Trace element abundances of high-mgo glasses from Kilauea, Mauna Loa and Haleakala volcanoes, Hawaii

Transcription

1 Contrib Mineral Petrol (1998) 131: 13±21 Ó Springer-Verlag 1998 T.P. Wagner á D.A. Clague á E.H. Hauri á T.L. Grove Trace element abundances of high-mgo glasses from Kilauea, Mauna Loa and Haleakala volcanoes, Hawaii Received: 4 February 1997 / Accepted: 27 August 1997 T.P. Wagner (&) 1 á T.L. Grove Department of Earth, Atmospheric and Planetary Science, Massachusetts Institute of Technology, Cambridge, MA 02139, USA D.A. Clague 2 USGS, Hawaiian Volcano Observatory, Hawaii National Park, Hawaii E.H. Hauri Department of Terrestrial Magnetism, 5241 Broad Branch Rd, NW, Washington, DC Present address: 1 Department of Geology, University of Papua New Guinea, Box 414 University PO, NCD, Papua New Guinea, @compuserve.com 2 Monterey Bay Research Aquarium, P.O. Box 628, 7700 Sandholdt Road, Moss Landing, CA , USA Editorial responsibility: J. Hoefs Abstract We performed an ion-microprobe study of eleven high-mgo (6.7±14.8 wt%) tholeiite glasses from the Hawaiian volcanoes Kilauea, Mauna Loa and Haleakala. We determined the rare earth (RE), high eld strength, and other selected trace element abundances of these glasses, and used the data to establish their relationship to typical Hawaiian shield tholeiite and to infer characteristics of their source. The glasses have trace element abundance characteristics generally similar to those of typical shield tholeiites, e.g. L(light)REE/ H(heavy)REE C1 > 1. The Kilauea and Mauna Loa glasses, however, display trace and major element characteristics that cross geochemical discriminants observed between Kilauea and Mauna Loa shield lavas. The glasses contain a blend of these discriminating chemical characteristics, and are not exactly like the typical shield lavas from either volcano. The production of these hybrid magmas likely requires a complexly zoned source, rather than two unique sources. When corrected for olivine fractionation, the glass data show correlations between CaO concentration and incompatible trace element abundances, indicating that CaO may behave incompatibly during melting of the tholeiite source. Furthermore, the tholeiite source must contain residual garnet and clinopyroxene to account for the variation in trace element abundances of the Kilauea glasses. Inversion modeling indicates that the Kilauea source is at relative to C1 chondrites, and has a higher bulk distribution coe cient for the HREE than the LREE. Introduction This study reports the trace element abundances of high- MgO tholeiite glasses from three Hawaiian volcanoes. These glasses are signi cant because some of them are nearly in Fe/Mg exchange equilibrium with mantle olivine, and thus these glasses provide direct constraints on the composition of primary magmas delivered to Hawaiian volcanoes. We determined the trace element abundances of these glasses by ion microprobe. We use the data to determine the relationship between high- MgO glasses and typical Kilauea and Mauna Loa shield lavas, and to gain insight into the tholeiite source. Description of glass grains All of the samples are fragments of tholeiitic basalt glass, 100 to 300 lm across. The Kilauea samples occur as sand grains in a turbidite deposit near Kilauea's submarine east rift zone, the Puna Ridge (Clague et al. 1991, 1995). The glasses are unique in that they have up to 15.0 wt% MgO, the highest magnesia contents reported for volcanic glass from Hawaii (Frey 1991). Microphenocrysts of Fo 90.7 olivine are present in some grains. The Mauna Loa and Haleakala glasses have moderately high MgO contents, 8 to 11 wt%. No phenocrysts are present in the grains. The Mauna Loa glasses were sampled from a sand core taken near Mauna Loa's

2 14 submarine southwest rift. These glasses are derived from both subaerially-erupted airfall and submarine-eruptions. Based on their location, we infer that the glasses are from Mauna Loa. If, however, the glasses were erupted from Kilauea, as some major element characteristics suggest (D. A. Clague, unpublished data 1997), our conclusions would not change. The Haleakala glasses were sampled from an Mn-cemented sandstone sampled by dredge #43 of Moore et al. (1990). Analytical methods Glasses were mounted in polished thin sections and analyzed for trace element abundances (Table 1) by secondary ion mass spectrometry using the Cameca IMS 3f ion microprobe at the Woods Hole Oceanographic Institution. Rare earth elements (REE: La, Ce, Nd, Sm, Eu, Dy, Er, and Yb) and Hf were analyzed on a single schedule using a 1- to 2-nA primary beam of O ) ions focused to a spot size of 20 to 30 lm. Other trace elements (Sc, Ti, V, Cr, Sr, and Zr) were analyzed on a separate, single schedule using a 0.2-nA primary beam, focused to a spot size of 8 to 15 lm. Positive secondary ions were collected and counted by an electron multiplier. Molecular interferences were excluded by energy ltering using 10V energy window, a )35 to )60 V o set for REEs, and a )90 V o set for other trace elements (Shimizu and Hart 1982). One sigma uncertainties based on counting statistics are approximately 4±7% relative for the REE, with the exception of Eu which has uncertainties of 10±14% relative due to the correction factors applied for BaO interferences. In addition, Er has NdO, SmO and BaO 2 interferences which can result in positive anomalies. Neither the Eu or Er data were used in the subsequent modeling. The other trace elements have uncertainties based on counting statistics of < 2% relative, with the exception of Nb and Rb which have uncertainties of 10% relative due to their low abundances. Kilauea glass sample KL-2 (Newsom 1986) was used as a standard. Replicate analyses of KL-2 and high-mgo Kilauea glass 57-7 show precision similar to that calculated from counting statistics. Each Kilauea glass was analyzed at 3 to 4 separate points for both REE and other trace element schedules. The Mauna Loa and Haleakala glasses were generally analyzed at 1 to 2 points for each schedule. Results for each glass were averaged for reporting in Table 1. All glasses are homogeneous, and precision of the glass averages were within counting statistic uncertainties. Results The trace element abundances of the glasses are similar to those of tholeiitic shield lavas from Kilauea and Mauna Loa (Fig. 1). The glasses display variable enrichments in the light (L) REEs and other less compatible elements relative to the heavy (H) REEs, as is Fig. 1 Comparison of Kilauea glass trace element abundances to published Kilauea and Mauna Loa abundances. Glass data from Table 1. Abundances normalized to C1 chondrite of Anders and Grevesse (1989). Data for Kilauea and Mauna Loa elds as follows: Kilauea eld from BVSP (1981) and Hofmann et al. (1984); Mauna Loa eld from BVSP (1981) and Budahn and Schmitt (1985) Table 1 Rare earth element and other trace element abundances of high-mgo glasses from various Hawaiian volcanoes; MgO in wt% from Clague et al. (1995) or D.A. Clague, unpublished data. All other elemental abundances in parts per million Sample # MgO La Ce Nd Sm Eu Dy Er Yb Rb Nb Sr Zr Hf Ti Sc V Cr Kilauea glasses 57± ± ± ± ± ± ± ± ± ± ± ± ± ± Mauna Loa glasses S6/T ± T ± T ± ± ± ± ± ± ± ± ± T ± ± ± ± ± ± ± ± ± Haleakala glasses 3a ± b ± ± ± ±

3 15 characteristic of shield tholeiites. On a more detailed level, however, the glasses show trace element abundance characteristics that vary between the values observed for Kilauea and Mauna shield tholeiites. The slopes of the REE patterns, as measured by the Ce/Sm and Ce/Yb ratios, can be used to discriminate between Kilauea and Mauna Loa shield lavas. The glass Ce/Sm and Ce/Yb ratios (Fig. 2) are of similar magnitude to those of the shield lavas, but vary between the Kilauea and Mauna Loa elds. The Kilauea glasses range from low, Mauna Loa-like values to higher, Kilauea-like values. At a given MgO content (12 wt%), the Kilauea glasses show signi cant variation in both of these ratios. In addition, the Kilauea glasses show some correlation of these ratios with MgO. The Mauna Loa glasses vary between the Kilauea and Mauna Loa elds in Ce/Sm, but are Mauna Loa-like in Ce/Yb. The Haleakala glasses are in the Mauna Loa eld for both ratios. The Sr/Nb, Zr/Nb and Rb/Sr ratios can also be used to discriminate between Kilauea and Mauna Loa shield lavas (Frey and Rhodes 1993). Again, the magnitude of Fig. 2 Ratios of REE showing di erences between Kilauea and Mauna Loa. Glass data from Table 1. 1 sigma error bars, 14% relative. Error bars are left o all glasses except Mauna Loa in the interest of clarity. References as in Fig. 1 Fig. 3 Trace element ratios plotted against MgO in wt%. Glass data from Table 1. Data for Kilauea and Mauna Loa as in Fig. 1 except for Kilauea/Mauna Loa dividing lines which are approximated from Frey and Rhodes (1993), and Puna Ridge lavas from Clague et al. (1995). 1 sigma error bars

4 16 these ratios for the glasses is similar to that observed for the shield lavas. The Kilauea glasses have ratios of these elements that are generally similar to those of the Kilauea shield lavas (Fig. 3), with the exception of and 57-27, which are at the Kilauea/Mauna Loa boundary due to their low-nb contents. The Mauna Loa glasses have Kilauea-like Sr/Nb and Zr/Nb, but have Mauna Loa-like Rb/Sr. The Haleakala glasses have Kilauea-like Sr/Nb and Zr/Nb, but are split between the elds in Rb/Sr. Discussion Characteristics of tholeiite primary magmas The high-mgo glasses have trace element abundance characteristics similar to the more abundant tholeiitic lavas that make up the Hawaiian shields (Figs. 1±3). We infer that the glass magmas formed by processes similar to those that produce the shield lavas. The glasses' high- MgO nature, however, indicates that glass magmas escaped much of the shallow level processing that a ects the magmas that produce shield lavas. The glasses can thus be used more directly to constrain the primary tholeiite composition, and to infer characteristics of the tholeiite source. The variable MgO contents (Table 1) and FeO/MgO ratios of the glasses imply that the glass magmas underwent some olivine fractionation since segregating from the mantle. Mixing of primitive and evolved liquids could also account for this variation, however, such mixing would fractionate the Sm/Sr and Ti/Eu ratios, sensitive indicators of plagioclase and Fe-Ti oxide crystallization. These ratios are nearly constant for the Kilauea glasses, and indicate that mixing has not occurred. Moreover, only olivine microphenocrysts are found in the glasses, and Clague et al. (1995) show that the Kilauea glasses follow olivine fractionation curves. Other phases are unlikely to have crystallized from the glass magmas, as Kilauea tholeiite with >7:5 wt% MgO has been shown to only crystallize olivine (Helz 1987; Helz and Thornber 1987). Fractionation correction We corrected the major element concentrations of all the glasses, except the two lowest in MgO, to their prefractionation levels (Table 2) by olivine addition calculations using the program MORBFRAC (Grove et al. 1992). Glass compositions were corrected to be in Fe/ Mg equilibrium with Fo 90.7 olivine, the most Fo-rich microphenocryst found in the glasses. Major element concentrations of glasses were normalized to 100%, volatile free. The Fe 2+ /Fe Total was assumed to be 0.9, based on the wet chemical determination for the glassy rim of a Puna Ridge lava (Moore 1965). Olivine was added in 0.25 wt% increments as stoichiometric (Mg, Fe) 2 SiO 4. The Fe/Mg ratio of the added olivine was recalculated at each step to maintain equilibrium with the liquid, assuming an exchange K Fe Mg D of 0.3 (Roeder and Emslie 1970). The glasses required the addition of 5±24 wt% olivine to equilibrate with Fo This calculation does not take into account the potential di erences in Fe/Mg ratio or f O 2 in the magma sources for each volcano. However, the variation caused by these di erences is likely to be smaller than the variation caused by crystallization, and considering that our simple olivine addition calculation reduces much of the variation in glass major element data, our assumptions seem justi ed. Correlation of REE ratios and MgO. Since the fractionation corrected glasses all have similar MgO contents, much of the correlation of the Ce/Yb and Ce/Sm ratios with MgO (Fig. 2) disappears after correction. The two Kilauea glasses lowest in MgO, which could not Table 2 Major element abundances corrected to be in equilibrium with Fo 90.7 olivine. Original analyses for Kilauea glasses are reported in Clague et al. (1995). Dilution factor equals N, where N is the ``No. steps'' of olivine addition Sample # No. steps % Dilution SiO 2 Al 2 O 3 TiO 2 FeO T P 2 O 5 MgO CaO Na 2 O K 2 O Kilauea glasses 57± ± ± ± ± ± ± Mauna Loa glasses S Haleakala glasses 3a b

5 17 be corrected accurately for fractionation, do have the highest values for the Ce/Sm and Ce/Yb ratios. It is likely, however, that these two glasses are not petrogenetically related to the more magnesian glasses since they are Kilauea shield-like in all trace element ratios (Figs. 2, 3), and we conclude that the correlation of the REE ratios with MgO is probably fortuitous. We argue below that the variation in REE ratios observed in the fractionation corrected glasses is produced by the mantle melting process. Composition of primary tholeiite The proposed MgO contents of primary Hawaiian tholeiite have ranged from 7 to >20 wt% (review in Eggins, 1992). When corrected for olivine fractionation, the glasses show that some Kilauea, Mauna Loa and Haleakala primary magmas have 16.6 to 17.6 wt% MgO (Table 2). The complete set of Kilauea glasses, studied by Clague et al. (1995), extends this range from 13.4 to 18.4 wt% MgO. The SiO 2 contents of the primary magma estimates are remarkably uniform, and almost all of them fall between 48±49 wt% SiO 2. This SiO 2 content is higher than that of experimentally produced melts of garnet lherzolite, and Wagner and Grove (this issue) suggest that these high-sio 2 contents re ect the reaction of primary tholeiite with harzburgite prior to segregation from the mantle. Blended chemical characteristics. The glasses also show that magmas with a blend of compositional characteristics are delivered to both the Kilauea and Mauna Loa magma systems, in contrast to studies of shield lavas that have shown distinct di erences between volcanoes (e.g., Frey and Rhodes 1993). The Kilauea glasses have Ce/Yb ratios similar to Mauna Loa shield lavas (Fig. 2), but are Kilauea-like for the other studied ratios (Figs. 2, 3). Kilauea shield lavas with similar chemical characteristics to the glasses are rare, but not unknown (Rhodes et al. 1989). The Mauna Loa glasses show variation similar to the Kilauea glasses. They are similar to the Kilauea glasses for all ratios except Rb/Sr, which is the only intervolcano discriminant based on shield lavas that holds for the glasses. Kilauea-like characteristics have also been observed in lavas from Mauna Loa's southwest rift (Garcia et al. 1995; Kurz et al. 1995) and some older lavas from the northeast rift (Hauri et al. 1996). Some overlap is also shown in glass major element concentrations (Fig. 4). Mauna Loa shield lavas are known to have distinctly lower K 2 O and CaO than Kilauea shield lavas (Frey and Rhodes 1993). When corrected for olivine fractionation, the Mauna Loa glasses similarly have lower K 2 O contents than the Kilauea glasses, but have higher CaO (Fig. 4). Source constraints. A straightforward contribution of magmas from one shield's source to the other volcano, as suggested by Rhodes et al. (1989), cannot account for Fig. 4 Major element abundances corrected for olivine fractionation plotted against Ce/Yb ratio. Arrows are schematic, representing general trends and pointing in direction of increasing degree of melting for the Ce/Yb ratio. 1 sigma error bars, left o all glasses except Mauna Loa in the interest of clarity the blend of characteristics observed. However, a complexly zoned source, as envisioned by Hauri et al. (1996), might be able to produce such magmas. These hybrid magmas may be small volume events, whose signatures are simply overwhelmed by contributions of the more distinct magma source. Eruption of these magmas may depend on the position of the volcano relative to the zoned source, and may explain why these hybrid magmas are found primarily associated with rift eruptions, rather than summit eruptions. Characteristics of the tholeiite source Incompatible behavior of CaO The fractionation corrected CaO contents of the high- MgO glasses are negatively correlated with the Ce/Yb ratio (Fig. 4). Assuming the Ce/Yb ratio is an indicator of degree of melting, CaO behaved like an incompatible element during formation of these magmas. Lime is known to behave compatibly during melting of spinel lherzolite (Kinzler and Grove 1992), hence this e ect was not produced by spinel- eld melting. The behavior of CaO during melting of garnet lherzolite is not well characterized (Wagner and Grove, this issue). However, if we assume these magmas segregated from a garnet lherzolite resid-

6 18 uum, as we discuss the evidence for below, these results imply that CaO behaves incompatibly during garnet lherzolite melting. This inference is also supported by Frey and Rhodes (1993), who showed that Kilauea shield tholeiites have higher incompatible element abundances and CaO contents than Mauna Loa shield tholeiites. Alternatively, some of this correlation could re ect heterogeneity in source composition (Hauri 1996). Mineralogy of the Kilauea tholeiite source Evidence for residual garnet. After correction for olivine fractionation, the Kilauea glasses show variation in their LREE abundances and nearly constant HREE abundances. To illustrate this variation, and for the inversion modeling discussed below, we plotted the REE data on a process identi cation diagram (Treuil and Joron 1975), using La as the highly incompatible element (Fig. 5). Rubidium is a better choice for this diagram from the point of incompatibility, however the analytical uncertainties on the Rb analyses are much higher than those for La. Interelement diagrams (Hanson 1989) identi ed La as the most incompatible element of the remaining choices. Samples related through partial melting should form linear trends in Fig. 5 (Treuil and Joron 1975; Minster and Allegre 1978), and the Kilauea glass data satisfy this criterion. The regression line for the Ce data (Table 3) has a very shallow slope, due to the similar, 25% relative variation in abundance displayed by both Ce and La. The Dy and Yb regressions, however, have very steep slopes due to the less than 8% relative variation in Dy and Yb abundances. The variation in Dy and Yb abundances is almost within analytical uncertainty. Variable LREE abundances with non-varying HREE abundances in cogenetic samples have been observed and discussed previously (e.g. Shimizu and Arculus 1975; Frey et al. 1980; Hofmann et al. 1984). The conclusion reached in all of these studies is that the variations in LREE abundances re ect variations in the degree of melting, while the lack of variation in the HREE abundances re ect their high compatibility in the source due to the presence of residual garnet. Frey et al. (1980) and Hofmann et al. (1984) showed that accessory phases with high distribution coe cients (D) for the HREE, such apatite and amphibole, cannot produce the observed variations. Fig. 5 Process identi cation diagram similar to that of Treuil and Joron (1975). La abundance on the horizontal axis is corrected for olivine fractionation using dilution factors from Table 2. Vertical axis normalized to C1 chondrite of Anders and Grevesse (1989). 1 sigma error bars Evidence for residual clinopyroxene. The Kilauea glasses also show evidence for residual clinopyroxene in their source. The Sm/Sr ratio of the Kilauea glasses is nearly constant while Sm and Sr each show variation in abundance (Fig. 6). Element pairs that maintain constant ratios, while individually showing variation in abundance, have partition coe cients that are either identical or negligibly small for the residual mineralogy of the source (Hofmann et al. 1984). The Sr and Sm abundances are not correlated with the highly incompatible elements, which implies that the controlling phase must have at least moderately high and similar Ds for these elements. This constraint rules out olivine and Table 3 Linear regression results for Kilauea glasses. Linear regressions for lines in Fig. 5; r 2 is the sum of the squared residuals Regression line Slope SE Intercept SE r 2 Ce ) Nd Sm Dy Yb ) Nb ) Sr Zr Ti

7 19 Fig. 6 Sm/Sr plotted against fractionation corrected Sr abundance in ppm for high-mgo Kilauea glasses. The Sr abundance corrected for fractionation using dilution factors from Table 2. Error bars are 5% based on uncertainty in ion-probe analyses orthopyroxene, which have very low Ds for Sr and Sm. The D for Sr in garnet is also very low, 0:01±0:02, and results in D Sm /D Sr of 65±111 (Gill 1981; Hauri et al. 1994), too high to bu er the Sm/Sr ratio. Clinopyroxene, on the other hand, has a D for Sr of 0.16, signi cantly higher than in garnet, and results in a D Sm /D Sr of 3 (Hofmann et al. 1984; Hauri et al. 1994). Clinopyroxene, then, is the only phase that could control the Sm/Sr ratio and must be in the source residue of the high-mgo Kilauea glass magmas. Inversion models Trace element abundances of the tholeiite source The Kilauea glass data were inverted to constrain their source abundances and partition coe cients relative to La using the techniques of Minster and Allegre (1978), as simpli ed by Hofmann and Feigenson (1983). These equations assume batch equilibrium partial melting, which may not be valid for melt generation in the Hawaiian plume, where fractional melting is likely to occur. However, the treatment does provide a simple way of inverting the data, and allows us to compare results with previous inversion models of Kilauea trace element data that employed similar techniques (e.g. Hofmann et al. 1984). The y-intercept of the each element's regression line (Fig. 5, Table 3) determines its abundance in the source relative to La, using Eq. (7 ) of Hofmann and Feigenson (1983): C i 0 =CH 0 ˆ 1 Pi =I i where C 0 is the source concentration of the element (i denotes the element of interest; H denotes the highly incompatible element, which is La in this case), P is the weighted partition coe cient of the melting reaction, and I is the y-intercept value of the regression line. I is a near zero value for the HREE regression lines, and this result greatly magni es uncertainties in the determined values of P. Hofmann et al. (1984) solved for P iteratively by assuming the source had a smooth pattern for the HREE; then, they solved for the melt reaction from P assuming: (1) only clinopyroxene (cpx) and garnet are involved in the melt reaction; (2) the reaction involves the maximum amount of cpx consistent with the P. Hofmann et al. calculate that cpx and garnet enter the melt in a ratio of 81:19. The uncertainties on our I values for Dy and Yb preclude our solving for P using this same technique. Instead, we used the Hofmann et al. melt reaction to get an initial estimate of the source abundances. The light and middle REE form a smooth, at pattern at 1, similar to the pattern shown in Fig. 7. If we assume that the relatively at pattern of the LREE supports a at overall source pattern, we can assume that the source has C0 Yb=CLa 0 ˆ 1. Calculation of P from the above equation using I ˆ 0:017 yields a P value of Using the same assumptions as Hofmann et al., this value of P constrains the melt reaction to 85 cpx:15 garnet. The inversion results plotted in Fig. 7 use this melt reaction. Experimental studies indicate that melting of garnet lherzolite involves a reaction relationship of garnet+cpx+olivine melting to produce orthopyroxene+liquid (Herzberg et al. 1990). In the garnet lherzolite melting reaction of Kinzler (1992), cpx and garnet enter the melt in nearly equal proportions. This reaction, however, was determined near the spinel to garnet transition. The much higher ratio of cpx to garnet entering the melt estimated in our inversion could re ect the e ect of increasing pressure on the mantle melting Fig. 7 La-normalized source concentrations and bulk distribution coe cients after inversion. Concentrations in mantle source are normalized to source concentration of La and C1 chondrite. Solid line shows best value based on inversion. Dashed lines are maximum and minimum values based on uncertainties in regression coe cients (Table 3). The Dy is not plotted because inversion equations became unde ned for uncertainty in regression coe cients. Bulk distribution coe cients of source are also normalized to source La concentration. Uncertainty discussed in text

8 20 reaction, and is similar to the melt reaction of Walter (1997) for garnet lherzolite melting at 5 GPa. Our inversion shows that the Kilauea source has chondritic relative abundances of the REEs. Other inversions of the REE abundances of Hawaiian tholeiites (e.g., Albarede 1983; Feigenson et al. 1983; Hofmann et al. 1984) had similar results, and concluded that the tholeiite source has a at to slightly LREE enriched REE pattern relative to chondrites. This result, however, is inconsistent with the Nd isotopic data from Hawaiian volcanics (e.g., DePaolo and Wasserburg 1976; O'Nions et al. 1977; Chen and Frey 1985) and speci cally from Kilauea tholeiites (e.g., Hofmann et al. 1984; Stille et al. 1986). These studies have shown the Nd isotopes re ect derivation from sources with time averaged Sm/Nd ratio greater than the chondritic value, supporting a LREE depleted source. This contradiction has generally been interpreted to mean that the source has recently been enriched in highly incompatible elements (review in Frey and Roden, 1987). Bulk distribution coe cients of the tholeiite source The La-normalized distribution coe cients were calculated using Eq. (10) of Hofmann and Feigenson (1983): D i 0 =CH 0 ˆ Si 1 P i =I i where D i 0 is the bulk partition coe cient for element i in the source, and S i is the slope of the regression line from Fig. 5 (other variables as in previous equation). The results of the inversions (Fig. 7) show a distinct increase in D Dy and D Yb relative to the other elements. The uncertainties on this inversion are greater than the small spikes displayed by adjacent elements, i.e. D Dy > D Yb and D Sm > D Eu are probably not real. The uncertainties are, however, smaller than the large increase in D for the HREEs relative to the lighter REEs. The higher Ds for the HREE are consistent with the interpretation that garnet is a residual phase in the glass magma source. Acknowledgements T.W. thanks F.A. Frey for helpful direction and reviews of this paper in various stages. H.J. Yang reviewed earlier versions of the manuscript, and discussions with him greatly improved it. N. Shimizu helped analyze some of the glasses, and reviewed an earlier draft of the manuscript. Thanks also go to A. Brandon and W. McDonough for helpful discussion. This paper was formally reviewed by S. Eggins and G. Sen, both of whose comments were very helpful. This work was supported by the US National Science Foundation. References Albarede F (1983) Inversion of the batch melting equations and the trace element pattern of the mantle. J Geophys Res 88: 10573± Anders E, Grevesse N (1989) Abundances of the elements: meteoritic and solar. Geochim Cosmochim Acta 53: 197±214 Budahn JR, Schmitt RA (1985) Petrogenetic modeling of Hawaiian tholeiitic basalts: a geochemical approach. Geochim Cosmochim Acta 49: 67±87 BVSP (1981) Basaltic volcanism on the terrestrial planets. Pergamon Press Inc, New York Chen C, Frey FA (1985) Trace element and isotopic geochemistry of lavas from Haleakala volcano, East Maui, Hawaii: implications for the origin of Hawaiian basalts. J Geophys Res 90(B10): 8743±8768 Clague DA, Weber WS, Dixon JE (1991) Picritic glasses from Hawaii. Nature 353: 553±556 Clague DA, Moore JG, Dixon JE, Friesen WB (1995) Petrology of submarine lavas from Kilauea's Puna Ridge, Hawaii. J Petrol 36(2): 299±349 DePaolo DJ, Wasserburg GJ (1976) Inferences about magma sources and mantle structure from variations of 143Nd/144Nd. Geophys Res Lett 3: 743±746 Eggins SM (1992) Petrogenesis of Hawaiian tholeiites. 1. Phase equilibria constraints. Contrib Mineral Petrol 110: 387±397 Feigenson MD, Hofmann AW, Spera FJ (1983) Case studies on the origin of basalt II. The transition from tholeiitic to alkalic volcanism on Kohala volcano, Hawaii. Contrib Mineral Petrol 84: 390±405 Frey FA (1991) Mantle yields its secrets. Nature 353: 500 Frey FA, Rhodes JM (1993) Intershield geochemical di erences among Hawaiian volcanoes: implications for source compositions, melting processes and magma ascent paths. Philos Trans R Soc London 342: 121±136 Frey FA, Roden MF (1987) The mantle source for the Hawaiian islands: constraints from the lavas and ultrama c inclusions. In: Menzies M, Hawkesworth C (eds) Mantle metasomatism. Academic Press, London, pp 423±463 Frey FA, Roden MF, Zindler A (1980) Constraints on mantle source compositions imposed by phosphorous and the rareearth elements. Contrib Mineral Petrol 75: 165±173 Garcia MO, Hulsebosch T, Rhodes JM (1995) Olivine-rich submarine basalts from the southwest rift zone of Mauna Loa volcano: implications for magmatic processes and geochemical evolution. In: Rhodes JM, Lockwood JP (eds) Mauna Loa revealed: structure, composition, history and hazards. Am Geophys Union, Washington, DC, pp 219±239 Gill JB (1981) Orogenic andesites and plate tectonics. Springer- Verlag Berlin Heidelberg New York Grove TL, Kinzler RJ, Bryan WB (1992) Fractionation of midocean ridge basalt (MORB). In: (eds). Mantle ow and melt generation at mid-ocean ridges. Phipps Morgan J, Blackman DK, Sinton JM Am Geophys Union, Washington, DC, pp 361 Hanson GN (1989) An approach to trace element modeling using a simple igneous system as an example. In: Lipin BR, McKay GA (eds) Geochemistry and mineralogy of the rare earth elements. Mineral Soc Am, Washington, DC, pp 79±97 Hauri EH (1996) Major element variability in the Hawaiian mantle plume. Nature 382: 415±419 Hauri EH, Wagner TP, Grove TL (1994) Experimental and natural partitioning of Th, U, Pb and other trace elements between garnet, clinopyroxene and basaltic melts. Chem Geol 117: 149± 166 Hauri EH, Lassiter JC, DePaolo DJ (1996) Osmium isotope systematics of drilled lavas from Mauna Loa, Hawaii. J Geophys Res 101(B5): 11793±11806 Helz RT (1987) Di erentiation behavior of Kilauea Iki lava lake, Kilauea volcano, Hawaii. In: Mysen BO (ed) Magmatic processes: physiochemical principles. The Geochem Soc, 241±258 Helz RT, Thornber CR (1987) Geothermometry of Kilauea Iki lava lake, Hawaii. Bull Volcanol 49: 651±668 Herzberg CT, Gasparik T, Sawamoto H (1990) Origin of mantle peridotite: constraints from melting experiments to 16.5 GPa. J Geophys Res 95(B10): 15779±15803 Hofmann AW, Feigenson MD (1983) Case studies on the origin of basalt. I. Theory and reassessment of the Grenada basalts. Contrib Mineral Petrol 84: 382±389 Hofmann AW, Feigenson MD, Raczek I (1984) Case studies in the origin of basalt. III. Petrogenesis of the Mauna Ulu eruption, Kilauea, 1969±1971. Contrib Mineral Petrol 88: 24±35

9 21 Kinzler RJ (1992) Mantle melting processes at the spinel-garnet transition (17±21 Kbar). EOS Trans Am Geophys Union 73(43): 615 Kinzler RJ, Grove TL (1992) Primary magmas of mid-ocean ridge basalts. 2. Applications. J Geophys Res 97(B5): 6907±6926 Kurz MD, Kenna TC, Kammer DP, Rhodes JM, Garcia MO (1995) Isotopic evolution of Mauna Loa volcano: a view from the submarine southwest rift. Rhodes JM, Lockwood JP (eds) Mauna Loa revealed: structure, composition, history and hazards. Am Geophys Union, Washington, DC, pp 219±239 Minster JF, Allegre CJ (1978) Systematic use of trace elements in igneous processes. III. Inverse problem of batch partial melting in volcanic suites. Contrib Mineral Petrol 68: 37±52 Moore JG (1965) Petrology of deep-sea basalt near Hawaii. Am J of Sci 263: 40±52 Moore JG, Clague DA, Ludwig KR, Mark RK (1990) Subsidence and volcanism of the Haleakala Ridge, Hawaii. J Volcanol Geothermal Res 42: 273±284 Newsom HE (1986) Oceanic basalts. Earth Planet Sci Lett 80: 299 O'Nions RK, Hamilton PJ, Evensen NM (1977) Variations in 143Nd/144Nd and 87Sr/86Sr ratios in oceanic basalts. Earth Planet Sci Lett 34: 13±22 Rhodes JM, Wenz KP, Neal CA, Sparks JW, Lockwood JP (1989) Geochemical evidence for invasion of Kilauea's magmatic plumbing system by Mauna Loa magma. Nature 337: 257±260 Roeder PL, Emslie RF (1970) Olivine liquid equilibrium. Contrib Mineral Petrol 29: 275±289 Shimizu N, Arculus RJ (1975) Rare earth element concentrations in a suite of basanitoids and alkali olivine basalts from Grenada, Lesser Antilles. Contrib Mineral Petrol 50: 231±240 Shimizu N, Hart SR (1982) Applications of the ion microprobe to geochemistry and cosmochemistry. Annu Rev Earth Planet Sci 10: 483±526 Stille P, Unruh DM, Tatsumoto M (1986) Pb, Sr, Nd, and Hf isotopic constraints on the origin of Hawaiian basalts and evidence for a unique mantle source. Geochim Cosmochim Acta 50: 2303±2319 Treuil M, Joron JL (1975) Utilisation des elements hygromagmatophiles pour la simpli cation de la modelisation quantitative des processes magmatiques: exemples de l'afar de la dorsale medio-atlantique. Soc Ital Mineral Petrol 31: 125± 174 Wagner TP, Grove TL (this issue) Melt/harzburgite reaction in the petrogenesis of tholeiitic magma from Kilauea volcano, Hawaii. Contrib Mineral Petrol Walter MJ (1997) Melting of garnet peridotite and the origin of komatiite and depleted lithosphere. Submitted to J Petrol