Journal of Petrology

This Article Abstract FREE Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Search for citing articles in: ISI Web of Science (68) Request Permissions Google Scholar Articles by EILER, J. M. Articles by STOLPER, E. M. Search for Related Content GeoRef GeoRef Citation Social Bookmarking          What's this?

Journal of Petrology Volume 41 Number 2 Pages 229-256 2000
© Oxford University Press 2000

Oxygen Isotope Geochemistry of Oceanic-Arc Lavas

JOHN M. EILER^1,*, ANTHONY CRAWFORD², TIM ELLIOTT³, KENNETH A. FARLEY¹, JOHN W. VALLEY⁴ and EDWARD M. STOLPER¹

¹DIVISION OF GEOLOGICAL AND PLANETARY SCIENCES, CALIFORNIA INSTITUTE OF TECHNOLOGY, PASADENA, CA 91125, USA
²DEPARTMENT OF GEOLOGY, UNIVERSITY OF TASMANIA, GPO BOX 252C, HOBART, TAS. 7001, AUSTRALIA
³FACULTEIT AARDWETENSCHAPPEN, VRIJE UNIVERSITEIT, DE BOELELAAN 1085, 1081 HV AMSTERDAM, NETHERLANDS
⁴DEPARTMENT OF GEOLOGY AND GEOPHYSICS, UNIVERSITY OF WISCONSIN, MADISON, WI 53706, USA

Received November 2, 1998; Revised typescript accepted July 15, 1999

	ABSTRACT

TOP ABSTRACT INTRODUCTION SAMPLES AND ANALYTICAL METHODS RESULTS DISCUSSION SUMMARY AND CONCLUSIONS APPENDIX REFERENCES

 
Variations of oxygen isotope ratios in arc-related lavas can constrain the contributions of subducted crustal igneous rocks, sediments, and fluids to the sub-arc mantle. We have measured oxygen isotope ratios in 72 arc and back-arc lavas from five ocean–ocean subduction zone systems using laser-fluorination analyses of olivine and other phenocrysts and glass. Eighty percent of our samples have ¹⁸O values for any given phase (olivine, plagioclase, glass, or biotite) within 0·2 of the average value for that phase in upper-mantle peridotites and mid-ocean ridge basalt (MORB); the range for each phase is 1·0. This result contrasts with previous studies of whole-rock samples (which are significantly more variable even after exclusion of samples believed to be altered or fractionated by magmatic differentiation) and demonstrates that most arc-related lavas contain 1–2% of ¹⁸O-enriched crustal oxygen from any source (i.e. assimilation or subducted contributions). Elevations in ¹⁸O that do occur in these basic, arc-derived magmas relative to values most common for mantle-derived lavas are associated both with ‘enriched’ radiogenic isotope signatures and, even more strongly, with chemical indices consistent with high integrated extents of melting of their peridotite sources. We interpret these relationships as evidence that melting in the sources of the high-¹⁸O lavas we have studied was fluxed by addition of high-¹⁸O aqueous fluid (or perhaps a hydrous melt) from the subducted slab, such that sources that contain relatively large components of slab-derived fluid or melt are both relatively ¹⁸O enriched and also experienced relatively large amounts of melting. We have developed a quantitative model linking the amount of melting to the extents of ¹⁸O, radiogenic isotope, and trace-element enrichment in a mantle source being fluxed by addition of aqueous fluid. Comparison of this model with observed variations in the geochemistry of lavas from the Vanuatu–Fiji–New Caledonia region (the suite of related samples showing the greatest range in ¹⁸O observed in this study) constrains the amounts and chemical and isotopic compositions of slab-derived phases in the sources of these arc-related lavas. Assuming a ¹⁸O value of 20 for the slab-derived fluid, 0·5–1·0 wt % is added to the sources of most mantle-derived arc magmas; the maximum amount of slab-derived flux in the sources of arc magmas according to our results is 2·5 wt %.

KEY WORDS: oxygen isotopes; arc volcanism

	INTRODUCTION

TOP ABSTRACT INTRODUCTION SAMPLES AND ANALYTICAL METHODS RESULTS DISCUSSION SUMMARY AND CONCLUSIONS APPENDIX REFERENCES

 
There is considerable evidence that igneous rocks, sediments, and fluids from subducted oceanic crust contribute to the formation of convergent margin magmas (e.g. Morris & Tera, 1989; McCulloch & Gamble, 1991; Stolper & Newman, 1994; Elliott et al., 1997; Hawkesworth et al., 1997). These contributions are generally assumed to be introduced into the overlying mantle wedge as aqueous fluids and/or silicate melts released from the downgoing slab as it heats up. Mixing of these water-rich fluids and melts with peridotite in the overlying mantle leads to the production of magmas with the distinctive petrology and geochemistry of convergent margin igneous rocks. Given the likely thermal structure of the subducting slab and overlying mantle, melting in the mantle wedge may in many cases require and/or be controlled by the introduction of these water- and incompatible-element-rich fluids and/or melts because they significantly lower the temperature of the peridotite solidus (Kushiro et al., 1968).

Efforts to identify and characterize the contributions of subducted materials to arc volcanism have emphasized the abundances and isotopic ratios of minor and trace elements that are relatively abundant in altered basaltic rocks and sediments and are concentrated into aqueous fluids and silicate melts relative to residual solids (e.g. K, B, Sr, U; Morris & Tera, 1989; Plank & Langmuir, 1993; Hawkesworth et al., 1994; Leeman et al., 1994; Ryan et al., 1995). These elements can be sensitive indicators of the presence of slab-derived components in the sources of arc lavas; they can also be used to discriminate among and characterize various slab-derived components (e.g. aqueous fluids vs silicate melts; Elliott et al., 1997) and to estimate the time required for these components to travel from the slab to the source regions of magmas in the mantle wedge (Hawkesworth et al., 1997; Regelous et al., 1997). However, such elements are poor monitors of the absolute amounts of slab-derived components in magmatic sources because the concentrations of these elements vary widely among subducted crustal rocks and sediments (Plank & Langmuir, 1998) and because the partitioning of these elements among minerals, melts, and aqueous fluids is strong, highly variable, and only partially constrained by experiment (Green, 1994; Brenan et al., 1995; Keppler, 1996; Johnson & Plank, 1999).

Oxygen isotopes can provide constraints on the sources of arc volcanism that are complementary to those of trace element abundances and their isotope ratios. Most ultramafic rocks and inferred sources of basaltic lavas have ¹⁸O [¹⁸O=(¹⁸O/¹⁶O_sample/¹⁸O/¹⁶O_VSMOW - 1) x 1000, where ¹⁸O/¹⁶O_VSMOW = 0·0020052] values of 5·5 ± 0·2 (Mattey et al., 1994). In contrast, ¹⁸O values of altered basalts in the uppermost oceanic crust are typically +10–20 (compared with values of 0–7 for the lower oceanic crust; Muehlenbachs, 1986; Staudigel et al., 1995); values for coarse clastic sediments are also typically +10–20 (Arthur et al., 1983); values for oceanic clays are typically +15–20 (Arthur et al., 1983); and typical values for carbonate-rich and siliceous pelagic sediments are +30–35 (Kolodny & Epstein, 1976; Arthur et al., 1983). The oxygen isotopic contrast between upper oceanic crust (i.e. sediments and altered basalts) on the one hand and typical upper mantle on the other therefore provides a potential indicator of the presence of subducted upper oceanic crustal materials in the sources of subduction-zone lavas (note, however, that the contrast in oxygen isotope ratios between the mantle and lower-crustal materials may be more subtle and in the opposite direction; Muehlenbachs, 1986). It is known that the distinctive ¹⁸O values of the oceanic crust can be preserved to depths of tens of kilometers in subduction zones (Bebout & Barton, 1989), can persist for long times in deeply recycled materials (Garlick et al., 1971), and can survive transport through the sub-arc mantle in a melt or fluid phase (Eiler et al., 1998). Moreover, there are only modest variations in the concentration of oxygen in most geological solids and fluids, and the isotopic composition of oxygen is only weakly and predictably fractionated by high-temperature exchange among solids, melts, and fluids (e.g. Chiba et al., 1989; Palin et al., 1996); therefore, variations in oxygen isotope ratios in arc-related magmas may more strongly constrain the absolute amounts of slab-derived components contributing to the mantle sources of arc-related magmas than do incompatible trace elements and their isotopes, i.e. uncertainties in such estimates should stem only from uncertainties of a factor of 2 in the ¹⁸O values of subducted upper-crustal materials, rather than from uncertainties of an order of magnitude or more in the abundances of incompatible trace elements. Such constraints on the amounts of fluid and/or melt added to the sources of arc lavas could in turn help to constrain the concentrations of incompatible trace elements in the slab-derived components, to distinguish between metasomatic fluids and melts (which may differ substantially in their abundances of certain elements relative to oxygen), and to define the processes by which slab-derived components interact with the mantle (e.g. bulk mixing, chromatographic exchange, fluid-driven melting).

The oxygen isotope geochemistry of arc-related lavas is known principally through measurements of whole-rock samples of andesite and more evolved rocks [see review by Harmon & Hoefs (1995)]. Such data are the basis of several early and fundamental constraints on subduction-related volcanism, including the discovery of the approximate (±1–2) correspondence between arc-related andesites on the one hand and peridotites and basalts from other tectonic settings on the other (suggesting an origin of the parental liquids of andesitic magmas by melting of peridotite rather than by direct melting of subducted crustal rocks—a significant finding at the time it was made; e.g. Matsuhisa et al., 1973) and of the importance of crustal assimilation and large-scale crustal melting in arc magmatism (e.g. Hildreth & Moorbath, 1988; Davidson & Harmon, 1989). However, the oxygen isotope ratios of the abundant evolved lavas from arcs have been modified from primitive, mantle-derived values by magmatic processes (e.g. fractional crystallization and crystal accumulation; Woodhead et al., 1987; Harmon & Hoefs, 1995), and fine-grained and glassy lavas are particularly susceptible to subsolidus alteration (Davidson & Harmon, 1989; Harmon & Hoefs, 1995). These high-level fractionation and alteration processes can produce shifts in ¹⁸O of 1 or greater in lavas and thus hinder the study of the smaller (<1) variations in oxygen isotope ratios known to characterize the mantle sources of ocean-island basalts (OIB) (Eiler et al., 1996) and that might be more typical consequences of the interactions of the sub-arc mantle with fluids or melts from the subducted slab. Previous efforts to circumvent these difficulties in applying oxygen isotope geochemistry to the study of arc magmatism have included correction or exclusion of measured ¹⁸O values of whole-rock specimens inferred to be significantly altered and/or unusually evolved (Woodhead et al., 1987; Davidson & Harmon, 1989; Harmon & Hoefs, 1995), or analysis of glass (Ito & Stern, 1985; Macpherson & Mattey, 1997) or phenocrysts (Singer et al., 1992; Smith et al., 1996; Thirlwall et al., 1996; Macpherson & Mattey, 1997; Macpherson et al., 1998) rather than whole rocks.

We describe in this paper the results of a study of oxygen isotope variations in relatively magnesian lavas from oceanic arcs (and associated back-arc basins) based primarily on measurements on olivine phenocrysts, supported with less extensive measurements on other phenocryst phases and glass. Olivine has a low rate of oxygen exchange in the absence of replacement by alteration minerals (Reddy et al., 1980); it is an early phase in the low-pressure crystallization sequences of primitive basaltic liquids; and it is thought to be present as a major phase in the mantle sources of basalts. It is therefore an obvious target for providing insights into the upper-mantle sources of basic magmas. We note, however, that our emphasis on olivine-bearing lavas may have prejudiced our sampling against magmas (and their fractionation products) produced by single-stage melting of non-ultramafic sources (e.g. partial melts of subducted crust unmodified by reaction with the mantle wedge in transit to the surface), if such magmas exist (Defant & Drummond, 1990). Approximately 20 laser-fluorination measurements of the ¹⁸O of olivines from arc-related magmas have been previously published, distributed among four studies of the Lau basin, the Lesser Antilles, and the Kermadec–Hikurangi margin (Smith et al., 1996; Thirlwall et al., 1996; Macpherson & Mattey, 1997; Macpherson et al., 1998). We have not integrated these data with our study because of both the small number and wide distribution of these previous measurements and the difficulty in interpreting small (several tenths of per mil) variations in ¹⁸O measurements from different laboratories without careful cross-standardization. However, we note that data from those studies are generally within the range of results presented here. Results for other phases (glass and clinopyroxene) from these recent studies will be discussed further below.

	SAMPLES AND ANALYTICAL METHODS

TOP ABSTRACT INTRODUCTION SAMPLES AND ANALYTICAL METHODS RESULTS DISCUSSION SUMMARY AND CONCLUSIONS APPENDIX REFERENCES

 
The suite of samples we have studied includes 72 basalts, basaltic andesites, shoshonites, boninites, and alkali basalts from five ocean–ocean subduction systems: the Mariana arc and back-arc trough, the South Sandwich arc and associated Scotia Sea back arc, the Vanuatu–Fiji arc and nearby New Caledonia arc (these are treated as a single suite in much of the subsequent discussion), and Papua New Guinea [alkaline lavas of the Tabar, Lihir, Tanga, and Feni (‘TLTF’) islands, derived from melting of the mantle wedge of a recently active subduction zone]. The sample suite was restricted to lavas from oceanic (as opposed to continental) arcs and to olivine-phyric lavas that are more primitive than those typically considered in previous studies of the oxygen isotope geochemistry of arc lavas (the mean MgO and SiO₂ contents of samples for which major element data are available are approximately 7 and 51 wt %), although samples were also chosen so as to include a significant range in minor and trace element geochemistry and radiogenic isotope ratios. Analyses are available for 58 of the 72 samples for major element concentrations, trace element concentrations, mineral chemistry, and/or Sr, Nd, Pb, and/or U-series isotope geochemistry, providing a petrological and geochemical context in which to interpret our data. The supplemental chemical data are reported using the same sample labels as in the literature (Hawkesworth et al., 1977; Saunders & Tarney, 1979; Hawkins & Melchior, 1985; Maillet et al., 1986; Kennedy et al., 1990; Eggins, 1993; Monzier et al., 1993; Sigurdsson et al., 1993; Rogers & Setterfield, 1994; Crawford et al., 1995; Pearce et al., 1995; Elliott et al., 1997; Peate et al., 1997) or are from unpublished studies of A. Crawford or R. Stern; a table including the supporting geochemical data is available on request from the authors. The 14 samples that have not been previously characterized are related to characterized samples and come from previously studied sequences of lavas, such that aspects of their geochemistry can be confidently inferred; these samples are indicated in Table 1 and in the figures. All of the significant variations and trends in ¹⁸O observed in this study have extremes exhibited by well-characterized samples.

All samples were crushed in a steel percussion mortar and dry-sieved to size fractions of 600–300 µm and 300–150 µm for hand picking of olivine, plagioclase, glass, and/or biotite separates. All separates were cleaned of dust by briefly blowing filtered compressed air over them while holding them in a 100 µm sieve. Olivine separates were prepared for 67 of the 72 samples; 24 plagioclase separates, 11 glass separates, and two biotite separates were also prepared.

Most mineral and glass separates were analyzed for ¹⁸O at the CO₂-laser-fluorination laboratory at the University of Wisconsin, Madison [‘UW’, described in detail by Valley et al. (1995)]. A subset of the data was measured by the same methods at the University of Southern California (‘USC’). Analyses at UW were made on 12 days over a period of 7 months, whereas those at USC were made on 3 days over a period of 1 month. Measurements of unknowns on each day in both laboratories were accompanied by 4–8 measurements of UWG-2 garnet (Valley et al., 1995) and 2–4 measurements of the SCO-1 olivine working standard (Eiler et al., 1995). The 1 uncertainty for a given standard on any given day averaged 0·07. In addition, 60 of the unknown samples were measured 2–3 times, and the mean of the 1 uncertainties of these multiple analyses is 0·06. On 12 of the 15 days on which analyses were performed, measurements of each standard averaged within 0·1 of the accepted value and no systematic correction was applied to the data. On 3 days, however, the averages for both UWG-2 and SCO-1 differed from their accepted values by 0·1–0·2, and a correction was applied to the data equal to the average of the differences between the mean-measured value and the accepted value for each of the two standards. Such corrections are required on 25% of the working days at the UW laboratory and have been inferred to be caused by small variations in the condition of the gas-purification and gas-transfer apparatus (Valley et al., 1995). Four olivine samples (i.e. unknowns) that had been measured on a day on which a correction was applied based on the analyses of the standards were reanalyzed on another day on which no correction was required; the average difference between the corrected and uncorrected measurements was 0·06. One analysis of an olivine separate (GUG-3) yielded an anomalous result (3·5) that was not reproduced by two further measurements of a new mineral separate from this sample. The anomalous value was discarded, and only the average of two additional measurements is listed in Table 1a.

View this table: [in this window] [in a new window]   Table 1a: Oxygen isotope data and select major elements

 

The accuracy of our analyses was assessed by measurement of seven silicates (three olivines, two basaltic glasses, one biotite, and one plagioclase; Table 1b), six of which had been previously analyzed multiple times by other laboratories using conventional fluorination procedures. The average difference between the mean value of ¹⁸O for these materials determined by laser fluorination at UW and the accepted values was 0·09; that is, comparable with our precision based on repeated analyses of the UWG-2 working standard. Moreover, the UWG-2 garnet standard and three other materials (SCO-1, KHO-1, and AH95-22 glass) were analyzed in both the UW and USC laser fluorination laboratories, and the differences between the average values for these materials from these two laboratories averaged 0·06. Overall, these demonstrations of reproducibility and of agreement between laser fluorination analyses at UW and USC and with conventional fluorination analyses allow us to infer that our data are accurate and precise to approximately ±0·07 (1) for the phases analyzed in this study.

View this table: [in this window] [in a new window]   Table 1b: Oxygen isotope standards

 

	RESULTS

TOP ABSTRACT INTRODUCTION SAMPLES AND ANALYTICAL METHODS RESULTS DISCUSSION SUMMARY AND CONCLUSIONS APPENDIX REFERENCES

 
Values of ¹⁸O determined in this study vary over a range of 1 or less for each of the phases analyzed: the ranges are 0·98 for the 67 olivine analyses; 0·38 for the 24 plagioclase analyses; and 0·14 for the 11 glasses (Table 1a). Average values of ¹⁸O for olivine, plagioclase, and glass are within 0·1 of those for fresh mid-ocean ridge basalts (MORBs) or their inferred sources (Ito et al., 1987; Eiler et al., 1996); likewise, the average for olivine in our sample suite is within 0·1 of the average for olivine from mantle peridotite xenoliths (Mattey et al., 1994). Even the largest range in ¹⁸O for a single phase (olivine) is small relative to expectations based on most previous studies of arc-related lavas (Harmon & Hoefs, 1995). The relatively small range in ¹⁸O is particularly striking given the number and diversity of samples that we have studied: our suite of samples is taken from five different arcs and three back-arc regions and includes much of the range in radiogenic isotope ratios (e.g. ⁸⁷Sr/⁸⁶Sr = 0·7026–0·7044), trace element compositions (e.g. Ba/Nb = 6–400), U-series radioisotope activity ratios (e.g. [U²³⁸/Th²³⁰] = 1·0–1·6), and petrologic diversity (i.e. island-arc basalts and basaltic andesites, back-arc-basin basalts, shoshonites, boninites, and alkali basalts) known from oceanic-arc lavas. It should be noted, however, that we have not included any samples characterized by exceptionally high ⁸⁷Sr/⁸⁶Sr ratios and Th concentrations, such as those common to the Philippines, Indonesia, and the Aeolian islands [i.e. the ‘high-Ce/Yb’ arcs of Hawkesworth et al. (1994)].

Approximately 80% of the ¹⁸O values for olivine are within the range for typical upper-mantle olivine based on peridotite xenoliths (5·0–5·4; Mattey et al., 1994). Only one sample (Charlotte 22 from Vanuatu) is significantly below this range (and this one by only 0·15); the remaining 20% of the samples are enriched in ¹⁸O relative to this range by up to 0·43 (Fig. 1). Relatively ¹⁸O-enriched samples include three island-arc basalts (all from western Vanuatu), three shoshonites (also from the Vanuatu region), three boninites (from Vanuatu and New Caledonia), and six alkali basalts from the TLTF islands (Papua New Guinea). In the following paragraphs we assess some of the systematics of our data, comparisons with previous work, and correlations of our data with other geochemical characteristics of the samples.

View larger version (46K): [in this window] [in a new window]   Fig. 1. Values of 18O for olivine from oceanic-arc lavas analyzed in this study, organized into groups defined by geographic location and petrologic type. The gray field indicates the range typical of olivine from upper-mantle peridotite xenoliths (Mattey et al., 1994) and inferred for olivine in the MORB source (5·2 ± 0·2; Ito et al., 1987; Eiler et al., 1996). Short vertical lines mark average values for each suite of related samples. Samples are labeled by location and by petrologic type: island-arc basalt (IAB, circles), back-arc-basin basalt (BABB, triangles), shoshonite (squares), boninite (stars), and alkali basalt (Alk B, diamonds). TLTF indicates lavas from the Tabar–Lihir–Tanga–Feni islands, Papua New Guinea. Symbols for Mariana arc and back-arc samples are open; for South Sandwich–Scotia Sea samples they are gray; for the Vanuatu–Fiji–New Caledonia suite they are filled; for TLTF samples they are open.

 

Distribution of oxygen isotopes among coexisting phases
We have made nine determinations of basaltic glass–olivine oxygen isotope fractionation (_{glass–olivine}; Fig. 2a) and 21 determinations of plagioclase–olivine fractionation (_{plagioclase–olivine}; Fig. 2b; where _i–j = ¹⁸O_i – ¹⁸O_j). Both of these fractionations are essentially constant given the precision of our measurements: _{plagioclase–olivine} = 0·65 ± 0·14 (1) and _{glass–olivine} = 0·36 ± 0·11 (1). The expected precision of a determination of _i–j is 0·10 (1), based on independent errors of 0·07 for a determination of ¹⁸O on each phase.

View larger version (42K): [in this window] [in a new window]   Fig. 2. Histograms illustrating the variability in glass–olivine (a) and plagioclase–olivine (b) measured in this study (dark gray boxes) and from previous studies (diagonally ruled boxes; data from Onuma et al., 1970; Clayton et al., 1971, 1972; Kyser et al., 1981, 1982; Macpherson & Mattey, 1997). Vertical dashed lines and light gray fields illustrate the range of fractionations between olivine and either An50–90 plagioclase or basaltic melt at 1300°C based on experimental studies of isotope partitioning and models of isotope partitioning among silicate phases (Muehlenbachs & Kushiro, 1974; Chiba et al., 1989; Rosenbaum, 1994; Matthews et al., 1998).

 
The average fractionation we determine between plagioclase and olivine phenocrysts compares favorably with that expected at magmatic temperatures based upon exchange experiments (e.g. 0·58 for An₉₀ plagioclase to 0·76 for An₅₀ plagioclase at 1300°C; Chiba et al., 1989). The fractionation between basaltic liquid and olivine is not known directly from experiment, but a value of 0·32 is predicted at 1300°C based upon the combination of experimentally determined melt–CO₂, forsterite–calcite, and calcite–CO₂ fractionations (Muehlenbachs & Kushiro, 1974; Chiba et al., 1989; Rosenbaum, 1994). However, the accumulated error in such an estimate is relatively large (±0·2–0·3, 1), so the close correspondence between this predicted value and our measured value may not be significant. An empirical model for oxygen isotope partitioning between minerals and melts (Matthews et al., 1998) predicts a value of 0·4–0·5 for _{melt–olivine} for the lava compositions considered in this study. Our results compare favorably with the range of values expected from these two independent estimates.

The fractionations of oxygen isotopes we observe among coexisting phases are within the ranges of but more restricted than fractionations observed in previous studies of phenocrysts and glass or groundmass in basic lavas (Fig. 2). Values of _{glass–olivine}, _{groundmass–olivine}, or _{whole rock–olivine} from previous studies range from 0·2 to 2·0, and values for equilibrium melt–olivine fractionation have been suggested to be as high as 0·9–1·0 (Onuma et al., 1970; Clayton et al., 1971, 1972; Hoernes & Friedrichsen, 1977; Kyser et al., 1981, 1982; Macpherson & Mattey, 1997). Our results indicate a value of 0·4 ± 0·1, significantly lower than the upper end of the range of previous data. Previous measurements of _{plagioclase–olivine} are similarly more widely distributed (0·4–1·3) and on average larger than observed in this study (Onuma et al., 1970; Clayton et al., 1972; Hoernes & Friedrichsen, 1977; Kyser et al., 1981, 1982). These differences may be due to the fact that ours is the first study to make a significant number of determinations of _{glass–olivine} and _{plagioclase–olivine} using laser-based measurements, and therefore we may have been more successful at avoiding sample alteration and the well-known analytical difficulties associated with conventional oxygen isotope measurements of olivine (Mattey et al., 1994).

Comparison with previous work
The range of ¹⁸O values we find for olivine phenocrysts (0·98) is a factor of 10 lower than that reported previously based on conventional fluorination measurements of whole-rock samples from oceanic arcs [summarized by Harmon & Hoefs (1995)]. Some of this difference could reflect our emphasis on olivine-bearing samples, which are primitive relative to the full range of arc-related lavas represented in the conventional data base. However, this is not likely to be the principal cause because much of the reported range in whole-rock measurements can be found in basalts, basaltic andesites, and andesites in the same arcs we have studied (Pineau et al., 1976; Ito & Stern, 1985; Woodhead et al., 1987; Harmon & Hoefs, 1995). We infer, as has previously been concluded from study of mantle peridotites and ocean-island basalts (Mattey et al., 1994; Eiler et al., 1995, 1996), that the oxygen isotope variations in olivine phenocrysts are more restricted than in whole rocks primarily because analysis of fresh phenocrysts (particularly olivines) avoids or minimizes the effects of low-temperature alteration; nevertheless, an additional factor may be that the increases in ¹⁸O that are known to build up in residual liquids from fractional crystallization (Matsuhisa et al., 1973) are minimized by focusing on relatively primitive, olivine-bearing samples.

Our data contradict a conclusion reached by some previous studies of oxygen isotope variations in oceanic-arc basalts that the average value of ¹⁸O in oceanic-arc magmas, even after correction for low-temperature alteration and magmatic fractionations, is 0·5 higher than for MORB and back-arc-basin basalts (Woodhead et al., 1987; Harmon & Hoefs, 1995) and that this difference reflects a common and relatively large amount of subducted crustal oxygen in the sources of these arc magmas. In contrast, we find that elevations in ¹⁸O of 0·5 with respect to MORB are an extreme found only in a minority of arc samples and that the sources of both arc and back-arc lavas are on average closely similar to estimates for the typical upper mantle based on measurements on MORBs and peridotite xenoliths (Fig. 1). We note that in this respect our results confirm the work of Ito & Stern (1985), who found that glasses from both the Mariana arc and trough are similar in ¹⁸O to MORB. Similarly, recent results for glasses from the Lau basin (Macpherson & Mattey, 1997) span a small range in ¹⁸O near the average value for MORB and display correlations with chemistry that are analogous to correlations displayed by our data (discussed below). Recent study of lavas from the Lesser Antilles and the Kermadec–Hikurangi margin have found larger ranges and higher average values of ¹⁸O than both MORB and the results of this study, but in both cases these differences were interpreted to reflect assimilation of continental material (Smith et al., 1996; Thirlwall et al., 1996; Macpherson et al., 1998); these studies confirm the conclusions of Harmon et al. (1981), Hildreth & Moorbath (1988), and Davidson & Harmon (1989) that variations in ¹⁸O of arc-related lavas erupted through thick sequences of continental rocks or sediments dominantly reflect assimilation–fractional crystallization processes.

Correlations between ¹⁸O and sample chemistry
When comparing our oxygen isotope data with other chemical indices, it is useful to focus on a single phase to remove the effects of oxygen isotope fractionations among different crystalline phases and glass. We have selected olivine for this purpose because it was measured in most of our samples and, for the reasons discussed above, we believe it to be a reliable monitor of ¹⁸O in unaltered and relatively primitive basaltic lavas. For the five out of 72 samples in which olivine was not analyzed, we calculated an ‘equivalent olivine’ value equal to the value measured in plagioclase or glass in that sample minus the average fractionation between olivine and plagioclase or glass observed in this study. Although these fractionations are well defined by our data, this correction introduces additional uncertainty for these samples; they are distinguished from the other data by parentheses in Table 1. None of the oxygen isotope variations or correlations observed in this study are significantly dependent on these five values, but they are provided for completeness. We emphasize at the outset of this discussion that all values of ¹⁸O in excess of 5·4 occur in lavas from the Vanuatu and New Caledonia arcs or the TLTF islands (i.e. Papua New Guinea), and we reiterate that such ¹⁸O-enriched samples are atypical (20% of our survey sampling); thus, the observed trends and our interpretations of them are biased toward processes relevant to these samples. However, the other studied suites are consistent with the trends defined by these extreme samples and therefore are consistent with the interpretation that such suites were influenced, though not as strongly, by similar processes.

Values of ¹⁸O in olivine phenocrysts in our samples are not correlated with the whole-rock SiO₂ or MgO contents of the host lavas when the data set is viewed as a whole (Table 1). We infer that the effects on ¹⁸O of magmatic fractionation and accumulation leading to such correlations for whole-rock data from arc settings (Matsuhisa et al., 1973; Woodhead et al., 1987) have been minimized by our emphasis on relatively primitive olivine-bearing lavas.

A key question is whether ¹⁸O values for our samples correlate with conventional chemical and isotopic monitors of slab components in the sources of arc-related lavas. To answer this question, our results for the oxygen isotope composition of olivine in arc-related lavas are compared in Fig. 3 with ⁸⁷Sr/⁸⁶Sr and various trace- and minor-element indices that plausibly monitor the contribution of subducted materials to the sources of arc lavas.

View larger version (30K): [in this window] [in a new window]   Fig. 3. Comparison of oxygen isotope ratios in olivine phenocrysts from arc and back-arc lavas with other geochemical indicators in the host lavas: (a) 87Sr/86Sr; (b–f) various minor and trace element abundance ratios (by weight). Sr isotope and element-abundance ratios are either measured on the same samples (small symbols) or estimated based on data for closely related samples (large symbols). (See text for sources of trace element and Sr isotope data.) Symbols are as in Fig. 1. Gray boxes mark typical NMORB compositions (Sun & Nesbitt, 1979; Ito et al., 1987; Michael, 1995). It should be noted that only lavas with MgO contents between 6 and 12 wt % are plotted in (c), to minimize the effects of fractionation on the K2O/TiO2 ratio. This range was selected based on the observed relationship between this ratio and MgO in Vanuatu and Mariana arc lavas.

 

Values of ¹⁸O in olivine span a restricted range near the average upper-mantle value in lavas with low-⁸⁷Sr/⁸⁶Sr ratios; elevated ¹⁸O values are restricted to samples at the high end of the distribution of ⁸⁷Sr/⁸⁶Sr (Fig. 3a). Radiogenic strontium in arc-related lavas is generally interpreted as a monitor of slab-derived contributions to their sources (Gill, 1981), and thus the relationship shown in Fig. 3a suggests that the elevated oxygen isotope ratios we observe are related to slab-derived components. The observed relationship between ¹⁸O and ⁸⁷Sr/⁸⁶Sr is consistent with a hyperbola that results from mixing between an Sr-poor, low-¹⁸O, low-⁸⁷Sr/⁸⁶Sr component (i.e. the approximately MORB-like mantle wedge) and an Sr-rich, high-¹⁸O, high-⁸⁷Sr/⁸⁶Sr component (i.e. slab-derived fluid and/or melt). It should be noted that the overall hyperbolic trend shown in Fig. 3a, although defined by the data set as a whole, is best defined by the single suite of island-arc basalts from the western Vanuatu arc (filled circles); this suite is the only group of related samples considered in this study that covers a sufficient range in ¹⁸O and ⁸⁷Sr/⁸⁶Sr to display the full trend. It should be noted also that correlations between ¹⁸O and ⁸⁷Sr/⁸⁶Sr as a result of assimilation of pre-existing crust are generally linear rather than hyperbolic, because of the similarity in Sr concentrations of arc-related lavas and older rocks in the arc crust (e.g. Davidson & Harmon, 1989); therefore, the hyperbolic relationship in Fig. 3a suggests processes other than crustal assimilation.

Arc-related lavas are characterized by distinctive enrichments and depletions in certain trace elements relative to MORB and OIB that are associated with (or defined as characteristics of) slab-derived components (Gill, 1981). Certain chemical characteristics are common to nearly all arc-related lavas and are rare elsewhere [e.g. enrichment of large ion lithophile elements (LILE) such as K, Ba, and Sr with respect to high field strength elements (HFSE) such as Ti and Nb], whereas others are thought to be diagnostic of the presence in the sub-arc mantle of distinctive components such as melts or aqueous fluids that may sample a specific part of the subducting slab (e.g. the basaltic crust vs the overlying sediments; Kay, 1980; Ellam & Hawkesworth, 1988; Elliott et al., 1997; Turner et al., 1997) or pre-existing ‘OIB-like’ enrichments in the mantle (Lin et al., 1989). The number and identity of such components differ according to the set of elements considered; for example, the classification of Hawkesworth et al. (1997) would define all lavas considered in this study as falling within the compositional field (Sr/Th > 200; Th < 5 ppm) interpreted to reflect dominantly addition of a slab-derived aqueous fluid; in contrast, Elliot et al. (1997) concluded that sediment-derived melt controls the budgets of many trace elements in the sources of samples we have studied from the Marianas. However, certain common themes have emerged in recent studies: uranium excesses (i.e. [²³⁸U /²³⁰Th] > 1) and enrichments of fluid-soluble elements (Ba, Sr, U) with respect to non-fluid-soluble but incompatible elements (Th, Nb, Zr) are generally identified with slab-derived aqueous fluids dominated by contributions from the basaltic portions of the subducted oceanic crust (Ellam & Hawkesworth, 1988; Elliott et al., 1997; Turner et al., 1997); in contrast, high concentrations of Th, elevated ratios of Th and rare earth elements (REE) to Nb, strong light rare earth element (LREE) enrichments, negative Ce anomalies, and/or U-series isotopic compositions near the equiline (i.e. [²³⁸U /²³⁰Th] 1·0) are associated with a component of sediment-derived melt (Ellam & Hawkesworth, 1988; Elliott et al., 1997; Turner et al., 1997), or possibly a pre-existing ‘OIB-like’ enrichment in the mantle (Lin et al., 1989).

Elevated values of ¹⁸O (= 5·4) observed in this study are generally associated with elevated ratios of highly incompatible LILE to moderately incompatible trace elements not believed to be highly fluid soluble (e.g. Sr/Yb and K₂O/TiO₂; Fig. 3b and c). These trends, although somewhat scattered, are consistent with the conclusion drawn from Fig. 3a that ¹⁸O enrichments in lavas are associated with distinctive, ‘enriched’ components in their sources. Relationships between ¹⁸O and trace-element ratios thought to discriminate among different varieties of enriched components are complex and fail to reveal a unique association between ¹⁸O enrichments and any one of the several distinctive slab-derived components currently believed to contribute to the sub-arc mantle (Fig. 3d–f). For example, arc lavas with high-¹⁸O olivines have Ba/La ratios that are high relative to NMORB, but they span a considerable range and are not significantly elevated in Ba/La relative to the lower-¹⁸O arc-related lavas; in particular, the most extreme elevations of Ba/La among lavas examined in this study are in samples from the South Sandwich arc, yet these have ¹⁸O values indistinguishable from average upper mantle (Fig. 3d). It should be noted that enrichment of Ba relative to other LILE such as La is one of the most widely used indicators of contributions of fluid from the basaltic portions of the subducted slab to the sub-arc mantle (e.g. Gill, 1981; McCulloch & Gamble, 1991; Hawkesworth et al., 1994), so our data do not support an exclusive association of ¹⁸O enrichments with such a component. Similarly, lavas from Guguan island in the Mariana arc have ¹⁸O values for olivine within the range of olivine in typical upper-mantle peridotites (Table 1), despite spanning a significant range in [²³⁸U/²³⁰Th] ratio from 1·29 to 1·56; such large U excesses are taken as evidence that fluids from the basaltic portions of the subducting slab dominate the U and Th budget of the mantle sources of Guguan lavas (Elliott et al., 1997).

High-¹⁸O lavas are generally LREE enriched relative to NMORB and to most lower-¹⁸O arc-related lavas (Fig. 3e). However, shoshonites from the northern Mariana arc have ¹⁸O values similar to average upper mantle, yet they display extreme LREE enrichments, demonstrating that LREE and ¹⁸O enrichments are not uniquely linked in arc environments. Similarly, enrichments in LREE with respect to HFSE (e.g. La/Zr; Fig. 3f) are generally positively correlated with ¹⁸O among our samples, but the Mariana shoshonites have extreme enrichments of La with respect to Zr without any elevation in ¹⁸O. Enrichments of LREE relative to heavy rare earth elements (HREE) and HFSE in arc lavas have been proposed as evidence that sediment-derived melts dominate the trace-element characteristics of enriched sub-arc mantle (Brenan et al., 1995; Elliot et al., 1997); the results shown in Fig. 3e and f indicate that ¹⁸O elevations are generally but not universally associated with such a signature.

Figure 3d–f thus demonstrates an important result of our study: extreme ‘slab-derived fluid’ signatures (e.g. Ba/La and U/Th enrichment usually identified with fluids from the basaltic component of the subducting crust) and ‘sediment-derived melt’ and/or ‘OIB-like’ signatures (e.g. LREE/HREE or LREE/HFSE enrichment) can each be present in the sub-arc mantle sources of arc-related magmas without a distinctive oxygen isotope signature. Collectively, the relationships illustrated in Fig. 3 suggest that, although elevated ¹⁸O values of arc-derived magmas are associated in general terms with commonly used trace element indicators of slab-derived components in their sources (Fig. 3a–c), the abundances of such components as measured by oxygen isotopes are not related one-to-one to the proportions of distinguishable ‘slab fluid’ and ‘sediment melt’ components based on diagnostic trace element ratios (Fig. 3d–f).

Although the relationships between ¹⁸O and commonly used trace element monitors of slab-derived components are scattered, there are several well-defined relationships between the ¹⁸O values from this study and other geochemical indices, which rather than being direct monitors of slab contributions to the sources of arc lavas are useful as monitors of the extent of melting and/or prior depletion of the peridotitic sources of basaltic melts. These relationships are illustrated in Fig. 4 and described in the following paragraphs.

View larger version (28K): [in this window] [in a new window]   Fig. 4. Comparisons of oxygen isotope ratios in olivine phenocrysts from arc and back-arc lavas with other geochemical indicators: (a) wt % TiO2 in the host lava; (b) TiO2(8·0) values of the host lava [calculated following methods described by Klein & Langmuir (1987); see text for details]; (c) the cr-numbers of coexisting chromite phenocrysts [i.e. the Cr/(Cr + Al) ratio on a molar basis]; (d) the Yb/Sc ratio (by weight) of the host lavas. Gray boxes mark typical NMORB compositions. Chemical and mineral-composition data are either measured on the same samples (small symbols) or estimated based on data for closely related samples (large symbols). (See text for data sources.) Symbols are as in Fig. 1.

 

TiO₂ contents
Values of ¹⁸O are compared with concentrations of TiO₂ (wt %) measured in whole-rock specimens of host lavas in Fig. 4a; samples plotted in this panel are restricted to those with 6–12 wt % MgO to minimize variations related to fractional crystallization. This figure shows that the ¹⁸O-enriched samples all have low TiO₂ contents, generally less than 0·8 wt %—approximately half the average concentration in NMORB (Sun & McDonough, 1989) and at the lower end of the range observed in arc lavas (Plank & Langmuir, 1988). The overall trend shows both systematically higher TiO₂ contents for back-arc magmas (triangles) than for arc magmas with comparable ¹⁸O values and a negative correlation between TiO₂ and ¹⁸O when only the arc lavas are considered. All other things (e.g. source composition, mineralogy of the residue, and extent of fractionation) being equal, lower values of TiO₂ in basalts indicate higher extents of single-stage melting and/or melting of a more depleted (i.e. previously melted) peridotite source.

Although Fig. 4a is restricted to lavas with 6–12 wt % MgO, some of the variation in TiO₂ could still reflect superposition of the effects of fractionation on source and degree-of-melting effects. We have evaluated this by calculating values of TiO_2(8·0) for suites of related samples to correct for low-pressure fractionation (Klein & Langmuir, 1987). Eleven suites of lavas were selected that were sufficiently well sampled to permit an estimation of TiO_2(8·0) values: two islands from the Mariana arc (Guguan and Agrigan), the Mariana trough as a whole, the Scotia Sea as a whole, one island from the South Sandwich arc (Zavodovski), three islands from the Vanuatu arc (Ambrym, Tanna, and Tongoa), boninites from New Caledonia and the Hunter fracture zone (Vanuatu), and one island from the ‘TLTF’ arc (Lihir). TiO₂ contents of individual samples from each suite and related samples from that suite not analyzed for ¹⁸O in this study were regressed with MgO content, and the TiO₂ value of the resulting line at 8·0 wt % MgO was taken as the TiO_2(8·0) value for that suite; this value applies to all members of that suite and the ¹⁸O value is the average for one or more members of that suite analyzed for oxygen isotope composition in this study. Figure 4b confirms the key feature of Fig. 4a: elevated ¹⁸O values are associated with low values of TiO_2(8·0) in the host lavas, generating an overall negative correlation. Scatter about the trend in Fig. 4a is reduced in Fig. 4b. The trend in Fig. 4b is particularly well defined by data other than alkali basalts from Papua New Guinea (the open diamond). Alkali basalts, including those in arc settings, are systematically higher in Ti than related tholeiitic lavas (e.g. Shimizu & Arculus, 1975), so this discrepancy is not unexpected.

Macpherson & Mattey (1997) presented data for oxygen isotope variations in Lau basin back-arc lavas that show a subtle trend of increasing ¹⁸O with decreasing Na₂O_(8·0). The behavior of Na during partial melting of peridotite is, to first order, similar to that of Ti, and thus this observation is analogous to that described above. Lavas examined in this study and found to have relatively high values of ¹⁸O (¹⁸O_olivine = 5·4) also generally have low concentrations of Na relative to the sample suite as a whole (2·1 vs 2·5 wt % Na₂O), although the overall correlation between ¹⁸O and Na₂O is not as well defined as those illustrated for TiO₂ in Fig. 4a and b.

Chrome number
Among samples containing chromite phenocrysts (approximately half of our samples; mostly from the Vanuatu suite), high ¹⁸O values are restricted to samples with chromites having chrome numbers (i.e. cr-number = Cr/[Cr + Al]) of 0·80 or greater (Fig. 4c). Such values are high compared with chromites in abyssal peridotites (0·1 < cr-number < 0·6) and most basaltic lavas (0·2 < cr-number < 0·7); comparable values are common only in alpine peridotites, boninites, high-Mg andesites, and, less commonly, oceanic plateau basalts (Dick & Bullen, 1984; Arai, 1994). Relatively high cr-number values in spinels from peridotites are interpreted as an indication that the rocks are residual to high integrated extents of melting (i.e. the total amount of melt extracted from the source, whether in single or multiple melting events, is high) or high extents of melt–peridotite reaction (Arai & Matsukage, 1996); this is because Cr is compatible and Al incompatible in peridotitic residues equilibrated with basaltic melts (Dick & Bullen, 1984). When found in basaltic phenocrysts, such spinels are similarly thought to reflect derivation of those lavas from sources that have undergone unusually high integrated extents of melting (Dick & Bullen, 1984; Arai, 1994). Thus, the association of high ¹⁸O values with high cr-number suggests that the residues remaining in the sources of lavas with elevated ¹⁸O are highly depleted.

Yb/Sc
There is a negative correlation between ¹⁸O and Yb/Sc among samples from this study, regardless of geographic location and petrologic type (Fig. 4d). The Yb/Sc ratio of unfractionated, mantle-derived magmas is predicted to decrease continuously with increasing integrated extents of melting of peridotitic sources because Yb is somewhat more incompatible than Sc during peridotite melting (Klein & Langmuir, 1987). Clinopyroxene fractionation will lead to progressive increases in this ratio in residual liquids, but it should be insensitive to fractional crystallization of other phases and particularly during fractionation of olivine from relatively primitive magmas. Although care must be taken to exclude highly fractionated magmas when examining variations in Yb/Sc, observed relationships between Sc and Yb concentrations and other indices of extent of melting in both MORB and arc-related magmas (Klein & Langmuir, 1987; Plank & Langmuir, 1993) confirm that they can be a valuable measure of integrated degree of melting. We have examined the Yb/Sc relationship shown in Fig. 4d for the effects of clinopyroxene fractionation; although this process probably influenced the two highest Yb/Sc ratios, the trend is unchanged if only samples with 8% MgO and/or 25 ppm Sc are plotted (a compositional range over which the effects of clinopyroxene fractionation are not expected to be significant). The correlation in Fig. 4d is therefore consistent with the relationships exhibited in Fig. 4a–c; that is, higher ¹⁸O is associated with higher extents of single-stage melting and/or higher degrees of depletion of the peridotitic sources of arc-related basalts as a result of previous melting events.

	DISCUSSION

TOP ABSTRACT INTRODUCTION SAMPLES AND ANALYTICAL METHODS RESULTS DISCUSSION SUMMARY AND CONCLUSIONS APPENDIX REFERENCES

 
Key results of this study are that variations in ¹⁸O among arc-related lavas, although unusual, are associated with known trace element and radiogenic isotopic tracers of slab input to their sources and, more importantly, they are correlated with chemical indices that are sensitive to the integrated extent of melting of those sources. In the following discussion we develop a model for the petrogenesis of arc-related lavas that both explains this observation and is consistent with the other known characteristics of these lavas. Much of this discussion is biased toward explanation of the relationship between ¹⁸O and other characteristics of lavas from the Vanuatu–Fiji–New Caledonia region because this is the suite of samples covering the full range of ¹⁸O observed in this study. However, the samples from the other suites fall on the same overall trends, suggesting that the model we develop could apply more broadly; more work will be needed to define whether these other suites (as well as suites not considered in this study) can be understood in the framework of the same model.

Oxygen isotope shifts associated with melting and magmatic evolution
Fractionations of oxygen isotopes among coexisting minerals and melts are small at the temperatures of peridotite melting and basaltic volcanism (Chiba et al., 1989; Palin et al., 1996; Fig. 2), but it is possible that even small fractionations could produce the 1 variations in ¹⁸O observed for olivine in this study by Rayleigh fractionation during melting or crystallization. This possibility can be tested using experimentally known and empirically estimated high-temperature fractionations of oxygen isotopes among minerals and melts.

Crystallization of olivine ± pyroxene will raise both the ¹⁸O values (Chiba et al., 1989; Matthews et al., 1998; Fig. 2) and TiO₂ contents of residual liquids, and therefore should produce trends opposite to that observed in Fig. 4a. Moreover, the correction of TiO₂ contents to a common MgO content should minimize the effects of fractional crystallization, and thus the variations in TiO_2(8·0) and the correlation with ¹⁸O shown in Fig. 4b would be difficult to account for by this process. Crystallization of Cr-rich spinel as part of a fractionating assemblage is expected to lead to progressive decreases in the cr-number of spinel phenocrysts in most circumstances (Arai, 1994); therefore, fractional crystallization can produce negative correlations between ¹⁸O and the cr-number of spinels but not trends with positive slope such as that shown in Fig. 4c for arc-related lavas. Fractionation of clinopyroxene rapidly removes Sc from magmas, such that Yb/Sc ratios in low-MgO lavas can be high (>0·1, based on data reported in the references in the ‘Samples and analytical methods’ section for evolved lavas in the arcs we have studied). However, the MgO and Sc concentrations of the samples defining the trend in Fig. 4d (average 8 wt % and 35 ppm) are substantially higher than those typical of such fractionated magmas (<5 wt % and <25 ppm). In any case, fractionation of clinopyroxene is expected to produce a positive trend in Fig. 4d rather than the negative one that is observed (i.e. because it raises both the ¹⁸O and Yb/Sc of residual magma). The relationships observed in Fig. 4 are therefore at odds with the expected effects of fractional crystallization, and we conclude that such processes are not major factors in the observed covariations between ¹⁸O and other geochemical and petrological indicators.

The difference in ¹⁸O between basaltic melt and a residual peridotitic mineral assemblage during mantle melting is expected to be a small positive number (0·2; Chiba et al., 1989; Matthews et al., 1998; Fig. 2), and thus even extensive fractional melting of peridotite will produce depleted residues that are only slightly lower in ¹⁸O than initial fertile sources (e.g. 20% fractional melting lowers ¹⁸O of the residue by only 0·04). This expectation is confirmed by analysis of olivine and pyroxene from highly depleted harzburgites of the Voykar massif, Polar Urals: these rocks are thought to be residual to 15% fractional melting (Sharma et al., 1997), yet they contain olivines that are indistinguishable in ¹⁸O from those in average fertile mantle peridotite (Eiler et al., 1996). Based on this and the predicted effects of melting on elemental abundances described above (i.e. residues will have lower TiO₂ and Yb/Sc, and their spinels will have higher cr-number), melting alone could therefore only produce correlations in Fig. 4 that are opposite in slope from those observed (and in any case so steep so as to appear vertical at the scale plotted). We therefore conclude that equilibrium partitioning of oxygen isotopes during simple batch or fractional fusion of peridotite cannot produce the trends in Fig. 4.

Given that fractionation of oxygen isotopes during melting or crystallization cannot explain the observed relationships between oxygen isotopes and sample chemistry, variations in ¹⁸O observed in this study are most simply interpreted in terms of mixing of two or more isotopically distinct reservoirs. High-level assimilation of oceanic crust and/or the island-arc volcanic edifice has been suggested as a way to introduce ¹⁸O-enriched material into mantle-derived magmas (Harmon et al., 1981; Hildreth & Moorbath, 1988; Davidson & Harmon, 1989). Although the effects of assimilation are widespread in arc-related environments, particularly in evolved magmas from arcs with thick crustal sections of continental rocks or sediments, it is difficult to explain by such a mechanism the relationships between ¹⁸O and sample chemistry observed in this study. In particular, this explanation would require preferential addition of crustal assimilants to lavas that are relatively high-degree mantle melts and/or are derived from relatively depleted mantle sources; moreover, such a process is not expected to generate the relationship between ¹⁸O and ⁸⁷Sr/⁸⁶Sr shown in Fig. 3a (i.e. assimilation of crustal rocks generally leads to more linear trends in these dimensions; Davidson & Harmon, 1989); finally, assimilation–fractional crystallization (AFC) processes generally produce the largest deviations in ¹⁸O in the most fractionated samples (DePaolo, 1981; Taylor, 1986) and therefore would produce trends generally opposite those observed in Fig. 4 (for reasons similar to those discussed above in reference to fractional crystallization). We conclude that although it would be possible to construct models of assimilation processes consistent with the trends in Figs 3 and 4, such models would contrast with the typical behavior of systems influenced by AFC processes.

Fluxed melting—a model for the interactions between slab-derived fluid or melt and the sub-arc mantle
A possible explanation for the correlations in Figs 3 and