Electrical conductivity during incipient melting in the oceanic low-velocity zone.

The low-viscosity layer in the upper mantle, the asthenosphere, is a requirement for plate tectonics. The seismic low velocities and the high electrical conductivities of the asthenosphere are attributed either to subsolidus, water-related defects in olivine minerals or to a few volume per cent of partial melt, but these two interpretations have two shortcomings. First, the amount of water stored in olivine is not expected to be higher than 50 parts per million owing to partitioning with other mantle phases (including pargasite amphibole at moderate temperatures) and partial melting at high temperatures. Second, elevated melt volume fractions are impeded by the temperatures prevailing in the asthenosphere, which are too low, and by the melt mobility, which is high and can lead to gravitational segregation. Here we determine the electrical conductivity of carbon-dioxide-rich and water-rich melts, typically produced at the onset of mantle melting. Electrical conductivity increases modestly with moderate amounts of water and carbon dioxide, but it increases drastically once the carbon dioxide content exceeds six weight per cent in the melt. Incipient melts, long-expected to prevail in the asthenosphere, can therefore produce high electrical conductivities there. Taking into account variable degrees of depletion of the mantle in water and carbon dioxide, and their effect on the petrology of incipient melting, we calculated conductivity profiles across the asthenosphere for various tectonic plate ages. Several electrical discontinuities are predicted and match geophysical observations in a consistent petrological and geochemical framework. In moderately aged plates (more than five million years old), incipient melts probably trigger both the seismic low velocities and the high electrical conductivities in the upper part of the asthenosphere, whereas in young plates, where seamount volcanism occurs, a higher degree of melting is expected.

The lithosphere is a chemically depleted and mechanically strong region of the uppermost mantle, overlying the chemically enriched and mechanically weak asthenosphere [1-3,8,16]. Volatile enrichments in the asthenosphere have long been shown to trigger incipient melting[10,13-15] (that is small degree of partial melting due to small amounts of CO2 and H2O) in the upper part of the asthenosphere and a link between incipient melting and seismic low velocity zone has also long been suggested[13,15,17]. In this article, we demonstrate that incipient melting of the mantle by the presence of small amounts of CO2 and H2O can also trigger high electrical conductivities. We herein assume that the low viscosity layer, the high electrical conductivity layer (ECL) and low seismic velocity layer (LVZ) are coincident, and use the term Asthenosphere for this layer.

The Asthenosphere is characterized by high electrical conductivities[4,5,18] and reduced S-wave velocities[8,16]. While the characteristics of the Asthenosphere are commonly related to water related defects in olivine[2-4], a number of multidisciplinary observations[5-8,10,13,15,17-20] and the discovery of petit spot volcanoes[21] indicate that the Asthenosphere most likely contains partial melt.

Two issues arise when the observed features of the Asthenosphere are attributed to partial melting: (1) a few volume percentages of basaltic melt are generally required to explain high electrical conductivities, which is problematic as melt would unavoidably tend to rise if present at such high amounts[11-12] and (2) the lithosphere-asthenosphere boundary (LAB) occurs at a near constant depth of 50-75 km for both warm/young and cold/ancient lithospheres[4,5,8,16,18], where the temperature may not be sufficiently high to produce such amounts of melt.

However, incipient CO2-H2O rich melts, that are stable under the P-T-fO2 conditions of the Asthenosphere[10,13-15,17,22-23], allow melting in both warm and cold regions of the Asthenosphere[17]. Low temperature carbonatite melts, composed of almost 50% CO2, are characterized by high electrical conductivities[19], but their stability is restricted to the coldest and driest regions of the Asthenosphere[17]. Increasing temperature or water content changes the composition of the prevailing melts to intermediates between basalts and carbonatites, often described as carbonated basalts[17,22]. Very little is known about the physical properties of such intermediate volatile-rich melts (CO2 and H2O). In particular, their electrical properties have never been measured. In order to address the issues regarding onset of partial melting at the LAB, and to permit a test of the incipient melting model suggested by petrological studies[15,17], we performed electrical conductivity measurements on CO2-H2O-rich melts.

We developed a new experimental set-up, specifically adapted for liquids with high conductivities (Extended Data Fig. 1). The high performance of this modified 4-wire method, adapted to ½ inches piston cylinders, is discussed in the Methods section and Extended data figures 1, 2 and 3a. Five melts, with CO2 and H2O contents ranging from 10 to 48 wt% and 0 to 10 wt%, respectively, were analysed by impedance spectroscopy in the temperature range 900 - 1500°C at a confining pressure of 3 GPa. We tested the reproducibility of the measurements by taking measurements during both cooling and heating of the samples (Extended Data Fig. 3a) and we verified that decarbonation and dehydration of samples at high temperature did not affect the conductivity results. Figure 1 reports the measured electrical conductivities as a function of reciprocal temperature. For similar water contents, the electrical conductivity of carbonated basalts is higher than that of hydrated basalts and the difference increases with an increase in the CO2 content of the melt to a maximum of nearly one log unit. The most CO2-rich melt has conductivities higher than 200 S.m−1. We develop a semi-empirical law that takes into account the two parallel conductive processes operating in carbonated basalts: conduction by covalent polymer-like hydrous silicate melts and ionic conduction by carbonate melts[19].

Figure 1

Electrical conductivity of hydrous carbonated basalts vs. hydrated basalts and hydrous olivine

The conductivities of the hydrous carbonated basalts experimentally measured in this study are by far the highest, reaching up to 200 S/m and being about one and four order of magnitude higher than hydrated basalts[7] and hydrous olivine[25], respectively. The fitting curves are calculated according to our conductivity model for CO2- and H2O-bearing melts (Eq. 1).

The calculated conductivities using equation 1, as shown in figure 1, reproduce our measurements and those of ref. 7 on CO2-free hydrated basalts with an average precision of 5% in σ (see Methods). The effect of CO2 on melt conductivity, predicted by equation 1, is negligible at low CO2 content, but increases sharply for CO2 content higher than 6 wt%. Such a change is most likely caused by an abrupt transition in the melt structure and properties from silicate-type to carbonate-type.

We calculate the mantle electrical conductivity for variable amounts of bulk H2O and CO2 contents in a partially molten peridotite. We assume that the interconnected melt is equally distributed between grain edge tubules and grain boundary melt films[18,20,24] (see Methods). The conductivity of hydrated olivine was calculated from ref. 25 and equation 1 was used for CO2-H2O-bearing melts. We assume that carbon is exclusively soluble in the melt[23] (carbonate units) and computed the partitioning of water between carbonated melt, pargasite, olivine, and peridotite combining ref. 9, 10 and 22. We report results for partially molten peridotite containing only H2O (Fig. 2a) and both CO2 and H2O (Fig. 2b and 2c). In all simulations, partitioning constraints for CO2 and H2O between solids and melts impose that (1) CO2-H2O-rich melts can only be produced at the onset of mantle partial melting and (2) that small melt fractions always contain far more CO2 than H2O (see top axis in Fig. 2). If > 1% melting is attained, the melt volatile contents drop to values that modestly impacts on their electrical conductivity.

Figure 2

The incipient melt effect on the electrical conductivity of depleted and enriched carbonated peridotites

The conductivity of partially molten peridotite (log values increasing from cold to warm colours) is reported as a function of melt content and temperature for a, CO2-free peridotite with 200 ppm H2O and b, c, depleted and enriched CO2-bearing hydrous systems. H2O partitions between minerals and melt, and CO2 distributes in melt only (Methods). Addition of CO2 triggers a peak in conductivity at 0.1-0.3 vol.% of melt, where the intergranular liquid is CO2-rich and therefore highly conductive. At higher degrees of melting, the bulk conductivity decreases since volatiles are diluted in the melt (melt H2O and CO2 are tabulated atop each panel), which becomes basaltic. A peridotite with 0.1 vol.% carbonated basalt is as conductive as with 10 vol.% basalt. Two sets of melt H2O contents are given for the bottom panel (500 ppm H2O and CO2), which correspond to pargasite-saturated (low T<1070°C, italics) and paragsite-undersaturated (T>1070°C, normal) melt water contents.

The CO2-free depleted mantle, containing ca. 200 ppm H2O[26-28], cannot be conductive at temperatures below 1350°C (i.e. σ ≥ 0.1 S.m−1), unless it contains more than ~5 vol.% basaltic melts (Fig. 2a). Only unreasonably high temperatures for the LAB (>1450°C) can make the mantle conductive with small amounts of melt (<1 vol%). Moreover, at high melt percentage, water has almost no effect on mantle conductivity, since its content in the melt remains small (i.e. <1 wt% H2O negligibly affects basalt conductivity). If an enriched mantle is considered (500 ppm H2O), a reasonably low melt content (1 vol.%) can trigger high conductivity but it still requires high temperature (>1325°C; Extended Data Fig. 4), and the enriched mantle is also CO2-rich[26-28].

In presence of CO2, the formation of incipient CO2-rich melts (<0.5 vol.%) disproportionately increases the effective electrical conductivity of the mantle (Fig. 2b and 2c). For example, in the depleted mantle, containing 200 ppm H2O and 200 ppm CO2[26] and fuelling the dominant part of MORBs[27-28], 0.1-0.15 vol.% of melt at 1325°C can explain the high electrical conductivity of the ECL reported in oceanic domains[5,18,29]. The melt is a carbonated basalt, typically containing 15-35 wt% CO2 and about 2-3 wt% H2O (Fig. 2b). Remarkably, the enriched mantle[27-28] with 500 ppm H2O and 500 ppm CO2 can produce high conductivities at temperature and melt fractions as low as 1050°C and 0.2 vol%. We also notice that incipient melting of the enriched mantle triggers conductivities that are 2.5 times greater than the depleted mantle making variations in electrical conductivity a powerful probe of the chemical enrichment in the upper mantle.

The stability of incipient melts in the upper part of the asthenosphere has long been shown by petrological constraints[13-15], that are not considered in figure 2. The P-T region of incipient melting in peridotite is shown in Extended Data Fig. 5 together with the stability domain of pargasite, that is the main solid host for water in peridotite containing more than 150-200 ppm[10]. We calculate that the presence of pargasite restricts the amount of water to 40-50 ppm in olivine, according to partition coefficient among peridotite minerals[9,10]. Pargasite is however unstable at T>1070°C[10] and its occurrence must be merely considered for the enriched mantle (>200 ppm H2O). Based on the P-T phase diagram of Extended Data Fig. 5 and considering oceanic geotherms at 23.5, 35 and 70 Ma, we have calculated 1D conductivity profiles illustrating the impact of several petrological discontinuities (Fig. 3). We have considered the depleted mantle (200 ppm H2O plus CO2 varying from 100 to 500 ppm) and the enriched mantle (500 ppm H2O and 500 ppm CO2). Variable CO2 contents in the depleted mantle account for the fact that MORBs have degassed their CO2 and the carbon content of their source is therefore highly uncertain[26-28].

Figure 3

Petrologically-based conductivity profiles across the incipient melting region under the lithosphere-asthenosphere boundary for various ages

Top axis indicates electrical conductivity and how it varies with depth during cooling of the lithosphere for ages of 23.5, 35 and 70 Ma (see choices of geotherm in Fig. 4). Conductivities were calculated according to the same model used in Fig. 2 (Methods). Several electrical discontinuities are predicted at variable depths based on the phase-equilibria relationships shown in Extended Data Fig. 5-6; the most striking conductivity jumps is related to the upper and lower boundaries of incipient melting (55-150 km). The volatile depleted and enriched mantles are considered and one can appreciate that the conductivity during incipient melting is strongly correlated to CO2 contents (grey dashed lines labelled from 100 to 500 ppm).

The upper discontinuity (Fig. 3) predicted by our model is the beginning of incipient melting at ~50 km depth for young/warm plates and at ~70 km for colder/older plates. This discontinuity marks the thermodynamic boundary between CO2-rich melt and CO2-rich vapour[14]: the melt being stable at greater depth. In the case of an enriched mantle, an additional discontinuity occurs due to the pargasite dehydration melting reaction (producing CO2-H2O rich, low SiO2 melt) that can be shallower than the previously described discontinuity for young plates (ie. 23.5Ma) and deeper for old plates (70 Ma). At 35 Ma, these two discontinuities occur at the same depth (60 km). The lowest discontinuity shown in figure 3 occurs in the depth interval 120-150 km and is described as the region of redox melting[22,23]; that is the boundary separating diamonds from CO2-rich melts, the melt being stable at shallower depth. Incipient melting, which triggers the conductive region of the asthenosphere, is therefore permitted between the redox melting lower boundary and the decarbonation upper boundary and this agrees well with electromagnetic observations in oceanic domains[4,5,18,22,29], though ref. 29 indicates slightly deeper ranges.

The increase in conductivity in the incipient melting region is major, being half a log-unit for the depleted mantle (200 ppm CO2) and more than one log-unit for the mantle containing 500 ppm CO2. High conductivities of 0.1 S/m or more can be reached for CO2 contents as low as 300 ppm in the case of young plates. Note that the surprising effect of water (Fig. 3), where incipient melting in a mantle with 500 ppm H2O and 500 ppm CO2 induces lower conductivities than in a mantle with 200 ppm H2O the same amount of CO2. The imaging of the 23.5 Ma old LAB by ref. 5 at ~50 km depth revealed conductivities of 0.1-0.2 S.m−1. These are definitely not explainable by melting of a CO2-free H2O-depleted mantle, since too high temperatures and/or too high melt contents are demanded (Fig. 2). They cannot be explained by pargasite dehydration melting in a CO2-free H2O-enriched mantle either, as this process cannot produce high enough conductivities (Fig. 3, Extended Data Fig. 4). Once deciphered in a petrological framework, the conductivities of the LAB in ref. 5 can be reached by incipient melting of a mantle containing 400 ppm CO2 (Fig. 3). We recall that ref. 5 introduced moderate electrical anisotropy in the inversion of their magnetotelluric data whereas we merely discuss here the geometric mean conductivity, which is much less model dependent.

The presence of CO2-rich melts in the asthenosphere not only better explains the electrical properties of the Asthenosphere, but also explains the weak dependence of the lithosphere thickness on the age of the oceanic crust (Fig. 4). The bottom of the lithosphere in figure 4 marks a seismic discontinuity characterised by a reduction in S-waves velocity of 5 to 15 %[8,16]. This discontinuity cannot be caused by partial melting of a dry or water under-saturated mantle, as this may only occur at higher depths and temperatures[9] (see blue melting curve on Fig. 4). Previously suggested melting reactions such as the dehydration melting of amphibole[10] also fail to reproduce the depth-age relationships of the LAB (Fig. 4). Remarkably, the CO2+H2O melting curve[14], which delimits the upper boundary of the incipient melting region already shown in figure 3, ranges from 50 km down to 80 km depths from the youngest to the oldest lithospheres (purple curve in Fig. 4). This correlates pretty well with the bottom of the lithosphere as imaged by the seismic discontinuity. The lower limit of incipient melting, i.e. the redox melting[22,23], matches also well the lower part of the seismic low velocity zone[8] at depth of about 140-180 km. At low pressures, above the incipient melting region, the decarbonation of the melt forms an impermeable layer in which buoyant CO2-rich melts are frozen into clinopyroxene-rich residue (with pargasite) and CO2-rich fluids (Fig. 4). Melting is therefore permitted in the Asthenosphere and the melt cannot rise through the LAB because of the existence of this melt-freezing boundary (see Methods). It is only where the mantle is hot enough to suppress the freezing reaction (that is melting does not anymore require CO2) and where melt fractions are large enough[12] (2-5 vol%) that melts can rise through the LAB. This occurs for young plates (<5Ma), where volcanic seamounts are observed, and this has also been related[6] to the electrical properties of the young LAB[4].

Figure 4

The oceanic seismic low velocity zone bracketed by the upper and lower boundary of incipient melting

a, Oceanic crustal ages versus depth of seismic discontinuities (Vs reductions) marking the LAB (grey circles[8]) beneath the Pacific Ocean. Colour curves designate the solidi for hydrated (200 ppm H2O; blue), carbonated (green) and H2O-undersaturated carbonated (purple) peridotites. Isotherms (grey curves) are calculated from a half-space sudden cooling model, assuming[8] ΔT = 1350°C, an average plate velocity of 8 cm•yr−1 and a thermal diffusivity of 1 mm2•s−1. Varying the plate velocity does not change the plot. b, A visual picture capturing the domain of incipient melting in the oceanic low velocity zone. The LVZ is lower-bounded by the redox melting[22,23] and upper-bounded by the decarbonation[14] leading to the freezing of incipient melts. This boundary constitutes an impermeable layer leaving a clinopyroxene-rich residue and a CO2-rich vapour phase.

Incipient melting has long been described as a key petrological process operating in the seismic low velocity region marking the upper asthenosphere[13,15,17]. The mantle geochemistry and petrology in this region argues for production of incipient CO2-enriched melts[9,10,13,15]. We demonstrate that these melts have conductivities of hundreds of S/m, much higher than CO2-free hydrated melts or hydrated minerals. Our modelling, despite unavoidable simplifications, considers geochemical and petrological constraints and indicates that mantle with small fractions of CO2-rich melts at 50-150 km reproduces the electrical properties as well as the depth of the LAB pretty well, whereas CO2-free systems yields too poor or no agreement with geophysical observations. The presence of CO2-rich incipient melts in the Asthenosphere has important implications for radiogenic heat production as such melts are enriched in heat-producing elements like K-U-Th[30]. Moreover, the involvement of CO2-rich incipient melts is also recognized in petrological processes occurring in the continental and in the cratonic LAB[30]. The Asthenosphere - incipient melting association we suggest here can therefore be extended to geodynamic settings other than the oceanic domains. It remains however to be defined how the mechanical strength of the Asthenosphere can be impacted by small amount of CO2-H2O-rich melts and how this can be connected to plate motions.

METHODS

Starting materials

(Extended Data Table 1a). Electrical measurements were performed on five mixtures: 2 dry carbonated melts (CO2=44 -48wt%), 1 hydrous carbonated melt (CO2 =25.9 wt%; H2O=10.2 wt%) and 3 hydrous carbonated basalts (CO2 = 10.39 to 23.32 wt%; H2O = 4.43 to 9.22 wt.%). Starting materials used to obtain these mixtures were natural dolomite (MgCa(CO3)2), a natural basalt (popping rock[31]), salt (NaCl), sodium carbonates (Na2CO3) and brucite (Mg(OH)2).

Experiments

All experiments were performed at 3 GPa in ½-inches piston cylinders (graphite-Pyrex-talc assemblages), which was connected to a 1260 Solartron Impedance/Gain Phase Analyzer for electrical conductivity measurements. The temperature was measured with a B-type thermocouple localized on sample top ((Extended Data Fig. 1a)). Oxygen fugacity (fO2) was not controlled during the measurements but the presence of graphite (furnace) and molten carbonates (sample) should imply an oxygen fugacity close to FMQ-2[23].

We have developed a new protocol specifically adapted for electrical conductivity measurements on highly conductive molten materials (Extended Data Fig. 1). The new design employs a pseudo-4-wire configuration, which removes the electrical contribution of the electrical cell itself (Extended Data Fig. 2a). Such a configuration previously adapted at 1 atm[19,32] is necessary for our measurements at pressure.

Cold pressed pellets (5 mm outer diameter) were cored in their centre in order to place an inner Pt electrode (1 mm). A Pt foil surrounding the sample was used as outer electrode. An alumina jacket isolated the entire electrical cell from the graphite furnace. The sample impedance was measured between the two electrodes arranged in a co-axial geometry[33,34]. The inner electrode was connected to the impedance spectrometer via the two wires of the thermocouple[34]. The outer electrode was connected to a nickel cylinder (located 5 mm above the sample) that was mounted in series with two additional wires (B-type thermocouples) (Extended Data Fig. 1a).

Impedance spectra, conductivity calculations and uncertainties

The impedance spectra were collected during heating and cooling cycles (Extended Data Table 2a) at different temperature plateaus in the frequency range 1 Hz to 1 MHz. Conversely to the spectra collected at low temperatures (i.e. solid samples showing impedance arcs), high temperature spectra (i.e. molten samples) indicated vertical lines (Extended Data Fig. 2b). These spectra correspond to inductance-dominated signals and the intercept of each spectrum with the X-axis yielded the resistance of the sample.

Reproducibility of electrical measurements was validated by performing the measurements during heating and cooling cycles. For the hydrated experiments (HC and HCB-9, -7, -4), a step of about 10 min was operated at 700°C (before brucite dehydration) and temperature was rapidly raised (< 10 sec) to 1300-1410°C (i.e. temperature of molten state), which limited sample dehydration.

Data reductions and uncertainties

The electrical conductivities of the samples were calculated from the measured resistances using the following relationship[33,34]: with σ being the electrical conductivity in S.m−1, r, respectively the outer radius, the inner radius and the height of the samples in m, and R, the resistance of the sample in Ω (Extended Data Fig. 1b and 2b).

Uncertainties in σ were calculated considering geometrical factors of the samples (Extended Data Fig. 1b) and propagated errors of each measured resistance. The uncertainties on σ are 7 % on average for all measurements and reach a maximum of 16% on HCB-4.

Sample characterization

Scanning electron microscope (SEM) imaging and electron microprobe analyses (EMPA) were systematically operated after each experiment. Determination of r and h by SEM imaging showed an average decrease of 20% compared to the initial geometry, most likely due to porosity loss during melting (Extended Data Fig. 1b). No melt leak was observed and the entire sample remained sandwiched between the MgO plugs and the electrodes.

EMPA were conducted at 15 KeV, 10 nA and 10 sec counting on peak elements. The beam size (100 μm × 100 μm) was adapted to obtain average chemical compositions, smoothing the heterogeneities due to quench crystallizations. Compositions before and after experiments indicates no contamination by the MgO surrounding the sample and no considerable volatile loss from the sample (Extended Data Table 1b). CO2 content were determined using the by-difference method[22] and indicate negligible decarbonation.

An elemental analyser, type Flash 2000 (Thermo Scientific), was used to measure H2O content of sample (before and) after experiments. Samples are heated to >1500°C and the released H2O is reduced into elemental H being finally detected by a highly sensitive thermal conductivity detector. This gives water content with a precision of +/− 0.5 wt%. We observed a negligible dehydration during conductivity measurements.

Conductivity Results

Extended Data Fig. 3a shows the good reproducibility of the electrical measurements during heating and cooling cycles. The conductivity-temperature relationships for each sample were fitted using an Arrhenius law: with σ, the conductivity of the sample (S.m−1), σ, the pre-exponential factor (S.m−1), E the activation energy (J.mol−1), R, the gas constant and T, the temperature (K).

Calculated pre-exponential factors and activation energies are presented in Extended Data Table 2b.

Increasing the CO2 concentration in the melt drastically increases its conductivity. Furthermore, we observed that CO2 tends to decrease both activation energies and pre-exponential factors.

Conductivity Modelling

The semi-empirical law that we have developed can be considered as the sum of two conductive processes operating in carbonated basalts: (i) conduction in the hydrous silicate melts by interstitial sodium mobility[7,33-34] and (ii) conduction in CO2-rich melts caused by the motion of all species in ionic liquids[19] (equation 1).

The pre-exponential factor σ0 and the activation energy Ea for both H2O and CO2 terms are related by a compensation law[35,36] (Extended Data Fig. 3b): the decrease of activation energy as a function of volatile content is exponential: C is the CO2 and H2O content in wt%.

Thus, the set of Eqs 1, 4 and 5 directly relate melt conductivities to melt H2O and CO2 contents. We have determined the Arrhenius parameters for the melt composition as a function of H2O contents using data in ref. 7 in the temperature range 1200-1500°C. a, b, c, d and e parameters for H2O were obtained by fitting these data with Eqs 4 and 5 (Extended Data Table 3). The best parameters for CO2 were obtained by minimizing the differences between our measured conductivities and the CO2-free values returned by equation 1 (Extended Data Table 3). In doing so, we assumed that the effect of water on electrical conductivity is similar and modest in both silicate and carbonate melts, which is indeed what our measurements show (see Extended Data Fig. 3a, samples C vs. HC). Our model reproduces the experimental measurements on σ, in S/m, (ours and that of ref. 7) within an average error of 5% (max 10%).

Figures 2 & 3

The bulk rock is considered as a peridotite containing a fraction of interconnected melt, where volatiles partition between the solid and the melt phase. The bulk H2O content is related to H2O in melt and in peridotite as: where is the mass fraction of melt and is the partition coefficient of H2O between peridotite and melt (0.007, i.e. average partition coefficient over 1.5-4 GPa[9,37]).

The concentration of H2O in olivine is: differs from that of peridotite since is about 0.002 according to ref. 9. (see also figure 6 of ref.9 for modal proportion of mineral phases in peridotite Ol58Opx28Cpx12Spn2).

Pargasite amphibole was considered to affect the distribution of water for bulk water content exceeding 200 ppm (enriched mantle in Fig. 2c, Fig. 3 and Extended Data Fig. 4). We computed that the water content exceeding 200 ppm bulk goes in pargasite and computed the partitioning of water among NAMs following ref. 9. If temperatures exceed ca. 1070°C (see Extended Data Fig. 5), then pargasite dehydration melting occurs and water partitions between melt and the solids as described above.

CO2 distributes exclusively in the liquid phase[23,38], i.e. . Therefore:

At small melt fractions, this can lead to CO2 concentrations higher than that of carbonate (>~45 wt. %) or, in other words, to CO2 saturation; Calculations performed in saturated-ranges are mentioned in Fig. 2b and 2c.

Conversion from mass to volume fraction of melt is done considering volume properties of silicate melts[39] and carbonate melts[40,41]: where d is the density of peridotite, i.e. 3.3, and d is the density of melt. As there are no data for the density of hydrous carbonated basalts, we estimate the density of these melts using a simple mixing law: where concentrations are expressed as wt. % d = 1.4 (considering 12 cc/mol for H2O partial volume in melts). The mass percent of carbonate (M-CO3, where M denotes cations such as Ca, Mg and others) in melt is approximated as . Carbonate density d was set to 2.4 according to the density of molten carbonates at 1 bar[40]. The use of CO2 partial molar volume calculated in ref. 41 yields similar density results. The density of d was taken as 2.8 (ref. 39).

The conductivity of the melt was calculated using equation 1. The conductivity of the peridotite was assumed to be controlled by that of hydrous olivine[25].

The bulk conductivity was calculated using the mean of tube[42-44]: and film[44-47]:

Geometries resulting in values almost similar to refs 20 and 46. Remarkably, our system is rather insensitive to melt geometry since the difference between both laws is mostly lower than 0.2 log units. We therefore averaged the two laws:

The bulk conductivity of partially molten peridotite reported in Fig. 2 and 3 as well as in the text was calculated using this last equation (13).

Melt fraction in Fig. 3 was approximated as (Extended Data Fig. 6).

Buoyant basalts versus incipient melts

An impermeable layer has been suggested to prevent the melt prevailing in the LAB from rising to the surface[49]. The rate of melt ascent due to buoyancy is otherwise expected to be of the order of several cm/year[11,12], if melt content is 3-5 vol. %. Our model of incipient melting implies an impermeable boundary that is caused by phase relationships[14], i.e. a thermodynamic boundary through which melt cannot rise. We furthermore emphasise the limited melt mobility[50] at the small melt fraction of interest as, in particular, surface tensions would unavoidably tend to retain the buoyant melt[51]. To conclude, if basalts, being anyway not thermodynamically stable in the asthenosphere, tend to migrate out of the asthenosphere[52], small melt fractions may in contrast be mechanically stable in the LAB.

a, Modified piston cylinder assembly for electrical conductivity measurements using a 4-wires configuration. The cored sample (in green) contains in its centre an inner electrode in platinum (in blue). A platinum foil (in blue) surrounds the sample, which extends upwards and downwards from the sample and corresponds to the outer electrode. The sample is sandwiched by machined MgO ceramics (in white). The electrode-sample assemblage is isolated from the graphite furnace by an Al2O3 jacket (in yellow). The 4-electrode wires are emplaced using a 4-hole Al2O3 tube (in orange). Two of these wires, i.e. the thermocouple, are in contact with the inner electrode, whereas the outer electrode is in contact with two other wires by means of a top Ni plug (in red). b, SEM image of the assemblage of sample C after experiments (up to 1463°C and 3 GPa). We observed an average decrease of 20% compared to the initial cell geometry (corresponding to the porosity loss during melting). Cell geometry parameters (h and r in equation SI2) are determined from SEM images for each sample.

a, The electrical cell resistance versus temperature. We show the resistance of a sample made of nickel measured using either a 2-wire setup (empty diamond) or a 4-wire setup (red diamond). There are several orders of magnitude of difference between the two measurements showing that the 2-wire setup is not suitable at all for conductive materials. We also show the resistance of carbonate in 4 wires setup (it is the sample C, molten at T>1230°C, green triangle). b, Impedance spectra obtained on molten carbonate (sample C) at 3 GPa as a function of temperature. Impedance spectra show vertical lines indicating an inductance-dominated signal for all temperatures. The resistance is taken from the intercept with the horizontal axis. Data are obtained at frequencies ranging from 19905 to 315479 Hz. The black line represents an impedance spectrum of a nickel sample (blank) with a 4-wire configuration obtained at 1464°C.

a, Electrical conductivity vs. reciprocal temperature measured on carbonated melts and hydrous carbonated basalts. Samples: a carbonated melt (C), a hydrous carbonated melt (HC) and 3 hydrous carbonated basalts with water contents ranging from 4.43 to 9.22 wt% (HCB-9, HCB-7 and HCB-4) and CO2 contents ranging from 10.39 to 23.32 wt%. In order to complete figure 1, we distinguished heating-cooling temperature cycles and reported error bars. Solid symbols = heating cycle (H1); open symbols = cooling cycle (C1); small solid symbols = second heating cycle (H2) (cf. Extended Data Table 2a). The error bars include uncertainties on the geometrical factors of the samples and on the measured resistance. b, Compensation plots showing the correlation between activation energy, Ea, and pre-exponetial terms, ln σ Hydrous basalts (HB) are from the experimental dataset of ref. 7 between 1200 and 1500°C and the data point for the dry basalt (B) is from ref. 32. The dry carbonated melt (C), the hydrous carbonated melts (HC) and the hydrous carbonated basalts (HCB) are from this study (see Extended Data Table 2b for the Arrhenius parameters).

The incipient melt effect on the electrical conductivity of an H This figure completes the scenarios illustrated in Fig. 2. The conductivity of partially molten peridotite, in which H2O partitions between minerals and melt, (Methods), is reported as a function of melt content and temperature for CO2-free peridotite with 500 ppm H2O (log values; conductivity increases from cold to warm colours). The discontinuity at T=1070°C is due to pargasite amphibole breakdown (Extended Data Fig. 5) that redistribute H2O between NAMs and the melt as explained in the Methods. Melt H2O contents (blue if pargasite out, green if pargasite in) are tabulated atop the panel.

Melting curves for different bulk peridotitic systems as a function of temperature and depth. The solidus of dry peridotite (black curve) is calculated from ref. 53. The dehydration solidus of nominally anhydrous peridotite at 200 ppm H2O (blue curve) is modelled from ref. 9. The dehydration solidus of pargasite lherzolite is after ref. 10. Nominally anhydrous carbonated, fertile peridotite solidus is after ref. 54 and references therein (green curve). The H2O-undersaturated carbonated, fertile peridotite curve (purple curve) corresponds to the solidus of a pyrolite with 0.5-2.5 wt.% CO2 and 0.3 wt.% H2O (ref. 14).

For pressures ≤ 1.7 GPa, carbonated melts are unstable and gaseous CO2 prevails. We connected the melting curve of CO2-bearing peridotite to that of the dry peridotite at low pressures, which slightly differs from previously published phase diagrams. We considered that, for P ≤ 1.7 GPa, gaseous CO2 must have a negligible influence on the peridotite solidus due to the small solubility of CO2 in basaltic melts[55]. Similarly, at low pressures, the H2O-undersaturated carbonated, fertile peridotite solidus was connected to the dehydration solidus of nominally anhydrous peridotite (considering peridotite with 200 ppm water), neglecting the presence of pargasite owing to the NAM’s water capacity storage.

Phase equilibria control on H We show changes in water content in olivine (left), melt fraction (centre) and melt CO2/H2O (right) for the 70 Ma age used for calculation in Fig. 3. We illustrate that for two compositions: bulk with 200 ppm H2O and 500 ppm CO2 and bulk with 500 ppm H2O and 500 ppm CO2.

a Bulk chemical composition of the starting materials (wt%). b Analysed chemical compositions (in wt%) of the bulk systems after each experimental runs. H2O contents were analysed using an elemental analyser (Flash 2000). CO2 contents were obtained by difference on EMPA.

a Heating and cooling cycles. Cycles of the different experimental runs with corresponding temperature ranges (in °C). Runs HC, HCB-9, -7 and -4 were heated up from room temperature to 700°C. A step of 10 min was operated at 700°C before a rapid increase to the temperature of measurement. b Arrhenius parameters and their errors determined for each melt studied.

Parameters used for .