## Introduction

Hydrous ringwoodite with about 1.5 wt.% H2O1 and phase egg2,3,4 has been discovered in the diamond inclusions of mantle xenoliths. Ringwoodite is a high-pressure polymorph of olivine and a significant mineral in the mantle transition zone. Phase egg is one of several hydrous aluminosilicate minerals which is stable under the conditions found in the mantle transition zone2,3. Identifying these hydrous minerals of the mantle transition zone would be natural evidence supporting the hypothesis that the mantle transition zone is a water reservoir.

As wadsleyite and ringwoodite are major constituent minerals of the mantle transition zone, their water capacities have been experimentally investigated. The water solubilities of wadsleyite and ringwoodite have been estimated to be about 1–3 wt.%5,6. As a basis of comparison, the water solubilities of major constituent minerals of the mantle, olivine and bridgmanite, are about 0.1 wt.%6 and 0.2 wt.%7, respectively.

The water-rich mantle transition zone plays an important role as a trigger of mantle melting. Owing to the large difference in water solubilities between olivine and wadsleyite, dehydration melting can occur at the 410 km discontinuity at the beginning of the transition zone8. Geophysical observations support the existence of melt at the base of the upper mantle based on low seismic wave velocity anomalies9,10 and high electrical conductivity10. In addition, the gravitational stability of hydrous magma was experimentally confirmed by Sakamaki et al.11. However, water-induced mantle melting can occur not only above the mantle transition zone but also below the transition zone. Significantly, a low-velocity anomaly has been detected at the top of the lower mantle12,13,14.

Melting experiments on mantle materials have been widely performed15,16,17. The composition of partial melt becomes magnesium (Mg)-rich and silicon (Si)-poor with increasing pressure15. In the case of melting under the lower mantle condition, the liquidus phase is bridgmanite, and the composition of the coexisting melt is richer in iron (Fe) and Calcium (Ca) than that in bridgmanite16,18,19,20,21. Kawamoto conducted experiments with a water-saturated KLB-1 peridotite melt at 14–24 GPa and 900 °C–1400 °C22. A clear difference was observed in crystal-melt partitioning between the dry and hydrous systems. The melt exhibited a Ca-rich and Si-, Mg- and aluminium (Al)-poor composition as compared with the anhydrous system. A problem with this experiment was the valence of iron in the starting materials. Although samples with divalent iron (Fe2+) have generally been used in previous studies, trivalent iron (Fe3+) is more dominant at the top of the lower mantle23,24. This is because the mineral structure of bridgmanite is such that it easily accepts an Fe3+ and at least 60–80% of iron cations are Fe3+24,25. Melting experiments of this type should be performed using the more dominant trivalent iron.

Herein, we conducted melting experiments with a hydrous peridotite composition with Fe3+ to more accurately reproduce the melting phenomenon that occurs at the boundary between the mantle transition zone and the lower mantle.

## Results

### Chemical compositions of partial melts and obtained crystals

Hydrous peridotite melting experiments were conducted at 1300 °C–1600 °C at 23.5 and 26 GPa. All recovered samples contained bridgmanite, CaSiO3-perovskite and ferropericlase with quench microcrystals from the melt. This implies that all experiments were performed between solidus and liquidus (Table 1 and Supplementary Fig. S1). A cross section of representative experimental products (26 GPa and 1400 °C for 120 min) is shown in Fig. 1. The longer the duration of the experiment, the larger the crystals grew. We evaluate the chemical equilibrium of the sample based on experiments with different retention times (60 min and 120 min) at the same experimental condition (P = 26 GPa and T = 1400 °C). Since the compositional difference fell within the standard error range, we regarded melting experiments as equilibrium (Supplementary Tables S1.5 and S1.6). The difference in crystal distribution can be clearly seen in Fig. 1a. Most crystals were bridgmanite. In Fig. 1b (=high-temperature side), microcrystals around larger bridgmanite crystals can be observed. On the lower temperature side, ferropericlase and CaSiO3-perovskite have crystallised (Fig. 1c).

Chemical mapping images of the recovered sample are shown in Fig. 2 and Supplementary Fig. S2. The brightest area in the Si mapping is bridgmanite; Mg is ferropericlase; Ca is CaSiO3-perovskite; Fe is bridgmanite and the melt; and Al is bridgmanite. In the Fe-mapping, the dark region is consistent with an existence of CaSiO3-perovskite. The ratios of each component are plotted in Fig. 3. The SiO2 content of the melt is less than half that of bridgmanite and CaSiO3-perovskite (Fig. 3a), and the MgO and FeO contents of bridgmanite and the melt are nearly identical (Fig. 3b). The melt contains a higher CaO concentration than the bridgmanite (Fig. 3c), and the bridgmanite exhibits the highest Al2O3 content (Fig. 3d) among all samples. The composition of bridgmanite was 53.5 ± 0.8 wt.% SiO2, 5.2 ± 0.1 wt.% Al2O3, 9.0 ± 0.4 wt.% FeO* and 35.5 ± 0.6 wt.% MgO. Total cations were calculated as Fe3+ except ferropericlase (Supplementary Table S1). The cation ratio of recovered sample was estimated based on stoichiometry of analyzed chemical composition using energy dispersive X-ray spectrometry (EDS). Based on the estimation, bridgmanite was mostly fitted with Fe3+, and ferropericlase was accommodated to Fe2+. This result implies that the bridgmanite26 contains little Fe2+, and ferropericlase comprises mainly Fe2+. The averaged composition of the partial melt was 15.8 ± 3.5 wt.% SiO2, 1.6 ± 0.3 wt.% Al2O3, 8.0 ± 0.5 wt.% FeO*, 34.8 ± 1.6 wt.% MgO and 9.8 ± 1.6 wt.% CaO. These values represent the average composition of each measurement of melt shown in Supplementary Table S1 and the uncertainties are calculated from the standard deviation. Comparison with anhydrous experiments, the partial melt had a CaO-rich and SiO2- and Al2O3-poor composition. Notably, the weight content of SiO2 was only one half that of MgO. The atomic ratio of Mg/Si for this study is 4.3, whereas that of the anhydrous melt is 1.018; Kawamoto22 calculated the atomic ratio of Mg/Si of the hydrous melt as only 1.9 in a previous hydrous study (Table 2). Such high-MgO and low-SiO2-content melt can influence density and compressibility under deep mantle conditions.

### Comparing the chemical composition of crystal and the partial melt

When comparing the composition of bridgmanite with that of melts, no difference was observed in MgO content (Fig. 3; Supplementary Table S1). This trend is similar to the results of previous experiments performed with anhydrous systems (Table 2). The important point is that less difference in iron content between melt and bridgmanite (Femelt/FeBrg = 0.9). Compared to anhydrous system (Femelt/FeBrg = 2–3)18,19, preferred concentration of Fe into melt is not observed. In the case of melting experiment of Fe2+-bearing sample22, the Fe content ratio of bridgmanite to melt is also about 0.9 (see Table 2). The Fe-Mg exchange partition $${K}_{D}^{(Fe/Mg)}$$ between bridgmanite and melt of anhydrous peridotite is about 0.418,19, while that of this study (hydrous system) is 1.1.

### Density of the melt

To discuss the gravitational stability of the melt in the lower mantle, the melt density was calculated using Eq. (1) proposed by Wakabayashi & Funamori27.

$$P=\frac{3}{2}{K}_{0,{T}_{0}}({(\frac{{V}_{0,{T}_{0}}}{{V}_{p,T}})}^{\frac{7}{3}}-{(\frac{{V}_{0,{T}_{0}}}{{V}_{P,T}})}^{\frac{5}{3}})+{\alpha }_{0,{T}_{0}}{K}_{0,{T}_{0}}(T-{T}_{0})$$
(1)

In Eq. (1), P, K, V, α and T represent pressure, bulk modulus, volume, the thermal expansion coefficient and temperature, respectively. Because water concentrates in the melt of run products, considering the effect of H2O on the density of silicate melts is necessary. The partial molar volume of H2O in magma at high pressure and temperature was calculated using the Eq. (2):

$$P=3{K}_{T}[1-{(\frac{{{\nabla }}_{{H}_{2}O}}{{{\nabla }}_{{H}_{2}O,0}})}^{\frac{1}{3}}]{(\frac{{{\nabla }}_{{H}_{2}O}}{{{\nabla }}_{{H}_{2}O,0}})}^{\frac{2}{3}}\exp \{\frac{3}{2}(K^{\prime} -1)[1-{(\frac{{{\nabla }}_{{H}_{2}O}}{{{\nabla }}_{{H}_{2}O,0}})}^{\frac{1}{3}}]\}\,$$
(2)

where $${{\nabla }}_{{H}_{2}O}$$ is the high-pressure partial molar volume of H2O, $${{\nabla }}_{{H}_{2}O,0}$$ is the zero-pressure partial molar volume as described by Bouhifd et al.28 and KT is the isothermal bulk modulus as described by Sakamaki29.

Considering the values in Supplementary Table 1, the total value of the melt is low (64.0–78.3 wt.%). Owing to the feature of EDS, measurement of light elements, such as H, is impossible; moreover, if H is concentrated in the sample, the average of the total value (wt.%) is estimated to be low. It is not conclusive what leads to the low total value because other influences such as vacancy must also be considered, but coexisting solid phases are anhydrous minerals; hence, we estimated that the remaining weight ratio of the melt was concentrated water (see Supplementary Fig. S4 for detail). We calculated the chemical compositions of hydrous magma based on the average composition of obtained results and the water content of 29.9 ± 4.1 wt.%. The compression curve of the obtained melt is calculated by the extrapolation of the Eqs (1) and (2). Figure 4a shows two calculated patterns of hydrous melt (29.9 ± 4.1 wt.% H2O) and dry melt (0 wt.% H2O). Comparing with the preliminary reference earth model (PREM)30, both melt densities of this study become lighter than the lower mantle. This means that the melt in this study would be gravitationally stable around the top of the lower mantle regardless of the melt’s water content.

## Discussions

The mantle transition zone can act as a water reservoir/source, and a hydrous condition can be created at the top of the lower mantle. The solidus of peridotite rocks under anhydrous conditions is much higher than that under the mantle geotherm16,18,19,21. However, the presence of water lowers the melting point of the rock, and as a result, the rock begins to produce a partial melt. This MgO-rich and SiO2-poor melt becomes lighter owing to the higher concentrations of water and lower concentrations of iron. Thus, the generated melt at the top of the lower mantle can be separated from the surrounding mantle (Fig. 4b) and be observed as a low velocity anomaly12,13,14. This melt is caused by a downward flow such as a slab13,14 and may react with the mantle transition zone. Even if this melt is absorbed by the mantle transition zone, the melt can be continuously generated at the top of lower mantle as long as the down-welling mantle continues.

The residual mantle rocks separate from the melt due to the large density difference. The downward mantle rock is poor in melt component, that is, the lower mantle composition below the melt can be MgO-poor and SiO2-rich, which is consistent with silicate perovskitic composition. The silicate perovskitic lower mantle has been proposed based on the comparison of experimentally determined sound velocity of lower mantle minerals with seismological observation31. The melt at the top of lower mantle plays as a filter and causes a chemical contrast between upper and lower mantles. Our results can provide a possible mechanism for formation of the silicate perovskitic lower mantle.

## Methods

### Starting material

Reagent powders (SiO2, Al2O3, Fe2O3, MgO and CaCO3) were mixed and heated for decarbonation of CaCO3 at 1000 °C for 16 h. Then, brucite (Mg(OH)2) was added to the decarbonated powder. The water content was 6.98 wt.% (Table 3), and #Mg = Mg/(Mg + Fe) = 0.907.

### High pressure and temperature experiment

High pressure and temperature experiments were conducted using a Kawai-type multi-anvil press (Hymag 1,000-tonne press) at Bayerisches Geoinstitut (BGI) in Germany. We used a 32-mm second-stage anvil (ha7%Co/hawedia) and a 7/3 cell assembly. The sample was enclosed in a gold capsule. Supplementary Fig. S4 shows a detailed design of the cell assembly.

Pressure and temperature conditions were 23.5 and 26 GPa and 1300, 1400, 1500 and 1600 (or 1590) °C, respectively (Table 1 and Supplementary Fig. S1). In the experiment where the duration was doubled (60 min became 120 min) at 26 GPa and 1400 °C, crystal growth was observed; however, no significant difference was observed in chemical composition. After maintaining the temperature for 15 to 120 min, the samples were recovered via quenching.

### Analysis of the recovered samples

Recovered samples were polished using sand papers and abrasives ranging from #80 to #3000, and then finished with 3-μm and 1-μm diamond paste. Field emission scanning electron microscope (FE-SEM; JSM-7001F/JEOL) and EDS (SN: 60154/Oxford Instruments) instruments were used to analyse the samples. The acceleration voltage was 15.0 kV, and the emission current was 77–80 μA.

We estimated the Fe3+/Fe2+ ratio of minerals in the recovered sample based on stoichiometry of analyzed chemical composition using EDS. Based on the estimation, bridgmanite is Fe3+-rich and ferropericlase is Fe2+-rich. However, we could not determine the Fe3+/Fe2+ ratio of the melt. The EDS measurements of melt composition have been performed under the same analysis conditions. The melt composition shown in the Supplementary Table S1 takes the average of ten times measurements.

### Calculations of melt density

The reference temperature T0 was 2500 K, and the temperature of the required density T was set to 1873 K. Consequently, calculations yielded that $${K}_{0,{T}_{0}}$$, $${V}_{0,{T}_{0}}$$ and $${\alpha }_{0,{T}_{0}}$$ are 64.4 GPa, 17.0 cm3/mol and 15.4 × 10−5/K, respectively. In the case of dry peridotitc melt, these parameters were calculated to be 54 GPa, 18.0 cm3/mol and 11.5 × 10−5/K, respectively27.

In Eq. (1), it is assumed that this is in an anhydrous system, and the Fe valence is perfectly Fe2+ (=FeO). However, we used Fe2O3 in this study.