New textural evidence on the origin of carbonado diamond: An example of 3-D petrography using X-ray computed tomography

Carbonado is an enigmatic polycrystalline diamond variety found in placer deposits in Brazil and the Central African Republic (Trueb and Butterman, 1969; Trueb and De Wys, 1969, 1971). Carbonados are commonly millimeter to centimeter sized, and the “Carbonado of Sergio” from Brazil is the largest diamond known (Haggerty, 1999) at 3167 carats. Age determinations range from 2.6 to 3.8 Ga, with large uncertainties (Ozima and Tatsumoto, 1997; Sano et al., 2002). The confined geographic distribution of carbonados, on land masses that were likely adjacent at times in the Archean, suggests that they may all have been created in a single event (Heaney et al., 2005), although they are in some respects similar to other polymineralic diamond varieties such as framesites and yakutites (McCall, 2009).

Carbonado is set apart from other diamond varieties by several unusual characteristics. The diamond material is texturally diverse, being dominantly porphyritic and consisting of small (10–250 µm) diamonds cemented together by even smaller (<1 µm) microdiamonds (De et al., 1998; Petrovsky et al., 2010). The smaller crystals can have subplanar dislocations that may be interpreted as defect lamellae, whereas the larger ones are mostly defect free. Other, rarer textures found in subregions are homogeneous, flow texture (Yokochi et al., 2008) and columnar crystals reminiscent of vein infill (Rondeau et al., 2008). Carbonado carbon isotopes are very light, with δ¹³C values ranging from –21‰ to –32‰, with major modes at –24‰ and –26‰, rare for diamonds and suggestive of organic carbon (De et al., 2001; Kagi et al., 2007), although the source for light carbon in the mantle remains an open question (Deines, 2002). The diamond material contains H (hydrogen) defects indicative of a hydrogen-rich environment (Garai et al., 2006; Nadolinny et al., 2003), and N (nitrogen) defects in various states of aggregation (Fukura et al., 2005; Garai et al., 2006). Carbonado hosts extensive porosity ranging from the nanometer up to the millimeter scale. Some carbonado material has been found to be enriched in isotopes characteristic of the products of spontaneous fission of uranium (Ozima and Tatsumoto, 1997), but cathodoluminescence imaging suggests that this may be a secondary feature caused by infiltration of the pore network by fluids enriched in radioactive elements (Kagi and Fukura, 2008; Kagi et al., 2007; Rondeau et al., 2008).

The inclusion suite in carbonado has also proven enigmatic. It hosts enclosed nano-inclusions of various metal alloys (Fe, Fe-Ni, Ni-Pt, Si, Ti, Sn, Ag, Cu, and SiC), as well as other minerals (calcite, sylvite, smithsonite) that may be original (De et al., 1998) and some (augite, ilmenite, phlogopite) that may predate the diamond-forming event (Sautter et al., 2011). It also hosts larger metal particles tens of micrometers in diameter within open pore space (De et al., 1998; Fitzgerald et al., 2006). Its macro-inclusion suite principally features minerals of ostensibly crustal origin, such as orthoclase, goethite, quartz, kaolinite, hematite, serpentine, and florencite, a hydrous rare-earth phosphate that commonly forms as an alteration product of monazite. The fact that extensive leaching can extract virtually all compositional impurities (Dismukes et al., 1988) and magnetic components (Fitzgerald et al., 2006; Kletetschka et al., 2000) suggests that all macro-inclusions are probably secondary products precipitated by fluids moving through a connected pore network. None of the inclusion phases found thus far is typical of kimberlitic diamond, such as pyrope, pyrrhotite, and chromian clinopyroxene (Heaney et al., 2005), although olivine has been reported (Trueb and De Wys, 1971). Ishibashi et al. (2012) report a 0.3 µm void with euhedral walls that they interpreted as a former fluid inclusion, though it was not preserved as such due to sample preparation.

The origin of these diverse features in carbonado has eluded consensus for over 40 yr. Hypotheses have included meteorite impact (Smith and Dawson, 1985); formation during hot Archean subduction of early organic material; volcanic hydrothermal systems with metal-catalyzed diamond growth (McCall, 2009); extraterrestrial origin (Garai et al., 2006); and irradiation of carbonaceous materials (Ozima and Tatsumoto, 1997). Recent work has favored formation at mantle conditions, likely in the presence of a C-O-H fluid supersaturated in carbon (Ishibashi et al., 2012; Petrovsky et al., 2010), a C-H–rich fluid in lower crust transiently metamorphosed at mantle conditions (Sautter et al., 2011), or associated with komatiite magma intrusion into the continental lithosphere (Cartigny, 2010).

Because carbonado is among the toughest substances in nature, it is difficult to study using traditional petrographic techniques. With the exception of early X-ray imaging (Trueb and Butterman, 1969; Trueb and De Wys, 1969, 1971), most studies have concentrated on microchemical analyses of small pieces of material that were powdered, fractured, Focused Ion Beam– (FIB) extracted, laser-cut, or polished with laborious effort, providing a spatially limited and largely two-dimensional perspective. Most macro-inclusions have been studied by disaggregating carbonado specimens. It is particularly challenging to examine the macro-inclusions and diamond in concert on anything but a fractured or laser-cut surface, due to the extreme range of material hardness. As a result, relating these features to each other has been problematic. Here, we report on a textural analysis using high-resolution X-ray computed tomographic (HRXCT) imaging of a relatively large specimen. Three-dimensional nondestructive imaging of pores and inclusions in situ enables us to look at various textural details in their spatial context, providing a holistic view, and essentially allowing “three-dimensional (3-D) petrography” to be performed.

Our sample is from the Central African Republic, and it was purchased from a dealer in Belgium. It is 23.45 carats (4.690 g), ∼18 mm × 13 mm × 10 mm, with a deltoid shape (Fig. 1). One broad side and a smaller adjacent edge have a silvery, vitreous patina, a common though not ubiquitous feature in carbonados that has been interpreted by some as a fusion crust (e.g., Shelkov et al., 1997), although overall the side is rough. The other side is a dull dark gray and rougher. Pores are visible to the naked eye on all surfaces.

We imaged the specimen using the Xradia MicroXCT scanner at the University of Texas High-Resolution X-Ray CT Facility (http://www.ctlab.geo.utexas.edu). The MicroXCT has a unique design for a micro–computed tomography scanner. Whereas most instruments utilize large detectors and geometric magnification from a small X-ray focal spot to achieve high resolution, the MicroXCT derives its resolution from a set of specialized detectors consisting of a 35 mm camera lens or various microscope objectives coated with scintillating material. This arrangement produces very sharp imagery and facilitates “zooming in” to image small subvolumes within larger specimens by simply switching detectors, in much the same manner one switches objectives with a petrographic microscope (Figs. 2A, 2B, and 2C). With each such increase in magnification, however, acquisition time rises considerably, due to lower signal and the need to acquire more densely sampled data (i.e., more views) to reduce the effect of parts of the sample not in the region of interest but still in the X-ray path.

The specimen was scanned at a range of resolutions with different detectors, using 29.2 µm voxels to image the whole sample and 5.4 µm and 1.0 µm voxels to image selected subvolumes within it. X-ray energies were varied from 40 to 140 kV, and some scans were repeated at different energies to help with discrimination of phases. Crystals of almandine, apatite, and calcite were also included in the field of view when imaging the entire specimen to aid in interpretation of computed tomography (CT) numbers (image gray levels).

Data were reconstructed as up to 900 ∼1000 × 1000 16 bit grayscale images per scan, and software corrections were applied to compensate for ring and beam hardening artifacts. The volumetric data were visualized with Avizo (VSG, Inc., versions 6.1 and 7.0) and VG Studio Max (Volume Graphics, Inc., version 2.0). Measurements of metal inclusions were carried out with Blob3D software, accounting for partial-volume effects to compensate for their small size (Ketcham, 2005a, 2006), and 3-D determination of inclusion preferred orientation was conducted using star volume analysis with Quant3D software (Ketcham, 2005b).

Following scanning, a single cut was made in the sample using a laser at the Carnegie Institute. After polishing the surface on a diamond wheel, scanning electron microscopy (SEM) images and chemical analyses were obtained using the JEOL JSM-6490LV at the University of Texas.

Visualization was a particularly important aspect of our methodology. It is not an exaggeration to say that many of the insights in this study can only be obtained by viewing the CT imagery in motion, either as slice animations stepping through the data or as rotating volume renderings allowing structures and fabrics to be viewed from many orientations. We thus provide a number of these animations as supplemental material. These animations are also available at http://www.ctlab.geo.utexas.edu/pubs/ketcham_koeberl/ketcham_koeberl.htm.

CT numbers reflect the linear X-ray attenuation coefficient of the materials being scanned, which are a function of density, atomic number (Z), and X-ray energy (Ketcham and Carlson, 2001). Due to the complexities of scanning, such as the polychromatic nature of the X-ray source and beam hardening and other artifacts, as well as the variability of natural materials, which can feature zoning, impurities, and microporosity, absolute mapping of CT number to phases is often not straightforward. As a result, CT data in general do not have information sufficient for independent mineral identification, necessitating prior knowledge of which phases are likely to be present. The effective attenuation coefficient for a mineral is essentially the integral of all attenuation coefficients over the range of X-ray energies used, weighted by the relative intensity of the X-ray flux at each energy, and other machine-specific factors such as detector efficiency as a function of energy. It is also affected by beam hardening within the sample and measures taken to correct for it.

The relationship between X-ray energy and linear attenuation coefficient for the comparison minerals used and some of the phases discussed in this study is shown in Figure 3. The step function for florencite is caused by the X-ray absorption K edge of Ce at 40.4 keV; heavier rare earth elements (REEs) have their K edges at progressively higher energies, up to 63.3 keV for Lu. By scanning below and substantially above 40 kV (Animations 1 and 2), the effective attenuation coefficient of REE-rich phases will change dramatically with respect to other minerals. In this case, Figure 3 suggests that when the entire X-ray spectrum used is below 40 keV, then florencite will have an effective attenuation coefficient similar to almandine, whereas when a large portion of the spectrum is above 40 keV, then florencite will be substantially more attenuating. Because previous carbonado work indicates that only REE-rich phases will have heavy elements in sufficient quantity to affect their X-ray attenuation, and other REE-rich minerals aside from florencite have also been reported (e.g., rhabdophane, xenotime; De et al., 1998), in this study we generically refer to minerals identified by reduced relative X-ray attenuation below 40 keV as REE rich.

At energies below 40 keV, the most attenuating macrophase we expect to find based on prior investigations of carbonado is native Fe. The diamond material itself has very low attenuation at low kV because of its extremely low mean atomic number relative to other phases. As X-ray energy increases, and the dominant attenuation mechanism transitions from photoelectric absorption to Compton scattering, the attenuation coefficient becomes a linear function of energy, and other phases with higher mean atomic numbers but lower density (i.e., spinel, kaolinite) come to have similar or lower attenuation coefficients compared to diamond. For example, a euhedral inclusion faintly visible in the lower right corner in Figure 2D, which was imaged at 40 kV, is almost indistinguishable from the diamond matrix when imaged at 140 kV in Figure 2C.

Figure 4 and Animations 3–6 show a series of volume renderings of the inclusion phases in the 29.2-µm-resolution data, in which the diamond is rendered transparent except for the sample boundary, and other phases are rendered transparent or partially or fully opaque based on CT number. The highest-attenuation phases in the data gathered at 140 kV (Fig. 4A; Animation 3) are expected to be REE-rich minerals such as florencite, and native Fe, the latter of which is far less abundant. REE-rich phases occur throughout the sample. When data are acquired at 40 kV (Fig. 4B; Animation 4), the REE-rich phases are indistinguishable from the almandine crystal included in the scan field, as predicted from the relations shown in Figure 3. The similarity of the inclusions rendered opaque in Figures 4A and 4B indicates that there is little material similar to almandine (i.e., a relatively dense Fe-bearing phase) in the carbonado, and this corroborates the interpretation that the phases shown in Figure 4A are REE-bearing due to their substantial change in attenuation relative to almandine. As the range of CT numbers rendered opaque is expanded to encompass the values for the apatite and calcite samples (Figs. 4C and 4D; Animations 5 and 6), more inclusion phases become visible.

Native metal inclusions were identified by their high CT numbers in the 40 kV, 29.2-µm-resolution data (Fig. 5A, upper left). In total, 16 were identified and measured in the complete data set, with sphere-equivalent diameters ranging from 80 µm to 148 µm, and a mean value of 107 µm. Every metal inclusion was directly adjacent to the boundary of the specimen, with the largest distance from sample exterior to inclusion being 300 µm (see Animation 7). All of these inclusions appear to be linked to the outside of the specimen via open porosity, though the local artifacts cast by the inclusions in the HRXCT data, as seen for example in Figure 5A, prevent this statement from being made unequivocally. No smaller metallic inclusions were observed within the specimen in any of the higher-resolution subvolume scans; the nano-inclusions observed in transmission electron microscopy (TEM) studies (De et al., 1998; Sautter et al., 2011) are below the resolution of the HRXCT data.

SEM analysis on the cut section showed kaolinite and possibly other clay minerals, with minor iron, to be by far the most common inclusion mineral in this specimen. Disseminated florencite was often distinguishable by significant Ce peaks. Other probable phases included ilmenite and quartz.

The volume directly beneath the patinaed side of the specimen has an evident rim texture (Fig. 5A, right side; also see Animations 1 and 2) with up to three components distinguished by the abundance of small inclusions: a 300–500-µm-wide inclusion-depleted outer rim; an ∼500-µm-wide inclusion-rich middle band that features overall slightly higher X-ray attenuation than surrounding material; and an inner inclusion-depleted zone with variable thickness. The outermost zone may correspond to the 200-µm-thick outer rind investigated with SEM by Shelkov et al. (1997), who found it to be free of microporosity, which they interpreted to be a result of annealing just below what they interpreted to be the fusion crust.

Large, millimeter-scale inclusions occur throughout the specimen, with no readily discernible pattern, even crossing the three-layer rim texture. These inclusions are responsible for most of the megaporosity observed on the patinaed surface, and in all other faces. However, smaller inclusions with maximum dimensions in the tens to hundreds of micrometers have an apparent gradation from sparse near the patinaed surface to increasingly abundant with increasing distance from it, as seen both in the example slice image (Fig. 5A) and most clearly in a 3-D volume rendering (Fig. 5B; Animation 6). There is also an apparent dearth of small inclusions along the non-patinaed surface, but closer inspection reveals both open porosity on the same size scale as the small inclusions, and many inclusions of lower attenuation that do not show up as clearly in the volume rendering, so this is most likely a surface leaching effect.

When the region near the patinaed surface is observed more closely with 5.4-µm-resolution data (Fig. 6), it is also clear that there are low-attenuation inclusions and open pores, but the overall grading pattern of lower-inclusion-number density toward the patinaed surface remains clear. The inclusion-rich band near the patinaed surface seen in Figure 5A can also be detected in some regions of the 5.4-µm-resolution data (Animation 8). The band has a higher density of small inclusions near the resolution limit of those data, but it retains an overall brightening that cannot be ascribed to individual inclusions. Based on the progression observed across data resolutions, however, it can be inferred that the brightening is probably caused by yet smaller inclusions.

Cursory viewing suggests that the inclusions have a preferred orientation that is oblique to the overall apparent grading direction. This is supported by 3-D fabric analysis that combines all phases significantly more attenuating than diamond in the 40 kV, 29.2-µm-resolution data (Fig. 6B). As suggested by the fabric analysis and corroborated by data visualization, the predominant shape of elongated pores corresponds to blades (ellipsoid axes a > b > c). Similar fabrics were also observed in early X-ray imaging (Trueb and Butterman, 1969; Trueb and De Wys, 1969, 1971).

The higher-resolution 3-D images (Fig. 6; Animations 8–10) of the inclusions show that nearly all of them (>99%) are composites with multiple minerals. They are in many cases a mix between low-attenuation and high-attenuation phases, the latter in most cases likely to be florencite. Grain boundaries of phases within inclusions are diffuse and intergrown or mottled, indicating alteration or replacement and suggesting disequilibrium and material exchange after the diamond had formed.

There are two distinct populations of inclusions evident in the 5.4-µm-resolution HRXCT data (Fig. 6C; Animations 9 and 10). One is smaller and irregular, but often roughly ellipsoidal and with a preferred alignment. The other is larger and features many perfectly euhedral boundaries in contact with the diamond matrix. In other words, there are within the diamond material perfect negative crystal forms much larger than the diamond crystallites, which host polymineralic assemblages with disequilibrium textures. These latter minerals are thus clearly pseudomorphs replacing an earlier phase. The most distinct pseudomorphed habit is rhombic dodecahedral {011}, which tends to occur in clumps of 0.2–1 mm crystal forms (Figs. 2E and 7; Animations 11–13).

Aside from the near-edge occurrences of inclusions interpreted to be native metal, all inclusions containing high-Z phases that we observed in detail were polymineralic; the only inclusions we observed that appeared to be monomineralic were relatively large and contained low-Z material. The 1-µm-resolution images, which include one low-Z inclusion (Fig. 2D, lower right; Animation 14), reveal extremely sharp grain boundaries and a rhombic dodecahedral habit. We positioned our laser cut to expose this latter inclusion and found it to consist of kaolinite, and thus another pseudomorph.

The polymineralic nature of many inclusions complicated 3-D visualization by volume rendering, and thus required segmentation by hand to fully discern pseudomorph faces. Selected examples are shown in Figure 7 and Animations 11 and 12. Among the 13 inclusions thus processed, some were undeterminable, but the only habit identified was rhombic dodecahedral.

SEM imagery of the laser-cut surface (Fig. 8) corroborates the inferences from the CT data concerning the two distinct morphologies of inclusions. Figure 8A shows a euhedral pseudomorph, in which the infilling minerals include florencite and kaolinite. Figure 8B shows a pair of elliptical, irregular inclusions that are also polymineralic.

Open megapores hundreds of micrometers and larger were relatively sparse, and these were principally observed near the rim of the specimen, though some are present in the region ∼1 mm inward from the patinaed surface. Some of these megapores have a similar size and shape to nearby inclusions (Fig. 6C), suggesting dissolution of previous grains or infilling of previous pores, or both. Similarly, open pores in a smaller size range (tens of micrometers) often have the same preferred orientation as nearby inclusions (Fig. 6C). With increasing HRXCT scan resolution, smaller pores become visible, and are observed in every 1-µm-resolution scan we obtained. However, their appearance is isolated, and the HRXCT data were not able to resolve directly continuous channels forming a connected network. The SEM imagery documented micropores of various sizes throughout the diamond matrix along the cut surface. Pores from tens of micrometers down to <1 µm width can be seen, and there were no broad areas that lacked pores at some level. Only the megapores were observed to have other minerals within them.

The polymineralic nature and disequilibrium texture common to all REE-containing inclusions indicate that they are exclusively secondary. Florencite is often a hydrothermal alteration product of monazite, which also tends to host high amounts of thorium and uranium. The occurrence of fission products (Fukura et al., 2005) and radiogenic lead in carbonado diamond matrix (i.e., Ozima and Tatsumoto, 1997) can be attributed to recoil implantation from these inclusions and pore fluids carrying components that have been mobilized. Overall, the observations in this study support the conclusion that all mega-inclusions in carbonado are secondary, and that the pore network is almost fully 3-D connected, as suggested previously based on leaching experiments (Dismukes et al., 1988).

However, the rhombic dodecahedral pseudomorphs almost certainly represent a phase that did form as an original inclusion in carbonado but has since been lost. Insofar as these pseudomorphs are many times larger than the diamond crystallites comprising their boundaries, we consider it unlikely that they could be some form of negative diamond crystal; we also consider it unlikely that they are themselves resorbed diamonds. The pristine faces and corners of the dodecahedra, combined with their presence throughout the sample volume, strongly suggest that they formed coevally with the diamond, and moreover did not suffer any resorption during diamond growth. Of the 30 or so minerals previously identified in carbonado, none has a rhombic dodecahedral crystal form. A variety of minerals may form rhombic dodecahedra (including, rarely, diamond or spinel), and this habit is commonly associated with garnet, a common inclusion phase in mantle diamonds (e.g., Spetsius and Taylor, 2008). However, it also seems evident that this phase must have subsequently become unstable, insofar as it has been entirely removed from this large specimen, and all other carbonados studied to date. Moreover, insofar as wide varieties of pseudomorphing minerals are present, from almost pure kaolinite in some to florencite mixed with other phases in others, at least some chemical components from the original phase must have migrated through the pore network.

The only large metallic inclusions were observed extremely close to the outside of the specimen, and probably connected to the exterior via relatively wide open porosity. The resolution of our data is not sufficient to image the nanoscale metallic inclusions documented by De et al. (1998), which were not observed to be in contact with any pore network. We thus infer that there are two populations of metallic particles in carbonado: Megaparticles are probably secondary pore fillings, whereas fully enclosed nano-inclusions are primary. The observation that the metal inclusions are confined to the very edge of this specimen suggests that they may even be a tertiary feature acquired after the other interior inclusions had already formed.

A striking feature of this sample is the spatial coherence of the patinaed surface paralleled with banding just underneath and apparent grading of pores extending throughout the specimen. This coherence implies that a single process or event was responsible for all of these textural features. The patinaed surface has been interpreted in other carbonados as possibly representing a fusion crust, which may be formed by passage through the atmosphere at high velocity, either as an incoming meteorite (Shelkov et al., 1997; Smith and Dawson, 1985), or as ejecta from an impact (Kletetschka et al., 2000). However, we consider it unlikely that either of these mechanisms would result in the observed bulk rearrangement of the internal volume of the diamond material, extending several millimeters inwards, i.e., much thicker than typical fusion crusts. For this to occur, there would have to be a transfer of carbon from the rim several millimeters into the interior through the confined pore network, and recrystallization there as diamond, without forming graphite, a similar physical argument to the one that rules out an impact origin (DeCarli, 1997).

A more likely alternative is that the small-inclusion patterns reflect an original pore network, which grades from almost nonporous near the patinaed surface to progressively more abundant and larger pores with increasing distance from this surface. The connectivity of the network implies that the pores were formed with a common, intercommunicating mechanism, such as a trapped grain-boundary fluid phase consistent with recent conclusions of carbonado forming in a fluid medium (Ishibashi et al., 2012).

Overall, the porosity in carbonado has a distinctive trimodal character. The first two modes consist of large euhedral pores up to millimeter scale caused by dissolution of originally included minerals, and preferentially elongated pores from 0.1 up to 0.3 mm in length with irregular boundaries, which appear to be isolated islands of original porosity. The third is micro-to nanoscale pores that constitute the connected network. Precipitation of secondary minerals appears to be almost exclusively within the large pore classes, probably owing to larger spaces being more favorable due to surface energy considerations. The principal exception is within the bright band adjacent to the patinaed surface, in which smaller pore spaces may be filled. This may be due to the possible presence of a locally pore-free, impermeable boundary immediately below the patinaed surface as identified by Shelkov et al. (1997), which could have formed a local trap for mineral-laden fluids, forcing them to precipitate their material in-place.

If the pore network indeed bears some genetic relation to the patinaed surface, the question arises of whether growth was toward or away from this surface. We favor the idea that growth was away from it. One possible explanation is that the initial layers of carbonado could more easily precipitate porosity free, and that as carbon was consumed from the fluid, residual fluid remained trapped inside the pores, inhibiting infill. An alternative possibility is that the large, irregular pores represent bubbles of fluid trapped in a diamond crystal mush during the early formation stages of carbonado, and that these bubbles buoyantly rose from the patinaed surface, which was down-facing. Also of note is that euhedral pores and elongate irregular pores both intersect the patinaed surface, and that the patinaed surface, while relatively smooth, is still rough in the manner of an impression on rock rather than rounded.

The elongation of the irregular inclusions and pores also requires explanation, as it implies that either these pores formed with this shape and orientation or were deformed by strain in the diamond matrix. We can speculatively link this texture with the two-stage carbonado growth model suggested by Petrovsky et al. (2010), in which the diamond material begins as a loose mush of 10–100 µm crystals, which is then sintered with smaller diamonds and cryptocrystalline material by a large increase in nucleation rate caused by a sudden change in some environmental factor, such as temperature decrease, pressure increase, or crystallization of a new phase. If this mush contained bubbles of liquid, which were then sheared during a relatively sudden event with an associated pressure increase that drove pressure significantly away from the diamond-graphite stability boundary, the deformation of the bubbles could be frozen in place by the rapid crystallization of microdiamonds. The contemporaneity of the shear strain and sintering provides an explanation for deformation lamellae (De et al., 1998) and high residual stress (Kagi et al., 2007) being confined to the smaller, later diamond population, which would be deforming even as it was crystallizing, transiently being a weaker and more mobile material, while the larger crystals remained undamaged. Another important constraint is provided by the pristine nature of the pseudomorphs, which show no signs of strain, suggesting that the diamond medium surrounding the dodecahedral phase was weaker than the dodecahedral phase itself. The two-stage hypothesis also explains the different carbon isotopes of the small and large diamond populations (De et al., 2001; Petrovsky et al., 2010). Opening of fractures in the crystalline mush during deformation and contemporaneous crystallization may also provide a mechanism for the formation of columnar diamond (Rondeau et al., 2008).

Even if the present-day mega-inclusions in carbonado are secondary, the unusual abundance of U, Th, REEs, and other incompatible elements in carbonado still calls for explanation. We suggest that for the sake of parsimony, this explanation should optimally be linked to the formation event. For this evidently rare diamond creation mechanism to be followed by a similarly rare incompatible-element enrichment event in a different environment and location, millions or billions of years later, in samples now distributed between two continents, is akin to lightning striking twice. On the other hand, carbonatitic high-density fluids trapped in fibrous diamonds are frequently rich in incompatible elements (Rege et al., 2010; Weiss et al., 2011).

An obvious place to look is to the now-absent dodecahedral phase. If that phase was sufficiently enriched in U and Th, over time it could become unstable due to metamictization from radiation damage, causing the crystal structure to collapse and new minerals to form in its place. As one possibility, garnet is capable of hosting high concentrations of U, Th, and REEs in hydrothermal conditions; Smith et al. (2004) document local U concentrations of up to 358 ppm and REE concentrations up to 4724 ppm in zoning bands within skarn garnets from the Beinn an Dubhaich granitic aureole in Skye, Scotland, which they attributed to periods of closed-system crystallization. Garnet retains all fission damage at temperatures below 200 °C (Haack and Potts, 1972), and this ability to accumulate radiation damage without annealing facilitates metamictization. After 1.7 b.y., the age of some conglomerates bearing carbonados (Sano et al., 2002), a garnet with 350 ppm U and Th/U ratio of 4 will accumulate a dose of 4.2 × 10¹⁸ alpha decays per gram, which is enough to cause significant swelling in zircon (Weber, 1990). However, whatever phase housed U and Th would have to grow in the highly reducing conditions implied by the metal nano-inclusions; Ishibashi et al. (2012) placed the likely oxygen fugacity of the diamond-forming fluid at ∼3 log units below the FMQ buffer, while Sautter et al. (2011) placed it even lower, at ∼15 log units below the IW buffer based on the presence of metallic Ti.

We thus postulate that the carbon-supersaturated fluid from which the diamond grew was also enriched in incompatible elements, perhaps in the manner of a pegmatite from the fluid residuum of a crystallizing carbonatitic magma body, albeit in a possibly very unusual setting, and that these elements partitioned into the dodecahedral phase, and perhaps others that we are not able to identify from the pseudomorphs. The enrichment may have also included other elements that are also represented in the present-day carbonado inclusion suite, such as K, P, and Al, with the first now helping comprise the kaolinite, the second the florencite, and so on. In other words, even though the present-day inclusion minerals in carbonado are almost certainly secondary, it does not necessarily follow that the materials that comprise them are. The radiogenic elements in the fluid may have also helped in diamond crystallization (e.g., Ozima and Tatsumoto, 1997), but it is not clear if such catalysis is necessary.

Figure 9 shows a schematic summary of carbonado formation as postulated here. In addition to incorporating all of the textural observations in this study, we believe that it is also consistent with the multitude of observations by other authors discussed in the foregoing text.

Left unspecified in this narrative is whether the carbonado formation process took place in the crust or the mantle. Most evidence points to a mantle origin, as the known processes that catalyze diamond formation at crustal pressures either include features not observed in carbonado or do not explain features that are observed (Petrovsky et al., 2010); an alternative possibility is a transient event producing mantle conditions in the deep crust (Sautter et al., 2011). However, the positing of a deep liquid so highly enriched in incompatible elements as to allow precipitation of millimeter-scale grains of a U-, Th-, and light (L) REE–rich phase is unconventional. A possible compensating circumstance is that carbonado formation occurred relatively early in the evolution of Earth, possibly at 3.8 Ga or even earlier, at a time when the segregation of incompatible elements into the crust was far from complete (Rollinson, 2006).

High-resolution X-ray CT enables “3-D petrography” to reveal a number of previously unrecognized features in a large specimen of carbonado diamond. These include a grading of oriented inclusions that is spatially associated with the patinaed surface, and the occurrence of dodecahedral pseudomorphs reflecting the first syngenetic inclusion phase above the nanometer scale recognized in carbonado, albeit in its absence. From these observations, we build upon recent work proposing that carbonado grew from a C-supersaturated fluid, and we propose that the irregular pores may correspond to trapped bubbles in a 10–100 µm loose or lightly bound diamond slurry, and that the observed fabric in these pores may reflect a strain event that sparked the growth and deformation of the smaller, <1 µm diamond population that holds carbonado together. We further postulate that this fluid may have also been highly enriched in a number of incompatible elements that were partitioned into primary inclusion phases, which subsequently broke down due to their high U and Th contents. While some rough edges and loose ends remain, this combined model encompasses a great number of the enigmatic observations that have made the origin of carbonado a mystery for several decades, and it does so with a single formation mechanism and setting. In addition, the proposed incompatible-element–rich fluid may provide a compact explanation for the observed differences between carbonado and other polycrystalline diamonds such as yakutites and framesites, suggesting that, in other respects, the mechanism for their origin may have been similar.

This research was supported in part by National Science Foundation grant EAR-0948842, and by the John A. and Katherine G. Jackson School of Geosciences and the Owen-Coates Fund of the Geology Foundation at the University of Texas at Austin. We thank Joseph Lai for performing the laser cut, Travis Clow for helping with inclusion segmentation, and J. Barnes, J. Lin, and D. Smith for helpful discussions. We appreciate the efforts of the editors and two anonymous reviewers in helping to improve the manuscript.