Laboratory experiments and seismology data have created a clear theoretical picture of the most abundant minerals that comprise the deeper parts of the Earth’s mantle. Discoveries of some of these minerals in ‘super-deep’ diamonds—formed between two hundred and about one thousand kilometres into the lower mantle—have confirmed part of this picture^1,2,3,4,5. A notable exception is the high-pressure perovskite-structured polymorph of calcium silicate (CaSiO₃). This mineral—expected to be the fourth most abundant in the Earth—has not previously been found in nature. Being the dominant host for calcium and, owing to its accommodating crystal structure, the major sink for heat-producing elements (potassium, uranium and thorium) in the transition zone and lower mantle, it is critical to establish its presence. Here we report the discovery of the perovskite-structured polymorph of CaSiO₃ in a diamond from South African Cullinan kimberlite. The mineral is intergrown with about six per cent calcium titanate (CaTiO₃). The titanium-rich composition of this inclusion indicates a bulk composition consistent with derivation from basaltic oceanic crust subducted to pressures equivalent to those present at the depths of the uppermost lower mantle. The relatively ‘heavy’ carbon isotopic composition of the surrounding diamond, together with the pristine high-pressure CaSiO₃ structure, provides evidence for the recycling of oceanic crust and surficial carbon to lower-mantle depths.

A key goal of solid-Earth geosciences is to establish the mineralogy of the Earth’s mantle throughout its depth, which acts as a primary control on mantle dynamics and chemistry. Diamonds are unique in this regard because they provide access to the deepest intact material from the Earth’s interior through the minerals contained within their volumes. Over the past three decades, a growing number of studies have used a class of diamonds known as super-deep diamonds to study mantle processes in the deep sublithospheric mantle, the transition zone and the lower mantle^1,2,3,4,5,6. Early studies^1,2,3 suggest that some of the assemblages included within super-deep diamonds represent samples of the lower mantle and the transition zone that variably retrogressed to lower pressures. Later studies indicate that some of these assemblages and minerals might originate from shallower depths^7,8, although still beneath the lithosphere.

The most common minerals found within super-deep diamonds are ferropericlase [(Mg,Fe)O] and CaSiO₃ (refs 1, 2, 3, 9). Ferropericlase is stable at most pressure and temperature conditions in the mantle; therefore, when found as a single inclusion within diamond, this mineral cannot be considered an unambiguous indicator of a super-deep origin⁷.

The CaSiO₃ phase found within super-deep diamonds typically has the crystal structure of walstromite (BaCa₂Si₃O₉)^1,6,8,9. Perovskite-structured CaSiO₃ (Ca-Pv) is considered one of the most important components in the Earth’s lower mantle, comprising approximately 7% of the peridotitic mantle and about 23% of the volume of a subducted mid-ocean ridge basalt slab^9,10,11. As such, it is likely to be the fourth most abundant terrestrial mineral. Within the peridotitic lower mantle, Ca-Pv is the dominant sink for Ca and for incompatible elements, including the key heat-producing elements K, U and Th (ref. 12). However, Ca-Pv has so far never been found in nature and even high-pressure laboratory experiments have failed to quench it to a metastable phase at the conditions of the Earth’s surface. Although early studies^1,2,3,4 of super-deep diamonds make a clear case for the presence of Ca-Pv in the transition zone and lower mantle, the structure of this phase was either undetermined or documented to be the lower-pressure polymorph—CaSiO₃ walstromite—and interpreted as a back-transformation of perovskite-structured CaSiO₃. The phase transformation from Ca-Pv to CaSiO₃ walstromite would require a volume change¹³ of about 28%, which is impossible for diamond to accommodate owing to its extremely high bulk modulus¹⁴. The absence of healed fractures in the diamond host reported in ref. 13 implies that CaSiO₃ walstromite is unlikely to represent inverted Ca-Pv. Plastic deformation of the diamond lattice could accommodate some of the volume change necessary for the phase transformation of the inclusion. However, although plastic deformation in super-deep diamonds has been well documented¹⁵ and is expected to be substantial, it has never been quantified. Some super-deep diamonds with documented phase assemblages that include ferropericlase, enstatite (inverted bridgmanite) or CaSiO₃ walstromite probably originate from lower-mantle depths^1,2,3,4,9, but ambiguity remains. Therefore, finding an un-retrogressed Ca-Pv would provide confirmation of lower-mantle sampling by super-deep diamonds.

In this study we investigated an inclusion within a diamond from the Cullinan mine in the Gauteng province of South Africa. The Cullinan kimberlite is a group I kimberlite; that is, its chemistry and Sr, Nd and Hf isotope signatures are thought to reflect a melt source beneath the lithospheric mantle, within the Earth’s convecting mantle¹⁶. The Cullinan mine is renowned for producing exceptionally large diamonds (such as the 3,107-carat Cullinan diamond^6,17), most of which have been suggested to be super-deep diamonds⁶.

The diamond examined here has a 31 μm × 26 μm × 10 μm CaSiO₃ inclusion, which was exposed by polishing. X-ray diffraction, Raman spectroscopy and electron backscatter diffraction (EBSD) reveal the CaSiO₃ in this inclusion to have a perovskite structure. To our knowledge, this represents the only finding of non-reverted Ca-Pv in nature and the first Ca-Pv, including those synthesized in the laboratory, to preserve its high-pressure structure at the surface of the Earth.

Cathodoluminescence imaging of the host diamond surrounding the Ca-Pv inclusion (Fig. 1) reveals multiple growth zones and a complex internal structure, typical of super-deep diamonds^4,18. Fourier transform infrared (FTIR) spectroscopy (Extended Data Fig. 1) of the diamond host indicates a nitrogen content of 34 p.p.m., with 97% in the B-aggregated form; that is, the diamond host is type IaB. The low nitrogen content and very high level of B aggregation are typical characteristics of super-deep diamonds^4,19, indicating prolonged exposure to the high temperatures that are prevalent at transition-zone and lower-mantle depths.

The Ca-Pv inclusion is shown in yellow. Carbon isotopic compositions measured at five locations (red circles) are expressed as δ¹³C values.

Full size image

• Download PowerPoint slide

The chemical composition of the Ca-Pv inclusion, determined by electron microprobe analysis, is almost pure CaSiO₃ (Ca_0.98Si_0.98O₃), with minor impurities of Ti, Al, Fe and Mg totalling 0.04 atoms per formula unit (Extended Data Table 1).

Backscattered-electron imaging and energy-dispersive X-ray spectroscopy (EDS) element maps (Fig. 2) show that the Ca-Pv crystal includes 14 irregular areas of CaTiO₃ perovskite with sizes between 1 μm and 7–8 μm and an approximate stoichiometry of Ca(Ti_0.92Si_0.07Al_0.02)O₃. The texture, size and abundance of these CaTiO₃ intergrowths are very similar to those of inclusions reported in CaSiO₃ walstromite phases in super-deep diamonds²⁰ from Juina, Brazil. The exposed surface of our Ca-Pv inclusion makes accurate estimation of its bulk composition difficult, but image analysis indicates that the host crystal in bulk may contain up to 6% by volume CaTiO₃. CaTiO₃ perovskite is a common mineral in nature and remains stable well into the lower mantle²¹. By contrast, a Ca-Pv sample that retains its perovskite structure at room temperature and pressure has no experimentally synthesized analogues, unless a considerable amount of CaTiO₃ (about 34 mol%; ref. 21) is dissolved within its structure, far more than the CaTiO₃ component observed here. However, our discovery of natural Ca-Pv with less than 2 mol% CaTiO₃ in the CaSiO₃-rich portion of the inclusion indicates that, unlike experiments, nature must provide pressure–temperature–time pathways that are capable of preserving this metastable phase.

a, Backscattered-electron image of the Ca-Pv inclusion (dark grey) surrounded by the diamond host (black), showing smaller inclusions of CaTiO₃ perovskite (light grey). b–d, Energy-dispersive X-ray spectroscopy elemental maps of Ca (b), Ti (c) and Si (d). The colour intensity (black within the grain outline through to saturation in a specific colour) is proportional to the element concentration.

Full size image

• Download PowerPoint slide

As stated above, X-ray diffraction data show that the CaSiO₃ inclusion has a perovskite structure. The small size of the inclusion (thickness ≤ 10 μm, as estimated by confocal Raman spectroscopy) and its entrapment within the diamond host resulted in only a limited number of measured diffraction reflections (n = 91), of which only nine were unique (Extended Data Table 2). All of the 91 reflections were used to refine the Ca-Pv unit-cell parameters: a = 5.397 ± 0.004 Å, b = 5.404 ± 0.004 Å, c = 7.646 ± 0.004 Å, volume 223.0 ± 0.03 ų

However, alternative unit-cell refinements using other numerical approaches could provide considerably different (by more than 1%) unit-cell parameters owing to the relatively poor accuracy and precision with which the spacings between crystal planes (d spacings) were measured in this study. These relatively large uncertainties are typical when studying minerals of this size and arise not only from the limited number of reflections, but also because the measurements were performed using an area detector, which provides lower precision in d-spacing determination than a point detector. Such relatively large uncertainty on the cell parameters makes any comparison with the unit-cell volume of CaTiO₃ perovskite unreliable, although we can establish that the structures of CaTiO₃ perovskite and Ca-Pv are very similar. Ewald projections along the three crystallographic axes (Fig. 3a) indicate an orthorhombic unit cell. The unit cell and the chemical composition confirm that the mineral is Ca-Pv. Recent numerical simulations on ‘host–inclusion’ systems²² indicate that an inclusion partly exposed to atmospheric pressure loses only a portion of its residual pressure, depending on the elastic properties of both the host mineral and the inclusion. The Ca-Pv inclusion studied here is partly exposed at the diamond surface, but with two-thirds of its volume still buried in the diamond host. Thus, any measurements on this grain would be affected by some residual pressure still acting on the inclusion, which in turn affects the X-ray diffraction data and Raman spectra.

a, Ewald projections along three different orientations for the Ca-Pv. b, Baseline-corrected Raman spectrum of the Ca-Pv inclusion compared with that of the CaTiO₃-perovskite intergrowth (Fig. 2a).

Full size image

• Download PowerPoint slide

Raman spectra (Fig. 3b) of the inclusion show that the spectrum of the CaTiO₃ perovskite is in excellent agreement with Raman data for CaTiO₃ perovskite from the RRUFF database²³ (Extended Data Fig. 2). The CaSiO₃ and CaTiO₃ spectra are similar. Small differences are evident because of the presence of two Raman peaks for the CaSiO₃ spectrum, which could belong to the lower-pressure CaSiO₃ polymorph²³ wollastonite-2M. This wollastonite polymorph is not stable at pressures higher than 3 GPa along a mantle geotherm²⁴, well below the diamond stability field. Therefore, its presence is probably due to minor partial inversion of the Ca-Pv phase caused by the polishing of the sample to expose the inclusion, as reported previously²⁵.

EBSD measurements conducted on several areas of the grain provide no evidence of amorphous portions (Fig. 4). The EBSD pattern of the CaSiO₃ area (red circle), shown as the non-indexed EBSD pattern in Fig. 4b, is complex and could not be indexed by a single phase. The observed pattern can be indexed by using a combination of reference EBSD patterns for CaTiO₃ perovskite (Fig. 4c) and CaSiO₃ wollastonite-2M (Fig. 4d), again confirming that CaSiO₃ is present in this diamond with a perovskite-type structure. Because EBSD measures surface responses (within tens of nanometres from the surface), this signal can only come from the CaSiO₃ phase.

a, Location (indicated by the red filled circle) from which the EBSD images were obtained. b, Non-indexed EBSD pattern relative to the CaSiO₃ inclusion. c, d, EBSD patterns indexed with the reference patterns of CaTiO₃ (c) and CaSiO₃ wollastonite-2M (d). The coloured lines in c represent the EBSD indexed pattern of CaTiO3, whereas the coloured lines in d represent the EBSD indexed pattern of wollastonite-2M. The numbers reported in both the figures at the intersections between the coloured lines represent the zone axes.

Full size image

• Download PowerPoint slide

We suggest that the natural Ca-Pv found trapped within the super-deep diamond was a result of the unmixing of the high-pressure solid solution Ca(Ti,Si)O₃. If the two phases exsolved from a homogeneous bulk composition, this phase would contain about 3.9% TiO₂.

We estimate that the stoichiometry of the original phase composition was (Ca_0.98Mg_0.01Fe_0.01)(Si_0.93Ti_0.06Al_0.01)O₃. This composition is consistent with that of CaSiO₃ samples crystallized in experiments²⁶ from a mid-ocean ridge basalt (MORB)-like bulk composition at about 26 GPa and is similar to that of CaSiO₃ walstromite and CaTiO₃ intergrowths found within Juina super-deep diamonds; these intergrowths have been suggested to originate from basalt-like compositions subducted to lower-mantle depths that later retrogressed during their ascent to the Earth’s surface²⁰. The preservation of the high-pressure perovskite structure in the case of the Cullinan inclusion supports the derivation of such compositions from lower-mantle depths.

The possible subducted basaltic protolith origin of the Cullinan Ca-Pv inclusion suggests that we might expect to observe some evidence of a crustal parentage in the carbon isotopic composition of the host diamond (Fig. 1; Extended Data Table 3), which has δ¹³C values ranging from −2.3‰ to −4.6‰, where δ¹³C = (¹³C/¹²C)_sample/(¹³C/¹²C)_PDB − 1 (PDB, Pee Dee Belemnite reference material). The core region of the diamond, defined by cathodoluminescence imaging (Fig. 1), contains the Ca-Pv inclusion and has an average δ¹³C value of −2.3‰ ± 0.5‰, considerably lower than the typical upper-mantle value²⁷ of −5.5‰. By contrast, the outer-rim region of the diamond has a composition (mean δ¹³C of −4.1‰ ± 0.5‰) that is closer to −5.5‰. Crustal carbon reservoirs have carbon isotopic compositions that are both ‘heavier’ and ‘lighter’ than the typical upper-mantle value. While ‘isotopically light’ carbon compositions (δ¹³C < −25‰) have been found in super-deep diamonds from Juina, which are thought to be derived from subducted basalt protoliths^20,28, ‘isotopically heavy’ (−3‰ to −0.5‰) carbon compositions, such as those measured in the core of the studied diamond, have also been reported in super-deep diamonds from Brazil (Sao Luis and Juina) and Guinea (Kankan)^18,19,20. If the δ¹³C value of −2.3‰ is compared with the median value (−4.91‰) of 1,473 published analyses of lithospheric diamonds containing peridotitic inclusions—a group of diamonds usually accepted to be minimally affected by subduction²⁷—it is found to be an outlier, beyond three times the median absolute deviation. Such anomalously high carbon isotopic compositions are thought to reflect a greater influence of subducted carbonate in the fluid that formed these super-deep diamonds^18,19. The carbon isotope compositions of the rim of the Cullinan diamond (Fig. 1) may represent an overgrowth that developed under upper-mantle conditions or from a distinct source of carbon in the lower mantle. Regardless, the high δ¹³C values of the portion of the diamond that contains the Ca-Pv inclusion supports the premise that it originates from a subducted basaltic protolith.

Our discovery of Ca-Pv in a super-deep diamond firmly establishes this phase as a component of the Earth’s deep mantle, confirming previous suggestions^1,2,3,4,9 that lower-pressure CaSiO₃ polymorphs included in these diamonds may represent retrogressed Ca-Pv. The estimated original bulk composition of the Cullinan Ca-Pv inclusion is consistent with compositions that are stable in subducted oceanic basalt protoliths at about 26 GPa, in the uppermost lower mantle²⁶. Our finding thus confirms the expectation from calculations¹⁰ and high-pressure experiments^21,25, that Ca-Pv is the main Ca-bearing phase in the lower mantle in both basic and ultrabasic compositions, reaching up to 23 vol% in MORB-like compositions²⁶. The combined bulk composition of the Ca-Pv phase found here provides overwhelming evidence of the return of recycled oceanic crust into the Earth’s lower mantle²⁰, whereas the relatively high carbon isotopic composition of the diamond in contact with the inclusion indicates the subduction of crustal carbon to lower mantle depths.

Micro-Raman spectroscopy

The Ca-Pv sample was analysed using an InVia Renishaw micro-Raman spectrometer installed at the Department of Chemical Sciences of the University of Padova. The spectra were baseline-corrected. A 632.8-nm-wavelength excitation laser was used at a power of 7 mW. The Raman spectrum of the Ca-Pv crystal was collected for 40 s using a 50× objective with a spatial resolution of 1.1 μm and a spectral resolution estimated to be about 3 cm⁻¹. The most intense Raman peaks observed for the Ca-Pv inclusion are (in order of decreasing intensity): 774, 247, 470, 337, 181 and 226 cm⁻¹.

Direct comparison between the Raman spectrum of natural Ca-Pv with that of CaTiO₃—both from the CaTiO₃ inclusions (Fig. 3) and from the RRUFF database²³—indicates that the two spectra are very similar. A small but important difference is the presence of limited traces of wollastonite-2M in the Raman spectrum of the natural Ca-Pv (see peaks at 971 cm⁻¹ and 637 cm⁻¹), which, as expected, are not evident in the spectrum of the CaTiO₃ perovskite inclusions.

On the basis of the results of ref. 29, the broad Raman bands in the 650–850 cm⁻¹ region correspond to second-order Raman scattering and only the sharp peaks in the 200–500 cm⁻¹ region are first-order Raman bands. However, we considered the entire Raman spectrum of natural Ca-Pv, regardless of first- or second-order scattering, for a direct comparison with CaTiO₃ perovskite and wollastonite.

The strong similarity between the Raman spectra of the CaTiO₃ and CaSiO₃ perovskites in Fig. 3b could indicate that the spectra are dominated by emission from a larger, underlying, unexposed volume of crystalline CaTiO₃ surrounded by a matrix of amorphous CaSiO₃. This possibility can be discounted for a number of reasons. First, the partially exposed inclusion is under some stress and this will affect the Raman band shift, depending on the elastic properties of the two perovskites. More importantly, the Raman spectrum of such a hypothetical large amorphous area of CaSiO₃ would be totally distinct from that measured here (Fig. 3b), and would be characterized by the presence of three very intense Raman bands at about 370 cm⁻¹, 640 cm⁻¹ and 970 cm⁻¹ (depending on the pressure and temperature conditions^30,31). These Raman bands are absent in the spectrum of the CaSiO₃ portion of the perovskite-structured inclusion. Also, because of the confocal nature of the Raman measurements and their small spot size, they cannot be substantially influenced by the spatially associated CaTiO₃ intergrowth. Last, the EBSD measurements rule out this possibility because EBSD is a surface technique (see Methods section ‘EBSD’).

Cathodoluminescence

The cathodoluminescence scanning electron microscopy image shown in Fig. 1 was obtained using a Philips XL 30 scanning electron microscope with a cathodoluminescence attachment consisting of a Hamamatsu R376 photomultiplier tube (EOAS UBC). The accelerating voltage was 20 keV and the electron beam current was 100 μA.

Infrared spectroscopy

Infrared spectra were collected on a Nicolet 6700 FTIR spectrometer. The absorbance spectra were measured at maximum light transmission for 40 s at a spectral resolution of 0.5 cm⁻¹. Background spectra were collected for 120 s before the analysis and were subtracted from each measured absorbance spectrum. The nitrogen concentration and aggregation were determined using the procedure described in ref. 32 using a spreadsheet (‘FTIR analyser 3d’) created by J. Chapman (Rio Tinto Diamonds Ltd). Preliminary processing and baseline determination were performed using the EssentialFTIR software. The analytical and processing error was ±10% (1σ, relative error). The FTIR spectrum of the Cullinan diamond studied here is shown in Extended Data Fig. 1.

Electron microprobe analysis

Quantitative chemical analyses were performed at the Department of Earth, Ocean and Atmospheric Sciences of the University of British Columbia, using a fully automated CAMECA SX-50 electron microprobe operating in the wavelength-dispersion mode with the following operating conditions: excitation voltage, 15 kV; beam current, 20 nA; peak counting time, 20 s; background counting time, 10 s; actual spot diameter, 5 μm. Data reduction was done using the PAP φ(ρZ) method³³. The detection limits for most oxides were lower than 0.08 wt% and those for Cr₂O₃, MnO₂ and NiO were lower than 0.12 wt%. Because of the small crystal size of the natural Ca-Pv and the presence of CaTiO₃ perovskite inclusions, we were able to perform only three reliable analyses; the results are reported in Extended Data Table 1. The Na and K contents were not analysed.

Scanning electron microscopy and EDS

We studied our sample using scanning electron microscopy and EDS to investigate the distribution of Ca, Si and Ti over the grain. We used a CamScan MX3000 electron microscope equipped with a LaB₆ source, a four-quadrant solid-state backscattered-electron detector and an EDAX EDS system for micro-analysis installed at the Department of Geosciences of the University of Padova. The measurement conditions were: accelerating voltage, 20 kV; filament emission, about 13 nA; working distance (the distance between the specimen and the lowest part of the electromagnetic lens in the column of the scanning electron microscope), 27 mm. The backscattered-electron image of the grain and its EDS maps for Ca, Si and Ti are shown in Fig. 2.

Single-crystal micro-X-ray diffraction

Single-crystal X-ray diffraction measurements were performed at the Department of Geosciences of the University of Padova, using a Rigaku Oxford Diffraction Supernova goniometer equipped with a Dectris Pilatus 200 K area detector and a Mova X-ray micro-source (Mo Kα radiation) operating at 50 kV and 0.8 mA. The sample-to-detector distance was 68 mm. Data reduction was performed using the CrysAlis software (Rigaku Oxford Diffraction) to obtain the Ewald projections shown in Fig. 3. The diffraction analysis results are reported in Extended Data Table 2 in comparison with those of a reference CaTiO₃ single-crystal sample³⁴ with the following unit-cell parameters: a = 5.388 ± 0.001 Å, b = 5.447 ± 0.001 Å, c = 7.654 ± 0.001 Å, volume, 224.63 ± 0.001 ų.

Carbon isotope analyses

The carbon isotope compositions (δ¹³C) reported in Extended Data Table 3 were determined using a Cameca IMS 7f-GEO secondary ion mass spectrometer. The polished diamond was pressed into an indium mount with a 1-inch-diameter aluminium holder. Natural reference diamonds with δ¹³C values between −13.6‰ (2σ = 0.3‰) and 2.6‰ (2σ = 0.3‰) were used to determine the instrumental mass fractionation and drift before and after sample analyses. The sample and reference diamonds were coated with gold (20 nm thickness). The measurements were conducted using ¹³³Cs⁺ at 10 keV impact energy and a beam current of about 4 nA. The 15-μm-diameter ¹³³Cs⁺ primary-ion beam was used for pre-sputtering. During the measurements, the ion beam diameter was reduced to 5 μm. Secondary ions of ¹²C and ¹³C were extracted at −9 keV with an energy bandwidth of 90 eV. No electron-gun charge compensation was required. The ¹³C/¹²C ratios were measured using dual Faraday cups. The mass resolving power M/ΔM was 2,900. The ¹²C and ¹³C ions were counted for 1 s in each cycle of the 30 cycles and the total measurement time for each spot was 8 min. The standard deviation of the analysis is estimated to be about 0.4‰ to 0.5‰ at the 2σ (95% uncertainty) level.

EBSD

EBSD analyses were performed at CNR-ICMATE in Padova, using a Quanta 200F FEG-ESEM system operating in high-vacuum mode with an accelerating voltage of 30 kV, emission current of 174 μA and beam spot of 4.5 μm, without any conductive coating. EBSD patterns were collected at a working distance of 10 mm and a specimen tilt of 75° using an EDAX Digiview EBSD system. The instrument was controlled by the OIMTM 5.31 software, which contains a large EBSD pattern database.

Statistical analysis of carbon isotope composition

We used a compilation of 1,473 carbon-isotope analysis datasets for diamonds containing inclusions of lithospheric peridotite paragenesis from ref. 35. We calculated the median absolute deviation³⁶ of the data using a b factor of 1.4826 and a very conservative threshold factor of 3.

Data availability

All relevant data are presented in Extended Data Tables 1, 2, 3 and Figs 1, 2, 3, 4. Original spectral data and electron microprobe data are available from the corresponding author.

We thank M. Regier for proofreading the paper. F.N. is supported by the European Research Council (ERC) Starting Grant number 307322. M.K.’s work and sample collection was possible thanks to an NSERC Discovery grant. N.K. acknowledges funding from the Dr. Eduard Gübelin Association through a 2015 research scholarship. D.G.P. was funded by an NSERC CERC award. M.A. was supported by the ERC under the European Union’s Horizon 2020 research and innovation programme (grant 714936) ‘TRUE DEPTHS’ and by the SIR-MIUR grant (RBSI140351) ‘MILE DEEp’. We thank L. Litti and M. Meneghetti of the Laboratory of Nanostructures and Optics of the Department of Chemical Sciences, University of Padova for their help in acquiring and interpreting the Raman data. F.N. and D.G.P. were supported by the Deep Carbon Observatory. M.G.P. was supported by NERC grant NE/M015181/1.

Affiliations

1. Dipartimento di Geoscienze, Università degli Studi di Padova, Via Giovanni Gradenigo 6, I-35131 Padova, Italy

   • F. Nestola

2. Department of Earth, Ocean and Atmospheric Sciences, University of British Columbia, Vancouver, British Columbia V6T 1Z4, Canada

   • N. Korolev
   •  & M. Kopylova

3. Institute of Precambrian Geology and Geochronology RAS, 199034 St Petersburg, Russia

   • N. Korolev

4. Dipartimento di Scienze della Terra, Università degli Studi di Milano, Via Botticelli 23, I-20133 Milano, Italy

   • N. Rotiroti

5. Department of Earth and Atmospheric Sciences, University of Alberta, Edmonton, Alberta T6G 2E9, Canada

   • D. G. Pearson

6. Department of Earth Sciences, University College London, Gower Street, London WC1E 6BT, UK

   • M. G. Pamato

7. Department of Earth and Environmental Sciences, University of Pavia, Via Ferrata 1, I-27100 Pavia, Italy

   • M. Alvaro

8. CNR-Istituto di Geoscienze e Georisorse, Sezione di Padova, Via Giovanni Gradenigo 6, I-35131 Padova, Italy

   • L. Peruzzo

9. University of Cape Town, Cape Town, South Africa

   • J. J. Gurney

10. Rhodes University, Grahamstown, South Africa

    • A. E. Moore

11. Petra Diamonds, Bryanston, South Africa

    • J. Davidson

Contributions

F.N. conceived the study, wrote the initial manuscript and performed X-ray diffraction and micro-Raman measurements. N.K. found the mineral, made original mineral identifications on a confocal Raman spectrometer, performed microprobe and cathodoluminescence measurements, prepared samples for secondary ion mass spectrometry measurements and assisted with the manuscript preparation. M.K. supervised the study of the Cullinan diamond collection, which was acquired by J.J.G., A.E.M. and J.D., and assisted with the manuscript preparation. D.G.P. made the geochemical interpretations and led the manuscript revisions. M.G.P. assisted with the manuscript preparation and crystallographic interpretations. N.R., M.G.P. and M.A. assisted with the X-ray data interpretation. L.P. collected and interpreted the EBSD data. J.J.G., A.E.M. and J.D. designed the sampling programme.

Competing interests

The authors declare no competing financial interests.

Corresponding author

Correspondence to F. Nestola.

Reviewer Information Nature thanks B. Harte and the other anonymous reviewer(s) for their contribution to the peer review of this work.

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

To obtain permission to re-use content from this article visit RightsLink.

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.