Main

Mars accreted and differentiated rapidly after the Solar System formed10, and both geophysical modelling and geochemical evidence support the presence of a global magma ocean at this early time7,11. As this magma ocean cooled, dissolved H2O and CO2 were outgassed, building up a steam or supercritical atmosphere (SSA) tens to hundreds of bars thick7,8 that could have survived for around 107 years (ref. 9) and would have reacted with the first-formed solid crust of Mars7. Chemical weathering beneath this atmosphere should have been efficient because of the high chemical activity of H2O, and because pressure (P)–temperature (T) conditions would have been near or beyond the H2O critical point. These conditions would have been much hotter for a given pressure than were later hydrologic systems on Mars. Clay minerals formed during this intense weathering episode may be preserved to this day, because Mars did not experience mature, subduction-driven plate tectonics12 to recycle the altered early crust. Extrusive volcanism and impact-ejected material (basin ejecta) would have acted to bury a primordial clay layer; however, clays would have been remobilized by both impacts and physical weathering to incorporate them into younger, Noachian-aged geologic units.

In order to evaluate water–rock interactions at the crust–SSA interface and below, we reacted synthetic crystalline Martian basalt with H2O and H2O–CO2 mixtures at relevant P–T conditions (see Methods for details). Our results show that the basalt is rapidly and extensively altered to form clay minerals (Fig. 1 and Extended Data Fig. 1), with other minor alteration phases also appearing, including nepheline and portlandite. Clay minerals were formed in experimental conditions near the H2O critical point in all three phase regions (vapour, liquid and supercritical). We identified clays on the basis of: first, morphology in scanning electron micrographs (Fig. 1b–d); second, vibrational absorptions of metal–OH bonds in visible/near-infrared (VNIR) reflectance spectra between 2,200 nm and 2,400 nm that are diagnostic of iron/magnesium clays (Fig. 1e, f); and third, X-ray diffraction peaks at d-spacings of 10–14 Å that are characteristic of basal [001] reflections of phyllosilicates (Fig. 1f). A variety of iron-rich, trioctahedral clays formed depending on the P–T conditions and the presence or absence of CO2. For example, in the liquid-water field, a platy, expanding 12 Å clay formed, consistent with interstratified trioctahdral illite-montmorillonite. In the supercritical H2O field, a fibrous, non-expanding 10 Å clay formed, consistent with palygorskite. Finally, when CO2 was added, an expanding 14 Å clay formed in the supercritical field, consistent with trioctahedral montmorillonite or vermiculite. Together these experiments demonstrate that P–T conditions at the crust–SSA interface were highly efficient at weathering basaltic rocks to produce clay minerals. Iron-rich dioctahedral species are the most common types of Martian clays identified by remote sensing13, with the specific species being controlled by the crystal chemistry of the protolith, by P–T conditions and by any subsequent modification. These dioctahedral species probably did not form directly through aqueous alteration, but instead could have originated as iron-rich trioctahedral clays that were later oxidized13,14. Therefore the clays formed in our experiments are plausible precursors of those observed on Mars today.

Figure 1: Analyses of altered basaltic samples.
figure 1

a, Backscatter electron micrograph of a thick section of our unaltered basalt, where from darkest to lightest the phases are: plagioclase; glass (with exsolved bright quench crystals); pyroxene; olivine (zoned); and spinels. b–f, Secondary electron images showing the surfaces of unaltered and altered basalt particles. b, Unaltered starting basalt c, Sample altered at 425 °C and 150 bar (where H2O is in the form of vapour). d, Sample altered at 325 °C and 300 bar (where H2O is liquid). e, Sample altered at 425 °C and 300 bar (where H2O is in a supercritical state). f, Sample altered at 425 °C and 300 bar (with H2O–CO2 being in a supercritical state). Arrows point to clay species. g, VNIR spectra of unaltered and altered samples, with key absorptions near 700 nm, 1,900 nm and 2,300 nm indicated by grey bars. The box shows the region that is expanded in panel h. h, VNIR spectra highlighting absorption features near 1,900 nm and 2,300 nm (dotted lines) that are diagnostic of clay minerals. i, X-ray diffraction patterns of altered samples, relative to the pattern for unaltered basalt. Characteristic peaks (grey bars) at 10 Å, 12.3 Å and 14 Å represent basal [001] phyllosilicate reflections.

PowerPoint slide

Full size image

Primordial clays could have formed at considerable depths beneath the SSA because of porosity in the primary crust. The crustal porosity could have taken the form of large-scale fractures that resulted from early impact bombardment, similar to those seen on the Moon15, and microporosity that formed following thermal contraction when the primary crust cooled16. The pore-closure depth for the Martian megaregolith is estimated to be 10 km (ref. 17), suggesting that the SSA could have created an altered, clay-rich layer with a maximum thickness of around 10 km, depending on the rate at which the porosity was created. Such an early-formed, clay-rich layer on the top of the Martian crust would have been reworked by intensive impact bombardment (such as that which produced the Borealis basin18), by the emplacement of extrusive volcanic material (that is, secondary crust), and by physical and chemical weathering.

To explore the fate of a primordial clay layer on Mars, we constructed a parameterized three-dimensional box model that considers the effects of SSA-driven alteration, impact excavation, impact melting, secondary crust emplacement and basal geothermal heating (Fig. 2; see Methods for details). The model spans the first billion years after the crust is likely to have solidified19 (4.538 billion years ago (Ga) to 3.538 Ga), capturing most of the major geologic activity in the history of Mars. In the reference model (Figs 2 and 3a), we consider a sawtooth, late heavy bombardment (LHB) impact flux (ref. 20; Extended Data Fig. 2), with an imposed lull in the formation of large basins between the Borealis impact and the LHB21, an initial 3-km-thick clay layer, an SSA lasting for 107 years, 10 km of ejecta/melt from the Borealis impact being emplaced instantaneously at 4.5 Ga, and 10 km of extrusive secondary crust being emplaced with an exponentially decaying rate. In additional runs, we model the effects of thicker (Fig. 3b) and thinner (Fig. 3c) layers of Borealis ejecta, a longer-lasting SSA that persists until the time of the Borealis impact (Fig. 3d), and an accretionary tail impact flux (Fig. 3e). While the SSA is present, we assume that alteration is instantaneous on the 105-year time step of the model, and that this alteration extends to a 3-km depth at which porosity rapidly decreases17.

Figure 2: Results from the reference run of the crustal evolution model.
figure 2

The figure shows perspective views of snapshots from three time steps in the reference model. a, After 107 years, the initial altered layer has thickened owing to impact-induced excavation of unaltered material. b, After 108 years, the primordial clay layer is buried by melt and ejecta from the Borealis impact, as well as by extrusive volcanics and other impact melt sheets. c, At the end of the model run (109 years), the clay-rich layer remains mostly intact at depth; patches of minimally altered rock are exposed at the very surface. V.E., vertical exaggeration.

PowerPoint slide

Full size image
Figure 3: Alteration profiles and surface exposures for different model parameters.
figure 3

Left-hand plots show the mean clay content as a function of depth at the end of the model runs, with the reference model represented by a grey dashed line in b–e. Images on the right show a top-down view of the model grid surface, with colours stretched to show smaller clay contents than in Fig. 2. a, Reference model (Fig. 2). b, Model with 20 km of Borealis material. c, Model with 5 km of Borealis material. d, Model with SSA lasting for 38 million years, until the time of the Borealis impact (at 4.5 Ga). e, Model with accretionary tail impact flux.

PowerPoint slide

Full size image

In all model runs, the primordial clay layer initially thickens while the SSA is present (Fig. 2a). The layer is quickly buried by Borealis melt and ejecta (Fig. 2b), then modified by subsequent impacts and volcanism (Fig. 2c). Changing the model parameters results in different amounts of clay being exposed at the surface, a different overall dilution of the initially pure clay layer, and different burial depths of that layer (Fig. 3 and Extended Data Fig. 3). With twice the reference amount of Borealis cover, there are fewer clay exposures at the surface, and the clay layer is buried much more deeply in the crust (Fig. 3b). The opposite is true of the model with half the reference amount of cover (Fig. 3c). When the SSA lasts for 3.8 × 107 years (until the time of the Borealis impact), the clay layer is less diluted, and shows enhanced surface exposures (Fig. 3d). Finally, using an accretionary tail impact flux results in greater dilution and burial of the clay layer, and a greater clay content in surface materials because of enhanced impact mixing (Fig. 3e). We developed two metrics for evaluating the surface distribution of clays in model runs. The first is the detection rate, defined as the areal fraction of surface grid cells with more than 10% clay (assumed to be detectable by orbital remote sensing). The second is the background clay content, defined as the median clay proportion in surface grid cells outside of detections. The actual areal clay coverage on Mars is estimated to be 3.4% (ref. 4), and the background clay content is probably 1% or less, based on the lack of crystalline clay in soils at Gale crater22 and in the regolith breccia Northwest Africa 7034, which is a good spectral analogue of typical low-albedo (that is, dust-poor) terrains on Mars23,24. Extended Data Fig. 3 shows that two model runs—the reference model and the model with twice the Borealis cover—are consistent with estimates of both of these metrics for Mars. Other model runs predict either too many surface exposures, too much clay in the background regolith, or both.

Our primordial clay model has a number of strengths in explaining observed crustal clay exposures on Mars. In this model, the water and heat needed to drive clay formation are a natural consequence of the Martian magma ocean, and there is no need to appeal to enhanced rainfall, hotter geotherms, or sustained deep groundwater in the younger Noachian period. A primordial clay layer excavated by impacts can explain the patchy distribution of Martian clays, as well a possible clustering of clay detections around the Hellas, Argyre and Isidis basins. These may have been among the small handful of post-Borealis basins21 that were capable of excavating beneath the thick cover of Borealis ejecta and secondary crust that buried the primordial clay layer. Impact emplacement of older material has been suggested to explain the clays observed deep in the walls of Valles Marineris, which did not form in situ as a coherent alteration horizon25. Other models for clay formation involve subsurface hydrothermal alteration3, surficial weathering profiles6, impact-induced hydrothermal alteration26, and precipitation directly from magmatic fluids27. None of these processes is mutually excluded by the primordial-clay model, but the volume of clays that they contributed might have been limited by the rapid secular cooling of Mars. Observed Martian clays probably contain contributions both from primordial clays formed by the SSA, and from clays formed by these later processes. In many cases it may be difficult to distinguish between the two by using orbital remote sensing alone: in situ investigations or returned samples are needed. The SSA would have had a distinct composition from later fluids, which might allow SSA-derived clay to be identified. For example, SSA-derived clay probably has higher noble-gas contents and distinct isotopic compositions compared with later clays. Other volatiles—such as chlorine, fluorine and iodine—may also be useful in distinguishing these clays.

The primordial clay model also has implications for our understanding of early Martian climate, water sequestered in the subsurface, and crustal density. In this model, clay formation is mainly decoupled from the surface climate and is therefore consistent with the mostly cold and dry Noachian conditions suggested by recent three-dimensional climate models28. Martian clay minerals are partially hydrated, as indicated by vibrational absorptions from structural and/or adsorbed H2O and OH1,2,3. Therefore, a buried clay layer could have sequestered substantial amounts of water in the subsurface of Mars29 (although both bound and adsorbed water become unstable with increasing pressures and temperatures at depth). A buried primordial clay layer could also explain the anomalously low density of the Martian crust. Topography and crustal density measurements support the presence of a buried low-density layer in the southern highlands crust30, which is consistent with our modelling results (Fig. 3). Clays were dismissed as a possible contributor to this low-density layer because they were not expected to exist at depths below 1 km (ref. 30), but they have since been discovered at depths of 10 km or more4. Similar clay-rich layers formed during magma ocean cooling on Earth and Venus would not be expected to be preserved at the surface owing to plate tectonics and global resurfacing, respectively, but they might nevertheless have been important to the early geodynamical evolution of those worlds.

Methods

Basalt synthesis

We synthesized a crystalline basaltic material to use in the alteration experiments. The composition of this sample was based on the bulk chemistry of the meteorite Northwest Africa 7034—a Martian regolith breccia that closely approximates orbital γ-ray spectroscopy data from the bulk crust of Mars31. Expressed as wt% oxides, this composition is: 50.57% SiO2, 13.38% FeO, 11.92% Al2O3, 9.5% CaO, 8.31% MgO, 3.98% Na2O, 1.04% TiO2, 0.36% K2O, 0.3% MnO and 0.19% Cr2O3. To re-create this composition, we homogenized reagent-grade powders of SiO2, Fe2O3, Al2O3, MgO, TiO2, MnO, Cr2O3, CaCO3, K2CO3 and Na2CO3 under ethanol and then dried them. The resulting powder was decarbonated by heating in a horizontal furnace in air for 3 hours at 950 °C. Approximately 2 g of decarbonated powder were loaded into a 2 ml cylindrical alumina crucible and melted in a horizontal gas-mixing furnace; the sample was first heated to 1,400 °C, above its liquidus, for 2 hours, and then cooled at 5 °C per hour to 1,000 °C, at which point the furnace was shut off and the sample removed and broken out of the crucible. A controlled mixture of CO2 and CO in the furnace was used to set the oxygen fugacity to 1 logarithmic unit below the quartz–fayalite–magnetite solid oxygen buffer. The mineralogy of the resulting synthetic rock consisted of plagioclase > pyroxene (mostly homogenous at Wo46En30Fs24) > olivine (zoned from Fo31 to Fo63) with accessory oxides and quenched glass (Fig. 1).

Alteration experiments

The synthetic basaltic rock was ground and dry-sieved to a 0.5–1-mm grain size, then washed with ethanol to remove clinging fines. Aliquots of this particulate sample were loaded in silver–palladium tubing and crimped at both ends, but not welded shut. Samples were then loaded into hydrothermal vessels attached to a water reservoir. The pressure vessels were run in a horizontal furnace. Experiments were run for two weeks, after which the silver–palladium tubing was opened to remove the samples. Experimental conditions were based on modelled magma ocean cooling pathways for the surface of Mars7 and included P–T conditions in the liquid field (325 °C, 300 bar), vapour field (425 °C, 150 bar) and supercritical field (425 °C, 300 bar). In order to test the effect of CO2 on alteration products and extent, we ran a separate experiment at 425 °C and 300 bar with a welded silver–palladium tube filled with 100 mg of oxalic acid, which breaks down to give equal molar proportions of H2O and CO2.

Sample analysis

We analysed unaltered starting material and altered materials by using X-ray diffraction (XRD), visible/near-infrared (VNIR) spectroscopy, and both energy-dispersive X-ray spectroscopy (EDS) and wavelength-dispersive spectroscopy (WDS) elemental mapping. All analyses were performed in house at Brown University. Samples were ground and dry-sieved to a 45–75-μm size fraction for both XRD and VNIR spectral measurements. XRD was performed on a Bruker D2 PHASER diffractometer. Initial characterization was done from 5° to 70° 2θ, using a step size of 0.02° and a counting time of 15 s. For clay identification, samples were solvated in ethylene glycol, and heated first to 400 °C and then to 550 °C. After each treatment, XRD measurements were taken from 5° to 13° 2θ, with a counting time of 3 s. Diffraction peaks were identified using the Bruker DIFFRAC.SUITE software package. VNIR spectral measurements were made using the custom-built ultraviolet–visible–near-infrared bidirectional spectrometer at the Reflectance Experiment Laboratory at Brown University32. Measurements spanned from 320 nm to 2,600 nm with a 10-nm sampling. Absorption features in the spectra were compared to laboratory measurements of pure phases, and specific attention was paid to the Al–OH, Fe–OH and Mg–OH vibrational absorptions that are diagnostic of phyllosilicate minerals and occur at 2,200 nm, 2,290 nm and 2,320 nm, respectively33. Samples were imaged using a LEO 1530 VP ultrahigh-resolution field emitter scanning electron microscope in variable pressure mode with a voltage of 20 kV (Fig. 1b–f). EDS spectra were used to measure the qualitative elemental chemistry of alteration phases. Basalt grains were also mounted in epoxy and polished to create cross-sections that were mapped on a Cameca SX-100 electron microprobe. Elemental mapping of silicon, aluminium, calcium, iron and magnesium was performed with spatial resolution of 1 μm per pixel (Extended Data Fig. 1).

Crustal evolution model

We constructed a parameterized three-dimensional box model in MATLAB to consider the competing effects of impact ejection, impact melting and secondary crust formation on the fate of a thick altered layer at the top of the primary Martian crust. The model consists of 500 × 500 × Z grid cells, where Z starts at 300 and grows with time as secondary crustal material is added. The horizontal resolution of each cell is 7.61 km and the vertical resolution is 0.5 km, such that the areal extent of the entire grid represents 10% of Mars’ surface area. Each grid cell has a continuous value representing its clay content (0%–100%). The model runs from 4.538 Ga to 3.538 Ga, with a 105-year time step, encompassing the time period from solidification of the primary crust (estimated to have occurred at about 30 million years or less after Solar System formation19) to the cessation of major impact and volcanic events near the end of the Hesperian period.

At each time step, if the SSA is still present then all unaltered material within a certain depth from the surface (that is, the initial depth of the clay layer) is instantly converted to fully altered material. This assumes that alteration is 100% efficient within the 105-year time step of the model. How long the SSA persisted is poorly constrained and estimated to be of the order of 107 years (ref. 9). We used a value of 1.0 × 107 years for all model runs except one, in which the SSA was assumed to last for 3.8 × 107 years (until the time of the Borealis impact at 4.5 Ga in the model). A linear geothermal gradient of 20 °C per km is imposed, which converts grid cells heated above 750 °C (that is, 37.5 km depth or greater) into unaltered material, representing thermal alteration and degradation of clays. In the first stage of the model (Fig. 2a), the SSA is present and unaltered rock excavated from depth is altered by the SSA. This acts to thicken the clay-rich layer. After the SSA is gone, no additional clay is formed in the model. Unaltered rock accumulates in the upper crust by melting of clay-rich rock from impacts, and by mantle-derived melts erupted at the surface. Impact cratering mechanically mixes both altered and unaltered crustal materials, diluting the clay-rich material and forming a layered volume of serial ejecta blankets and lava that contains an increasing proportion of unaltered rock over time.

Impact craters are generated using a Monte Carlo approach that is based on cratering rates and size–frequency distributions appropriate for Mars. Craters are selected from the following final crater diameters (DF; all in km): 50, 60, 70, 80, 90, 100, 115, 130, 150, 170, 200, 250, 300, 350, 400, 450, 500, 600, 700, 800, 900, 100, 1,100, 1,200, 1,300, 1,400 and 1,500. The crater size–frequency distribution (CFSD) follows a uniform power-law relation with a –2.2 slope, appropriate for these large diameters34. This CFSD determines the relative probability of a crater of a given size being chosen. The total number of craters per time step (that is, the impact flux) is modelled in two ways (Extended Data Fig. 2): using a sawtooth LHB that is caused by a dynamic instability at around 4.1 Ga (ref. 20) and for a single model run (Fig. 3e) a simple exponential decay (that is, an accretionary tail) that is extrapolated from the N50 values of the five basins dated in ref. 35. In the LHB runs, we prescribe a lull in large basins (DF > 1,000 km) after the Borealis event21, then explicitly add a single 1,500-km basin at 3.9 Ga as a proxy for an Isidis/Argyre-type event. Extended Data Table 1 gives an example crater population for the reference model run.

The crater population for each time step is applied in a random order to the grid at randomized locations, such that each crater interacts with the target in succession. For craters, we calculate the diameter of the transient cavity to be DTC = 60.15 × DF0.85 (ref. 36). Toroidal flow streams of ejecta extend from the point of impact to the edge of the transient cavity with a maximum excavation depth of 0.1 × DTC. Grid cells within this torus are categorized into three vertical sections (shallow, intermediate and deep), and the material in each section is homogenized in terms of its clay content. Ejected grid cells are then distributed one at a time outside the crater rim with a random azimuth, and a random distance r is drawn from a probability function with the form (r / DF)−2.3 (ref. 37). The material sourced from the deep part of the ejecta torus is ejected farthest by slightly biasing r for ejected parcels. The shallow-sourced ejecta ends up closest to the crater rim, with the intermediate section in between. Impact melt volumes are calculated as 3.6382 × 10−4 × DTC3.85 (ref. 38), and melt is assumed to be a half-sphere (truncated at the top) extending downwards from the point of impact. Grid cells within the melt radius are reduced to a clay content of 0%. Melt within the ejecta flow streams is added to the ejecta, and the remaining melt is then settled at the bottom of the final crater, whose initial (melt-free) void space is calculated as a paraboloid with depth 0.286 × DF0.582 (ref. 39).

Erupted volcanic material is added at each time step as unaltered grid cells to the upper surface of the grid in a fractal-like pattern. The thickness (T) of erupted volcanic material, averaged over the grid, diminishes with time as a simple exponential function, T = A × exp(−bt), where A is the initial volumetric crustal production rate at t = 0 (20 km3 per year; ref. 40), distributed evenly over the grid; b is fixed by the total erupted thickness over the model duration; and t is time. The total volume of secondary crust added to Mars is estimated to be about 30 km (ref. 40). We use a conservative extrusive/intrusive ratio of 1/2, such that the amount of extrusive volcanics for a model run would equate to a 10-km-thick layer if it were evenly spread over the grid. This leads to values of A = 1.38 × 10−4 and b = −1.62 × 10−27 for calculating T in metres per year.

Code availability

We have opted not to make the code for the three-dimensional box model available, because it was written solely for this project and does not have broad utility for other analyses.

Data availability

The data that support the findings of this study are available from the corresponding author on reasonable request.