Large gem diamonds from metallic liquid in Earth’s deep mantle

See allHide authors and affiliations

Science  16 Dec 2016:
Vol. 354, Issue 6318, pp. 1403-1405
DOI: 10.1126/science.aal1303

Diamonds rock their metal roots

Massive diamonds are rare, expensive, and captivating. These diamonds now appear to be distinctive not only in their size but also in their origin. Smith et al. probed mineral inclusions from these very large diamonds and found abundant slivers of iron metal surrounded by reducing gases. This suggests that the large diamonds grew from liquid metal in Earth's mantle. The inclusions also provide direct evidence of a long-suspected metal precipitation reaction that requires a more reducing mantle.

Science, this issue p. 1403


The redox state of Earth’s convecting mantle, masked by the lithospheric plates and basaltic magmatism of plate tectonics, is a key unknown in the evolutionary history of our planet. Here we report that large, exceptional gem diamonds like the Cullinan, Constellation, and Koh-i-Noor carry direct evidence of crystallization from a redox-sensitive metallic liquid phase in the deep mantle. These sublithospheric diamonds contain inclusions of solidified iron-nickel-carbon-sulfur melt, accompanied by a thin fluid layer of methane ± hydrogen, and sometimes majoritic garnet or former calcium silicate perovskite. The metal-dominated mineral assemblages and reduced volatiles in large gem diamonds indicate formation under metal-saturated conditions. We verify previous predictions that Earth has highly reducing deep mantle regions capable of precipitating a metallic iron phase that contains dissolved carbon and hydrogen.

Earth has a metallic liquid outer core, and it has been predicted from theory and experiments that the deep mantle could precipitate iron alloys. The presence of such metallic iron phases, especially in high enough abundance, would have profound effects on the physical and chemical properties of Earth’s deep mantle. However, the inaccessibility of the deep Earth makes it challenging to observe. Upwelling mantle regions melt adiabatically, producing prolific basaltic volcanism and silicate mantle residues, both of which appear too oxidized to have originated from a deeper region of metal saturation. Thus, physical evidence for such reducing regions—essential for understanding mantle evolution—has been virtually absent.

We have identified a genetically distinct diamond population that samples these metal-saturated regions of Earth’s mantle. These diamonds are typified by the 3106-carat (1 carat = 0.2 g) Cullinan diamond and are almost always classified as type II, referring to their minimal nitrogen content (<5 to 20 parts per million). More specifically, diamonds that are Cullinan-like tend to be large, inclusion-poor, relatively pure, irregularly shaped, and resorbed (13). Combining the traits into an acronym leads us to term them CLIPPIR diamonds to distinguish them from other populations of diamond, especially other varieties of type II diamonds lacking these features (2, 4).

Examination of 53 CLIPPIR diamonds with inclusions shows that a magnetic, metallic inclusion is the most common trapped material (Fig. 1 and fig. S1), appearing as the only inclusion in 38 of the 53 diamonds. Similar inclusions have been mistakenly identified as graphite on the basis of appearance (1, 2) because they are camouflaged by prominent graphitic decompression cracks. Five metallic inclusions from the 38 diamonds were exposed by polishing for chemical microanalysis (Fig. 2). We found a multiphase assemblage composed primarily of cohenite [(Fe,Ni)3C], an interstitial Fe-Ni alloy, iron sulfide (pyrrhotite) segregations, and some more minor accessory phases (Fe-phosphate, Cr-Fe-oxide, and Fe-oxide) (Fig. 2 and table S1). X-ray diffraction confirms the identification of the carbide as cohenite (5). Small amounts of graphite occur at the diamond-inclusion interface and the fractures radiating from inclusions. We detected a thin fluid jacket of CH4 around most of the inclusions by Raman spectroscopy (5). In 13 samples, H2 was also detected, accompanying intense CH4 signals (Fig. 1). Taking these observations together, we interpret the inclusions as former Fe-Ni-C-S melt with minor dissolved H, P, Cr, and O, indicating a reducing environment.

Fig. 1 Representative Fe-Ni-C-S melt inclusion.

Raman maps show CH4 and H2 concentrated at the inclusion nucleus, with warmer colors representing higher intensities (sample 100517599181). Equal vertical scaling is applied to both the CH4 and H2 spectra. The vertical axis is intensity in arbitrary units.

Fig. 2 Scanning electron microscopy x-ray maps of Fe-Ni-C-S melt inclusions.

(A and B) Cohenite [(Fe,Ni)3C] (green) is surrounded by interstitial Fe-Ni alloy (pink) and segregations of Fe-rich sulfide (teal), likely pyrrhotite. Detailed Ni, Fe, and C maps for the dashed area in (B) show that the cohenite is relatively Ni-poor but Fe- and C-rich. (C) Fe-phosphate bleb within sulfide and a large cohenite grain (no Fe-Ni alloy intersected). The samples shown are Letseng_889, inclusion E (A); Letseng_890, inclusion A (B); and OC2, inclusion D (C).

The remaining 15 diamonds contain inclusions of silicate minerals from a high-pressure origin, such as Cr-poor majoritic garnet (5) and CaSi-perovskite (CaPv) inverted to lower pressure phases (figs. S2 and S3). Some of these silicates also have coexisting metal and CH4 ± H2 fluid trapped in the same inclusion (figs. S2, S3, and S4). We therefore infer that these silicate inclusion assemblages were trapped under similar reducing conditions as the metal-only population at minimum pressures of 12 GPa (360 km depth). The presence of garnet precludes a deeper origin than 750 km because of the maximum stability of the mineral in eclogite (6, 7). These constraints suggest that CLIPPIR diamonds may form within the mantle transition zone at 410 to 660 km depth.

The rare previous reports of various native Fe, Fe-Ni, and Fe-carbide inclusions in diamond were interpreted as unusual and isolated occurrences (810). Though such inclusions are indicative of reducing conditions, these reports have fallen short of establishing systematic genetic relationships. The physical characteristics of these previously studied diamonds, and the textures and mineral assemblages of their metallic inclusions, are distinctly different from those of CLIPPIR diamonds.

We did not observe wüstite or ferropericlase—normal products of carbonate reduction reactions (11)—in the assemblage, which rules out previous models proposed for other sublithospheric diamonds (12). Instead, multiple lines of evidence support a model of diamond growth from metallic liquid. Previous experiments have demonstrated diamond growth from a Fe-Ni-C-S melt, which solidifies to a Fe-carbide, Fe-Ni alloy, sulfide, and graphite assemblage that mimics the natural Fe-Ni-C-S inclusions in CLIPPIR diamonds (13). The preserved CH4 and H2 characteristic of CLIPPIR diamonds are likewise observed in synthetic diamonds grown in molten Fe-Ni alloy. Inclusions in synthetic diamonds are often carbides or other alloys with fluid CH4 and H2 from hydrogen inadvertently dissolved in the metallic liquid (14). In general, a metallic liquid is a favorable medium for growing large diamonds such as CLIPPIR diamonds, with few inclusions and little or no chemical zonation, because the carbon supply is well buffered and carbon diffuses rapidly (13).

The growth of large diamonds (>5 cm) in the mantle might be accommodated by a liquid pocket that provides the unobstructed space necessary at deep mantle pressure. The high temperature of the sublithospheric mantle could enhance dislocation mobility, allowing formation of the dislocation networks commonly seen in CLIPPIR diamonds (5). The characteristically low nitrogen content in CLIPPIR diamonds might also be explained by N partitioning into the metal (4) or high-pressure mantle nitride phases (15).

The mantle likely becomes more reducing with depth, and thus capable of Fe-Ni precipitation (16). This hypothesis is supported by multiple lines of theoretical and experimental evidence suggesting Fe-Ni metal saturation below 250 ± 30 km (1721). The increasing stability of Fe3+ in subcalcic pyroxene, majoritic garnet, and eventually (Fe,Mg)SiO3 (bridgmanite) with pressure causes progressive disproportionation by the reaction 3Fe2+ → 2Fe3+ + Fe0. Whereas Fe3+ is maintained within the silicates, Fe0 exsolves as a stable metallic phase. The reduction of the liquidus temperature from the dissolution of C and S into the metallic phase allows it to be fluid at mantle conditions (21).

We suggest that the Fe-Ni-C-S melt inclusions reported here are samples of this metallic liquid. The metallic inclusions may be direct evidence of charge disproportionation and the resulting limited activity of oxygen in the transition zone. These redox conditions may be widespread and persistent through time, given that CLIPPIR diamonds are found in kimberlites from different continents with emplacement ages spanning at least 1 billion years. For example, the Cullinan diamond was recovered from the 1.18-billion-year-old Premier kimberlite in South Africa (22), whereas several of our specimens come from the 90-million-year-old Letseng kimberlite in Lesotho (1).

Dissolved carbon must reach supersaturation to crystallize diamond from a metallic liquid at compositions in which diamond is the liquidus phase. In a purely Fe-C system, the increase in liquidus temperature with increasing pressure outpaces mantle geotherms and suggests that a C-rich metallic melt may intersect the liquidus and begin crystallizing diamond if it is transported downward (23). An alternative mechanism that could trigger C crystallization is a change in the melt composition, such as an increase in S concentration (24) or a decrease in Fe content. Direct assimilation of introduced C and other minor components such as H and O, perhaps during subduction, could also drive diamond crystallization at deeper levels in the mantle. As diamond growth proceeds within the metallic liquid, occasional droplets of the Fe-Ni-C-S melt are included, as shown by the CLIPPIR metallic inclusions.

Other aspects of the CLIPPIR suite of diamonds and their metallic inclusions potentially constrain the host rock for the metallic liquid, suggest a storage capacity for other volatiles, and lend insight into carbon subduction. The modest Ni/Fe ratios of the metal, though also a function of oxygen fugacity, suggest either a lower mantle origin or an association with subducted eclogite, because they are too low to be derived from more Ni-rich upper mantle peridotite (21). The abundance of CaPv, sometimes with a CaTiO3 component, and the presence of Cr-poor majoritic garnets support an association with subducted eclogite at the depth of the transition zone or uppermost lower mantle (Fig. 3).

Fig. 3 Model of CLIPPIR diamond formation.

Formation of metallic iron proceeds in subducting eclogite by disproportionation. Metal segregation may be aided by deformation of the subducting slab in the transition zone. The liquid metal composition evolves to Fe-Ni-C-S, also dissolving P and H. Diamond crystallization occurs within metallic liquid pockets, likely in the pressure range of 12 to 25 GPa. Pocket walls become a site of CaSi-perovskite (CaPv) crystallization, where it can be included in diamond. Carbon saturation is achieved by increasing pressure, assimilating further C, or another mechanism such as increasing S content. After growth, diamonds are physically separated from the growth environment and may be transported and entrained by a kimberlite eruption.

Variably light carbon isotopic compositions measured in seven of the CLIPPIR diamonds range in δ13C from –26.9 to –3.8 per mil (table S2), consistent with the similarly large range of values observed in eclogitic diamonds and one previous study of type II diamonds (25). Such isotopically light compositions are thought to result from the recycling of crustal carbon. Intuitively, one might consider eclogite to be too oxidized to permit the precipitation of an iron liquid, but at depth, subducting eclogite can be comparable to or better than peridotite at generating metallic iron (5, 26, 27). In an eclogite host, mantle metals should remain in small, isolated intergranular pockets (21), but localized rock deformation associated with convection can encourage metal segregation (28) that could further localize strain. Interconnectivity and agglomeration of the metallic liquid may be essential for efficient C scavenging and, ultimately, diamond growth. Partitioning of Ni, C, S, P, H, and other elements from the eclogite into the metal would accompany the creation of a liquid metallic iron alloy (24).

The common occurrence of cohenite in these inclusions shows that Fe-Ni metal can dissolve substantial C, supporting a role of Fe-Ni metal in reduced portions of the mantle in the deep carbon cycle (24, 29). Furthermore, the growth of CLIPPIR diamonds themselves indicates that the metal can facilitate extreme local excesses of pure C in the transition zone or deeper mantle. A high solubility of H in the high-pressure liquid metal is confirmed by hydrogen in the Fe-Ni-C-S inclusions, which suggests that metallic Fe-Ni may also contribute to the deep Earth hydrogen budget. In this regard, the storage and cycling of any element that is soluble in the metallic liquid merit consideration. Furthermore, if present in high enough concentrations in mantle rocks, Fe-Ni metallic liquid could affect mantle rheology. It might act as a weak phase, like ferropericlase in the lower mantle, that accommodates more strain than neighboring strong phases and leads to shear localization (30).

Our observations verify and further constrain the prediction that the mantle has regions that are reducing enough to precipitate iron alloys. We know that there must be large variations in the redox conditions of the mantle, given the contrast between Earth’s Fe-Ni core and the oxidized silicate lithosphere. These direct, diamond-based observations of regionally metal-saturated conditions in the upper mantle imply similar reducing conditions elsewhere in the mantle. Previous experiments and theory suggest comparably reducing conditions in the D′′ layer, within large low shear-wave velocity provinces, at higher levels in the lower mantle, and in the transition zone (16, 17, 1921, 27, 29, 31). The presence of metal has implications for the seismic velocity, thermal and electrical conductivity, rheology, and volatile element cycling in Earth’s deep mantle (16, 29, 32).

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S4

Tables S1 to S2

References (3340)

Data S1 to S4

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
Acknowledgments: Sincere thanks go to C. Locke, K. S. Moe, U. D’Haenens-Johansson, P. Johnson, and T. Moses for help with sample selection; A. Chan and J. Lai for help with sample preparation; J. Liao for photography; and K. Smit, L. Loudin, A. Balter, A. Shahar, and D.G. Pearson for valuable discussion and feedback. J. Butler is thanked for donating a sample to this study. Carbon standards were generously lent by C. Jackson and Z. Du. Anonymous reviewers are thanked for their constructive comments. The Deep Carbon Observatory is acknowledged for support to S.B.S. and F.N., the National Science Foundation for support to S.B.S. and J.W. (grant no. EAR1049992), and the European Research Council for support to F.N. (INDIMEDEA, grant no. 307322). Data files with a sample catalog, microprobe analyses, Raman spectra, and x-ray diffraction analyses are available in the supplementary materials.

Stay Connected to Science

Navigate This Article