Blue Boron-Bearing Diamonds from Earth's Lower Mantle

Home , Almandine, Fracture (mineralogy), Kyanite

Letter https://doi.org/10.1038/s41586-018-0334-5

Blue boron-bearing diamonds from Earth’s lower mantle Evan M. Smith1*, Steven B. Shirey2, Stephen H. Richardson3, Fabrizio Nestola4, Emma S. Bullock5, Jianhua Wang2 & Wuyi Wang1

Geological pathways for the recycling of Earth’s surface materials in sublithospheric diamonds tend to destabilize during ascent in the into the mantle are both driven and obscured by plate tectonics1–3. mantle and break down to lower-pressure minerals, often unmixing Gauging the extent of this recycling is difficult because subducted into composite assemblages14,17. Many inclusions described here are crustal components are often released at relatively shallow depths, multiphase assemblages, as is the case with previously studied inclu- below arc volcanoes4–7. The conspicuous existence of blue boron- sions in super-deep diamonds and their high-pressure experimental bearing diamonds (type IIb)8,9 reveals that boron, an element analogues14,16,17. It is implausible that the same multiphase assemblages abundant in the continental and oceanic crust, is present in could be coincidentally replicated by random sampling of lower- certain diamond-forming fluids at mantle depths. However, both pressure mineral aggregates at shallower, lithospheric depths14. the provenance of the boron and the geological setting of diamond The most abundant inclusion identified, in 31 of 46 samples, was crystallization were unknown. Here we show that boron-bearing Ca-silicate dominated by CaSiO3 walstromite, sometimes with larnite diamonds carry previously unrecognized mineral assemblages (β-Ca2SiO4) and other phases of CaSiO3 composition (Extended Data whose high-pressure precursors were stable in metamorphosed Table 1). These inclusions are commonly interpreted as retrogressed 15,17,18 oceanic lithospheric slabs at depths reaching the lower mantle. We CaSiO3 perovskite (Ca-Pv) . As retrogression of pure Ca-Pv alone propose that some of the boron in seawater-serpentinized oceanic should maintain a bulk Ca:Si ratio of 1, the presence of (Ca-rich) larnite lithosphere is subducted into the deep mantle, where it is released in some inclusions may indicate that diamond growth occurred in a with hydrous fluids that enable diamond growth10. Type IIb chemically evolving system with variable calcium enrichment, as seen diamonds are thus among the deepest diamonds ever found and in other super-deep diamonds16,19,20. indicate a viable pathway for the deep-mantle recycling of crustal Other observed inclusions also correspond to retrogressed high- elements. pressure minerals (Extended Data Table 1). For example, inclusions Type IIb diamonds—including the Hope, a renowned blue of orthopyroxene, with sharp Raman spectra matching enstatite, and diamond—are mantle-derived minerals that contain boron at 0.01– minor amounts of coexisting olivine are interpreted as retrogressed 10 p.p.m. levels and show a lack of nitrogen absorption in infrared bridgmanite14,15, the lower-mantle Mg-silicate perovskite phase. spectroscopy8. Boron imparts their blue colour and p-type semi- Multiphase inclusions containing ortho- or clinopyroxene, coexisting conductivity, although they may not always appear blue when boron with jeffbenite (Mg3Al2Si3O12) or spinel ((Mg,Fe)Al2O4), are inter- concentrations are low or additional defects are present9. Because preted as aluminous bridgmanite14, although some bearing clinopy- boron is a quintessential crustal element with a low concentration in roxene may represent retrogressed majoritic garnet17. Earth’s mantle11, blue diamonds and their formation have long been a One inclusion provides a convincing example of retrogressed geochemical enigma. majorite. Being fortuitously exposed on a facetted diamond, the Type IIb diamonds have been recovered from worldwide local- two-phase assemblage of NaAl-clinopyroxene and jeffbenite (Fig. 1b) ities, including southern and central Africa, India, South America was confirmed with microanalysis by energy-dispersive X-ray spectros- and Borneo9, having been brought to the surface in kimberlite copy (Extended Data Fig. 3) and was interpreted as a former low-Ca, volcanoes ranging in age from the 1.15-billion-year-old Premier pipe12 high-Na majoritic garnet17. A separate inclusion of orthopyroxene in to the ~90-million-year-old Letseng deposit13. They can reach large this diamond, interpreted as former bridgmanite, would then make sizes, such as the 176.2-carat Brazilia diamond, the 122.5-carat rough a putative majorite–bridgmanite pair that would restrict its origin to diamond that yielded the (24.18-carat) Cullinan Dream, which was within17 ~660–750 km. Other observed inclusion phases are coesite examined as part of this study, and the 112.5-carat rough diamond (with accessory kyanite, interpreted as former stishovite) as well as from which the Hope was cut9. ferropericlase (found as relatively small, brown inclusions; see Extended The geological origin of blue diamonds has nevertheless remained Data Fig. 4 and Supplementary Table 1). unknown owing to their rarity (≤0.02% of mined diamonds; Another diamond contains a multiphase inclusion dominated by see Methods), high value and general lack of mineral inclusions. To over- nepheline and spinel, interpreted as former calcium-ferrite-type (CF) come this problem, prospective samples were screened from the extensive phase or possibly new aluminous (NAL) phase, which is compelling grading operations of the Gemological Institute of America. Over two evidence of derivation from host rocks of basaltic composition at years, this approach allowed examination of 46 type IIb diamonds with lower-mantle depths (Fig. 1)14,16,17. The same diamond also contains a inclusions, an invaluable suite for analysis (Extended Data Fig. 1). multiphase inclusion of Fe carbide, Fe sulphide and wüstite (Extended Inclusions were characterized using Raman spectroscopy (Fig. 1) and Data Fig. 5, Supplementary Table 2) that does not correspond to a were found to differ substantially from the common minerals found known mineral but may represent a former metallic melt similar to in diamonds from the cratonic lithosphere (<200 km), such as olivine those recently discovered in (boron-lacking) CLIPPIR (Cullinan-like, and Cr-rich pyrope from peridotite or grossular-almandine-pyrope and Large, Inclusion-Poor, relatively Pure, Irregularly shaped and Resorbed) omphacitic clinopyroxene from eclogite. Instead, the inclusion mineral- diamonds21. Three other type IIb samples also contain metallic-looking, ogy is typical of super-deep diamonds originating from the mantle tran- magnetic inclusions that may be similar to those of CLIPPIR diamonds, sition zone to the lower mantle (Extended Data Fig. 2)14–17. Inclusions although it should be stressed that these are a minor part of the type IIb

1Gemological Institute of America, New York, NY, USA. 2Department of Terrestrial Magnetism, Carnegie Institution for Science, Washington, DC, USA. 3Department of Geological Sciences, University 4 5 of Cape Town, Rondebosch, South Africa. Department of Geosciences, University of Padova, Padua, Italy. Geophysical Laboratory, Carnegie Institution for Science, Washington, DC, USA. *e-mail: [email protected]

663 a a CH4 990 2,917 cm-1 CaSiO3 walstromite 1,049 y

Reference Intensit

684 100 μm NaAl-pyroxene 1,023 357 b 200 μm 2,850 2,9002,950 Raman shift (cm-1) b 866 H 927 CH4 2 Jeffbenite 642 990 2,918 cm–1 588 cm–1 506 320 H2 4,157 cm–1 y 100 μm y 531 c 550 600 Intensit Intensit 195 Coesite 139 273 430

Kyanite 2,9002,950 4,100 4,1504,200 100 μm Raman shift (cm–1)

Fig. 2 | Inclusions jacketed by thin films of fluid CH4 and H2, revealed 100 μm 430 Nepheline by Raman spectroscopy. a, CaSiO3 walstromite (former Ca-Pv), 400 with the main part of the inclusion circled, in sample 110208780369. 471 987 d b, Orthopyroxene (former bridgmanite) in sample 110208773706. Olivine Lobate sprays of small inclusions are thought to reflect expansion and proliferation of inclusion material into its own decompression crack.

560 645 753 404 Spinel of at least ~9 GPa in the mantle (we note that these entrapment pressures are severe underestimates due to diamond deformation; Extended 100 μm Data Fig. 6)22. Similarly, a coesite inclusion with its main Raman peak shifted from 520.6 cm−1 up to 537.9 ± 0.5 cm−1 (Extended Data Fig. 6) 200 400 600 800 1,000 1,200 indicates extreme remnant pressure, far exceeding the 4 GPa highest- -1 Raman shift (cm ) pressure benchmark for coesites observed in lithospheric diamonds Fig. 1 | Selected Raman spectra of inclusions in type IIb diamonds. (see Methods). X-ray diffraction also reveals remnant pressures of a, Former Ca-Pv, now CaSiO3 walstromite, in sample 110205945970. ~1.8 GPa in ferropericlase inclusions, requiring a minimum entrap- b, Former majoritic garnet, now a composite of NaAl-pyroxene and ment pressure of 10.3 to 14.1 GPa, calculated at 1,200 K and 2,000 K, jeffbenite (verified by scanning electron microscopy/energy-dispersive respectively, corresponding to a depth beyond 300 km, well below the X-ray spectroscopy; see Extended Data Fig. 3) in sample 880000037816. base of the deepest continental lithospheric mantle keels23. c, Former stishovite, now coesite, in sample 101024478345. Also shown on Additional pressure observations come from the silicate inclusions the left is a composite coesite plus kyanite spectrum (green) from sample 890000180201. d, Former CF, now composite of nepheline and spinel in blue diamonds, which typically have large decompression cracks (with CH4 fluid, not shown), in sample 110208245246. Dashed lines are and often lobate sprays of tiny droplet-like satellite inclusions in 19 30 reference spectra of CaSiO3 walstromite , jeffbenite , nepheline and healed fractures emanating from the main inclusion (Figs. 1, 2). These spinel14, plus omphacite R061129 and coesite X050094 from the RRUFF satellite inclusions formed during exhumation, as high internal inclu- database. Spectra are stacked vertically for clarity. Source Data. sion stresses20 relative to decreasing external confining pressure led to rupturing of the host diamond. Pervasive dislocation networks in diamond inclusion suite, whereas CLIPPIR diamonds are dominated type IIb diamonds (Extended Data Fig. 7) from plastic deformation and by metallic Fe–Ni–C–S inclusions21. recovery24 are also consistent with a high-temperature, sublithospheric The observed inclusion mineralogy and the absence of typical silicate mantle origin. Equivalent dislocation networks also appear in sublith- inclusions that characterize diamonds from the continental lithospheric ospheric type IIa, CLIPPIR diamonds21, but thus far have not been mantle advocate for type IIb diamond growth in host rocks of basaltic- reported in diamonds from the cooler, shallower lithospheric mantle. to-peridotitic bulk composition, consistent with the lithologies in Methane, and often hydrogen, (CH4 ± H2) were detected by Raman subducted oceanic lithosphere reaching lower-mantle depths. Although spectroscopy in 13 diamonds (28% of samples) as a thin fluid layer some samples, particularly those containing Ca-Pv alone, may have around one or more inclusions of varying mineralogy (Fig. 2). The fluid grown in the mantle transition zone, the inclusion assemblages with is a result of hydrogen escaping the inclusion and accumulating at the former bridgmanite, ferropericlase and CF phase require an origin in inclusion–host interface. When hydrogen is surrounded by diamond, the lower mantle14,17. it can form methane by reacting with the surrounding carbon. Similar In addition to inclusion mineralogy, a sublithospheric origin for blue CH4 ± H2 fluid jackets have been found around metallic melt inclu- diamonds is physically required by the extreme remnant pressure in sions in natural and synthetic diamonds, formed when formerly some inclusions (see Methods). Using the pressure-induced shift in dissolved atomic hydrogen diffused out of the inclusions upon cooling 21 Raman spectral features, a 4.4 ± 0.1 GPa remnant pressure was meas- and decompression . In the present mineral inclusions, these CH4 ± H2 ured in CaSiO3 walstromite, which requires entrapment at a pressure fluids are a strong indication that at least some of the retrograde

Kimberlite eruption Seawater boron in serpentinite geotherms could successfully stabilize dense hydrous magnesium sili- cates (DHMSs) before serpentine breakdown, thereby creating a min- l lithosphere Continenta 1 eralogical pathway for boron, water and other components to reach phere 1,10,28 Serpentinite os the mantle transition zone and beyond . The depth range of blue th to DHMS 2 li ic diamond genesis may therefore relate to DHMS breakdown, releasing n a 10 e 5 c 410 km hydrous diamond-forming fluid as well as liberating boron (Fig. 3). O Eventual transport of the blue diamonds to the surface is thought to Upwelling Transition zone mantle involve upwelling mantle and kimberlite volcanism, as with other 660 km super-deep diamonds. Growth of 4 Blue diamonds point to a subduction-driven geochemical pathway on-bearing 3 bor DHMS Lower mantle diamond extending from serpentinized oceanic lithosphere at Earth’s surface to breakdown the lower mantle. This picture is consistent with key features of blue diamonds: mineral inclusions of basaltic-to-peridotitic bulk com- 14 Fig. 3 | Formation of type IIb diamond. (1) Seafloor hydrothermal position with affinity to subducted slabs ; high-pressure mineral circulation causes serpentinization, introducing boron into oceanic assemblages indicating a depth of origin reaching the lower mantle, lithosphere. (2) Subduction and metamorphism of serpentine to DHMS. coinciding with the projected breakdown of DHMS minerals10; (3) Breakdown of DHMS yields hydrous, boron-enriched fluid that CH4 ± H2 fluids suggestive of hydrous media; and, most remarkably, migrates and evolves. It may gather carbon from the altered oceanic the boron content of the diamonds themselves. If metamorphosed lithosphere. (4) Crystallization of boron-bearing diamond, triggered serpentinite ultimately provides the boron for type IIb diamonds, as by redox reactions or in response to changing pressure, temperature or proposed here, then it is also implicated as a carrier of water to the composition of the fluid. (5) Vertical transport may involve localized deep mantle28,29. The recognition that blue diamonds originate from buoyancy associated with diamond-related metasomatism or an external mechanism such as a plume, with ultimate exhumation to the surface due the lower mantle highlights a possible major pathway for ultra-deep to kimberlite volcanism. water recycling on Earth. Online content mineral assemblage is hydrogen-saturated, implying that the original Any Methods, including any statements of data availability and Nature Research high-pressure minerals interacted with hydrous media. reporting summaries, along with any additional references and Source Data files, Blue diamonds exhibit an inclusion mineralogy that requires are available in the online version of the paper at https://doi.org/10.1038/s41586- diamond formation in host rocks of basaltic-to-peridotitic bulk com- 018-0334-5. position in the lower mantle, much like inclusions described in other Received: 12 January 2018; Accepted: 30 April 2018; lower-mantle diamonds that have been linked to subducted oceanic Published online 1 August 2018. lithosphere14–18. Inclusions of basaltic association, such as former stishovite or CF phase, clearly point to the involvement of subducted 14,16 1. Kendrick, M. A. et al. Seawater cycled throughout Earth’s mantle in partially ocean crust , whereas those of peridotitic association, such as serpentinized lithosphere. Nat. Geosci. 10, 222–228 (2017). ferropericlase or low-Al bridgmanite, may be more closely related to the 2. Tackley, P. J. Mantle convection and plate tectonics: toward an integrated peridotitic portion of the oceanic lithospheric slab15. These subduction- physical and chemical theory. Science 288, 2002–2007 (2000). 3. Hofmann, A. W. Mantle geochemistry: the message from oceanic volcanism. related mineral assemblages are supported by a range of light-to-heavy Nature 385, 219–229 (1997). carbon isotope signatures16,25 in type IIb diamond, with three samples 4. Leeman, W. P., Tonarini, S. & Turner, S. Boron isotope variations in Tonga- having a δ13C of −13.4‰, −3.4‰ and −1.8‰ (δ13C is the parts-per- Kermadec-New Zealand arc lavas: implications for the origin of subduction 13 12 components and mantle infuences. Geochem. Geophys. Geosyst. 18, thousand deviation of C/ C from the Pee Dee belemnite standard; 1126–1162 (2017). see Supplementary Table 3), which complement three diamonds from a 5. Deschamps, F., Godard, M., Guillot, S. & Hattori, K. Geochemistry of subduction previous study with a δ13C range of −20.8‰ to −14.5 ‰26. zone serpentinites: A review. Lithos 178, 96–127 (2013). 6. Plank, T. & Langmuir, C. H. The chemical composition of subducting sediment Boron is scarce in the convecting mantle, being 100 times more and its consequences for the crust and mantle. Chem. Geol. 145, 325–394 5,11 depleted compared to Earth’s surface . Its occurrence in blue dia- (1998). monds implies an anomalously boron-enriched mantle source, espe- 7. Konrad-Schmolke, M., Halama, R. & Manea, V. C. Slab mantle dehydrates beneath Kamchatka—yet recycles water into the deep mantle. Geochem. cially considering that boron behaves incompatibly in diamond growth Geophys. Geosyst. 17, 2987–3007 (2016). experiments. Although present observations do not rule out boron con- 8. Gaillou, E., Post, J. E., Rost, D. & Butler, J. E. Boron in natural type IIb blue tribution from the ambient convecting mantle, the clear subducted- diamonds: chemical and spectroscopic measurements. Am. Mineral. 97, 1–18 (2012). host-rock signatures strongly suggest that the boron is surface-derived, 9. King, J. M. et al. Characterizing natural-color type IIb blue diamonds. Gems even if only small amounts of the full boron budget of the slab are Gemol. 34, 246–268 (1998). subducted. Boron isotopic measurements may clarify its origin, but 10. Harte, B. Diamond formation in the deep mantle: the record of mineral the prerequisite for non-destructive analysis, the lack of standards and inclusions and their distribution in relation to mantle dehydration zones. Mineral. Mag. 74, 189–215 (2010). the low boron concentrations in type IIb diamonds have been barriers 11. Grew, E. S. Boron: from cosmic scarcity to 300 minerals. Elements 13, 225–229 to the use of current techniques. Nevertheless, support for a crustal (2017). boron source comes from the actual distribution of boron on Earth11, 12. Wu, F. Y. et al. In situ U-Pb age determination and Sr-Nd isotopic analysis of perovskite from the Premier (Cullinan) kimberlite, South Africa. Chem. Geol. the clear geophysical evidence of subducted slabs reaching the lower 353, 83–95 (2013). mantle2 and the already established link between certain lower-mantle 13. Skinner, E. & Truswell, J. in The Geology of South Africa (eds Johnson, M. R. et al.) diamonds and subducted oceanic crust14. 651–659 (Geological Society of South Africa, Johannesburg, 2006). 14. Walter, M. J. et al. Deep mantle cycling of oceanic crust: evidence from In deeply subducted oceanic lithosphere, the most likely boron diamonds and their mineral inclusions. Science 334, 54–57 (2011). source is metamorphosed serpentinized peridotite, with boron orig- 15. Stachel, T., Harris, J. W., Brey, G. P. & Joswig, W. Kankan diamonds (Guinea) II: inally introduced by hydrothermal seawater circulation4,5,7 (Fig. 3). lower mantle inclusion parageneses. Contrib. Mineral. Petrol. 140, 16–27 (2000). Other boron carriers, such as sediments, organic matter or white 16. Thomson, A. et al. Origin of sub-lithospheric diamonds from the Juina-5 micas in the ocean crust, are destabilized at relatively shallow depths, kimberlite (Brazil): constraints from carbon isotopes and inclusion but serpentinized peridotite in the lithospheric-mantle portion of the compositions. Contrib. Mineral. Petrol. 168, 1081 (2014). slab can serve as an effective vehicle and is expected to be the largest 17. Harte, B. & Hudson, N. C. F. In Proc. 10th International Kimberlite Conference Vol. 4,5,7 1 (eds Pearson, D. G. et al.) 235–253 (Springer, New Delhi, 2013). boron reservoir within the slab . Like water, boron is structurally 18. Nestola, F. et al. CaSiO3 perovskite in diamond indicates the recycling of oceanic hosted in serpentine27. Under the right conditions, cool subduction crust into the lower mantle. Nature 555, 237–241 (2018).

19. Brenker, F. E. et al. Detection of a Ca-rich lithology in the Earth’s deep provided by the Deep Carbon Observatory (DCO). The European Research (> 300 km) convecting mantle. Earth Planet. Sci. Lett. 236, 579–587 (2005). Council supported F.N. (INDIMEDEA, number 307322). Sincere thanks to 20. Anzolini, C. et al. Depth of formation of CaSiO3-walstromite included in K. S. Moe, T. Moses, M. Breeding, U. D’Haenens-Johansson, P. Johnson, K. Smit, super-deep diamonds. Lithos 265, 138–147 (2016). J. Liao, S. Persaud, E. Myagkaya, A. Balter and B. Torres for analytical/logistical 21. Smith, E. M. et al. Large gem diamonds from metallic liquid in Earth’s deep assistance, to N. Renfro for the micrograph of Fig. 1a, to Ascot Diamonds for mantle. Science 354, 1403–1405 (2016). lending rough samples and to M. Alvaro for discussions about geobarometry. 22. Anzolini, C. et al. Depth of formation of super-deep diamonds: Raman barometry of CaSiO3-walstromite inclusions. Am. Mineral. 103, 69–74 (2018). Reviewer information Nature thanks E. Gaillou and T. Stachel for their 23. Angel, R. J., Mazzucchelli, M. L., Alvaro, M. & Nestola, F. EosFit-Pinc: a simple GUI contribution to the peer review of this work. for host-inclusion elastic thermobarometry. Am. Mineral. 102, 1957–1960 (2017). 24. Hanley, P. L., Kifawi, I. & Lang, A. R. On topographically identifable sources of Author contributions E.M.S. led the research, characterized the samples, cathodoluminescence in natural diamonds. Phil. Trans. R. Soc. Lond. A 284, conducted Raman analyses, interpreted results and wrote the initial 329–368 (1977). manuscript. S.B.S. and S.H.R. contributed scientific interpretations and 25. Stachel, T., Harris, J., Aulbach, S. & Deines, P. Kankan diamonds (Guinea) III: substantive manuscript writing. F.N. conducted X-ray diffraction and assisted d13C and nitrogen characteristics of deep diamonds. Contrib. Mineral. Petrol. with geobarometry. E.S.B. provided support for electron microprobe analysis. 142, 465–475 (2002). J.W. conducted mass spectroscopy for carbon. W.W. helped guide the project 26. Milledge, H. J. et al. Carbon isotopic variation in spectral type II diamonds. and ensured access to samples and analytical resources. Nature 303, 791–792 (1983). 27. Pabst, S. et al. Evidence for boron incorporation into the serpentine crystal Competing interests The authors declare no competing interests. structure. Am. Mineral. 96, 1112–1119 (2011). 28. Ohtani, E., Litasov, K., Hosoya, T., Kubo, T. & Kondo, T. Water transport into the Additional information deep mantle and formation of a hydrous transition zone. Phys. Earth Planet. Extended data is available for this paper at https://doi.org/10.1038/s41586- Inter. 143–144, 255–269 (2004). 018-0334-5. 29. Pearson, D. G. et al. Hydrous mantle transition zone indicated by ringwoodite Supplementary information is available for this paper at https://doi.org/ included within diamond. Nature 507, 221–224 (2014). 10.1038/s41586-018-0334-5. 30. Nestola, F. et al. Tetragonal Almandine-Pyrope Phase, TAPP: fnally a name for it, Reprints and permissions information is available at http://www.nature.com/ the new mineral jefbenite. Mineral. Mag. 80, 1219–1232 (2016). reprints. Correspondence and requests for materials should be addressed to E.M.S. Acknowledgements This research was supported by a GIA Liddicoat Publisher’s note: Springer Nature remains neutral with regard to jurisdictional Postdoctoral Research Fellowship to E.M.S. Support to S.B.S. and F.N. was claims in published maps and institutional affiliations.

Methods electron microscopy revealed it to be a Fe–S–C–O multiphase inclusion, with Raman spectroscopy. Raman spectroscopy was performed at the Gemological three main phases: Fe sulphide, Fe carbide, and Fe oxide (Extended Data Fig. 5). Institute of America (GIA), New York, with a Renishaw InVia Raman Microscope, These phases were also analysed using an electron microprobe (Supplementary −1 using a grating with 1,800 lines mm , a 50× lens with a numerical aperture of Tables 1 and 2). The Fe sulphide matches pyrrhotite by composition, Fe1−xS, 0.55, and the 514.5 nm (and 488 nm for additional analyses) output of an argon- with minor variable surface tarnish smeared across the polished surface of the Fe ion laser set to 150 mW output power. Calibration was done using the 520.5 cm−1 sulphide, thought to be an artefact of the polishing process (see mottled texture of Raman peak of a polished silicon wafer. Fe sulphide phase in the oxygen X-ray map; Extended Data Fig. 5). The Fe oxide Energy dispersive X-ray spectroscopy, electron microscopy and cathodolumi- composition matches well with wüstite, Fe1−xO. The Fe carbide phase was verified nescence imaging. Qualitative scanning electron microscopy/energy-dispersive to contain carbon, but quantification was complicated by the additional carbon X-ray spectroscopy (SEM-EDS) examination was carried out at GIA, New York, signal from the carbon coat (and potentially the underlying and neighbouring host using a Zeiss EVO MA10 SEM system with an Oxford X-MaxN 20 EDS detector, diamond material). The carbide phase is tentatively interpreted as an Eckstrom– in variable-pressure mode with chamber pressure of 20 Pa and electron beam Adcock carbide (ideally Fe7C3), although cohenite (Fe3C) is not ruled out without conditions of 10–20 kV and 1 nA. The same instrument was used to capture further analysis. This metallic Fe–S–C–O multiphase inclusion is similar to, but secondary-electron and backscatter images of exposed inclusions, with a chamber still distinct from, the Fe–Ni–C–S inclusions recently reported from CLIPPIR pressure of 20 Pa and electron beam conditions of 15 kV and 1 nA, as well as diamonds (a variety of large, inclusion-poor, often type IIa, sublithospheric dia- cathodoluminescence images (black and white, panchromatic), with a chamber monds)21. In the samples examined, metallic Fe–S–C–O inclusions are a relatively pressure of 20 Pa and electron beam conditions of 15 kV and 0.5 nA. rare occurrence, as opposed to Fe–Ni–C–S inclusions, which are the most common Electron microprobe analysis. Electron microprobe analyses were performed at inclusion type seen in CLIPPIR diamonds. the Geophysical Laboratory of the Carnegie Institution for Science on a JEOL JXA- Sample containing ferropericlase (plus olivine and nyerereite), verified by X-ray 8530F field-emission electron probe. The operating conditions were 15 kV and diffraction. Sample 110208425476 contains multiple dark inclusions. These inclu- 30 nA. Samples were coated with carbon. For the carbide analyses, Fe7C3, NiS and sions are unusual compared to those in the rest of the samples because they are an Al-bearing enstatite glass were used as standards. The Fe7C3 standard was a syn- very irregular in shape, distribution and abundance within the sample. These dark thetic high-pressure iron carbide sample prepared at the University of Michigan31. inclusions are accompanied by smaller (down to micrometre size) inclusions defin- X-ray diffraction. X-ray diffraction analyses were done in the Department of ing curviplanar features interpreted as healed cracks. Where these features intersect Geosciences, University of Padova using a custom instrumentation setup including the polished surfaces, the polish is interrupted slightly, as would be expected for the a high-brilliance MoKα micro-beam source operating at 50 kV and 0.8 mA, a slight crystallographic misalignment of a healed crack. In cathodoluminescence Rigaku Oxford Diffraction SuperNova goniometer and a Dectris Pilatus 200K imaging, these healed cracks appear as bright lines, which are intimately connected detector. The setup allows measurement of inclusions of very limited size, down within the diamond’s strong overall dislocation network pattern (Extended Data to 5–10 µm, that are still in situ within diamonds. Although two samples were Fig. 4). The cathodoluminescence image reveals complex features texturally con- examined (110208245246 and 110208425476), owing to the difficulty in aligning sistent with intense bulk plastic/brittle deformation. the microbeam with individual inclusions, only the latter sample successfully Initial Raman spectroscopy revealed some inclusions to be (or at least contain) produced reflections from inclusion phases. ferropericlase, according to two broad Raman bands at around 650–690 cm−1 and Mass spectroscopy. Carbon isotopic analysis was done at the Department of 190 cm−1, which correspond to ferropericlase in other diamonds in our experience. Terrestrial Magnetism of the Carnegie Institution for Science on a CAMECA 6F Later, polishing down to some of the inclusions revealed examples of ferropericlase, ion microprobe using an extreme energy filtering method. A 10 kV primary Cs which were analysed with an electron microprobe (Supplementary Table 1). beam of 5 nA was focused to ~15 µm and the beam centre was rastered across a Many inclusions, however, did not yield any clear Raman signal. The ferroper- 20 µm × 20 µm area on the sample, held at −4,650 V. Secondary ions of 12C and iclase inclusions (Extended Data Fig. 4) may have been modified during diamond 13C were extracted at −5 kV with an energy window of 100 V. The 12C ions were deformation, especially if large cracks developed that allowed melt or fluid to counted for 1 s and the 13C ions for 5 s, and counting was repeated 200 times. The penetrate and alter or augment the inclusion mineralogy. Examination with X-ray 2σ counting precision for 13C was about 0.5 ‰. The diamond standards N198, diffraction (University of Padova) revealed the presence of not only ferropericlase 4139, 1013, ‘Mao’ and JWH-NS were used to determine the instrumental mass (unit cell edge length, 4.209(8) Å; volume, 74.5(3) Å3; uncertainties are estimated fractionation and drift before and after sample analyses. standard deviations per convention), but also olivine (unit cell dimensions, Prevalence of type IIb diamonds. Type IIb diamonds are recognized as being 4.771(4) Å, 10.266(6) Å, 6.004(3) Å; angles, 90.0°, 90.0°, 90.0°; volume, 294.0(3) Å3), relatively rare, but their rarity is seldom—if ever—defined quantitatively. To deter- and nyerereite, Na2Ca(CO3)2 (observed d-spacings are 3.22, 6.42, 1.79, 1.52 and mine a quantitative figure, a review was conducted for a sample set of 13.8 million potentially others that overlap with diamond reflections). Some additional reflec- natural gem-quality diamonds submitted to GIA. This sample set excludes syn- tions were observed that could not be conclusively assigned: 1.44 (very intense), thetic diamonds, which are often of type IIb and might otherwise skew the results. 2.04 (very intense, not overlapped with diamond), 2.40 (medium intensity), 2.86 Within the sample set of natural gem diamonds, 0.02% were recorded as being (low intensity) and very-low-intensity peaks at 1.84, 1.70 and 1.75. It is possible type IIb. We note that this figure is considered a robust cross-section of natural that the first four reflections (1.44, 2.04, 2.40 and 2.86) correspond to ringwoodite, gem diamonds, but it comes with some uncertainty. A small number of the dia- a polymorph of olivine stable in the lower half of the mantle transition zone. The monds surveyed may be duplicates, submitted for grading multiple times within relatively large measured olivine unit cell certainly suggests Mg-poor compositions the sample period; this may disproportionately affect the count of type IIb samples, (Mg/(Mg+Fe) < 0.80). Accounting for even a modest remnant pressure of 32,33 especially for higher-valued blue gems. Also, the tally of type IIb diamonds may be 0.5–0.6 GPa would imply an olivine composition of Fo73 (Mg/(Mg+Fe) = 0.73), inflated because they are more likely to be submitted to GIA for grading because which is far from typical olivine-inclusion compositions, but overlaps with the 02. 4 of their colour, and thus their higher gem value. Conversely, a small proportion composition reported for a ringwoodite inclusion with Mg(/Mg + Fe)0=.7502. 1 of type IIb diamonds may not have been officially recorded as type IIb in the (ref. 29). It is therefore considered possible that this sample may contain both database used, and by default would be counted as non-type IIb. Lastly, and most olivine from inverted ringwoodite, as well as preserved ringwoodite. Again, it importantly, the above value is expressly for gem diamonds, and does not account should be noted that the healed cracks raise the possibility that some inclusion for industrial goods. Given the colour, morphology and clarity characteristics of material could have been introduced post-growth, meaning that ferropericlase known type IIb diamonds, it is considered unlikely that many industrial-quality may not necessarily have been in equilibrium with other phases, nyerereite and natural diamonds are of type IIb. Because about 70%–80% of mined diamonds are olivine. For this reason, and given that ferropericlase was observed in other of industrial quality, the proportion of all diamonds (gem plus industrial) that are samples, only ferropericlase is discussed and considered as a primary mineral in of type IIb may be as low as 0.004%. the main text. Samples. A list of all samples examined and the inclusion phases observed within Raman- and X-ray-diffraction-based inclusion barometry. Within diamonds, them is given in Extended Data Table 1. We note that a diamond may contain the pressure inside an inclusion is often elevated. Although the extreme confining multiple inclusions of the same mineralogy and that the list does not necessarily pressure of the deep mantle is removed as the diamond is carried to the surface, reflect the number of inclusions found in each sample. However, the abundance the inclusion remains confined within the diamond host and may still have some of a certain kind of inclusion within individual diamonds compares very well with remnant pressure. The expected remnant pressure is primarily dependent on the frequency of those inclusions across the whole suite. More important is the the inclusion mineral species and depth (pressure and temperature conditions) inclusion assemblage and the potential host rock paragenesis that it portrays. A of entrapment. In reality, most inclusions have at least some reduction in their visual overview of the sample suite is given in Extended Data Fig. 1. remnant pressure due to diamond deformation (brittle/plastic) relieving some Metallic Fe–S–C–O multiphase inclusion. Sample 110208245246 contains an built-up pressure. Determining the remnant pressure in an inclusion can give a opaque inclusion (inclusion B) whose mineralogy was unclear from the initial physical indication of the minimum depth of entrapment, constraining the depth Raman analysis. Subsequent polishing to expose this inclusion to examine it by of diamond growth20,22,23,32,34–40.

Raman spectroscopy is based on the energy associated with inducing molec- For all three inclusions, CaSiO3 walstromite, coesite and ferropericlase, the ular vibrations, which in some instances can be sensitive to stresses acting upon estimated entrapment conditions are considered as minimum pressures. Much of the material. This means that the Raman spectrum of some materials exhibits the inclusion pressure has been relieved by brittle and plastic deformation of the measurable, predictable changes under pressure. The Raman peaks of minerals, surrounding diamond. Taken together, these results require that the diamonds for example, can be slightly shifted from their expected values when the minerals originated from below the lithosphere. are under pressure. Data availability. All relevant data are presented in Extended Data Figs. 1–7, In the suite of type IIb diamonds, two inclusions were found to preserve extreme Extended Data Table 1 and Supplementary Information Tables 1–3. Original remnant pressure. The first inclusion is a CaSiO3 walstromite inclusion, which spectral data and electron microprobe data are available from the corresponding although it is interpreted to be an inversion product from Ca-Pv, still preserves author. remnant pressure that can help retrace the final stages of exhumation from the mantle. The inclusion, in sample 110203744064, has its 977.3 cm−1 peak (uncon- −1 31. Liu, J., Li, J. & Ikuta, D. Elastic softening in Fe7C3 with implications for Earth’s fined position) shifted upwards to 1,000.4 ± 0.5 cm (Extended Data Fig. 6), deep carbon reservoirs. J. Geophys. Res. Solid Earth 121, 1514–1524 (2016). which corresponds to a remnant pressure of 4.4 ± 0.1 GPa, using the mineral- 32. Howell, D., Wood, I. G., Nestola, F., Nimis, P. & Nasdala, L. Inclusions under and peak-specific experimental calibration factor22 of 5.16 ± 0.9 cm−1 GPa−1. remnant pressure in diamond: a multi-technique approach. Eur. J. Mineral. 24, Assuming any reasonable mantle temperature within 1,200–2,000 K, this rem- 563–573 (2012). nant pressure would have required a minimum inclusion entrapment pressure of 33. Nestola, F. et al. First crystal-structure determination of olivine in diamond: 22 composition and implications for provenance in the Earth’s mantle. Earth ~9 GPa (approximately 240–280 km deep) . Planet. Sci. Lett. 305, 249–255 (2011). Similarly, in sample 890000180198, the main Raman peak of coesite is shifted 34. Mazzucchelli, M. L. et al. Elastic geothermobarometry: corrections for the from its unconfined value of 520.6 cm−1 to 537.9 ± 0.5 cm−1 (Extended Data geometry of the host-inclusion system. Geology 46, 231–234 (2018). Fig. 6). Using the pressure calibration factor41 of 2.9 cm−1 GPa−1 gives a nominal 35. Angel, R. J., Alvaro, M., Miletich, R. & Nestola, F. A simple and generalised P–T–V remnant pressure of 6.0 ± 0.4 GPa, but this is not considered a reliable pressure EoS for continuous phase transitions, implemented in EosFit and applied to quartz. Contrib. Mineral. Petrol. 172, 29 (2017). determination. Other peaks in the spectrum do not exhibit the same magnitude 36. Gonzalez-Platas, J., Alvaro, M., Nestola, F. & Angel, R. EosFit7-GUI: a new of pressure-induced shift, which suggests anisotropic stresses. Caution is needed graphical user interface for equation of state calculations, analyses and in the interpretation of remnant pressure in coesite owing to the complex effects teaching. J. Appl. Cryst. 49, 1377–1382 (2016). of anisotropy, anomalies in its elastic behaviour, plus the lack of reliable high- 37. Angel, R., Alvaro, M., Nestola, F. & Mazzucchelli, M. Diamond thermoelastic pressure and low-to-high-temperature experiments for coesite. A non-cubic mineral properties and implications for determining the pressure of formation of diamond-inclusion systems. Russ. Geol. Geophys. 56, 211–220 (2015). trapped in a cubic (almost elastically isotropic) host such as diamond will be 38. Angel, R., Nimis, P., Mazzucchelli, M., Alvaro, M. & Nestola, F. How large are subject to anisotropic strains during exhumation from the mantle and will develop departures from lithostatic pressure? Constraints from host–inclusion elasticity. deviatoric stresses. Currently there is no reliable method among existing analytical J. Metamorph. Geol. 33, 801–813 (2015). techniques and hydrostatic calibrations to determine strains on minerals subject 39. Angel, R. J., Alvaro, M. & Gonzalez-Platas, J. EosFit7c and a Fortran module to deviatoric stresses and interpret them in terms of pressure. X-ray diffraction (library) for equation of state calculations. Z. Kristallogr. Cryst. Mater. 229, could be a useful method to supplement Raman spectroscopy for this purpose, but 405–419 (2014). 40. Angel, R. J., Mazzucchelli, M. L., Alvaro, M., Nimis, P. & Nestola, F. Geobarometry unfortunately it cannot be performed on small inclusions to the required level of from host-inclusion systems: the role of elastic relaxation. Am. Mineral. 99, precision and accuracy. These factors limit the interpretation of the Raman results 2146–2149 (2014). from this inclusion. A minimum entrapment pressure is therefore not derived 41. Hemley, R. J. in High-Pressure Research in Mineral Physics Vol. 39 (eds here. Nevertheless, the remnant pressure suggested by the main peak of coesite far Manghnani, M. H. & Syono, Y.) 347–359 (American Geophysical Union, exceeds those of other coesite inclusions found in lithospheric diamonds, which Washington DC, 1987). 42–44 42. Sobolev, N. V. et al. Fossilized high pressure from the Earth’s deep interior: the reach upwards of 3.6–4.3 GPa . coesite-in-diamond barometer. Proc. Natl Acad. Sci. USA 97, 11875–11879 In addition to Raman spectroscopy, X-ray diffraction revealed high remnant (2000). pressure inside ferropericlase inclusions in one of the studied type IIb diamonds, 43. Smith, E. M., Kopylova, M. G., Frezzotti, M. L. & Afanasiev, V. P. Fluid inclusions in sample 110208425476. X-ray diffraction permitted the calculation of the unit cell Ebelyakh diamonds: evidence of CO2 liberation in eclogite and the efect of H2O (74.5(3) Å3), which, when combined with the composition obtained later by an on diamond habit. Lithos 216-217, 106–117 (2015). + = 44. Smith, E. M., Vendrell, C. & Johnson, P. Coesite inclusions with flaments in electron microprobe (Mg/(Mg Fe) 0.91), gives a remnant pressure of 1.8 GPa. diamond. Gems Gemol. 52, 410–412 (2016). At mantle temperatures of 1,200–2,000 K, this is estimated to require entrapment 45. D’Ippolito, V., Andreozzi, G. B., Bersani, D. & Lottici, P. P. Raman fngerprint of at pressures of at least 10.3–14.1 GPa23, not accounting for the stress relieved by chromate, aluminate and ferrite spinels. J. Raman Spectrosc. 46, 1255–1264 brittle and plastic strains. (2015).

110208245246 880000597423 101024478345 890000180201 110206933429 880001287015 890000508981 2.76 mm long 3.07 mm long 7.38 mm long 2.86 mm max diam. 7.29 mm long 6.64 mm long 5.08 mm long

110208425476 100521978876 890000180198 110208086814 110205945970 880002502964 880000037816 1.95 mm long 4.77 mm long 2.42 mm max diam. 3.22 mm max diam. 4.23 mm max diam. 3.97 mm max diam. 4.76 mm long

110208122379 110108200990 110207972434 890000188195 110203744064 890000562519 110208093607 12.92 mm wide 12.58mm long 14.88 mm long 9.86 mm long 11.15 mm long 18.46 mm long 19.99 mm long

110206861587 100717827346 110208258915 890000700302 110207965506 110207872703 110208241567 5.54 mm wide 5.46 mm max diam. 7.66 mm long 7.73 mm long 4.28 mm long 16.69 mm long 9.03 mm long

110208768050 110208948788 110208247369 110208579703 110208423120 DVBT 5.24 mm long 6.08 mm long 11.03 mm long 10.76 mm long 13.83 mm long 2.78 mm wide

110208084323 110208780369 110208773706 110208135763 890000076656 110207974941 9.82 mm long 14.50 mm long 4.21 mm max diam. 9.85 mm long 5.00 mm max diam. 11.67 mm field of view

110207974942 110207974945 110207974947 110207974948 110207974949 110207974950 11.67 mm field of view 11.67 mm field of view 11.67 mm field of view 11.67 mm field of view 11.67 mm field of view 11.67 mm field of view Extended Data Fig. 1 | Suite of 46 type IIb diamonds studied. Images are not to scale. Refer to noted dimensions (max diam., maximum diameter).

Peridotitic bulk composition Basaltic bulk composition mineral proportions (wt%) mineral proportions (wt%) 100 100 20 40 60 80 20 40 60 80

Opx Cpx Upper mantle 200

hpcpx olivine garnet garnet clinopyroxene 10 400 Tr Pressure (GPa ) ansision zone Depth (km) wadsleyite majoritic garnet 20 600 ringwoodite majoritic garnet

1 Lower mantle

7 4 NA L 800 ferropericlas 6 7 25 12 1 stishovit e

bridgmanite 30 Ca-Pv

e Ca-Pv bridgmanite Extended Data Fig. 2 | Mineralogy of mantle rocks with peridotitic and between the left and right panel is for illustrative purposes, and in reality basaltic bulk composition as a function of depth. Numbers in boxes some samples (for example, with Ca-Pv alone) are not firmly categorized. denote the number of diamonds observed with inclusions of the given See Extended Data Table 1 for a breakdown of inclusions by sample. phase, and blue shading in the boxes indicates that a thin fluid CH4 ± H2 Adapted from ref. 17. jacket was found with the phase. We note that the division of samples

NaAl-pyroxene

Jeffbenite

100 µm 50 µm

Jeffbenite NaAl-pyroxene

keV keV Extended Data Fig. 3 | Multiphase inclusion interpreted as former with their Raman identification as jeffbenite and NaAl-pyroxene low-Ca, high-Na majoritic garnet. a, Optical microscope image of (monoclinic, with composition between enstatite and jadeite). High Na the inclusion, exposed on a polished facet of sample 880000037816. content suggests a metabasaltic paragenesis, while low Ca content may b, Secondary-electron image of the same inclusion, grooved with nearly reflect Ca partitioning into coexisting Ca-Pv at the base of the mantle horizontal polishing lines. c, d, EDS spectra of the two phases, consistent transition zone or the uppermost lower mantle.

a b

500 µm 100 µm

200 µm

Extended Data Fig. 4 | Diamond sample 110208425476. a, Optical is a small amount of iron inadvertently deposited on the surface during microscope image of the whole diamond, showing multiple dark polishing on a conventional cast iron scaife. c, Cathodoluminescence inclusions of ferropericlase. b, Polishing down the table facet slightly image of the whole diamond, revealing a complex dislocation network exposed this group of four ferropericlase inclusions, shown here in an pattern, with interspersed healed fractures, that records a combination of electron backscatter image. The smeared texture on the largest inclusion both plastic and brittle deformation.

a b

Fe-oxide

Fe-carbide

10 µm 10 µm Fe-sulfide 10 µm

Fe-carbide Fe-sulfide Fe-oxide

d Fe Kα1S Kα1O Kα1

10 µm 10 µm 10 µm

Extended Data Fig. 5 | Multiphase Fe–S–C–O metallic inclusion in all spectra owing to the diamond host, diamond particles embedded in sample 110208245246 (inclusion B). a, Optical microscope image of the polished surface (black specks, especially in the sulphide phase), as inclusion B, while still contained within the diamond host. b, Electron well as the carbon inherent to the Fe-carbide phase. d, X-ray elemental backscatter image of the inclusion, after polishing to expose a cross- maps obtained with EDS, showing the spatial variation in signal in the section through it. The three main phases are colour-coded in the right region of Fe, S and O peaks (Kα1). Sulphur delineates the Fe-sulphide panel. c, X-ray spectra obtained with EDS, showing the qualitative phase. Oxygen marks the Fe-oxide phase, while also showing the variable elemental composition of each of the three phases. Carbon is present in oxidation/tarnish layer on the sulphide portion of the inclusion.

668.8

a CaSiO3 walstromite

1,000.4 1,062.1

100 µm 656.3 977.3 reference 1,037.6 y

537.9 Intensit b coesite

520.6

reference 10 µm

200400 600800 1,000 1,200 Raman shift (cm-1) Extended Data Fig. 6 | Two inclusions exhibiting a large pressure- crack, along with other related ‘satellite’ inclusions that presumably induced shift in Raman features. a, CaSiO3 walstromite (thought to surrounded a nucleus coesite inclusion that was polished away when this be former Ca-Pv) in sample 110203744064, inclusion A, with the three diamond was facetted. Neighbouring coesite inclusions in b also have main peaks shifted to higher wavenumbers compared to a zero-pressure high, but variable, remnant pressures, as reflected by the Raman spectra. reference spectrum. This inclusion also contains CH4. b, Coesite (SiO2, Reference spectra are from ref. 19 and RRUFF-X050094, and zero-pressure thought to be former stishovite) in sample 890000180198. The inclusion reference peak positions are taken from refs22,41. analysed (circled) is about 2 µm wide and lies in a planar lobate healed

200 µm

Extended Data Fig. 7 | Dislocation network pattern in sample cathodoluminescent boundaries surround darker, low-strain domains. The 110208245246, as seen in panchromatic cathodoluminescence. Each dark curved feature on the right of the centre is a crack (not healed). of the bright web-like lines are made up of many dislocations, and these

Extended Data Table 1 | Suite of 46 type IIb diamond samples and inclusion summary

Sample Mass Observed single-ormulti-phase inclusions,interpreted as: [Ca-Pv, Ca-silicate perovskite] (carats) *[Bridgmanite, Mg-silicate perovskite] [Stishovite] [CF, calcium ferrite-type phase] [Ferropericlase] [sodic majoritic garnet] †[Bridgmanite, aluminum bearing] [metallic alloy/mixture] (Mineral abbrev. in footnote)

Basaltic compositional association, containing CF, Stishovite, or sodic majoritic garnet 110208245246 0.08 [Wal], [Opx + Jeffbenite + Spinel + Ilmenite], [Nepheline + Olivine+ Spinel + CH4], [FeS + Fe-carbide + FeO] 880000597423 0.13 [Pyroxene+ Spinel + Corundum + Olivine], [Coesite + trace Px + CH4] 101024478345 0.57 [Cpx + Opx + Jeffbenite], [Coesite], ‡[possible sulfide],§[small unidentified opaque], §[Calcite, in healed cracks] 890000180201 0.08 [Pyroxene+ Jeffbenite], [Coesite + Kyanite] 110206933429 1.01 [Wal + Larnite], [Cpx + Jeffbenite + Spinel], [Coesite], ‡[possible sulfide], ‡[Perovskite, may be CaSiO3 or CaTiO3] 100521978876 0.13 [Wal + Larnite + Wollastonite], [Coesite] 890000180198 0.05 [Wal + Larnite], [Coesite] 880002502964 0.24 [Cpx + Jeffbenite], [Coesite], [small unidentified opaque] 880000037816 0.21 [Wal], [NaAl-pyroxene + Jeffbenite; examined with SEM-EDS], [Opx], ‡[possible sulfide] May associate with basaltic or peridotiticbulk compositions, not firmly categorized 110205945970 0.27 [Wal + Larnite], [Opx + Jeffbenite ], ‡[possible sulfide] 110208086814 0.13 [Wal + Wollastonite], [Pyroxene+ Jeffbenite + Spinel + Ilmenite + Olivine], [Opx + Spinel], ‡[possible sulfide] 100717827346 0.61 [Cpx + Opx + Spinel + Olivine] Raman clearly shows two spectrally/spatially distinct pyroxenes 110208579703 2.01 [Wal + Larnite + CH4], [Opx], [Unidentified opaque, magnetic + CH4], §[lobe of small unidentified inclusions + CH4, magnetic, suspected metallic alloy] 890000188195 3.46 [Wal + Larnite + CH4] 890000700302 0.61 [Wal + Larnite + CH4 + H2], §[lobe of small unidentified inclusions + CH4, magnetic, suspected metallic alloy] 110203744064 2.70 [Wal + CH4] 110208093607 24.18 [Wal + Larnite] Cullinan Dreamdiamond, 122.52 carat rough mass 890000562519 17.09 [Wal + Larnite] 110108200990 3.81 [Wal + Larnite] 110208258915 2.26 [Wal + Larnite] 110207965506 0.46 [Wal + Larnite] 110207872703 2.46 [Wal + Larnite] 110208241567 2.15 [Wal + Larnite] 110206861587 0.73 [Wal + Larnite] 110208948788 0.32 [Wal + Larnite] 110208247369 2.08 [Wal] 110207972434 10.67 [Wal] DVBT 0.03 [Wal], §[small Fe-Ni-S inclusions, examined with SEM-EDS, in healed lobate crack] 110208423120 4.06 [Wal + Pseudowollastonite] 880001287015 0.35 [Wal + Larnite + Pseudowollastonite], [Unidentified opaque, magnetic + CH4 + H2] 890000508981 0.30 [Opx + Jeffbenite] 110208122379 6.08 [Unidentified opaque, 655 band like wüstite] 110208768050 0.42 ‡[possible sulfide] 110208084323 1.02 ‡[possible sulfide] 110208135763 1.71 [graphitic fracture rosette hides inclusion; suspected Wal or sulfidebased on appearance] 110207974950 2.97 [graphitic fracture rosette hides inclusion; suspected Wal or sulfidebased on appearance] Peridotitic compositional association, with fPer and/or Opx having distinctly sharp enstatite Raman spectrum and no Al-phases 110208773706 0.32 [Wal + Larnite], [Opx + CH4 + H2] 110207974945 0.78 [Wal], [Opx + Olivine+ CH4 + H2, plus weak 253+376 may be lepidocrocite γ-FeOOH], [fPer], [fPer + unidentified opaque + CH4], §[lobe of small unidentified inclusions + CH4] 110207974941 0.92 [Opx + CH4], [Unidentified opaque, magnetic + CH4 + H2, suspected metallic alloy] 890000076656 0.48 [Wal + Larnite], [Opx + Olivine+ CH4] 110207974949 2.80 [Wal + Larnite], [fPer] 110207974948 2.17 [Wal + Larnite], [fPer], §[lobe of small unidentified inclusions + CH4, plus weak 253+376 may be lepidocrocite γ-FeOOH] 110208780369 5.02 [Wal + Pseudowollastonite + CH4], [fPer], [fPer+ minor sulfate or phosphate, and sulfide phases] 110207974947 1.45 [fPer+ Unidentified opaque], [graphitic fracture, inclusion nucleus not visible] 110207974942 2.31 [fPer+ Unidentified opaque], §[Dolomite, in healed cracks] 110208425476 0.03 §[fPer, irregularly shaped inclusions,pervaded by healed fractures; XRD shows fPer + Olivine+ nyerereite]

−1 Wal, CaSiO3 walstromite; Cpx, clinopyroxene; Opx, orthopyroxene (resolvable doublet in the 660–680 cm region); fPer, ferropericlase. Pyroxene not specifed as clino- or ortho-pyroxene was not −1 45 −1 confdently subcategorized from its asymmetric main peak at 660–690 cm . The identifed spinel has Raman spectral features like those of MgAl2O4 and FeAl2O4 , a prominent peak at ~750 cm , and other peaks matching spinel reported in other retrogressed CF-phase and Al-bridgmanite inclusions14. The possible sulphide phase has a Raman spectrum resembling arsenopyrite and other −1 As or Sb sulphides, with a Raman spectrum having weak, sharp, variable peaks in the region 100–350 cm . Pseudowollastonite is a high-temperature CaSiO3 polymorph, found here as an accessory within three calcium silicate inclusions (main peaks: 985, 580, 374 and 142 cm−1). −1 *Bridgmanite identifcation based on the presence of orthopyroxene, with sharp Raman peaks matching enstatite (Mg2Si2O6) with a clearly resolved doublet in the 660–680 cm region. The absence of jefbenite/spinel suggests low-Al bridgmanite. †Aluminous bridgmanite is the likely precursor for these multi-phase assemblages. However, some of these inclusions may actually represent original majoritic garnet if the pyroxene phase were to be identifed as NaAl-pyroxene, which can be difcult to resolve in some cases. It should be stressed that even with chemical analysis, interpretation of these multi-phase retrogressed inclusions is not straightforward, and the signifcance of jefbenite in retrogressed assemblages remains a matter of debate17. ‡Small (<10 µm, often <5 µm) and rare inclusion, regarded as a minor accessory inclusion. §Inclusions in healed cracks, of mantle origin but trapped or modifed post-growth.