Petrology and Geochemistry of Boninites from the North Termination of the Tonga Trench : Constraints on the Generation Conditions of Primary High-Ca Boninite Magmas

We report here a detailed mineralogical, geochemical, and experimental study of a high-Ca boninite suite from the northern termination of the Tonga trench. Most samples are strongly olivine porphyritic and show a significant range of phenocryst compositions including a very refractory olivine-spinel assemblage Fo94-CrN = 87. They are also characterized by a wide range of incompatible-element contents, e.g., (La/Yb)y varies from 05 to 16, whereas compatible major-element concentrations (A12O3, FeO, CaO, SiO2, and MgO) remain essentially the same. Primary melt compositions for the suite were established on the basis of an experimental study of melt inclusions in phenocrysts and numerical modelling of the reverse of fractional crystallization. Tongan primary melts are characterized by high MgO contents (22-24 wt.%) and originated in the mantle wedge at pressures of 20-25 kbar and temperatures of 1450-1550 °C. H2O contents in primary melts were estimated from direct measurements of melt inclusions by ion probe, and range from 2-0 to 10 wt.%, and a strong correlation exists between H2O and other incompatible element contents. The primary melts crystallized in the presence of an H2O-rich fluid in the temperature range 1390-1150 °C and pressures of 1-7-O15 kbar. Continuous degassing of melts took place during crystallization. Trace-element concentrations in primary melts were estimated using protonand ion-probe analyses of melt inclusions in olivine, and whole-rock analyses. Our data suggest that three independent components (D, El, and E2) were involved. Component D was a refractory mantle depleted in incompatible elements, likely to be hot 'dry' lherzolite produced by previous melting within a mantle plume. Component El was an H2O-rich fluid containing LILE and Th, and had an H2O/K2O value of 20; it was probably produced by dehydration of the subducted slab. Component E2 is thought to have been an incompatible-element-enriched silicate melt of plume origin. Formation of high-Ca boninites requires interaction of hot 'dry' residual mantle, associated with plumes, with a subduction-related H2O-bearing component.

It is generally agreed that boninites are the result of melting of depleted peridotite in a mantle wedge in the presence of a hydrous fluid enriched in incompatible elements (Meijer, 1980;Hickey& Frey, 1982;Cameron etal., 1983).However, compositions of primary melts and P,T conditions of their origin are not well known.Consequently, several important aspects of boninite genesis remain controversial, including the nature and composition of components involved in boninite petrogenesis, and the scenarios of component interactions in boninitic sources.
The source of compatible mantle-derived elements such as Ni, Mg, and Cr in boninites is believed to be highly refractory mantle peridotites, which are also responsible for high Si, low Al, and variable Ca in these rocks (Crawford et al, 1989).The geodynamic environment in which this mantle source was formed is critical for understanding both boninite genesis and island-arc evolution in general.Following Meijer (1980), the dominant opinion is that this mantle source is essentially residual peridotite formed after extraction of N-MORB (midocean ridge basalt) type magmas in a mid-ocean ridge environment.An alternative point of view was formulated by Sharaskin et al. (1983a) and Falloon & Crawford (1991), who showed that isotope geochemistry of the most depleted boninites from the northernmost Tonga trench indicate an ocean island basalt (OIB) type nature for their mantle reservoir.If true, this interpretation could more closely constrain the origin of boninites and imply a close connection with mantle plumes.However, Stern et al. (1991) argued that Nd-Sm, Rb-Sr, and Pb-U isotopic data for boninites provide information only on the enriched components, even for the most depleted varieties, and thus do not provide any information on the nature of their mantle-derived component.
The nature of enriched component(s) in boninite sources is generally related to subduction processes.The following enriched components have been proposed to take part in boninite genesis: fluids from the subducted slab (Meijer, 1980), silicate melt, either from the slab or from the mantle wedge (Hickey & Frey, 1982;Pearce et al., 1992), and ephemeral carbonatitic melt (Falloon & Crawford, 1991).However, the exact composition and the number of components are not well known and differ between localities (Crawford et al., 1989).
Two contrasting models can be put forward to describe boninite petrogenesis.The first argues for boninite generation by reheating of cold, metasomatized H 2 O-bearing mantle (Crawford et al., 1989).The second suggests melting of relatively hot, depleted dry mantle during metasomatism by H 2 O-bearing fluids (Meijer, 1980).These models have different consequences for the modelling of boninite genesis.The latter requires a very hot mantle wedge but no extra heating for melting.The former involves normal sub-arc mantle, but needs a mechanism of extra heating, such as subduction of a spreading centre (Crawford et al, 1989).
In this paper, we present new experimental, mineralogical, and geochemical data for high-Ca boninites from the north Tonga trench and use these, together with other published data (Sharaskin et al, 1983b;Falloon el al, 1989;Falloon & Crawford, 1991) to determine the composition of primary melts (including H 2 O content) and the P,T conditions of their formation.We show that the primary magmas of these high-Ca boninites, with >21 wt.% MgO, require very high temperatures (~ 1500 °C), segregated around 20-25 kbar, and have relatively low H 2 O content (~2 wt.%) compared with 5-7 wt.% H 2 O suggested from high-pressure experiments on boninite petrogenesis (Umino & Kushiro, 1989;Van der Laan et al, 1989).Our data suggest that petrogenesis of these high-Ca boninites requires mixing of an H 2 O-bearing slab-derived component with hot, dry OIB-type mantle.
Minerals and glasses were analysed at the Vernadsky Institute with a CAMEBAX MicroBeam electron microprobe and at the University of Tasmania with a CAMECA SX/50 microprobe, using 15 kV accelerating voltage and 20 nA beam current.Counting time was 10 s.The beam diameter was 1-2 fim for minerals and 10 //m for glasses.Clinopyroxenes USNM 122142, plagioclase USNM 115900, basalt glass USNM 111240/52 (Jarosewich et al., 1980), olivine CH-1, and spinel UV-126 (Lavrentev et al., 1974) were used as standards for major elements.
Trace elements in melt inclusions were analysed by an IMS-3F ion microprobe at Massachusetts Institute of Technology, Cambridge, and by proton microprobe at the Commonwealth Scientific and Industrial Research Organisation, Sydney, Australia.Technique details have been given by Shimizu & Hart (1982) and Sieet al. (1990).Accuracy of both methods is considered to be 10-20 rel.%.
The molecular composition of fluid inclusions in olivine was obtained using Raman spectroscopy at high temperatures (Sobolev et al., 1983) at CREGU, Nancy, France.
H 2 O contents were measured in quenched glasses and melt inclusions by Fourier transform infra-red spectroscopy (Danyushevsky et al., 1993) and an ion microprobe technique (Sobolev & Chaussidon, in prep.).The accuracy of both methods is estimated to be better than 10 rel.%.

PETROGRAPHY AND MINERALOGY
The petrography of the Tongan boninites was presented in detail elsewhere (Sharaskin et al., 19836;Falloon et al, 1987;Visotsky, 1989).Most samples studied in this paper are strongly porphyritic and contain up to 50% phenocrysts, which are euhedral prismatic olivine crystals typically 1-10 mm long.Small (up to 3 mm long) phenocrysts of ortho-and clinopyroxene are also present.Groundmass consists of glass with variable amounts of olivine and pyroxene microphenocrysts and microlites.Spinels are present as inclusions in all phenocryst phases, but rarely occur as discrete crystals.Sample 28-1 is a volcanic glass with rare microphenocrysts of orthopyroxene, plagioclase, and magnetite.Rare microlites consist of the same minerals.There is no petrographic evidence for secondary alteration in the boninite samples studied.

Olivine
Olivine crystals in all samples are unzoned and very magnesian, up to Fo 93 for the Western group and up to Fo 94 for the Eastern group (Fig. 2, Table 1).A coherent CaO vs. Fo trend (Fig. 2), which is more Ca rich than that of mantle olivines [CaO < 0-1 wt.%, Simkin,& Smith (1980)], and the presence of melt inclusions suggest that olivines in these rocks, including the most magnesian crystals, crystallized from the boninite magma and are not xenocrysts of mantle origin.Histograms of Fo content (Fig. 2) show a unimodal distribution for the Western group boninites, and a bimodal distribution for the Eastern group.The range of Fo contents of the Tongan olivines is similar to that of boninites in general, and they are among the most Ca-rich olivines so far analysed from boninites.
by relatively lower Cr^ [Cr/(Cr + Al)], and higher Mg^ and TiO 2 contents than the Eastern group spinels (Fig. 3, Table 1).Spinels with Fe 2 + /Fe 3 + > 3 are absent in the Western group.
In both groups the large range of TiO 2 contents in spinels in each sample cannot have been produced during simple fractional crystallization.Tongan high-Ca boninite spinel populations do not contain the Cr-rich spinels (Cr w >88) present in low-Ca boninites.However, they cover the total range of boninitic spinel TiO 2 contents and Fe 2+ /Fe 3 + values.

Compositions of crystalline inclusions in phenocrysts from the Western group Tongan boninites
inclusion walls and can be seen only during heating at temperatures >200°C by the movement of mineral particles.The main component of the fluid is H 2 O. CO 2 is present in trace amounts (~ 0-1 mol%).Other components, notably CO and CH 4 were not detected.In quenched glasses, fluid inclusions at room temperature also consist mainly of a gas phase, but a liquid phase is visible.Sobolev & Danyushevsky (1986) demonstrated the magmatic nature of these inclusions.The presence of primary H 2 O-rich fluid inclusions in all phases of Tongan boninites argues for H 2 O saturation of the melt during the entire course of its crystallization history.

Rocks and glasses
The compositions of glasses and rocks (Tables 2 and 3) are plotted in Fig. 4 together with available data for whole-rock, groundmass, and glass compositions from the Tongan boninites.Significant differences exist in FeO*, K 2 O, and P 2 O 5 contents between the Western and Eastern group boninites (Fig. 4), the Western group being characterized by generally lower FeO* and higher K 2 O and P 2 O 5 .Broad variations within each group in the contents of the most incompatible elements (TiO 2 , P 2 O 5 , and K 2 O) at constant MgO content cannot be explained by fractionation, and appear to reflect the variable composition of primary melts.However, the tight trends of SiO 2 , A1 2 O 3 , CaO and FeO* vs. MgO indicate that major-element compositions of these primary melts were probably fairly similar.

Melt inclusions
To determine the melt composition at various stages of its crystallization history, together with temperature of crystallization, a heating stage study of melt inclusions in olivines and orthopyroxenes from the Western group sample 26-2 was carried out (see Appendix A for the details of technique).Sixty runs were done with inclusions in olivine, and five with inclusions in orthopyroxene.Homogenization of melt inclusions was established in 15 runs with olivine, and in every run with orthopyroxene.The range of homogenization temperatures of inclusions in olivines is 1230-1330 °C; in orthopyroxenes it is 1150-1185 °C.
The correlation between the composition of a mineral and its crystallization temperature (correlation coefficient r = 0-70), a feature demanded by fractional crystallization, is shown in Fig. 5b.Temperatures obtained for orthopyroxene are in agreement with those calculated using two-pyroxene equilibria [Lindsley (1983), Fig. 5a].For reasons given in Appendix A, experiments with the lowest homogenization temperatures at a fixed mineral composition were deemed to be most useful and correct.Compositions of quenched inclusions homogenized during these runs are listed in Table 4.
Compositions of homogenized melt inclusions from sample 26-2 fall within the field of rocks and natural glasses of the Western group (Fig. 4).The only observed difference is systematically lower FeO* concentrations in the melt inclusions in olivine (Fig. 4).tf D (Fe 2+ -Mg) values between host olivine and melt inclusion of 0-36-0-39 (Table 4) are significantly higher than those measured experimentally for dry and H 2 O-bearing systems [0-3±003 (Roeder & Emslie, 1970;Ulmer, 1989)].The high Mg* values and low FeO* contents of these inclusions together argue for the re-equilibration of Fe 2+ and Mg between the inclusions and the host olivines after trapping (Gurenko et al., 1991).To achieve the initial equilibrium between melt inclusion and host olivine, the redistribution of Fe into inclusions and Mginto host olivinescan be calculated using equations of Ford etal. (1983).The calculated compositions are listed in Table 4, and FeO contents of the recalculated melts correspond to the field of rocks and glasses (Fig. 4).
Results obtained for sample 26-2 demonstrate the evolution of Western group melts during crystallization within the temperature range 1300-1150°C, which corresponds to ~ 30% of fractionation (Table 4).(1987,1989), and Falloon & Crawford (1991).•, Western group rocks; A, Eastern group rocks; +, homogenized melt inclusions; O, compositions of homogenized melt inclusions corrected for Fe-Mg exchange with host mineral (see text for discussion); A, compositions of naturally quenched melt inclusions in ohvine Fo, 4 from the Eastern group, used by Falloon & Green (1986) for calculation of primary melt composition.

H 2 O content
The data (Tables 2 and 4) show that H 2 O contents in naturally quenched melt inclusions are significantly higher than those in quenched volcanic glasses.The former are trapped primitive melts, affected by olivine crystallization on the walls (15-30%, Table 4), whereas the latter are the result of > 30% of fractional crystallization of these melts (see the above paragraph).The data show the degassed nature of Tongan glasses and relatively undegassed H 2 O contents of inclusions, and argue for a significant loss of H 2 O from the melt during crystallization.This is also supported by the H 2 O content of homogenized melt inclusions, which represent primitive melts without significant crystallization, but have H 2 O contents equal to or even higher than evolved glasses.Magma degassing is directly supported by the presence of H 2 O-rich fluid inclusions in olivines and glasses.

Trace elements
Trace-element concentrations are listed in Tables 3 and 5.A comparison of chondritenormalized REE patterns between whole rocks and melt inclusions for samples 26-1 and 26-2 (Fig. 6) demonstrates a general similarity of pattern shapes for inclusions and host rocks.This suggests that, at least for the studied samples, inclusions are representative of melt composition not only in terms of major elements, but trace elements as well.The observed difference in the absolute concentration of REE in melt inclusions and in the whole-rock for sample 26-2 is due to the phenocryst-rich nature of the rock.Ratios between large ion lithophile elements (LILE) Rb, Ba, K, and Sr, and REE, are reasonably close for rocks and melt inclusions in olivines for the same samples.This indicates that whole-rock LILE concentrations are not affected by post-magmatic processes.
The normalized REE patterns of the rocks and the melt inclusions of the Tongan boninites are plotted in Fig. 6.Whereas the Western group boninites are characterized by V-shaped or enriched REE patterns with very high La/Sm values [(La/Sm) JV = 3-4], the Eastern group rocks have slightly enriched or depleted patterns.The absolute R.EE concentrations vary significantly, and mimic the behaviour of Ti, K, and P. Other aspects of trace-element geochemistry are discussed in the following sections.

PRIMARY MELTS
We consider primary melt as a melt which has not undergone any compositional changes since the moment of its last equilibrium with the mantle source.Parental melt is the leastevolved melt which can be established from a study of the products of crystallization (rocks, quenched glasses, minerals, etc.).

Parental melt composition
True fractional crystallization of high-Ca boninites implies that the melt which was in equilibrium with the most magnesian olivine is the parental melt for the entire series.The composition calculated from composition of correspond ing homogenized melt inclusions in olivine by Fe-Mg exchange (see text for details), 11-homogenized mdt inclusion, major-ckroenl compositions of inclusions 13-15 are from Falloon & Green (1986); No -unnque inclusion number, T^a*-homogenization temperature.Host-type of the host mineral.FeO*-total iron as FeO H2O measured using ion microprobe; FeO and Fc2Oj calculated using spinel compositions in the same hotf rainerab after Maurd & Maurcl (1982a); Mg>B-composition of the host-mineral; ATDforFe-Mg partitioning between olivine and mdt; TfQ]-calculated temperature of'dry' I atmliquidm of olivine using equations of Ford et at.(1983 V aisuming oil Fe as FeO; 01, Opx, and Cpx-percent of mineral amounts which have crystallized from the moil primitive melt (O-32) to the given composition, obtained by mmng calculations using a tauusquarcs method (Wright & Dohcrty, 1970), these are calculated for homogenized induwoni only; Diff.-calculation error, i e .theium of squares of residuaU.%OI-the calculated amount of olivine (wt %) crystallized from trapped mdl on the walb of inclusion i estimated by model Ling of the reverse of olivioe crystallization up to equilibrium with the host olivine, this was calculated for un heated inclusions only; ohvioe-rndt equihbria during calculations were calculated after Ford et at (1983).parental melt compositions were calculated from whole-rock compositions using the method of Sobolev et al. (1993) for olivine-phyric rocks.The essence of this method is (1) calculation of groundmass composition on the basis of mass balance: groundmass + average olivine = rock (assuming equilibrium of the groundmass with the least magnesian olivine phenocryst), and (2) calculation of the parental composition by modelling the reverse of olivine fractional crystallization from the groundmass composition up to the moment of equilibrium with the most magnesian olivine found.Olivine-liquid equilibria were calculated using the model of Ford et al. (1983).The Fe 2+ /Fe 3 + value in melts was calculated using correlations between composition of spinel inclusions and host olivines (Fig. 2a) and the model of Maurel & Maurel (1982a) for Fe 2+ /Fe 3+ distribution in spinel and equilibrium melt (see Table 6 for equations).Additionally, this calculation for sample 5-25 was performed directly from the groundmass composition reported by Falloon et al. (1989), and for sample 26-2 from 10:  compositions of homogenized melt inclusions in magnesian olivine.All these calculations assume that Fe-Mg partitioning between olivine and melt is independent of the presence of H 2 O at the level of a few weight per cent, as was demonstrated by Ulmer (1989).The results obtained are shown in Table 6.Temperatures obtained via these calculations (T iry , Table 6) correspond to equilibrium of these melts with olivine for 'dry' conditions.Each group is characterized by a relatively narrow range of variation of major elements (SiO 2 , A1 2 O 3 , FeO*, MgO, and CaO).Standard deviations for these elements are apparently within total errors of the technique used.In contrast, relatively incompatible elements (TiO 2 , Na 2 O, K 2 O, and P 2 O 5 ) show a very large range, reflecting differences in primary melts.
The H 2 O contents of parental melts were estimated using ion-probe data on the melt inclusions in high-Fo olivines from both groups of Tongan boninites (Table 4).The measured H 2 O contents of these inclusions were corrected for olivine crystallization by a least-squares method (Wright & Doherty, 1970) using major-element compositions (SiO 2 , A1 2 O 3 , FeO*, MgO, and CaO) of the average parental magmas (Table 6).The results from both unheated inclusions and inclusions quenched after high-temperature experiments are similar for the Western group.The absolute values of H 2 O content in the Eastern group lavas are significantly higher than those for the Western group, in striking contrast to the behaviour of other incompatible elements.

Primary melt compositions
A requirement of any established parental melt to be primary is that it has a high Mg,y (>70), to be in equilibrium with mantle olivine (Green, 1970).However, the wide range of mantle olivine compositions (Fo 94 _ 87 ) precludes the use of this criterion in quantitative calculations.Arai (1987Arai ( , 1990) ) has shown that the olivine-spinel liquidus assemblage of mantle-derived primitive melts allows the correlation between compositions of coexisting olivines and spinels in the mantle to be used as a suitable criterion of primary magma candidacy (Fig. 7).The existence of such a correlation was shown theoretically [and experimentally by Jaques & Green (1980)] to result from mantle depletion owing to progressive partial melting.
The most primitive olivine-spinel associations observed in boninite and boninite-like magmas fall within the mantle array (Fig. 7), indicating possible equilibrium of parental melts with the mantle.To apply this criterion, however, it is necessary to know, and take into account, the dependence of mineral compositions in the liquidus assemblage on temperature and pressure, because the generation and crystallization of a melt are characterized by different P, T conditions (Appendix B).Available data show a small increase of olivine Mgŵ ith decreasing temperature (~05 Mg^ per 100°C for T> 1300°C) and its independence from pressure, and an increase of spinel CT N with decreasing pressure (10-15% in the range 15-1 kbar in the boninitic compositional range) and its independence of temperature.Using these estimations, it is possible to predict the change in composition of the primary liquidus assemblage with changing pressure if the magma P-T path is known.The latter can easily be obtained from numerous data on the olivine-saturated liquidus slope for high-Mg magmas (~5°C/kbar, Ford et al., 1983).These calculations show that the slope of the shift in the Fo-Cr N plot for the liquidus assemblage of derived parental melts is close to that of the mantle array (Fig. 7).Thus, despite a significant potential difference between temperature and pressure of melt segregation and crystallization, liquidus assemblages of parental melts can be used as a criterion of mantle origin.
The most primitive liquidus assemblage for the Eastern group plots at the magnesian border of the mantle array, implying that observed olivine Fo 94 can be regarded as liquidus   (Arai, 1990).Cape Vogel, Papua New Guinea, data from Walker & Cameron (1983), Victoria data from Crawford (1980), New Caledonia data are our unpublished data; Troodos UPL data from Sobolev el al. (1993).•, estimated shift in a liquidus assemblage composition (Fo 94 , Cr x = 85) during 15 kbar ascent to the surface.(See text for discussion.)Horizontal dashed line indicates approximate border between Iherzolites and harzburgites (Arai, 1987).
olivine for the primary melt for this suite and, correspondingly, that the established parental melt is identical to the primary melt.The most primitive liquidus assemblage of boninites from New Caledonia, Cape Vogel, and Victoria, and of Troodos Upper Pillow Lavas (UPL) also plot at the magnesian border of the mantle array, suggesting that it is a general feature of boninites.In contrast, the most primitive liquidus assemblage of the Tongan Western group has significantly less magnesian olivine for the corresponding spinel.This implies that more magnesian liquidus olivines (up to Fo 94 ) are likely to have been present in these suites, but are not recorded in studied samples.Thus we assume that the primary melt of the Western group boninites was in equilibrium with olivine Fo 94 .The primary melt composition was calculated from the parental melt by modelling of reverse olivine fractional crystallization (Table 6).

H 2 O contents of primary melts
The established degassing of Tongan boninites during crystallization raises the question of whether the H 2 O content of Tongan primary melts can be derived from that of parental melts, or whether some H 2 O was lost before crystallization began.As H 2 O content was measured directly in melt inclusions in Fo 94 olivines from Eastern group lavas, the potential H 2 O loss by degassing is limited, as H 2 O decreases crystallization temperature.Therefore, an olivine-saturated melt should crystallize immediately after H 2 O loss.However, H 2 O loss  (Sobolev el al., 1993).Fields in C: 1, back-arc basin glasses; 2, MORB glasses; 3, arc tholeiites glasses; solid field, Hawaiian glasses.All data are from Danyushevsky el al. (1993) and reference therein.
before crystallization is still possible if the melt was superheated during ascent.Estimation of this effect is not quantitative at present, owing to lack of experimental data for ultramafic melts.Therefore, an independent criterion must be used to prove whether data obtained for parental melts are valid for primary melts.Correlation between H 2 O and other incompatible element contents could be used as such a criterion.Figure 8 presents data on H 2 O in melt inclusions in high-Fo olivines and glasses in the form of incompatible element ratios, to avoid effects of olivine and pyroxene fractionation.A significant difference between glasses and melt inclusions is obvious.Whereas glasses have nearly constant H 2 O/A1 2 O 3 values independent of variations in other incompatible element ratios, inclusions show a strong correlation of these parameters.Such systematic H 2 O behaviour in melt inclusions indicates that they are relatively undegassed.Thus, H 2 O contents of these inclusions may be taken as an estimation of H 2 O contents in primary melts (Table 6).

Trace-element contents of primary melts
Chemical features of Tongan boninites suggest that a set of primary melts with distinctly different incompatible element contents but close major-element composition was involved in the formation of each suite.Therefore, we have calculated incompatible element contents in the primary melt from each individual whole-rock or melt-inclusion analysis.The procedure used involved addition/subtraction of a certain amount of olivine as determined by a least-squares mixing calculation method (Wright & Doherty, 1970).Contents of SiO 2 , A1 2 O 3 , FeO*, MgO, and CaO only were used in calculations.K D values for all incompatible elements between olivine and melt were assumed equal to zero.Results are listed in Table 7.

P,T conditions of primary-melt formation and crystallization
The demonstrated saturation of primitive melts by an H 2 O-rich fluid allows calculation of crystallization pressure, which is estimated from melt H 2 O content using the Burnham (1979) model.Results for melt inclusions in high-Fo olivines are 04-0-9 kbar for the Western group lavas and 0-9-1-7 kbar for the Eastern group.These estimates correspond to the beginning of crystallization.
Crystallization temperature for the Western group primary melts can be extrapolated from the experimentally derived correlation between crystallization temperature and olivine composition (Fig. 5b).Only Fo-r pairs for runs with minimum temperature (see Table 4) were used to establish the Fo-rregression: T(°C) = (23-84 x Fo)-903, which gives 1340°C for Fo 94 .This value is significantly lower than calculated 'dry' temperatures for this composition (Table 6), consistent with the presence of H 2 O in these melts.The crystallization temperature can be also estimated from the difference between the measured and calculated 'dry' temperatures for homogenized melt inclusions in magnesian olivines (Danyushevsky et al., 1992).The maximum difference determined, 95 °C, is the most reliable because of the demonstrated degassing of Tongan boninites during crystallization and possible H 2 O loss during experiments in some runs (Appendix A).The result of such an estimation is 1360 °C, and 1340-1360 °C was accepted as the crystallization temperature for primary melts of the Western group (Table 6).Crystallization temperature of the Eastern group primary melt (1390°C) was estimated from the 'dry' temperature and a correction (95°C) for H 2 O content, obtained for the Western group melts (Table 6).This estimate is the maximum value, because H 2 O content in the Eastern group melts is higher than in the Western group (Table 6).
Pressure of the last equilibrium of primary melts with their mantle sources was estimated using an approach originally proposed for 'dry' melting conditions by Jaques & Green (1980) and Takahashi & Kushiro (1983) and developed by Falloon & Green (1987, 1988) and Sobolev et al. (1992) (Fig. 9).The shift of isobars in this projection owing to the presence of 2 wt.%H 2 O in the melt was estimated using experimental data of Green (1973Green ( , 1976) ) and Kushiro (1989Kushiro ( , 1990)).Results are shown in Table 6.Temperatures of primary melts segregation (Table 6) were calculated from established crystallization temperatures corrected for their pressure of formation, assuming an oli vine-saturated liquidus P-T slope of ~5°C/kbar (Ford et al., 1983).

MgO content ofhigh-Ca boninite primary melts
One of the principal results of this study is a demonstration of very high MgO contents (21-24 wt.% MgO) of primary melts of Tongan high-Ca boninites.Similar high MgO contents have been also proposed for primary magmas of Troodos UPL (Sobolev et al., 1993), which are another example of a high-Ca boninite suite, and for Cape Vogel (Walker & Cameron, 1983) and Victorian (Crawford, 1980) low-Ca boninites.However, our results are somewhat more magnesian than those estimated by Falloon & Green (1986) for primary melts of Tongan boninites (~ 17 wt.%MgO).The reason for this discrepancy for Tongan boninites is that Falloon & Green (1986) calculated the primary melt from the composition of naturally quenched melt inclusions in olivine Fo 94 , which very likely have undergone a significant loss of iron after trapping.As we have shown, iron loss in inclusions is a widespread process for Tongan boninites and is explained by re-equilibration of inclusions with host olivine after trapping.As Falloon & Green (1986) . Dashed lines-ol + opx±cpx cotectics for dry mantle melting after Sobolev et al. (1992); solid lines-approximate position of these cotectics for 2 wt.%H 2 O in melt (see text for references).M PY-composition of MORB pyrolite less 40% olivine (Falloon & Green, 1987); TQ-composition of Tinaquillo lherzolite less 40% olivine (Falloon & Green, 1988).
composition from a modified melt inclusion with an unrealistically low FeO/MgO value (Fig. 4), and calculated this up to equilibrium with olivine Fo 94 , their primary magma has a significantly lower MgO content than our calculated 21-24 wt.%.

H 2 O content of primary high-Ca boninite melts
The H 2 O contents in the primary Tongan melts (from 11 to 2-0 wt.%) are among the first direct estimates for primary boninite melts.They are close to similar estimates for Troodos UPL primary melts (Sobolev et al., 1993) and are in the range of those proposed for Western Pacific boninites on the basis of geochemical studies (Walker & Cameron, 1983;Dobson & O'Neil, 1987).Particularly significant is the demonstrated undegassed nature of these H 2 O contents.Our estimates are significantly lower than those proposed for various primary boninite melts on the basis of recent experimental studies (Umino & Kushiro, 1989;Van der Laan et al., 1989, and references therein).The reason for this overestimation of H 2 O content in the latter studies is that relatively evolved samples were used as starting compositions for experimental runs.Underestimation of the normative olivine content of the starting compositions leads to overestimation of H 2 O pressure necessary to produce olivine on their liquidus (Kushiro, 1972).
Our data suggest significantly H 2 O-undersaturated conditions for generation of primary high-Ca boninite melts in the mantle.However, it is emphasized that these values are among the highest estimated for various mantle-derived magmas (Fig. 8C).

Temperature of primary high-Ca boniriite melts and nature of their mantle source
Both the high MgO and relatively low H 2 O contents established for primary high-Ca boninite magmas suggest exceptionally high temperatures for their origin.These values (1400-1550 °C) are close to similar estimates for Troodos UPL primary melts (Sobolev et al., 1993) and far exceed those proposed previously for boninites [up to 135O°C, Crawford et al. (1989)].These are critical for distinguishing between alternative scenarios of boninite generation, the nature of the high-Ca boninite mantle source, and the nature of interaction between the H 2 O-bearing component and mantle source.
Postulated temperatures of primary high-Ca boninite melt formation strongly suggest that their mantle source is too hot to be a MORB mantle reservoir (Falloon & Green, 1987,1988;McKenzie & Bickle, 1988;Sobolev & Dmitriev, 1989).Such temperatures occur in mantle with normal oceanic geotherms at the depths corresponding to the lower mantle [deeper than 700 km, e.g.Richter & McKenzie (1981)], and, if no major heating of diapirs occurs at shallow levels (unlikely, in our opinion), the high-Ca boninite source mantle must have originated from these depths.The temperature of the high-Ca boninite source is close to that of the OIB mantle reservoir, for which the same deep origin has been proposed (Wyllie, 1988;McKenzie & O'Nions, 1991).Refractory peridotite residual after extraction of plume-related melts could be the source of high-Ca boninites.In this framework, our data strongly suggest that primary high-Ca boninite melts are produced by melting an initially dry and hot source, triggered by introduction of some H 2 O-bearing component(s).

Number and composition of components involved in high-Ca boninite petrogenesis
A number of established features of primary high-Ca boninite melts, including their location on an Ol-Q-Pl projection (Fig. 9) (which argues for a source more refractory than that of MORB), the composition of their primary liquidus assemblage (olivine Fo 94 , spinel Cr N = S7), and their low Ti contents, imply that the composition of their mantle source was significantly depleted.High CaO/Al 2 O 3 values of primary melts (0-81-0-83) argue for clinopyroxene in the mantle source (Falloon et al., 1989).Falloon & Crawford (1991) have inferred that the mantle source of these melts was a depleted lherzolite, consistent with data obtained here.The mantle residue after extraction of these melts is harzburgite.Formation of these melts via melting of a depleted lherzolite is shown in Fig. 9, where all estimations of primary melts lie above an olivine control line that passes through Tinaquillo lherzolite [see Falloon & Green (1987) for detailed discussion].
Despite this refractory source, the presence of strongly incompatible-element-enriched primary melts among high-Ca boninites, and their high H 2 O contents, indicate that in addition to depleted lherzolite (referred to herein as component D), incompatible-elementenriched component(s) took part in their petrogenesis.Increasing absolute contents of the most incompatible elements (we have used La) lead to increases of La/K, La/Rb, La/Ba, La/Th, Ce/Sr, Sm/Hf, and Eu/Ti values for both groups of Tongan boninites (Fig. 10).These trends may be explained by mixing between two end-members.The first end-member (component El), characterized by very low LREE contents, has low LREE/LILE values and is best represented by the composition of sample 5-24.The correlation between calculated La and K/Ti values in primary melts together with the strong inverse correlation between H 2 O/A1 2 O 3 and K/Ti (Fig. 8) argue that this component has a high H 2 O content relative to the second end-member, as all primary Tongan boninite melts have similar A1 2 O 3 contents.The second end-member (component E2) is characterized by high LREE, high La/Sm and La/Yb, and strong relative depletion in Zr, Hf, and Ti.It is best represented by sample 7-18.
Thus, the geochemical features of the Tongan high-Ca boninites can be explained by the interaction of an incompatible-element-depleted mantle source with two independent incompatible-element-enriched components.Geochemical features of the mantle source itself are probably reflected by middle to heavy REE contents in most LREE-depleted  Component E2 was most likely an incompatible-element-enriched silicate melt.The presence of this component is a diagnostic feature of Tongan boninites, and those with a significant contribution of this component have lower H 2 O contents than others, suggesting that it contains a relatively low H 2 O content.This suggests, in turn, that it may not be related to subduction processes.Falloon & Crawford (1991) suggested, on the basis of isotope geochemistry and incompatible-element ratios, that this component could be an OIB-like melt, which is generally consistent with our data.Involvement of an incompatible-elementenriched silicate melt in Troodos UPL petrogenesis has been also demonstrated by Rogers et al. (1989) and Sobolev et al. (1993).
Our geodynamic model for the origin and interaction of components D, E1, and E2 will be published elsewhere (Danyushevsky et al., in prep.).
(2) H 2 O contents in the primary melts vary from 2-0 to 1-3 wt.% for the Eastern group, and from 1-4 to 1-0 wt.% for the Western group.
(3) The primary melts crystallized in the presence of an H 2 O-rich fluid in the temperature range 1390-1150 °C and pressures 1-7-015 kbar for the Eastern group and 0-9-0-3 kbar for the Western group.
(4) The data suggest that three independent components are involved in Tongan high-Ca boninite petrogenesis.Component D was a refractory mantle depleted in incompatible elements, likely to be hot 'dry' lherzolite produced by previous melting with a mantle plume.Component El was an H 2 O-rich fluid containing LILE and Th, which has an H 2 O/K 2 O value of 20 and was produced by dehydration of a subducted slab.Component E2 is thought to be an incompatible-element-enriched silicate melt of plume origin.
(5) Primary high-Ca boninite primary melts form as a result of interaction between hot dry mantle and H 2 O-bearing subduction-related fluid.

Experimental study of melt inclusions
Theoretical aspects of melt and fluid inclusions study have been discussed by Roedder (1984).At room temperature, primary melt inclusions consist of several phases (i.e., quench glass, gas bubble, crystals), which precipitate during cooling owing to crystallization inside inclusions and thermal skrinkage.(Note that the host-mineral always crystallizes on the inclusion walls rather than as discrete crystals inside the inclusion volume, so inclusions at room temperature always consist at least of two phases-quenched glass and crystals on the walls.)The central point of melt inclusions practice is the homogenization of an inclusion.This is the moment of reverse transformation into a homogeneous melt which is fixed by the dissolution of the last phase inside an inclusion, which is normally a shrinkage gas bubble (Roedder, 1984).In the ideal case, fora gas-saturated system, the inclusion composition at this moment is equal to the composition of the trapped melt and the temperature of homogenization matches the temperature of trapping.In real cases, several problems arise as a result of kinetics of melting and behaviour of H 2 O inside inclusions.

Kinetic effects on the temperature of homogenization
Kinetic problems (in our case, the relationship between rate of heating and rate of melting of crystalline phases within the inclusion volume and from inclusion walls) are well known in the literature.For melt inclusions, the dependence of homogenization temperature on heating velocity was shown by Bakumenko (1975).An increase of heating velocity in the high-temperature range (> 1100 °C) causes an exponential increase in the homogenization temperature, whereas its decrease causes an asymptotic decrease of the homogenization temperature to a constant value.This value is accepted as the true temperature of homogenization.The appropriate velocity in each case depends on the melt composition, the size of the inclusion, the temperature of the experiment, and the type of the host mineral.For ultramafic to basaltic compositions of the melt, the most suitable range of velocities is from 1 °C/s to 1 °C/min (Danyushevsky et a/., 1992).The dependence of suitable velocity on the inclusion size is shown in Fig. 5D for a MORB olivine from the Atlantic.The size increase of the inclusion leads to the necessity of decreasing the heating velocity, so that 40-/im and larger MORB inclusions in olivine should be heated significantly slower than 1 °C/min to obtain the true homogenization temperature.To choose the appropriate velocity in the high-temperature range it is necessary to carry out similar kinetic experiments in each case.
The above pertains to cases where exchange processes between an inclusion and its environment during the run are not significant.However, such cases are basically those where disssociation of H 2 O inside inclusions is insignificant and the method of homogenization then allows establishment of the mineral crystallization temperatures with an accuracy of 10-20 °C.Determination of the composition of the quenched homogenized melt inclusion then allows determination of the composition of the melt with an accuracy of 5 rel.% for most major elements (e.g., Sobolev et al., 1991).

The influence of H 2 O dissociation
As was shown by Sobolev et at. (1983), in some cases a continuous increase of the homogenization temperature occurs at high temperatures (> 1100°C) during the run.This process is accompanied by precipitation of magnetite inside such inclusions and is explained by H 2 loss reactions such as or H 2 Om + 2FeOm = H 2 + Fe 2 O 3 m (at temperatures higher than magnetite stability), which proceed owing to the strong diffusive ability of H 2 and the difference in its fugacity inside and outside the inclusion during the experiments.This difference occurs because runs are carried out at 1 atm external pressure of inert gas, whereas the pressure inside the inclusions is, in general, close to the pressure of trapping near homogenization temperature.The reaction is limited by the ability of the melt to oxidize.The main compound of natural melts which can be oxidized is Fe 2 + .It is a simple calculation to determine the maximum amount of H 2 O which may dissociate: H 2 O (wt.%) = 0-125 x C (Fe0) , where C (FeO) = wt.%FeO in the melt.For basaltic melts, the maximum amount of'lost' H 2 O is < 1 wt.%.In acid melts, with FeO content <2 wt.%, possible 'lost' H 2 O contents are <0-25 wt.% and so the effect must be minimal.Experimental data of Clocchiatti (1975) confirm this suggestion.
Two different effects will occur owing to H 2 O dissociation: a volume reduction of melt as a result of disappearance of the component with large molar volume, and the known influence of H 2 O on the temperature of melting.Both of these factors will change the homogenization temperature.Calculations using data of Nelson & Carmichael (1979) for molar volumes of major elements in melts and data of Burnham & Davis (1974) for molar volume of H 2 O in albite melt show that, at low pressures (< 2 kbar) and at high temperatures (> 1100 °C), the first factor is the most significant.This effect could have a dramatic influence on the homogenization temperature.As will be shown below, the homogenization temperature of small inclusions may increase by 150°C during the first few minutes of the run.
If pressure inside an inclusion is high during an experiment, this effect lessens, owing to the volume reduction of H 2 O in melt (Burnham & Davis, 1974), and the second factor is the main one.This effect is pressure independent and comparatively small.For inclusions in olivine and clinopyroxene it does not exceed 40°C (if 1 wt.%H 2 O dissociates).This means that under high-pressure conditions the homogenization temperature should not increase strongly.This suggestion is confirmed by experimental results of a study of melt inclusions trapped by minerals from alkaline Hawaiian suites at 5 kbar (Nikogosyan, 1990).During these experiments, the homogenization temperature increases only slightly, in spite of the H 2 O dissociation which is confirmed by the appearance of magnetite.
In summary, the measurement of crystallization temperatures using the method of homogenization is relatively difficult for H 2 O-bearing melts which crystallized at low pressures.Dissociation of H 2 O can increase the homogenization temperature by more than 100 °C.However, the experimental study ofhawaiite magmas from Etna volcano (Sobolev eta/., 1991) which contain ~1 wt.% H 2 O, shows that the use of the optimal experimental conditions, based on the kinetics experiments, makes it possible to establish the temperature of crystallization and the melt compositions with an accuracy close to that obtained for 'dry' melts.Independent criteria to test measured temperatures are necessary in each case.

Experiments with melt inclusions during this study
The experimental study of the melt temperatures in olivine and orthopyroxene from sample 26-2 was carried out using a high-7" heating stage in an atmosphere of high-purity He under visual control.The heating stage has been described by Sobolev et al. (1980).Primary melt inclusions of homogeneous trapping were used for the experiments.During the runs, a rapid increase of the temperature of homogenization at high temperatures (>1100°C), and magnetite crystallization in inclusions in orthopyroxenes were noted.Suitable experimental conditions involving appropriate heating velocity, run duration, and size of the inclusions were determined on the basis of 15 kinetics experiments summarized in Fig. 5C.Runs with olivine as host mineral were carried out under the following conditions: size of inclusions 100-150 /im; duration of runs at high temperatures < 1 min, heating velocity near the homogenization temperature 3O-4O°C/min.During runs with inclusions in orthopyroxene, these conditions were 50-70 /im, <3-4 min and 10-20 "C/min, respectively.

Olivine
Treatments of experimental data on equilibrium between olivine and dry melt at low pressures and temperatures below 1300 °C (e.g., Roedder & Emslie, 1970;Ford et al, 1983) suggest that K D of Fe-Mg partitioning is independent of temperature.The data at higher pressures, and therefore at higher temperatures, show increasing K D values, which are usually interpreted as the effect of pressure (Ford et al., 1983;Takahashi & Kushiro, 1983;Ulmer, 1989), but the possibility of the hidden effect of temperature has also been mentioned (Takahashi & Kushiro, 1983).To clarify the K D behaviour, a statistical treatment of available data at high temperatures (> 1300 °C) was carried out.The following criteria were used for data selection: (1) to avoid problems of quenching, only data from 'sandwich' and liquidus experiments were used; (2) to be sure that the main form of Fe is FeO, only experiments in graphite or iron capsules were chosen; (3) experiments were anhydrous.Some 88 selected pairs (Grover et al., 1980;Stolper, 1980;Takahashi & Kushiro, 1983;Falloon & Green, 1987, 1988;Falloon et al., 1988)  The result is the opposite of that obtained by Ulmer (1989), who argued for K D independence of temperature.However, his data at high temperatures were obtained in the presence of unmeasured amounts of H 2 O. Dissociation owing to the diffusion of hydrogen could significantly increase the oxygen fugacity in the melt and hence the Fe 2 O 3 content, leading to a systematic decrease of K D values.Because of this possible problem, we prefer to use 'dry' experiments in our treatment.We believe that the presence of small amounts of H 2 O in the melt (2-3 wt.%) will not strongly affect the derived correlation.

Spinel
To estimate the effect of pressure and temperature on the composition of liquidus spinel, available data on compositions of coexisting melts and spinels were selected over a wide range of temperature (1200-1400 C C), pressure (1 atm-15 kbar)and melt compositions (Fig. Bl.B).Data from experiments and from melt inclusions in minerals were used.The data were divided into three groups: P<2 kbar, P= 10 kbar and P= 15 kbar.The first group consists of experiments at 1 atm (Fisk & Bence, 1980;Sack, 1982;Grove & Bryan, 1983) and results of melt inclusion studies, in which pressure was determined independently by the study of fluid inclusions (Nikogosyan, 1990, and our data).The other two groups consist of data from Jaques & Green (1980) and Fujii & Scarfe (1985).In all cases, TiO 2 content in spinels were <0-5 wt.%.A1 2 O-, content in melts was chosen as a melt composition indicator, because the experimental study of Maurel & Maurel (1982b) showed a strong positive correlation between A1 2 O 3 content in coexisting spinels and melts.From Fig. Bl.B, the decrease of Cr N of spinel with increasing pressure is obvious.Data from melt inclusion studies at 8 and 3 kbar (Nikogosyan, 1990;Gurenko et al., 1991) were used as independent criteria of the correlation obtained.
FIG. 4. Composition of rocks, natural glasses, and homogeni2ed melt inclusions from Western and Eastern groups of the Tongan boninites.Data from this study andSharaskin el al. (1983ft),Falloon et al. (1987Falloon et al. ( ,1989)),and Falloon & Crawford (1991).•, Western group rocks; A, Eastern group rocks; +, homogenized melt inclusions; O, compositions of homogenized melt inclusions corrected for Fe-Mg exchange with host mineral (see text for discussion); A, compositions of naturally quenched melt inclusions in ohvine Fo, 4 from the Eastern group, used byFalloon & Green (1986)  for calculation of primary melt composition.

FIG. 5
FIG.5Homogenization temperatures of melt inclusions in olivines and orthopyroxenes from sample 26-2.A, comparison of homogenization temperatures of inclusions in pyroxenes with temperatures, calculated from hostmineral compositions using two-pyroxene thermometer ofLindsley (1983); dashed lines indicate the accuracy of the two-pyroxene thermometer.B, correlation between homogenization temperatures and host-mineral composition.The elongate cross shows error bars for B. l.Tongan olivines; 2, Tongan orthopyroxenes; 3, olivines from Troodos high-Ca boninite UPL.The correlation between olivine compositions and its crystallization temperature fits well to that demonstrated bySobolev el al. (1993)  for Troodos high-Ca boninites.C, D, the correlation of homogenization temperature with heating velocity and run duration.A7", )OM is the difference between experimentally obtained homogenization temperature and true homogenization temperature for melt inclusions; numbers in /im represent inclusion size; in C solid lines stand for olivine, dashed lines for orthopyroxene; in D solid line stands for melt inclusions in MORB olivines, dashed line for inclusions in Tongan olivines Comparison with MORB (D, our unpublished data) shows relatively low viscosity and, respectively, high velocity of processes in boninite melts.
FIG.8.A, B, correlation between H 2 O and other incompatible elements in melt inclusions in high-magnesian olivines from Tongan high-Ca boninites.Absence of any correlation in Tongan glasses demonstrates their degassed nature.C, H 2 O and K 2 O contents in natural quenched glasses and melt inclusions from lavas of various tectonic settings.Normalization to TiO 2 used to avoid effects of crystallization.1, Tongan boninite melt inclusions in high-magnesian olivines; 2, Tongan boninite glasses; 3, Troodos UPL glasses(Sobolev el al., 1993).Fields in C: 1, back-arc basin glasses; 2, MORB glasses; 3, arc tholeiites glasses; solid field, Hawaiian glasses.All data are fromDanyushevsky el al. (1993)  and reference therein.
FIG.10.Incompatible-element contents in primary Tongan high-Ca boninite melts.El and E2 represent two incompatible-element-enriched components involved in their petrogenesis (see text for discussion).Normalization to primitive mantle (Table3).
FIG. Bl.A, correlation between K D of Fe-Mg exchange between olivine and melt and the temperature of equilibria.B, pressure dependence of spinel composition in equilibrium with melt.

TABLE 2
Compositions of quenched glasses

TABLE 3
Major-and trace-element concentrations in high-Ca boninites(wt.% and ppm, respectively)

TABLE 4
Compositions of the melt inclusions in minerals HOW

TABLE 5
Trace-element concentrations in melt inclusionsTrace elements in inclusions O-32 and Op-9 were analysed by proton microprobe, and those in other inclusions by ion microprobe.

TABLE 6
Calculated parental and primary melts -pressure of the last equilibrium with the mantle source.Estimations of primary melt for the Western and the Eastern group were calculated using the following correlations between Fe 2+ (Ford et a/., 1983)tm dry olivine liquidus temperature(Ford et a/., 1983); T cr -liquidus temperature estimated from the experimental study of melt inclusion; P cr -pressure of the beginning of crystallization; T m -temperature of the last equilibrium with the mantle source; P m