Geochemistry and Geochronology of the Society Islands: New Evidence for Deep Mantle Recycling

William M. White

Dept. of Geological Sciences, Cornell University, Ithaca, New York

Robert A. Duncan

College of Oceanic and Atmospheric Sciences, Oregon State Universty, Corvalis, Oregon

We report new K-Ar ages, Sr, Nd, and Pb isotope ratios and concentrations of 34 trace elements in volcanic rocks from the Society Islands. Maximum ages of volcanism at Mehetia, Tahiti, and Maupiti indicate that volcanism in the Society Islands progresses southeastward at a rate of 11 cm/yr, which is consistent with current estimates of Pacific Plate motion. Tahaa has the largest range in ages, 2.3 million years. This large range reflects the eruption of highly undersaturated lavas during a post- erosional phase that followed a volcanic hiatus of 1.2 million years. The entire range of isotope ratios observed in the Society Islands is nearly encompassed by the range observed on Tahaa: the two post- erosional lavas have among the most depleted isotopic signatures while the shield series lavas have the most enriched isotopic signatures. The isotopic and chemical evolution of Tahaa thus follows the Hawaiian pattern. There is a strong correlation between Sr and Nd isotope ratios and weaker correla- tions involving Pb isotope ratios. Compared to other oceanic islands, Sr and Nd ratios range to extremely high and low values respectively, but Pb isotope ratios are intermediate. The isotope correla- tions indicate the Society plume can be approximately modeled as consisting of two slightly hetero- geneous end members. All magmas are moderately to strongly enriched in light rare earth elements as well as other incompatible elements. After exclusion of weathered (as indicated by anomalous Ba/Rb ratios) and differentiated lavas, there are strong correlations between Hf/Sm, Pb/Ce, Zr/Nb, Nb/U, 87 86 ε 207 204 and Rb/Sr and Sr/ Sr (and Nd and Pb/ Pb) and weaker correlations between Nb/La and La/ 87 86 207 204 Sm and Sr/ Sr (and εNd and Pb/ Pb). This indicates the variation in Hf/Sm, Pb/Ce, Zr/Nb, Nb/ U, and Rb/Sr is largely due to variations in the composition of the source. This inference is confirmed by factor analysis. On the other hand, variance in the La/Sm ratio appears dominated by varying degrees of partial melting, perhaps because differences between the two end members in La/Sm are small. The high 87Sr/86Sr component in the plume has high 207Pb/204Pb and Pb/Ce and low Nb/U. These characteristics are consistent with this component consisting partly of sediment that has been subducted into the deep mantle. The low 87Sr/86Sr component in the Society plume appears to be also richer in incompatible elements than depleted mantle and is isotopically similar to the common com- ponent of plumes postulated by others. This component is most prevalent in late and very early stage Society magmas and thus probably constitutes the sheath of the Society plume. This sheath was most likely viscously entrained in the deep mantle during ascent of the plume.

INTRODUCTION • Mantle plumes are not characterized by a single com- position, each plume is somewhat unique [e.g., Sun, 1980; It now seems well established that oceanic island basalts Zindler et al., 1982]. are derived from mantle plumes. In the nearly 25 years • Despite this, plumes can be divided on the basis of their since the papers of Morgan [1971] and Schilling [1973] isotope compositions into just a few classes [White, 1985]. that introduced and popularized the concept, a limited un- Schilling [1973], and later Wasserburg and DePaolo [1977], derstanding of the physics and chemistry of mantle plumes proposed that plumes consisted of “primordial”, or primi- has been achieved. This understanding may be summa- tive mantle. Although many plumes contain some primor- rized as follows: dial He, we now recognize primitive mantle cannot be the • Mantle plumes differ fundamentally in composition primary constituent of any of these plume classes. from the depleted upper mantle, the presumed source of • Fluid dynamic studies have demonstrated that plume- mid-ocean ridge basalts [Schilling, 1973; Hart et al., 1973]. like features arise from thermal boundary layers in con- vecting systems that are heated from below. Thus mantle plumes must arise from a thermal boundary layer in the Copyright 1995 by the American Geophysical Union. deep mantle. The most likely candidates are the core– Reprinted from: Isotope Studies of Crust-Mantle Evolution, mantle boundary or, if the γ-olivine–perovskite transition an AGU Monograph honoring M. Tatsumoto and G. Tilton. ed- is a barrier to whole mantle convection, the 660 km seis- ited by A. Basu and S. R. Hart. In press. mic discontinuity. WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 2

• The plume heat flux is small compared to the heat loss and sediment. through plate creation, implying plumes are a second order We report on an isotopic and trace element study of eight feature of mantle convection [Davies, 1988]. Neverthe- of the Society Islands (Figure 1). Two of the islands, Ta- less, assuming the plume mass flux (presently about 2.5 × hiti and Tahaa, were selected for more detailed study. Ta- 1014 kg/yr) has been constant through geologic time, about hiti is the largest and one of the youngest of the Society 25% of the mass of the mantle has been cycled through Islands, and the results of our study are reported in Duncan plumes, hence plumes may have played a significant role et al. [1994]. We found that volcanic evolution there fol- in the chemical evolution of the Earth. If material resides lows the Hawaiian pattern in which late state magmas are in the thermal boundary layer for 109 yrs before being in- more alkalic and have less enriched isotopic signatures than corporated in plumes, then this boundary layer must con- magmas of the earlier shield building phase. In this report, stitute 6% of the mass of the mantle. This volume is an we show that post-erosional magmas from Tahaa, which order of magnitude larger than the continental crust. If in- were erupted after a hiatus of 1.3 Ma, have dramatically compatible element abundances in plumes are an order of lower Sr and Pb isotope ratios and higher Nd isotope ratios magnitude lower than in the continental crust, then the in- than magmas erupted during the main shield building stage. compatible element inventory in the plume source layer is We also show that Pb isotope and trace element ratio varia- comparable to that of the continents. tions, particularly Pb/Ce and Nb/U ratios, in the Society • The geochemistry of plumes, however, bears no im- Island magmas as a whole are consistent with the hypoth- print of their deep mantle origin. There is, for example, no esis that the high 87Sr/86Sr component in the Society plume evidence that plumes lost or gained siderophile elements consists in part of recycled marine sediment or continental through reaction with the core [Newsom et al., 1986]. There crust. The low 87Sr/86Sr component in the Society plume is no obvious imprint of trace element fractionation between may be the common component in plumes, which has been silicate liquid and deep mantle phases such as majorite or variously called FOZO [Hart et al.,1992], PHEM [Farley perovskite [Kato et al. 1987; Kato et al. 1988]. However, et al., 1992], and “C” [Hanan and Graham, 1994]. Since no studies have been specifically designed to test the latter. this latter component is most prevalent in early and late Because of the absence of this deep mantle imprint, stage lavas, we speculate that it constitutes the entrained geochemical studies have not contributed to resolving the sheath, whereas the recycled component, which dominates question of the depth from which plumes originate. the shield-building stage, constitutes the core of the plume. • Two principal theories have been proposed to describe how plumes may have acquired their peculiar geochemical ANALYTICAL METHODS features. First, Chase [1981] and Hofmann and White [1980, 1982] proposed that subduction carried oceanic crust Isotopic Analyses into the deep mantle, perhaps the core mantle boundary, where it is reheated and eventually risies buoyantly as To avoid sample contamination during grinding, rock mantle plumes. McKenzie and O’Nions [1983] proposed chips rather than powders were used for isotopic analysis. subcontinental lithosphere might sometimes delaminate, Two to three hundred mg of chips were weighed into teflon sink into the deep mantle, and eventually rising buoyantly beakers or capsules, and leached for approximately 30 min- as a mantle plume. Most current hypotheses for the geo- utes in 1 to 2 ml of hot 6N HCl prior to dissolution. The chemical evolution of plumes are variations on these two leaching process serves not only to remove Pb contamina- main ideas. Alternatively, delamination might remove and tion introduced during handling, but also to remove Sr con- “recycle” lower continental crust as well as mantle lithos- phere [Kay and Kay, 1991]. 153° 152° 151° 150° 149 148° 147° W The present study was undertaken with the specific ob- 15° 2000 m jective of testing the hypothesis that plumes, or at least some 3000 m of them, consist of deeply recycled oceanic crust and sedi- 16° 3000 m 4 ment. Earlier studies had shown that basalts from the So- Tupai 000 m Bora Bora ciety Islands have particularly radiogenic Sr [Duncan and Maupiti 4000 m Tahaa Hauhine Compston, 1976; Hedge, 1978]. White and Hofmann [1982] 17° Raiatea Tetiaroa 3000 m proposed on the basis of Sr and Nd isotope ratios that the Moorea 2000 m Teahetia Maiao Rocard Society mantle plume acquired its unique geochemical char- TahitiNui ° TahitiIti acteristics from subduction and recycling of ancient oce- 18 CyanaMehetia 4000 anic sediment. The work of Devey et al. [1990] on volcanics m Moua Pihaa from young seamounts of the Society chain provided some support of this idea, as some of these volcanics had high 19° S Pb/Ce ratios. Thus the Societies seemed a particularly Figure 1. Map of the Society Archipelago. Teahetia, Rocard, promising locality to test and investigate further the hy- Cyana, and Moua Pihaa are seamounts, all of which, along with pothesis that mantle plumes contain recycled oceanic crust the young island of Mehetia, are probably volcanically active. Tetiaroa and Tupai are atolls with no exposed basalt. WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 3 tamination, which has been observed in even very fresh dard was 0.710223. Based on the reproducibility of this oceanic island basalts and is probably derived from seasalt standard, the 2σ analytical uncertainty is estimated at [e.g., White et al., 1993]. Many samples analyzed in the ±0.000028. Purified Nd was loaded onto single Re fila- present study were subareally altered to varying degrees, ments with Biorad AG resin beads (which serves as a re- and this leaching also served to remove Sr in secondary ductant) and phosphoric acid and analyzed as Nd+ using a phases. Experiments showed, however, that leaching may dynamic multicollection technique. Correction for mass remove in excess of 50% of the Sr, Nd and Pb in basalts, fractionation was made by normalizing to 146Nd/144Nd thus relatively large sample quantities were used. =0.72190. The 145Nd/144Nd and/or the 150Nd/144Nd ratio All chemical procedures were carried out in a clean envi- were also monitored for data quality assurance. Measured ronment (Class 1000 to Class 100), and all reagents used mean 143Nd/144Nd for the La Jolla Nd standard was were purified through multiple subboiling distillation. 0.511852. Based on reproducibility of this standard, the Samples were digested for 2 to 4 hours with 24N HF. Fol- 2σ analytical uncertainty of the 143Nd/144Nd ratios is esti- lowing that, approximately 2 ml of 6N HBr were added mated at ±0.000016. and the sample evaporated to dryness. The sample was taken up in 3 ml 0.5 M HBr, centrifuged, and the supernate Trace Element Analyses loaded on an ion exchange column consisting of 50 µl of BioRad AG1-X8 (100-200 mesh). All elements except Pb Trace element analyses were performed using a V.G. were eluted with 3 ml of 0.5N HBr and 0.2 ml of 2N HCl. Plasmaquad 2+ inductively coupled plasma mass spec- These solutions were combined and converted to perchlo- trometer (ICP-MS) at Cornell University. Two hundred mg rate by drying with 1 ml HClO4. Pb was eluted from the of rock powder or fine chips were digested for 24-48 hours columns with 1 ml of 6N HCl, evaporated to dryness and with a 3:1 HF:HNO3 mixture in teflon capsules. After evapo- treated with HNO3 to destroy residual organics. Further rating to dryness, samples were dissolved in dilute HNO3 details of this procedure may be found in White and Dupré and the solutions diluted to 200 ml with distilled water. This [1986]. procedure differs from that previously used in this labora- An aliquot equivalent to 50 mg of rock was taken from tory [Cheatham et al., 1993] in that the digestion proce- the residue after Pb extraction, dissolved in 1 ml 2.5 N HCl dure did not employ HClO4. HClO4 was abandoned to 139 141 and loaded on a column consisting of 5 ml BioRad AG50W- eliminate CaClO4 interferences on La and Pr X12 (200-400 mesh) resin, from which Sr and rare eatrth [Longerich, 1993]. The HF-HNO3 digestion has the addi- (REE) fractions were extracted. Nd was purified from the tional benefit that Zr, Nb, Hf, and Ta remain dissolved in- REE fraction on a 1.5 ml column consisting of di-2-ethyl- definitely in the resulting solutions, presumably because hexyl orthophosphoric acid absorbed onto teflon beads. enough F remains after digestion to form stable soluble fluo- Further details of these techniques may be found in White ride complexes with these elements. With the HF–HClO4 and Patchett [1984]. dissolution, we found that the concentrations of these ele- Isotopic analysis in most cases was carried out on a V.G. ments decreased with time, presumably due to adsorption Sector mass spectrometer at Cornell University. A few on container walls. On the negative side, the presence of samples were analyzed using a new Fisons V.G. Sector 54 residual fluoride makes dissolution of the digested samples mass spectrometer. Methods were fundamentally similar more difficult. in both cases. Purified Pb was loaded onto Re filaments Elements were analyzed in 5 groups: first transition met- with silica gel and phosphoric acid. Data was acquired als; Rb through Nb; light rare earths (La, Ce, Pr, Nd); heavy used a static multicollection technique during which the rare earths (Pr through Lu), Hf and Ta; and U, Th, and Pb. 208Pb ion beam intensity was maintained at about 1.5 × The Cornell ICP-MS was upgraded in 1993 to employ a 10–11 A. A mass fractionation correction was determined high-performance interface and a spray chamber cooled to by making repeated measurements of the NBS 981 Pb iso- 2° C. These improvements produce a typical analytical tope standard assuming true values for this standard of sensitivity of 50 million ions/sec/ppm for 115In. For ele- 206Pb/204Pb=16.937, 207Pb/204Pb=15.493 and 208Pb/ ments with concentrations exceeding 50 to 100 ppm in the 204Pb= 36.705. As the analyses were carried out over a rock (50 to 100 ppb in solution), this results in detector period of years, the corrections applied varied, but were saturation effects when the detector is run in pulse count typically 1.10‰/dalton. The 2σ analytical uncertainties mode. As a result, analyses of the first transition metals, were estimated from repeated analysis of NBS 981 at ±0.011 Rb through Nb, and the light rare earth groups were made for 206Pb/204Pb, ±0.014 for 207Pb/204Pb and ±0.045 for using the detector in analog mode (Extended Dynamic 208Pb/204Pb. Purified Sr was loaded onto single W fila- Range mode). The remaining elements were analyzed us- ments with TaF activator and analyzed using a dynamic ing pulse count mode. Since Pr and Nd were analyzed by multicollection technique. Correction for mass fraction- both analog and pulse count methods, they provided a cross ation was made by normalizing to 86Sr/88Sr=0.11940. The check. No significant loss of precision or accuracy was 84Sr/88Sr ratio was also monitored for data quality assur- observed in analog mode for these elements. Analyses were ance. Measured mean 87Sr/86Sr for the NBS 987 Sr stan- performed in peak jumping mode with a 5120 to 10240 µs WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 4 dwell time, 3 steps per peak separated by 6 DAC steps, and 150-500 sweeps. Washout time was set to 3 minutes and TABLE 1. New K-Ar Age Determinations for Volcanic Rocks sample uptake time to 1.5 minutes. Typical analysis runs from the Society Islands consisted of 1 blank, 5 external calibration standards, 1 data %K Radiogenic Radiogenic Age ±1σ quality standard (a USGS or in-house standard run as an 40Ar, 40Ar, % Ma unknown), 6 drift correction standards, and 14 unknowns 10-7 cc/g and required 2 to 3 hours to complete. Data was reduced Maiao using the drift-corrected external calibration technique de- MAO-6 0.499 0.3206 8.5 1.65 0.07 scribed in Cheatham et al. [1993]. Data reported here are MAO-13 2.724 1.9542 53.7 1.85 0.02 MAO-27 0.492 0.2211 11.4 1.16 0.07 the means of 2 to 5 replicate analyses of each solution run MAO-61 0.998 0.6235 22.4 1.61 0.06 on different days. MAO-65 1.189 0.8482 29.7 1.84 0.04 MAO-102 0.565 0.3576 11.1 1.63 0.08 K-Ar Age Analyses MAO-103 1.751 1.3820 15.1 2.03 0.05 MAO-105 1.468 1.0649 17.0 1.89 0.02 Radiometric ages were determined on whole rock sam- MAO-111 0.457 0.3539 5.4 1.99 0.11 MAO-113 1.003 0.6878 13.1 1.76 0.11 ples by the K-Ar method. Rocks determined to be unal- MAO-114 1.041 0.7576 26.9 1.87 0.07 tered on the basis of petrographic examination were crushed MAO-118 1.006 0.6948 43.9 1.78 0.03 to 0.5-1.0 mm size fraction, ultrasonically washed in dis- MAO-123 0.910 0.5911 5.5 1.67 0.18 tilled water, and dried. A split of this cleaned, crushed frac- Mehetia tion was removed and powdered for K analysis by atomic MHT-101 0.905 0.000 0.0 0.000 0.000 MHT-103 1.023 0.1049 8.2 0.264 0.019 absorption spectrophotometry. Approximately 5 g of the MHT-106 1.060 0.1273 10.3 0.309 0.013 crushed rock was loaded into a Mo-crucible for each argon MHT-120 1.726 0.0056 0.6 0.008 0.002 analysis and a 1 × 10-8 torr vacuum was achieved by over- MHT-121 1.254 0.1428 3.9 0.293 0.006 night baking (175°C) of the glass extraction line. Samples MHT-151 1.269 0.0989 9.5 0.201 0.013 were then fused by radio frequency induction heating of Raiatea RA-008 1.337 1.016 56.4 2.53 0.03 the crucible and active gases were gettered over hot Ti- RA-144 1.213 1.156 17.3 2.45 0.04 TiO2 metal sponge. The isotopic composition of argon was RA-193 1.087 1.018 51.7 2.41 0.04 then measured mass spectrometrically with an AEI MS- RA-219 4.418 3.973 32.8 2.31 0.03 10S instrument at Oregon State University, connected on- RA-225 3.498 3.256 23.4 2.39 0.03 line with the extraction system. RA-227 1.764 1.588 21.5 2.32 0.04 RA-257 4.526 4.033 21.6 2.29 0.03 The majority of samples were analyzed in the conven- RA-71A 0.990 1.025 25.4 2.66 0.05 tional manner, introducing a known amount of 38Ar spike Tahaa during the fusion to determine the concentration of radio- Ta-1D 0.846 0.992 39.8 3.02 0.06 genic 40Ar [Dalymple and Lanphere, 1969]. For very young Ta-2J 1.297 1.707 40.3 3.39 0.14 basaltic samples from Mehetia, the analytical procedures Ta-2R 0.632 0.270 37.5 1.10 0.02 Ta-2V 0.574 0.315 21.7 1.41 0.03 of Cassignol and Gillot [1982] were adopted. This proce- Ta-3F 1.517 1.856 24.5 3.15 0.09 dure yields improved precision for samples with low con- Ta-4H 1.764 2.081 32.2 3.03 0.06 centrations of radiogenic argon and is different in several Ta-8B 1.033 1.276 17.5 3.18 0.09 ways. First, 38Ar spike was not added because of small but Ta-8G 5.000 5.313 13.5 2.73 0.11 significant amounts of 40Ar and 36Ar in the spike. Instead, Ta-9G 1.320 1.402 74.6 2.73 0.03 40 Ta-10H 0.506 0.611 23.3 3.10 0.10 sample Ar concentration was determined from instrument Ta-15R 1.321 1.346 73.6 2.62 0.03 sensitivity, which varied only 1-2% over long-term mea- Ta-2R 0.632 0.270 37.5 1.10 0.02 surements of 38Ar peak heights in spiked analyses and 40Ar Ta-2V 0.574 0.315 21.7 1.41 0.03 peak heights in atmospheric argon calibrations. Second, Age calculations based on the following decay and abundance λ × -10 -1 λ × -10 -1 aliquots of air argon were routinely measured both before constants: e = 0.581 10 yr ; β = 4.962 10 yr , 40 × -4 and after each sample analysis in order to make the impor- K/K = 1.167 10 mole/mole tant correction for atmospheric 40Ar contamination as pre- reported in Table 1. These data, together with data of cisely as possible. This approach results in lower errors in Duncan and McDougall [1976], Duncan et al. [1994], and correcting for atmospheric argon than spiking. Proportions Binard et al. [1993] are represented in Figure 2 as a series of radiogenic 40Ar as small as a few tenths of one percent of histograms on an age versus distance plot. Based on the of the total 40Ar signal are thus detectable and reproducible maximum ages of volcanism at Mehetia, Tahiti, and [Levi et al., 1990; Gillot and Nativel, 1989; Prestvik and Maupiti, the motion of the Pacific Plate over the hotspot is Duncan, 1991]. 11 cm/yr, the same value estimated by Duncan and McDougall [1976]. RESULTS On Hauhine, the range in ages is about 1 million years. On Maiao, the range in ages is about 900,000 years. How- New K-Ar ages on Society Islands volcanic rocks are WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 5

600 ever, all but one sample were erupted within 400,000 years. The range in ages from Raiatea is about 350,000 years; the 10 ages reported by Duncan and McDougall [1976] fall within 500 8 Maupiti 6 this range (after recalculation using the decay constants in Bora Bora 4 Tahaa Table 1). The range in ages from Maupiti fall between that 2 400 Frequency Raiatea of Hauhine and Raiatea, while those of Bora Bora and Hauhine Moorea are somewhat smaller, perhaps due to limited sam- pling or preservation. 300 Maiao 11.3 cm/yr Though no historic subareal eruptions are known, Mehetia appears volcanically active, judging from a 1981 200 Moorea volcanoseismic crisis and probable submarine eruption on Distance along Strike, km Tahiti Nui the southeast flank [Binard et al., 1993] as well as its youth- Tahiti Iti ful appearance. Our new data confirm the occurrence of 100 recent volcanic activity on Mehetia: there have been at least 2 subareal eruptions in the last 10,000 years. However, Mehetia our results also show that Mehetia, though small (just over 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 400 m in elevation and 1.5 km in diameter), has been emer- Age, Ma gent for at least 300,000 years, and thus substantially older Figure 2. Histograms of K-Ar ages of individual Society Islands than inferred by Binard et al., [1993]. arranged on a time vs. distance plot. The Figure includes data Tahaa clearly stands out in that there is a total range in from Duncan and McDougall [1976], Duncan et al. [1994], and ages of 2.3 million years, the new ages completely brack- Binard et al. [1993] as well as Table 1. Maximum ages indicate the Pacific Plate has been moving at 11.3 cm/yr over the plume. eting those reported by Duncan and McDougall [1976].

12 Pacific MORB J Tahiti (This Work) ? Tahaa E Tahiti (Others) C Maiao 10 B Moorea (This Work) } Hauhine Hawaii G Moorea (Others) 8 Maupiti G F a Mehetia (This Work) 2 Tehetia 8 la A p Mehetia (Others) 5 a { Rocard g { H Bora Bora os { Moua Pihaa Tubuai { ?{{ 6 {{ C J E ε { J Nd Rurutu { J JCJEJH {?CFEJEJ FAC AE CC EJE JE H JE E}J A A EJE AA2 4 AEJ2FJ?A H AJ JEH2 22 Mangaia 22EA22JC ? FF2JABA2H 22 J2EAA 5? Marquesas B2 2E } 5 J E2 2 Samoa J 2 2 Macdonald Smt E H8 G 82} Rapa E 555 ? 5 }{? 0 Raratonga B Heard Is. J 555} B5 ? 5?? Pitcairn Smts ? ? ? -2 Kerguelen 0.702 0.703 0.704 0.705 0.706 0.707 87Sr/86Sr

ε 87 86 Figure 3. Nd vs. Sr/ Sr for the Society Islands. Data for Tahiti (Others), Moorea (Others), Mehetia (Others), Tehetia, Rocard, and Moua Pihaa are from Devey et al. [1990], Cheng et al. [1993], and Hémond et al. [1994]. Fields for other oceanic islands and Pacific MORB are based on data compiled from the literature. WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 6

TABLE 2. Isotope Ratios in Society Islands Magmas Rock Type Series 87Sr/86Sr 143Nd/144Nd 206Pb/204Pb 207Pb/204Pb 208Pb/204Pb Bora Bora USNM 101282 alkali basalt 0.70522 0.512719 18.916 15.593 38.634 USNM 101286 alkali basalt 0.70453 0.512824 19.270 15.637 39.087 USNM 101287 alkali basalt 0.70409 0.512909 USNM 101288 alkali basalt 0.70445 0.512861 73-257 alkali basalt 0.70473 0.512797 19.032 15.599 38.727 73-332a alkali basalt 0.70495 0.512829 19.092 15.611 38.800 Hauhine USNM 101211 alkali basalt 0.70527 0.512768 19.212 15.605 38.899 USNM 101198 hawaiite 0.70580 0.512657 19.190 15.636 38.95 73-83 basanite 0.70428 0.512855 18.976 15.574 38.684 73-95a alkali basalt 0.70567 0.512704 19.030 15.600 38.839 73-111 trans. basaltb 0.70607 0.512639 19.092 15.607 38.873 Maiao MAO-6 hawaiite 0.70370 0.512930 19.228 15.526 38.700 MAO-13 mugearite 0.70394 0.512902 19.127 15.529 38.652 MAO-27 alkali basalt 0.70481 0.512809 19.161 15.551 38.687 MAO-61 alkali basalt 0.70481 0.512883 19.170 15.540 38.706 MAO-65 alkali basalt 0.70383 0.512869 19.164 15.529 38.693 MAO-113 alkali basalt 0.70387 0.512863 19.136 15.543 38.683 MAO-114 hawaiite 0.70382 0.512891 19.161 15.536 38.644 Maupiti 73-201 trans. basalt 0.70523 0.512726 19.063 15.615 38.822 73-204c alkali basalt 0.70530 0.512706 19.068 15.586 38.826 Mehetia MHT-101 alkali basalt 0.70390 0.512890 19.050 15.534 38.565 MHT-103 alkali basalt 0.70391 0.512894 19.045 15.526 38.553 MHT-120 alkali basalt 0.70425 0.512802 18.797 15.515 38.329 MHT-121 alkali basalt 0.70449 0.512802 19.038 15.553 38.657 MHT-151 alkali basalt 0.70452 0.512825 19.043 15.547 38.650 MHT-155 alkali basalt 0.70388 0.512883 19.043 15.525 38.544 Moorea USNM 101159 mugearite 0.70593 0.512621 19.317 15.631 39.108 73-233 mugearite 0.70491 0.512637 19.323 15.651 39.157 73-234 alkali basalt 0.70461 0.512802 19.068 15.586 38.826 93MO-1 alkali basalt 0.70459 0.512773 19.210 15.588 38.916 93MO-4 mugearite 0.70492 0.512643 19.311 15.637 39.118 Tahaa 73-185a alkali basalt shield 0.70691 0.512587 19.288 15.656 39.032 73-186a alkali basalt shield 0.70505 0.512804 73-189 alkali basalt shield 0.70552 0.512672 73-190 alkali basalt shield 0.70619 0.512603 19.103 15.622 38.908 73-192 trans. basalt shield 0.70629 0.512613 19.237 15.664 39.063 Ta-1D alkali basalt shield 0.70631 0.512565 19.250 15.594 38.804 Ta-2J alkali basalt shield 0.70487 0.512781 18.976 15.529 38.523 Ta-3F alkali basalt shield 0.70605 0.512653 19.106 15.569 38.721 Ta-4H hawaiite shield 0.70608 0.512608 19.092 15.573 38.744 Ta-8B alkali basalt shield 0.70458 0.512829 18.941 15.531 38.512 Ta-8G trachyte shield 0.70621 0.512597 19.238 15.587 38.811 Ta-2R basanite p. e.c 0.70368 0.512958 18.757 15.479 38.302 Ta-2V basanite p. e. 0.70371 0.51289 18.766 15.470 38.301 Tahitid USNM 100793 alkali basalt 0.70446 0.512782 19.124 15.606 38.861 USNM 100801 alkali basalt late stage 0.70405 0.512873 19.056 15.585 38.753 USNM 101030 phonolite Tahiti Iki 0.70404 0.512904 19.214 15.609 38.856 USNM 101044 olivine basalt late stage 0.70425 0.512836 19.229 15.603 38.839 74-416a alkali basalt late stage 0.70434 0.51285 74-422a alkali basalt late stage 0.70419 0.512918 74-437 alkali basalt shield 0.70452 0.512826 19.103 15.538 38.641 T85-7 alkali basalt shield 0.70438 0.512828 19.101 15.531 38.624 T85-12 alkali basalt late stage 0.70400 0.512895 18.961 15.524 38.490 WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 7

TABLE 2. (continued) Rock Type Series 87Sr/86Sr 143Nd/144Nd 206Pb/204Pb 207Pb/204Pb 208Pb/204Pb Tahitid T85-19 alkali basalt shield 0.70448 0.512793 19.034 15.542 38.568 T85-24 alkali basalt late stage 0.70424 0.512866 18.975 15.512 38.482 T85-33 alkali basalt late stage 0.70408 0.512859 19.031 15.534 38.571 T85-34 alkali basalt shield 0.70393 0.5129 18.981 15.520 38.536 T85-35 trans. basalt shield 0.70417 0.512818 19.194 15.575 38.790 T85-40 trans. basalt shield 0.70520 0.512728 19.062 15.570 38.705 T85-41 trans. basalt shield 0.70518 0.512752 19.050 15.555 38.655 T85-43 alkali basalt late stage 0.70471 0.51281 19.061 15.528 38.587 T85-44 hawaiite shield 0.70581 0.512642 19.182 15.565 38.815 T85-67 hawaiite shield 0.70397 0.512904 18.961 15.505 38.491 T85-72 alkali basalt shield 0.70375 0.512904 19.120 15.519 38.611 T85-93 alkali basalt shield 0.70439 0.512814 19.005 15.530 38.570 T85-102 mugearite shield 0.70397 0.51293 18.977 15.516 38.521 aSr and Nd isotope ratios fromWhite and Hofmann [1982]. bTrans. basalt indicates transitional basalts with little or no nepheline in the norm, c”p.e.” indicates post-erosional lavas. dData from Tahiti are also available from the American Geophysical Union as a microfliche suppliment to Duncan et al. [1994].

40.5 years, the shield series. The post-erosional phase appar- Macdonald Smt St. Helena Galapagos ently followed a volcanic hiatus of 1.2 million years. In 40 Easter Pitcairn Smts Atiu this respect, and in the highly undersaturated nature of the Kerguelen post-erosional lavas, this pattern of activity is similar to 39.5 Rurutu Pb Local Sediment the Hawaiian one described by Macdonald and Katsura Samoa5{ BB 204 HB 2 ?? Marquesas A2 E 39 ?}22}HBE [1964]. The Hawaiian-style post-erosional volcanism of }8BA}5J5 JJ Pb/ 2 J? JAAH J{? J Z EH2{JE??{C{C Tahiti (This Work) ? 8E}FJJCC Tahaa Tahaa appears to be unique in the Societies. While Tahiti Raratonga Z EJJFEJJ E 208 JF Tahiti (Others) C 38.5 ?J?J Maiao A Mehetia (Others) } Heard ?F Rapa Hauhine had a very long period of activity (1.4 Ma) and lavas be- F Mehetia (This Work) 8 Maupiti B Moorea 2 Tehetia came increasingly undersaturated with time [Cheng et al., 38 { Moua Pihaa Hawaii 5 Rocard 1993; Duncan et al., 1994], there is no obvious hiatus in Pacific MORB H Bora Bora 37.5 activity. There may have been a significant hiatus between 18 18.5 19 19.5 20 20.5 21 the main period of activity on Maiao (2.03–1.61 Ma) and the eruption of MAO-27 at 1.16 Ma. However, MAO-27 15.8 St. Helena Local Sediment is a rather typical, moderately differentiated alkali basalt Atiu and does not have the highly undersaturated character of 15.7 Pitcairn Smts Rapa Samoa Tahaa or Hawaiian post-erosional volcanics. Kerguelen ?? Pb 5{ B HB ?2 }E Rurutu Isotope ratios are reported in Table 2. As may be seen in Z 82 E 204 Z H}2 J H}}J HJ Easter 15.6 8 A525 ? 2JB B? Figures 3, 4, and 5, the Society Islands define a particu- }E2{JE? J Pb/ A? Macdonald Smt Heard E JAA{{J F CC JFJCC{ Marquesas

207 ??JEJCC EJJF J C 0.708 F J Galapagos J 15.5 J Tahiti (This Work) ? Tahaa Raratonga E ? Tahiti (Others) C Maiao Hawaii A Mehetia (Others) } Hauhine Pacific MORB F 8 ? Mehetia (This Work) Maupiti 2 15.4 B Moorea Tehetia 5 18 18.5 19 19.5 20 20.5 21 { Rocard Pitcairn ?? Moua Pihaa 206 204 ? ? H Pb/ Pb Smts }?? Bora Bora 0.706 5 B J}{5

Samoa } Figure 4. Pb isotope ratios in Society Islands volcanics. Data 2 Sr Heard }8 E sources are the same as in Figure 3. Also shown are the Pb isoto- Kerguelen 8 JJ H 86 5 2 H B pic compositions of samples of local sediment reported by Devey 0.705 ? 2 B H5 C Sr/ EA et al. [1990]. ?E BAA2 B JF2AJJ HE 87 J J2 Raratonga A Atiu F }J E J Marquesas J J EJEJ CJ JJF C Brousse et al. [1986] inferred the existence of two over- 0.704 F CC ?? {J C{ Rapa { lapping calderas on Tahaa and recognized four phases of Hawaii { Macdonald Smt volcanism: a pre-caldera collapse phase, a post-caldera Rurutu 0.703 Galapagos Easter phase, a phase corresponding to the filling of the eastern St. Helena (Haamene) caldera, and a reactivation phase, which we will Pacific MORB 0.702 call the post-erosional phase. Our data show that while all 18 18.5 19 19.5 20 20.5 21 the older magmas are from the first phase, the first three 206Pb/204Pb phases overlap in time. We will refer to magmas of this Figure 5. 87Sr/86Sr vs. 206Pb/204Pb for Society Islands volcanics. period of activity, which had a duration of about 800,000 Data sources are the same as in Figure 3. WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 8

0.707 > ment samples from the region analyzed by Devey et al. [1990]. > > > > > 0.706 > Figure 6 illustrates the variation in isotope ratios with time in Tahaa. There is no obvious distinction in isotope > geochemistry among the first three phases of Brousse et al. > 0.705 > [1986], nor any obvious systematic geographic or tempo- Sr > 86 ral variation among the shield lavas. Post-erosional lavas, Sr/ which occur as two small eruptive centers on the north coast 87 0.704 ? ? and one on the south coast, are clearly both much younger than other magmas from the island and have much more 0.703 Post- Shield depleted isotopic signatures. Erosional Trace element concentrations are reported in Table 4. 0.702 Table 3 also reports the means and standard deviations of 1 1.5 2 2.5 3 3.5 Age, Ma analyses of 3 solutions of USGS Standard BHVO-1 (each solution was analyzed 3 or more times), from which accu- Figure 6. 87Sr/86Sr vs. age for Tahaa volcanics. Eruption of post- racy and precision of the analyses can be judged. Preci- erosional lavas with dramatically lower 87Sr/86Sr followed a vol- canic haitus of 1.2 million years. 500 larly large range in Sr and Nd isotope ratios, but a rather Mehetia Bora Bora Tahitian B J limited range in Pb isotope ratios. New data reported here Basalts MHT-151 101282 Maiao Maupiti J F H F J MAO-65 73-201 fall within the range of Sr and Nd isotope ratios previously B A F J > AB F Hauhine Moorea reported [White and Hofmann, 1982; Devey et al., 1990; 100 H B J A > H> A F 73-95 73-234 > B H A > Cheng et al., 1993; Hémond et al., 1994], but extend the H F JB F range of Pb isotope ratios to both higher and lower values. A> B H J F A> B The entire range of isotope ratios observed in the Society H >J F HA B >J F Sample/Chondrites HA Islands is nearly encompassed by the range observed in the >JB F HA F J>B F HA F single island of Tahaa: the two post-erosional lavas (TA- BJ> F 10 HA J> HBA A>J >J 2V and TA-2R) have among the most depleted isotopic sig- BH HA B natures while the older ones (most notably 73-185), which 5 we will refer to as shield-series lavas, have the most en- La Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb Lu riched isotopic signatures. Also apparent in these figures Figure 8. Representative rare earth patterns of volcanics from is the strong correlation between Sr and Nd isotope ratios, Bora Bora, Hauhine, Maiao, Mehetia, and Maupiti compared with and the weaker correlations involving Pb isotope ratios, a field for basalts from Tahiti. which is due in part to the small range of the latter. Com- sion ranges from a high of 0.8% for Y and Sc to a low of pared to other oceanic islands, Sr and Nd ratios range to 4.6% for Th. The data we report in Table 3 are means of 2 extremely high and low values respectively, but Pb isotope to 5 analyses, so the precision of these analyses should be ratios are intermediate. Two samples from Moorea, 72- comparable to the standard deviations listed for BHVO-1. 233 and 93MO-4, plot distinctly off the Sr-Nd isotope ar- Representative rare earth patterns are shown in Figures ray. Figure 4 also shows the Pb isotope ratios of two sedi- 7 and 8. All magmas are moderately to strongly light rare earth-enriched. The steep slopes of the heavy rare earths 500 J suggest these magmas were produced primarily within the J H garnet stability field (deeper than 60 km). This is in turn J Tahaa F H J F consistent with the production of these magmas beneath AH Ta 8G Ta 2V F JH A H A 100 F Ta-4H Ta 2R relatively thick 60 Ma old lithosphere. For the most part, A F H A H Society magmas have reasonably parallel rare earth pat- JF H A F H terns, [La/Sm]EF ranges only from 2.3 to 3.7 in basalts and JA F H JA H hawaiites (some differentiated lavas have higher La/Sm). F Sample/Chondrites Tahaa Shield JA H F H Highly differentiated magmas, such as rhyolite Ta 8G, have JA F HH JA JF steep light rare earth patterns. Tahaa sample Ta-4H, a AJJF J 10 AF AF hawaiite with 5.26% MgO, shows a strong negative Ce A anomaly. Whether this is inherited from the source, a prod- 5 La Ce Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb Lu uct of unusual fractional crystallization, or a result of alter- ation is unclear. Ce anomalies were not observed in other Figure 7. Rare earth patterns of post-erosional basalts from Tahaa (Ta-2R and Ta-2V) compared with a field for shield series basalts. samples. Perhaps surprisingly, the rare earth patterns of Ta-8G is a trachyte and Ta-4H is a hawaiite, both of the shield the two post-erosional magmas from Tahaa, which have series. WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 9 TABLE 3. Trace Element Analyses of Society Islands Volcanics Analyses of Society Islands Element Trace 3. TABLE 2.84 3.53 3.03 2.58 2.59 2.90 3.71 2.04 3.85 4.03 3.79 3.10 3.34 2.97 24.72 27.02 26.30 24.00 24.29 26.01 20.70 3.24 25.44 24.68 23.53 25.47 σ 73−332 101286 101282 73−83 73−95 101211 ΜΑΟ−6 ΜΑΟ−13ΜΑΟ−27ΜΑΟ−61ΜΑΟ−65 ΜΑΟ−113 ΜΑΟ−114 73−201 1 ± 0.5% 0.8% Mean 314.20 USGS Std BHVO-1 Bora Hauhine Maiao Maupiti 2 TiO ScV 2.73 CrCo 288.0Ni 1.6% 119.68 0.2Cu 1.3% 136.77 259.8Zn 0.7% 109.41 0.9% 328.3Ga 60.60 602.8 1.1%Rb 21.8 413.94 263.6 43.16 134.5 126.08Sr 21.85 63.30 1.1% 303.85Y 257.3 50.79 1.1% 527.8 9.12 74.58 401.2Zr 106.2 514.78 1.0% 222.1 76.52 797.8 18.45Nb 126.5 58.70 326.54 1.5% 29.15 179.2Ba 22.66 278.5 47.79 156.96 518.5 32.85 0.8% 117.4 57.89 482.1 19.10La 1.9% 31.99 20.27 134.5 587.4 290.1 43.49 96.63 211.6Ce 2.8% 114.3 46.69 26.47 243.9Pr 52.68 16.22 17.42 2.1% 668.2 32.29 124.9 41.01 146.8 313.3 121.1 45.26 27.52 4.59Nd 40.40 2.8% 268.0 31.54 20.58 44.17 667.1Sm 29.20 360.2 159.95 2.1% 392.4 120.6 36.17 9.25 346.0 5.69 1.4Eu 25.05 25.08 33.30 21.91 49.90 588.0 24.16 347.1 82.65 3.9% 203.2 135.0 57.44Gd 48.74 35.30 3.4% 405.4 6.30 3.32 130.3 29.03 24.08Tb 80.67 34.66 629.5 27.68 326.3 47.41 2.4% 296.3 123.9 2.00 37.32 51.74 481.5 32.55Dy 71.58 7.71 114.6 123.33 1080.9 6.18 25.43 2.4% 45.36 28.85 6.18 42.84Ho 245.8 130.9 10.53 37.14 295.8 25.90 71.68 476.1 1056.5 1.02 1.6% 56.59 6.82 57.34 113.34Er 23.53 31.1 40.15 39.09 5.39 57.42 322.0 1.8% 125.0 15.59 302.5 41.76 93.58 2.01 513.8 817.1Tm 37.73 8.93 37.12 88.40 1.02 11.78 1.9% 25.54 6.11 54.58 38.65 369.48 Yb 30.19 617.1 130.9 266.3 2.66 33.96 461.0 821.8 1.4% 28.1 2.48 7.25 10.10 36.98 0.97 73.20 102.27Lu 24.01 24.39 27.85 0.35 7.79 35.37 42.20 4.92 315.9 1.6%Hf 54.53 121.57 943.8 710.9 181.5 130.6 10.84 2.4% 59.14 1.25 2.77 2.04 68.97 35.31 0.91 6.73 47.29Ta 38.42 164.80 38.01 23.87 6.28 7.94 344.5 0.30 77.59 2.1% 592.0 700.8 126.4U 2.20 60.56 694.0 131.27 1.15 9.44 61.84 1.24 2.06 4.53 38.01 0.32 2.8% 31.43 64.96 8.15 23.44 315.1Th 66.08 456.1 5.96 6.14 106.18 117.5 2.75 1.17 50.14 2.3% 752.7 1.75 15.81 0.38 12.32 1.03 78.90 18.83 0.96 2.35 102.67 36.72 48.23 4.2% 433.2 8.21 0.48 0.26 275.1 526.5 2.25 4.63 6.85 20.41 1.54 2.43 41.01 38.78 5.54 58.42 4.3% 0.33 43.66 47.36 0.82 0.33 13.18 1.09 2.50 318.6 92.77 4.6% 355.3 1.84 15.51 48.05 1.89 5.30 7.27 7.64 53.66 1.94 108.33 24.56 13.74 257.7 0.27 0.77 40.52 0.93 0.28 433.6 2.82 1.18 3.98 13.60 3.27 32.78 11.11 54.48 70.34 1.45 5.77 9.42 11.06 2.19 1.26 52.80 357.5 13.56 0.30 5.65 1.03 0.22 3.22 1.71 3.97 10.33 11.05 51.27 1.72 8.22 4.81 30.16 2.43 2.47 0.33 12.51 9.26 1.44 0.25 2.15 1.63 3.26 9.31 11.35 55.60 1.92 7.82 6.72 3.23 0.84 14.00 0.45 37.60 3.90 1.38 0.28 2.63 1.45 3.23 9.44 10.83 2.41 7.05 6.05 3.38 1.62 9.34 0.50 11.56 9.08 1.23 0.33 2.36 1.48 3.39 9.97 2.89 7.37 7.44 2.89 0.93 0.39 7.78 6.40 1.33 0.42 3.52 1.56 12.39 3.27 9.17 2.20 7.75 3.16 1.70 0.43 6.33 1.40 0.31 6.36 10.37 1.46 7.05 3.57 2.46 3.31 2.59 7.39 13.46 0.46 0.36 3.69 6.48 1.33 1.63 7.69 2.22 2.62 10.59 2.07 8.10 3.20 0.37 3.48 0.48 1.49 1.04 7.65 6.60 1.04 2.69 5.02 3.54 3.03 0.51 0.89 0.38 5.90 6.11 1.25 2.76 2.09 0.29 0.40 2.67 7.16 1.63 4.34 1.11 0.24 2.72 6.42 6.08 1.61 2.13 4.65 1.24 Pb 2.05 1.1% 2.89 3.65 5.52 2.60 4.92 4.24 4.28 7.75 4.46 3.06 3.09 2.44 3.34 3.91 WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 10 TABLE 3. (continued) TABLE -101 -120 -151 -155 2.61 3.21 3.83 3.63 3.40 1.62 2.97 3.15 1.59 0.57 3.68 3.28 3.29 2.66 3.02 3.59 3.60 73-204 MHT MHT MHT MHT 73-233 73-234 93MO-1 93MO-4 101159 73-185 73-186 73-189 73-190 73-192 Ta-1D Ta-2J Maupiti Mehetia Moorea Tahaa 2 ScTiO 22.37 26.55 20.31 24.91 25.37 21.05 25.99 23.24 22.95 26.98 21.41 25.02 VCr 239.4Co 673.8Ni 275.0 61.80Cu 699.7 229.8 355.55Zn 66.44 447.87 95.3 278.2 49.38Ga 121.5 70.99 88.16 381.8Rb 297.8 65.24 372.93 60.39 19.05 126.7Sr 105.7 35.80 682.0 39.55 380.75Y 148.9 62.01 19.39 150.71 575.4 249.3 270.5 71.36Zr 244.19 143.3 25.69 28.29 26.10 463.5Nb 23.37 610.8 261.2 74.88 347.97 55.39 227.5 52.90 22.34 1025.3 127.7Ba 22.19 575.3 137.09 37.97 34.00 26.39 103.8La 187.4 717.9 247.2 62.84 410.4 20.54 59.79 22.74 46.03Ce 237.6 437.8 37.93 123.6 21.30 17.3 29.23 180.97 671.0 36.69 26.67 304.6Pr 115.50 30.34 58.54 65.32 300.9 19.9 22.26 212.4 353.65 78.86 335.9 124.3Nd 661.5 28.24 33.29 5.63 45.67 28.00 297.3 299.01 22.94Sm 599.8 261.2 389.5 20.63 271.8 75.61 9.86 189.3 62.42 37.82 40.96 42.62 22.67Eu 133.82 324.24 913.0 40.87 9.63 606.6 237.7 596.7 42.78 344.9 29.27 238.1 7.71Gd 30.36 116.73 40.42 93.76 253.75 59.15 93.96 262.8 9.96 40.35 158.19 136.13 171.85Tb 287.3 37.82 551.6 2.29 349.2 65.94 37.14 101.93 212.4 16.28 38.05 27.92 83.29 53.01Dy 8.92 260.6 6.60 115.07 367.10 370.6 25.83 59.89 83.84 177.37 48.20 593.0 11.96 99.0Ho 37.08 1.03 14.06 242.5 33.78 2.59 124.32 47.36 60.95 988.3 34.67 75.54 309.7 19.98 46.41Er 4.95 930.2 7.53 43.09 226.4 1660.6 462.2 10.36 10.61 66.49 119.69 80.56 4.16 98.37Tm 0.85 50.68 1.15 12.15 78.34 31.42 70.85 379.6 285.7 21.48 162.44 21.26 39.48 683.7 33.54 124.48 638.6 274.1 Yb 5.47 1.97 39.76 116.92 33.79 9.43 3.06 38.38 187.61 73.21 1.88 0.26 8.90Lu 88.95 91.6 0.92 612.6 19.96 9.56 45.97 306.5 33.93 283.61 722.7 126.33 14.79 137.91 8.99 62.38 Hf 41.39 1.46 174.84 37.65 556.6 34.25 2.04 46.57 1.36 2.85 1.58 267.74 22.91 363.0 29.90 77.81 543.1 0.26Ta 8.21 0.21 8.44 9.09 6.37 90.80 81.07 24.39 3.56 811.3 62.17 U 59.48 3.51 1.41 11.50 5.74 38.24 1.06 237.7 26.04 1.26 0.49 43.51 114.48 542.5 22.19Th 2.37 0.20 8.36 98.45 5.98 20.03 563.2 2.33 32.31 2.68 833.8 82.38 2.49 7.51 33.22 1.86 291.6 29.57 46.51 0.30 6.00 1.00 1.63 0.37 16.06 9.01 65.02 6.88 370.5 519.3 22.31 2.21 42.05 35.64 632.7 1.63 47.78 2.21 1.19 29.99 8.73 2.46 7.32 1.60 20.80 0.28 69.78 0.22 1.35 78.68 5.87 453.1 4.68 284.5 12.15 26.75 47.60 4.96 3.88 13.72 152.44 1.54 35.03 12.79 6.97 1.04 1.14 0.58 3.72 607.6 44.17 2.30 2.51 12.35 0.21 68.43 16.02 33.66 5.52 8.63 2.42 80.64 3.32 9.49 3.02 0.33 6.20 480.6 0.96 1.89 2.05 0.50 35.91 40.06 9.93 7.80 2.27 8.30 1.86 19.88 9.96 2.19 3.48 71.19 8.97 2.71 0.30 0.27 1.78 1.44 13.12 5.84 8.02 6.26 5.21 9.94 1.60 40.44 4.26 1.52 2.80 2.43 3.72 7.12 7.05 0.64 15.50 19.24 2.26 0.23 7.61 6.03 6.17 3.61 1.19 1.17 1.26 0.97 2.58 6.15 10.03 8.32 6.24 2.12 0.54 20.78 2.85 5.57 6.32 12.32 0.38 1.13 0.91 34.91 1.19 2.14 4.31 0.84 5.74 6.22 2.25 2.69 16.11 7.26 8.21 0.36 16.77 10.21 0.32 1.00 4.06 1.01 2.40 4.96 2.09 9.42 30.59 2.38 7.31 4.53 6.52 3.25 0.32 0.30 0.89 14.09 1.15 5.75 1.84 7.03 3.35 2.13 8.21 2.80 1.46 2.44 0.30 0.26 1.04 6.96 6.33 1.72 1.57 2.53 5.45 1.21 3.37 1.10 0.35 0.25 5.30 8.31 2.91 2.05 1.97 6.92 0.39 0.92 1.87 0.30 5.77 11.98 2.29 2.15 1.37 0.29 0.33 2.46 7.14 6.63 1.64 4.28 1.70 12.07 0.23 3.14 2.55 5.18 1.39 Pb 4.41 2.88 4.73 4.14 3.07 10.96 3.61 2.97 11.85 18.95 11.22 4.17 5.73 3.92 4.68 7.69 3.36 WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 11 TABLE 3. (continued) TABLE 3.55 3.05 1.51 3.19 3.15 Ta-4H Ta-8B Ta-8G Ta-2R Ta-2V 74-416 74-422 74-437 T85-7 T85-12 T85-19 T85-24 T85-33 T85-34 T85-35 T85-40 T85-41 Tahaa Tahiti 2 VCr 239.4Co 89.9Ni 279.2 35.14Cu 730.4 78.2 72.10Zn 60.11 47.89 318.1 10.3 385.35Ga 10.88 135.50 31.80 561.8Rb 58.51 330.2 128.26 57.15 243.16 27.42 115.19Sr 10.80 589.8 118.91 43.97 221.94 56.44 Y 21.81 63.91 728.5 117.43 Zr 39.79 35.04 156.76 65.05 757.0Nb 83.67 21.08 365.4 37.69 671.5Ba 26.68 48.36 22.73 233.9La 844.8 551.8 62.25 1071.3 32.15Ce 42.14 19.80 260.8 924.8 130.73 76.96 29.41 429.6Pr 116.02 36.82 815.5 694.9 66.61Nd 285.3 32.91 30.19 73.72 127.25 815.5 18.82 625.4Sm 51.62 256.3 26.48 239.20 71.14 77.61 44.56Eu 318.4 848.3 38.01 33.31 714.1 15.44 92.67 39.11 9.21Gd 37.81 304.8 49.56 52.87 102.71 34.69 379.3 24.49 878.5 35.58Tb 4.68 77.33 8.06 31.99 14.68 74.10 409.2 23.48 594.6Dy 11.21 69.64 44.11 307.7 10.75 35.73 43.52 99.58 160.99Ho 814.5 2.35 2.34 621.4 226.2 26.39 7.10 12.41 11.54 55.07 127.84 48.32Er 9.08 1440.2 59.24 37.16 283.8 2.84 46.80 33.91 462.9 38.93 101.42Tm 2.15 7.81 9.59 1.09 5.27 115.65 10.14 873.0 430.1 45.40 45.66 77.97Yb 51.49 596.0 12.82 2.75 49.18 4.95 1.25 7.74 35.44 0.64 108.39 394.4Lu 0.90 5.82 86.79 21.88 8.27 965.9 60.71 327.3 15.61 38.01 38.17Hf 3.42 44.21 174.25 3.06 1.19 10.75 2.06 60.57 745.0 1.02 5.80 8.53 466.4 0.27Ta 50.57 0.49 360.8 107.66 12.39 35.66 82.67 32.74 2.37 2.61 12.08 404.8 81.74 597.2U 1.53 38.41 1.00 7.04 8.30 1.32 128.72 0.38 6.40 270.4 9.71 31.08Th 50.58 0.21 2.62 43.98 3.16 10.54 459.5 2.33 42.45 717.3 2.37 1.10 9.31 313.8 81.85 36.90 11.67 0.31 1.12 5.37 5.37 1.50 39.89 61.89 0.35 80.27 8.12 87.59 367.0 8.12 1.76 2.30 19.32 10.01 3.37 298.9 1.44 2.58 34.20 21.55 0.94 6.85 0.35 38.40 540.9 35.25 0.25 55.55 86.25 8.94 1.13 7.41 5.66 2.15 13.78 4.61 1.99 38.85 1.17 9.31 3.02 1.54 0.29 10.07 460.8 7.42 57.72 15.54 40.95 0.28 3.68 2.61 15.34 1.62 38.07 7.00 6.64 2.40 0.34 1.53 1.28 1.45 42.53 11.41 0.22 7.17 5.55 10.58 1.85 4.10 6.02 7.86 2.62 44.50 2.91 1.09 10.98 0.38 1.71 11.31 1.25 5.34 0.25 6.77 2.28 7.25 12.64 43.75 2.16 4.43 1.25 2.94 1.23 10.76 0.93 6.28 8.94 0.39 4.12 0.30 3.15 8.74 9.86 2.14 1.45 2.22 1.16 3.32 1.96 9.58 0.29 9.62 5.44 0.32 4.13 2.74 1.62 7.17 9.21 0.38 1.97 1.72 3.21 1.51 9.54 0.22 7.36 8.12 2.21 3.09 5.38 4.09 7.79 0.55 1.38 1.30 2.40 1.43 0.31 6.98 5.44 2.74 6.33 3.30 8.61 2.98 1.22 2.84 0.40 1.24 4.64 1.24 0.48 2.63 12.39 6.19 1.29 2.17 8.34 2.89 2.83 1.37 5.02 0.39 1.11 6.95 0.31 6.58 7.72 2.77 2.25 10.85 2.62 1.29 1.35 0.36 6.80 0.32 3.48 7.78 3.07 1.69 2.03 6.80 0.43 1.24 0.28 4.64 6.79 2.47 2.96 2.03 8.66 0.41 0.36 7.60 2.60 2.34 1.06 4.56 0.34 2.39 7.41 1.14 5.75 2.32 1.53 5.41 ScTiO 22.57 26.99 10.78 25.81 27.58 Pb 6.54 3.34 22.15 3.13 3.09 3.06 3.48 5.14 3.41 2.78 3.32 6.16 3.57 4.83 3.53 4.04 3.76 WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 12

TABLE 3. (continued) al. [1990] are representative of sediment in the region and Tahiti the oceanic crust consists of typical MORB. T85-43 T85-44 T85-67 T85-72 T85-93 T85-102 Rb 54.38 44.20 72.41 20.46 35.67 TRACE ELEMENT AND ISOTOPIC VARIATIONS IN Sr 592.4 583.3 1043.5 585.1 750.3 SOCIETY VOLCANICS Y 35.15 35.84 46.63 29.51 36.79 Zr 384.1 381.5 458.5 258.8 318.7 Nb 58.86 46.84 96.47 38.76 52.36 Basalts from the Society Islands have some of the most Ba 537.6 551.6 745.3 257.7 477.4 1262.3 radiogenic Sr known from oceanic islands and show an 87 86 La 46.33 50.12 70.95 33.95 41.85 124.85 unusually strong correlation between Sr/ Sr and εNd. Ce 104.06 111.10 153.83 74.25 94.52 258.12 The latter suggests the Society Islands source consists pri- Pr 13.35 13.69 18.63 10.13 11.95 28.51 marily of two components: one incompatible element-en- Nd 53.73 53.19 70.49 41.62 48.43 91.21 Sm 11.06 10.85 13.31 9.04 10.04 15.56 riched with very radiogenic Sr and unradiogenic Nd, and Eu 3.15 3.11 3.83 2.69 2.89 4.68 the other with more typical depleted isotopic signatures. Gd 9.29 8.83 10.77 8.03 8.67 12.44 These observations raise two questions which we wished Tb 1.46 1.46 1.70 1.27 1.36 1.90 to address with this data set. First, are there variations in Dy 7.05 7.26 8.45 6.23 6.82 9.21 trace element ratios that can be related to these isotopic Ho 1.23 1.27 1.54 1.07 1.23 0.24 Er 2.83 3.10 3.80 2.45 2.94 4.54 variations, and second, what is the nature of these two com- Tm 0.37 0.45 0.53 0.34 0.40 0.63 ponents and how have they evolved? The data set used in Yb 2.08 2.44 3.18 1.77 2.35 3.61 the subsequent analysis and discussion includes not only Lu 0.29 0.35 0.47 0.24 0.34 0.52 the data reported here but also data reported by Devey et Hf 8.98 8.81 8.94 6.43 7.25 al. [1990], Cheng et al. [1993], and Hémond et al. [1994]. Ta 3.45 2.63 5.47 2.34 3.16 U 1.77 2.08 2.36 1.04 1.46 3.85 Th 6.56 9.63 9.24 4.04 5.80 15.60 Data Filters Pb 4.35 5.60 5.18 2.94 3.43 11.05 Note: all concentrations are in ppm, except for TiO2, which is Before pursuing questions of source heterogeneity, we in weight percent. Tahiti data are also available from AGU as a wished to eliminate the effects of weathering. While es- microfische suppliment to Duncan et al. [1994]. sentially all the samples we analyzed appeared fresh in hand dramatically more depleted isotopic signatures than the specimen, all, with the exception of the very young Mehetia shield magmas, are indistinguishable from the shield mag- samples, have nevertheless been exposed to tropical mas. weathering for considerable periods. This weathering af- The two Moorea samples that plot off the Sr-Nd isotope fects chemistry, particularly the concentrations of the more array are both mugearites. Both were collected from the mobile elements such as the alkalis and uranium. Thus in same general area (the southwest coast; 73-233 was col- our analysis of the data, we excluded samples that had ex- lected in 1973 by Duncan and 93MO-4 was collected in perienced significant chemical change due to weathering. 1993 by White) and they have nearly identical isotope ra- We used the Ba/Rb ratio as an indicator of weathering ef- tios. 93MO-4 has incompatible element concentrations fects as Hofmann and White [1983] had shown that histori- only a few percent higher and compatible element con- cally erupted oceanic island basalts always had Ba/Rb ra- centrations only a few percent lower than 73-233. It is tios of about 11, whereas older samples had more variable quite possible that both samples come from the same flow Ba/Rb. We excluded all samples with Ba/Rb greater than and that the compositional differences reflect very minor 17 or less than 7. These cutoff points are arbitrary and are fractional crystallization. Since these samples have experi- based on the simple empirical observation that samples enced extensive fractional crystallization, assimilation of outside these limits had apparent “anomalous” ratios of Nb/ material within the oceanic crust might reasonably be sus- U, Rb/Sr, and Pb/Ce whereas those within the limits did pected as the cause of their anomalous Sr and Nd isotope not. Relatively few samples were excluded (10 of 95). geochemistry. These two samples do have slightly elevated Examples are Ta-1D, MAO-61 and 73-186 from our data Pb/Ce ratios compared to basalts with similar 87Sr/86Sr. set and TH-24 from Hémond et al. [1994], which have Ba/ However, elevated Pb/Ce ratios appear to a general char- Rb of 2.3, 50, 21.9, and 77, respectively. acteristic of all the highly differentiated (mugearites, tra- We also wished to exclude samples that had experienced chytes, phonolites) Society volcanics and probably result extensive fractional crystallization. A preliminary exami- from crystallization of minor phases, such as apatite, nation of the data revealed that very fractionated rocks, sphene, or zircon, that have high Ce partition coefficients. mugearites, trachytes, and phonolites, had anomalous in- There is no other obvious feature of their trace element compatible element ratios when compared to basalts and geochemistry that might relate to their apparently anoma- hawaiites. Accessory minerals such as apatite, sphene, bi- lous Sr and Nd isotope geochemistry. These two samples otite, etc., often precipitate during the extreme phases of have particularly high 206Pb/204Pb, whereas assimilation fractional crystallization. These minerals can have high of either basalt or sediment in the oceanic crust would lower partition coefficients for many incompatible elements and 206Pb/204Pb, assuming the sediment analyses of Devey et thus will fractionate incompatible element ratios in differ- WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 13

87Sr/86Sr 87Sr/86Sr 0.702 0.703 0.704 0.705 0.7060.702 0.703 0.704 0.705 0.706 0.707 1.20 0.07 ? ? 1.10 ?? ? 0.06 88 } 1.00 5 2 ? MORB A A H2 J 0.05 H Pitcairn AB ? J 0.90 Smts } FH?2 2 H JJ 2 J Pb/Ce Hf/Sm Hawaii H J BJ? H J EPR 8}?❋ J 0.04 ° J?HJ F J 0.80 9 30'N J FJ J J? C JF J ? ? J J 8 JB J}JJB ?C 0.03 0.70 J Heard Australs CFJJJFMarquesas C ?CFJ? FPitcairn Smts Hawaii 0.60 C 18 AVE MORB Hawaii 16 Zr/Nb = 30 Hawaii { 14 Marquesas ? ? 1.5 12 E 2Heard J }22 Heard JH?J ? 10 JEJ ? ? JJJJJ2A 22? Zr/Nb H 22 2 Nb/La 25 2 JF 2 2 5JH8 J F B8}?? 8 BAAA ? C F FBA H 2A2HABA GA ? C AAJ8 J{CF JFJF222J2?2 } ? Marquesas C 22 2 5 CC J}JJ 8 ? A2 GHJ ?1.0 6 EJ JJ? 2 J E A 4 ?? Kerguelen

2 Pitcairn Smts EPR 9° 30'N Pitcairn Smts 0 0.5 AVE Hawaii 0.14 MORB Heard 50 ❋ 0.12 Marquesas J?? ?Pitcairn Smts ?} 0.10 CJJ J 40 J J?B CJJ ? H JJJHHJ 8J?0.08 J E222HJ8? Nb/U CB? ?G J AAH22? ? JA? Rb/Sr 30 F J25} 0.06 FF2} F22HA?2AA 28 {CEJA222A5 5J? ?CJ}JFB2 2F8?AVE FJ 5J? JEBPitcairn Smts 0.04 20 H2 MORB FJ J Kerguelen 0.02 ❋ CHawaii 10 0.00 0.702 0.703 0.704 0.705 0.7060.702 0.703 0.704 0.705 0.706 0.707 87Sr/86Sr 87Sr/86Sr Figure 9. Relationship between Hf/Sm, Zr/Nb, Nb/U, Pb/Ce, Nb/La, and Rb/Sr and 87Sr/86Sr in Societies volcanics. Symbols are the same as in Figure 5. Data from Devey et al. [1990] and Hémond et al. [1994] are also shown. Fraction- ated and weathered samples, as defined in the text, often plotted off the correlations and are not shown. Average MORB values are from Hofmann [1988]. EPR 9° 30’N field is from Harpp [in press]. Fields for other islands and MORB are based on literature compilations. Factor Analysis ent ways and to a greater extent than will minerals that typically crystallize from basalt. We therefore excluded Figure 9 shows trace element ratios plotted against 87Sr/ all samples with less than 4% MgO. 86Sr for the filtered data set and compares these variations WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 14 with those from a number of other oceanic island groups. Pb Pb There are apparent strong correlations between Hf/Sm, Pb/ Pb Sr 204 204 87 86 204 Ce, Zr/Nb, Nb/U, and Rb/Sr and Sr/ Sr and a weak cor- EF 86 3 O 2 Pb/ Pb/ Pb/ O

87 86 2 Sr/ relation between Nb/La and Sr/ Sr. We used the statis- 2 Nd La/Sm ε 87 207 208 Al MgO FeO* CaO Na Rb/Sr Zr/Nb Pb/Ce Nb/U Mg # 206 tical technique of factor analysis to explore further the re- 1 Rb Ba Nb La Sr Pb Zr Sm Yb Ni SiO lationship between trace elements and isotopes in this large 0.8 and complex data set. We particularly wished to determine 0.6 the relative contributions of source heterogeneity, partial 0.4 melting and fractional crystallization to the variance in in- 0.2 compatible element ratios. The factor analysis model as- Factor 1 0 sumes that variance in a data set is due to p underlying common factors and some component of additional ran- -0.2 -0.4 dom variation in each variable. This may be expressed 1 mathematically as: 0.8 p 0.6 0.4 Xj = Σ ajr fr + εj (1) r = 1 0.2 0

th th Factor 2 -0.2 where Xj is the j original variable, ƒr is the r common -0.4 factor, ajr is the coefficient relating Xj and ƒr, or the “load- ε ε -0.6 ing of Xj on ƒr”, and j is random variation unique to Xj. j -0.8 might include both geological factors and analytical un- 1 certainty, but their meaning and significance are, by def- 0.8 0.6 inition, not addressed in factor analysis. Such a model 0.4 seems well suited to the problem at hand. For example, 0.2 source heterogeneity, partial melting, and fractional crys- 0 -0.2 Factor 3 tallization might be the common factors in the factor analy- -0.4 sis model. The problem might be more complex, however, -0.6 and several factors might be required to describe source -0.8 -1 heterogeneity; in addition to the extent of partial melting, 0.8 depth of equilibration, length of the melting column, etc. 0.6 might also be factors. Thus more than three factors might 0.4 be necessary to fully describe the variance in the data. 0.2 Factor analysis is performed by computing eigenvalues and 0

eigenvectors from the correlation matrix. It differs from Factor 4 -0.2 principle component analysis in that factor analysis attempts -0.4 to account maximally for intercorrelations of variables, -0.6 -0.8 2 3 O whereas the principle component analysis attempts to ac- EF Sr Sr Zr La Ni Pb Pb Pb Pb Ba Yb Rb Nd Sm O 2 Nb ε 2 86 CaO SiO MgO count maximally for the variances of all observed variables. Mg # FeO* Rb/Sr Nb/U 204 204 204 Na Pb/Ce Al Zr/Nb Sr/ The mathematical details are described in a number of texts La/Sm Pb/ Pb/ Pb/ [e.g., Davis, 1986; Jöreskog et al.; 1976]. 87 207 206 208 We chose a subset of variables for the factor analysis experiment using two criteria. First, to assist in identify- ing factors, we included some variables that could unam- Figure 10. Factor loadings of 27 variables deterimined by factor analysis following varimax orthogonal rotation. We associate biguously be associated with source heterogeneity, frac- the first factor, which accounts for 36.4% of the variance in the tional crystallization, and partial melting. Since radiogenic data set to partial melting, the third factor, which accounts for isotope ratios are insensitive to partial melting and frac- 11.2% of the variance, with fractional crystallization, and the first and fourth factors, which account for 25.0% and 7.4% of the vari- tional crystallization (in the absence of assimilation), a fac- ance respectively, with source heterogeneity. tor related to source heterogeneity should show high load- ing for isotope ratios. Major element and compatible trace incompatible element such as La/Sm, and trace element element concentrations in magmas, on the other hand, are concentrations are sensitive to the extent of partial melt- generally dominated by the effects of fractional crystalli- ing, as is the Na2O concentration [Klein and Langmuir, zation, thus a “fractional crystallization factor” should show 1987], but source heterogeneity and fractional crystalliza- high loadings for major elements and major element ratios tion might also contribute to their variance. Other major such as Mg number. Our expectations for a partial melting element concentrations are also somewhat sensitive to ex- factor were more vague. Trace element ratios, particularly tent and depth of partial melting. Second, we wished to ratios of a highly incompatible element to a moderately maximize the number of samples included in the data set WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 15 and therefore chose variables for which we had the most complete data (only samples for which all variables have Sr been determined can be included in a factor analysis ex- 86 periment). Other variables, such as incompatible element a Sr/ 87 Heard ratios, were chosen because we wished to determine the n dominant control on their variance. Computations were e Pitcairn Smts l done using STATVIEW 4.0™. Twenty-seven variables and e J u g J 38 observations were used. Factors were extracted using r JJJJ e J J K J J the principle components method, and the number of fac- Raratonga SamoaJJ tors to retain was determined using the root curve method. J J J J J JJ J JJJ An orthogonal factor rotation was performed using the JJ J J 206 Varimax method. JJJJ JJJJ Pb/ JJ J 204 Four factors that together account for 80% of the vari- JJJ J JJJ Marquesas Pb J J ance in the data set were extracted. Factor loadings are JJJJ J J illustrated in Figure 10. The first factor, which has high J loading for incompatible elements, Na2O, La/Sm, and Rb/ Hawaii Sr, as well as small negative loadings for Ni, MgO, and Galapagos FeO, seems fairly clearly associated with extent of partial ε Rurutu melting. The second factor, with its high loadings for iso- Nd Mangaia tope ratios, can clearly be associated with source heteroge- Pacific MORB Tubuai neity. The third factor, with high loadings for Ni, MgO, and Mg number, is clearly associated with fractional Sr crystallization (olivine fractionation in particular). The 86 fourth factor, which accounts for only 7.4% of the variance Sr/ 87 in the data set, is somewhat more difficult to interpret. It 87 86 Kerguelen has high loadings for Sr/ Sr, εNd, major elements, par- Σ ticularly SiO2 and FeO, and incompatible element ratios. b It is possible this factor has no geological significance; al- Heard J ternatively it may suggest that some of the variance in ma- Pitcairn J JJ JJ jor elements is related to source heterogeneity. A third al- Smts J J JJ Samoa ternative is that it reflects a hidden correlation between the Raratonga JJ JJ J melting process and source heterogeneity. 206 J J JJJJ Though merely extracting 3 factors that can be associ- Pb/ J J JJJ 204 JJJJJ JJJ ated with source heterogeneity, partial melting, and frac- JJJJ J ε J J J Nd Pb JJJ J tional crystallization provides no significant new informa- JJJ JJJJ tion, this experiment does provide several important in- JJ J sights. First, partial melting appears to dominate the vari- Marquesas Hawaii ance in the La/Sm ratio, and to a lesser extent the Rb/Sr Galapagos ratio, whereas variations in Zr/Nb, Pb/Ce, and Nb/U ap- Tubuai Rurutu pear to be primarily related to source heterogeneity. Sec- Pacific MORB ond, with the exception of Pb concentrations, source het- Mangaia erogeneity is not an important control on incompatible ele- Figure 11. Two three-dimensional views of the Society Islands ment concentrations. Third, fractional crystallization is of isotope data shown in the oceanic basalt isotope tetrahedron of secondary importance for highly incompatible element con- Hart et al. [1992]. In 11a, the intersection of the axes is in front centrations, which are controlled mainly by extent of par- of the plane of the paper. In 11b, the 206Pb/204Pb axis is oriented ε toward the viewer, the Nd axis is oriented away from the viewer. tial melting. Conversely, fractional crystallization is a more 11b is rotated approximately 90° clockwise around a vertical axis important control on the variance of moderately in- from 11a. Fields for other oceanic islands and MORB are based compatible elements such as Yb than is partial melting. on literature compilations. Finally, the source heterogeneity factor (Factor 2) has a surprisingly high negative loading for CaO. This loading for source heterogeneity, we performed an additional fac- reflects a highly significant correlation between CaO and tor analysis experiment using only isotope ratios. Two fac- isotope ratios. Whether this is spurious, reflects real source tors account for 88.1% of the variance: the first factor alone variations in CaO, or reflects a hidden correlation between accounted for over 68% of the variance. Thus two thirds the melting process and source heterogeneity that factor of the variance in the isotope data may be successfully ex- analysis could not resolve remains unclear. plained in a simple two component model, in which case To determine how many factors were needed to account the primary factor is related to the proportion of the second end member in the mixture. A simple 3 component model WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 16 O Rb/Sr 2 .038 1.000 .176 -.125 1.000 Nb/U Ba/Nb Na EF .308 -.226 1.000 .389 .387 -.391 .349 1.000 1.000 -.079 .183 -.117 1.000 1.000 .690 .523 .538 .407 -.504 -.653 -.603 .796 Pb Hf/Sm Zr/Nb Pb/Ce Nb/La La/Sm 204 -.456 1.000 Pb/ 208 Pb 204 .437 .366 Pb/ -.533 1.000 207 Pb 204 .455 -.517 -.481 -.514 -.820 -.559 .243 - -.138-.313 -.131 .041 .139 -.014 .081 -.054 .089 .305 -.074 -.092 -.027 1.000 1.000 Pb/ 206 Nd 1.000 ε TABLE 4. Correlation Coefficients for Isotope and Trace Element Ratios in Society Islands Volcanics Element Ratios in Society Islands Trace for Isotope and 4. Correlation Coefficients TABLE Sr 86 386 -.361 .535 -.514 .895 -.847 .587 .869 .825 .694 .662 .771.714 -.736 -.685.666 .718 .867 -.641 .588 .943 .717 .651 .739 . .679 -.632 .347 -.365 -.612 .653 -.969 -.644 .682 Sr/ 87 Pb Pb Pb Sr 1.000 EF 204 204 204 86 O -.192 .153 -.138 -.229 -.213 -.105 -.217 -.287 .145 Pb/ Pb/ Pb/ 2 Sr/ Nd ε 87 206 Pb/Ce Nb/La La/Sm Hf/SmZr/Nb .754 -.686 .586 .742 .691 1.000 207 208 Based on 32 samples. Correlations that are statistically significant at the 0.5% level shown in bold, those 5% italics. Rb/Sr Nb/U Ba/Nb Na WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 17 would explain 88% of the variance in the isotope data. the correlation matrix and these factor analysis experiments: Alternatively, the second factor could be interpreted as ran- (1) source heterogeneity, rather than partial melting or frac- dom heterogeneity within one of the principle end mem- tional crystallization, is the cause of most of the variance bers. Three dimensional views of the isotope data, illus- in Hf/Sm, Pb/Ce, Zr/Nb, Nb/La, Nb/U, and Rb/Sr; (2) even trated in Figure 11, reveal a linear rather than planar array with the inclusion of 8 incompatible element ratios, two so we adopt a model of a two component mixture with factors explain most of the variance attributed to source heterogeneous end members. heterogeneity; (3) the variation in La/Sm and in most in- We performed a third factor analysis experiment to ex- compatible element concentrations appears to be primarily plore further the relative roles of source heterogeneity and due to differences in extent of partial melting; and (4) 207Pb/ partial melting in controlling incompatible trace element 204Pb correlates better with trace element ratios and with 87 86 206 204 207 204 ratios. The data set for this experiment consisted of only Sr/ Sr and εNd than do Pb/ Pb and Pb/ Pb. trace element and isotope ratios and Na2O concentrations. The importance of source heterogeneity in the variations Though our first factor analysis experiment shows the ef- of ratios such as Hf/Sm, Pb/Ce and Nb/U is not surprising fect of fractional crystallization on incompatible element since these element pairs all have similar incompatibilities. ratios was limited, we wished to minimize or eliminate its That variations in Zr/Nb appear to be almost exclusively effect and therefore excluded all samples with Mg# less due to source heterogeneity is, however, surprising, since than 50 in this third experiment. Fourteen variables and 32 these elements have quite different incompatibilities. It is samples were used in the analysis. Three factors that ac- also surprising that Zr/Nb is positively correlated with 87Sr/ counted for 76.4% of the variance were extracted. Factor 86Sr, since Nb is more incompatible than Zr. A similar loadings in this experiment were essentially similar to those positive correlation is present in Hawaiian data, while Heard of Factors 1, 2 and 4 shown in Figure 10; i.e., there were and the Marquesas show the expected negative correlation two factors with high loadings for isotope ratios that we (Figure 9). associate with source heterogeneity and one with a high loading for Na2O that we associate with extent of partial NATURE OF THE SOCIETY ISLANDS MANTLE melting. Hf/Sm, Pb/Ce, Zr/Nb, Nb/La, Nb/U, and Rb/Sr SOURCES all showed high loadings on the source heterogeneity fac- tors; La/Sm and Ba/Nb showed high loadings on the par- White and Hofmann [1982] speculated that the incom- tial melting factor. patible enriched component in the Society Islands source The correlation matrix used in this third experiment is was deeply subducted oceanic crust and sediment that had shown in Table 4. Statistically significant correlations oc- become incorporated into the Society Islands mantle plume. cur betweeen Sr and Nd isotope ratios and Hf/Sm, Pb/Ce, This idea adequately explains the observed radiogenic Sr Zr/Nb, Nb/La, Nb/U, and Rb/Sr. The correlations with Hf/ and unradiogenic Nd, but a number of other models could Sm are controlled to some degree by the exceptionally high explain this observation equally well. The question then Hf/Sm of the sample with the most radiogenic Sr (T73- becomes, are there specific predictions that can be made 185), as may be seen in Figure 9. When this sample is from the White and Hofmann [1982] and alternative mod- excluded, however, the correlation between isotope ratios els that would allow their testing? and Hf/Sm remains statistically significant at the 0.5% level. The isotope geochemistry of modern marine sediment is With the exception of Rb/Sr, all these ratios also show sig- reasonably well characterized. In addition to radiogenic 207 204 nificant correlations with Pb isotope ratios; Rb/Sr shows a Sr and unradiogenic Nd, it has radiogenic Pb/ Pb, but 206 204 weak correlation with 207Pb/204Pb and 208Pb/204Pb, but Pb/ Pb that typically overlaps oceanic basalt values. 208 204 206 204 none with 206Pb/204Pb. 207Pb/204Pb correlates better with Pb/ Pb ratios are also higher for a given Pb/ Pb Sr and Nd isotope ratios and trace element ratios than do than MORB, but the distinction is somewhat smaller than 207 204 the other Pb isotopes. La/Sm and Ba/Nb correlate only for Pb/ Pb. The incompatible element geochemistry 87 86 of sediments are somewhat less easily characterized, in part weakly with Sr/ Sr and εNd and do not correlate with Pb because the concentrations of some elements vary greatly isotope ratios. Na2O shows a weak correlation only with La/Sm. between sediment types and locations, and in part because A number of important conclusions can be drawn from there are few published data sets on which all key concen-

Table 5. Incompatible Element Ratios in Sediments, Mid-Ocean Ridge Basalts, and Continental Crust La/SmEF Rb/Sr Nb/La Ba/La Ba/Nb Zr/Nb Hf/Sm Nb/U Pb/Ce Ave. Marine Sedimenta 4.64 0.303 0.309 27.0 88.9 14.67 0.707 5.26 0.405 1 standard deviation 1.09 0.575 0.092 43.9 88.9 7.15 0.503 4.31 0.348 n 109 132 52 94 107 107 106 51 58 Ave. N-MORBb 0.64 0.011 0.900 3.6 4.0 29.723 0.793 49.32 0.041 T & M Cont. Crustc 2.82 0.123 0.688 15.6 22.7 9.091 0.857 12.09 0.242 aCompiled from BenOthman et al. [1989], Plank and Ludden [1992], and Vroon et al. [in press]. b from Hofmann [1988] c from Taylor and McLennan [1985] WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 18 trations have been measured. Table 5 compares a number of 15.7 incompatible element ratios in sediments with those ratios > in MORB and continental crust. The incompatible element > 15.65 B 5 esas 5 geochemistry of sediments may be summarized as follows. u B H } q B r 2 Rb/Sr, Ba/La, Ba/Nb, and Hf/Sm ratios are extremely vari- a E > E J = M J 2 H J JH } able. Ba/Nb, Ba/La, and Rb/Sr are nevertheless almost al- Galapagos H } 15.6 2 > Pb AA 5 = J BB > ways higher than in MORB, and therefore presumably higher 2 204 J } > than in the upper mantle, while the average Hf/Sm ratio in / E E A J > A J Pb EA 5 Heard Is. F J sediments is close to that of MORB. Zr/Nb and Nb/La in 15.55 F C 207 { C C J CF J J sediments are systematically lower than in MORB, while C C J > > C FFJ J J JE La/Sm is systematically higher. Sediments are most J FJ EF J 15.5 Hawaii readily distinguished from MORB by their Nb/U and Pb/ Pitcairn Smts > 2σ Ce, which are on average an order of magnitude lower and > higher, respectively, than the MORB values. 15.45 We must at this point admit that the composition of sedi- 0.703 0.7035 0.704 0.7045 0.705 0.7055 0.706 0.7065 0.707 ment carried into the mantle is not likely to have the same 87Sr/86Sr composition as modern marine sediment due to geochemical 207 204 87 86 processing in subduction zones. There is abundant evidence Figure 12. Pb/ Pb vs. Sr/ Sr for Society volcanics. Fields for other islands and Pacific MORB are based on literature com- that some of the Pb, alkalis, and other incompatible elements pilations and are shown for comparison. The positive correla- in subducting sediment and oceanic crust finds its way into tion between 207Pb/204Pb and 87Sr/86Sr is consistent with the sedi- island arc magmas. This must inevitably change the com- ment recycling hypothesis of White and Hofmann [1982]. position of the sediment that remains, probably by decreas- ing Pb/Ce, La/Sm, Rb/Sr, etc. and perhaps increasing Nb/U, Sm are less valuable because these ratios are extremely Nb/La, etc. This process remains poorly understood. It ap- variable in sediments. pears to involve dehydration, but may also involve melting The test of the recycling hypothesis may be formulated as well (at least of sediment). Under these circumstances, as follows. If the “enriched” end member in the Society we think it preferable to base the following discussion on Island mantle plume (i.e., the end member with radiogenic the values in Table 5 rather than speculate wildly on what Sr and unradiogenic Nd) contains recycled sediment, this the composition of deeply subducted sediment might be. In end member should also have high 207Pb/204Pb and Pb/Ce this respect we note that mass balance and flux calculation and low Nb/U and Nb/La. Figure 12 shows that, consis- in arcs suggest that only a small fraction of the available tent with the prediction of the recycling hypothesis, the end subducted sediment and oceanic crust is needed to account member with most radiogenic Sr has the highest 207Pb/ for incompatible elements in arc magmas (e.g., White and 204Pb. Figure 9 and Table 4 show that this component also Patchett, 1984], hence the values in Table 5 should be valid has highest Pb/Ce and lowest Nb/U and Nb/La, which is for the purely qualitative arguments below. On the other again consistent with the recycling hypothesis. There is 87 86 hand, we feel that because of these chemical processes in also a weak correlation between Sr/ Sr and εNd and La/ subduction zones, the values in Table 5 are not useful for Sm, which is also consistent with this hypothesis. Our fac- quantitative purposes, such as estimating the amount of re- tor analysis experiments suggest La/Sm may be strongly cycled sediment in the plume. affected by differences in degree of melting, and this may These observations on the geochemistry of sediments al- partly explain why this correlation is weak. We suggest an low the formulation of a specific test of the recycling hy- additional reason below. Pb/Ce, 87Sr/86Sr, and 207Pb/204Pb pothesis of White and Hofmann [1982]. Of the incompatible are lower than typical marine sediment even in the most element ratios mentioned above, Pb/Ce, Nb/U and Nb/La extreme samples, and Nb/U and Nb/La are higher. This is are the most valuable. Pb/Ce and Nb/U are uniform in consistent with the proposition that recycled sediment is a MORB, and indeed in many oceanic island basalts as well component in the Society mantle plume that is diluted by [Newsom et al., 1986; Hofmann et al., 1986]. It appears other material. therefore that these ratios are not easily changed by mag- Figure 9 and Table 4 also show a relatively strong posi- matic processes. The high loadings of these ratios on the tive correlation between 87Sr/86Sr and Hf/Sm and Zr/Nb, “source heterogeneity” factors and low loadings on “partial neither of which is predicted by the recycling hypothesis. melting” factors in our factor analysis experiments support Zr/Nb ratios in all Society basalts are well below those in this contention. Nb/La is valuable because magmatic frac- both MORB and are within the range typical of oceanic tionation (particularly partial melting) should increase this island basalts. Maximum Zr/Nb ratios (≈10) are similar to ratio (i.e., Nb is more incompatible than La), while sedi- those of continental crust and remain about 30% below the ments have a low Nb/La ratio. While there is a clear dis- average value in sediment. The Zr/Nb variations in Soci- tinction between MORB and modern marine sediments in eties basalts would be consistent with the recycling hypoth- La/Sm, this ratio is easily changed by magmatic processes, esis if the unradiogenic end member is not MORB but rather as are the Zr/Nb and Rb/Sr ratios. Ba/La, Ba/Nb, and Hf/ material with low Zr/Nb. As noted above, Hf/Sm ratios WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 19 are quite variable in sediments and encompass the entire ment recycling. Indeed, any form of crustal recycling ap- range observed in oceanic island basalts and MORB. pears consistent with our data. However, because there is Patchett et al. [1984] concluded that terrigenous sediments, clear and convincing evidence from both geophysical stud- that is those sediments rich in continent-derived lithics, had ies of forearcs and geochemical studies of arc magmas that high Hf/REE ratios whereas pelagic sediments, sediment sediment is subducted into the mantle, and evidence of other composed of fine grained continent-derived material as well forms of recycling is weaker, we prefer the sediment sub- as authigenic and biogenic material, had low Hf/Sm ratios. duction mechanism. The difference apparently relates to the concentration of Other possible explanations are less plausible. For ex- Hf in zircon and the behavior of zircon during weathering ample, heterogeneity produced by accretion of material of [Patchett et al., 1984]. These results suggest that if re- differing volatility during formation of the Earth would not cycled sediment is present in the Society mantle plume, result in varying Hf/Sm, Zr/Nb, or Nb/U ratios as these this sediment is more similar to modern terrigenous sedi- elements are all refractory. Crystallization of a primordial ment than to modern pelagic sediment. magma ocean could efficiently fractionate Hf/Sm, Zr/Nb, We conclude that the new Pb isotope and trace element and Nb/La, judging from the partition coefficients for data we report, as well as other data in the literature, most majorite and Mg-perovskite determined by Kato et al. notably Woodhead et al., [1993], are consistent with the [1988], however, it would produce correlations between recycled sediment hypothesis of White and Hofmann these three ratios that are different than those observed, so [1982], though clearly these data do not prove the hypoth- this possibility may also be rejected. esis. There are, however, few obvious alternatives. Based Finally, assimilation of sediment by magmas ascending on the similarity of Sr and Nd isotope ratios in oceanic through the oceanic crust may be ruled out on several basalts and xenoliths from continental basalts, McKenzie grounds. First, assuming the sediment analyzed by Devey and O’Nions [1983] suggested that mantle plumes are de- et al. [1990] is representative of that in the region (since rived from delaminated subcontinental lithosphere that has there are no other sediments analyzed from the region, this sunk to the deep mantle. McKenzie and O’Nions [1983] is the only reasonable assumption), sediment from the re- did not speculate on the mechanism by which the subcon- gion does not plot on the Society Pb-Pb arrays or exten- tinental lithosphere becomes enriched in incompatible el- sions of them as they must if sediment assimilation were ements, but at least two such mechanisms have been sug- important. Second, there is no correlation between MgO gested by others. In the first, melts generated in the un- or Mg-# and isotope ratios, as might be expected if assimi- derlying asthenosphere or mantle plumes might freeze lation-fractional crystallization were important. Further- within the lithosphere. In the second, fluids released by more, highly differentiated lavas, which having experienced subducting oceanic lithosphere might carry fluid-mobile apparent long residence times within high level magma elements into overlying continental lithosphere. Subconti- chambers are the best candidates for having assimilated nental lithosphere enriched by the first mechanism is not a crust, have isotope ratios that plot entirely within the field satisfactory source for the enriched component in the So- defined by primitive basalts (with the exception of 73-233 ciety mantle plume as partial melting of the mantle appar- and 93MO-1 as noted above). Such lavas do tend to have ently does not significantly fractionate Pb from Ce and Nb elevated Pb/Ce, Rb/Sr, and La/Sm and low Nb/U ratios, from U, as pointed out by Newsom et al. [1986] and but these features are readily explained by fractional crys- Hofmann et al. [1986]. Subcontinental lithosphere enriched tallization alone. Thus we find no evidence to support the by the second mechanism is a plausible source for the en- hypothesis that assimilation of sediment has affected Soci- riched component in the Society mantle plume. Island arc ety magmas in any significant way. volcanics, whose genesis appears closely linked to dehy- In summary, the sediment recycling hypothesis appears dration of subducting lithosphere, have low Nb/La and Nb/ to best explain the geochemistry of the enriched compo- U and this Nb depletion may be due to the low solubility of nent in Society magmas. While this hypothesis has passed Nb in fluids released from the subducting lithosphere. Is- an important test, additional tests that are even less am- land arc volcanics, including those for which Pb isotope biguous, for example oxygen isotope analyses, will be re- ratios suggest essentially no contribution from subducted quired before the hypothesis can be entirely accepted. sediment, also have high Pb/Ce [Miller et al., 1994]. The An additional important question is the nature of the low 207 204 87 86 high Pb/ Pb in the enriched Society end member, how- Sr/ Sr, high εNd component in the Society plume. White ever, requires that either this enriched lithosphere be ex- and Hofmann [1982] implicitly assumed it was depleted tremely ancient (>3 Ga) or that subducted sediment con- mantle, i.e., MORB source mantle. Figures 3 through 5 tributed to the enriching fluid. show that this is plausible for isotope ratios provided this Yet another alternative is delamination of lower conti- depleted mantle component has 206Pb/204Pb close to the nental crust [Kay and Kay, 1991]. As key trace element high end of normal MORB range. However, Figure 9 sug- ratios in the lower crust may be similar to those of both gests that this “depleted” end member has Nb/La that is sediments and the upper crust [Taylor and McLennan, higher than MORB and Zr/Nb that is substantially lower 1985], this hypothesis is not easily distinguished from sedi- than MORB. As we discuss in the next section, some WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 20 volcanics with low 87Sr/86Sr were produced by low de- grees of melting, e.g., the Tahaa post-erosional samples. It Shield-building Phase could be argued therefore, that the apparent low Zr/Nb in the low 87Sr/86Sr end member is due to Zr-Nb fraction- ation during production of small degree melts. This view is inconsistent with the results of our statistical analysis, which shows that Zr/Nb does not correlate with variables, such as incompatible element and sodium concentrations, that do appear to be controlled by degree of melting, and a low loading of Zr/Nb on the factor we associated with par- tial melting. These statistical results do not imply that Zr/ Nb is not affected by partial melting, rather that this effect accounts for a much smaller fraction of the variance than source heterogeneity. The same may be said of Nb/La. We therefore conclude that the low 87Sr/86Sr component does indeed have Zr/Nb lower than, and Nb/La higher than, the depleted upper mantle (MORB source), indicating it is sig- nificantly less depleted in incompatible elements than is Post-erosional Phase the depleted upper mantle. Judging from the correlation with isotope ratios among the samples that we believe rep- resent the larger degree melts, this end member has 87Sr/ 86Sr ~ 0.7037, Zr/Nb ~ 7, Nb/La ~ 1.2, Rb/Sr ~ 0.02, and Nb/U ~ 40, and Pb/Ce and Hf/Sm that are indistinguish- able from MORB (these estimates are, of course, highly speculative). The incompatible element-rich nature of this low 87Sr/ 86Sr component may explain why our factor analysis ex- periments indicated the variance in La/Sm was largely due to partial melting rather than source heterogeneity, whereas the opposite was true of Zr/Nb. Since the difference in incompatibility between Nb and Zr is greater than between La and Sm, this result was unexpected. If however, the difference in La/Sm between the two main components in the Society source (the low 87Sr/86Sr and high 87Sr/86Sr components) is relatively small whereas the differences in Figure 13. Cartoon illustrating how volcanic evolution of Tahaa is controlled by passage over a plume that is surrounded by a Zr/Nb are relatively large, then the variance in La/Sm would sheath of viscously entrained deep mantle. Shading is used to be largely controlled by partial melting whereas that of Zr/ suggest variations in temperature, and consequent extent of melt- Nb would not. Thus there is indirect evidence that the La/ ing. The central core of the plume, consisting in part of deeply 87 86 subducted sediment and oceanic crust, is hottest. Temperatures Sm ratio in the low Sr/ Sr component is higher than in in the sheath decrease radially away from the central core. In the MORB. Judging from the correlation between La/SmEF main shield phase, magmas contain a relatively large fraction of and εNd, the La/SmEF may be as high as 1.8 at the low high degree melts derived from the plume core. In the post- 87 86 ε ε ≈ erosonal stage, the contribution from the core is small and the Sr/ Sr–high Nd end of the array ( Nd +7). We ad- bulk of the magmas are derived by small extents of melting of dress the question of the nature of this component in the the cooler sheath, which consists of less incompatible element- next section. enriched material entrained from the deep mantle.

VOLCANIC EVOLUTION OF THE SOCIETY IS- LANDS AND THE STRUCTURE OF THE SOCIETY ing stage [e.g., Macdonald and Katsura, 1964; Chen and MANTLE PLUME Frey, 1985]. Chen and Frey [1985] proposed that the com- positional evolution of Hawaiian volcanoes reflects pas- Both Cheng et al. [1993] and Duncan et al. [1994] ob- sage of the volcanoes over the plume. Duncan et al. [1994] served a systematic variation with time in isotope and ma- proposed a somewhat similar model to explain the varia- jor element geochemistry of Tahitian lavas. As lava chem- tions in Tahitian lavas. Here we propose a slight variation istry becomes increasingly silica-undersaturated, the Sr and of these models, which we believe explains the relatively Nd isotope ratios shift toward more depleted, more MORB- depleted nature of both late stage and post-erosional like, signatures. This pattern is highly reminiscent of the volcanics on Tahiti and Tahaa and of the lavas of the young Hawaiian one, where alkalic capping lavas have more de- Moua Pihaa seamount. Following Duncan et al. [1994], pleted isotopic signatures than tholeiites of the shield-build- we propose the relatively depleted isotopic signature of the WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 21 late and early stage magmas primarily reflects melting of [1992] and Hanan and Graham [1994] than that estimated material entrained by the Society plume. Duncan et by Hart et al. [1994]. Indeed, it appears to us that oceanic al. [1994] suggested this entrained material was depleted arrays focus on the center of the tetrahedron in Figure 11, upper mantle. However, from the discussion in the previ- rather than on the base, which is also consistent with the ous section and the isotope–trace element ratio correlations common component composition estimated by Farley et in Figure 9, it is apparent the source of these late and early al. [1992] and Hanan and Graham [1994]. Following stage magmas must be enriched in incompatible elements Hanan and Graham [1994], we adopt the term “C” for this relative to depleted mantle. We suggest it is material en- component. We note, however, that Hanan and Graham trained from some deeper region of the mantle, perhaps at believe “C” represents the core material of plumes with or below the 660 km discontinuity. other components being entrained. Temporal evolution of The model is illustrated in Figure 13. It is based on the isotope ratios in the Society Islands suggests the opposite observation that plumes can have viscosities no more than is the case; in this respect we agree with Hart et al. [1992]. 1 to 2 orders of magnitude lower than surrounding mantle. We also agree with Hart et al. [1992] that the common Thus plumes cannot rise freely through the mantle; instead, component is somewhat variable in composition and is not surrounding mantle will be viscously coupled to the plume easily characterized by a single isotopic composition. and rise with it as a sheath-like boundary layer. The thick- Hart et al. [1992] argued that FOZO represented the iso- ness of this entrained layer could reach hundreds of km topic composition of the mantle beneath 660 km. The en- [Hauri et al., 1994]. This produces, in plan view, a con- tire lower mantle having this composition poses serious centric zonation with the hottest material in the center sur- problems in performing a Pb isotope mass balance for the rounded by cooler entrained material. Because it has risen Earth, as FOZO has quite radiogenic Pb (206Pb/204Pb ≈ from shallower depth, the peripheral material is cooler than 19.3). Serious mass balance problems remain even if the the plume core and so will melt to a lesser degree. We as- lower mantle has Pb only as radiogenic as the least radio- sume that the volcano is located directly over the plume genic Society volcanics (206Pb/204Pb ≈ 18.7). The upper core during its most vigorous period of growth in the shield crust and all well-characterized mantle reservoirs (depleted stage. During the earliest stage of growth, which we sug- mantle, plumes) have Pb isotope compositions that are more gest Moua Pihaa is now in, and in late and post-erosional radiogenic than that of the bulk Earth (which must lie on or stages, the volcano is located over the entrained sheath. very near a 4.55 Ga isochron called the Geochron). The Small degree melts of this cooler sheath dominate the vol- lower continental crust appears to have a mean Pb isotopic canic activity in these stages. composition that is less radiogenic that the bulk earth. Hart et al. [1992] observed that most oceanic basalt iso- Whether it is sufficiently unradiogenic to balance the crust tope data plot within a tetrahedron in 87Sr/86Sr–143Nd/ and characterized mantle reservoirs remains unclear [e.g., 144Nd-206Pb/204Pb space and that most arrays converge on Rudnick and Goldstein, 1990]. It is highly unlikely that it a region in the base of this tetrahedron that they referred to could be sufficiently unradiogenic to balance these reser- as the “Focus Zone” or “FOZO”. Since this component voirs as well as the lower mantle if the latter has 206Pb/ seems to be common to many plumes, Hart et al. [1992] 204Pb as high as 18.7. If the lower mantle does have a ra- argued it represents the composition of the lower mantle diogenic Pb isotopic composition, either the existing data and is entrained by plumes rising through it from the core- is not representative of the crust and mantle, or the core mantle boundary. There have been other suggestions of a must be the compensating unradiogenic reservoir, and hence common component in plumes: the PHEM component of must have formed substantially after 4.55 Ga. It remains, Farley et al. [1992] and the “C” component of Hanan and therefore, an open question whether the entire mantle be- Graham [1994]. Conceptually, all refer to a component low 660 km could have the isotopic composition of this that is common to many plumes and that has intermediate common component in plumes. We feel any speculation Pb isotope ratios and high 3He/4He. Both PHEM and “C” on how the composition of the common component in differ from FOZO in that the former have intermediate Sr plumes has evolved would be premature before its trace and Nd isotope ratios whereas FOZO has depleted Sr and element composition is better constrained. Nd isotope signatures as defined by Hart et al. [1992]. There is no clear evidence for systematic variation in the We argue here that the low 87Sr/86Sr component in the composition of the Society plume over the time span rep- Society plume is located on the periphery of the plume since resented by our sampling (4.5 Ma). Chemical and isotopic it is more prevalent in early and late stage melts, whereas variations within a single island (Tahaa) equal that of the shield-building lavas generally have higher 87Sr/86Sr. Lo- whole archipelago. While it is true that the most “enriched” cation on the periphery of the plume would be consistent isotopic signatures occur on one of the older islands, 87Sr/ 86 with this material having been entrained, rather than being Sr nearly as high and εNd nearly as low occur on the inherent to the original plume. If the proposed common young seamounts of Moua Pihaa and on Tahiti. Further- 87 86 87 86 component of mantle plume is the low Sr/ Sr compo- more, the oldest island (Maupiti) has Sr/ Sr and εNd nent of the Society plume, then it appears to have an isoto- among the lowest in the archipelago. We suspect most of pic composition closer to that estimated by Farley et al. the apparent inter-island differences reflect incomplete sam- WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 22 pling. Establishment of systematic temporal variation in petrology of the emerged part of the Society hot spot, J. Society plume composition would require much more ex- Volcanol. Geotherm. Res., 55, 239-260, 1993. Brousse, R., T. Gisbert, and C. Léotot, L’ile de Tahaa: un tensive sampling than we have carried out. volcan à deux caldeiras successives, C. R. Acad. Sc. Paris II, 303, 247-250, 1986. SUMMARY AND CONCLUSIONS Cassignol, C. and P.-Y. Gillot, Range and effectiveness of unspiked potassium-argon dating: Experimental ground- Our most important findings may be briefly summarized work and applications, in Numerical Dating and Stratig- as follows: raphy, edited by G. S. Odin, pp. 159-179, John Wiley and Sons (New York), 1982. • New K-Ar ages for the Society Islands are consistent Chase, C. G., Oceanic island lead: two-stage histories and with a Pacific plate motion relative to the Society plume of mantle evolution, Earth Planet. Sci. Lett., 52, 277-284, 11 cm/yr. 1981. • Eruption of highly undersaturated, relatively MgO-rich Cheatham, M. M., W. F. Sangrey, and W. M. White, Sources lavas on Tahaa occurred after a volcanic hiatus of 1.2 mil- of error in external calibration ICP-MS analysis of geo- logical samples and an improved non-linear drift correc- lion years that followed the end of the shield building stage. tion, Spectrochimica Acta, 48B, E487-E506, 1993. These lavas have markedly less enriched isotopic signa- Chen, C.-Y., and F. A. Frey, Trace element and isotopic tures than Tahaa shield building lavas. In this respect, the geochemistry of lavas from Haleakala Volcano, east Maui, volcanic evolution of Tahaa appears to follow the Hawai- Hawaii: implications for the origin of Hawaiian basalts, ian pattern. However, unlike the Hawaiian case, rare earth Jour. Geophys. Res., 90, 8743-8768, 1985. Cheng, Q. C., J. D. McDougall, and G. W. Lugmair, patterns of the shield building and post-erosional lavas are Geocheical studies of Tahiti, Teahetia and Mehetia, Soci- similar, though other incompatible element ratios are dif- ety Island Chain, J. Volcanol. Geotherm. Res., 55, 155- ferent. 184, 1993. • Linear correlations among trace element and isotope Dalymple, G. B., and M. A. Lanphere, Potassium-Argon ratios in the Society volcanics indicate the source consists Dating, W. H. Freeman, San Francisco, 1969. Davies, G. F., Ocean bathymetry and mantle convection, 1, of a mixture of two components, each of which is slightly large-scale flow and hotspots, J. Geophys. Res., 93, 10467- heterogeneous. The isotopic and trace element composi- 10480, 1988. 87 86 nd tions of the high Sr/ Sr, low εNd component, particu- Davis, J. C., Statistics and Data Analysis in Geology, 2 larly the high 206Pb/204Pb and Pb/Ce and low Nb/U are ed., 646 pp., John Wiley and Sons, New York, 1986. consistent with this component containing recycled oce- Devey, C. W., F. Albarède, J.-L. Cheminée, A. Michard, R. 87 86 Mühe, and P. Stoffers, Active submarine volcanism on anic crust and sediment. The low Sr/ Sr component also the Society Hotspot Swell (West Pacific): a geochemical appears to be enriched in incompatible elements relative to study, J. Geophys. Res., 95, 5049-5066, 1990. MORB. This low 87Sr/86Sr component may be common Duncan, R. A., and I. McDougall, Linear Volcanism in to many plumes. French Polynesia, J. Volcanol. Geotherm. Res., 1, 197- • The geochemical evolution of Society volcanoes re- 227, 1976. Duncan, R. A., and W. Compston, Sr-isotope evidence for flects their passage over a concentrically zoned plume. The an old mantle source region for French Polynesian volca- plume consists a hot core containing recycled oceanic crust nism, Geology, 4, 728-732, 1976. and sediment surrounded by a sheath consisting of the Duncan, R. A., M. R. Fisk, W. M. White, and R. L. Nielsen, plume common component (variously called “C”, “PHEM” Tahiti: geochemical evolution of a French Polynesian or “FOZO”) entrained from the deep mantle. volcano, J. Geophys. Res., 99, 24341-24357, 1994. Farley, K. A., J. H. Natland, and H. Craig, Binary mixing of enriched and undegassed (primitive ?) mantle components Acknowledgements: This work was supported by NSF grants EAR85-09012 to RAD, and EAR87-07484 to WMW. We thank (He, Sr, Nd, Pb) in Samoan lavas, Earth Planet. Sci. Lett., R. Brousse, T. Gisbert, and H.G. Barsczus for providing many of 111, 183-199, 1992. the samples used in this study, and M. M. Cheatham for analytical Gillot, P.-Y. and P. Nativel, Eruptive history of the Piton de assistance and general support. We also thank H. Staudigel, F. la Fournaise volcano, Reunion Island, Indian Ocean, J. Albarède, D. Weis, and C. Hémond for assistance in the field and Volcanol. Geotherm. Res., 36, 53-66, 1989. J. Bischert-Toft for careful proofreading of the manuscript. This Hanan, B., and D. Graham, A common deep source for paper benefited from constructive reviews by Ken Farley and Barry mantle plumes: evidence from lead and helium isotopes, Hanan. WMW is grateful for the privilege of having worked with EOS, 75, No. 16, 67, 1994. both Mitsunobu Tatsumoto and George Tilton and would particu- Harpp, K. S., Geochemical Variations In a Single Basaltic larly like to acknowledge the support of “Tats” at an early stage in ° his career. In recognition of their enormous contributions to iso- Flow At 9 30’N On the East Pacific Rise, in et al. (eds) tope geochemistry, we dedicate this work to them. Proceed. Ocean Drill. Prog., Sci. Results , vol. 142, ed- ited by R. Batiza et al., Ocean Drilling Program, College Station, in press, 1995. REFERENCES Hart, S. R., E. H. Hauri, L. A. Oschmann, and J. A. Whitehead, Mantle Plumes and entrainment: isotopic evi- Ben Othman, D., W. M. White, and J. Patchett, The geo- dence, Science, 256, 517-520, 1992. chemistry of marine sediments, island arc magma gene- Hart, S. R., J. G. Schilling, and J. L. Powell, Basalts from sis, and crust-mantle recycling, Earth Plan Sci Lett, 94, Iceland and along the Reykjanes Ridge: Sr isotope 1-21, 1989. geochemistry., Nature Physical Science, 246, 104-107, Binard, N., R. C. Maury, G. Guille, J. Talandier, P. Y. Gillot, 1973. and J. Cotten, Mehetia Island, South Pacific: geology and Hauri, E. M., J. A. Whitehead, and S. R.Hart, Fluid dy- WHITE AND DUNCAN: NEW EVIDENCE FOR DEEP MANTLE RECYCLING 23

namic and geochemical aspects of entrainment in mantle tionation in the sedimentary system and crustal recycling plumes, J. Geophys. Res., 99, 24275-24300, 1994. into the Earth’s mantle, Earth. Planet. Sci. Lett., 69, 365- Hedge, C. E., Strontium isotopes in basalts from the Pacific 378, 1984. Ocean Basin, Earth Planet Sci. Lett., 38, 88-95, 1978. Plank, T., and J. N. Ludden, Geochemistry of Sediments in Hémond, C., C. W. Devey, and C. Chauvel, Source com- the Argo Abyssal Plain at Site 765: a continental margin positions and melting processes in the Society and Aus- reference section for sediment recycling in subduction tral plumes (South Pacific Ocean): element and isotope zones, in Proceed. Ocean Drill. Prog., Sci. Results, vol. (Sr, Nd, Pb, Th) geochemistry, Chem. Geol., 115, 7-46, 123, edited by F. M. Gradstein Ludden, J. N., et al., pp. 1994. 167-189, Ocean Drilling Program, College Station, 1992. Hofmann, A. W. and W. M. White, Ba, Rb, and Cs in the Prestvik, T. and R.A. Duncan, The geology and age of Peter Earth’s mantle, Z. Naturforsch., 38, 256-266, 1983. I Oy, Antarctica, Polar Res., 9: 89-98, 1991. Hofmann, A. W. and W. M. White, Mantle plumes from Rudnick, R. L., and S. L. Goldstein, The Pb isotopic com- ancient oceanic crust, Earth Planet. Sci. Lett., 57, 421- positions of lower crustal xenoliths and the evolution of 436, 1982. lower crustal Pb, Earth Planet. Sci. Lett., 98, 192-207, Hofmann, A. W. and W. M. White, The role of subducted 1990. oceanic crust in mantle evolution, Carnegie Inst. Wash- Schilling, J.-G., Iceland mantle plume: geochemical study ington Yearb., 79, 477-483, 1980. of the Reykjanes Ridge, Nature, 242, 565-571, 1973. Hofmann, A. W., K. P. Jochum, M. Seufert, and W. M. White, Sun, S. S., Lead Isotopic study of young volcanic rocks from Nb and Pb in oceanic basalts: new constraints on mantle mid-ocean ridges, ocean islands and island arcs, Phil. evolution, Earth. Planet. Sci. Lett., 79, 33-45, 1986. Trans. R. Soc. London, A297, 409-445, 1980. Jöreskog, K. G., J. E. Klovan, and R. A. Reyment, Geologi- Taylor, S. R., and S. M. McLennan, The Continental Crust: cal Factor Analysis, 178 pp., Elsevier, Amsterdam, 1976. Its Composition and Evolution, 312 pp., Blackwell Sci- Kato, T., A. E. Ringwood, and T. Irfune, Experimental de- entific Publications, Oxford, 1985. termination of element partitioning between silicate Vroon, P. Z., M. J. van Bergen, G. Klaver, and W. M. White, perovskites, garnet and liquids: constraints on early dif- Sr, Nd, and Pb isotopic and trace-element signatures of ferentiation of the mantle, Earth Planet. Sci. Lett., 89, the East Indonesian sediments: Provenance and implica- 123-145, 1988. tions for Banda Arc magma genesis, Geochim. Kato, T., T. Irifune, and A. E. Ringwood, Majorite partition Cosmochim. Acta, in press, 1995. behavior and petrogenesis of the Earth’s upper mantle, Wasserburg, G. J., and D. J. DePaolo, Models of earth struc- Geophys. Res. Lett., 14, 546-549, 1987. ture inferred from neodymium and strontium isotopic Kay, R. W., and S. M. Kay, Creation and destruction of lower abundances, Proc. Natl. Acad. Sci. USA, 76, 3594-3598, continental crust, Geol. Rund., 80, 259-278, 1991. 1977. Klein, E. M., and C. H. Langmuir, Ocean ridge basalt chem- Woodhead, J. D., P. Greenwood, R. S. Harmon, and P. istry, axial depth, crustal thickness and temperature varia- Stoffers, Oxygen isotope evidence for recycled crust in tions in the mantle, J. Geophys. Res., 92, 8089-8115, 1987. the source of EM-type oceanic island basalts, Nature, 362, Levi, S., H. Audunsson, R.A. Duncan, L. Kristjansson, P.- 809-813, 1993. Y. Gillot, and S.K. Jakobsson, Late Pleistocene geo- White, W. M., A. R. McBirney, and R. A. Duncan, Petrol- magnetic excursion in Iceland lavas: Confirmation of the ogy and Geochemistry of the Galapagos: Portrait of a Laschamp excursion, Earth Planet. Sci. Lett., 96, 443- Pathological Mantle Plume, J. Geophys. Res., 93, 19533- 457, 1990. 19563, 1993. Longerich, H. P., Oxychlorine ions in inductively coupled White, W. M., and A. W. Hofmann, Sr and Nd isotope plasma-mass spectrometry: Effect of chlorine speciation geochemistry of oceanic basalts and mantle evolution, - as Cl and ClO4. Journal of Analytical Atomic Spectrom- Nature, 296, 821-825, 1982. etry, 8, 439-444, 1993. White, W. M., and B. Dupré, Sediment subduction and Macdonald, G. A., and T. Katsura, Chemical composition magma genesis in the Lesser Antilles: Isotopic and trace of Hawaiian lavas, J. Petrol., 5, 82-133, 1964. element constraints, J. Geophys. Res., 91, 5927-5941, McKenzie, D. P., and R. K. O’Nions, Mantle reservoirs and 1986. ocean island basalts, Nature, 301, 229-231, 1983. White, W. M., and P. J. Patchett, Hf-Nd-Sr and incompat- Miller, D. M., S. L. Goldstein, and C. H. Langmuir, Ce- ible-element abundances in island arcs: implications form rium/lead and lead isotope ratios in arc magmas and the magma origins and crust-mantle evolution, Earth. Planet enrichment of lead in the continents, Nature, 368, 514- Sci. Lett., 67, 167-185, 1984. 519, 1994. White, W. M., The sources of ocean basalts: Radiogenic Morgan, W. J., Convection plumes in the lower mantle, isotopic evidence, Geology, 13, 115-118, 1985. Nature, 230, 42-43, 1971. Zindler, A., E. Jagoutz, and S. Goldstein, Nd, Sr and Pb- Newsom, H. E., W. M. White, K. P. Jochum, and A. W. isotopic systematics in a three component mantle: a new Hofmann, Siderophile and chalcophile element abun- perspective., Nature, 298, 519-523, 1982. dances in oceanic basalts, Pb isotope evolution and growth of the Earth’s core, Earth Planet. Sci. Lett., 80, 299-313, 1986. W. M. White, Department of Geological Sciences, Snee Hall, Patchett, P. J., W. M. White, H. Feldmann, S. Kielinczuk, Cornell University, Ithaca, NY 14853 and A. W. Hofmann, Hafnium/rare earth element frac- R. A. Duncan, College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis, OR 97331