High concentrations of manganese and sulfur in deposits on Murray Ridge, Endeavour Crater, Mars Open Access

Recent observations by the Opportunity Mars rover on the Cape York rim segment of the Noachian-age 22 km wide Endeavour Crater (Fig. 1) revealed evidence for aqueous mobilization of Zn and precipitation of gypsum in veins (Squyres et al. 2012), together with the formation of phyllosilicate minerals in and along fractures (Arvidson et al. 2014). In parallel, the Curiosity rover in Gale Crater has uncovered morphologic, compositional, and mineralogic evidence for a broadly coeval fluvial-deltaic-lacustrine system and associated diagenetic alteration of sedimentary rocks of basaltic composition (Grotzinger et al. 2014, 2015). These discoveries add to the growing evidence from orbital and landed missions that early Mars supported extensive water-related alteration of crustal materials in both surface and subsurface environments (e.g., Poulet et al. 2005; Ehlmann and Edwards 2014). In this paper we describe measurements acquired by Opportunity while exploring the Murray Ridge rim segment of Endeavour Crater (Figs. 2–4, Table 1). We first provide an overview of Opportunity’s traverses and measurements on Murray Ridge and consider the implications for past aqueous processes based on exposed bedrock measurements. We then focus on rocks named Pinnacle Island and Stuart Island that were serendipitously excavated from a fracture by Opportunity’s wheels. These rocks have unique coatings with the highest Mn and S concentrations found thus far at Meridiani Planum. The data imply aqueous precipitation of sulfate-rich salts, followed by introduction of a strong oxidant that led to precipitation of one or more Mn oxide(s).

Opportunity is a six-wheeled, solar-powered rover (Squyres et al. 2003) equipped with mast-based Pancam multispectral stereo cameras with 13 filters covering the 0.432 to 1.009 μm spectral region (Bell et al. 2003), a robotic arm (instrument deployment device or IDD) with a microscopic imager (MI) that can acquire panchromatic images with 31 μm pixel sizes (Herkenhoff et al. 2003), an alpha particle X-ray spectrometer (APXS) to determine the target chemical compositions (Gellert et al. 2006), and a rock abrasion tool (RAT) to brush loose dust and sand from targets and/or to grind into rocks to remove indurated coatings or weathering rinds (Gorevan et al. 2003). In addition, the rover carries mast-mounted stereo cameras for navigation and terrain context measurements (Navcams), and front and rear body-mounted stereo cameras (Hazcams) used for hazard avoidance during traverses and fine-scale placement of IDD-based instruments onto rock and soil targets. The science instrument payload also includes the Mini-TES thermal emission spectrometer and a Mössbauer spectrometer, but these two instruments were no longer functioning during the time period covered by this paper.

Endeavour is a complex impact crater that is largely buried by younger sulfate-rich sandstones of the Burns formation (Squyres et al. 2012; Crumpler et al. 2015; Grant et al. 2015). The exposed rim is divided into discrete segments separated by relatively low regions covered by Burns formation rocks. The presence of extensive breccia outcrops of the Shoemaker formation on Cape York and Murray Ridge confirms that the crater formed by a bolide impact (Squyres et al. 2012; Arvidson et al. 2014; Crumpler et al. 2015). Comparison to martian impact craters of similar size indicates that only ~100 to 200 m of rim material has been removed by erosion, mostly by fluvial activity before deposition of Burns formation materials (Grant et al. 2015). Thus, exploration and characterization of Endeavour’s rim by Opportunity provide detailed ground truth information about the lithologic nature and extent of alteration by aqueous fluids for the rim of a Noachian-age complex impact crater.

Structural observations of complex terrestrial impact craters similar in size to Endeavour provide insight into the types of fractures expected on the rim segments and surrounding terrains explored by Opportunity. In particular, the Ries Crater, which is ~26 km in diameter (Stöffler et al. 2013), the ~23 km diameter Haughton Crater (Osinski and Spray 2005), and the ~22 km wide Gosses Bluff impact structure (Milton et al. 1972) have been studied in detail and exhibit extensive fractures that formed both radially and concentrically to the crater centers. Osinski and Spray (2005) and Kenkmann et al. (2014) provide schematic views of structural patterns generated during the collapse stage of complex crater formation. Their models, supported by observations, imply that concentric and radial fractures should cut through rim structures of complex craters. These fractures should also propagate upward through deposits generated after the impact event by reactivation during later local- to regional-scale tectonic activity and/or increased stresses associated with loading as later deposits accumulate.

Mars Reconnaissance Orbiter HiRISE images offer 0.25 m/pixel ground resolution, with an imaging system modulation transfer function that preserves fine spatial details of the martian surface (McEwen et al. 2007). Examination of HiRISE images for Endeavour rim and inter-rim segments traversed by Opportunity show evidence for both concentric and radial fractures, including systems that have propagated upward into Burns formation materials. For example, the Burns formation bedrock in Botany Bay hosts fractures that are approximately parallel or perpendicular to the rim segments (Fig. 3). In addition, several fractures are evident extending westward from Murray Ridge into the surrounding Burns formation materials (Fig. 4). As will be shown in the next section of this paper, Opportunity-based images demonstrate that the impact breccia outcrops on Murray Ridge exhibit extensive fracturing that is consistent with the formation of Endeavour Crater and later readjustments.

Opportunity traversed from outcrops of Burns formation sandstones onto the Shoemaker and underlying Matijevic formation rocks (Squyres et al. 2012; Arvidson et al. 2014; Crumpler et al. 2015) exposed on the Cape York rim segment of Endeavour Crater in 2011. The rover explored this rim segment until it was commanded to head south and enter Botany Bay. Opportunity then crossed the Burns formation outcrops in Botany Bay and approached its first outcrop on the northeastern corner of Murray Ridge (Fig. 1). After conducting several measurements on this portion of the rim, Opportunity was commanded to drive around the northern nose of Murray Ridge (Solander Point) and begin a thorough exploration and characterization of its western slopes. The western side of Murray Ridge was chosen because of extensive outcrops and less steep terrain than the interior portion of the Ridge. Opportunity is a solar-powered rover located in the southern hemisphere. As a consequence, the rover is subject to low solar power during the southern winter season, and its activities are more limited than during other seasons. For its sixth winter season Opportunity was directed to Cook Haven, a gentle swale on Murray Ridge (Fig. 4) with requisite north-facing slopes that would provide enough solar power for the rover to survive and gather some science data.

Opportunity’s exploration and characterization of outcrops on Murray Ridge included imaging using Pancam and the engineering cameras, with several stops for measurements of breccia and soil targets using the MI and APXS instruments. Table 1 provides APXS compositions for Murray Ridge targets covered in this paper. Two key stops for remote sensing and compositional measurements to the north of Cook Haven were the Spinifex outcrop target and, farther south, the Moreton Island area, with its Mount Tempest and Tangalooma outcrop targets (Fig. 5). At both sites these breccia outcrops exhibit extensive fractures and contain relatively dark rock clasts several centimeters in diameter embedded in a relatively brighter, fine-grained matrix. The targets were too rough to brush or grind using the RAT.

After leaving Moreton Island, Opportunity traversed into Cook Haven from the south, thereby avoiding the need to cross south-facing slopes and low insolation values associated with the northern portion of this gentle swale. Imaging of Cook Haven shows low-lying, relatively bright outcrops cut by fractures that are partially filled with wind-blown soil deposits (Fig. 6–7). During the downhill traverse into Cook Haven Opportunity stopped on a soil-filled fracture, later executing a 146° turn, with the rear and middle wheels unintentionally excavating both the soil and underlying rocks (Figs. 7–8). Two rocks, subsequently named Pinnacle Island and Stuart Island (PI and SI), were fortuitously excavated and overturned to reveal unusually dark and bright material on the newly exposed rock surfaces. PI slid within reach of Opportunity’s IDD during the drive to the Cape Elizabeth target after completing measurements on the Cape Darby outcrop, requiring no additional rover motions to deploy the MI and APXS onto the newly arrived ~3.5 cm wide rock. Opportunity conducted measurement campaigns on both PI and SI, the soil (named Anchor Point) from which they were excavated, and a loose, relatively dark rock cobble in the vicinity named Sledge Island. Bedrock outcrops Cape Elizabeth and Green Island, both flat outcrops that were brushed before MI and APXS data were acquired (Fig. 8–9), illustrate the nature of outcrops in Cook Haven in that they are relatively bright compared to other Murray Ridge outcrops, with small rock clasts set in an extensive, fine-grained matrix. On the way out of Cook Haven during the ensuing southern hemisphere spring season, Opportunity conducted measurements on one last Cook Haven outcrop, a target named Turnagain Arm, and then exited Cook Haven to continue exploring the western portion of Murray Ridge.

A breccia outcrop named Bristol Well located to the south of Cook Haven exhibited bright veins within fractures (Fig. 10). Three overlapping APXS and MI measurements were acquired to either side and centered on one of the veins. In situ measurements were also made on impact breccia targets to the south of the Bristol Well targets. These targets are the Sarcobatus outcrop matrix and two overlapping measurements on a relatively dark rock clast (Fig. 11). The matrix target was flat and smooth, enabling brushing before acquiring MI and APXS data. The last two breccia targets on Murray Ridge were Tuscaloosa and Sodaville. The former represents outcrop and the latter grus-like debris within a fracture just uphill of Tuscaloosa. After these measurements the rover continued south, conducting its last in situ measurements on a soil target named Barstow, and left Murray Ridge to begin its ascent of the Cape Tribulation rim segment (Fig. 1).

APXS compositional data in Table 1 show that Murray Ridge rocks and soils are basaltic in composition, with the notable exceptions of the Bristol Well vein and the relatively dark and bright materials associated with PI and SI rocks excavated by Opportunity’s wheels. To evaluate further the extent to which there are deviations toward smectite or other phyllosilicate compositions similar to what was found on Cape York (Arvidson et al. 2014), APXS data for outcrops and soils (excluding PI and SI) are shown in a ternary plot of mole fraction Al₂O₃–(CaO+Na₂O+K₂O)–(FeO_T+MgO) (Fig. 12). Also plotted are laboratory-based measurements for various phyllosilicates and other APXS data for Endeavour’s rim. The Espérance data are for APXS measurements at Cape York in a fracture zone where progressively deeper RAT grinds showed evidence for a compositional trend toward montmorillonite (Arvidson et al. 2014). With the exception of the Bristol Well and Sledge Island measurements, no significant deviations from a narrow range of basaltic compositions are evident. Examination of data shown in Table 1 demonstrates that Sledge Island has a slightly different composition as compared to outcrops on Murray Ridge and may be an erratic added to Cook Haven (e.g., as impact ejecta).

The Bristol Well vein is ~1 cm wide and did not fill the field of view of the APXS. Three overlapping in situ measurements were made in a direction perpendicular to the vein in attempt to determine vein composition, using the methodology implemented for measurements over the Homestake vein on Cape York (Squyres et al. 2012). The Bristol Well_2 target was centered over the vein and shows slightly enhanced values of Ca and S and lower values of Fe, Si, Al, and Mg as compared to Bristol Well 1 and 3 targets (Table 1). Similar compositional patterns were found for the Homestake vein that cuts the Grasberg bench deposits surrounding Cape York. Both the Homestake and Bristol Well data, and the Ortiz vein measurements on Matijevic Hill (Cape York) (Arvidson et al. 2014), indicate that sulfate-rich aqueous fluids moved through fractures and precipitated a relatively insoluble Ca sulfate mineral or minerals during one or more episodes.

The Cook Haven bedrock exposures are brighter, relatively flat-lying, and have smaller lithic clasts than other breccia outcrops on Murray Ridge. These outcrops have more S and Cl than the other breccia outcrop targets on the ridge (Fig. 13). This is the case even after brushing to remove loose dust and sand. In addition, the Sarcobatus brushed breccia matrix target shows an enrichment of Cl relative to other Murray Ridge rocks, including the Sarcobatus clast. The increase in S and Cl is interpreted to be due to the slight addition by aqueous processes of sulfate and chloride salts to selected Murray Ridge rocks, primarily to Cook Haven bedrock. The compositional trends shown in Figure 13 also highlight the very high S content of the PI and SI rocks. We consider in detail textures and compositions, together with inferred mineral phases, and explanations for the high S content of these two rocks in the next several sections of this paper.

Pancam false color images, together with MI anaglyphs generated from stereo observations by varying the IDD incoming angle to produce image pairs with parallax (Herkenhoff et al. 2006), show that the PI and SI rocks are thinly coated by dark materials (Figs. 14–17). In addition, PI shows evidence for a thin, bright coating interpreted to lie directly on top of relatively fresh rock surfaces and beneath the dark coating (Fig. 14–17). The dark coating is concentrated in the concave-upward center part of PI and has a lumpy texture. Pancam false color images, combined with MI data (Figs. 16–17), indicate that SI does not exhibit the same type of areally extensive, relatively bright coating found on PI. Instead in the false color images and MI data, SI exhibits an extensive and variably colored dark coating, together with areas interpreted to be relatively fresh or dusty rock surfaces. The former have a bluish-gray and the latter a reddish hue. MI anaglyphs covering SI do show several oval-shaped areas that are interpreted to be rock clasts, with one of them surrounded by a very thin, relatively bright annulus (Fig. 16). The bright annulus, in turn, is surrounded by a relatively dark annulus. Thus both PI and SI exhibit dark coatings, with a bright coating well exposed on PI.

Pancam 13-band multispectral observations (0.432 to 1.009 μm) acquired for PI, SI, and surrounding areas provide quantitative colorimetric and spectral reflectance constraints on coating mineralogy. Pancam raw image data were calibrated to surface radiance factor (I/F, where I is the measured radiance and πF is the incident solar radiance) divided by the cosine of the incidence angle at the time of image acquisition, with absolute reflectance levels accurate to within ~10% (Bell et al. 2006). Spectral end-members for PI were retrieved using the sequential maximum angle convex cone methodology in which spectral extremes are located in multi-dimensional space and separated from shadow values (Gruninger et al. 2004). Four statistically significant end-members were retrieved: relatively fresh rock, dusty rock, bright coating, and dark coating (Fig. 18).

The dusty rock end-member spectrum has a broad ferric edge absorption (~0.43–0.75 μm) interpreted to result from Fe³⁺ in nanophase iron oxides found in martian dust (Morris et al. 1993). The fresh rock end-member spectrum resembles the spectral properties of rocks on Murray Ridge not covered or only thinly covered by dust. The ferric edge is subdued relative to the dusty rock spectrum, and at longer wavelengths the spectrum shows a shallow dip, consistent with electronic absorptions due to the presence of one or more ferrous silicates, most likely pyroxene(s) (Clark 1999). The bright coating end-member has the highest reflectance values of the four PI end-members. The ferric edge is present, but not as prominent as observed for the dusty rock. The bright coating end-member spectrum also exhibits a relatively steep downturn in reflectance between 0.9 to 1.0 μm, consistent with the presence of H₂O and/or OH in the mineral structure (e.g., Rice et al. 2010).

Spectral end-members were retrieved using the sequential maximum angle convex cone methodology on Pancam 13-band data for SI. Four end-members were entered as a constraint, although only three could be retrieved in a statistically viable manner. The first is similar to the PI dusty rock end-member, and the second is similar to the relatively fresh rock end-member. The third is statistically indistinguishable from the PI dark coating end-member (Fig. 18). The PI and SI dark coating end-member spectra both have very low reflectance values that increase monotonically with increasing wavelength, with the exception of a shallow dip centered at 0.754 μm, and a flattening of the slope for wavelengths longer than 0.934 μm. Munsell color values (Kelly and Judd 1976) for the dark coating end-members for both rocks are 2.5R2.5/2, where 2.5R designates a slight deviation from a red hue, 2.5 indicates a dark surface, and 2 indicates a near gray appearance. The dark coating end-member color and spectral properties are unique for Mars, which ubiquitously show ferric absorptions at shorter wavelengths, increasing rapidly with increasing wavelength to ~0.77 μm, as shown for the other end-member spectra. We revisit the likely mineralogy of the dark and bright coatings after first considering constraints from the APXS-based compositions for PI and SI rocks.

Five overlapping APXS observations were acquired for PI (Fig. 14) and four for SI (Fig. 16) (Table 1). The field of view (FOV) of the instrument is comparable to the ~3.5 cm PI width, although most of the signal comes from the inner portion of the FOV. The situation is slightly better for the ~12 cm long SI rock. APXS placements were positioned to concentrate on the spectral end-member locations for PI and to cover the breadth of SI. For the PI APXS measurements the highly overlapping nature of the observations requires that care be taken in interpretation. Examination of the compositional data shown in Table 1 shows enrichments in Mg, Mn, Ni, and S for both PI and SI, relative to surrounding basaltic bedrock. The enrichment patterns are also distinctly different than found for the sulfate-rich sandstones that dominate the Burns formation (Clark et al. 2005). The APXS FOVs containing the highest concentrations of the dark coating end-member (PI_3, PI_5) have the greatest enrichments in these elements. PI_3 and PI_5 also show minor enrichments in Ca and P, and overall the PI targets have higher concentrations of Cl, relative to SI targets (Fig. 13, Table 1). PI_3 has the highest concentration of Mn (3.48 wt% MnO, Table 1) measured thus far by the Spirit and Opportunity rovers, and SI_3 has the highest S concentration (38.21 wt% SO₃, Table 1) measured by either rover. PI and SI coating compositions are thus unique among the many hundreds of APXS measurements collected by the Spirit and Opportunity rovers (Gellert et al. 2006).

Because the APXS FOV is large relative to the size of PI, it is not possible to use the observations alone to retrieve the compositions of the purest dark and bright coatings that are well-exposed on this rock and localized using Pancam spectral analysis. On the other hand, if we assume that the compositions follow the areal concentrations of Pancam-based end-members, various techniques can be used to retrieve compositional estimates for the purest end-member locations. To pursue these retrievals, the location of each APXS PI measurement was derived from examination of MI frames pointed toward the APXS target center, together with projections of the IDD motions toward the target. Locations were refined using the predicted total signal as a function of terrain topography and derived APXS stand-off distances. Overall the methodology for localization of the APXS measurements followed the procedures described in VanBommel et al. (2016).

For each PI APXS observation the Pancam-based end-member phase abundance maps were spatially convolved with the APXS FOVs at the Pancam pixel scale, given the APXS measurement location, the lateral distance from the detector center, the APXS detector stand-off distance, and topography (Fig. 19). To ensure more observations than unknowns in retrieving compositions for the Pancam-based purest end-member locations, the dusty and fresh rock end-members were combined, and SI, Anchor Point soils, and Cook Haven bedrock observations were added to the data matrix. Thus, 16 measurements were used together with the phase abundance matrix (3 end-member columns and 16 observation rows) to solve for the compositions of 3 end-members (16 oxide columns and 3 end-member rows): rock, bright coating, and dark coating. In matrix notation the phase abundance matrix was post-multiplied by the end-member composition matrix to generate the matrix of oxide compositions.

An iterative minimization algorithm with a non-negativity constraint was used in which both the phase abundance and end-member composition matrix values were allowed to vary. Initial phase abundances for PI observations were set by the Pancam-based end-member map convolutions with the APXS FOVs, whereas the other initial phase abundances were initially set to random numbers, including those for SI. The sums of squared deviations between the model predictions and measured values for the 16 elements for the 16 APXS measurements were minimized and used to compute statistical errors of the retrieved end-member compositions and phase abundances. Results are presented in Table 2 for retrieved end-member compositions and Table 3 for phase abundances. Major elements were retrieved with small errors for the dark coating and rock end-members. The bright coating end-member composition retrievals have relatively large errors that are a consequence of the small areal extent of this end-member and thus relatively poor APXS statistics. Low concentration elements were also difficult to retrieve for all end-members and some zero values were retrieved, which is compositionally incorrect, and a limitation of the retrieval procedure. Phase abundance retrievals are consistent with the Pancam-based images for PI; e.g., examinations of APXS locations over the end-member abundance maps are consistent with retrievals shown in Table 3. In addition, SI phase abundances show high concentrations of bright and dark end-members in all four observations, consistent with the more complicated color patterns evident in Pancam data for SI than for PI (Figs. 14 and 16).

To pursue how the end-member retrievals match and/or extend trends in compositions, a correspondence analysis (CA) was run for all measurements on Murray Ridge and included the three end-member compositions (Fig. 20). CA is a row- and column-normalized principal component analysis used for understanding correlations among samples and variables and has been used to explore patterns for the intrinsically multivariate APXS data acquired by both the Spirit and Opportunity rovers (Arvidson et al. 2010, 2011). The first two CA factor loadings, which carry ~98% of the data variance, demonstrate that the rock end-member composition has a close affinity to bedrock. The bright and dark coating end-members extend the differences in compositions between the measurements centered over the bright and dark coating exposures on PI. The bright coating end-member is characterized by an affinity for magnesium and sulfur, and modest amounts of manganese, whereas the dark coating also has an affinity for manganese, calcium, and phosphorus, in addition to magnesium and sulfur.

We also use the trends evident in the CA factor loadings plot to consider bivariate correlations between oxide compositions; e.g., Si and S are clearly negatively correlated (Fig. 20). This correlation is evident in a plot of the two oxides and by the high correlation coefficient of these two elemental abundances (Fig. 21). Mg and S are positively correlated as are Ca and P (Figs. 20–21). The retrieved rock and dark coating end-members set upper and lower bounds on the first three bivariate plots, whereas the Ca vs. P plot shows a much greater spread of data and end-members. Ni is not well estimated in the end-member retrievals. However, a strong Mn vs. Ni positive correlation is evident in the bivariate plot shown in Figure 22, using only the actual APXS observations.

The bright end-member coating spectral reflectance is consistent with the presence of a pure Mg-sulfate end-member, although the composition, low overall reflectance relative to Mg sulfate powders and rocks (e.g., Cloutis et al. 2006), and the presence of the short-wavelength ferric absorption imply a more complex mineral assemblage. The dark coating end-member spectral characteristics are clearly indicative of the presence of one or more minerals that are intrinsically dark because of multiple, overlapping charge transfer and/or electronic transition absorptions associated with the presence of transition metals such as Mn (Sherman 1984). We explore a range of possible candidates to show that the low reflectance and lack of ferric edge are consistent with the presence of Mn oxides. Several synthetic samples were generated and used to pursue this comparison, including hausmannite [(Mn²⁺,Mn³⁺)₂O₄], bixbyite ( $α−Mn23+O3$⁠), pyrolusite (β-Mn⁴⁺O₂), and a series of phyllomanganates: triclinic and hexagonal forms of birnessite [(Na,Ca,K)_x(Mn⁴⁺,Mn³⁺)₂O₄·nH₂O)], and vernadite $(Na0.2Mn0.954+O2⋅nH2O)$⁠. These phases were synthesized in the laboratory under controlled conditions, with phase identification verified using X-ray diffraction (Hinkle 2015). Spectral reflectance data were acquired for silt-sized portions of these minerals with lighting and viewing conditions similar to those for which the Pancam data were acquired (Fig. 23). All of the Mn oxide spectra have low overall values and either are spectrally flat, or increase modestly in reflectance with increasing wavelength, consistent with the overlapping nature of the charge transfer and electronic transition absorptions for these oxides. To our knowledge no other Mn-bearing minerals (e.g., Mn sulfates, which are bright pink) exhibit these spectral characteristics. Mn sulfides were not considered to be viable matches for the dark coating end-member spectra because these minerals are exceptionally rare in nature. The reason is that the electronic structure of Mn favors the maintenance of localized 3d orbitals in a high-spin configuration, rather than hybridization into molecular orbitals shared with sulfur and typical of other sulfide minerals (Vaughan and Rosso 2006).

To further pursue possible mineral assemblages we calculated the phases that would have been produced if the bright and dark coating end-member compositions formed via equilibrium precipitation from an aqueous fluid. Calculations were based on a geochemical reaction model using The Geochemist’s Workbench version 10.0.6 (Bethke 2007). Oxide components reported in Table 2 were modeled as reacting with 1 kg of water, converting to mineral phases based on fluid saturation state. Cr₂O₃, Br, Ni, and Zn were excluded from the model as these occur at minor to trace levels and will largely occur as substituting elements in other minerals. Reaction of 1.5 kg of oxide components with 1 kg of water yielded a stable configuration, i.e., an incremental addition of oxide components yielded a proportional increase in existing minerals. This reaction left a residual brine containing primarily Mg and sulfate; this fluid was then evaporated to obtain the full mineralogy. The initial reaction employed a previously described thermodynamic database (Catalano 2013). The evaporation step required use of a Pitzer-style activity model. A previously compiled database (Tosca et al. 2005, 2007) was modified with more recent compilations of ion-interaction parameters (Marion et al. 2003, 2008, 2009, 2010) and revised solubility data (Grevel and Majzlan 2009, 2011; Kobylin et al. 2011; Majzlan et al. 2004a, 2004b). Pitzer models are not parameterized for P and Ti, and thus these elements were removed for the evaporation step; >99.99% of the P and Ti added to the system were precipitated in minerals in the initial reaction. All calculations were performed at 25 °C because the thermodynamic data available are most robust at this temperature.

The geochemical modeling results yield plausible mineral assemblages, with both end-members dominated by Mg sulfate, as expected from the bulk compositions (Fig. 24). The dark coating end-member retrieval also contains gypsum, ferric hydroxysulfates, and ferric phosphate, with the bright coating retrieval containing nontronite (as the main host of SiO₂), and minor gibbsite and gypsum. Mn in both coatings is predicted to occur as a phyllomanganate (birnessite), with the dark coating containing a substantially larger mass fraction of this mineral. These calculations assume all phases were in equilibrium and that the end-members contained no detrital material. Neither assumption is fully valid for real systems. Thus these calculated assemblages simply demonstrate that end-member compositions correspond to realistic mineral assemblages that would form by precipitation from aqueous solutions. Additional calculations (not shown) explored possible paragenetic sequences associated with closed system chemical processes during and after deposition, but were unable to produce the dark coatings by alteration or leaching of the bright coatings, and vice versa, as both contain mixtures of soluble and insoluble phases and have distinct compositions. The model results suggest that the formation of these coatings involved multiple stages of fluid flow and coating formation.

The morphologic, stratigraphic, spectral, and compositional patterns evident on the PI and SI rock surfaces are interpreted to indicate two episodes of authigenic mineral deposition, both dominated by precipitation of sulfates from subsurface fluids that were largely neutralized by reactions with basaltic bedrock. The initial precipitation generated bright coatings that were subsequently altered to form a thin layer of what is likely Mg-sulfate-dominated mineralogy that also includes Mn oxides. We note that the presence of Mn oxides would also suggest scavenging of Ni (Post 1999), thereby helping to explain the positive correlation between these two elements (Fig. 22). The precipitation of Mn oxides requires the presence of a high concentration of potential oxidants (e.g., O₂, although this is by no means a unique identification), most likely associated with fluids in direct contact with the atmosphere. To account for the observed elemental correlations, incorporation of Ca, P, Cl, and Br into one or more phases [e.g., chloroapatite (Klein 2002)] must also have occurred for PI.

The PI and SI coating compositions are unique among the many hundreds of APXS measurements acquired by the Opportunity and Spirit rovers. This is evident in a plot of Fe vs. Mn for all data collected through Murray Ridge (Fig. 25). It is also evident when the S- and Cl-free PI and SI compositions are placed in context with all of the soil and bedrock measurements. Specifically, projecting the data onto an S- and Cl-free ternary diagram illustrates the decreasing importance of silicates and Fe-bearing phases as PI and SI measurements move from rock to bright coating to dark coating compositions (Fig. 26). PI and SI coating trends are distinctly different from trends related to hematitic concretions in the Burns formation sulfate-rich sandstones (Morris et al. 2006), Ca-sulfate veins on Cape York and surrounding bench deposits (Squyres et al. 2012), Bristol Well Ca-sulfate veins, and aluminous phyllosilicates (Espérance) in a fracture on Cape York (Arvidson et al. 2014).

Endeavour’s highly fractured rim is interpreted to have provided a conduit for subsurface fluid flow, and this would have particularly been the case in the immediate aftermath of the crater-forming impact event and associated heating of groundwater (e.g., Osinski and Pierazzo 2012). Based on Opportunity observations, most of the impact breccia outcrops on the western rim of Endeavour’s Murray Ridge rim segment do not show major element compositional deviations from a basaltic composition. Minor fracture-filling Ca-sulfate veins have been encountered, implying modest flow of fluids and regional-scale precipitation of a relatively insoluble sulfate phase or phases. In addition, relatively high S and Cl concentrations associated with Cook Haven bedrock outcrops imply modest addition of these mobile elements. On the other hand, Pinnacle and Stuart Island rocks, serendipitously excavated from a soil-filled fracture by Opportunity’s wheels, provide strong evidence for movement of fluids through the subsurface, and formation of a unique sulfate-rich deposit overlain by a sulfate and Mn oxide-rich coating.

On the basis of inferred mineralogy, the aqueous fluids that deposited coatings on Pinnacle and Stuart Island rocks exhibited temporally varying redox conditions governed by subsurface rock-water interactions in contact with an oxidizing surface environment. Discovery of these rare deposits on the rim of Endeavour Crater complements the discovery of equally rare Mn-oxide deposits formed by aqueous flow in subsurface fractures by the Curiosity rover in Gale Crater (Lanza et al. 2015). These two discoveries demonstrate that Mn-oxides must have been part of the planet’s secondary mineral repertoire that required a stronger redox gradient in some near-surface environments than previously recognized. In contrast to arid regions on Earth, where Mn oxides are widely incorporated into coatings on surface rocks (e.g., Liu and Broecker 2008), our results demonstrate that on Mars the most likely place to deposit and preserve Mn oxides was in fracture zones where migrating fluids intersected surface oxidants, forming precipitates shielded from subsequent physical erosion.

We thank the Opportunity Project Team at the NASA/Caltech Jet Propulsion Laboratory and scientists from many institutions who made possible the collection of data included in this paper. We thank Paolo Bellutta for help in localizing APXS fields of view and Susan Slavney and Jennifer Ward for careful review and editing of text and figures. Bonnie Redding, U.S. Geological Survey, kindly generated the MI anaglyphs. We also thank NASA for the support needed to operate Opportunity and collect and analyze the data included in this paper and the Mars Fundamental Research Program support to J.G. Catalano. The NASA Planetary Data System Geosciences Node houses the data included in this paper and we thank them for their efforts. See http://pds-geosciences.wustl.edu/. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.