Boninites are an important ‘end-member’ supra-subduction zone magmatic suite as they have the highest H₂O contents and require the most refractory of mantle wedge sources. The pressure–temperature conditions of boninite origins in the mantle wedge are important to understanding subduction zone initiation and subsequent evolution. Reaction experiments at 1·5 GPa (1350–1530°C), 2 GPa (1400–1600°C) and 2·5 GPa (1450–1530°C) between a model primary high-Ca boninite magma composition and a refractory harzburgite under anhydrous and H₂O-undersaturated conditions (2–3 wt % H₂O in the melt) have been completed. The boninite composition was modelled on melt inclusions occurring in the most magnesian olivine phenocrysts in high-Ca boninites from the Northern Tongan forearc and the Upper Pillow Lavas of the Troodos ophiolite. Direct melting experiments on a model refractory lherzolite and a harzburgite composition at 1·5 GPa under anhydrous conditions (1400–1600°C) have also been completed. Experiments establish a P, T ‘melting grid’ for refractory harzburgite at 1·5, 2 and 2·5 GPa and in the presence of 2–3 wt % H₂O. The effect of 2–3 wt % dissolved H₂O produces a liquidus depression in primary boninite of ∼112 ± 19°C at a given temperature. The H₂O-bearing melts, recalculated to 100 wt % anhydrous, are ∼2–6 wt % higher in MgO, ∼1–2 wt % higher in SiO₂ and ∼1–1·5 wt % lower in FeO, compared with nominally anhydrous melts at the same P and T. These differences are consistent with a change in the melting reaction, resulting in a higher contribution of orthopyroxene to the melt phase, compared with anhydrous conditions. We conclude that high-Ca boninite petrogenesis requires temperatures as high as ∼1480°C at depths of ∼45 km in the mantle wedge; these are constraints for any proposed model of intra-oceanic subduction zones. A comparison of the results from the boninite–harzburgite reaction experiments with the direct melting experiments on refractory lherzolite and harzburgite indicates that the influence of subduction components (included in the composition of the added model boninite) is to cause high-pressure melting cotectics to move towards the olivine apex (i.e. to relatively higher pressures) of the molecular normative projection from diopside onto the base of the ‘basalt tetrahedron’ [Jd + CaTs + Lc–Qz–Ol] compared with anhydrous melting of normal mantle in the absence of a subduction component. The subduction component involved in high-Ca boninite petrogenesis in addition to H₂O has relatively high Al₂O₃ and Na₂O contents. The experimental data from this and other studies empirically quantify the absolute effect of dissolved H₂O (0·2–21 wt %) on the liquidus depression of olivine-saturated basaltic melts with ∼<3 wt % total alkalis as follows:

The role of water in the uppermost mantle has been explored to 6 GPa (∼200 km) by a novel experimental approach in which the silicate melting solidus, the stability of hydrous phases and the H₂O contents in nominally anhydrous minerals (NAMs) were determined. The composition studied is a fertile lherzolite modelled as a source for mid-ocean ridge basalts (MORB). The use of crushed olivine as traps for melt or fluid inclusions allows a distinction to be made between quenched hydrous silicate melt and quench material from water-rich vapour phase. The vapor-saturated solidus (water-rich vapor) of fertile lherzolite increases in temperature (T) from a minimum of 970°C at 1·5 GPa (∼50 km) to 1375°C at 6 GPa. The Ca-rich amphibole pargasite is stable to the vapour-saturated solidus to 3 GPa (∼100 km). Based on normative components, at 2·5 GPa the near-solidus melt (1–2%) in mantle with very low H₂O content is transitional between sodic–dolomitic carbonatite and olivine melilitite. With higher melt fraction (∼5%) at higher T or higher H₂O content it is olivine-rich basanite. Both immediately below and above the solidus, the H₂O content in residual lherzolite is ∼200 ppm retained in NAMs at 2·5 and 4 GPa. The experimentally determined vapour-saturated solidus corrects recent numerical models of melting of lherzolite + H₂O based on inferred high solubilities of H₂O in NAMs and accounts for a discrepant experimental determination of the vapour-saturated solidus in which very high water/rock ratios were used. At 2·5 ± 0·1 GPa, the water content of experimental charges was varied from 0·05 to 14·5 wt %. Below the solidus and with increasing water content from 0·05 to 2·9 wt %, pargasite decreases in K₂O and Na₂O content and is absent in experiments with 7·25 and 14·5 wt % H₂O. Also with increasing water content from 0·05 to 14·5 wt % H₂O, the Na₂O content of clinopyroxene decreases from 1·6 wt % to below the limit of detection (0·2 wt %). The destabilization of pargasite and change of clinopyroxene composition at 2·5 GPa and 1000°C are attributed to the leaching role (Na₂O and K₂O particularly) of the water-rich vapour at high water/rock ratios. The hydrous mineral pargasite is the major site of H₂O storage in fertile uppermost mantle lherzolite but pargasite is unstable at pressures (P) >3 GPa (∼100 km depth), causing a sharp drop in the water storage capacity of the upper mantle from >2000 to ∼200 ppm. For small H₂O contents (500 ppm H₂O) overlying the ‘depleted’ MORB source (∼200 ppm H₂O) in the deeper asthenosphere. Water also plays a significant role at convergent margins, where hydrous silicate melting in the mantle wedge is initiated at the vapour-saturated solidus. Melting of lherzolite at or near the vapour-saturated solidus does not fully dehydrate residual lherzolite or harzburgite. Residual lithosphere returned to the upper mantle may carry ∼100–200 ppm H₂O. At 6 GPa the low K/Na model mantle composition (MORB-source mantle) with >200 ppm H₂O has normal rather than supercritical melting behaviour with the solidus at 1375°C, which is ∼350°C below the C + H-free solidus.