N-type MORB sample NAM-7 [18] was used as the eclogite component in our experiments, whereas spinel lherzolite sample KLB-1 was used as the fertile mantle peridotite component [4] . The chemical compositions and representative mineral constituents of each starting material are presented in Table 1. Both rock types were powdered and dried at 1 atmosphere under a CO-H₂ buffer, which is close to an NNO buffer, and were kept in a furnace at 200˚C in order to prevent atmospheric hydration.

Our melting experiments were performed at pressures of 2.9 GPa, 5 GPa, and 8 GPa, temperatures from 1300˚C to 1650˚C, and for durations of 1 - 24 hours (Figure 1 and Table 2). All run conditions and the resultant phases produced are listed in Table 2. Anhydrous solidi for KLB-1 and N-MORB NAM-7 are shown in Figure 1 (after [4] and [18] , respectively), where it can be seen that the latter is situated at temperatures ~200˚C below than that of the former. In order to investigate wall-rock reactions between molten eclogite and ambient mantle peridotite in an ascending plume, most of the experiments in this study were conducted at temperatures below the dry peridotite solidus.

All experiments were carried out at the Magma Factory in the Tokyo Institute of Technology, Japan [19] . Each individual rock powder was pressed into a separate pellet, and both were then placed into graphite/platinum double capsules to avoid Fe-loss. Figure 2 shows the capsule assembly with the N-MORB layer placed on the top of peridotite layer near to the thermocouple.

“tr” represents for trace amount of minerals or melts which cannot be analyzed by EPMA.

Experiments performed at 2.9 GPa were conducted using a Boyd-England-type piston-cylinder apparatus, and were based on a half-inch calibration of talc/Pyrex- glass/graphite assembly [4] with W-5%Re, W-26%Re thermocouples (a cross section of this configuration is shown in Figure 3(a)). In those runs, the two- lithology starting material was kept at 1200˚C and 2.9 GPa for 20 hours before the final experiment in order to convert low-pressure mineral assemblages to subsolidus high-pressure eclogite and garnet peridotite assemblages. Conversely, 5 GPa and 8 GPa experiments were performed using a 1000-ton end-loaded Kawai-type multi-anvil apparatus [19] . We used 18 mm edge-length MgO octahedra as the pressure medium, and FUJILLOY (TN-05) tungsten carbide anvils with truncation edge lengths of 11 mm. LaCrO₃ was employed as the heater, and W-5%Re and W-26%Re were used as thermocouples (Figure 3(a)). The pressure calibration of Aoki and Takahashi [20] was used.

Run products were collected, mounted in epoxy, polished, and carbon-coated for electron microprobe analysis (EPMA). Textures and chemical compositions were analysed with a JEOL-JXA8530F electron microprobe housed at the ELSI, Tokyo Institute of Technology. The acceleration voltage, beam current, and peak measurement time were 15.0 kV, 1 × 10⁻⁸ A and 20 s, respectively. Na was analysed for 10 s in order to avoid count loss. Mineral grains were analysed using a focused electron beam, while a defocused beam (~5 - 30 μm) was used for silicate melt.

All run conditions and the phases produced in each lithology are listed in Table 2. The eclogite layer contained clinopyroxene (Cpx), plagioclase (Pl), garnet (Grt), and coesite (Coe) at subsolidus temperatures, and Cpx, Grt, and silicate melt at suprasolidus temperatures (only a low degree of partial melting was observed). Basaltic melts coexisted with Cpx and Grt in isolated subsolidus peridotite layers. The peridotite failed to melt in most of our experiments, where it consisted

of olivine (Oliv), orthopyroxene (Opx), Cpx, and Grt. An additional Al-rich phase (Cr-rich spinel) was present in 2.9-GPa run products, though this transformed to garnet at higher pressures. Significant melting took place in runs conducted at temperatures above the peridotite solidus―particularly near the basalt melt layer―with Oliv and Opx present as restitic phases in the suprasolidus regime. Melt pockets spread along matrix Opx-Oliv grain boundaries, while Cpx, Grt, and spinel were consumed.

Cpx and Grt produced at subsolidus basalt temperatures were mostly subhedral to euhedral (Figure 4(a), Figure 4(b)), although traces of melt occurred along their grain boundaries at temperatures slightly above the dry eclogite solidus (Figure 4(d), Figure 4(e)). In the peridotite layer, Opx was mostly euhedral (Figure 4(c), Figure 4(f)), olivine occurred along Opx-Cpx-garnet grain boundaries (i.e., within the peridotite matrix), and garnet had a poikilitic texture.

Since partial melt formed in eclogite at pressure?temperature conditions below the peridotite solidus would not have been in chemical equilibrium with Oliv, thin Opx reaction bands formed between both rock types via chemical reaction between silica-rich partial melt of eclogite and Oliv in the peridotite matrix. At subsolidus peridotite conditions, the thickness of the Opx reaction band positively correlated with the run temperature and durations. As partially melted eclogite layers were separated from peridotite by thin Opx bands, chemical reactions between each basalt and peridotite layer were thus hindered and could only proceed via solid-state diffusion. At higher temperatures, substantial melting occurred both in basalt and peridotite layers, and fully molten basalt melts merged with molten pockets in peridotite. Extensive chemical mass transport was able to take place through chemical diffusion and liquid percolation under these conditions, and so basalt melting consistently allowed chemical equilibrium to be established with ambient peridotite, even in short runs.

Representative chemical compositions of individual solid and melt phases are listed in Table 3. Melts produced at all experimental conditions exhibited compositions that plotted in the basalt to basaltic andesite domains on a Harker diagram (Figure 5). Black crosses represent data from Mallik and Dasgupta [17] . Basaltic andesite melts dominated peridotite subsolidus runs, while relatively

Si-rich tholeiitic melts dominated at temperatures near the dry peridotite solidus, the compositions of which were buffered by the aforementioned Opx reaction band along the interface of each rock type.

The relative K₂O and TiO₂ contents of melts formed within each starting material (yellow star in Figure 6) can be used as a proxy for the degree of melting of N-MORB. Lower K₂O and TiO₂ contents mean that the volume of melt increased by melting of peridotite, which essentially lacks K₂O and is very TiO₂- poor (Figure 6(c), Figure 6(d), Figure 6(f)). The presence of an Opx band caused chemical exchanges between both rock types to proceed solely by solid- state diffusion for as long as the peridotite lacked melt.

Because of the different starting materials used for layered experiments, our melt compositions (e.g., Ti, Na, Ca, and Fe) were systematically different from those reported by Mallik and Dasgupta [17] . In our study, all major elements except for Fe showed a close fit with a Hawaiian tholeiite composition. Partial melts of anhydrous run products contained ~55 - 57 wt% SiO₂ (i.e., andesite) at low temperature and ~50 - 53 wt% SiO₂ (i.e., tholeiite) at high temperature; the latter melt compositions being similar to those observed in Hawaii during its shield-building stage (Figure 6(b)). Composition of melts obtained at temperatures close to the dry peridotite solidus closely matched the composition of Hawaiian tholeiite (Figure 6(a)). Notably, N-MORB had almost entirely melted at these conditions. These results and their comparison with Hawaiian tholeiite (Figure (6)) show that the chemical composition of recycled eclogite in the sub-Hawaiian plume is more enriched in FeO than N-MORB is.

A schematic model for melting processes in a heterogeneous Hawaiian plume is shown in Figure 7 (after Takahashi and Nakajima [7] ). We have demonstrated that the entire compositional spectrum of Hawaiian tholeiites (basalt to picrite) can be formed by basalt-peridotite hybrid melting near to the dry peridotite solidus. In other words, our results indicate that the temperature at the centre of the plume head may be close to that of the dry solidus of peridotite KLB-1 (1480˚C at 3 GPa, Takahashi et al. [19] ), or else slightly lower. This implies that the PMT (Potential mantle Temperature) of the Hawaiian plume may be much lower than has been previously estimated (e.g., 1558˚C, Watson and McKenzie, [3] ).

The size of the entrained eclogite component plays an essential role in the melting processes that occur within a heterogeneous plume. If the eclogite components were small (such as mafic dyke or mafic bands, which are common in peridotite massifs), partially molten eclogite layers could achieve equilibrium with the surrounding peridotite over geological timescales relevant to plume melting (i.e., 10⁴ - 10⁶ yr). Conversely, if eclogite blocks were large (e.g., km-scale or larger), melt formed in the eclogite blocks would be isolated from surrounding peridotite due to the formation of Opx bands at their interfaces. The timescale of melt segregation from the partially molten eclogite may be faster

than that for reaction rate across the Opx band; thus, melts derived from eclogite domains would have geochemical signatures that record entrainment and would contaminate the surrounding peridotite less extensively.

Takahashi and Nakajima [7] proposed that magma produced near the axis of the plume head may be a mixture of two types of melts: 1) basaltic andesite melt formed by eclogite melting, and 2) picritic melts formed by the reactive melting of eclogite and peridotite. According to their study of the Koolau volcano, an eclogite block larger than 1000 km³ supplied silica-rich magma at the last stage the volcano’s formation, and remained in disequilibrium with peridotite during its lifetime. Hence layered eclogite-peridotite experiments are necessary in order to study the dynamics of the recycled oceanic crust in the ascending plume.

Tholeiite to basaltic andesite magmas could be generated in an upwelling mantle plume containing a recycled oceanic component (Figure 7) by hybridised melting of basalt and peridotite. High-silica tholeiitic OIB, such as occurs in Hawaii, may be formed under nearly dry conditions in a heterogeneous plume. The relative stability of Opx caused it to form in our experiments via chemical reaction between silicate melt and SiO₂-undersaturated olivine from ambient peridotite. This Opx separated basaltic domains and peridotite domains during melting. Major-element compositions of melt derived from layered N-MORB and spinel lherzolite closely matched those of Hawaiian tholeiite. Very depleted heavy rare earth element (HREE) patterns in Hawaiian tholeiite might be attributed to percolation of partial melt in vertically elongated eclogite blobs in the centre of the plume. Enrichments in Fe, K, Ti, and light rare earth elements (LREE) in the Hawaiian tholeiite compared with modern MORB may indicate that the source of the recycled components in the Hawaiian plume is more enriched in Fe and incompatible elements than modern N-MORB is. A likely source material here is Fe-rich tholeiite, which was common in the Archaean and early Proterozoic [21] . This conclusion is consistent with model ages of Hawaiian magma sources determined by Pb isotopes [21] .