Mineralogy and Composition of theOceanic Mantle

KEITH PUTIRKA1*, F. J. RYERSON2, MICHAEL PERFIT3 ANDW. IAN RIDLEY4

1CALIFORNIA STATE UNIVERSITY, FRESNO, DEPARTMENT OF EARTH AND ENVIRONMENTAL SCIENCES,

2576 E. SAN RAMON AVE., MS/ST25, FRESNO, CA 93740-8039, USA2INSTITUTE OF GEOPHYSICS AND PLANETARY PHYSICS, LAWRENCE LIVERMORE NATIONAL LABORATORY, L-396,

LIVERMORE, CA 94550, USA3UNIVERSITY OF FLORIDA, DEPARTMENT OF GEOLOGICAL SCIENCES, BOX 112120, GAINESVILLE, FL 32611-2120, USA4US GEOLOGICAL SURVEY, DENVER FEDERAL CENTER, MS 973, DENVER, CO 80225, USA

RECEIVED DECEMBER 4, 2009; ACCEPTED NOVEMBER 9, 2010ADVANCE ACCESS PUBLICATION JANUARY 8, 2011

The mineralogy of the oceanic basalt source region is examined by

testing whether a peridotite mineralogy can yield observed whole-rock

and olivine compositions from (1) the Hawaiian Islands, our type

example of a mantle plume, and (2) the SiqueirosTransform, which

provides primitive samples of normal mid-ocean ridge basalt. New

olivine compositional data from phase 2 of the Hawaii Scientific

Drilling Project (HSDP2) show that higher Ni-in-olivine at the

Hawaiian Islands is due to higher temperatures (T) of melt gener-

ation and processing (by c. 3008C) related to the Hawaiian

mantle plume. DNi is low at high T, so parental Hawaiian bas-

alts are enriched in NiO. When Hawaiian (picritic) parental

magmas are transported to shallow depths, olivine precipitation

occurs at lower temperatures, where DNi is high, leading to high

Ni-in-olivine. Similarly, variations in Mn and Fe/Mn ratios in

olivines are explained by contrasts in the temperatures of magma pro-

cessing. Using the most mafic rocks to delimit Siqueiros and

Hawaiian Co and Ni contents in parental magmas and mantle

source compositions also shows that both suites can be derived from

natural peridotites, but are inconsistent with partial melting of nat-

ural pyroxenites.Whole-rock compositions at Hawaii and Siqueiros

are also matched by partial melting experiments conducted on perido-

tite bulk compositions. Hawaiian whole-rocks have elevated FeO

contents compared with Siqueiros, which can be explained if

Hawaiian parental magmas are generated from peridotite at 4^5

GPa, in contrast to pressures of slightly greater than 1 GPa for melt

generation at Siqueiros; these pressures are consistent with olivine

thermometry, as described in an earlier paper. SiO2-enriched

Koolau compositions are reproduced if high-Fe Hawaiian parental

magmas re-equilibrate at 1^1·5 GPa. Peridotite partial melts from

experimental studies also reproduce the CaO and Al2O3 contents of

Hawaiian (and Siqueiros) whole-rocks. Hawaiian magmas have

TiO2 contents, however, that are enriched compared with melts

from natural peridotites and magmas derived from the Siqueiros

depleted mantle, and consequently may require an enriched source.

TiO2 is not the only element that is enriched relative to melts of nat-

ural peridotites. Moderately incompatible elements, such asTi, Zr,

Hf,Y, and Eu, and compatible elements, such asYb and Lu, are all

enriched at the Hawaiian Islands. Such enrichments can be ex-

plained by adding 5^10% mid-ocean ridge basalt (crust) to

depleted mantle; when the major element composition of such a mix-

ture is recast into mineral components, the result is a fertile peridotite

mineralogy.

KEY WORDS: mineralogy; mantle composition; partial melting;

Hawaii; Siqueiros

I NTRODUCTIONThere is a contradiction in our understanding of themineralogy and bulk composition of mantle componentsthat partially melt to produce oceanic basalts. Recent

*Corresponding author. E-mail: kputirka@csufresno.edu

� The Author 2011. Published by Oxford University Press. Allrights reserved. For Permissions, please e-mail: journals.permissions@oup.com

JOURNALOFPETROLOGY VOLUME 52 NUMBER 2 PAGES 279^313 2011 doi:10.1093/petrology/egq080 at U

niversity of Florida on M

arch 17, 2011petrology.oxfordjournals.org

Dow

nloaded from

petrological studies (Sobolev et al., 2005, 2007; Herzberg,2006; Dasgupta et al., 2010) suggest that a peridotite min-eralogy is incapable of producing the whole-rock majorelement compositions of many oceanic basalts, nor thecompositions of olivine crystals contained therein. Thesestudies amplify a model by Hauri (1996), who was perhapsthe first to propose pyroxenite (or partial melts fromsuch) as an admixture into the Hawaiian mantle. Thesenew studies suggest that pyroxenite-derived melts form asmuch as 30% of erupted mid-ocean ridge basalt (MORB)compositions, and that pyroxenite is the dominant lith-ology beneath the Hawaiian Islands (Sobolev et al., 2007),and is otherwise ever-present in undisclosed amountsbeneath other ocean islands (Dasgupta et al., 2010).However, these petrological arguments for a pyroxenitesource are at odds with a tide of isotopic and geochemicaldata that have long indicated that (1) only small amounts(510%) of enriched components occur in the mantlesources of Hawaiian and other ocean island basalt (OIB)magmas (e.g. White & Hofmann, 1982; Weaver, 1991;Bennett et al., 1996; Stracke et al., 1999; Willbold &Stracke, 2006; Prytulak & Elliott, 2007), and (2) a pyroxen-ite mineralogy in particular is inconsistent withmineralogy-sensitive trace element and isotopic systemat-ics (Norman & Garcia, 1999; Putirka, 1999; Stracke et al.,1999; Pietruszka et al., 2006; Elkins et al., 2008). We exam-ine this contradiction by testing whether Hawaiian bas-alts, the apparent type-example of a pyroxenite-fedsystem, can indeed be derived by partial melting of perido-tite, and by comparing Hawaiian lavas with relativelyprimitive MORB from the Siqueiros Transform, a near-by segment of the East Pacific Rise spreading ridgesystem.The islands of Hawaii contain one of the largest volca-

noes on the planet, and may represent Earth’s most intensethermal mantle plume. To the extent that thermal mantleplumes reflect mantle convection, primitive Hawaiian vol-canic rocks (see Putirka et al., 2007) may represent ourclearest window into the composition and mineralogy ofEarth’s deep interior. Olivine compositions from picriticrocks from phase 2 of the Hawaii Scientific DrillingProject (HSDP2) (Fig. 1a; Rhodes & Vollinger, 2004;Stolper et al., 2004) provide a means to test models ofmantle plume mineralogy and composition. Volcanicrocks from the Siqueiros Transform, a first-order discon-tinuity at the East Pacific Rise, provide a valuable com-parison with HSDP2, as Siqueiros samples include

Fig. 1. (a) Variation of wt % MgO vs wt % SiO2 for HSDP2whole-rocks. High- and low-SiO2 HSDP2 groups have been recog-nized in both whole-rock (Rhodes & Vollinger, 2004) and glass(Stolper et al., 2004) compositions. High- and low-SiO2 groups areseparated using: SiO2¼ ^0·2407(MgO)þ 51 (see text). (b) Peridotitewhole-rock compositions from W. McDonough (personal com-munication) recast as mineral components (ol¼olivine;

Opx¼orthopyroxene; Cpx¼clinopyroxene), and plotted usingStreckeisen’s (1976) triangular diagram for ultramafic rocks; gt¼ gar-net; sp¼ spinel. (c) Cation sums on the basis of four oxygens vs NiO(wt %) for olivines from lherzolites from the GEOROC database.

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

280

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

some unusually mafic MORB (Perfit et al.,1996), generatedat typical upper mantle temperatures (Putirka et al.,2007), and because they contain high forsterite olivines.To compare the viability of pyroxenite and peridotite

source mineralogies, we examine the extent to whichvariations in the pressure^temperature (P^T) conditionsof partial melting (and crystallization) may explain someof the compositional parameters that have been used tosupport a pyroxenite model. We also examine whetherlava compositions from these localities can be generatedfrom natural peridotite compositions. For this latter com-parison, we make no restriction on potential peridotitesources. This strategy is in contrast to that of Dasguptaet al. (2010), who concluded that pyroxenite must be an im-portant source for OIB because ‘volatile-free peridotite’cannot explain OIB compositions. However, peridotitesare not singular in composition, and the sources ofHawaiian and other OIB are known to be volatile-bearing(e.g. Dixon et al., 1997, 2002).As to the specific arguments for pyroxenite, olivine

compositions from the Hawaiian Islands play a key rolein the recent debate, which motivates our collection ofhigh-precision olivine compositions from Siqueiros andHSDP2 samples. Clague et al. (1991) were perhaps thefirst to recognize the high Ni contents of olivines fromthe Hawaiian Islands, and Garcia (2002) showed thatNi contents are especially high at Koolau. Sobolev et al.(2005, 2007) suggested that for most, if not all,Hawaiian volcanoes, Ni-in-olivine is so high and Mncontents are so low as to preclude a peridotite mantlesource. Similarly, it has been argued that low Ca inwhole-rock compositions (Herzberg, 2006) and highSiO2 (Dasgupta et al., 2010) and Fe/Mn (Huang et al.,2007) require that the mantle source is rich in pyroxen-ite. Sobolev et al. (2005, 2007) and Herzberg (2006) sug-gested that the source for nearly all Hawaiian volcanoesis pyroxenitic, and is possibly olivine-free (Sobolev et al.,2005). We examine each of these petrological and geo-chemical issues.To be clear about the meaning of rock names, we rely

on the ternary diagram of Streckeisen (1976) (Fig. 1b).In this classification scheme, a pyroxenite is any ultramaficrock containing less than 40% olivine. A garnet clino-pyroxenite is a garnet-bearing rock falling within theclinopyroxenite corner of the diagram; this same rockcould be called an ‘eclogite’ if the pyroxene is rich inomphacite (CaMgSi2O6^NaAlSi2O6 solid solution).Most ‘peridotites’ referred to in this study fall within thefields of ‘lherzolite’ or ‘harzburgite’. When we use theterms pyroxenite and peridotite, we generally mean allmembers of these rock families, unless we specificallyrefer to samples as being ‘eclogite’, ‘harzburgite’ or ‘lherzo-lite’, etc. (Fig. 1).

BACKGROUND AND PR IOR WORKPeridotite vs pyroxenite as theHawaiian mantle sourceWashington (1925) may have been the first to propose thatthe upper mantle consists of ‘peridotite’, and Bowen (1928)discussed the possibility of generating basalt by decompres-sion melting of such. Ringwood (1962a, 1962b) refined theperidotite model [Washington’s (1925) ‘peridotite’ was,however, mineralogically very different from our modernconception of the rock], and developed the term pyrolite(1 part basaltþ 4 parts depleted peridotite) that is stillused today as a reference for the bulk composition andmineralogy of the upper mantle (e.g. McDonough & Sun,1995). This work was followed by partial melting experi-ments of Yoder & Tilley (1962) and Green & Ringwood(1967), which provided support to the peridotite model,and since then there has been much agreement that ocean-ic basalts derive by decompression melting of peridotite.This is not to say that the mantle is homogeneous.

Hawkesworth et al. (1979) may have been the first to suggestthat subducted oceanic crust is sampled at ocean islands.White & Hofmann (1982) and Hofmann & White (1982)envisioned that subduction causes oceanic lithosphere toaccumulate at or near the core^mantle boundary, andthat with heat supplied from the core, subducted materialcan regain buoyancy and rise to the surface as a compo-nent of thermal upwellings or mantle plumes (Morgan,1971). This component, if it is more fertile than ambientperidotite, could partially melt at greater depths andshould be preferentially sampled where the degree of par-tial melting is comparatively low (Pickering-Witter &Johnston, 2000). Zindler & Hart (1986) interpreted theisotopic signatures of hotspot volcanic rocks in terms ofthree different subducted materials, and subsequent iso-topic studies have shown that the mantle sources of oceanislands are indeed heterogeneous, and must containrecycled crustal materials (e.g. Stracke et al., 2005;Willbold & Stracke, 2006). White & Hofmann (1982) esti-mated that the isotopic compositions of ocean island vol-canic rocks can be explained by addition of 1^2%subducted sediment to mantle peridotite. Later iterations(e.g. Willbold & Stracke, 2006) still require no more thanabout 2% of subducted crust, added to a dominantly peri-dotite mantle, to explain isotopic variations [although insome variants of the model (e.g. Willbold & Stracke(2006), 90% of the source may be recycled when depletedmantle lithosphere is included].Comparing Hawaiian lava compositions with experi-

mentally generated partial melts of peridotite, and notingthat SiO2 contents (at a given FeO content) were too highto be explained by partial melting of peridotite, Hauri(1996) proposed a significant departure from these models,positing a source with significant eclogite. Koolau has thehighest SiO2 among the Hawaiian Islands (Frey et al.,

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

281

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

1994; Haskins & Garcia, 2004; Huang & Frey, 2005), andHauri (1996) postulated that the Koolau mantle containsup to 5% eclogite, which by selective partial melting(yielding high-SiO2 dacitic melts) contributes up to 20%of the total melt fraction (14% on average). By these samearguments, 8% of Mauna Kea and 9% of Mauna Loamagmas contain an eclogite-derived component (Hauri,1996).Additional support for a pyroxenite model derives

from observed olivine compositions. Sobolev et al. (2005,2007) suggested that the high NiO contents of olivines inHawaiian shield-building tholeiites can be produced onlyif the source has a mineralogy whereby Ni is incompatiblecompared with a peridotite source. Sufficiently low bulkdistribution coefficients for Ni are achieved in their modelby a pyroxenite source that is nearly olivine-free. Sobolevet al. (2007) also suggested that low MnO contents inHawaiian olivines provide further evidence of a pyroxenitesourceça consequence of higher mineral^melt partitioncoefficients for Mn in orthopyroxene and clinopyroxenecompared with olivine.The pyroxenite model of Sobolev et al. (2005, 2007) con-

trasts with the model of Hauri (1996) in that all Hawaiianshield-building lavas are generated from varying amountsof a pyroxenite source. This source is estimated to form upto 30% of the Hawaiian mantle, contributing as much as81% of the melt erupted at Koolau, up to 51% at MaunaLoa, and 45% at Mauna Kea. The Sobolev et al. (2005)model is also multi-staged. Recycled basalt (as eclogite orpyroxenite) partially melts at great depth, and these par-tial melts rise to react with overlying peridotite to producea high-MgO (18^20% MgO) olivine-free pyroxenitesource, which then partially melts at shallow depths toyield Hawaiian magmas. Herzberg (2006) further sug-gested that Hawaiian olivine Ni contents are not only toohigh, but that CaO contents for nearly all HSDP2whole-rocks are too low to be explained by partial meltingof peridotite alone. He concluded that the high-SiO2 andlow-SiO2 groups, and all but a few high-CaO glasses re-ported by Stolper et al. (2004), are derived from a pyroxen-ite source. Huang et al. (2007) also required a pyroxenitecomponent in the source of the Koolau volcanic rocks,based on their high Fe/Mn ratios; however, they calledupon much smaller contributions of pyroxenite-derivedpartial melts, up to 30% for Koolau and less than 10% forother Hawaiian volcanoes.Several studies, however, have called into question the

pyroxenite model. For example, Norman & Garcia (1999)showed that magmas parental to Hawaiian tholeiites arehigh in MgO (at least 13^17 wt % MgO), and could notbe simply generated by partial melting of eclogite (garnetpyroxenite) that represented subducted oceanic crust.They also noted that various trace element ratios are

inconsistent with generation from an eclogite source. Inan added complication, Hirschmann & Stolper (1996) fur-ther suggested that pyroxenite might be present beneathmid-ocean ridges. Putirka (1999) tested the pyroxenite-source model at both a mid-ocean ridge (Siqueiros) and aplume setting (Hawaii) and showed that for most pyroxen-ite bulk compositions, partial melts would have Sm/Yband Na/Ti ratios that are too high for natural Hawaiianor MORB compositions. In addition, existing isotope datafor Hawaiian volcanic rocks (e.g. Bennett et al., 1996;Stracke et al., 1999) constrain the amounts of recycled ma-terials to be very small. Finally, Waters et al. (in press)showed that fertile garnet peridotite is no less suitablethan garnet pyroxenite to explain enriched MORB at theEast Pacific Rise.These studies have placed doubts on the role of pyroxen-

ite, but did not account for high SiO2 at Koolau (Hauri,1996), although other studies have attempted to addressthe issue. Submarine lava samples show that Koolau iscomposed mostly of high-MgO picrites (Garcia, 2002),just as at other Hawaiian volcanoes, although at any givenMgO content these picrites are still higher in SiO2 andlower in CaO than other Hawaiian lavas. SiO2 contentscan be increased through melt^peridotite reactions, where-by pyroxene is dissolved and olivine is precipitated, produ-cing a SiO2-enriched liquid residue (Eggins, 1992; Wagner& Grove, 1998; Stolper et al., 2004). Huang & Frey (2005)disagreed with this mechanism, at least for HSDP2samples, because olivine precipitation should decrease Niconcentrations, and Rhodes & Vollinger (2004) showedthat the high-SiO2 HSDP flows do not have lower Ni con-centrations than low-SiO2 flows. But experiments byMatzen et al. (2009) showed that Ni contents in olivinecan be controlled by the conditions of melt generation.Norman et al. (2002) allowed that dacitic partial meltsfrom eclogite may be important at Koolau (so as to explainhigh SiO2), but that trace element ratios for otherHawaiian volcanoes are inconsistent with an eclogiticsource for Hawaiian volcanoes in general. Stolper et al.(2004) allowed for a pyroxenite component for a smallsubset of HSDP2 glass compositions, which have a combin-ation of intermediate SiO2, and elevated CaO, TiO2 andAl2O3.Perhaps the most compelling objection to the pyroxenite

hypothesis derives from isotope ratios. Studies by Strackeet al. (1999), Pietruszka et al. (2006), and Elkins et al.(2008) showed that a pyroxenite mineralogy is unlikelyto be important at Hawaii, or at other oceanic islands.Nevertheless, for the peridotite hypothesis to remainviable, it must be possible to explain the apparently highNiO and low MnO in Hawaiian olivines, and high SiO2,high Fe/Mn and low CaO contents for some Hawaiianwhole-rocks.

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

282

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

Olivine compositions of Hawaiianvolcanic rocksIt is generally accepted that olivine plays an important rolein the evolution of Hawaiian magmas, and several decadesof study have led to some very important findings aboutthe magmatic activity in general. Macdonald (1944) notedthat during the 1840 Kilauea eruption, olivine phenocrystswere present in low-altitude eruptions but absent at higherelevations, thus providing evidence for the importance ofolivine settling in shallow magma reservoirs (Powers,1955). Murata & Richter (1966) showed that the amountof olivine carried by any particular flow is correlated tothe vigor of eruption, implying that high-energy eruptionsentrain olivine residues.Powers (1955) and Wright & Fiske (1971) also demon-

strated that the compositions of many Hawaiian lavas aregoverned by the addition or removal only of olivine, a pro-cess termed ‘olivine control’. Such work led to an importantquestion: might some olivines represent xenocrysts derivedfrom disaggregated peridotite, rather than representingcrystals precipitated from a liquid? In a key study of the1959 Kilauea eruption, Helz (1987) showed that various‘petrographic classes’ established on the basis of olivinezoning or deformation patterns are effectively part of acompositional continuum. She concluded that deformed orkink-banded textures do not provide evidence of xenocrys-tic origin, but rather that most deformed olivines formedas liquid precipitates, forming deep-seated aggregatesalong the walls of the Hawaiian magma plumbing system.Baker et al. (1996) and Garcia (1996) showed that most oliv-ines from Mauna Kea (HSDP1 core samples), whetherstrained or unstrained, are closely related to their hostmagmas, and so Helz’s (1987) conclusions appear to be gen-erally applicable. Baker et al. (1996) also showed throughmass-balance calculations that Mauna Kea magmasappear to entrain much of the olivine that is precipitatedby parental liquids. Kurz et al. (2004) confirmed the rela-tionship between olivine and host-rocks by showing that3He/4He in olivine correlates with the Nd and Pb isotopicratios of the host lavas, consistent with a genetic relation-ship between magmas and their entrained olivine. Finally,Garcia et al. (1995) showed that CaO contents in olivinescan be an effective discriminant between those that precipi-tated from a melt, which have high CaO contents, andthose that have equilibrated at subsolidus conditions,which have lower CaO contents (see Putirka et al., 2007).Similarly, Norman & Garcia (1999) showed that olivineCaO contents generally follow the CaO and CaO/Al2O3

contents of their host whole-rocks, consistent with the viewthat deformed olivines are magmatic precipitates.We attempt to build on this foundation with a study of

olivine compositions from HSDP2, and high-forsterite oliv-ines from the SiqueirosTransform.

METHODSNew olivine compositions are from (1) the HSDP2 core,which sampled an upper portion of Mauna Loa flows, butmostly sampled subaerial and submarine Mauna Keaflows (Garcia et al., 2007), and (2) submarine flows fromthe Siqueiros Transform, which offsets the East PacificRise at �8825’N (Perfit et al., 1996).We also examine pub-lished data from the subaerial Makapuu stage of Koolauvolcano (Frey et al., 1994; Haskins & Garcia, 2004; Huang& Frey, 2005) at Oahu and HSDP2 whole-rocks (Rhodes& Vollinger, 2004). Electronic Appendix A contains tablesof data and calculations. Appendix B provides additionaldetails for calculation methods, and sample calculations.Electron microprobe analyses of HSDP2 olivines

(Electronic Appendix, Supplementary Tables A1a andA2a, available at http://www.petrology.oxfordjournals.org) were performed on a five-spectrometer JEOL 8200at Lawrence Livermore National Laboratory (LLNL).Initially, data were collected using an accelerating voltageof 15 kV and primary beam current of 20 nA, using afocused beam (Supplementary Table A1). Counting timeswere 20 s on peak and 10 s at each background position,yielding standard deviations from counting statistics of3^7% for Ni, depending upon concentration. X-ray inten-sities were converted to concentration using the CITZAFcorrection program (Armstrong, 1995) using the followingstandards: diopside (Si), anorthite (Al), MgO (Mg),Fe2O3 (Fe), wollastonite (Ca), spessartine (Mn) and a syn-thetic Ni-olivine (Ni). Olivine from high-MgO SiqueirosTransform lavas (Supplementary Tables A1b and A2b)were initially analyzed using a JEOL 8800 at the USGSlaboratory in Denver, CO, with an accelerating voltage of15 kV, sample current of 20 nA and a spot size 10 mm.Counting times were as at LLNL, except for Ni, withcounting times of 200 s on peak and 100 s on background.To compare precision, we also reanalyzed a subset ofboth HSDP2 and Siqueiros samples at LLNL using beamconditions of 200 nA, and counting times of 200 s on peakand 100 s on background (Table A2a and b). Our compari-sons of these low-current, ‘low-precision’ (20 nA) andhigh-current, ‘high-precision’ (200 nA) datasets indicatethat Mn and Ca data are much more precise when ana-lyzed at the higher beam currents and longer countingtimes, but that there are no statistically significant norother discernible differences with regard to NiO and for-sterite (Mg2SiO4, or Fo) contents between these datasets.For comparisons of HSDP2 and Siqueiros CaO and MnOcontents in olivines we thus use only electron microprobedata collected at the higher beam currents, whereasall other comparisons make use of all data in Tables A1and A2.Tests for mineral^whole-rock equilibrium make use of

whole-rock compositions from Rhodes & Vollinger (2004).

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

283

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

For the comparison of olivine compositions with theirwhole-rock hosts, we use the methods outlined by Putirkaet al. (2007).Rhodes & Vollinger (2004) recognized that HSDP2

samples can be divided into four geochemical types basedon major oxides and trace elements. The most striking div-ision involves high- and low-SiO2 groups (Fig.1); a bimodaldistribution verified by the analyses of HSDP2 glasses(Stolper et al., 2004). Because the high- and low-SiO2

groups might be generated as the product of partial melt-ing of two dissimilar source materials (e.g. Herzberg,2006), or by the fractionation of a single parent, followingdivergent liquid lines of descent (LLD) or contrasting evo-lutionary paths (e.g. Stolper et al., 2004), an approximatelyequal number of samples were selected from both thesegroups, for the determination of olivine compositions.In an attempt to quantify the relationships, and to easeinter-sample comparisons, the two SiO2-based groups aredistinguished using the following equation:

SiO2 ¼ �0 � 2407ðMgOÞ þ 51 ð1Þ

where SiO2 and MgO are in weight per cent (Fig. 1a).HSDP minor and trace element abundances are fromHuang & Frey (2003) and Rhodes & Vollinger (2004).

Calculating mantle partial melts andsource compositions, subsolidus equilibriaand crystal lines of descentThe trace element contents of hypothetical mantle sourcesare calculated from the trace element contents of parentalliquids. Mineral^melt distribution coefficients for mostelements are from Salters & Longhi (1999) and Salters &Stracke (2004). Partition coefficients forVand Cr are fromUlmer (1989), Hauri et al. (1994), Canil (1999), and Adam& Green (2006). Partitioning relationships of TiO2, Ni,Co and Mn are given below. Additional calculation detailsare presented in Appendix B. Table 1 shows calculated par-ental melt compositions, and Table 2 and Appendix Bshow the mantle source compositions, calculated using dif-ferent values for melt fraction (F) and assumed mantlemineralogies.The trace element contents of HSDP2 paren-tal magmas (Table 1) are calculated from the mean ofwhole-rock samples with 15%4MgO421% (Huang &Frey, 2003; Rhodes & Vollinger, 2004), and from HSDP2sample SR641, which are thought to approximate primitive(nominally parental) liquids (see Putirka et al., 2007). Forthe Siqueiros MORB source, we use trace element datafrom Hays (2004), averaging all samples with 49·5%MgO (also in equilibrium with phenocryst olivine) to de-termine the composition of a parental Siqueiros magma.Except for Ni and Na, which are sensitive to temperature(T) and pressure (P) (see below) respectively, the samemineral^melt partition coefficients are used to calculateboth HSDP2 and Siqueiros mantle sources (Salters &

Stracke, 2004). Our model peridotite mineralogy atHawaii is 60% olivine, 20% orthopyroxene,15% clinopyr-oxene and 5% garnet. For Siqueiros, we use a source min-eralogy of 57% olivine, 28% orthopyroxene, 13%clinopyroxene, and 2% spinel (Workman & Hart, 2005).We also consider other mineral proportions, based on ex-perimental partial melting studies (Walter, 1998). Modelsfor a pyroxenite source assume a mineralogy of 21·4%garnet and 57·1% clinopyroxene [an average fromSobolev et al. (2005)].To evaluate the role of temperature in determining trace

element contents of magmas and crystals, we use partitioncoefficients for MnO and NiO that are calibrated fromexperimental studies. Many partition coefficients arehighly correlated with the partition coefficient of MgO,

Table 1: Trace element compositions of parental liquids

Element Parental liquids

Hawaii* MORBy

(HSDP2) (Siqueiros)

Rb 3·28 3·36

Ba 72·40 23·40

Th 0·66 0·53

K 0·20 0·10

U 0·22 0·27

Nb 9·45 4·73

La 7·79 6·26

Ce 20·30 16·00

Sr 237·30 135·00

Nd 14·10 11·91

Sm 3·80 3·72

Na2O 1·52 2·16

Zr 115·53 91·69

Hf 2·65 2·85

Eu 1·33 1·33

TiO2 1·87 0·97

Y 21·48 29·40

Yb 1·59 3·45

Lu 0·20 0·51

V 237·00 123·70

Co 78·6 71·2

Cr 905·00 854·10

Ni 830·00 342·00

*HSDP2 parental liquid is derived from HSDP2 sampleSR641, corrected by olivine addition to 21% MgO.yMORB parental liquid is derived from mean concentrationsof Siqueiros lavas with49·5% MgO (Hays, 2004) correctedto 14·6% MgO by addition of olivine. Elements are in ppm;Na2O and TiO2 are in wt. %.

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

284

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

DMg (Jones, 1984), and it is common to express trace elem-ent partition coefficients as a function of DMgO (e.g.Wang& Gaetani, 2008). However, DMg is a proxy for the truethermodynamic control on partitioning (i.e.T ), as changesin partitioning behavior, including that for the highly T-sensitive DMg, are determined by differences in partialmolar entropies between liquid and crystalline phases.When partition coefficients are used as explicit functionsof T, they allow a direct analysis of the effects of partialmelting and crystallization driven by changes in tempera-ture. Wang & Gaetani (2008) calibrated the partitioningof NiO and MnO between olivine and coexisting liquid(DNiO

ol^liq, DMnOol^liq) at 1 atm and relatively low tem-

peratures (1201^13258C).Variations in DNiOol^liq have long

been known to be especially sensitive to T (Arndt, 1977),as well as bulk composition, and Wang & Gaetani (2008)also showed a dependence on SiO2 (expressed as a ratioof bridging to non-bridging oxygens). Because we suspectthat Hawaiian magmas are generated at higher

temperatures than those explored by Wang & Gaetani(2008), we recalibrate DNiO

ol^liq and DMnOol^liq using

high-temperature experimental data from the Library ofExperimental Phase Relations (LEPR: Hirschmann et al.,2008, appendix C), so as to minimize model extrapolation,obtaining

Dol�liqNiO ¼ exp½�3 � 257þ 6800=Tð�CÞ� ð2aÞ

Dol�liqNiO ¼ exp½�4 � 75þ 0 � 033ðSiOliq

2 Þ þ 6829=Tð�CÞ�

ð2bÞ

Dol�liqNiO ¼ exp½ � 1 � 78þ 0 � 04ðSiOliq

2 Þ

� 0 � 04ðMgOliqÞ þ 3236=Tð�CÞ�

ð2cÞ

Dol�liqMnO ¼ exp½ � 2 � 76þ 3583=Tð�CÞ�: ð2dÞ

In these equations, DNiOol^liq and DMnO

ol^liq are the con-centration ratios of NiO or MnO (in weight per cent)

Table 2: Inferred Hawaiian mantle compositions as a function of melt fraction (F) and residual mineralogy (trace elem-

ents are in ppm; oxides are in wt %)

Mantle compositions for HSDP2 (15–21% MgO) Mantle compositions for Siqueiros (14·6% MgO)

Peridotite Peridotite Wehrlite Wehrlite Peridotite Peridotite Harzburgite Harzburgite

F: 0·1 0·2 0·1 0·2 0·08 0·1 0·08 0·1

Rb 0·329 0·657 0·329 0·657 0·028 0·056 0·017 0·028

Ba 7·248 14·487 7·244 14·484 1·096 2·190 0·657 1·095

Th 0·069 0·135 0·067 0·134 0·006 0·012 0·004 0·006

K2O 0·021 0·041 0·020 0·040 0·001 0·002 0·001 0·001

U 0·024 0·046 0·023 0·046 0·003 0·006 0·002 0·003

Nb 1·099 2·028 1·046 1·980 0·054 0·104 0·033 0·053

La 0·813 1·588 0·803 1·579 0·101 0·199 0·060 0·099

Ce 2·934 4·864 2·496 4·474 0·434 0·794 0·224 0·371

Sr 29·5 52·6 26·7 50·1 6·4 12·5 3·7 6·2

Nd 2·604 3·881 2·007 3·351 0·506 0·957 0·275 0·458

Sm 1·085 1·387 0·738 1·078 0·235 0·408 0·109 0·181

Zr 33·9 42·9 24·2 34·4 7·6 11·8 3·0 4·8

Hf 0·812 1·016 0·536 0·771 0·331 0·442 0·080 0·133

Eu 0·466 0·562 0·370 0·477 0·111 0·177 0·073 0·100

TiO2 0·310 0·483 0·269 0·447 0·145 0·236 0·068 0·106

Y 14·5 15·3 8·2 9·6 6·1 8·3 5·0 5·9

Yb 1·5 1·5 0·8 0·9 0·3 0·5 0·2 0·2

Lu 0·294 0·284 0·147 0·153 0·059 0·085 0·031 0·043

V 181·4 187·6 113·6 127·3 38·5 53·6 18·1 24·8

Co 65·6 66·1 60 61·1 52·5 52·7 50·9 51

Cr 984·5 975·7 1371·7 1319·9 1524·4 1456·1 1515·6 1489·9

Ni 1871 1756 1726 1627 1957 1804 2381 2304

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

285

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

between olivine and liquid; SiO2liq and MgOliq are the

weight per cent values for SiO2 and MgO in the coexistingliquid;T is in degrees Celsius. Equation (2b) is calibratedusing only high-Texperiments (T¼1300^20208C) whereasequations (2c and 2a) are calibrated over the entireT range available from the LEPR database (1139^20208C).For equation (2a), the the standard error of estimate(SEE)¼�0·36, R2

¼0·85 and n¼ 81. For equation (2b)the SEE is �0·88, R2

¼0·88 and n¼ 49; for equation (2c)the SEE is �1·23, R2

¼0·93 and n¼ 81. For equation (2d)the SEE is �0·14, R2

¼0·87 and n¼ 93, where T rangesfrom 1020 to 20208C. Experimental data are listed in theElectronic Appendix. The models are derived from linearleast-squares regression analysis, which iteratively excludesdata points falling outside 3s variation of model error.These models are not necessarily better than theT-inde-pendent models published byWang & Gaetani (2008) andothers; equations (2a)^(2c) are useful in that they allowus to quantify directly the effects of Tas a potential causeof differences in Ni concentrations, and because by usinghigh-Tdata for calibration there is no model extrapolationwhenT520208C.For Na partitioning, the following T- and P-sensitive

expression is used:

Dcpx�liqNa2O ¼ ½ � 3 � 525þ 1046 � 1=Tð�CÞ

þ 1062 � 4PðGPaÞ=Tð�CÞ�ð3Þ

where the partition coefficient, DNa2Ocpx^liq, expresses

the weight per cent ratio of Na2O between clinopyroxeneand liquid, CNa2O

ol/CNa2Oliq. Equation (3) is calibrated

from 1235 partial melting studies (Putirka, 1999;Putirka et al., 2007). The SEE is �0·21, R¼ 0·68. We alsocalculate TiO2 in the Hawaiian source, here usingpartition coefficients derived from data presented byWalter (1998) (mean values are DTiO2

garnet^liquid¼ 0·291;

DTiO2clinopyroxene^liquid

¼ 0·073; DTiO2orthopyroxene^liquid

¼

0·077) and from the LEPR database (DTiO2olivine^liquid

¼

0·03; DTiO2orthopyroxene^liquid

¼ 0·028; DTiO2clinopyroxene^

liquid¼exp[^4·95þ4907·7/T(8C)].

To illustrate the effects of fractionation on various com-positional parameters, we plot ‘crystal lines of descent’(CLD), the crystal complements to liquid lines of descent(LLD). These calculations use the partitioning relation-ships noted in this section, with curves that track the evolu-tion of olivine over a range of T. For our CLD curves weuse a Hawaiian parental liquid composition as calculatedby Putirka et al. (2007), with an initial NiO content of 0·14wt % and MnO content of 0·17 wt %, where T andDMg

ol^liq are calculated using the models of Beattie (1993)and Putirka et al. (2007). Equilibrium olivine is subtractedin 1·5% increments [assuming that KD(Fe^Mg)ol^liq¼0·33; we use 0·33 as a middle ground between ourmass-balance considerations and estimates by Matzenet al. (2011); see the section on ‘Olivine^whole-rock

equilibrium’, below], withTand olivine compositions recal-culated at each increment (calculated temperature rangesare c. 1200^15008C). NiO contents for CLDs are calculatedusing equation (2a) and MnO contents using equation(2d). Additional calculation methods, and sample calcula-tions and for the 1 atm CLD, are given in Appendix B.For subsolidus equilibria, we use the olivine^ortho-

pyroxene partitioning relationship for NiO/MgO derivedby Witt-Eickschen & O’Neill (2005). Their work showedthat CNiO

ol/CMgOol¼ (CNiO

opx/CMgOopx)exp[1419/T(K)

�0·241].

Whole-rock and olivine databasesTo test whether the composition inferred for the Hawaiiansource (based on lava flow compositions) is permissiblyencompassed by natural peridotite compositions, we usegarnet-peridotite (n¼ 519) and spinel-peridotite (n¼ 503)compositions generously provided by W. McDonough(Electronic Appendix Table A3a provides mean, medianand standard deviation values for a number of elementsfor these peridotites, as well as several quantile values);these data are compared with data from GEOROC(Electronic Appendix Table A3b). All plots of peridotitewhole-rock trace element contents use data fromW. McDonough (personal communication; ElectronicAppendix Table A3b). Pyroxenite whole-rock data andlherzolite olivine compositions are from GEOROC.For some of our tests, we recast the major element con-tents of peridotites into mineral components (olivineþclinopyroxeneþ orthopyroxeneþAl-bearing phase,garnet or spinel), using the methods of Thompson (1982);details are given in Appendix B.For GEOROC olivine compositions, we check the qual-

ity of the analyses, and whether NiO contents are affectedby such quality, by comparing NiO contents against olivinecation sums (ideally 3·0 on a four-oxygen basis). Ofthe 495 olivine compositions where major elements arereported, only two have cation sums outside the range2·95^3·05 (0·4%) (Fig. 1c), and NiO contents are uncorre-lated. We conclude that olivine compositions reported inGEOROC are only very rarely of mistaken identity orpoorly analyzed, and that the reported NiO contents faith-fully represent the NiO contents of olivine; 493 olivinecompositions reported in the GEOROC database fromlherzolites are used.

RESULTSHSDP2 (and Koolau) olivine compositions:Fo, Ca and Ni contents, Fe/Mn ratiosThe ranges of Ni, Mn and Fo (forsterite, Mg2SiO4)contents in olivines recovered from the HSPD2 core are re-markably similar, regardless of the depth at which theywere recovered, or whether they derive from low- orhigh-SiO2 suites (Fig. 2). There is also no correlation

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

286

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

Fig. 2. (a) Forsterite (Fo) contents of olivines from the HSDP2 core as a function of core depth and host-whole-rock classification (see Fig. 1for definition of high- and low-SiO2 compositions); both high- and low-precision data are plotted (Electronic Appendix A: Tables A1 and A2).(b) NiO in olivine vs Fo content (Tables A1 and A2) for HSDP and Koolau olivines. (c) High-precision (200 nA) data (Table A2) showingCaO vs NiO in HSDP and Siqueiros olivines HSDP2; also plotted are lower precision (20 nA) data for olivines from Koolau (Garcia, 2002).(d) The same data as in (c); CaO content in olivine vs Fo content (mol %). (e) FeO/MnO in olivine vs Fo content. Two crystal lines of descent(CLDs; see Methods) are shown, calculated at 1 GPa and 1 atm (T, not P is the important intrinsic parameter), to illustrate how variations inFeO/MnO can be affected by fractional crystallization conditions. (See text for discussion.) Symbols as in (c).

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

287

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

between mean or maximum Fo contents and d18O as deter-mined byWang et al. (2003), just as there is no discernibledifference between d18O in either of the high- or low-SiO2

suites (Wang et al., 2003). Maximum Fo contents may bereflective of mineral^liquid equilibration temperatures(Roeder & Emslie, 1970; Putirka, 2005); however, max-imum Fo contents for both the high- and low-SiO2 suitesare identical at 90·5 (exclusive of post-shield flows fromRuns 120^152), as are the distributions of Fo contents.Maximum Fo contents also do not vary systematicallywith depth for either suite (Fig. 2a), and most flow unitscontain olivines that approach the global maximum of90·5 recovered for Mauna Kea. This maximum appearsto be highly reproducible as it matches the maximumvalue of 90·5 determined by Baker et al. (1996).The only evident vertical change in maximum Fo con-

tent is associated with the change from Mauna Kea toMauna Loa volcanic rocks, as the latter yield a highermaximum Fo content of 91·0 (sample SR23), slightlyhigher than the maximum value (90·7) obtained by Bakeret al. (1996) (Fig. 2a). Although the differences in maximumFo values between Mauna Loa and Mauna Kea are small,they are sustained through a large number of olivine ana-lyses, and so appear to be real.CaO, MnO and NiO contents in olivines also do

not vary systematically with core depth, and high-and low-SiO2 HSDP suites have similar NiO contents(Fig. 2b). Compared with HSDP2 samples, Siqueiros oliv-ines have similar NiO contents, but have higher Fo. TheHSDP2 suites define diverging trends with regard to CaOcontents. Both high- and low-SiO2 suites have similarCaO contents at high NiO and high Fo (Fig. 2c and d),and both evolve along what appear to be a high- andlow-Ca path, but with a tendency for low-SiO2 HSDP2lavas to fall on the higher CaO path (Fig. 2c), similar toSiqueiros olivines. These diverging evolutionary paths forCaO are not matched by differences in olivine concentra-tions for MnO, Al2O3 or P2O5 for either the high- orlow-SiO2 suites. Because CaO contents are identical athigh Fo and high NiO contents, these differences arealmost certainly unrelated to any differences in sourceregion or parent magmas, but instead reflect contrastingliquid and crystal evolution. One possibility is that thehigher CaO contents reflect lower pressures for partialcrystallization for low-SiO2 magmas. Finnerty & Rigden(1981), for example, showed that CaO contents in olivinesare higher at lower pressures; using their model, low-SiO2

suite magmas would have crystallized olivine at an averagepressure that is 3·3 kbar (19·7 kbar vs 16·4 kbar) lowerthan for high-SiO2 HSDP2 magmas. However, anothermore likely possibility is that differences in CaO contentreflect differences in the SiO2 activity; Davidson &Lindsley (1994) showed that CaO contents in olivine arelower in systems with high SiO2, as Ca2SiO4 reacts with

SiO2 to form a wollastonite (Ca2Si2O6) component inclinopyroxene. The Davidson & Lindsley (1994) reactionmechanism is qualitatively consistent with the model ofStolper et al. (2004), whereby high-SiO2 magmas are gener-ated by the reaction of low-SiO2 magmas with wall-rocks,through the dissolution of pyroxene and precipitation ofolivine. Koolau olivines (from Garcia, 2002) appear totrend to lower CaO contents (see Garcia, 2002), but thesedata were collected at low beam current (20 nA), and sodifferences in CaO may reflect differences in analyticalprecision.Among the Hawaiian Islands, pyroxenite has been

called upon to explain elevated whole-rock SiO2 (Hauri,1996; Takahashi & Nakajima, 2002) and olivines that havehigh FeO/MnO (Sobolev et al., 2007), but no correlationof these parameters occurs among high-precision data.For HSDP2 samples with Fo486, low-SiO2 HSDP2 flowshave a mean FeO/MnO of 74�2·8 (n¼ 352; range¼ 67^82·6) (Fig. 2e), statistically equivalent to the mean FeO/MnO of 75·8�1·6 for high-SiO2 Koolau flows (Sobolevet al., 2007; data are converted to weight ratios).High-SiO2 HSDP2 flows in contrast have intermediateFeO/MnO, at 72�2·8 (n¼ 238; range¼ 64·6^80·4).The range of FeO/MnO ratios at any given Fo content at

Fo586 is substantial, and because partition coefficientsfor FeO and MnO are not precisely unity, it is possible toaffect FeO/MnO ratios by fractionation. We illustrate thispotential by showing CLD paths for olivines precipitatingfrom a Kilauea primitive magma (Putirka et al., 2007) con-taining 0·17 wt % MnO, and precipitating olivine at 1GPa and 1 atm (Fig. 2e); these curves match the observedFeO/MnO trends. Small differences in FeO or MnOwithin a peridotite source can almost certainly also affectFeO/MnO in olivine (peridotites have a median FeOt/MnO of 61·3, and a 99·5% quantile of 112·6). In any case,our high-precision data do indicate significantly lowerFeO/MnO ratios for high-Fo olivines at Siqueiros, whichhave a mean of 65�2·4 (n¼ 82; range¼ 59·4^72·1;Fo ranges from 89·3 to 91·1). This matches the patternof lower FeO/MnO in MORB shown by Humayun et al.(2004), and is consistent with crystallization at lower P^Tconditions.

Olivine^whole-rock equilibriumThe Rhode’s diagram (Fig. 3) provides a convenient test forolivine^liquid equilibrium, providing a graphical portrayalof Fo contents of olivines [Fo¼XMgO

ol/(XMgOolþXFeO

ol),where Xi

ol is the mole fraction of i in olivine] comparedwith 100�Mg-number [where Mg-number¼XMgO

liq/(XMgO

liqþXFeO

liq) and Xiliq is the mole fraction of i in

the liquid] for putative liquid compositions. To calculatethe equilibrium curve we assume that Hawaiian liquidsand olivines equilibrate at an fO2 near the magnetite^wu« s-tite oxygen buffer (although there is aT-sensitivity, whichcan be addressed through iterative calculations, the

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

288

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

70

75

80

85

90

95

35 45 55 65 75 85 95

Mg# Whole Rock

Fo

in O

livin

e

65

70

75

80

85

90

95

30 40 50 60 70 80 90

Mg# Whole Rock

Fo

in O

livin

e

HSDP-2Maximum Fo Contents

KD(Fe-Mg)ol-liq = 0.32±0.03 fO2 = MW

KD(Fe-Mg)ol-liq = 0.32±0.03 fO2 = MW

HSDP-2 OlivineCore Compositions

HSDP-2 Mean(of Max Fo)

SR641

(a)

(b)

Closed-

system

fraction-

ation

Fig. 3. (a) Fo contents of olivine core compositions vs host whole-rock Mg-number. All such olivines for sample R0641 (SR641) fall within 1sof the anticipated equilibrium values. The equilibrium curve is calculated assuming that fO2 is buffered at magnetite^wu« stite (MW) andthat the Fe^Mg exchange coefficient, KD(Fe^Mg)ol^liq¼ 0·32�0·03. (b) Maximum Fo contents in olivine vs the Mg-number of each hostwhole-rock. The HSDP-2 mean is consistent with a minimum value for KD(Fe^Mg)ol^liq of 0·32.

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

289

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

relationship is effectively FeO¼x[Fe2O3total], where

x¼ 0·90^0·91, and where FeO and Fe2O3 are in weightper cent). We also assume that most olivines equilibrateat 10 kbar, and so calculate that the Fe^Mg exchangecoefficient between olivine and liquid, KD(Fe^Mg)ol^liq

(¼ ([XFeOol][XMgO

liq])/([XMgOol][XFeO

liq])) is 0·32(Putirka, 2005). Composition-sensitive models of Herzberg& O’Hara (1998) and Toplis (2005) yield similar results.When whole-rock compositions are compared with oliv-

ine core compositions, the data cluster to the right of theequilibrium curve defined by KD(Fe^Mg)ol^liq¼ 0·32 forthe data as a whole, and for nearly all individual flowunits. However, olivine core compositions from sampleSR641 fall very near the equilibrium curve; all are withinthe 1s error bounds of 0·32�0·03, indicating that thisparticular whole-rock is a plausible liquid composition(Fig. 3a). On this diagram, closed-system equilibrium frac-tionation would lead to a vertical array owing to the in-variant whole-rock composition, with the maximum Focontent plotting on the equilibrium curve, and subsequentlower temperature precipitates falling directly below themaximum Fo. An additional test, then, is to compare onlythe maximum Fo contents for each flow (Fig. 3b) (seeBaker et al. 1996). By this test, it appears that, on thewhole, Hawaiian picrites (Fig. 4) carry a representativefraction of their precipitates as suggested by Baker et al.(1996). This argument implies that the KD(Fe^Mg)ol^liq isindeed 0·32, but as noted by Putirka et al. (2007), it may behigher, as Fig. 3b probably places a minimum value onKD(Fe^Mg)ol^liq. The reason for this is that as KD(Fe^Mg)ol^liq decreases, the equilibrium curve shifts to the left.With such a shift, more samples would then lie to the rightof the curveçin the direction of olivine accumulation. Toallow the mean of all erupted products to fall to the right ofthe curve implies that high-density, olivine-rich slurries are

preferentially erupted over their less dense slurry-free coun-terparts, which seems unlikely. This analysis is consistentwith the value for KD(Fe^Mg)ol^liq of 0·34 determined byMatzen et al. (2011). Table 3 shows estimates for mantle po-tential temperatures for volcanoes from the island ofHawaii and maximum forsterite contents (Garcia et al.,1995; Norman & Garcia, 1999); here we assume equivalentpressures of 3·0 GPa, although observed variations in SiO2

may well represent variations in equilibration P.

DISCUSS IONNi and Mn contents in olivineA key issue in support of pyroxenite in the mantle sourceof oceanic basalts hinges on the interpretation of high Niin Hawaiian olivines, and their comparison with olivinesfrom lherzolites. As we show below, the comparisonbetween peridotite and magmatic olivine compositions isirrelevant, as these suites are affected by different kindsof equilibria, and the high Ni in Hawaiian olivines canbe generated by precipitation of olivine from picritic(high-MgO, high-NiO) liquids (generated from typicalperidotite) at low P andT.To illustrate the reason for high NiO in olivine, we

plot three CLDs calculated at 1 atm, 10 kbar and 20 kbar(Fig. 5a); initial NiO content is 0·14 wt % and the calcu-lated T range is c. 1200^15008C. The HSDP2 NiO^Foarray is consistent with olivine fractionation at about 10kbar on average; Siqueiros olivines with Fo contents of

0

5

10

15

20

25

0 5 10 15 20 25 30

MgO

FeO

KSDP (Koolau)

Mauna Loa

Mauna Kea

Kohala

Kilauea

Hualalai

Fig. 4. MgO vs FeOt (FeO total) for volcanoes from the island ofHawaii and for Koolau (source: GEOROC; see text for calculation).At elevated MgO, FeOt contents are effectively constant among vari-ous Hawaiian volcanoes; this similarity indicates that the most maficof samples from the Hawaiian Islands are produced at similarmantle potential temperatures, unless olivines vary in composition orfO2 varies in their respective source regions.

Table 3: Temperature estimates for Hawaiian basalts

Volcano Fomax Tol–liq (8C) Tp (8C)

Kohala 90·1 1530 1640

Mauna Kea 90·3 1538 1648

Hualalai 89·9 1522 1632

Kilauea 90·6 1550 1660

Loihi 90·3 1538 1648

Mauna Loa 91·3 1580 1690

Fomax is the maximum forsterite content observed at eachvolcano; Tol–liq (8C) is the temperature of olivine–liquidequilibration at 3 GPa, assuming KD(Fe–Mg)ol–liq¼ 0·32,and constant FeO in the liquid (Fig. 4). Tp (8C) is mantlepotential temperature, assuming an approximate 1108Ccombined correction for the heat of fusion, and adiabaticascent (Putirka, 2008a).

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

290

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

89·5^91·0 fall on a 1 atm crystallization trend precipitatingfrom liquids that contain 0·030^0·046 wt % NiO (or 235^360 ppm Ni). As we discuss below, these contrasts in NiOcontents between Siqueiros and Hawaii are unlikely to berelated to differences in source materials, but are insteadmore probably due to contrasts in the conditions of partialmelting.The CLD curves of Fig. 5a were calculated using equa-

tion (2a), as this composition-independent model bettercaptures that actual curvature of the natural Fo^NiOtrends. Most importantly, olivines with the highest NiOand Fo contents can be generated by precipitation fromliquids with modest NiO contents, derived from typicalperidotite (Fig. 5b; see below). For example, at 1 atm and1400^14108C, a Hawaiian liquid with 0·125 wt % NiOand 18·8% MgO will precipitate an Fo90·5 olivine withNiO¼ 0·60 wt % (see Appendix B).The high NiO contentresults because when olivine precipitation occurs at low P,theTof equilibration is lower, and DNi is higher; as a con-sequence, basaltic magmas can precipitate olivine withNiO contents much higher than the NiO-in-olivine oftheir mantle source materials. Most Hawaiian olivineshave much lower NiO contents, and can be generated byfractionation at the modest P inferred by Garcia (2002)and indicated by Putirka (2008c). A few HSDP2 olivinesdisplay very low Fo contents at a given NiO content(Fig. 5a), but NiO enrichment in the source is not required.Clinopyroxene-rich ankaramites are not uncommon atMauna Kea (Frey et al.,1991), and by adding clinopyroxeneto the fractionating assemblage (a 50^50 mix of oliv-ineþ clinopyroxene is shown) olivines that fall to thelow-Fo (high-NiO) side of the Hawaiian CLD trend canbe explained. High-NiO olivines from Koolau plot veryclose to the 1 atm olivine CLD, indicating that somesubset of Koolau picritic magmas (but by no means all)have precipitated olivine at relatively low temperaturescompared with other Hawaiian magmas.If we take the GEOROC lherzolite olivine data at face

value, lherzolite olivines are actually not much differentin NiO compared with Hawaiian olivines (Fig. 5a). A keydifference in olivine compositions is not so much thatHawaiian olivines are enriched in NiO at a given Fo(their ranges in Ni are effectively equivalent), but ratherthat Hawaiian olivines have lower Fo for a given NiO(lherzolites have higher Fo contents overall, especially atlow Ni). The reason for this most probably is that magmat-ic differentiation more effectively drives Fo and NiO con-tents to lower values compared with subsolidus processes(magmatic effects notwithstanding; e.g. Kelemen et al.,1998). To illustrate this, we use the NiO/MgO partitioningrelationships of Witt-Eickschen & O’Neill (2005) andobserved orthopyroxene compositions from Hawaii to cal-culate the expected sub-solidus equilibrium of NiO con-tents for coexisting Hawaiian lherzolitic olivines. We

allow olivine compositions to vary from Fo85 to Fo95 (tomatch the cluster of observed Fo contents for lherzolites),and then calculate NiO in olivine at 8008C, 11008C and13008C, using the mean of observed orthopyroxene NiO/MgO ratios (¼ 0·0028). The Witt-Eickschen & O’Neill(2005) relationships show that NiO contents in olivine arehigher at lower T. For a fixed amount of MgO in thesystem, decreasingTdrives NiO in olivine upwards, whichcan yield the near-vertical array of lherzolite compositions.In addition, data from GEOROC show that mean NiO/MgO in orthopyroxenes from Hawaiian lherzolites are ef-fectively equivalent to all orthopyroxenes in theGEOROC lherzolite database, which average 0·0029, indi-cating no particular NiO enrichments or depletions in theHawaiian source. Natural variations in lherzolite bulkNiO/MgO undoubtedly also affect lherzolite olivine NiO/MgO. Hawaiian lherzolitic orthopyroxenes, for example,have NiO/MgO ratios that range from 0·0018 to 0·004,which are used to calculate the two 9008C isotherms ofFig. 5a. These 9008C isotherms bracket nearly the entirerange in lherzolite olivine NiO contents.As a further test of Ni contents in mantle sources, we cal-

culate the NiO contents of olivines that can precipitatefrom a liquid with a given amount of NiO; for example,963 ppm as in HSDP2 sample SR641, which we infer to ap-proximate a liquid (Putirka et al., 2007). The gray field inFig. 5a shows that if a liquid with a fixed amount of Ni isallowed to pond at different depths, hence forcing olivineprecipitation at different temperatures, the resulting oliv-ines can have very different Ni contents. Olivines fromHSDP2 samples have NiO contents that are broadly similarto those of lherzolites, but are also consistent with precipita-tion from liquids that have NiO contents equivalent tothose of their host whole-rocks (which, as we show below,can in turn be derived from typical peridotites).The highestNiO contents in both Koolau and HSDP2 olivines mostprobably reflect the partial crystallization of mafic(high-MgO, high-NiO) magmas at low temperatures.Another argument that has previously been used to sup-

port a pyroxenite source for Hawaiian magmas is thatHawaiian olivines have low MnO contents (Sobolev et al.,2007), and that such MnO depletion, coupled with Ni en-richment, is a consequence of partial melting of a pyroxen-ite source. As a test of this argument, we compare MnOand Fo contents for HSDP olivines with the compositionsof olivines from Koolau, natural lherzolites and volcanicrocks from the Siqueiros transform (Fig. 5c). The pyroxen-ite model predicts lower MnO contents for magmas withhigh SiO2 (such as at Koolau and for the high-SiO2

HSDP2 samples) and high NiO (as at Koolau). However,our high-precision data show that olivines from the high-and low-SiO2 HSDP2 groups have identical MnO con-tents, and both suites are broadly similar to samples fromKoolau and Siqueiros. The similarity between MnO

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

291

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

70 75 80 85 90 95 100

NiO

(w

t. %

)

Fo

20 kbar

1 atm

10 kbar

0

20

40

60

80

100

120

140

160

180

200

0 500 1000 1500 2000 2500 3000 3500

Ni (ppm)

Co

(pp

m)

Garnet peridotites

Spinel peridotites

Clinopyroxenites

(a)

(b)

Range of NiO in ol

precip. from liq.

with 963 ppm Ni

at 1375-1500oC

10 kbar 1 atm

1500oC

1375oC

1300oC

1100oC

800oC

900oC (max)

900oC (min)

50% ol + 50% cpx

Sub-solidus

equilib

Siqueiros, 1-atm

HSDP2 - Hi Si

HSDP2 - Low Si

Koolau

Siqueiros high Fo

Lherzolites

Hawaii (1500oC)

Hawaii (1665oC)

MORB (1200oC)

MORB (1350oC)

Fig. 5. (a) Comparison of NiO and Fo contents for MORB (Siqueiros), Hawaiian (HSDP2)- and lherzolite-hosted olivines. Curved lines(concave-up) are crystal (as opposed to ‘liquid’) lines of descent (CLD) calculated for Hawaii at 1 atm, and 10 and 20 kbar, assumingolivine-only fractionation, using the DMg

ol^liq from Putirka et al. [2007, their equation (2)], the geothermometer of Beattie (1993) and the as-sumption that the Fe^Mg exchange coefficient, KD(Fe^Mg)ol^liq¼ 0·33. Parental Hawaiian magmas begin with 0·14 wt % NiO (see Appendixfor sample calculations). A 1 atm CLD for Siqueiros is also shown. Dashed near-horizontal curve represents the effect of adding clinopyroxeneto the Hawaiian fractionating assemblage. Nearly horizontal light and dark gray continuous lines represent subsolidus equilibration betweenolivine and orthopyroxene; light gray curves are isotherms calculated using NiO/MgO in coexisting opx of 0·0028; the two 9008C isothermsuse minimum (0·0018) and maximum (0·004) values for NiO/MgO for Hawaiian lherzolitic opx. The gray field ranging from 0·63 to 0·41NiO, represents the NiO contents of olivines in equilibrium with a liquid containing 963 ppm Ni (e.g. sample SR641) crystallized at tempera-tures from 1375 to 15008C respectively.

(continued)

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

292

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

Fig. 5. Continued.(b) Comparison between calculated mantle source concentrations of Co and Ni for Siqueiros (MORB) and Hawaii, and natural peridotitewhole-rocks fromW. McDonough (personal communication) and natural pyroxenites from GEOROC. The partitioning of Ni between olivineand liquid is highly sensitive to T and is calculated using equations (2a), (2b) and (2c), and equation (10) of Wang & Gaetani (2008) (seeAppendix for a complete list of calculations), using the Hawaiian and MORB parental magmas of Putirka et al. (2007). Ni contents for equilib-rium liquids are estimated from the maximum and mean NiO contents of olivines (this study), using a DNi calculated from the mean of equa-tions (2a), (2b) and (2c), and assuming equilibration temperatures of 15008C for Hawaii and 13008C for MORB; Co and Ni partitioncoefficients are highly sensitive to mineralogy. For MORB, we use the source mineralogy of Workman & Hart (2005). For Hawaii, the assumedsource mineralogy is 60% olivine, 20% orthopyroxene, 15% clinopyroxene, and 5% garnet. If the equilibrium residual mantle mineralogy,at 20% melting, is 56·2% olivine, 31·4% clinopyroxene and 12·3% garnet (e.g. see Walter, 1998, experiment 50.6), then Ni contents forthe Hawaiian source are higher. The mantle source regions beneath mid-ocean ridges and Hawaii have effectively equivalent Co contents; Nicontents are also effectively equivalent, provided that equilibration in the Hawaiian mantle occurs at temperatures that are 250^3008C greaterthan for MORB (Ni in MORB liquids is almost certainly underestimated). (c) Comparison of MnO and Fo contents for olivines fromKoolau, HSDP2, the Siqueiros Transform (Perfit et al., 1996; Hays, 2004) and natural peridotites (GEOROC), using all available publisheddata, and data fromTable A1. (d) MnO and Fo contents are compared for high-precision HSDP and Siqueiros olivine analyses (Table A2) andhigh-precision olivine analyses from Koolau [Sobolev et al., 2007; table S2; ppm(X in XnO)¼ (n)(104)(mol wt X)(wt % XnO)/(mol wt XnO)].For both (c) and (d), gray lines show olivine-only liquid lines of descent, calculated at 10 kbar using equation (2c).The gray fields show how dif-ferences in temperature of equilibration can affect the content of MnO in coexisting olivines [equation (2c)]: olivines from MORB may haveslightly higher MnO contents as in (d) (Sobolev et al., 2007), but these can be explained if MORB-derived olivines equilibrate from liquidswith similar MnO contents, but crystallize at lower temperatures.

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

293

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

contents for the high- and low-SiO2 suites is especially evi-dent in our high-precision data (Fig. 5d).Some Kooalu olivines trend to lower MnO contents

(Fig. 5c), but high-precision data from Sobolev et al. (2007)(Fig. 5d) show that MnO (and FeO/MnO; see Fig. 2e) con-tents at Koolau are equivalent to our high-precision datafrom the HSDP2. Both Koolau and Siqueiros alsoshow MnO contents that trend to slightly higher MnO athigh Fo. However, as shown in Fig. 5c and d, such mildlyelevated MnO contents (if real) can be just as easily ex-plained by fractionation at lower temperatures (and prob-ably also lower pressures).A key observation is that liquids containing the same

amount of MnO can precipitate olivines with very differ-ent MnO contents, depending upon the depth (i.e. tem-perature) at which the olivines crystallize. Because NiOand MnO are less compatible at higherT, the presence ofthick crust or lithosphere can result in olivines with lowerMnO and NiO contents, as liquids trapped at greaterdepth should precipitate olivines at higher temperatures.This is illustrated in Fig. 5c and d, where the gray fieldsshow the range of MnO contents for olivines equilibratedwith a liquid containing a constant amount of MnO(0·17%, as in SR641) but that precipitates olivine over arange of temperatures. The gray curves in Fig. 5c and dtrace the paths of olivine-only CLDs using this calculatedrange of MnO contents, and initial olivine compositionsof Fo91·5.We conclude that differences in MnO contents re-flect temperature control on olivine precipitation, ratherthan differences in mantle source composition, and thatthe sub-parallel trends shown in fig. 1a of Sobolev et al.(2007) are proxies for thermal contours.

Ni contents in HSDP2 whole-rocks, andnatural peridotitesAs with olivines, HSDP2 whole-rock compositions alsoshow that the Hawaiian mantle is not enriched in Ni rela-tive to natural peridotites. To illustrate this, we calculatethe Ni content of the Hawaiian mantle source using the as-sumption that sample SR641 (18·3% MgO, 963 ppm Ni)and HSDP2 whole-rocks with similar bulk compositions(�3 wt % MgO) are plausible primary melt compositions.For SR641, olivine^melt equilibria at 20 kbar yield tem-peratures of equilibration of 1483^15438C [Beattie, 1993;Putirka et al., 2007, equations (2) and (4)] (assumptionswith regard to P have little intrinsic effect on the Ni calcu-lations, but have an effect on calculated T, which in turnaffects DNi). Models A1 and A2 from Putirka et al. (2007)yield estimates of F (melt fraction) of 0·10^0·17 for SR641.Data from the laboratory in Mainz yield a mean Ni con-tent of 741 ppm for HSDP whole-rocks with 15%5MgO521%, whereas the laboratory at UMassAmherst (Rhodes & Vollinger, 2004) yields an average of831 ppm for such samples. Using Ni¼ 831 ppm, F¼ 0·15,and modal melting with a bulk distribution coefficient

of DNi¼ 2·85 (DNiolivine^liq

¼ 3·90, where T¼15008C,SiO2¼47·2%; DNi

clinopyroxene^liquid¼1; DNi

garnet^liquid¼

0·5; DNiorthopyroxene^liquid

¼1·7), the Ni concentration in theHawaiian peridotite source would be 2137 ppm. Thisvalue is remarkably close to, but less than, the median(2198 ppm) and mean Ni (2320� 644 ppm; Appendix B,Table B3) contents in peridotites, and is within 1svariationof the peridotite mean. Using the Ni content of SR641yields a mantle source with 2477 ppm Ni. Although thereis clearly some uncertainty in determining Ni contents inthe Hawaiian source, these estimates show that theHawaiian source has Ni contents that are very close, if notidentical, to those of typical peridotites.We should emphasize that we use SR641 as a primary

melt because we suspect that its whole-rock compositionrepresents that of a liquid, and so we can estimate theTofolivine^liquid equilibration, the Ni content of the putativeprimitive liquid, and the melt fraction, F. We suspect thatHawaiian liquids are generated at higher temperatures(and pressures) than the olivine^liquid equilibrationT forSR641, and that will decrease our estimates of Ni in thesource region. But even if we hold Ni constant at 831 ppm,and F constant at 0·2, and then increaseT to the highestvalue obtained from olivine^liquid equilibria (16658C;Putirka, 2008a), calculated Ni contents for the sourceregion are still peridotite-like (1421^2381 ppm; Fig. 5b).And of course, magmas generated at higher T will havehigher MgO and Ni, and will be generated at lower F asP will be higher (they will be closer to the solidus), andthese changes would have the effect of increasing estimatesfor Ni in the mantle source region, offsetting the effectsof increasedT.

Is the MORB source depleted in Nirelative to Hawaii?Sobolev et al. (2005) used MORB as an example of apartial melt of a mantle source that is free of pyroxenite, al-though this was revised in a later paper (Sobolev et al.,2007), where pyroxenite was inferred to be present beneathMORs. In any case, Sobolev et al. (2005) showed that Nicontents in olivines from MORB are much lower than atHawaii, and indeed, MORB whole-rocks and glasses con-tain far less Ni than Hawaiian whole-rock samples. Thiscomparison holds for picrites from Siqueiros and HSDP2.Three samples from Siqueiros have MgO49·5% and con-tain 574�167 ppm Ni (Perfit et al., 1996; Hays, 2004), butthe remaining 36 samples from Siqueiros that average7·93�0·54% MgO have just 83�22 ppm Ni (Hays,2004). In contrast, HSDP2 rocks with 7%5MgO59%have 136�36 ppm Ni, much higher than the bulk of themafic samples at Siqueiros with similar MgO contents.However, in both cases, Ni contents are undoubtedly lowcompared with parental magmas owing to fractionationof olivine (Fig. 5).

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

294

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

To illustrate further, we compare both Co and Ni con-tents for the mantle sources for HSDP2 lavas and MORB,as represented by high-MgO samples from Siqueiros(Hays, 2004) (Fig. 5b). The value for DCo is assumed inde-pendent of temperature (DCo

olivine^liq¼ 2; DCo

clinopyroxene^

liquid¼ 3; DCo

garnet^liquid¼ 2; DCo

orthopyroxene^liquid¼ 4·2).

We assume that F¼ 0·08 for MORB and F¼ 0·2 atHawaii (Putirka, 2008a). Mantle mineralogies are as dis-cussed in the Methods section and Appendix B. For theNi contents of primitive MORB magmas, we averagethe maximum (0·382 wt % NiO) and mean (0·326 wt %NiO) NiO contents of Siqueiros high-Fo olivines (0·354 wt% NiO; Table A2), and assume a temperature of olivine^liquid equilibration of 13008C to obtain a bulk distributioncoefficient of 8·0 [using equations (2a)^(2c)]; this yields aNi content of 350 ppm for the Siqueiros parental liquids(see Appendix B), which is what we derive from our CLDanalysis of Fig. 5a.We use 831 ppm Ni for Hawaiian primi-tive magmas. For calculated Ni contents for the source re-gions in Fig. 5b, we use equations (2a)^(2c) and equation(10) from Wang & Gaetani (2008); all calculations areshown in Appendix B.There is considerable uncertainty in calculating Co and

Ni contents in basalt source regions, as Co and Ni contentsin primary liquids are not easily obtained. However, asillustrated in Fig. 5b, Siqueiros and Hawaiian basalts canpotentially be derived from sources with nearly identicalCo and Ni contents, provided that the Hawaiian source is�3008C hotter than MORB. The thermal contrasts arelessened if primitive Siqueiros picrites have greater NiOcontents (the thermal contrasts are zero if the highest Nicontents for Siqueiros picrites are used as input), and areincreased if Hawaiian magmas are generated at lower F.Perhaps more importantly, Co and Ni contents for boththe MORB and Hawaiian sources match peridotite com-positions very well, but are a very poor match for pyroxen-ite. This result is valid regardless of whether we useequations (2a)^(2c) or theWang & Gaetani (2008) model[their equation (10)] to calculate DNi

ol^liq, or whether weadopt higher Ni contents for primary Siqueiros magmas.Of course, the temperatures used in Fig. 5b are not mantlepotential temperatures, but rather they represent a rangeof temperatures at which olivine may have equilibratedwith primitive Hawaiian and MORB magmas. But if wecan accept that Hawaii is underlain by a thermal plumewith an excess temperature of c. 3008C (e.g. Putirka et al.,2007) then similar thermal contrasts in their respectivemagma plumbing systems should be expected, in whichcase their mantle sources may have equivalent mineralo-gies and very similar Ni contents.

HSDP whole-rock compositions comparedwith peridotite partial meltsIf NiO contents provide evidence against a pyroxenitesource, what can be said for CaO or SiO2? Here, we

compare Hawaiian lava compositions with the compos-itions of 285 experimentally produced liquids that areeither equilibrated with a peridotite mineral assemblageor ultimately derived as partial melts from a peridotitebulk composition (Fig. 6). The peridotite-derived liquidsused for comparisons involve both fertile and infertile, an-hydrous and hydrous bulk compositions, and liquidsderived at low F (with a lherzolite mineralogy) and highF (with a harzburgite or ‘dunite’ residue). We use thesecompositions as we suspect that (1) the mantle is not singu-lar in composition, (2) high values of Fmay be appropriateat Hawaii, and (3) most lava compositions, even when cor-rected to high MgO (e.g. Dasgupta et al., 2010) or to be inequilibrium with a given mantle olivine composition,are not necessarily direct mantle partial melts, and arenot necessarily co-saturated with olivineþ clinopyrox-eneþ orthopyroxene� garnet or spinel (see Kinzler &Grove, 1992). We also compare natural Hawaiian compos-itions with experimental pyroxenite-derived liquids fromKogiso & Hirschmann (2001) and Keshav et al. (2004),and we compare experimentally produced liquids withhigh-SiO2 compositions from Koolau (Frey et al., 1994).We wish to emphasize that melting is undoubtedly incre-

mental and polybaric, and so there is no expectation thatbecause a natural whole-rock composition matches a par-ticular experiment that a primary mantle-derived magmacomposition has been identified. However, primarymagmas are almost certain to lie along, or on extensionsof, natural geochemical trends and so a comparison of nat-ural and experimental compositions, especially at highMgO, should be useful.A key aspect of this comparison is that it is only

necessary that one or a few experimental liquids lie alongthe main HSDP2 compositional trend, preferably at highMgO (417% MgO), where we assume that parental li-quids will be situated (Norman & Garcia, 1999; Putirkaet al., 2007). The reason for this is that it is well establishedthat once on that trend, the remainder of the HSDP2compositions can be generated through the addition orsubtraction of (largely) olivine (Rhodes & Vollinger,2004). Our comparison shows that with regard to CaO,FeOt, SiO2, Al2O3 and other major oxides (Fig. 6),Hawaiian parental magmas can be produced as partialmelts of peridotite.The SiO2 contents of peridotite-derived liquids easily

encompass the entire range of SiO2 contents observedat the Hawaiian Islands, including the high-SiO2 subaerialKoolau volcanic rocks (Fig. 6a). Indeed, model peridotite-derived liquids calculated by Longhi (2002) at 1·0 GPaencompass the SiO2^MgO variation of high-SiO2 Koolausamples well. Moreover, although Herzberg (2006)argued that CaO contents at Hawaii are too low to be pro-duced from a peridotite source, Hawaiian liquids are onlylow in CaO relative to liquids produced by the

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

295

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

SiO

2

0 5 10 15 20 25 30 35MgO

Peridotite-derived liquidsWalter (1998)Longhi (2002) 1 GPaLonghi (2002) 3 GPa

Pyroxenite-derived liquids

Takahashi et al (1993) 4.6 GPa,1750oC

40

42

44

46

48

50

52

54

56

58

60

KoolauHSDP2 low-SiHSDP2 high-Si

(a) Herzberg & Zhang (1996)5 GPa,1726oC

Walter (1998)5 GPa,1680oC

Hawaiian Lavas

PM-K/O

CaO

2

4

6

8

10

12

14

(b)

Siqueiros

Fig. 6. (a)^(h) Major element compositions for natural whole-rocks from HSDP2 and Koolau compared with the compositions of experimen-tally generated partial melts (Falloon et al., 1988; Kinzler & Grove, 1992; Takahashi et al., 1993; Baker & Stolper, 1994; Longhi, 1995; Herzberg& Zhang, 1996; Kushiro, 1996; Kinzler, 1997; Robinson et al., 1998; Walter, 1998; Falloon & Danyushevsky, 2000; Pickering-Witter & Johnston,2000; Longhi, 2002). Experimental liquids are derived using peridotite starting compositions, or equilibrated with a peridotite (sensu stricto) min-eral assemblage. Several experimental partial melts are highlighted in the figure; PM^K/O is a mixture of three melt compositions:(0·225[K&G92]þ 0·525[F88]þ 0·25[F&D00], where PM^K/O signifies the parental magma for Koolau, K&G92 is experiment H154 fromKinzler & Grove (1992), F88 is experiment T-762 from Falloon et al. (1988) and F&D00 is experiment T-3572 from Falloon & Danyushevsky(2000). In (e) and (g) we also show data for Siqueiros (for picrites with 10·34^12·78 wt % MgO; Perfit et al., 1996; Hays, 2004) to comparepressures of partial melting. In (g) and (h), we compare FeOt [and MnO in (h)] contents for Hawaiian rocks and liquids from partial meltingexperiments; total Fe in HSDP2 rocks is reported as Fe2O3t (Rhodes & Volinger, 2004), which we recalculate to FeOt (a very close approxima-tion for HSDP2 samples is FeOt¼ 0·914[Fe2O3t]).

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

296

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

experiments of Herzberg & Zhang (1996) and Walter(1998). Other experimental studies, such as those ofFalloon et al. (1988; 3·0 GPa), Takahashi et al. (1993; 4·6^6·5 GPa), Kushiro (1996; 1·5^3·0 GPa, 1350^14808C),Falloon & Danyushevsky (2000; 1·5^2·0 GPa, 1350^15608C) and Longhi (2002; 2·8 GPa, 15738C) yield liquidsthat match CaO contents at Hawaii remarkably well(Fig. 6b). Ironically, pyroxenite-derived liquids have CaO

contents that are too high compared with Hawaiian sam-ples, although this difference might reflect the small data-set and consequent narrow range of starting compositionsexplored for these rock types. (Because pyroxenites are ef-fectively crystalline basalts, presumably one can easilyfind one to match any particular basaltic rock.) At lowMgO, Hawaiian rocks tend to have lower Al2O3 thanperidotite-derived experimental liquids (Fig. 6c), but

MgO

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

TiO

2

0

2

4

6

8

10

12

14

16

18

Al 2

O3

(c)

(d)

0 5 10 15 20 25 30 35

Fig. 6. Continued.

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

297

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

Rhodes & Vollinger (2004) showed that the Hawaiiantrend to lower Al2O3 is controlled by fractionation andmixing processes involving high-MgO magmas; the rele-vant comparison is that at high MgO, Hawaiian lavashave Al2O3 contents that are identical to those derivedfrom peridotite partial melting. If parental Hawaiianmagmas have 15% MgO, natural and experimentalAl2O3 contents are very similar.

Dasgupta et al. (2010) noted that OIB, including those ofHawaii, have high FeOt and SiO2, and argued for a pyrox-enite source; however, as with CaO and other majoroxides, SiO2 and FeOt contents at Hawaii can both be ex-plained by peridotite partial melting. Peridotite-derivedexperimental liquids that fall on the Hawaiian FeO^SiO2

trend (Fig. 6g) are from Falloon et al. (1988; 1·0^1·5 GPa,1330^14008C), Kinzler & Grove (1992; 1·1^1·6 GPa,

Fig. 6. Continued.

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

298

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

1280^13208C), Takahashi et al. (1993; 4·6 GPa, 17508C),Kushiro (1996; 2·5^3·0 GPa, 1470^15008C), Walter (1998;4·5^7·0 GPa, 1620^18108C) and Longhi (2002; 2·8 GPa,1510^15558C). Although we did not restrict our selection ofperidotite-based experiments, these peridotite-derivedexperimental liquids are furthermore nominally anhyd-rous, and so at Hawaii contradict the premise of

Dasgupta et al. (2010). It is perhaps worth noting that ourhighlighted experiments in Fig. 6 fall on a single line inFig. 6e, but not in Fig. 6g, because unlike FeOt, the SiO2

contents in peridotite partial melts are not a monotonicfunction of P; when using the Si-activity model of Beattie(1993) and additional compositional corrections to formu-late a barometer, however, global OIB fall on a single

FeO

tF

eOt/M

nO

SiO2

10

30

50

70

90

110

130

40 45 50 55 60

(h)

0

2

4

6

8

10

12

14

16

18

(g)

Global OIB Array

Siqueiros

Fig. 6. Continued.

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

299

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

curvilinear P vs FeOt trend (Putirka, 2008c), indicatingthat most OIB compositions can be related by partial melt-ing of a peridotite source at varying P andT, with P ran-ging from 1·7 to 3·6 GPa, when using the Si-activitybarometer of Putirka [2008c; equations (41) and (42)] andparental OIB magma compositions from Putirka (2008a).Although it is still true that most experimental liquids inFig. 6g yield lower FeOt contents at a given SiO2 thanHSDP samples, this represents less the failure of peridotitesto generate such liquids, and more the lack of experimentsconducted at either sufficiently high pressures or on pre-cisely the right types of peridotite bulk compositions.Returning to HSDP2 compositions, experiments by

Falloon et al. (1988; runs T-762 and T-781, at 1 GPa and1350 and 14008C), Takahashi et al. (1993; 4·6 GPa and17508C) and Kushiro (1996; 3 GPa and 14708C) each pre-cisely explain both FeOt and CaO contents at theHawaiian Islands, and collectively bracket HawaiianSiO2, Al2O3, and Na2O contents. Indeed, all but one(Takahashi et al., 1993) reproduce even the high TiO2 atHawaii, although the Takahashi et al. (1993) experiment,at 19·8% MgO, comes closest to reproducing the highMgO contents expected for Hawaiian parental magmas(e.g. Norman & Garcia, 1999; Putirka, 2005). It is not pre-cisely clear why these experiments more precisely explainHawaiian geochemical trends than, say, experiments byWalter (1998). The peridotite starting compositions,KLB-1 used by Takahashi et al. (1993) and PHN1611 usedby Kushiro (1996), have slightly lower Al2O3 (by 1% onaverage) and SiO2 (by 0·8% on average), and higherNa2O (0·1% on average), compared with the KR4003peridotite used by Walter (1998). However, even Walter’s(1998) partial melts approach Hawaiian compositions,with only slightly lower Al2O3 and slightly higher CaO,and all three peridotite types have very similar CaO con-tents (KLB-1, 3·44%; PHN1611, 3·26%; KR4003, 3·45%).Peridotites are in any case not singular in composition, andpartial melting does not occur at singular values of P andT, and partial melting experiments have associated errors.The clear implication of Fig. 6 is that peridotite-derived li-quids can reproduce Hawaiian compositions on average;for most oxides such experiments are perhaps most inform-ative collectively, rather than individually.Our comparisons nevertheless still reveal some import-

ant systematics. For example, regardless of experimentalbulk composition or the laboratory in which the experi-ments were performed, high FeO contents at Hawaiiare explained by peridotite partial melting only at high P

and T (i.e. between 3 and 5 GPa, and 1530^17508C) asbracketed by experiments and models by Takahashi et al.(1993), Kushiro (1996), Walter (1998) and Longhi (2002)(see Klein & Langmuir, 1987; Langmuir et al., 1992). Incontrast, FeOt at Siqueiros is best explained if most partialmelting occurs at pressures slightly greater than 1·0 GPa,

and less than 3·0 GPa (and F¼ 0·1) (Fig. 6e and g), whichis consistent with experiments that show that high-MgOSiqueiros glasses are saturated at 1·2^1·3 GPa with a peri-dotite residue (Wendlandt & Ridley, 1994), the work ofPerfit et al. (1996), which indicates partial melting condi-tions of 1240^13408C and 1 GPa, and estimates by Gregget al. (2009) indicating a mantle potential temperature of13508C. The significantly higher P estimates for partialmelting at Hawaii compared with Siqueiros are in agree-ment with our results for high temperatures of partial melt-ing to explain the contrasts in NiO contents, and arefurthermore in agreement with the high temperatures cal-culated at Hawaii (Putirka et al., 2007) from olivine^liquid equilibria (16658C) (mantle source independent),and models based on partial melting experiments byMaal�e (2004), which show that Hawaiian picrites with23% MgO should be saturated with a harzburgite residueat 36�5 kbar and 1680�508C. We emphasize that noneof these estimates are precise in detail, but collectivelythey provide a firm picture of the P^T conditions of meltgeneration at the Hawaiian Islands and the mid-oceanridge system.

A peridotite source for high-SiO2Koolau lavasVolcanic rocks erupted at Koolau (Frey et al.,1994; Haskins& Garcia, 2004; Huang & Frey, 2005), and exposed on thesubmarine flanks of the volcano (Garcia, 2002) have dis-tinct compositions: not only are SiO2 contents higher atany given MgO content compared with other Hawaiianlavas (Fig. 6a), but CaO and TiO2 are distinctlylower (Fig. 6b and d) and Al2O3 is subtly higher (Fig. 6c);such differences have motivated the development ofpyroxenite-bearing source models (Hauri, 1996; Huang &Frey, 2005). For a peridotite source to be valid at Koolau,it must be possible to simultaneously explain its SiO2,Al2O3, CaO and TiO2 contents, as well as the high FeOt

contents which generally characterizes Hawaiianmagmas. No particular experimental liquid has all thesecompositional characteristics, but several experimentsbracket the Koolau composition, and an accumulationof such liquids, as might be expected during polybaricmelting, can produce the Koolau compositions precisely.For example, although by no means a unique solution,the following mixture of peridotite-generated liquids wassuccessful in this regard:

PM�K=O ¼ 0 � 225½K&G92� þ 0 � 525½F88�

þ 0 � 25½F&D00�ð4Þ

where PM-K/O signifies the parental magma for Koolau.In equation (4) K&G92 is experiment H154 from Kinzler& Grove (1992; 1·3 GPa, 12808C; 47·8% SiO2; 7·38%CaO), F88 is experiment T-762 from Falloon et al. (1988; 1GPa, 14008C; 50·36% SiO2; 8·9% CaO) and F&D00 is

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

300

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

experiment T-3572 from Falloon & Danyushevsky (2000;1·5 GPa, 15008C; 51·9% SiO2; 6·74% CaO). These experi-ments were not conducted using identical bulk compos-itions or P^T conditions, but each of these liquids is theproduct of peridotite partial fusion, within a broadly simi-lar P range of 1^1·5 GPa, over a range inT, 1280^15008C.Other bulk compositions and P^Tconditions intermediatebetween these three experiments could yield the sameresult. This is not to imply that 1^1·5 GPa is the P rangeat which the Koolau parental melts were created. Senet al. (2005) used isotopic data to argue that the Koolaumagmas did not interact with the oceanic lithosphere(75^80 km, Bock, 1991); they concluded that rising Koolaumagmas traveled through a chemically insulated conduit,jacketed by an inert zone up to 2 km in thickness. Our re-sults indicate that the high-SiO2 samples at Koolau can beexplained by generation from a peridotite source, providedthat such liquids re-equilibrated at pressures of 1^1·5GPa(c. 45^50 km), within this insulated conduit, which ineffect is the model of Wagner & Grove (1998).Finally, although partial melts of a pyroxenite source can

yield the high FeO/MnO ratios (and high SiO2) observedfor whole-rocks at Koolau (Huang et al., 2007), Fig. 6hillustrates that a pyroxenite source is an unnecessary ex-planation for elevated FeO/MnO. First, partial melts fromperidotites span a very broad range of FeO/MnO ratios,from as low as 10 to4150. Many of the experiments shownin Fig. 6h have lower FeO/MnO compared with Hawaii(in large part because of the very high FeO in Hawaiianlavas, rather than Mn depletion; see Fig. 6e and g); how-ever, these peridotite-derived liquids have MnO contentsthat range from nearly zero to 40·4 wt %. Second, theinter-island variation among the Hawaiian Islands is negli-gible, trending orthogonal to the trend of experimentalpartial melts of pyroxenites (although, again, we acknow-ledge that this is a small pyroxenite dataset). Althoughsome Koolau olivines are Mn depleted (Fig. 5c), and somewhole-rocks have elevated FeO/MnO, neither of thesecharacteristics at Koolau exceeds observed variationsfrom lherzolite-derived olivines or peridotite-derived li-quids. Finally, even if high-precision olivine compositionsdo indeed yield variations in FeO/MnO ratios, the magni-tude of these differences, when compared with the rangeof FeO/MnO in olivines that can be generated by fraction-ation (Fig. 5e), is far too low to require anything butminor differences in source region composition, let alonesource mineralogy.

TiO2 enrichments in the Hawaiian sourcePrytulak & Elliott (2007) showed that highTiO2 is a char-acteristic of many ocean islands, including Hawaii, andconcluded that 1^10% of mafic crust, sediment or meta-somatic veins may contribute highTiO2 to mantle-derivedOIB magmas. Jackson et al. (2008) observed a positive cor-relation between TiO2 and 3He/4He ratios, and suggested

that this correlation is evidence for an isotopic mantlecomponent that is also distinct with respect to its majorelement composition. This correlation is not entirely clear.Putirka (2008a, 2008b) observed a correlation between3He/4He and mantle potential temperature (Tp), butTiO2 contents in the parental compositions of Putirka(2008a) do not correlate with isotope ratios orTp. Below,we examine the TiO2 and trace element contents ofHSDP2 lavas to test what kind of mantle source is requiredto produce primitive Hawaiian magmas.In the context of mantle mineralogy, two questions

are important. (1) Can peridotites dissolve sufficient TiO2

(or any other given element) to explain Hawaiian mag-matic TiO2 contents? If so, then (2) do natural peridotitescontain sufficient TiO2? As to the former question, experi-ments by Longhi (1995) (the highest TiO2 data in Fig. 6d)indicate the affirmative. These partial melts have up to6·25 wt % TiO2 at low F, and more than 4 wt % whenF40·2, well beyond the 1·86% TiO2 observed in HSDP2rocks with 15^21% MgO, or even the maximumTiO2 forparental OIB magmas of 2·8 wt % at Heard Island, as cal-culated by Putirka (2008a). Clearly, more than sufficientTiO2 can be dissolved within a peridotite mantle sourceto explain the TiO2 contents at any ocean islandçhighTiO2 does not intrinsically require a different mineralogy.However, the Longhi (1995) experiments used startingcompositions with as much as 2·17% TiO2, much higherthan that of most peridotites.To address question (2), we compare natural peridotite

TiO2 contents with those calculated for the Hawaiianmantle source, assuming that F¼ 0·1^0·2 (see Putirkaet al., 2007), using two contrasting mineralogies. We con-sider a garnet wehrlite residue [experiment 45.03 ofWalter (1998) yields a renormalized mineralogy of oliv-ine¼ 59%, clinopyroxene¼ 29·4%, garnet¼11·6%], asexperiments by Walter (1998) indicated that at high pres-sure (4·5^5·0 GPa) and at F¼10^20% [Feigenson et al.(2003) estimated F¼ 0·08^0·15 for HSDP2 mafic lavas],high-MgO melts may be in equilibrium with such mater-ials.We also use a nominally fertile peridotite source (oliv-ine¼ 60%, clinopyroxene¼15%, orthopyroxene¼ 20%and garnet¼ 5%). For the partition coefficients noted inthe Methods section, sample SR641 requires a source with0·26 wt % TiO2 at F¼ 0·1, and 0·43 wt % TiO2 atF¼ 0·2, using a peridotite residue, and 0·30% and 0·46%TiO2 for F¼ 0·1 and 0·2 respectively using a wehrlite resi-due. These values are 3^5 times higher than the medianperidotite TiO2 content of 0·09 wt %, and are higherthan estimates for depleted mantle (Salters & Stracke,2004, 0·2% TiO2; Workman & Hart, 2005, 0·13% TiO2).The Hawaiian source, however, lies between the 90% and99·5% quantiles for natural peridotites, which contain 0·2and 0·55 wt % TiO2. The Hawaiian source is thus en-riched in TiO2 compared with typical natural peridotites

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

301

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

and estimates of depleted mantle. In contrast, the sourcefor the Siqueiros magmas appears to require less or per-haps no enrichment in TiO2. Parental Siqueiros magmascontain 0·97% TiO2 and the calculated Siqueiros mantlesource contains 0·07^0·08% TiO2 if the melting residue isharzburgite, or 0·14^0·23% TiO2 if the residue is peridotite(Workman & Hart, 2005).As for natural peridotites, the nature of the TiO2

enrichment is not clear. Peridotite samples that containTiO240·2 wt % and that have been analyzed for La andYb have (La/Yb)N (chondrite-normalized La/Yb) ratiosthat exceed 2·0, the value that McDonough & Sun (1995)used to filter peridotites affected by metasomatism.However, spinel peridotites indicated as ‘wet’ in theMcDonough database contain a mean of 0·118 wt %TiO2 and a median of 0·087 wt %, whereas those labeled‘dry’ have a mean of 0·111 wt % and a median of 0·091 wt%; TiO2 is furthermore uncorrelated with either La/Yb(R¼ 0·074) or La/Nb (^0·113) for the garnetþ spinel peri-dotite dataset as a whole. Therefore it is not clear thatmetasomatism is the appropriate enrichment mechanism.Among spinel peridotites, however, TiO2 contents aremost strongly correlated with Al2O3 or, when recast asmineral components, with calculated spinel or garnet con-tents (R¼ 0·57 for calculated garnet contents andR¼ 0·59 for spinel). By these mass-balance calculations, asource with 0·3 wt % TiO2 could be achieved from onethat has c. 10% garnet or spinel (as opposed to the averagecalculated spinel content of 4·0%, or the average of 4·4%garnet for the natural peridotites).ThereforeTiO2 contentsin the natural peridotite database may reflect a degree offertility more than any effect related to metasomatism.To better constrain the fertility of the Hawaiian source,

we calculate the trace element compositions of Hawaiianand Siqueiros parental magmas (Table 1), and from these,their mantle sources (Table 2; see Appendix B for detailsof the calculations). Our Hawaiian mantle source, whencompared with our Siqueiros mantle or published depletedmantle sources is enriched in trace elements from Rb toV,including moderately incompatible elements (Ti, Hf, Eu,and Y) and several compatible elements (Yb and Lu).Trace element and isotopic enrichments at Hawaii havebeen explained previously by the addition of smallamounts of subducted sediment or MORB crust emplacedinto ambient mantle (e.g.Weaver, 1991; Prytulak & Elliott,2007; Jackson et al., 2008). To obtain our calculatedHawaiian mantle source, we were unable to add subductedcontinental crust (as sediment) from Hofmann (1988) todepleted mantle (Salters & Stracke, 2004), as mixing ofcontinental crust yields higher than observed enrichmentsin highly incompatible elements such as Rb, Ba and Th,and would result in lower than observed concentrationsfor Co, Cr, and Ni (unless large ion lithophile elementsare stripped from down-going subducted plates; Weaver,

1991). However, we find that a simple two-component mix-ture involving mean MORBþ depleted mantle (Salters &Stracke, 2004) reproduces our calculated Hawaiian sourcemantle remarkably well (Fig. 7a). For the subductedMORB composition, we use the mean of all compositionsreported in the PetDB database (http://www.petdb.org/petdbWeb/index.jsp) (Electronic Appendix A: TableA4). To obtain a mixing ratio, we searched for theratio that best matches our calculated Hawaiian source(Table 2), assuming a wehrlite residue and F¼ 0·1. Themixing ratio that best describes the Hawaiian Source(HS) is HS¼10·8% MORBþ 89·2% DM (DepletedMantle) (Salters & Stracke, 2004) (Electronic AppendixA: Table A5). The coefficients are not highly precise. If weuse only the elements from Nb to Lu (thus assuming thatwe know their concentrations and partitioning behaviorsbetter than those of other elements), the mixing ratio isHS¼12·9% MORBþ 87·1% DM; if we use TiO2 as theonly constraint, the mixing ratio is HS¼ 5%MORBþ 95% DM (Table A5). These estimates are never-theless consistent with Os and Pb isotopic constraints thatplace the amount of an enriched component in theHawaiian source at about 10% or less (Bennett et al., 1996;Hauri et al., 1996).By adding10·8%MORB to a depleted mantle, we calcu-

late by mass balance that the bulk mineralogy would be60·1% olivine, 15·4% clinopyroxene, 13·9% orthopyroxeneand 10·6% garnet (Table A5). In contrast, Sobolev et al.(2007) and Jackson et al. (2008) proposed that the amountsof MORB crust that might be added to plume sourcesmay be 20%. The problems associated with adding20% MORB to DM, however, are several. Besides thecontradiction with Os^Pb isotope data at Hawaii(Bennett et al., 1996; Hauri et al., 1996), such a mantlesource would be depleted in Ni, and so would predict thatHawaiian basalts would have lower Ni than MORB, acharacteristic that is not observed. Additionally, Norman& Garcia (1999) noted that addition of large amounts ofbasaltic crust (they considered a 25% MORB^75% peri-dotite mixture) would lead to a ‘hyperfertile’ mantle. Ourmass-balance calculations support this view. Adding 20%of MORB (7·6% MgO,15% Al2O3, 11·6% CaO) to a DM(39% MgO, 4% Al2O3, 3·2% CaO; Workman & Hart,2005) would lead to a mantle source with low MgO(32·7%) and high Al2O3 (6·2%) and CaO (4·8%). Usingolivine, orthopyroxene and clinopyroxene mineral com-positions fromWorkman & Hart (2005) and a garnet com-position from PHN-1611 (Takahashi & Ito, 1987),the mineralogy of such a source would contain 38·1% oliv-ine, 27·5% orthopyroxene, 18·5% clinopyroxene and15·9% garnet. Such a source can be rejected, as it wouldhave higher than observed Sm/Yb or La/Lu ratios.Appropriate Sm/Yb ratios can be achieved from thissource only if F 0·35, but increasing F beyond 0·25

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

302

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

No

rma

lize

d H

awa

iian

& S

iqu

ieo

rs M

an

tle

Sou

rce

s

Siqueiros/Med. Perid.

Hawaii/SiqueirosHawaii/(10%MORB + 90%DM)(a)

Rb Ba Th K 2O U Nb La Ce Sr Nd Sm Zr Hf Eu Y Yb Lu V Co Cr NiTiO

2

0.01

0.1

1

10

100

Hawaii/Med. Perid.

Hawaii/DM (S&S)

0.1

1

10

100

Rb Ba Th U Nb La Ce Sr Nd Sm Zr Hf Eu Y Yb Lu V Co Cr NiK2O

TiO2

Ro

ck/P

rim

itiv

e M

an

tle

Mean HSDP with

15%<MgO<21%

(Sm/Yb = 2.39; La/Lu = 39.0;

Cr/Ni = 1.22)

Partial melts from:

hyperfertile source (= MORB (20%) + DM (80%) mix)

(F=0.2 Sm/Yb = 2.94; La/Lu = 68.4; Cr/Ni = 0.75

F=0.35: Sm/Yb = 2.2; La/Lu = 37.0; Cr/Ni = 0.76)

(b)Mean Koolau with

15%<MgO<21%

(Sm/Yb = 2.5; La/Lu = 34.8;

Cr/Ni = 0.96)

F=0.35F=0.20

Siq. normalized to median peridotite

Haw. normalized to median peridotite

Haw. normalized to Siqueiros mantle

Haw. normalized to DM (Salters & Stracke, ‘04)

Haw. normalized to enriched mantle

Fig. 7. (a) Normalized trace element characteristics of mantle sources. Dashed lines: calculated Hawaiian and Siqueiros mantle sources com-pared with median (Med.) peridotite compositions (W. McDonough, personal communication). Continuous gray lines: Hawaiian source com-pared with Siqueiros source (this study) and the depleted mantle (DM) source of Salters & Stracke (2004) (S&S). Continuous black line:Hawaiian source compared with an enriched mantle source composed of 89·2% DMþ10·8% MORB. (b) Mafic (nominally parental) lavacompositions from HSDP2 and Koolau (both with 18%MgO) normalized to primitive mantle (Lyubetskaya & Korenaga, 2007) and comparedwith similarly normalized partial melts of a hyperfertile mantle source (38·1% olivine, 27·5% orthopyroxene, 18·5% clinopyroxene and 15·9%garnet, derived by mass balance from a chemical mixture of 20% MORBþ 80% DM; see text).

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

303

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

yields abundances for incompatible elements that are toolow (Fig. 7b). Adding more MORB crust simply createsmore garnet in the source and adds to the overall hyperfer-tility of the mantle. Adding mafic gabbro (e.g. Perk et al.,2007; Pito Deep at the East Pacific Rise) to the subductionpackage exacerbates the problem, as such rocks have toolittle MgO to affect the mineral mass balance, and theyhave so little TiO2 as to require adding more MORBcrust into the mix to match the observed Hawaiian TiO2

contents.

CONCLUSIONSWe have demonstrated that a mantle source with a perido-tite mineralogy is a flexible source that is perfectly capableof generating Hawaiian and Siqueiros magmas. Our mostimportant findings are as follows.

(1) The high NiO contents in Hawaiian olivines have lessto do with the mantle source composition and moreto do with the conditions under which olivine crystal-lizes. High-NiO olivines from any of the HawaiianIslands can form when picritic magmas (high MgO,high NiO) are transported to shallow depths, andthus crystallize at low T, where DNi is high. In thisway, olivine phenocrysts may have NiO contents thatexceed the NiO content of olivine in the mantlesource region, where the latter exhibit Fo^NiO sys-tematics that are affected by subsolidus equilibration.Comparisons of Hawaiian olivine compositions withthose from natural peridotites are thus irrelevant.We nonetheless find that NiO contents of olivines atHawaii and of natural lherzolites both range continu-ously to 0·6 wt %, but that at a given NiO content,Hawaiian Fo contents are lower, owing to the effectsof magmatic differentiation.

(2) Our calculated NiO and Co contents for theHawaiian and Siqueiros mantle sources are similar toone another, and NiO contents at both localitiesmatch estimates for the mean and median of naturalperidotites (and primitive mantle) very well, but area poor match to natural pyroxenites.

(3) Experimental partial melts of peridotite can simultan-eously explain the FeOt, SiO2, CaO, and Al2O3

contents of Hawaiian basalts. High FeOt contents inHawaiian Islands basalts indicate that partial meltingoccurs at high P (4^5 GPa). High-SiO2 Koolau com-positions are reproduced if FeOt contents are set athigh pressures, and SiO2 is affected byre-equilibration at lower pressures (1^1·5 GPa), assuggested by Wagner & Grove (1998) and Stolperet al. (2004). Lower FeO contents at Siqueiros are ex-plained by peridotite partial melting at just above 1·0GPa, and mostly below 3 GPa. High P at Hawaii fur-thermore leads to elevated FeO/MnO, whereas later

cooling and partial crystallization lead to minor butmeasurable increases in FeO/MnO ratios.

(4) As shown in previous studies, TiO2 (Takahashi et al.,1993; Prytulak & Elliott, 2007) and many trace elem-ents (Hofmann, 1997) are enriched in the Hawaiiansource compared with both depleted MORB-sourcemantle and ‘typical’ natural peridotites. The TiO2

content in the Hawaiian mantle is in the range 0·2^0·4 wt %, above estimates for depleted mantle, andbetween the 90% and 99·5% quantile values of nat-ural peridotites. Both theTiO2 and trace element con-tents at Hawaii can be explained by the addition of5^12% of a crustal component identical to modernMORB, mixed into an ambient mantle that is similarto the Depleted Mantle (DM) of Salters & Stracke(2004) or Workman & Hart (2005). We do not advo-cate this as the sole mantle replenishment mechanism,but it is at least consistent with the Pb-isotope studyof Eisele et al. (2003). Ironically, if TiO2 is indeed asignal of mantle enrichment, Koolau has less of thisenrichment than HSDP2 samples.

We should perhaps emphasize that our work does not pre-clude the addition of pyroxenite or eclogite to ambientmantle. The MORB recycling that we advocate to explaintheTiO2 and trace element contents at Hawaii almost cer-tainly involves conversion of MORB to eclogite duringsubduction. Rather, we question whether large amounts ofeclogite or pyroxenite are preserved so as to provide theprincipal mineralogy during later partial melting. Hightemperatures and potentially long residence times forrecycled components subducted to the base of the lowermantle may well result in their mixing with ambientmantle before their return upwards. Blichert-Toft et al.(1999), for example, suggested that complete mixing couldtake place in50·4 Gyr, although isotopic data assure usthat some such heterogeneities are preserved. Perhapsmore importantly, hypotheses that attribute the variousgeochemical characteristics of oceanic basalts to mineral-ogical heterogeneity in the mantle source, may, by theirvery nature, minimize real and important contrasts in theconditions of magma generation and evolution (as empha-sized here) or contrasts in mantle permeability or magmatransport rates (e.g. Sims et al., 1999). The limits to whichthese process-based models fail to explain observedmagma compositions is where we must look to delimitminima in mantle source differences.

ACKNOWLEDGEMENTSWe thank Bill McDonough for sharing his compilationof spinel and garnet peridotite compositions, as well hisinsights into peridotite geochemistry. Thoughtful andthorough reviews by Cin-Ty Lee, J. Michael Rhodes, andan anonymous reviewer were greatly appreciated. Their

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

304

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

suggestions greatly improved this paper. We also thankDominiqueWeiss for her editorial efforts.

FUNDINGThis work was supported by the National ScienceFoundation grants to K.P. (NSF-EAR 0337345) and M.P.(OCE-90-19154). W.I.R. publishes with the permission ofthe Director of the US Geological Survey.

SUPPLEMENTARY DATASupplementary data for this paper are available at Journalof Petrology online.

REFERENCESAdam, J. & Green, T. (2006). Trace element partitioning between

mica- and amphibole-bearing garnet lherzolite and hydrousbasanitic melt: 1. Experimental results and the investigation ofcontrols on partitioning behavior. Contributions to Mineralogy and

Petrology 152, 1^17.Armstrong, J. T. (1995). CITZAFça package of correction programs

for the quantitative electron microbeam X-ray-analysis of thickpolished materials, thin-films, and particles. Microbeam Analysis 4,177^200.

Arndt, N. T. (1977). Partitioning of nickel between olivine and ultra-basic komatiite liquids. Carnegie Institution of Washington Yearbook 76,553^557.

Baker, M. B. & Stolper, E. M. (1994). Determining the composition ofhigh-pressure mantle melts using diamond aggregates. Geochimicaet Cosmochimica Acta 58, 2811^2827.

Baker, M. B., Alves, S. & Stolper, E. M. (1996). Petrography andpetrology of the Hawaii Scientific Drilling Project lavas: inferencesfrom olivine phenocryst abundances and compositions. Journal ofGeophysical Research 101, 11715^11727.

Beattie, P. (1993). Olivine^melt and orthopyroxene^melt equilibria.Contributions to Mineralogy and Petrology 115, 103^111.

Bennett, V. C., Esat, T. M. & Norman, M. D. (1996). Two mantle-plume components in Hawaiian picrites inferred from correlatedOs^Pb isotopes. Nature 381, 221^223.

Blichert-Toft, J., Frey, F. A. & Albarede, F. (1999). Hf isotope evidencefor pelagic sediments in the source of Hawaiian basalts. Science

285, 879^882.Bock, G. (1991). Long-period S to P converted waves and the onset of

partial melting beneath Oahu, Hawaii. Geophysical Research Letters

18, 869^872.Bowen, N. L. (1928).The Evolution of the Igneous Rocks. NewYork: Dover,

334 p.Canil, D. (1999). Vanadium partitioning between orthopyroxene,

spinel and silicate melt and the redox states of mantle sourceregions for primary magmas. Geochimica et Cosmochimica Acta 63,557^572.

Clague, D. A.,Weber,W. S. & Dixon, J. E. (1991). Picritic glasses fromHawaii. Nature 353, 553^556.

Dasgupta, R., Jackson, M. G. & Lee, C.-T. (2010). Major elementchemistry of ocean island basaltsçconditions of mantle meltingand heterogeneity of mantle source. Earth and Planetary Science

Letters 289, 377^392.

Davidson, P. M. & Lindsley, D. H. (1994). Effect of Ca content andSiO2 activity on augite þ olivine equilibria. American Mineralogist

79, 1123^1124.Dixon, J. E., Clague, D. A.,Wallace, P. & Poreda, R. (1997).Volatiles

in alkalic basalts from the North Arch Volcanic Field, Hawaii:extensive degassing of deep submarine-erupted alkalic series lavas.Journal of Petrology 38, 911^939.

Dixon, J. E., Leist, L., Langmuir, C. & Schilling, J.-G. (2002).Recycled dehydrated lithosphere observed in plume-influencedmid-ocean-ridge basalt. Nature 420, 385^389.

Eggins, S. M. (1992). Petrogenesis of Hawaiian tholeiites: 1, phaseequilibria constraints. Contributions to Mineralogy and Petrology 110,387^397.

Eisele, J., Abouchami, W., Galer, S. J. G. & Hofmann, A. W. (2003).The 320 kyr Pb isotope evolution of Mauna Kea lavas recordedin the HSDP-2 drill core. Geochemistry, Geophysics, Geosystems 4,doi:10.1029/2002GC000339.

Elkins, L. J., Gaetani, G. A. & Sims, J. W.W. (2008). Partitioning ofU andTh during garnet pyroxenite partial melting: constraints onthe source of alkaline ocean island basalts. Earth and Planetary

Science Letters 265, 270^286.Falloon, T. J. & Danyushevsky, L. V. (2000). Melting of refractory

mantle at 1·5, 2 and 2·5 GPa under anhydrous and H2O-undersaturated conditions: implications for the petrogenesis ofhigh-Ca boninites and the influence of subduction components onmantle melting. Journal of Petrology 41, 257^283.

Falloon, T. J., Green, D. H., Hatton, C. J. & Harris, K. L. (1988).Anhydrous partial melting of a fertile and depleted peridotitefrom 2 to 30 Kb and application to basalt petrogenesis. Journal ofPetrology 29, 1257^1282.

Feigenson, M. D., Bolge, L. L., Carr, M. J. & Herzberg, C. T. (2003).REE inverse modeling of HSDP2 basalts: evidence for multiplesources in the Hawaiian plume. Geochemistry, Geophysics, Geosystems4, doi:10.1029/2001GC000271.

Finnerty, A. A. & Rigden, S. M. (1981). Olivine barometry: applica-tion to pressure estimation for terrestrial and lunar rocks.Proceedings of the 12th Lunar and Planetary Science Conference. Geochimica

et Cosmochimica Acta Supplement, pp. 279^281.Frey, F. A., Garcia, M. O., Wise, W. S., Kennedy, A., Gurriet, P. &

Albarede, F. (1991). The evolution of Mauna Kea volcano, Hawaii:petrogenesis of tholeiitic and alkalic basalts. Journal of GeophysicalResearch 96, 14347^14375.

Frey, F. A., Garcia, M. O. & Rhoden, M. F. (1994). Geochemicalcharacteristics of Koolau Volcano: implications of inter-shieldgeochemical differences among Hawaiian volcanoes. Geochimica et

Cosmochimica Acta 58, 1441^1462.Garcia, M. O. (1996). Petrography and olivine and glass chemistry

of lavas form the Hawaii Scientific Drilling Project. Journal of

Geophysical Research 101, 11701^11713.Garcia, M. O. (2002). Submarine picritic basalt from Ko’olau volcano,

Hawai’i: implications for parental magma composition andmantle source. In: Takahashi, E., Lipman, P. W., Garcia, M. O.,Naka, J. & Aramaki, S. (eds) Hawaiian Volcanoes: Deep Underwater

Perspectives. American Geophysical Union, Geophysical Monograph 128,391^402.

Garcia, M. O., Hulsebosch,T. P. & Rhodes, J. M. (1995). Olivine-richsubmarine basalts from the southwest rift zone of Mauna Loavolcano: implications for magmatic processes and geochemicalevolution. In: Rhodes, J. M. & Lockwood, J. P. (eds) Mauna Loa

Revealed. American Geophysical Union, Geophysical Monograph 92,219^240.

Garcia, M. O., Haskins, E. H., Stolper, E. M. & Baker, M. (2007).Stratigraphy of the Hawai’i Scientific Drilling Project core

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

305

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

(HSDP2): anatomy of a Hawaiian shield volcano. Geochemistry,

Geophysics, Geosystems 8, doi:10.1029/2006GC001379.Green, D. H. & Ringwood, A. E. (1967). The genesis of basaltic

magmas. Contributions to Mineralogy and Petrology 15, 103^190.Gregg, P. M., Behn, M. D., Lin, J. & Grove, T. L. (2009). Melt gener-

ation, crystallization, and extraction beneath segmented oceanictransform faults. Journal of Geophysical Research 114, doi:10.1029/2008JB006100.

Haskins, E. R. & Garcia, M. O. (2004). Scientific drilling revealsgeochemical heterogeneity within the Ko’olau shield, Hawai’i.Contributions to Mineralogy and Petrology 147, 162^188.

Hauri, E. H. (1996). Major element variability in the Hawaiian mantleplume. Nature 382, 415^419.

Hauri, E. H.,Wagner, T. P. & Grove, T. L. (1994). Experimental andnatural partitioning of Th, U, Pb and other trace elements betweengarnet, clinopyroxene and basaltic melts. Chemical Geology 117,149^166.

Hauri, E. H., Lassiter, J. C. & DePaolo, D. J. (1996). Osmium isotopesystematics of drilled lavas from Mauna Loa, Hawaii. Journal ofGeophysical Research 101, 11793^11806.

Hawkesworth, C. J., Norry, M. J., Roddick, J. C. & Vollmer, R. (1979).143Nd/144Nd and 87Sr/86Sr ratios from the Azores and theirsignificance in LIL-element enriched mantle. Nature 280, 28^31.

Hays, M. R. (2004). Intra-transform volcanism along the SiqueirosFracture Zone 8820’N to 8830’N, East Pacific Rise, PhD thesis,University of Florida, Gainesville, 251 pp.

Helz, R. T. (1987). Diverse olivine types in lava of the 1959 eruptionof Kilauea volcano and their bearing on eruption dynamics.In: Decker, R.W.,Wright, T. L. & Stauffer, P. H. (eds) Volcanism at

Hawaii. US Geological Survey, Professional Papers 1350, 691^722.Herzberg, C. (2006). Petrology and thermal structure of the Hawaiian

plume from Mauna Kea volcano. Nature 444, 605^609.Herzberg, C. & O’Hara, M. J. (1998). Phase equilibrium constraints

on the origin of basalts, picrites and komatiites. Earth-Science

Reviews 44, 39^79.Herzberg, C. & Zhang, J. (1996). Melting experiments on anhyd-

rous peridotite KLB-1; composition of magmas in the uppermantle and transition zone. Journal of Geophysical Research 101,8271^8295.

Hirschmann, M. M. & Stolper, E. M. (1996). A possible role for garnetpyroxenite in the origin of the ‘garnet signature’ in MORB.Contributions to Mineralogy and Petrology 124, 185^208.

Hirschmann, M. M., Ghiorso, M. S., Grove, T. L., Krawczynski, M.,Medard, E. & Till, C. B. (2008). Library of Experimental PhaseRelations (LEPR): a database and web portal for experimentalmagmatic phase equilibria data. Geochemistry, Geophysics, Geosystems9, doi:10.1029/2007GC001894.

Hofmann, A. W. (1988). Chemical differentiation of the Earth: therelationship between mantle, continental crust, and oceanic crust.Earth and Planetary Science Letters 90, 297^314.

Hofmann, A. W. (1997). Mantle geochemistry: the message fromoceanic volcanism. Nature 385, 219^229.

Hofmann, A.W. & White,W. M. (1982). Mantle plumes from ancientoceanic crust. Earth and Planetary Science Letters 57, 421^436.

Huang, S. & Frey, F. A. (2003). Trace element abundances of MaunaKea basalt from phase 2 of the Hawaii Scientific Drilling Project:petrogenetic implications of correlations with major elementcontent and isotopic ratios. Geochemistry, Geophysics, Geosystems 4,doi:10.1029/2002GC000322.

Huang, S. & Frey, F. (2005). Recycled oceanic crust in the Hawaiianplume: evidence from temporal geochemical variations withinthe Koolau shield. Contributions to Mineralogy and Petrology 149,556^575.

Huang, S., Humayun, M. & Frey, F. A. (2007). Iron/manganeseratio and manganese content in shield lavas fromKo’olau Volcano, Hawai’i. Geochimica et Cosmochimica Acta 71,4557^4569.

Humayun, M., Qin, L. & Norman, M. (2004). Geochemical evi-dence for excess iron in the mantle beneath Hawaii. Science 306,91^94.

Jackson, M. G., Hart, S. R., Saal, A. E., Shimizu, N., Kurz, M. D.,Blusztajn, J. S. & Skovgaard, A. C. (2008). Globally elevated titan-ium, tantalum, and niobium (TITAN) in ocean island basaltswith high 3He/4He. Geochemistry, Geophysics, Geosystems 9, doi:10.1029/2007/GC001876.

Jones, J. H. (1984).Temperature and pressure-independent correlationsof olivine^liquid partition coefficients and their application totrace element partitioning. Contributions to Mineralogy Petrology 88,126^132.

Kelemen, P. B., Hart, S. R. & Bernstein, S. (1998). Silica enrichmentin the continental upper mantle via melt/rock reaction. Earth and

Planetary Science Letters 164, 387^406.Keshav, S., Gudfinnsson, G. H., Sen, G. & Fei, Y. (2004). High-

pressure melting experiments on garnet clinopyroxenite and thealkalic to tholeiitic transition in ocean-island basalts. Earth and

Planetary Science Letters 223, 365^379.Kinzler, R. J. (1997). Melting of mantle peridotite at pressures

approaching the spinel to garnet transition: Application tomid-ocean ridge basalt petrogenesis. Journal of Geophysical Research102, 853^874.

Kinzler, R. J. & Grove, T. L. (1992). Primary magmas of mid-oceanridge basalts 1. Experiments and methods. Journal of Geophysical

Research 97, 6885^6906.Klein, E. M. & Langmuir, C. H. (1987). Global correlations of ocean

ridge basalt chemistry with axial depth and crustal thickness.Journal of Geophysical Research 92, 8089^8115.

Kogiso, T. & Hirschmann, M. M. (2001). Experimental studyof clinopyroxenite partial melting and the origin of ultra-calcic melt inclusions. Contributions to Mineralogy and Petrology 142,347^360.

Kurz, M. D., Curtice, J., Lott, D. E., III & Solow, A. (2004). Rapidhelium isotopic variability in Mauna Kea shield lavas from theHawaiian Scientific Drilling Project. Geochemistry, Geophysics,

Geosystems 5, doi:10.1029/2002GC000439.Kushiro, I. (1996). Partial melting of a fertile mantle peridotite at high

pressures: an experimental study using aggregates of diamond.In: Basu, A. & Hart, S. (eds) Reading the Isotopic Code. American

Geophysical Union, Geophysical Monograph 95, 109^122.Langmuir, C. H., Klein, E. M. & Plank, T. (1992). Petrological

systematics of mid-ocean ridge basalts: constraints on melt gener-ation beneath ocean ridges. In: Morgan, J. P., Blackman, D. K. &Sinton, J. M. (eds) Mantle Flow and Melt Generation at Mid-

Ocean Ridges. American Geophysical Union, Geophysical Monograph 71,183^280.

Longhi, J. (1995). Liquidus equilibria of some primary lunar and ter-restrial melts in the garnet stability field. Geochimica et CosmochimicaActa 59, 2375^2386.

Longhi, J. (2002). Some phase equilibrium systematics oflherzolite melting: I. Geochemistry, Geophysics, Geosystems 3, doi:10.1029/2001GC000204.

Lyubetskaya, T. & Korenaga, J. (2007). Chemical composition ofEarth’s primitive mantle and its variance: 1. Method and results.Journal of Geophysical Research 112, doi:10.1029/2005JB004223.

Maal�e, S. (2004). The PT-phase relations of an MgO-rich Hawaiiantholeiite: the compositions of primary Hawaiian tholeiites.Contributions to Mineralogy and Petrology 148, 236^246.

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

306

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

Macdonald, G. A. (1944). The 1840 eruption and crystal differenti-ation in the Kilauean magma column [Hawaii]. American Journalof Science 242, 177^189.

Matzen, A. K., Bajer, M. B., Beckett, J. R. & Stolper, E. M. (2009).The temperature and pressure dependence of Ni partitioningbetween olivine and MgO-rich silicate melt. Goldschmidt ConferenceAbstracts A851.

Matzen, A. K., Bajer, M. B., Beckett, J. R. & Stolper, E. M. (2011).Fe^Mg partitioning between olivine and high-magnesian meltsand the nature of Hawaiian parental liquids. Journal of Petrology

(in press).McDonough,W. F. & Sun, S-s. (1995). The composition of the Earth.

Chemical Geology 120, 223^253.Morgan, W. J. (1971). Convection plumes in the lower mantle. Nature

230, 42^43.Murata, K. J. & Richter, D. H. (1966). The settling of olivine in

Kilauean magma as shown by lavas of the 1959 eruption. AmericanJournal of Science 264, 194^203.

Norman, M. D. & Garcia, M. O. (1999). Primitive magmas andsource characteristics of the Hawaiian plume: petrology and geo-chemistry of shield picrites. Earth and Planetary Science Letters 168,27^44.

Norman, M. D., Garcia, M. O., Kamenetsky, V. S. & Nielsen, R. L.(2002). Olivine-hosted melt inclusions in Hawaiian picrites: equili-bration, melting and plume source characteristics. Chemical Geology183, 143^168.

Perfit, M. R., Fornari, D. J., Ridley, W. I., Kirk, P. D., Casey, J.,Kastens, K. A., Reynolds, J. R., Edwards, M., Desonie, D.,Shuster, R. & Paradis, S. (1996). Recent volcanism in the Siqueirostransform fault: picritic basalts and implications for MORBmagma genesis. Earth and Planetary Science Letters 141, 91^108.

Perk, N.W., Coogan, L. A., Karson, J. A., Klein, E. M. & Hanna, H.D. (2007). Petrology and geochemistry of primitive lower oceaniccrust from Pito Deep: implications for the accretion of the lowercrust at the southern East Pacific Rise. Contributions to Mineralogy

and Petrology 154, 575^590.Pickering-Witter, J. & Johnston, A. D. (2000). The effects of variable

bulk composition on the melting systematics of fertile peridotiticassemblages. Contributions to Mineralogy and Petrology 140, 190^211.

Pietruszka, A. J., Hauri, E. H., Carlson, R. W. & Garcia, M. O.(2006). Remelting of recently depleted mantle within theHawaiian plume inferred from the 226Ra^230Th^238U disequilibriaof Pu’u ‘O’o eruption lavas. Earth and Planetary Science Letters 244,155^169.

Powers, H. A. (1955). Composition and origin of basaltic magma of theHawaiian Islands. Geochimica et Cosmochimica Acta 7, 77^107.

Prytulak, J. & Elliott, T. (2007). TiO2 enrichment in ocean islandsbasalts. Earth and Planetary Science Letters 263, 388^403.

Putirka, K. D. (1999). Melting depths and mantle heterogeneitybeneath Hawaii and the East Pacific Rise: constraints from Na/Tiand rare earth element ratios. Journal of Geophysical Research 104,2817^2829.

Putirka, K. D. (2005). Mantle potential temperatures at Hawaii,Iceland, and the mid-ocean ridge system, as inferred from olivinephenocrysts: Evidence for thermally driven mantle plumes.Geochemistry, Geophysics Geosystems 6, doi:10.1029/005GC000915.

Putirka, K. D. (2008a). Excess temperatures at ocean islands:implications for mantle layering and convection. Geology 36,283^286.

Putirka, K. D. (2008b). Hot arguments to cool off plume debate?:Comment. Geology doi:10.1130/G25165C.1.

Putirka, K. D. (2008c). Thermometers and barometers for volcanicsystems. In: Putirka, K. D. & Tepley, F. (eds) Minerals, Inclusions

and Volcanic Processes. Mineralogical Society of America and Geochemical

Society, Reviews in Mineralogy and Geochemistry 69, 61^120.Putirka, K. D., Perfit, M., Ryerson, F. J. & Jackson, M. G. (2007).

Ambient and excess mantle temperatures, olivine thermometry,and active vs passive upwelling. Chemical Geology 241, 177^206.

Rhodes, J. M. & Vollinger, M. J. (2004). Composition of basaltic lavassampled by phase-2 of the Hawaii Scientific Drilling Project:geochemical stratigraphy and magma series types. Geochemistry,

Geophysics, Geosystems 5, doi:10.1029/2002GC00434.Ringwood, A. E. (1962a). A model for the upper mantle. Journal of

Geophysical Research 67, 857^867.Ringwood, A. E. (1962b). A model for the upper mantle 2. Journal of

Geophysical Research 67, 4473^4477.Robinson, J. A. C.,Wood, B. J. & Blundy, J. D. (1998). The beginning

of melting of fertile and depleted peridotite at 1·5 GPa. Earth and

Planetary Science Letters 155, 97^111.Roeder, P. L. & Emslie, R. F. (1970). Olivine^liquid equilibrium.

Contributions to Mineralogy and Petrology 29, 275^289.Salters, V. & Longhi, J. (1999). Trace element partitioning during the

initial stages of melting beneath mid-ocean ridges. Earth and

Planetary Science Letters 166, 15^30.Salters, V. J. M. & Stracke, A. (2004). Composition of the

depleted mantle. Geochemistry, Geophysics, Geosystems 5, doi:10.1029/2003GC000597.

Sen, G., Keshav, S. & Bizimis, M. (2005). Hawaiian mantle xenolithsand magmas: composition and thermal character of the litho-sphere. American Mineralogist 90, 871^887.

Sims, K. W.W., DePaolo, D. J., Murrell, M. T., Baldridge, W. S.,Goldstein, S. & Clague, D. (1999). Porosity of the melting zoneand variations in the solid mantle upwelling rate beneath Hawaii:inference from 238U^230Th^226Ra and 235U^231Pa disequilibria.Geochimica et Cosmochimica Acta 63, 4119^4138.

Sobolev, A. V., Hofmann, A. W., Sobolev, S. V. & Nikogosian, I. K.(2005). An olivine-free mantle source of Hawaiian shield basalts.Nature 434, 590^597.

Sobolev, A. V., Hofmann, A. W., Kuzmin, D. V., Yaxley, G. M.,Arndt, N. T., Chung, S.-L., Danyushevsky, L. V., Elliott, T.,Frey, F. A., Garcia, M. O., Gurenko, A. A., Kamenetsky, V. S.,Kerr, A. C., Krivolutskaya, N. A., Matvienkov, V. V.,Nikogosian, I. K., Rocholl, A., Sigurdsson, I. A.,Sushchevskaya, N. M. & Teklay, M. (2007). The amount of recycledcrust in source of mantle-derived melts. Nature 316, 412^417.

Stolper, E., Sherman, S., Garcia, M., Baker, M. & Seaman, C. (2004).Glass in the submarine section of the HSDP2 drill core,Hilo Hawaii. Geochemistry, Geophysics, Geosystems 5, doi:10.1029/2003GC000553.

Stracke, A., Salters, V. J. M. & Sims, K. W.W. (1999). Assessingthe presence of garnet-pyroxenite in the mantle source of bas-alts through combined hafnium^neodymium^thorium isotopesystematics. Geochemistry, Geophysics, Geosystems 1, article number1999GC000013.

Stracke, A., Hofmann, A. W. & Hart, S. R. (2005). FOZO, HIMU,and the rest of the mantle zoo. Geochemistry, Geophysics, Geosystems 6,doi:10.1029/2004GC000824.

Streckeisen, A. (1976). To each plutonic rock its proper name.Earth-Science Reviews 12, 1^33.

Takahashi, E. & Ito, E. (1987). Mineralogy of mantle peridotitealong a model geotherm up to 700 km depth. In: Manghnani, M.H. & Syono, Y. (eds) High-Pressure Research in Mineral

Physics. American Geophysical Union, Geophysical Monograph 39,427^437.

Takahashi, E. & Nakajima, K. (2002). Melitng process in theHawaiian plume: an experimental study. In: Takahashi, E.,

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

307

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

Lipman, P. W., Garcia, M. O., Naka, J. & Aramaki, S. (eds)Hawaiian Volcanoes: Deep Underwater Perspectives. American Geophysical

Union, Geophysical Monograph 128, 391^402.Takahashi, E., Shimazaki, T., Tsuzaki, Y. & Yoshida, H. (1993).

Melting study of a peridotite KLB-1 to 6·5 GPa, and the origin ofbasaltic magmas. Philosophical Transactions of the Royal Society of

London 342, 105^120.Thompson, J. B. (1982). Composition space: an algebraic and geomet-

ric approach. In: Ferry, J. M. (ed.) Characterization of Metamorphism

Through Mineral Equilibria. Mineralogical Society of America, Reviews

in Mineralogy 10, 1^31.Toplis, M. J. (2005). The thermodynamics of iron and magnesium

partitioning between olivine and liquid: criteria for assessing andpredicting equilibrium in natural and experimental systems.Contributions to Mineralogy and Petrology 149, 22^39.

Ulmer, P. (1989). Partitioning of high field strength elementsamong olivine, pyroxenes, garnet and calc-alkaline picrobasalt:Experimental results and applications. Carnegie Institution of

Washington Yearbook, Annual Report of the Director of the Geophysical

Laboratory 89, 42^47.Wagner, T. P. & Grove, T. L. (1998). Melt/harzburgite reaction in the

petrogenesis of tholeiitic magma from Kilauea volcano, Hawaii.Contributions to Mineralogy and Petrology 131, 1^12.

Walter, M. J. (1998). Melting of garnet peridotite and the origin ofkomatiite and depleted lithosphere. Journal of Petrology 39, 29^60.

Wang, Z. & Gaetani, G. A. (2008). Partitioning of Ni between olivineand siliceous eclogite partial melt: experimental constraints onthe mantle source of Hawaiian basalts. Contributions to Mineralogy

and Petrology 156, 661^678.Wang, Z., Kitchen, N. E. & Eiler, J. M. (2003). Oxygen isotope geo-

chemistry of the second HSDP core. Geochemistry, Geophysics,

Geosystems 4, doi:10.1029/2002GC000406.Washington, H. S. (1925). The chemical composition of the Earth.

AmericanJournal of Science 9, 351^378.Waters, C. L., Sims, K. W. W., Perfit, M. R., Blichert-Toft, J. &

Blusztajn, J. (2010). Perspective on the genesis of E-MORB fromchemical and isotopic heterogeneity at 98-108N East Pacific Rise.Journal of Petrology (in press).

Weaver, B. L. (1991). The origin of ocean island basalt end-membercompositions: trace element and isotopic constraints. Earth and

Planetary Science Letters 104, 381^397.Wendlandt, R. F. & Ridley,W. I. (1994). Melting phase relations of a

glassy high-MgO basalt from the Siqueiros Transform Domain,East Pacific Rise (EPR). EOS Transactions, American Geophysical

Union, Fall Meeting 720.White,W. M. & Hofmann, A.W. (1982). Sr and Nd isotope geochem-

istry of oceanic basalts and mantle evolution. Nature 29, 821^825.Willbold, M. & Stracke, A. (2006). Trace element composition of

mantle end-members: implications for recycling of oceanic andupper and lower continental crust. Geochemistry, Geophysics,

Geosystems 7, doi:10.1029/2005GC001005.Witt-Eickschen, G., O’Neill, H. & St., C. (2005).The effect of tempera-

ture on the equilbrium distribution of trace elements betweenclinopyroxene, orthopyroxene, olivine and spinel in upper mantleperidotite. Chemical Geology 221, 65^101.

Workman, R. K. & Hart, S. R. (2005). Major and trace element com-position of the depleted MORB mantle (DMM). Earth and

Planetary Science Letters 231, 53^72.Wright, T. L. & Fiske, R. S. (1971). Origin of the differentiated and

hybrid lavas of Kilauea volcano, Hawaii. Journal of Petrology 12,1^65.Yoder, H. S. & Tilley, C. E. (1962). Origin of basalt magmas: an

experimental study of natural and synthetic rock systems. Journalof Petrology 3, 342^532.

Zindler, A. & Hart, S. (1986). Chemical geodynamics. Annual Review of

Earth and Planetary Sciences 14, 493^571.

APPENDIX A: ELECTRONIC DATATABLESTables A1a, b: Standard precision (see Methods section)mineral compositions for HSDP2 and Siqueiros olivines.Tables A2a, b: High-precision mineral compositions forHSDP2 and Siqueiros olivines.Tables A3a, b: Peridotite compositions from GEOROCand fromW. McDonough.Table A4: Mean MORB composition.Table A5: Model mantle compositions derived by mixingMORB and Depleted Mantle.

APPENDIX B: EXPLANATIONOF CALCULAT IONS ANDADDIT IONAL TABLESWITH SAMPLECALCULAT IONSCalculating trace element contents ofmantle source regionsThe trace element contents of hypothetical mantle sourcesare calculated from the trace element contents of parentalliquids. Trace element compositions are determined as fol-lows. Mantle source compositions are calculated using thebatch partial melting equation (Ci

MS¼Ci

P[F ^ Di(1 ^F)]),where Ci

MS is the concentration of i in the mantle source,Ci

P is the concentration of i in the parental magma, F

is melt fraction and Di is the bulk distribution coefficientfor component i). Mineral^melt distribution coefficientsfor most elements are from Salters & Longhi (1999) andSalters & Stracke (2004). Partition coefficients for V andCr are from Ulmer (1989), Hauri et al. (1994), Canil (1999),and Adam & Green (2006). Partitioning relationships ofTiO2, Ni, Co and Mn are explained in the main text.Although more complex partial melting models might beemployed, Langmuir et al. (1992) have shown that batchpartial melting yields similar results to those obtained bythe summing of fractional melts when calculating pooledprimary mantle melt composition (i.e. parental magmas).Our tests of more complex models showed such to be un-warranted, given the uncertainty involved in determiningpartition coefficients, parental magma compositions, andthe range of values of F at which parental magmas maybe generated.Values for Ci

P are derived assuming that parentalmagmas have MgO contents as determined by Putirkaet al. (2007) and Putirka (2008a, 2008b). The trace elementcontents of HSDP2 parental magmas (Table 1) are cal-culated from the mean of whole-rock samples with

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

308

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

15%4MgO421% (Huang & Frey, 2003; Rhodes &Vollinger, 2004), which are thought to approximate primi-tive (nominally parental) liquids (see Putirka et al., 2007).As we show in the Results section, tests for olivine^liquidequilibration indicate that sample SR641 (or R0641), with18·3% MgO, is the most mafic sample where thewhole-rock composition can be treated as a liquid(Putirka et al., 2007). Trace element concentrations forHSDP2 samples display little variation for rocks with simi-lar MgO contents, therefore we average the trace elementcontents of all HSDP2 samples with 18�3% MgO.Similar results are obtained if only trace element concen-trations for SR641 are used. For the Siqueiros MORBsource, we use trace element data from Hays (2004), aver-aging all samples with49·5% MgO (where MgO rangesto 10·12%). Except for Ni and Na, which are sensitive totemperature (T) and pressure (P) respectively, the samemineral^melt partition coefficients are used to calculateHSDP and Siqueiros mantle sources (Salters & Stracke,2004); for Ni and Na, we use P^Tconditions calculated byPutirka et al. (2007) and partitioning relationships asshown in the Methods section.To explore trace element enrichments, we compare nat-

ural peridotites and calculated mantle sources at Hawaiiand Siqueiros using a number of trace elements (Table 2;Fig. 7a and b). These calculated source mineralogiesuse the depleted mantle mineralogy of Workman &Hart (2005) and a nominal harzburgite mineralogy(70% olivine, 30% orthopyroxene) at Siquieros.At Hawaii we consider a garnet wehrlite residue [Walter’s(1998) experiment 45.03 yields a renormalized min-eralogy of olivine¼ 59%, clinopyroxene¼ 29·4%, gar-net¼11·6%], as experiments by Walter (1998) indicatethat at high pressure (4·5^5·0 GPa) and at F¼10^20%,high-MgO melts may be in equilbrium with such.We alsouse a nominally fertile peridotite source (olivine¼ 60%,clinopyroxene¼15%, orthopyroxene¼ 20% and gar-net¼ 5%). We assume that F¼ 0·1^0·2 at Hawaii and0·08^0·15 at Siqueiros. The results (at least on a log-scale)are robust against these ranges in F, and the noted min-eralogical uncertainty.

Recasting whole-rock compositions intomineral componentsTo convert whole-rock compositions into mineral com-ponents, we first convert a whole-rock compositionexpressed on a weight per cent to a mole fraction basis.We use only the following components, SiO2, Al2O3,FmO (¼ FeOþMgO) and CaO, which are written asa row matrix. These components are converted into themineral components olivine (forsteriteþ fayalite, or FoFa),clinopyroxene (diopsideþhedenbergite, or DiHd), ortho-pyroxene (enstatiteþ ferrosilite, or EnFs) and garnet(pyropeþ almandine, or PyAl), using the methods ofThompson (1982). The conversion is performed by

multiplying this row matrix of whole-rock mole fractionswith the following matrix:

FoFa DiHd EnFs PyAl

Fm2SiO4 CaFmSi2O6 Fm2Si2O6 Fm3Al2Si3O12

SiO2 �1 0 1 0

Al2O3 0 0 �1·5 1

FmO 1 0 �0·5 0

CaO 1 1 �1·5 0

and then renormalizing to unity the row matrix that is theproduct. This matrix assumes that garnet is the majorAl-bearing phase. For spinel peridotites, the second row(labeled Al2O3) becomes [0 0 0 1] and the heading to thelast column becomes ‘Spinel MgAl2O4’. We recast spinelperidotites as garnet peridotites, assuming that the spinel^garnet transition is not accompanied by a change in bulkcomposition. We accept only those samples that plotwithin the peridotite field; that is, the upper 3/5 ofStreckeisen’s (1976) triangular diagram for ultramaficrocks (Fig. 1b). Some samples within this peridotite data-base may be metasomatized, and McDonough & Sun(1995) used only samples with chondrite-normalized La/Yb [(La/Yb)N] 2 to avoid these.We apply no such filtersbecause, for trace element contents, we are curious notonly whether typical peridotites can explain Hawaiian orSiqueiros compositions but also, if not, whether any peri-dotite can match our calculated source compositions.

Olivine crystal lines of descent (Fig. 5a)Following the idea of a liquid line of descent, we calculatea ‘crystal line of descent’ (CLD) (Fig. 5a) for olivines,using a Kilauea parental magma composition. The paren-tal magma composition is shown in Table B1 and is fromPutirka et al. (2007).For the liquid of Table B1 (and each succeeding liquid),

we calculate an equilibrium olivine composition, assumingthat KD(Fe^Mg)ol^liq¼ 0·33; we then calculate T usingBeattie (1993) [and, as a check, equation (4) from Putirkaet al. (2007)], using an arbitrary value for pressureas input. The calculated T, and (if needed) liquid com-position, can then be used to calculate DNi, using various

Table B1: Kilauea parental magma composition (wt %)

SiO2 TiO2 Al2O3 FeOt MnO MgO CaO Na2O NiO

Kilauea 47·45 1·91 9·72 10·44 0·17 19·19 8·02 1·65 0·15

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

309

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

Table B2: Sample calculations for olivine crystal line of descent (CLD)

Step Input parameters

Olivine added KD(Fe–Mg)ol–liq P (kbar)

0 0·000 0·33 0·01

1 �0·015 0·33 0·01

2 �0·015 0·33 0·01

3 �0·015 0·33 0·01

4 �0·015 0·33 0·01

Step Primitive (step 0) and fractionated (steps 1–4) liquid compositions (wt %)

SiO2 TiO2 Al2O3 FeO MnO MgO CaO Na2O NiO

0 47·45 1·91 9·72 10·44 0·17 19·20 8·02 1·65 0·140

1 48·45 1·97 10·04 10·66 0·17 19·15 8·28 1·70 0·132

2 48·95 2·01 10·26 10·77 0·18 18·88 8·46 1·74 0·125

3 49·46 2·06 10·48 10·87 0·18 18·60 8·65 1·78 0·118

4 49·98 2·10 10·71 10·98 0·18 18·32 8·84 1·82 0·111

Step Calculated olivine compositions

Fo NiO (wt %)

0 90·9 0·64

1 90·7 0·62

2 90·5 0·60

3 90·2 0·58

4 90·0 0·56

Step Ancillary calculations

DNiO* DMgOy DMgOz T (8C)§ T (8C)� (XFe/XMg)liq** (XFe/XMg)

ol**

0 4·57 2·33 2·34 1423 1414 0·305 0·101

1 4·68 2·37 2·38 1417 1407 0·312 0·103

2 4·79 2·42 2·42 1410 1400 0·320 0·106

3 4·91 2·46 2·47 1402 1393 0·328 0·108

4 5·04 2·51 2·52 1395 1386 0·336 0·111

*DNiO is calculated using equation (2a) (see main text) and T from Beattie (1993).yDMgO is the ‘observed’ value, derived from the calculated equilibrium liquidþ olivine pair.zDMgO is calculated using Beattie’s (1993) equation (10).§T is calculated using Beattie (1993).�T is calculated using Putirka et al.’s (2007) equation (4) and the ‘observed’ DMgO.**Cation fraction ratios of Fe to Mg are given for coexisting liquid (liq) and olivine (ol) compositions.

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

310

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

Table B3: Calculated Co and Ni contents for Hawaiian and MORB source mantle

(1) MORB; 13508C, 10 kbar

Input data*

SiO2 in liquid (wt %) 48

MgO in liquid (wt %) 13

F (melt fraction) 0·08

DMg 3·52

CNiliq (ppm) 350

CColiq (ppm) 69·91

Minerals in mantle source Min. prop. DNi Eqn. for DNiy DCoz Bulk DNi§ Bulk DCo§ Ni in source (ppm)� Co in source (ppm)�

olivine 0·57 5·93 2a 2·00 3·99 1·72 1312 116·1

6·64 2b 4·39 1·72 1442 116·1

7·52 2c 4·89 1·72 1603 116·1

8·06 W&G 3·03 1·72 1004 116·1

orthopyroxene 0·28 1·70 1·50

clinopyroxene 0·13 1·00 1·21

spinel 0·02 0·00 0·00

(2) MORB; 12008C, 10 kbar

SiO2 in liquid (wt %) 48·00

MgO in liquid (wt %) 11·5

F (melt fraction) 0·08

DMg 3·94

CNiliq (ppm) 350

CColiq (ppm) 60·25

Minerals in mantle source Min. prop. DNi Eqn. for DNiy DCoz Bulk DNi§ Bulk DCo§ Ni in source (ppm)� Co in source (ppm)�

olivine 0·57 11·13 2a 2·00 6·95 1·72 2266 100·0

12·49 2b 7·72 1·72 2514 100·0

10·77 2c 6·74 1·72 2198 100·0

9·18 W&G 3·36 1·72 1110 100·0

orthopyroxene 0·28 1·70 1·50

clinopyroxene 0·13 1·00 1·21

spinel 0·02 0·00 0·00

(3) Hawaii; 16658C, 20 kbar

SiO2 in liquid (wt %) 47·20

MgO in liquid (wt %) 21·00

F (melt fraction) 0·20

DMg 2·20

CNiliq (ppm) 831·00

CColiq (ppm) 78·59

(continued)

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

311

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

forms of equation (2). To create the LLD curves, we sub-tract an arbitrarily small amount of the calculated olivinecomposition from the liquid and repeat the calculations.Sample calculations and inputs are inTable B2.In Fig. 5a, a CLD calculated at 10 kbar provides a curve

that mimics the Kilauea trend. If equation (2b) or (2c) isused, the values for DNi are higher, leading to steepercurves, which tend to underestimate NiO at a given Fo

content. Higher pressures are then needed to reproducethe Kilauea trend (because increased pressure leads tohigher temperature estimates, which in turn decreaseDNi, leading to a flattened CLD trend in Fig. 5a). ForFig. 5a, pressures were chosen arbitrarily so as to mimicthe Kilauea trend. Low P^T fractionation (1 atm, 1400^14108C), of mafic Hawaiian lavas allows for the precipita-tion of high-NiO olivines.

Table B3: Continued

Minerals in mantle source Min. prop. DNi Eqn. for DNiy DCoz Bulk DNi§ Bulk DCo§ Ni in source (ppm)� Co in source (ppm)�

olivine 0·60 2·29 2a 2·00 1·89 1·71 1422 123·5

2·48 2b 2·00 1·71 1496 123·5

3·36 2c 2·53 1·71 1848 123·5

4·69 W&G 3·33 1·71 2380 123·5

orthopyroxene 0·20 1·70 1·50

clinopyroxene 0·15 1·00 1·21

garnet 0·05 0·50 0·66

(4) Hawaii; 15008C, 20 kbar

SiO2 in liquid (wt %) 48·5

MgO in liquid (wt %) 20·15

F (melt fraction) 0·20

DMg 2·31

CNiliq (ppm) 831

CColiq (ppm) 76·48

Minerals in mantle source Min. prop. DNi Eqn. for DNiy DCoz Bulk DNi§ Bulk DCo§ Ni in source (ppm)� Co in source (ppm)�

olivine 0·60 3·58 2a 2·00 2·66 1·71 1934 120·2

4·07 2b 2·96 1·71 2134 120·2

4·53 2c 3·23 1·71 2314 120·2

4·97 W&G 3·50 1·71 2493 120·2

orthopyroxene 0·20 1·70 1·50

clinopyroxene 0·15 1·00 1·21

garnet 0·05 0·50 0·66

*Input data for each of the four sets of calculations are derived as in Tables B1 and B2, to calculate equilibrium olivineand liquid pairs, for a given parental composition (Table B1, and Putirka et al., 2007). Values for DMg [on which the Wang& Gaetani (2008) DNi calculation depends], are from calculated ol–liq pairs, but calculated values from Beattie (1993) arenearly identical. Other inputs (i.e. mineral and proportions) are nominal values presumed for peridotite (see text).yThe equations noted for the calculation of DNi are from the text, except for those labelled W&G, which are calculatedusing Wang & Gaetaini’s (2008) equation (10): ln[DNi]¼ 1·15ln[DMg]þ 0·64.zDCo values are representative values from GERM (http://earthref.org/GERM/).§For each example, four different bulk distribution coefficients are calculated corresponding to each of the four equationsused to calculate DNi

olivine–liquid; each uses the same values for DNi for the remaining minerals, using the indicated mineralproportions; DCo calculation methods do not vary, but the values are repeated for reporting consistency.�Ni and Co concentrations for mantle sources are calculated using the batch melting equation, Co¼Cl FþClD(1 – F),where Co is the concentration of an element in the mantle source, F is the melt fraction, Cl is the concentration of anelement in an equilibrium liquid, and D is the bulk distribution coefficient. Each value for Ni and Co in the source is forcorresponding values of DNi, as calculated in column 3. Because DCo was not calculated so as to vary with T or DMg, itsvariation is tied only to mineralogy.

JOURNAL OF PETROLOGY VOLUME 52 NUMBER 2 FEBRUARY 2011

312

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from

Other forms of equation (2) tend to yield higher valuesfor DNi. Using equation (2b), for example, the mainKilauea trend is best reproduced at 15^20 kbar. If the low-T calibrated equation (2c) is used, the Kilauea trend canbe reproduced only at pressures that are almost certainlytoo high to represent sub-Hawaiian differentiation depths(65 kbar), although it is perhaps worth noting that theseCLDs are probably not accurate barometers.

Co vs Ni calculations (Fig. 5b)To compare the Co and Ni contents of MORB andHawaiian source regions, we first consider the potentialCo and Ni contents of respective parental liquids. Ourhigh-precision olivine compositions and partition coeffi-cients for Ni can be used to estimate the Ni contents ofprimitive liquids at Hawaii and Siqueiros. At Siqueiros,high-Fo olivines (maximum¼ 91·1; mean¼ 90·3) have anaverage NiO content of 0·326 wt % and a maximum of0·382 wt %. If equilibration temperatures are close to13008C, then DNi

ol^liq¼ 8·0 [using an average of equations

(2a), (2b) and (2c)], which yields a magma NiO contentof 0·048 (maximum) and 0·041 (average) wt % NiO, or375 and 320 ppm respectively. These values are higherthan the Ni contents of whole-rocks and glasses fromSiquieros, indicating that most such compositions are frac-tionated. At Hawaii, HSDP2 olivines with Fo contents90 have a maximum NiO content of 0·528 wt %, andan average content of 0·461 wt %. If these olivines equili-brated at an averageT of 15008C, then DNi

ol^liq¼ 4, and

primitive liquids have a maximum of 0·132 wt % NiOand an average of 0·115 wt % NiO, or 1038 and 905 ppmNi, respectively. These Ni contents bracket the Ni contentof SR641 (965 ppm Ni) indicating that this whole-rocksample may approximate a liquid, as also indicated by oliv-ine^liquid Fe^Mg exchange considerations (Putirka et al.,2007).These values are higher than the Ni contents of aver-age HSDP2 whole-rocks with 15%5MgO521% (831ppm), indicating, as at Siqueiros, that most whole-rocksare fractionated with respect to Ni. We do not havehigh-precision data for Co in olivine, and so we use the fol-lowing equations to predict Co for a parental magma of agiven MgO content (based on whole-rock systematics): inHawaiian magmas, Co (ppm)¼ 2·4886(MgO)þ 26·33; inSiqueiros magmas, Co (ppm)¼ 6·4438(MgO) ^ 13·8553.These equations can be used to estimate the amounts ofNi and Co in high-MgO rock compositions. In Table A3,we show details of calculations presented in the text andin Fig. 5b.These calculations show that Ni and Co concen-trations for putative Hawaiian and MORB mantles aremuch more similar to peridotite than to clinopyroxenite,and also that both sources are similar, provided we allowthat Hawaiian magmas are generated and/or differentiateat higher temperatures than MORB. The required tem-peratures differences are about 3008C, which are verysimilar to the differences in mantle potential temperaturesbetween these sources (Putirka et al., 2007).

PUTIRKA et al. OCEANIC MANTLEMINERALOGY

313

at University of F

lorida on March 17, 2011

petrology.oxfordjournals.orgD

ownloaded from