A universal concept for melting in mantle upwellings

Abstract

Deep mantle melting marks the onset of Earth differentiation¹, yet a unifying framework for how buoyancy-driven mantle upwellings initiate melting and how such incipient melts evolve within the asthenosphere has remained elusive. Here we show that the first melts generated in any solid-state mantle upwelling are kimberlitic CO₂-rich silicate melts that form at about 250 km depth through oxidation of elemental carbon to CO₂ (refs. ^2,3). Our experiments force a range of surface melts, derived from mantle plumes⁴ or broad upwellings⁵ (kimberlites, ocean island basalts and mid-ocean ridge basalts), into equilibrium with fertile mantle at adiabatic and super-adiabatic conditions at 7 GPa. The results define a framework in which redox melting at depth universally yields kimberlitic melts, which, while ascending through the asthenosphere by reactive porous flow^6,7, evolve to higher degrees of melting, lesser volatiles and incompatible elements, but higher SiO₂. Channelized flow⁷ in the lithosphere may then enable direct extraction of these melts, leading to kimberlites, where the lithosphere commences just above the C → CO₂ redox front, to alkaline Si-undersaturated intraplate magmas where lithospheric thicknesses are 150–100 km, and to tholeiitic basalts below mid-ocean ridges where voluminous ‘dry’ melting becomes overwhelming. This framework is consistent with the widespread seismic low-velocity zone at about 250 km beneath mid-ocean ridges^8,9 and aligns with ocean island and mid-ocean ridge basalts sampling the various geochemical mantle components at different degrees of melting in different proportions^10,11.

The first melts generated in any solid-state mantle upwelling are kimberlitic CO₂-rich silicate melts that form at about 250 km depth through oxidation of elemental carbon to CO₂.

Main

Silicate Earth primarily differentiates through mantle-derived magmatism¹, which shapes the modern surface of our planet and generates many economic resources. Mantle-derived magmas are grouped into three main suites that occur (1) at constructive plate margins, in which mid-ocean ridge basalts (MORB) form oceanic crust; (2) within tectonic plates, in which they span a wide spectrum from diamond-carrying kimberlites to ocean island basalts (OIB) to flood basalts; and (3) at destructive plate margins, in which predominantly calc-alkaline magmas generate continental crust^1,12. The latter series is intrinsically linked to subduction, in which descending plates provide surface-derived volatiles that lead to H₂O- and SiO₂-rich arc magmas. The other main magma suites form in the asthenosphere in deeply rooted mantle upwellings, which are fed by either giant convection cells below mid-ocean ridges⁵ or finger-like plumes that may originate as deep as the core–mantle boundary^4,13. Buoyancy in these upwellings arises principally from temperature differences¹⁴. On an adiabatic pressure–temperature trajectory with a potential mantle temperature T_P of 1,350 ± 50 °C, representative of the modern ambient mantle^15,16, the dry (volatile-free) mantle solidus is intersected only at shallow depths of 40–70 km (refs. ^17,18). Yet analyses of basaltic glasses and melt inclusions in olivine have shown that oceanic basalts contain variable, in part elevated concentrations of CO₂ and H₂O (refs. ^19,20), which enable melting at higher pressures.

A redox origin of CO₂ in deep mantle melts

In the asthenospheric mantle, CO₂ and H₂O are not readily available as neither carbonates nor hydrous phases are stable at adiabatic temperatures^2,21,22, and hence volatile-assisted melting requires a mechanism different from overstepping a carbonate-saturated solidus. Instead, melting is triggered by a carbon-related redox reaction^2,3. Below approximately 230 km depth, the ambient mantle contains Fe^III-rich majoritic garnet, a metal alloy with dissolved carbon and diamond^23,24. In upwellings, the majoritic garnet component destabilizes on decompression²⁵, which releases Fe^III, causing first any Fe-metal to comproportionate to FeO^olivine and immediately after causing carbon to oxidize through C⁰ + 2Fe₂O₃^majorite = CO₂^melt + 4FeO^olivine (refs. ^2,3). This reaction consumes excess Fe^III, leaving the mantle at the C⁰–CO₂^melt oxygen fugacity buffer and completes between 9 GPa and 7 GPa, that is, at latest at about 230 km depth^2,3. Most importantly, this process oxidizes reduced carbon (in metal alloy and diamond), which is insoluble in silicate melts, to CO₂, which lowers the peridotite solidus by 400–450 °C, from about 1,750 °C to 1,300 °C at 7 GPa (refs. ^26,27), that is, from well above to below an average adiabatic or super-adiabatic²⁸ temperature (1,420 °C and ≤1,670 °C at 7 GPa for a T_P of 1,350 °C and 1,600 °C, respectively).

Forced multiple saturation experiments

Here we show through a new type of ‘forced multiple saturation’ high-pressure experiments that any mantle upwelling will produce carbonated silicate melts of kimberlitic composition through the above redox reaction, regardless of the buoyancy flux and depth of origin of such upwellings. These melts transform thermal buoyancy-driven solid-state upwellings into melt-bearing ones and represent a rather uniform beginning of melting for intraplate and mid-ocean ridge settings. To demonstrate the general validity of this concept, multi-anvil experiments (Extended Data Table 1) were undertaken at 7 GPa, equivalent to the top of the redox-front depth. These experiments identify the precursor melts to kimberlites, OIBs and MORBs. We select compositions representing (1) a primitive silica-undersaturated OIB (basanite) with 5.4 wt% CO₂ and 1.9 wt% H₂O based on melt inclusion studies²⁰, confirmed by model calculations²⁹; (2) an MORB composition (tholeiite) with 1 wt% CO₂ as in the Atlantic popping rocks¹⁹, and 0.5 wt% H₂O, characteristic of enriched MORB³⁰; and (3) a kimberlitic melt with 10 wt% CO₂ and 2.5 wt% H₂O (ref. ³¹), including a correction for alkali loss (Methods and Supplementary Table 1). To perform reverse-engineering of the mantle melting process, we use an experimental strategy that forces these melts to equilibrate and saturate with the four principal mantle phases—olivine, orthopyroxene, clinopyroxene and garnet—maintaining mineral compositions akin to those of the mantle (Supplementary Table 2). As the resulting carbonated silicate melts are unquenchable and do not form glasses, large melt pools are required for measurements averaging across the quench crystallites. This meant that experimental mineral to melt ratios could not exceed 2–3, and mineral proportions, melt fraction and melt composition had to be iteratively adjusted such that the final melts were in equilibrium with all four mantle minerals (Supplementary Fig. 1). Highly incompatible elements were set so that equilibrium clinopyroxene compositions correspond to those observed in fertile mantle rocks (Methods). As mantle upwelling temperatures are adiabatic or hotter, we use 1,400–1,420 °C at 7 GPa (corresponding to a T_P of 1,350 °C) and super-adiabatic temperatures of 1,480–1,630 °C at the same pressure, covering most of the range proposed for mantle plumes^32,33.

Extended Data Table 1.

Run Table

¹ all at 7 GPa, Re-Pt double capsule, graphite layer in the inner capsule.

² as given in Supplementary Table 1.

³ four-phase saturated key experiments in bold.

⁴ phase compositions in Supplementary Table 2.

The principal result is that all three surface melt types—kimberlites, alkaline basanites (OIBs) and tholeiitic basalts (MORBs)—equilibrate to carbonated silicate melts of broadly kimberlitic composition at 7 GPa when saturated with the four mantle phases (Fig. 1). Specifically, these melts have about 23–27 wt% SiO₂, 21–23 wt% MgO, 13 wt% CaO, 4.4–6.1 wt% total alkalis and 14–24 wt% H₂O + CO₂ at 1,400–1,420 °C; and about 35–39 wt% SiO₂, 23 wt% MgO, 9 wt% CaO, 3.7–4.9 wt% total alkalis and 3–9 wt% H₂O + CO₂ at 1,630 °C—that is, at 210 °C excess temperature compared with a T_P of 1,350 °C (Fig. 2, Table 1 and Extended Data Fig. 1). Al₂O₃ contents are only 1–3 wt% for all temperatures. This fundamental major element characteristic of these incipient mantle melts can be related to a self-regulatory mechanism in peridotites, the dominant upper mantle lithology, in which the olivine–orthopyroxene pair buffers SiO₂, FeO and MgO contents (for a given temperature), MgO–CaO is buffered by the pyroxene pair, and Al₂O₃ is controlled by the garnet–pyroxene equilibrium, limiting Al₂O₃ to remarkably low values. Na and K are regulated by partitioning between melt and clinopyroxene, and only the minor elements P (phosphorus) and Ti are controlled by mantle heterogeneities.

Fig. 1. Major element variation diagram discriminating the effects of temperature, pressure and volatiles on mantle melts.

The experiments of this study (large diamonds) simulate redox melting at 7 GPa for potential mantle temperatures T_P of 1,350–1,560 °C, yielding a narrow field of redox melts irrespective of whether a kimberlite, basanite (OIB) or tholeiite (MORB) starting melt is used. The redox melts overlap with the low-Mg, low-Si end of the kimberlite field (Methods), in which higher Mg and Si mirror assimilation of lithospheric mantle wall rocks. Kimberlites then form from asthenospheric melts extracted directly from 6 GPa to 7 GPa following redox melting. Asthenospheric melt evolution with decreasing pressure is modelled to 3 GPa (grey lines ending in a star) using appropriate melting reactions²⁶ and mineral compositions (Methods). The expected increase in the degree of melting F from 7 GPa to 3 GPa from about 1.5% to about 10% for T_P values of 1,560 °C and from 0.5% to 3–5% for T_P values of 1,350 °C (tick marks are increments of 1% and 0.5%, respectively). Volatile-free experimental peridotite melts at 1–4 GPa are also plotted (red crosses indicate those with residual clinopyroxene and white crosses without residual clinopyroxene; Methods), the dotted curve labelled ‘cpx’ marks the stability limit of clinopyroxene, which exhausts at F = 20–25%, much higher than the melting degrees permissible for any OIB composition. A comparison with primitive (X_Mg > 0.65) OIBs worldwide (n = 747), including primitive rejuvenated (n = 123) and shield stage (n = 225) Hawaiian lavas, shows that these can stem from neither low pressure nor dry melting, but require high-pressure CO₂-rich (kimberlitic) carbonated silicate melts that evolve through porous reactive flow to the asthenosphere–lithosphere boundary. For reference, high-Mg MORBs (X_Mg > 0.65, black circles, n = 715) are given; pyroxenite melts have even lower MgO and higher Al₂O₃ contents. Yet any small pyroxenite contribution would be eradicated in terms of major elements through equilibration with peridotite. Bars are 1σ s.e.

Fig. 2. Major element variation diagrams illustrating the alkaline and low-Al nature of redox melts and their evolution towards OIB magmas.

a,b, Diagrams of total alkalis (a) and Al₂O₃ (b) compared with SiO₂. The redox melts evolve to the observed array of OIB melts by progressive equilibration during asthenospheric rise (see Fig. 1 caption for details of plotted symbols and curves). The high alkalis in Si-undersaturated OIB melts testify to low-degree melting (a), whereas the low Al₂O₃ contents indicate high pressures of melting (b). The largely tholeiitic Hawaiian shield stage lavas also have low Al₂O₃ contents, such that parent melts akin to our 7 GPa melts are required (see also Extended Data Fig. 1). Their higher SiO₂ compared with OIBs elsewhere is likely to be attributed to partial equilibration in the lithospheric mantle. Note that kimberlites do not preserve magmatic Na₂O contents³⁶; reconstructed values are of the order of 3–5 wt% (ref. ⁵⁴) and total alkalis accordingly higher. Bars are 1σ s.e.

Table 1.

Major element composition of experimental and natural melts

	Ocean islandsa			Mid-ocean ridgesa			Kimberlitea		Primitive kimberlite Letsengc	Hawaii shield parentd
T (°C)	1,420	1,480	1,630	1,420	1,480	1,630	1,400	1,580
	aob12	aob11	aob5	m11	m10	m2	j15-73b	j16-1
SiO2	23.1	26.1	35.3	26.6	33.7	39.1	23.1	27.1	23.7	42.9
TiO2	4.0	3.2	1.9	4.1	4.5	2.5	1.6	1.2	3.2	0.5
Al2O3	1.3	1.6	3.0	2.1	2.4	3.5	1.5	1.7	2.6	8.3
Cr2O3	0.0	0.2	0.3	0.1	0.1	0.2	0.4	0.1	0.1	0.0
FeOtot	11.9	12.3	11.9	11.4	13.2	14.7	8.0	8.2	10.0	13.9
NiO	0.1	0.1	0.1	0.1	0.2	0.1	–	0.1	–	–
MnO	0.3	0.3	0.2	0.4	0.3	0.3	0.2	0.2	0.2	0.1
MgO	21.2	21.4	22.8	22.2	22.6	23.8	23.1	24.7	19.9	20.5
CaO	13.3	12.9	9.3	12.4	9.6	9.0	13.0	12.4	16.4	8.1
Na2Oe	3.6	3.6	3.1	3.5	2.9	2.8	3.2	3.2	4.3	3.8
K2Oe	2.5	2.2	1.8	1.9	1.4	0.9	1.2	1.2	2.0	0.3
P2O5	0.5	0.5	0.3	1.0	0.7	0.4	0.4	0.2	1.1	0.0
CO2e	15.4	11.5	7.3	9.4	5.5	1.7	16.4	15.7	12.0	1.1
H2Oe	2.7	4.1	2.6	4.7	2.8	0.8	7.9	4.0	4.5	0.4
Total	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0	100.0
CO2 + H2O	18.1	15.6	9.9	14.2	9.3	2.5	24.3	19.7	16.5	1.5

^aStarting composition. ^bFrom ref. ³¹. ^cModified after ref. ⁵⁴. ^dModelled asthenospheric parent melt to the Hawaiian shield stage at F = 0.10, 90 km depth. ^eFrom mass balance.

Extended Data Fig. 1. Further major element variation diagrams.

The green field represents the redox melts formed at 7 GPa, i.e. ca. 230 km depth. Melt evolution with decreasing pressure is modeled to 3 GPa (grey lines ending in a star) via stepwise equilibration with ambient mantle using appropriate melting reactions²⁶ and mantle mineral compositions (see Methods). The open star represents the expected increase in the degree of melting at 3 GPa to ≈10% at a T_P of 1560 ^oC (as likely for the shield stage of Hawaii) and to 4% for a T_P of 1350 ^oC (as characteristic for the Cape Verde and Canary hot spots). Tick marks are 1 and 0.5 % melting increments, respectively. Volatile-free experimental peridotite melts at 1-4 GPa (crosses, for details see Methods) are also plotted. The dotted curve labelled ‘cpx’ marks clinopyroxene exhaustion at melting degrees of 20–25%, much higher than permissible for any OIB composition. A comparison to primitive (X_Mg > 0.65) OIBs worldwide (pale green circles), including primitive rejuvenated (blue) and shield stage (orange) Hawaiian lavas, shows that the natural OIB melts can neither stem from low pressure nor dry melting, but require high-pressure CO₂-rich kimberlitic melts that evolve through porous reactive flow to the asthenosphere-lithosphere boundary. For reference, high-Mg MORBs (X_Mg > 0.6, black circles) are also plotted⁵⁷. Note that kimberlites do not preserve magmatic Na₂O contents³⁶, reconstructed values are in the order of 3-5 wt%⁵⁴. For further explanation see also main Figs. 1 and 2 and Methods.

Volatiles in incipient melts

Experimental melt volatile contents partly depend on the initial starting compositions. Nevertheless, in nature, H₂O in the carbonated silicate melt arises from partitioning with the nominally anhydrous minerals. With a bulk partition coefficient D_H^melt/mantle of about 0.004 and average mantle H₂O contents of 80–200 ppm (ref. ³⁴), 2–5 wt% H₂O are expected in the 7 GPa melts, which we match in the experiments (Table 1). Because CO₂ behaves perfectly incompatible in the adiabatic mantle, the main role of CO₂ is then to control melt fraction (F), with higher bulk CO₂ corresponding to higher F. The amount of carbon available in the convective mantle is 100–400 ppm (ref. ²⁹). Although most of this carbon is primordial, some carbon is added through redox freezing above deeply subducted plates, where slab-derived CO₂-rich melts and fluids migrate into metal-bearing mantle and CO₂ is reduced to C⁰ according to CO₂^{melt or fluid} + 2Fe⁰ = C⁰ + FeO^olivine (ref. ²). This process, which is widely documented in the formation of sub-lithospheric diamonds based on their carbon isotope systematics typical of subducted biogenic carbon³⁵, adds carbon and net-oxidizes these carbon-enriched mantle domains. These metasomatized volumes would be the first to undergo the inverse process of redox melting as described above. The range of 100–400 ppm carbon for depleted to enriched mantle can then be combined with 9–16 wt% CO₂ in the incipient melts (for a T_P of 1,350 °C), which results in 0.3–0.9 wt% incipient melting, which probably redistributes towards a characteristic average value of 0.4–0.7 wt%.

The first carbonated silicate melts form locally in the most oxidized volumes with the highest carbon and the lowest metal contents. These melts would then begin to migrate upwards, but will undergo redox freezing if they encounter metal-bearing, C-poorer reduced rocks during upward percolation. Multiple redox-melting and redox-freezing cycles are, therefore, likely and may continue until either the Fe^III-bearing majorite garnet component or the elemental iron and carbon are entirely consumed. Only when no metal-bearing mantle hinders further percolation, these carbonated silicate melts can begin their way to the surface. This depth corresponds to 7 ± 1 GPa (ref. ²⁵) and is the shallow limit of majoritic garnet stability. The lithosphere–asthenosphere boundary at kimberlite eruption sites occurs at about this depth³⁶, which explains why the melts forming natural kimberlites correspond to the experimental incipient carbonated silicate melts identified here (Fig. 3).

Fig. 3. Conceptual diagram of the beginning of melting in mantle upwellings and melt evolution within an adiabatic melting column.

Redox melting, which completes with the exhaustion of metal at about 7 ± 1 GPa, yields (kimberlitic) carbonated silicate melts at a degree of melting F of about 0.4–0.7 wt% for a potential temperature of 1,350°C and about 1.5 wt% for 1,550 °C. Experimentally, these melts equilibrate with 0.5 mm mantle minerals within hours, suggesting that porous flow in the asthenosphere leads to full equilibration at increasing degrees of melting until the rising melts reach the lithosphere–asthenosphere boundary. With lithospheric thicknesses of 6–7 GPa (ref. ³⁶), kimberlitic melts may be directly extracted from the redox melting depth. Ocean islands with highly Si-undersaturated primitive melts (for example, Cape Verde, Canary and rejuvenated Hawaii) remain equilibrated to lithospheric thicknesses corresponding to about 3 GPa (ref. ³⁹), hence evolving in the asthenospheric mantle towards higher SiO₂ and lower alkali and CO₂ contents than kimberlitic melts. Within the lithosphere, channelled flow may shield these melts from interaction with wall rocks and avoid lower-pressure equilibration. However, this isolation is much less likely for the higher-temperature melts of the most intense plumes, such as for the Hawaiian shield stage. Below mid-ocean ridges, the melting column reaches to <10 km depth. At about 2 GPa, the dry solidus is overstepped and the degree of melting increases substantially, leaving only a faint signature of deeper melting. The pressure of 7 GPa also corresponds to the depth of 230 km of the low-velocity zone below mid-ocean ridges and ocean islands^8,9, linking redox melting to a worldwide geophysical anomaly.

Asthenospheric melt evolution

In all other cases (that is, with a thinner lithosphere), the incipient kimberlitic melts will evolve when rising in the asthenosphere by reaction with peridotitic ambient mantle (Fig. 3). This modification explains how the rather uniform incipient melt composition proposed here transforms into the very diverse surface magmas forming above mantle upwellings. As asthenospheric melts migrate by grain boundary percolation in porosity waves^6,7, melt compositions will equilibrate with decreasing pressure while melt fractions increase. First, the major element systematics of mantle peridotite melting is such that decreasing pressure and CO₂ shift equilibrium melts to higher SiO₂ and lower (Mg,Fe)O (refs. ^37,38). Second, the dT/dP slopes of high-pressure silicate-melt liquidi²⁶ are steeper than the mantle adiabat (of about 10 °C GPa⁻¹; ref. ¹⁶), which implies that rising melts overheat and dissolve further material, leading to an increase in the degree of melting. The combined effect is that melt CO₂ and alkali contents decrease while the melts become less Si-undersaturated and their incompatible trace elements more diluted. Melts are thus expected to evolve from kimberlitic to basanitic and then to tholeiitic basalts, spanning compositions from strongly alkaline and Si-undersaturated when erupting over thick oceanic lithosphere (for example, the Cape Verde and Canary hotspots with a lithosphere of about 90 km thickness³⁹), to largely tholeiitic above thin oceanic lithosphere (20–30 km at Galapagos and Ascension³⁹). Finally, in the extreme case of shallow asthenospheric melting below mid-ocean ridges, the dry solidus is overstepped, and the approximately 0.5 wt% incipient kimberlitic melts are completely blurred out by the 10–15% melting at depths of <30 km (ref. ⁷), which yields tholeiitic, comparatively Si-rich MORBs.

Furthermore, contrasting the Cape Verde and Canary plumes, for some plumes, excess temperatures of up to 250 °C have been proposed²⁸, most prominently for Hawaii. The ensuing higher degrees of initial melting, mirrored by our 1,630 °C OIB experiment, cause higher initial silica (about 35 wt%), and more dilute volatiles (CO₂ + H₂O = 9.9 wt%; Table 1) and alkalis (Na₂O + K₂O = 4.9 wt%) with respect to the 1,420 °C OIB melts. The tholeiitic character of the Hawaiian shield stage, thought to represent the highest potential temperature in plumes²⁸, has been related to various factors, including increased H₂O contents, thinned lithosphere and consequent shallower continuation of melting, or higher degrees of asthenospheric melting due to higher excess temperatures^28,40. Our experiments define the incipient melt compositions arising from redox melting for a T_P of 1,560 °C (or 1,630 °C at 7 GPa). We then model their evolution to 3 GPa, the pressure at the base of the lithosphere below Hawaii³⁹, by using appropriate melting equations²⁶ (Methods). This modelled melt constitutes a suitable parent for the Hawaiian shield stage basalts (Figs. 1 and 2). By contrast, high-degree melting at 1–3 GPa does neither fit the characteristically high MgO/CaO ratios of the Hawaiian shield stage nor their SiO₂/Al₂O₃ ratios or alkali contents (Figs. 1 and 2). Thus, based on major elements, we postulate that the tholeiitic melts produced during the Hawaiian shield stage probably form through a three-stage process: (1) low-degree melting at the redox front near 7 GPa of about 1.5% (that is, threefold that of an upwelling with a T_P of 1,350 °C), (2) melt evolution during upward percolation from 7 GPa to 3 GPa to form a suitable parent for the shield stage magmas within the asthenosphere (Figs. 1 and 2 and Table 1); and (3) partial re-equilibration within the lithosphere, driven by the large excess temperature. Yet the post-shield or rejuvenated magmas from Hawaii are similar to the highly Si-undersaturated plume melts of, for example, Cape Verde, that is, melts produced at a T_P of about 1,350 °C, suggesting that these tap the margins of the main plume.

The geochemical context

A complementary perspective on the common parentage of kimberlites and oceanic basalts is provided by the Sr–Nd–Hf–Pb isotope geochemistry and by trace element patterns, which show that kimberlites, intracontinental mafic magmas and OIBs share (1) the same sources in the convective mantle^10,41,42 and (2) a continuum in trace elements with melt fractions increasing from kimberlites to OIBs^41,42. Furthermore, it is well established that the higher melting degrees associated with MORBs compared with OIBs diminish the role of mantle heterogeneities in the trace element and isotopic compositions of MORBs, which are geochemically more depleted and less variable than OIBs¹⁰. The similarities and differences in isotopic composition between kimberlites, OIBs and MORBs testify to sampling several mantle components in different proportions, mainly a dominant peridotite with minor pyroxenite. Pyroxenites are generally considered to play an important part in deep melting¹⁰ as they are enriched in incompatible trace elements and may carry distinct isotopic signatures such as enriched mantle and high U/Pb compositions prominent in OIBs. These signatures could be acquired during incipient melting or en route to the surface when the incipient melts percolate through heterogeneous mantle containing pyroxenites.

Owing to higher alkalis and a lower X_Mg, dry pyroxenites have slightly lower melting temperatures (by around 30 °C) than dry peridotitic mantle^43,44. However, fertility in the deep mantle is not governed by these subtle differences but by redox potential and carbon content because CO₂ lowers the melting point of peridotites and pyroxenites by about 400 °C. The subducted igneous oceanic crust, ultimately forming pyroxenites, is more oxidized than the ambient mantle, in particular, when carbonate is added through hydrothermal seawater circulation. Yet the oceanic crust loses most of its carbonates during ongoing subduction at depths of 300–500 km (refs. ^45,46), at the latest when relaxing to ambient mantle temperatures⁴⁷, leaving a carbon-poor recycled crust to enter into large-scale mantle convection at greater depths^48,49. As both metal and carbon are unlikely to be substantial in typical pyroxenites (that represent recycled oceanic crust) entrained in mantle upwellings, a fertility inversion is probable. Consequently, metal- and carbon-bearing peridotites are expected to melt earlier and to a larger degree than C-free pyroxenites. This does not preclude a contribution of pyroxenite melts (as documented by specific traces and isotopes¹⁰). However, in terms of major elements, this contribution will rapidly vanish because of equilibration with the overwhelmingly peridotitic mantle during porous flow.

The effect of hydrogen versus carbon

Our experiments indicate that kimberlitic melts also feed the mantle column, which ultimately generates MOR tholeiites at low pressure, a conclusion supported by independent petrological and geophysical arguments. First, it has long been known that MOR basalts contain a melt component that stems from the garnet stability field⁵⁰, corresponding to >50–70 km at adiabatic conditions (Fig. 3). A solution often proposed is that melting is triggered by H₂O that is shed from the nominally anhydrous minerals on decompression³⁴. However, for typical mantle H contents of 100–200 ppm (ref. ²⁵), this does not happen at depths greater than 60–90 km (refs. ^34,51), and hence H-assisted melting will be preceded by C-assisted redox melting. Second, the geophysical low-velocity zone, thought to be caused by a small melt fraction, reaches down to 200–250 km depth^8,9, which is consistent with the depth of redox melting triggered by carbon. We hence posit that incipient melting beneath mid-ocean ridges is due to carbon oxidation^2,3, with hydrogen playing a passive role controlled by partitioning between nominally anhydrous minerals and the carbonated silicate melt. Also, H₂O-assisted melting causes melt compositions to shift towards higher SiO₂ compared with volatile-free melting³⁷ (as characteristic for arcs), whereas the compositions of kimberlites and alkaline basalts require the opposite effect, namely, a strong decrease in SiO₂ as caused by CO₂ (refs. ^36,38).

A universal concept of melting in mantle upwellings

In summary, we postulate that there is no principal difference in the initial melting mechanism of mantle-derived magmas from intraplate settings and divergent plate margins. The concept of similar kimberlitic incipient melts that evolve as a function of depth while rising in the asthenosphere provides an overarching theoretical framework for melting in mantle upwellings, regardless of buoyancy and potential temperature. Decompression of carbon-bearing peridotites induces majoritic garnet to liberate Fe^III that oxidizes C⁰ to CO₂, triggering the formation of kimberlitic melts. The widespread occurrence of this process is illuminated by the occurrence of seismically detected low-velocity zones below mid-ocean ridges at depths of 200–250 km (refs. ^8,9), which corresponds to the conditions of carbon oxidation along a typical mantle adiabat²¹. The occurrence of kimberlites, that is, carbonate-rich silicate melts extracted from the convective mantle from depths of 200–250 km before being channelized along translithospheric fractures and erupted at the surface, provides direct evidence for the formation of these melts by carbon oxidation³⁶.

The formation of compositionally similar incipient kimberlitic melts that vary as a function of the excess temperature of the upwelling provides a unifying concept of deep mantle melting on a global scale and explains the various surface expressions of intraplate magmatism as well as the precursor melts to mid-ocean ridge magmatism. The variability in erupted magmas ultimately arises from the equilibration of this incipient melt to the uppermost limit of the asthenosphere, varying from about 250 km beneath continental cratons to <10 km depth at mid-oceanic ridges. Although a similar lid effect has been previously postulated to explain correlations between lithospheric thickness and Si-content or alkalinity of both OIBs^39,52 and continental intraplate magmas⁵³, our study shows the origin of these melts and the role of incipient kimberlitic melt as the starting composition of the vast range of magma types associated with mantle upwellings. Increasing melt fraction with decreasing depth and equilibration with peridotite minerals control the major element composition of erupted magmas, whereas part of the incompatible element and radiogenic isotope variability is dictated by mantle heterogeneities, including those arising from subducted material.

Methods

Experimental apparatus

All experiments were conducted in a 6–8 multi-anvil apparatus with 32 mm edge length WC cubes featuring 11 mm truncations and 18 mm edge octahedra, calibrated against the transitions of quartz–coesite, garnet–perovskite in CaGeO₃, and coesite–stishovite. The assembly consists of a Cr-doped MgO-octahedron, a zirconia sleeve, a stepped graphite furnace, inner MgO parts and a PtRh B-type thermocouple. Full details, including pressure calibration, are provided in ref. ²⁴. Based on previous experience with similar temperatures and melt compositions, equilibrium is typically achieved within 1–2 h. Consequently, our run times ranged from 2 h to 8 h, increasing with decreasing temperature. The starting mixtures were loaded into Re capsules folded from foil, placing a thin graphite layer at the bottom and top. These were then inserted into a Pt capsule, which was welded shut to contain the volatiles. The graphite–CO₂^melt pair yields the appropriate oxygen fugacity for redox melting, in which C⁰ oxidizes to CO₂.

Starting compositions

The experiments (Extended Data Table 1) equilibrate erupted surface melts with a lherzolite mantle at 7 GPa and temperatures of 1,420–1,630 °C. These conditions correspond to mantle potential temperatures (T_P) of 1,350–1,560 °C, encompassing a typical mid-ocean ridge adiabat^15,16 and an excess temperature of 210 °C, similar to the hottest mantle plumes^32,33. Initial starting materials (Supplementary Table 1) include an average primitive ocean-island basanite (APB, see below) and an average of close-to-primitive MORBs, corrected for olivine fractionation (MBB). The OIB composition contains 5.4 wt% CO₂ and 1.9 wt% H₂O, based on melt inclusion data²⁰. For the MORB, we selected the maximum observed CO₂ content of 1.0 wt% as in the popping rocks¹⁹ and 0.5 wt% H₂O, which is in the upper range of the MORB average (0.28 ± 0.24 wt% H₂O; ref. ³⁰). As erupted melts are largely degassed, reconstituting their volatile concentrations was necessary. Moreover, we extended our previous study on kimberlites³¹ to 1,580 °C, using a 1,400 °C experimental melt³¹ derived from a starting bulk composition with 7.5 wt% CO₂ and 3.5 wt% H₂O (JER1555).

All starting compositions were mixed by weight from previously dried chemicals, that is, SiO₂, TiO₂, Cr₂O₃, NiO, MgO, Al₂O₃, MnO and Na₂SiO₃, and from synthetic leucite, wollastonite, apatite and fayalite. All iron was present as Fe^II in the starting material. H₂O and CO₂ were introduced as Mg(OH)₂, synthetic CaCO₃ and natural magnesite. Powders were mixed in an agate mortar and homogenized in alcohol in a planetary mill. The starting materials were then kept in a desiccator and again dried at 110 °C before use.

Iterative forced multiple saturation experiments

The starting melt compositions were iteratively equilibrated with a four-phase lherzolite (olivine, orthopyroxene, clinopyroxene and garnet) modelled after peridotite KLB-1. Capsules were loaded either in a sandwich configuration with the melt placed in between two layers of synthetic powders representing mantle peridotite or as homogeneous mixtures of both components. Multiple iterations were necessary to (1) saturate the melt in all four lherzolite mantle phases and (2) to obtain large melt pools rather than interstitial melts (see Supplementary Fig. 1 for back-scattered electron images of the experiments). Melts rich in CO₂ + H₂O rarely quench into glass but rather form intergrown crystallites, requiring measurement areas larger than the crystallites to ensure accurate melt compositions. To address these challenges, subequal melt/peridotite starting proportions are required, and a suite of 13 and 11 experiments were conducted on the MORB and OIB compositions, respectively, to obtain equilibrium melt compositions at 1,420 °C, 1,480 °C and 1,630 °C (7 GPa). During these iterations, the peridotite component was adjusted by modifying the relative proportions of olivine, orthopyroxene, clinopyroxene and garnet, counteracting complete dissolution of any single mineral. Simultaneously, the melt composition was progressively refined (MBB → MORB2 → MORB10 and APB → OIB5 → OIB11; Extended Data Table 1 and Supplementary Table 1) until saturation in all four mantle minerals was reached and mineral compositions closely resembled those found in the mantle. A further experiment was run on a kimberlitic composition at 1,580 °C, which is complemented by the 1,400 °C melt composition previously obtained for kimberlites using a similar method³¹.

Preparation of experimental run products

The CO₂ + H₂O-bearing melts quench into a mixture of clinopyroxene, calcite and Na-rich interstitial carbonate-rich material and some interstitial voids, well observable (see detailed images in Supplementary Fig. 1). When left in the open, this quench material grows whiskers of several mm length within hours or days, leading to the destruction of the experimental charge. Thus, all preparation of the experimental run products was done minimizing the presence of humidity and avoiding water. Samples were exclusively handled in a purpose-built glove box that maintains a <2% relative humidity environment. Grinding and polishing were achieved using first dry SiC abrasive paper and then dry diamond powder. The very final polish was done using kerosene, after which samples were immediately coated and loaded either into the scanning electron microscope (for first imaging) or electron microprobe (for element quantification). Immediately after measurement, samples were impregnated with an epoxy droplet, protecting them from humidity. For any re-measurement, the polishing procedure was then repeated.

Experimental textures

To measure melt compositions, large melt pools are required. The low-viscosity melts migrate into the warmest part of the capsule, such that they form a girdle or belt in the equatorial region of the capsule or, when melt fractions are large, the crystals are found in an hour-glass-shaped zone at the colder ends of the capsule. These textures resulted irrespective of whether a homogeneous mix or a layered melt-peridotite starting configuration was used.

Melt interstitial to equilibrium minerals quenches into a combination of growth rims on equilibrium silicate minerals and sparitic intergrowth, leading to an overestimation of carbonate content when measuring the interstitial quench crystallites only. Large melt pools quench to fine-grained (with crystallites of typically 3–30 μm) or even glassy quench textures in their centre, and the margins of the melt pool develop coarse quench with abundant voids between the larger quench crystals (typically 50–150 μm). Whenever possible, we thus measured the fine quench. Experiments without any melt pools were discarded. Supplementary Fig. 1 provides an overview of the entire capsule for each experiment (polished to the centre axis, oriented E–W in the images) and detailed images of the quenched melt and the mineral-rich region.

Electron microprobe measurements

Mineral and melt compositions were measured with a JEOL JXA-8230 electron microprobe equipped with five wavelength-dispersive spectrometers and one energy-dispersive spectrometer. For minerals, 15 kV and 20 nA and a focused beam were used. For melt analyses, a maximum size defocused electron beam (20 μm) and 4 nA beam current were used. Most mineral and melt compositions were averaged from 5 to 25 measurements (Supplementary Table 3). Calibration standards included albite (Si and Na), anorthite (Al and Ca), forsterite (Mg), fayalite (Fe), synthetic rutile (Ti), chromite (Cr), microcline (K), synthetic pyrolusite (Mn), synthetic bunsenite (Ni) and apatite (P). Furthermore, natural hornblende and basaltic glass standards were used to monitor the accuracy of the calibration.

Melt compositions in terms of volatiles and alkalis

Melt compositions are given as measured by electron microprobe in Supplementary Table 2. CO₂ and H₂O cannot be directly measured in the quenched melt, and, despite completely water-free preparation of the experimental charges, measured alkali concentrations in the melt fall 20–50% below the values expected from mass balance. Thus, CO₂, H₂O, Na₂O and K₂O concentrations in the melt were determined from mass balance. Using the perfect incompatibility of CO₂, H₂O and K₂O, we calculate their contents in the melt using, for example, CO₂^melt = CO₂^bulk/melt fraction. For Na₂O, compatible in clinopyroxene and present in minor amounts in garnet, the mineral Na concentrations were considered when mass balancing for Na₂O^melt. This approach yields experimental melt alkali contents, from which we determined D_Na^cpx/melt. Using this partition coefficient, we then calculated the Na₂O^melt that would be in equilibrium with a lherzolitic clinopyroxene containing, on average, 1.27 wt% Na₂O (ref. ⁵⁵), thus simulating the incipient melts as they would occur in fertile mantle (Table 1; original measurements in Supplementary Table 2). In general, CO₂ and H₂O estimates from mass balance closely match those derived from the difference-to-100 method from electron probe microanalysis, with discrepancies mostly within ±2 wt%. Discrepancies up to −5 wt% were encountered when either the melt quench was relatively coarse-grained or quite porous, or the surface was irregular—all factors leading to lower microprobe totals. Finally, the mass balance assumes a closed system, although some leakage of hydrogen through the capsule walls is possible. Nevertheless, this leakage leads to a net Fe oxidation and hence increase of X_Mg in all phases, which is not observed. We thus posit that hydrogen loss is minor.

Compilation of natural and experimental data

The experimental melts are compared with 747 primitive OIB compositions selected from the GEOROC (Geochemistry of Rocks of the Oceans and Continents) database, using 0.65 < X_Mg < 0.75 and 200–700 ppm Ni as main criteria⁵⁶, singling out the rejuvenated (n = 123) and the shield stages of Hawaii (n = 225). For the starting material, an average was taken from the 13 ocean islands, which have >10 primitive melt compositions, excluding the tholeiitic shield stage of Hawaii. The MORB starting composition represents a composite average using analyses with X_Mg > 0.6 from the largest available MORB compilation⁵⁷, corrected for equilibrium with mantle olivine of X_Mg = 0.89 by stepwise addition of olivine (MBB). In detail, the fine-tuning of the initial starting compositions is inconsequential, as these compositions were equilibrated with the four-phase mantle lherzolite—that is, they adjust themselves to equilibrium with the peridotitic mantle. Yet, the closer the starting material to the equilibrium melt composition, the fewer iterations are required.

Kimberlitic rocks do not represent melt compositions, but the product of low-degree asthenospheric mantle melting, lithospheric contamination, degassing and alteration on emplacement, the latter two removing almost all Na₂O (refs. ^36,41). The kimberlite fields in Figs. 1 and 2 and Extended Data Fig. 1 are derived from a compilation of coherent kimberlites⁵⁸. We use the 50% envelope of the kernel density distribution of these data to exclude the most altered and modified samples.

Apart from our own experimental data, we plot the volatile-free peridotite melts at 1 GPa (refs. ^59,60) (MgO/CaO ≈ 1) and at 3 GPa (ref. ²⁶) (MgO/CaO > 1.5) to degrees of melting at which clinopyroxene melts out, which exceeds by far the degrees of melting relevant for OIBs and MORBs (Fig. 1).

Modelling asthenospheric melt evolution from 7 GPa to 3 GPa

We model the evolution of the 7 GPa OIB-type melts to the characteristic asthenosphere–lithosphere boundary (LAB) depth of about 90 km (ref. ³⁹) for oceanic crust aged >120 Ma (as, for example, applicable to Hawaii, the western Pacific, Cape Verde and the Canary Islands). For a potential temperature T_P of 1,560 °C, representing plumes with 200 °C excess temperature, we start with the 1,630 °C OIB melt with 7.3 wt% CO₂ (corresponding to about 1.5 wt% melting) and model melt evolution to a melt fraction of 10 wt%, as estimated for the Hawaiian shield stage⁶¹. This increase would reduce CO₂^melt to about 1 wt%, consistent with the upper bound found in Hawaiian shield melt inclusions⁶². For comparatively cooler plumes (for example, Cape Verde and Canary) with a T_P of 1,350 °C, we model the evolution of the 7 GPa, 1,420 °C OIB-type melt with 15.4 wt% CO₂ (corresponding to about 0.5 wt% melting) to a melt fraction of 4 wt%, based on CO₂ contents of 3–5 wt% in primitive melts of these hotspots^29,63.

For a T_P of 1,560 °C, we apply the melting equations and mineral compositions for dry peridotite melting²⁶ obtained at similar temperatures, which are, for melt fractions ≤10% (in wt units): 26 olivine + 50 clinopyroxene + 24 garnet = 100 melt (7 GPa) and 7 olivine + 68 clinopyroxene + 25 garnet = 84 melt + 16 orthopyroxene (3 GPa).

Using these equations for our moderately volatile case is justified as CO₂ and K₂O in the melt remain fully incompatible from 7 GPa to 3 GPa. Na₂O^melt concentrations are calculated using melt–clinopyroxene partitioning, with adjustments at melt fractions >4% for the decreasing Na₂O^cpx. For a T_P of 1,350 °C, we use the same melting equations, as no others are available, but mineral compositions as adequate for 1,420–1,350 °C, 7–3 GPa (this study; refs. ^21,64). The results are aggregate melts that approximate the evolution of redox melts in the asthenospheric melting column from 7 GPa to an LAB at 3 GPa.

As melt fraction increases continuously from the depth of redox melting to the base of the lithosphere, we linearly combine the two melting reactions to reach 10 wt% and 4 wt% melt, respectively, calculating melt compositions stepwise in 0.01% increments. This approach is justified because melt compositions and melting reactions must evolve continuously with depth. In the absence of more information, a linear combination remains the best approximation.

Clinopyroxene Ca contents are a considerable uncertainty as the conditions for a T_P of 1,350 °C straddle the flat pigeonite–high-Ca clinopyroxene saddle on the pyroxene miscibility gap. We therefore use different CaO contents of 13–16.5 wt% in clinopyroxene for 7–3 GPa (this study; refs. ^21,64), producing a range of model melt compositions diverging in CaO and MgO/CaO (Fig. 1 and Extended Data Fig. 1).

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-025-10065-3.

Author contributions

M.W.S. developed the concept of this paper, completed early experiments and wrote the first draft of the manuscript. N.P. performed most of the experiments and analytics. M.W.S., N.P. and A.G. contributed equally to elaborating and finalizing this study and the paper.

Peer review

Peer review information

Nature thanks the anonymous reviewers for their contribution to the peer review of this work.

Data availability

Data are given in the Supplementary Tables and are available at 10.5281/zenodo.17901938.

Competing interests

The authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s41586-025-10065-3.

Supplementary information

The online version contains supplementary material available at 10.1038/s41586-025-10065-3.