magma. The melt inclusions in rims appear to have been trapped at a later stage of magmatic evolution (again, all at approximately the same time), in response to the change of conditions that led to overgrowth of the rims onto cumulus cores (i.e., fractional crystallization from local intercumulus melt rather than equilibrium crystallization from the larger body of magma).

The melt inclusions in augite cores show simpler phase assemblages and textures compared with the melt inclusions in olivine, making spread beam analyses or grid analyses unnecessary (see S3). A reasonable estimate of the PTL represented by the inclusions in augite cores can be reconstructed from compositions of the observed phases in the inclusions, combined with estimated modal abundances. The inclusions that we studied show evidence of Fe/Mg re-equilibration with their hosts—FeO and MgO concentrations in the glass are very low, and the halo augite (average mg# approximately 46), which we interpret to be the wall augite, is relatively magnesian. The re-equilibration was not complete, as it was in the melt inclusions in olivine (if it had been, wall augite would be mg# 62, the same as core augite), which is consistent with slower Fe-Mg interdiffusion in pyroxenes (Gaetani and Watson 2000). Nevertheless, none of the inclusions that we studied are unequilibrated.

Observed proportions of the two major phases in the inclusions, halo augite and glass, range from approximately 40:60 to approximately 75:25 pyroxene:glass by volume (areal proportions corrected for concentric structure). However, preliminary calculations using MELTS constrain the amount of halo augite to be ≥50% (from the requirement that the PTL be augite-saturated), and furthermore suggest that higher values are more likely (because the lower values lead to a dominance of pigeonite over augite). We use a range from 60:40 to 70:30 pyroxene:glass, and average compositions for both the glass and the wall augite. This is justified for the glass by the lack of compositional trends among the glasses. The wall augite crystals are probably zoned to various degrees in Fe/Mg ratio, but they are too small to obtain analyses documenting this zonation. The analyses that were obtained were centered midway in the crystals (between glass and surrounding core augite) and encompassed most of their width, and thus can be taken as a good approximation to the integrated composition of the zoned crystals. The proportion of pigeonite in the inclusions is uncertain, but relatively small (estimated approximately 5 vol%), and this uncertainty will not have a large effect on the obtained bulk compositions. Proportions of the opaque minerals are very likely minor, based on BEI observations, on the order of 1–2%. However, we note that the presence of these minerals can have a significant effect on the bulk composition because they are all major carriers of FeO, hercynite is additionally a major carrier of Al2O3, and all 3 phases are SiO2-free. In the inclusions in which opaque grains are large, relative to the total volume of the inclusion (e.g., Fig. 9f), it is likely that the opaques are cognate phases, trapped along with melt, rather than grains that crystallized from the melt itself. The amount of Fe-Ti oxide can be reasonably constrained by the requirement that TiO2 content not be too high (≤1%). Given these considerations, we calculate a range of possible PTL compositions as 70:20 to 50:40 Fe-Al halo augite:glass, plus 5% pigeonite, 1% Ti-magnetite, 2% pyrrhotite, and 2% hercynite. For pigeonite, we use the composition from Varela et al. (2001) from inclusions in augite in Nakhla (their table 2, column 4). For pyrrhotite, Ti-magnetite, and hercynite, we use idealized compositions (note that S is not included in the PTL as it was not reported in any of the experimental rehomogenization studies). The resulting range of compositions (PTLaug) is given in Table 10, columns 4–5. We note that these compositions have an mg# (41) that is too magnesian to be in equilibrium with core composition augite, confirming our inference that the inclusions have at least partially re-equilibrated Fe/Mg with their hosts. They are adjusted for equilibrium with augite of mg# 62 in columns 6–7 of Table 10, using KD(Fe/Mg)aug/liq = 0.23 (from the experiments of Longhi and Pan 1989 for a Nakhla-like bulk composition).

In Table 10 and Fig. 12, this range of PTLaug is compared with the three proposed NPM compositions that were based on experimental rehomogenization of melt inclusions in augite, GH6 (Varela et al. 2001), NA03 (Stockstill et al. 2005), and NPM-05 (Sautter et al. 2012). The latter were not adjusted for Fe/Mg re-equilibration because Stockstill et al. (2005) and Sautter et al. (2012) assumed that the inclusions had not experienced Fe/Mg re-equilibration. Therefore, we use our unadjusted compositions (Table 10, column 4–5) in this comparison. Our estimated PTLaug have notably higher SiO2, lower FeO, and higher K2O contents (as well as higher mg# and K2O/Na2O ratio) than NA03 (Stockstill et al. 2005) and NPM-05 (Sautter et al. 2012), but in all these respects are similar to GH6 (Varela et al. 2001). GH6 was derived as the average composition for all (4) inclusions that were rehomogenized by Varela et al. (2001). However, both NA03 and NPM-05 represent single rehomogenized melt inclusions that were selected by Stockstill et al. (2005) and Sautter et al. (2012) as their “preferred” NPM compositions, based on various criteria. All rehomogenized compositions of these two authors (eight from Stockstill et al. 2005 and five from Sautter et al. 2012) are shown in Fig. 12. Our PTLaug are similar to several of the rehomogenized inclusions of Stockstill et al. (2005), as well as several of the individual rehomogenized inclusions of Varela et al. (2001).

Figure 12.

Plots of wt% SiO2 versus wt% Al2O3 a), wt% FeO b), mg# c), and wt% K2O d) for composition of primary trapped melt represented by melt inclusions in augite cores from this work (PTLaug), compared with compositions of rehomogenized melt inclusions in augite cores from Varela et al. (2001), Stockstill et al. (2005), and Sautter et al. (2012). Also shown are compositions of glasses and halo augite in inclusions (this work), pigeonite in inclusions (Varela et al. 2001), and core augite (this work).

Of particular note are the low SiO2 contents of NA03 and NPM-05 (approximately 47 and 49 wt%, respectively), compared with PTLaug (approximately 53–56 wt%) and with most of the other rehomogenized compositions from all three studies. The two major components of the inclusions are glass with approximately 72 wt% SiO2, and pyroxene with approximately 50 wt% SiO2 (Fig. 12). Thus, bulk inclusion compositions could have SiO2 contents less than 50% only by virtue of containing some of the minor, SiO2-free phases Fe-Ti oxide, hercynite, and/or pyrrhotite. Our calculated compositions show that, unless pyroxene constitutes virtually 100% of the inclusion (obviously impossible), unreasonably high contents of these phases would be required to lower SiO2 to the low values in NA03 and NPM-05. We suggest that these inclusions contained large cognate grains, such as those we observed in some inclusions (Fig. 9f), of one or more of these minor phases. These grains did not crystallize from the trapped melt and therefore should not have been included in the rehomogenized composition, but their presence was undetected and so they were included. If this is the case, it would simultaneously explain the higher FeO contents (and lower mg#) of NA03 and NPM-05 compared with PTLaug and most of the other rehomogenized compositions from all previous studies (Figs. 12b and 12c). Again, the compositions of the two major phases in the inclusions, glass and pyroxene, would indicate that the bulk silicate composition could not have FeO > ~20 wt% (the average FeO content of the halo augite). However, all of the minor phases are major carriers of FeO, and their presence in excess amounts in rehomogenized inclusions could have a large effect. Excess Fe-Ti oxide could be detected in unreasonably high TiO2 contents (which NA03 and NPM-05 do not appear to have), but excess pyrrhotite would be undetectable compositionally if S was not measured. Excess hercynite might be detectable in extremely high Al2O3, but as both the glass and the halo augite contain significant Al2O3, moderate excesses might go unnoticed. As most of the opaques we observed in the inclusions were pyrrhotite, it seems likely that this phase was the culprit. NA03 and NPM-05 were “preferred” by Stockstill et al. (2005) and Sautter et al. (2012) largely because of their low SiO2 and high FeO. The low SiO2 content was taken as an indication of the most primitive composition, and the high FeO was taken as an indication of minimal (or no) Fe/Mg re-equilibration (note that the actual compositions of the phases in these rehomogenized inclusions cannot be determined, as the technique requires that the inclusions be under the surface of the sample). However, we suggest that both of these compositional features are unrepresentative of the trapped melt, and that more representative compositions are given by the rehomogenized compositions with higher SiO2 (GH6 of Varela et al. [2001] and most of the compositions of Stockstill et al. 2005) and our calculated PTLaug, after correction of Fe/Mg ratio for partial re-equilibration.

The K2O content of NA03 is only 0.39 wt% (this low value could also be partially due to dilution with excess minor phases), but the majority of the rehomogenized compositions of Stockstill et al. (2005) have higher K2O (0.33–3.6 wt%). The three highest of these values (2.0–3.7%) are for inclusions that have 56–64% SiO2, which suggests that an insufficient amount of wall augite was homogenized (i.e., glass:pyroxene ratio was too high; see S2). However, even eliminating these three inclusions, several of their compositions are similar to PTLaug in SiO2 content and have K2O contents of approximately 0.7–0.8 wt%. Sautter et al. (2012) selected NPM-05 as their proposed NPM partly because its low K2O content (0.32 wt%) was similar to that of NA03. However, they also had higher K2O contents up to approximately 0.8 wt% in some of their rehomogenized inclusions. Three of the four rehomogenized compositions of Varela et al. (2001) have K2O contents of 0.8–0.9 wt%, which are within the range of values in PTLaug (Table 10). Thus, we argue that the average K2O content of the melt trapped in augite cores in Nakhla was significantly higher than suggested by NA03 (Stockstill et al. 2005) or NPM-05 (Sautter et al. 2012), and was probably in the range approximately 0.8–1.1 wt%. A further indication that high K2O is a general feature of nakhlite parent magmas is that the PTL derived by Imae and Ikeda (2007) for five melt inclusions in augite in MIL 03346 (see S2) has 1.2 wt% K2O (Table 10, column 15).

Relationship between PTLoliv and PTLaug

Our estimates of the composition of PTLaug (Table 10, column 6–7) are not olivine-saturated, and, based on MELTS calculations, do not evolve to early olivine saturation. This result is in agreement with Varela et al. (2001), who noted that GH6 was silica-normative and therefore could not have been in equilibrium with olivine. Sautter et al. (2012) also found that NPM-05 was not olivine-saturated according to calculations with MELTS and PETROLOG (Danyushevsky and Plekhov 2011), despite its lower SiO2 content. Stockstill et al. (2005) stated that NA03 was olivine-augite cosaturated, and used this as one argument supporting their choice of this rehomogenized composition as the most representative of the parent magma. However, our MELTS calculations at QFM and QFM-1 indicate that NA03 is only saturated with augite.

In contrast, PTLoliv was clearly saturated with olivine, and either olivine-augite cosaturated or close to it (i.e., it evolved to olivine-augite cosaturation after some crystallization of olivine). Thus, PTLaug cannot evolve to PTLoliv (or vice versa). This can also be seen in the fact that PTLaug has lower MgO content than PTLoliv (Table 10; Fig. 11), despite having, in principle, the same mg#. PTLoliv was reconstructed under the assumption that Nakhla's olivine cores were originally in equilibrium with its augite cores, i.e., they were originally Fo approximately 53 and later re-equilibrated to their present Fo approximately 40 (Longhi and Pan 1989). However, the composition of PTLaug argues against this. One possibility is that the olivine cores have retained their original composition and PTLoliv should have been calculated to be in equilibrium with Fo approximately 40 (as carried out in Table 10, column 2). In that case, could the olivine have crystallized late from PTLaug? This is unlikely, because, although PTLaug does crystallize late olivine (according to MELTS), that olivine is significantly more ferroan (Fo approximately 20–23) than the olivine cores.

Furthermore, although the K2O content of our estimated PTLaug is significantly higher than that of NA03 (Stockstill et al. 2005), NPM-05 (Sautter et al. 2012), or even GH6 (Varela et al. 2001), it is still lower than that of PTLoliv, and cannot evolve to that of PTLoliv even if the latter was in equilibrium with Fo 40 (Fig. 11). Thus, PTLaug and PTLoliv cannot be related by fractional crystallization, and if PTLaug is considered to be the NPM (because Nakhla is defined as an augite cumulate), then PTLoliv is not the NPM.

This leads to the conclusion that the olivine cores are xenocrystic, either entrained from surrounding wall rock or added as phenocrysts in an admixed magma. Treiman (1986) originally suggested that this was the case, but his argument was based only on the observation that the present Fo 40 composition of the olivine is not in equilibrium with the mg# 62 augite cores. Later, he (and most other workers) rejected this interpretation under the assumption that the olivine had originally been Fo 53 (Longhi and Pan 1989). In fact, if the olivine cores did not cocrystallize with the augite cores, as we now argue, then we have no actual constraint on the mg# (equilibrium Fo value) of PTLoliv other than that its equilibrium Fo was ≥40.

The conclusion that the olivine is xenocrystic was also reached by Sautter et al. (2012). However, in contrast to their interpretation, we suggest that PTLoliv and PTLaug were in some way related. The main basis for this conclusion is that both melts are notably K2O-rich (and in this regard, distinguished from typical shergottite-like melts). It is also supported by the common occurrence of rounded grains of augite within olivine cores. These augite inclusions are homogeneous and identical in composition to Nakhla primary augite cores. They could not have become encased in solid olivine crystals after the olivine was added to the Nakhla magma. Therefore, they must have been present in the incorporated olivine grains. Along the same lines, PTLoliv and PTLaug have similarly high CaO contents (Table 10) and both crystallize augite in preference to low-Ca pyroxene, which among the Martian meteorites is a unique feature of nakhlites. Thus, we suggest that PTLoliv and PTLaug were generated from the same source region, within a short time of one another. PTLoliv was generated first and emplaced to form olivine-rich cumulate rocks. Shortly thereafter, PTLaug was generated and ascended through these olivine-rich cumulates, incorporating fragments of wallrock that became the xenocrystic olivine cores in Nakhla. In the final section below, we discuss a possible scenario for the generation of these two magmas from the same source region.

Extension to Other Nakhlites

The conclusion that the Nakhla parent magma was not saturated with olivine can be extended to all nakhlites. Nakhla, Lafayette, Governador Valadares, NWA 998, the Yamato nakhlites, and (to a slightly lesser degree) NWA 817 are so similar to Nakhla in modal abundances, mineral compositions, bulk rock composition, melt inclusion properties, and degree of equilibration (Table S1), that they are likely to have had the same petrogenesis as Nakhla. It was already concluded by Imae and Ikeda (2007) that the olivine in MIL 03346 was not cumulus, based on its low abundance and textures. And NWA 5790, which has been inferred to be much less equilibrated than the other nakhlites (Jambon et al. 2010), has olivine core compositions very similar to the others (Table S1). If these cores are not re-equilibrated, then they almost certainly are xenocrystic. Thus, we would argue that the parent magmas of all nakhlites were saturated only with augite. This conclusion is consistent with, but does not require that they have, a common parent magma.

The conclusion that the olivine in nakhlites is xenocrystic also reopens the question of whether the Nakhla parent magma solidified as a closed system. Treiman (1993) concluded that if NK93 (with its very high K2O content) accurately represented the Nakhla parent magma, then Nakhla must have lost a late, fractionated (and K2O-rich) melt. Based on our new composition for the Nakhla parent magma (PTLaug), it might appear that loss of late melt is no longer required, as subtraction of only augite from the bulk rock can lead to the MgO-K2O composition of PTLaug (Fig. 11). However, this interpretation is not correct, because the bulk composition of the rock includes the olivine cores. As discussed in S2, this olivine represents an addition to the parent magma (PTLaug) regardless of whether it is cumulus or xenolithic. Therefore, in a mass balance calculation it must be subtracted, which returns us to the conclusion (Fig. 11) that Nakhla (and the similar nakhlites) has lost a late melt component. This is hardly surprising, given that these nakhlites most likely represent the lower levels of a cumulate pile. Similarly, NWA 5790, which has been suggested to represent the top of the cumulate pile (Jambon et al. 2010), may contain an excess of fractionated melt relative to the parent magma of the whole sequence.

Generation of K2O-rich Melt and Implications for Differentiation of the Martian Mantle

The melt compositions derived in this work, PTLaug and PTLoliv, are distinctive in their alkali element properties. As shown on a TAS (total alkalis versus SiO2) diagram (Fig. 13a), their total Na2O + K2O contents are higher than those of primitive shergottite magmas (Greshake et al. 2004; Gross et al. 2011; Filiberto et al. 2012), but within the range of those of basalts analyzed at Gusev crater on Mars by the Martian Exploration Rover Spirit (McSween et al. 2009). However, it is not their total alkalis that make these compositions so unusual, it is their K2O/Na2O ratios (Fig. 13b), which are much higher than those of any primitive Martian meteorite magma, the Gusev basalts, or even the highly alkaline Jake_M basalt analyzed by the Mars Science Laboratory (MSL) mission (Stolper et al. 2013). Only a few unusual samples like Bathurst_Inlet analyzed by MSL have K2O/Na2O approaching unity (Schmidt et al. 2013), as in PTLaug, and no known Martian samples have K2O/Na2O ≈ 2, as in PTLoliv (Table 10).

Figure 13.

a) Total alkalis versus silica (TAS) diagram, showing PTLol (yellow star) and PTLaug (yellow box) obtained in this study compared with primitive shergottite magmas (solid cyan boxes), midocean ridge basalts (MORB; solid black boxes), Gusev crater basalts (open cyan boxes) analyzed by the Mars Exploration Rover Spirit, and the rock Jake_M (open blue box with X) analyzed by Mars Science Laboratory. Solid magenta boxes are terrestrial ferrobasalts. Black curve is tholeiitic fractionation trend. Magenta curve is potassic silica-undersaturated alkalic trend (Whitaker et al. 2007). The dashed black line is the division between alkalic and subalkalic compositions. b) K2O versus Na2O for PTLol and PTLaug obtained in this study compared with primitive shergottite magmas, MORB, Gusev crater basalts, Jake_M, and terrestrial ferrobasalts with 45–54 wt% SiO2. Symbols and colors as in (a). Data for bulk compositions of primitive shergottites Y-980459 from Greshake et al. (2004) and Misawa (2004), NWA 5789 from Gross et al. (2011) and Irving et al. (2010), and NWA 6234 from Filiberto et al. (2012); Gusev basalts (renormalized on a volatile-free basis) from Squyres et al. (2007), Gellert et al. (2006), and Ming et al. (2008); Jake_M from Stolper et al. (2013); terrestrial ferrobasalts from Leeman et al. (1976) and Mitchell et al. (1996).

These enrichments in K2O over Na2O in primary nakhlite magmas raise the question of what mantle compositions and magmatic processes could generate such magmas. Mars is inferred to be richer in alkali elements (and other volatiles) than is the Earth (Wänke and Dreibus 1988), and so might be expected to generate more alkaline magmas (Dunn et al. 2007; Nekvasil et al. 2007). On Earth, production of alkali-rich basaltic magmas is commonly ascribed to one of three processes: melting of crustal materials recycled through plate tectonics (e.g., eclogite) ± mantle peridotite (e.g., Kogiso et al. 2003; Keshav et al. 2004); high-pressure melting of normal mantle lherzolite with high volatile contents (e.g., Dasgupta et al. 2007); and/or melting of metasomatized mantle, where the metasomatic fluid arises as very low degree partial melt of (or volatile fluxes from) volatile-bearing lherzolite (e.g., Pilet et al. 2008). For Mars, we can also consider the assimilation of an incompatible element–rich component that arose as the last fractionate of a magma ocean (e.g., Elkins-Tanton 2012; Mezger et al. 2013). This component would be similar to lunar KREEP (e.g., Jones 2003), but different in that it does not contain much phosphorus (Treiman 2003).

The first process, melting of recycled crustal material, is probably not important for Mars in the absence of plate tectonics. Crustal material could be recycled into the Martian mantle by delamination or by convective overturn from the base of the crust (e.g., Elkins-Tanton 2005; Elkins-Tanton et al. 2005), but that is likely to be minor. The second process, melting of volatile-bearing “normal” lherzolite, is also unlikely because it cannot produce magmas sufficiently rich in K2O (Pilet et al. 2008). That leaves mantle metasomatism and assimilation of “KREE(P)” as most likely to be relevant to the nakhlite source region. Mantle metasomatism has been postulated for Mars based on trace element chemistry of the shergottites and analogy with the Earth (e.g., Blichert-Toft et al. 1999; Treiman 2003). And assimilation of late magma ocean differentiates has been suggested based on radiogenic isotope chemistry and analogy with the Moon (Borg and Draper 2003; Jones 2003; Debaille et al. 2008; Greenough and Ya'acoby 2013).

The Sm-Nd radioisotope systematics of the nakhlites suggests that mantle metasomatism is the most likely process leading to their K-enrichment. The nakhlites are significantly enriched in the light rare earth elements (LREE), including Nd (Treiman 2005 and references therein), and the geochemical behavior of Nd and K is sufficiently similar in basaltic Martian meteorites (Treiman et al. 1986; Treiman 2003) that enrichment/depletion of Nd can stand as a proxy for K. Thus, the inferred enrichment in K in the nakhlite magmas is, in general, consistent with enrichment in Nd and other LREE. Results from the short-lived 146Sm-142Nd and long-lived 147Sm-143Nd radioisotope systems show that the source region of the Nakhla (and other nakhlite) parent magma was “…modified through the addition of a LREE-enriched component,” which was “…derived from an ancient LREE-depleted source” (Borg et al. 2003). A late fractionate from a magma ocean would have been LREE-enriched, but would not have been derived from a LREE-depleted source (similar to the case of lunar KREEP: Warren and Wasson 1979). Thus, the crustal assimilation model appears to be ruled out. However, the Sm-Nd data are entirely consistent with mantle metasomatism of the Nakhla source region shortly before melt generation: “A possible source of this component is an incompatible element–enriched fluid or melt derived from a depleted mantle source” (Borg et al. 2003). Such melts could be generated by very low degrees of melting, which would strongly concentrate whatever incompatible elements are available. Significant enrichments in incompatible elements are seen in metasomatized samples of the Earth's mantle, and have been ascribed to reactive transport of such fluids (e.g., Navon and Stolper 1987; Arai et al. 1997; Reiners 1998). Furthermore, enrichment in K2O over Na2O is observed. Potassic metasomatism resulting in Na2O/K2O fractionation occurs in the Earth's mantle, as evidenced by xenoliths with veins rich in biotite and K-amphibole (O'Reilly and Griffin 2013) and by potassic basalts (Elkins-Tanton and Grove 2003). For example, a potassic metasomatic fluid could be mobilized from eclogite (e.g., pressure > ~1.5 GPa) or plagioclase peridotite (pressure < ~1 GPa), with Na2O being retained in the residue in omphacitic pyroxene or plagioclase, respectively (e.g., Papike et al. 2013), while K2O is concentrated in the melt and fluid (e.g., Keshav et al. 2004).

We suggest that the nakhlite mantle source region was pyroxenitic in composition (rich in augite and containing some olivine), as previously suggested (Treiman 2005 and references therein), and had been enriched in K2O and K2O/Na2O via metasomatism. A high degree of melting of this source produced the silica-poor parent magma PTLoliv. Shortly thereafter, continued ascent and decompression of this source region led to generation of a silica-rich magma, PTLaug. This scenario could explain the mineralogy and major and minor element properties of PTLoliv and PTLaug, and would generate the two melts in the order required by the rocks. Decompression melting of a clinopyroxene-rich lherzolitic mantle source has been proposed to explain the progression from alkalic to subalkalic rocks in Hawaiian volcanism (Chen and Frey 1983; Chen et al. 1991). Alkaline magmas are produced at significantly higher pressures than tholeiitic magmas, but from a similar mantle source (e.g., Green 1970; Chen and Frey 1983). This is caused by a shift in the Di-Fo-An phase boundaries and, at pressures below approximately 4 kbar, the Di-Fo-An join represents a thermal divide between alkaline and tholeiitic basalts (Presnall et al. 1978). For a pyroxenite source (such as we are suggesting for the nakhlite source region), melting studies have shown that a wide range in magma compositions can be produced depending on the percent and pressure of melting (Hirschmann et al. 2003; Lambart et al. 2009). Deeper mantle melting produces alkali-rich, silica-poor magmas and shallow mantle melting, with smaller percents of melting, produces silica-rich and alkali-poor magmas (Hirschmann et al. 2003; Lambart et al. 2009).

Finally, we note that alkali-rich and potassic rocks are common near the landing site of the MSL Curiosity rover in Gale crater, Mars. The rock Jake_M has the composition of a mugearite basalt, with 2.2% K2O and 6.4% Na2O (Stolper et al. 2013). Other rocks in the area, although their parentage is less clear, contain up to 3.5% K2O (Schmidt et al. 2013). Thus, potassium-rich magmas, like those involved in the formation of the nakhlites, represent a part of the recognized diversity of Martian igneous rocks and point to the wide range of mantle compositions and processes that should be expected from the complex evolution of a planet like Mars.

Acknowledgments

We thank Tim McCoy, Linda Welzenbach, Joe Boesenberg, Phillipp Heck, and Carl Agee for assistance with the loan of samples. We thank Tomohiro Usui, Molly McCanta, and the Associate Editor Christine Floss for helpful reviews. This work was supported by a subcontract to C. A. Goodrich (Planetary Science Institute) from NASA grant NNX09AL25G to A. H. Treiman and J. Filiberto. LPI Contribution #1745.

Editorial Handling

Dr. Christine Floss

Ancillary