Capture of Helium and Other Volatiles during the Growth of Olivine Phenocrysts in Picritic Basalts from the Juan Fernandez Islands

Transcription

1 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 PAGES 421± Capture of Helium and Other Volatiles during the Growth of Olivine Phenocrysts in Picritic Basalts from the Juan Fernandez Islands JAMES H. NATLAND* ROSENSTIEL SCHOOL OF MARINE AND ATMOSPHERIC SCIENCE, UNIVERSITY OF MIAMI, MIAMI, FL 33149, USA RECEIVED DECEMBER 10, 1998; ACCEPTED AUGUST 30, 2002 Olivine crystals in basalt contain helium that can be extracted for isotopic analysis. Helium-bearing olivine phenocrysts in picritic tholeiites from the Juan Fernandez Islands, SE Pacific, crystallized from moderately differentiated liquids. None are xenocrysts of mantle peridotite. The helium occurs in cavities or bubbles in inclusions best seen at high magnification in the olivine. The olivine grew around spinel, sulfide, and bubbles that preferentially nucleated on crystal surfaces. Inclusions within these phases range from large cavities to tiny, faceted pits arranged in rows. Many crystals contain curving trains of inclusions along annealed features. The inclusions formed during mixing between cooler differentiated magma and hotter olivinecharged magma, which accelerated vesiculation. Bubbles nucleated on the olivine as they do on ice when stirred in carbonated water. Mixing also induced thermal stress fracturing, like the cracking of ice dropped into water. Cracking, irregular extinction, and subgrain formation occurred when faceted crystals collided with each other or with conduit walls. Boundary layer melts and bubbles were drawn quickly into the fractures. Thus few inclusions contain equilibrium proportions of minerals and vapor. Mantle-derived helium clearly permeated into shallow storage reservoirs, including rift zones where magmatic differentiation, mixing, and vapor exsolution were fairly extensive. KEY WORDS:Juan Fernandez; volatiles; inclusions; olivine; picrite INTRODUCTION Farley et al. (1993) presented He, Sr, and Nd isotopic data for picritic tholeiitic and alkalic basalts from the Juan Fernandez Islands, which are located on the Nazca plate west of Chile in the SE Pacific. Helium isotopes were measured on gases extracted from olivine separates crushed in vacuo, and then passed into a mass spectrometer. This is the usual way of measuring isotopes of noble gases in subaerial ocean island basalts, as they almost always lack the more favorable glass component characteristic of abyssal tholeiites. There are two major islands in the Juan Fernandez group (Fig. 1), Robinson Crusoe (also called Isla Mas a TierraÐliterally, island nearer land) and Alexander Selkirk (Isla Mas AfueraÐisland further away). Because the English names, derived from Defoe's (1719) fictional castaway and his historical inspiration who spent some years alone in these islands, are now official, I use them here. The islands are youthful, with radiometric ages in the range of 40±31 Ma for Robinson Crusoe and 13±085 Ma for Alexander Selkirk [as summarized from various sources by Baker et al. (1987)]. They are probably still active (Darwin, 1840). Their age progression is to the west and they are built on Eocene crust (32±37 Ma) of the Nazca plate (Corvalan, 1981). The islands are remote and rarely visited by geologists. Nevertheless, the presence there of picritic basalts uncommonly rich in olivine phenocrysts has long been known. Bowen (1928, p. 164), after Quensel (1912), published a photomicrograph of basalt from these islands, unusually rich in large olivine crystals, which he described as `having the highest proportion of normative olivine (53%) of any rock termed basalt by the author describing it'. Johanssen (1937) gave one such *Corresponding author: jnatland@msn.com Journal of Petrology 44(3) # Oxford University Press 2003; all rights reserved.

2 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 Fig. 1. Generalized bathymetric chart of the Juan Fernandez Ridge, SE Pacific, near Chile, showing the location of principal islands and seamounts, after Mammerickx et al. (1974), from Farley et al. (1993). uncommonly olivine-rich basalt from the islands a type-locality name, masafuerite. The most accessible sources of recent geochemical data for Juan Fernandez basalts are the studies by Gerlach et al. (1986), Baker et al. (1987) and Farley et al. (1993). The last includes analyses from two seamounts, one of them named Friday after the only other principal character in Defoe's novel. I address the question of whether minerals containing a vapor phase with mantle-derived noble-gas isotopic characteristics actually are mantle-derived mineral phases (xenocrysts) in their host basalts, or whether they are phenocrysts that happened to incorporate a vapor phase carrying a mantle isotopic signature. The Juan Fernandez samples studied by Farley et al. (1993) have 3 He/ 4 He ranging from 78 to 180 R A, a range spanning values they attributed to mid-ocean ridge basalt (MORB) mantle sources at the lower, and plume-related (ocean island) sources at the higher end. Because subaerial ocean island basalts are substantially degassed, this question relates to the efficiency of degassing and both where in the volcanic edifice and when in the course of magmatic differentiation degassing and entrapment of He took place. I have investigated several of the samples analyzed for He, Sr, and Nd isotopes by Farley et al. (1993). I examined olivine and associated Cr-spinel in polished thin sections using transmitted and both regular and interference-contrast reflected light, acquired highmagnification scanning electron micrographs of the surfaces of magnetically separated olivine, and determined the compositions of the two mineral phases by electron microprobe. I show that in typical Juan Fernandez picrites of tholeiitic composition, mantle He is contained in olivine phenocrysts, not xenocrysts, and that most of it was evidently incorporated during episodes of combined magma mixing and vesiculation, the two being closely related. Late-stage post-shield eruptives on Robinson Crusoe are basanites and these do contain xenocrysts of four mantle mineral phases (spinel, olivine, clinopyroxene, and orthopyroxene) that probably contributed to their noble-gas isotopic signature. These are not considered in this paper. He and other noble gases are only several among various volatile species that were present in the cavities in the olivine, and probably not the most abundant. Roedder (1965, 1983, 1984) pioneered the study of fluid inclusions in olivine phenocrysts and mantle xenoliths, and showed that CO 2 is the principal volatile constituent in them, being present both as liquid and gas under high confining pressures. His petrographic techniques were developed mainly to observe and conduct experiments on the fluid inclusions. The petrographic techniques used here tell nothing directly about volatile inclusions, certainly not precisely where they were in the minerals before the olivines were crushed and their volatiles extracted. However, as Roedder also observed, many inclusions may also contain glass, minerals, and globular sulfides. These can tell a great deal about the characteristics of associated bubbles, even when the bubbles themselves cannot be directly analyzed. PETROLOGICAL BACKGROUND Figure 2 summarizes the normative characteristics of basalts from the Juan Fernandez Islands reported by Baker et al. (1987) and Farley et al. (1993), and compares them with picritic Hawaiian tholeiites from Kilauea's submarine extension, Puna Ridge (Clague et al., 1991, 1995; Dixon et al., 1991). Tholeiitic basalts from the two places are very similar. Investigations of 422

3 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS Fig. 2. Comparison of principal normative constituents of basalts from the Juan Fernandez Islands and Kilauea±Puna Ridge. Juan Fernandez data are from Baker et al. (1987) and Farley et al. (1993). Juan Fernandez symbols are large, as follows: Trend 1, filled gray and black circlesðhypersthene-normative tholeiites from Robinson Crusoe and Alexander Selkirk, with gray representing samples of this study; Trend 2, open squaresðnepheline-normative alkalic olivine basalts from Robinson Crusoe and Alexander Selkirk; half-filled diamonds enclosed in fieldðbasanites from Robinson Crusoe. Kilauea±Puna Ridge symbols (dashed trend KP) are small filled right-pointing triangles (Clague et al., 1995). Trends 1 and 2 converge along the Ol±Di tie-line, toward the Ol apex. volatiles in basalts from Kilauea bear directly on this study at many points, and together with results here point to general processes that act in the formation of picrites. Farley et al. (1993) noted that their Juan Fernandez sample suite is biased toward olivine-rich samples deliberately collected for noble-gas measurements. Partly as a result of this sampling, two trends are evident in Fig. 2, one extending from the olivine corner first to more strongly hypersthene-normative or tholeiitic compositions, then ultimately to quartz tholeiite, the other to successively more strongly nephelinenormative undersaturated alkalic compositions. The trends diverge from a common point near the olivine corner, among picritic basalts. There, individual samples vary only slightly in degree of silica saturation or undersaturation. The normative distinction among the picritic basalts may therefore be small, having significance only for more evolved rocks. Indeed, the normative distinction between the two trends stems almost entirely from about 1±2 wt % difference in SiO 2 contents, this widening with decrease in MgO contents. Otherwise, rocks from the two trends are similarly sodic and potassic and they have similar concentrations and proportions of incompatible trace elements (Farley et al., 1993). The samples studied lie along the tholeiitic trend in Fig. 2 and encompass about two-thirds of the range in normative hypersthene, with the exception of one sample, a picrite that is very slightly Ne-normative. The extended tholeiitic trend parallels that for Puna Ridge. However, whereas the latter reaches strongly quartz-normative compositions, the Juan Fernandez tholeiitic trend is shifted in its entirety to the left, lying almost exclusively in the Di±Ol±Hy ternary. This is a consequence of somewhat higher total alkalis, especially Na 2 O, and somewhat lower SiO 2 and Al 2 O 3 contents. A closer match in all these characteristics is provided by tholeiites of the Honomanu volcanic series of Haleakala volcano, island of Maui (see Chen et al., 1991). The Juan Fernandez silica-undersaturated sequence is more extensively developed on Robinson Crusoe than on Alexander Selkirk. On Robinson Crusoe, a post-shield basanitic series followed that is even more strongly silica undersaturated (Fig. 2). The diversity of basalts sampled on Robinson Crusoe is undoubtedly enhanced by the greater age and more advanced stage of erosional dissection of this volcano than of Alexander Selkirk (Baker et al., 1987). Projection of the tholeiitic and alkalic trends toward the olivine apex in Fig. 2 is an indication that the compositions of many of the basalts are olivinecontrolled. That is, the proportion of olivine phenocrysts they contain largely determines variations in rock compositions. The olivine is accumulative, as Bowen (1928) argued, and the bulk compositions of the rocks therefore do not represent liquids. Indeed, some samples reported by both Baker et al. (1987) and Farley et al. (1993) contain 19±22% MgO, with 35± 45% of olivine in the norm. Olivine control is especially evident on a plot of Ni vs MgO (Fig. 3a). Ni partitions strongly into olivine, but varies as a function of forsterite content (Hart & Davis, 1978), being lower in more iron-rich olivines. Separate 423

4 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 Fig. 3. Comparison of chemical attributes of tholeiitic and alkalic olivine basalts from the Juan Fernandez Islands and Kilauea±Puna Ridge. Symbols are as in Fig. 2. Oxides are given in weight per cent, trace elements in parts per million. (a) MgO vs Ni; (b) MgO vs total iron as FeO; (c) MgO vs K 2 O; basanites are plotted on this diagram only; (d) MgO vs Zr; (e) TiO 2 vs CaO. Schematic trends: 1, olivine-controlled basalts; 2, olivine±plagioclase±clinopyroxene-controlled basalts; dashed, hybrids produced by mixing. In (a), the dashed and shaded trends show effects of control of olivine with different compositions on trends for Kilauea±Puna Ridge (KP) and Juan Fernandez (JF). In (b), lines of constant FeO T show the effects of control of olivine having different compositions, as indicated. In (c) and (e), lines with arrows indicate `olivine-control' trends. In (e), low-cao Juan Fernandez tholeiites at given TiO 2 and lying between Trends 1 and 2 are encircled by a continuous line. These appear to be hybrid lavas produced by mixing between primitive, olivine-rich basalt along Trend 1, and strongly differentiated, high-tio 2 basalt well along Trend 2. trends for Kilauea and Juan Fernandez suggest that the latter basalts in most cases carry more iron-rich olivine, as does a plot of total FeO vs MgO (Fig. 3b). Horizontal trends in Fig. 3b through the principal data clusters for Kilauea±Puna Ridge and Juan Fernandez indicate the nominal FeO contents of olivine of representative compositions controlling the trends. At Kilauea±Puna Ridge, olivine megacrysts have an average composition of about Fo 88±89 ; such olivine has about 2500±2800 ppm Ni (Clague et al., 1995). 424

5 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS Juan Fernandez basalts have a spectrum of higher FeO contents, giving a range in olivine compositions of Fo 88±85 ; these olivines have 2000±2500 ppm Ni (electron-microprobe measurements). The control of olivine on Ni in rock compositions is shown in Fig. 3a. Plots of incompatible elements vs MgO (Fig. 3c and d) are similar, showing slightly higher concentrations of these elements in Juan Fernandez tholeiites and alkalic olivine basalts than in the Puna Ridge picritic glasses. K 2 O contents in a few Juan Fernandez samples are low, perhaps because of groundwater leaching. Trends for tholeiitic glasses from Puna Ridge and for aphyric tholeiites from Juan Fernandez have inflections departing from olivine control at about 7% MgO, at which point at Juan Fernandez plagioclase joins the liquidus, and clinopyroxene shortly thereafter. This is shown schematically in Fig. 3c and d, with the effects of olivine separation on Kilauea glasses and aphyric Juan Fernandez basalts occurring along Trend 1, and multiphase control at lower MgO content along Trend 2. The MgO value at the inflection in both diagrams corresponds to liquid Mg L [ ˆ 100Mg/ (Mg Fe 2 )] of 60. The value of MgO (or Mg L ) for this inflection among Juan Fernandez aphyric basalts is somewhat imprecise, because fewer analyses are available to define it than for the Puna Ridge glasses; there is a wider range of Juan Fernandez tholeiitic and alkalic compositions, and, as discussed next, some of the rock compositions have been influenced by mixing. The olivine-control trends along which each sample lies also probably have slightly different inflection points. Some of the elevated concentrations of incompatible elements in Juan Fernandez picritic tholeiites appear to have resulted from mixing between primitive and porphyritic olivine-rich basalts originally lying at the high-mgo end of Trend 1 in Fig. 3c and d, and highly differentiated basalts, lying near the end of Trend 2. Several of the tholeiites I have studied in detail, for example, fall along such hypothetical mixing trends. Mixing between two such magmas would produce a hypersthene-normative hybrid, but with concentrations of alkalis and other incompatible elements reaching those in nepheline-normative alkalic olivine basalts. That mixing probably involved a primitive, porphyritic end-member is indicated for some samples especially by a plot of TiO 2 vs CaO (Fig. 3e). The effects of addition of just magnesian olivine to various compositions along Trend 2 are indicated. However, olivine is not a liquidus mineral along Trend 2, and it was not simply added to those magmas, like adding salt to soup. Instead, it was added as part of a porphyritic primitive magma slurryðby magma mixingðwith the highest proportion of the slurry being in those samples labeled `mixed', with about 2% TiO 2 content and containing about 30% olivine phenocrysts. Wright & Fiske (1971) and Clague et al. (1995) demonstrated that mixing between primitive and differentiated magmas was an important process in tholeiite petrogenesis at Kilauea±Puna Ridge. However, comparison of bulk-rock and glass compositions shows that mixing there occurred primarily between primitive magmas and only somewhat more differentiated magmas, both lying along or very close to the olivine-controlled portion of the differentiation sequence. Only a few Juan Fernandez picrites fall along Trend 1. There mixing was more often between more strongly contrasting primitive and differentiated magmas. In summary, one of the most significant attributes of the basalts from Juan Fernandez is the close resemblance, but not identicality, to sequences of Kilauea and Puna Ridge. Tholeiitic picrites from both Robinson Crusoe and Alexander Selkirk have slightly higher total alkalis, lower silica and alumina, and somewhat greater concentrations of highly incompatible elements. The influence of olivine control, and the point during crystallization when olivine control ceases, are much the same. Magma mixing was also important, but among the several Juan Fernandez tholeiites studied, it occurred between more extreme magma types, which enhanced alkali concentrations among the hybrid rocks. The Juan Fernandez volcanoes are nowhere nearly as large as any in the Hawaiian archipelago, yet this appears to be the only other island group built on old ocean crust in the Pacific where low-alkali tholeiitic basalts have erupted in any abundance. Detailed field relationships between tholeiites and silica-undersaturated alkalic picrites are still uncertain (Farley et al., 1993), but the basanites, which contain upper-mantle xenocrysts, clearly are the youngest eruptives on Robinson Crusoe (Baker et al., 1987). In this respect, they are comparable to the xenolith-bearing undersaturated lavas of the post-shield rejuvenated, once termed post-erosional, stage of Hawaiian volcanism. The mixing pattern of Juan Fernandez tholeiites is apparent in the compositions of olivine and spinel, as discussed below, and is important to the mechanism of incorporation of volatiles into olivine that I propose. ANALYTICAL PROCEDURES Polished thin sections were examined using both transmitted and reflected light at magnifications from 63 to Best results in reflected light were obtained using differential interference contrast, which highlighted residual relief from polishing around both 425

6 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 mineral inclusions and bubbles. Oil immersion at high magnification heightened reflectivity contrasts between iron±titanium oxide minerals and spinel, and reveals myriad minute sulfides. The study is concentrated on the phenocrysts themselves, the shapes of their inclusions, and the other phases besides volatile constituents those inclusions contain (minerals and/or glass). This is dictated by petrographic technique. Solid phases are the only ones that can be seen on polished or broken surfaces. Accordingly, I identify varieties of cavities, assuming that these were the vessels that once contained volatiles, whether fluid or vapor, assuming that rupture of these cavities released all original volatiles, including the noble gases that were measured for isotopes. In discussion below, cavities that lack mineral phases and either glass or devitrified glass are termed `empty', although they originally contained volatiles. Individual olivine grains originally concentrated by magnetic separator as splits for the isotopic analyses were mounted on stubs, coated with Au±Pd, and studied by scanning electron microscopy (SEM) at various magnifications at the Rosenstiel School, University of Miami. Photomicrographs were scanned at 300 dpi into a computer and then scaled and adjusted for contrast using Adobe Photoshop. In some cases, enhancements were applied to selected portions of photomicrographs, or to individual objects on them. Mineral compositions were determined on six tholeiites, two of them from Robinson Crusoe and four from Alexander Selkirk, selected from the samples analyzed for major oxides by Farley et al. (1993). One additional sample from Alexander Selkirk was studied by electron microprobe, but it has no rock analysis. On the basis of mineralogy and petrography, it is a tholeiite. Three additional samples, one of them analyzed, from Alexander Selkirk were examined in thin section only. Three basanites were also studied by electron microprobe, but the data are not reported here. Mineral analyses were normalized to the same mineral standards in all cases. The standard for olivine was US National Museum (USNM) San Carlos olivine. For spinel, USNM New Caledonia chromite was used except for TiO 2, total iron as FeO, and MnO, which were normalized to the standard values for USNM Tiebaghi ilmenite. PETROGRAPHIC SUMMARY In addition to containing olivine and rare skeletal plagioclase phenocrysts, tholeiitic picrites from the Juan Fernandez Islands have fine- to medium-grained groundmasses consisting at least of plagioclase, clinopyroxene, and either titanomagnetite or ilmenite and titanomagnetite. The phenocryst±groundmass sequence gives the order of crystallization in all samples regardless of grain size, based on idiomorphic relationships. This differs slightly but systematically from Kilauea tholeiites in that plagioclase in all cases precedes clinopyroxene instead of joining the liquidus at the same time or shortly after (see Wright & Fiske, 1971; Helz & Thornber, 1987; Clague et al., 1995). The distinction does not hold for melt inclusions in olivine phenocrysts, as described below. The lavas are variably vesicular. The more crystalline ones, which presumably came from flow interiors, have scattered, round, pinhole-sized vesicles. The more obviously quenched ones from flow tops have a greater percentage of vesicles that are irregular in shape and up to 3 mm in longest dimension. Flow-top plagioclases are small, acicular, and separated by a dark cryptocrystalline mesostasis; flow-interior plagioclases are larger, usually tabular, and are generally separated by more coarsely crystallized clinopyroxene and oxide minerals. Oxide minerals vary greatly in morphology, tending to be extremely elongate with elaborate cellular skeletal and dendritic morphologies in quickly crystallized flow tops, and more nearly equant but far less cellular and skeletal in flow interiors. In flow interiors ilmenite tends to elongate rhombic cross-sections and titanomagnetite to equant morphologies bounded by silicate crystal margins. Neither orthopyroxene nor primary sulfide occurs in the groundmass of any sample. The rock groundmasses indicate stages of differentiation that are nearly independent of bulk MgO contents; the latter depend more on the circumstantial extent of olivine accumulation. In less differentiated groundmasses, skeletal olivine is intergrown with euhedral plagioclase and encloses minute Cr-spinel. More differentiated groundmasses have no skeletal olivine. Tabular or microlitic plagioclase predominates, and it is partly encased by clinopyroxene. Two such samples also have fairly large anhedral clots of intergrown ilmenite and titanomagnetite, in addition to much smaller crystals of both minerals scattered throughout the groundmass. Some of the clots are attached to rims of olivine phenocrysts. Another sample has a small, felsic, and probably andesitic clot lacking clinopyroxene, and having only sparse and very tiny oxide minerals. Several of the rocks have swirls of different color, crystallinity, and vesicularity. Oxide minerals have different morphologies, being skeletal in the coarser-grained portions of the rocks, and both smaller and more equant in the finer-grained portions. The darker colors and contrasting crystal morphologies are probably an indication of a higher proportion of oxide minerals and clinopyroxene, and also different crystallinity, crystal morphology, and grain size. The 426

7 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS swirls therefore probably result from compositional heterogeneity and are a consequence of incomplete magma mixing. Among these lavas, the contrast between those with olivine crystallizing in the groundmass and those lacking olivine in the groundmass probably reflects a contrast in the proportions and compositions of liquids that originally crystallized and accumulated olivine, and those that now host it, an indication of mixing between primitive and differentiated magmas. The extreme contrast is the rock in which olivine phenocrysts enclosing Cr-spinel are intergrown with ilmenite and magnetite at their rims. A secondary contrast has to do with the proportion of ilmenite to titanomagnetite, and even the absence of ilmenite in some samples. This may be related to different Fe/Ti ratios in the crystallizing liquids (Wright & Fiske, 1971). Olivine phenocrysts are present in various proportions (up to 28%) and sizes; many are rather large (09 cm). In general, these conform to the five classes identified in tholeiites of the 1959 eruption of Kilauea volcano by Helz (1987), namely: (1) large, irregular blocky grains, often with planar extinction discontinuities, or kink bands; (2) euhedral and skeletal grains; (3) resorbed grains with original morphologies not evident; (4) angular to conchoidal fragments; (5) grains with swarms of inclusions of opaque minerals and pale glass. However, I do not attach the same genetic significance to the classification as did Helz (1987), nor are the distinctions between the classes necessarily sharp. For example, we may compare two examples of blocky, irregular crystals (Fig. 4a and b) with two euhedral crystals (Fig. 4c and d). The olivine in Fig. 4a is faceted on every side, even though it is irregular in shape. The olivine in Fig. 4b is not faceted only because of small chips and indentations in its outline. Otherwise, it would be almost identical to the olivine of Fig. 4d. Almost all of the `blocky and irregular' olivine in our samples is like this. The original outlines have simply been chipped away by some mechanical process, but in many cases are still partly present. Thus, the faceted, or nearly faceted, aspect of both classes is their most important feature, with the precise shape depending mainly on the cross-section of the mineral, or of intergrown grains, in the slide, less whatever has been nibbled from the edges. Therefore no fundamental distinction exists between the two classes, the irregular grains are not xenocrysts, nor did they necessarily crystallize in some other magma than euhedra of the same general size in the same sample. Both types clearly crystallized while being surrounded by melt. As to broken grains (Class 4), these typically conform in shape to the fracture-bounded internal segments of any of the olivine crystals shown in Fig. 4a±e, and thus more than likely originally were parts of phenocrysts as well. There are very few strikingly skeletal grains (Class 2) or those with amoeboidal outlines (Class 3). The latter are simply unusual sections through strongly skeletal crystals rather than the result of resorption. A few grains are combined in two- to three-crystal aggregates. Some grains are rounded, or have rounded interiors surrounded by a normally zoned rim with a faceted margin. In the most porphyritic samples, about 10% of the Class 1 and Class 2 olivine crystals exhibit coarse sub-grain development, irregular extinction, or kink banding (Fig. 4e and f). Such grains, and rounded grains, may be common in one sample but absent in another. One sample contains a dunitic xenolith about 1 cm in diameter. Helz (1987) suggested that isolated Class 1 crystals in basalts from Kilauea `were produced by disaggregation of coarse, mildly deformed dunite' (p. 712). Among these grains are those with planar deformation surfaces such as kink bands and rectangular subgrains. Helz (1987) cited experimental data (Raleigh, 1968; Carter & Ave Lallement, 1970; Raleigh & Kirby, 1970) to the effect that such deformed olivine forms in a crystalline matrix subject to non-hydrostatic stress. She concluded, `kink-banded crystals surrounded by melt must have been deformed in a very different environment' (p. 712). Partly from this suggestion, Clague & Denlinger (1994) argued that Class 1 olivine in Kilauea tholeiites, particularly that with kink banding, was probably scavenged from dunitic cumulates residing deep within the volcano. The faceted, euhedral but deformed crystals such as those in Fig. 4e and f, were not scavenged from dunite. Dunite is a cumulate, indeed an adcumulate, in which an original loose packing of minerals has been transformed by a process of grain boundary dissolution, reprecipitation, and textural equilibration (Hunter, 1987, 1998) to a collection of intergrown grains almost none of which retain their original euhedral morphology (Fig. 4g). The presence of subgrains in faceted olivine means instead that euhedral phenocrysts can indeed experience non-hydrostatic forces within a magma; that their edges may become chipped or even rounded by mechanical abrasion; and that some of the grains can break into fracture-bounded segments during magma transport and flow. Later I suggest a mechanism for this that also has bearing on the origin of certain types of melt and fluid inclusions. In all samples, olivine typically hosts one to several much smaller (503 mm 02 mm, and typically 01 mm 007 mm), euhedral, black crystals of chromian spinel. Rarely, some dozens of spinel grains and other inclusions occur in a single grain. These correspond to Class 5 olivines of Helz (1987), but rather 427

8 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 Fig. 4. Features of olivine phenocrysts. Olivine compositions (Fo) are indicated. (a) Blocky, euhedral olivine with almost no inclusions, sample PIN-8, Robinson Crusoe; transmitted light. (b) Block, irregular olivine enclosing some spinel. Sample MF-C4, Alexander Selkirk. Lines are drawn to indicate the original euhedral shape of the crystal. Transmitted light. (c) Euhedral olivine phenocryst enclosing a few spinels, and containing numerous ellipsoidal cavities parallel to crystal faces at upper right and lower left. The lower left face is partly broken. Sample MF-C4, Alexander Selkirk. Transmitted light. (d) Euhedral olivine similar in shape to that in (b), but with a partly broken edge at lower right. Sample PIN-12, Robinson Crusoe. Crossed nicols. (e) Euhedral olivine with subgrains. Sample PIN-12, Robinson Crusoe. Crossed nicols. (f) Two-crystal aggregate, having partly faceted and partly broken edges, and containing parallel deformation lamellae with irregular extinction. Sample PIN-12, Robinson Crusoe. Crossed nicols. (g) Dunite cumulateða xenolith from Ta'u Island, American Samoa. Absence of euhedral outlines, and the olivine with faint subgrains at lower right, should be noted. Many crystals have 120 triple junctions, indicating textural equilibrium in a monomineralic cumulate (Hunter, 1996). Sample 82MT-X2. Crossed nicols. 428

9 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS than considering such grains as a distinctive category, they are only one extreme in a population of olivine grains having from `none' to `many' such inclusions. A few spinel grains occur at olivine rims where they may be skeletal. In samples having the less differentiated groundmass, spinel also occurs as euhedral or skeletal isolated phenocrysts, generally smaller than spinel within olivine. Some rim and groundmass spinel has more reflective titanomagnetite margins; in the basalt with most differentiated groundmass, the rim spinellid mineral is actually chromian titanomagnetite intergrown with ilmenite. In the two samples with the most olivine phenocrysts, a few olivine grains enclose spherical sulfides either as isolated or multiphase inclusions. The larger sulfides comprise both pentlandite [(Fe, Ni) 9 S 8 ] and pyrrhotite. One other more differentiated sample has a trace of sulfide in the inclusions in one part of one olivine grain. Absence of sulfides in phenocrysts of most samples suggests that their S was degassed before the olivine crystallized. In all samples, degassing reduced S concentrations to well below sulfide saturation before crystallization of the groundmass. PETROGRAPHY OF INCLUSIONS Photomicrographs and scanning electron micrographs are presented generally in the order of increasing degree of magnification, but are to some degree interspersed to illustrate aspects of the same phenomena at different magnifications. General characteristics of inclusions in olivine phenocrysts As mentioned above, olivine phenocrysts have varying percentages of mineral±melt±vapor inclusions, ranging within the same sample from none to many (Fig. 4a; Fig. 5a±c). The olivine ranges from blocky and euhedral (Fig. 4a, c, and d) to fairly rounded (Fig. 5a±c). In transmitted light, both melt inclusions and spinel appear as black inclusions in photomicrographs. Many grains have central cores that are fairly densely charged with inclusions, and local concentrations of spinel-bearing inclusions elsewhere. The grains in Fig. 5a±c illustrate a common modality of inclusion arrangement, that of linear or curving arrays. These are the features that Roedder (1965) originally described as secondary inclusions, and that he interpreted as annealed fractures along which inclusion material was brought into the crystal. In some cases these are parallel. In Fig. 5a, three linear arrays (near arrows) converge on a point outside the boundary of the present olivine. In all the olivine, such arrays do not correspond to the current generation of fractures crossing the grains. The olivine of Fig. 5c combines a dispersed arrangement of inclusions with curvilinear arrays. The olivine of Fig. 5d is distinctive within the sample in having an extraordinarily dense concentration of small inclusions and some large ones carrying spinel. Some rows of spinel parallel the upper and lower crystal boundaries, and there is one large and rounded melt inclusion now crystallized to spherulitic basalt at center left. This grain has a younger generation of fractures slightly oblique to the linear arrays of spinel and the upper crystal faces. The inclusion-bearing region is bounded by a fairly wide rim, free of inclusions, but not of uniform width. At high magnification in reflected light (Fig. 5e) the dense array of small inclusions is seen to consist of elliptical to rectilinear cavities in the olivine; these are variously filled with devitrified melt (dark gray), spinel (light gray) and sulfide (bright). Dark gray shadowed pits in some of these are bubbles. The larger of these are oriented parallel to the linear arrays of large spinels in Fig. 5d. Although the orientation of these inclusions exhibits a strong crystallographic control, there is no indication that they were incorporated into the olivine along fractures. This particular olivine corresponds to perhaps the extreme variant of class 5 (grains that contain a very high density of tiny inclusions) in the classification of Helz (1987), even to the extent of containing sulfides. She noted the rarity of such grains in the 1959 Kilauea eruptives and in samples from the Kilauea Iki lava lake. Schwindinger & Anderson (1989) and Clague et al. (1995) illustrated similar olivine densely charged with inclusions from Hawaii. Farley & Craig (1994) found highly variable yields of Ar in several splits of olivine grains from the same sample (PIN 12; Robinson Crusoe). This apparently reflects the existence of variable inclusion densities in different olivine crystals seen during petrographic examination. Inclusion arrays and indications of deformation Figure 6a±c shows several olivine grains containing both inclusions and subgrains or planar extinction discontinuities bounding kink bands. Usually there is no relationship between curvilinear arrays of inclusions and subgrain or kink boundaries (Fig. 6a, near arrows, and Fig. 6c). In Fig. 6c, the rounded and deformed interior of the olivine is carapaced with an undeformed outer zone. Three trains of inclusions converge toward the upper right portion of the crystal, reaching the outer zoned margin. The mineral was first deformed, then incorporated into magma, then rounded, and then the curvilinear arrays of inclusions 429

10 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 Fig. 5. Features of inclusions in olivine phenocrysts photographed in transmitted light. Olivine compositions (Fo) are indicated. (a)±(c) show olivine phenocrysts in sample MF-C2, Alexander Selkirk. (a) Olivine phenocryst with several linear inclusion arrays trending toward the upper right. (b) Olivine phenocryst with nearly orthogonal inclusion arrays. (c) Rounded olivine phenocryst with many inclusion arrays. (d) and (e) show sample MF-S1, Alexander Selkirk. (d) An olivine phenocryst with a core densely charged with inclusions, and a mantle with no inclusions. (e) Detail of a portion of the same olivine phenocryst, photographed using differential interference-contrast reflected light. In the inclusions, which are recessed by polishing, dark gray is glass, light gray is spinel, and bright is sulfide. formed. Figure 6b provides an example in which several spinel-bearing inclusions lie along a planar deformation boundary (near line with arrowheads). In this case, the deformation occurred after incorporation of the inclusions, and the location of the planar deformation boundary may have been guided by internal stresses in the olivine associated with the inclusions. Incorporation of inclusions along curvilinear arrays proceeded even into the eruptive history of some of the rocks. Some olivine is decorated with triangular dendrites in optical continuity with the mineral interiors (Fig. 6d). Growth of such dendrites rather than continued uniform growth of the exterior of the crystals indicates subjection to suddenly higher cooling rates, or undercooling (Donaldson, 1976). In these examples, formation of dendrites was the last growth the olivine experienced before fairly rapid cooling of the groundmass following eruption. Scanning electron micrographs show that the dendrites have the typical faceted pyramidal terminations of single olivine crystals (Fig. 6e and f ), and that they budded preferentially on the curving edges of flat, stepped surfaces that are bounded by fractures orthogonal to the surface of the olivines (Fig. 6e). In detail the orthogonal fracture surfaces form sets of fractures with small offsets, but curving in total 90. The stepped morphology of the crystal edge is seen in thin sections to be particularly common on crystal faces paralleling the c-axis in rocks with a fine-grained groundmass, in which it is evidently a consequence of undercooling. The same crystals usually have skeletal projections at terminations on 101 or 021 crystal faces. The growing olivine was first 430

11 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS Fig. 6. Deformed olivine and olivine with epitaxial dendritic overgrowths. Olivine compositions (Fo) are indicated. (a) Olivine phenocryst with subgrains. Crossed nicols. Sample MF-S1, Alexander Selkirk. (b) Kink-banded olivine in the same sample. Crossed nicols. (c) Olivine with irregular extinction. Same olivine as in Fig. 4c. Crossed nicols. Sample MF-C2, Alexander Selkirk. (d) Olivine phenocryst with a dendritic rim and an inclusion train crossing one dendritic projection. Transmitted light. Sample MF-C2, Alexander Selkirk. (e) Scanning electron micrograph of dendritic projections growing on a stepped olivine crystal surface. Sample MF-3, Alexander Selkirk. (f) Scanning electron micrograph showing detail of olivine elsewhere on the same olivine. subjected to stresses that allowed the orthogonal fractures to form, defining the small stepped offsets on the mineral surface. At still higher undercooling, the dendrite buds then nucleated preferentially at the locations of strongest curvature, perhaps because this is where surface tension of the growing mineral surface was most disrupted. The olivine with a dendritic surface in Fig. 6d is complexly fractured, and also has a number of linear or curvilinear trains of inclusions, some of them fairly obviously arranged along fractures, some of them not. One of these arrays, however, is along a fracture (near line with arrowheads) cutting through the pyramidal cross-section of one of the dendrite studs on the mineral boundary. This is shown enlarged in the inset. The inclusion array thus formed after the dendrite began growing, that is, when cooling rates abruptly increased sometime just before eruption, or perhaps even while the lava was flowing. In one sample, an inclusion array formed late in the groundmass, within a clinopyroxene that subophitically encloses tabular plagioclase and that is adjacent to an olivine phenocryst. Varieties of inclusions To this point I have mainly described curvilinear arrays of inclusions that apparently define fracture surfaces, or annealed fracture surfaces, in the olivine. Several other varieties occur, including cavernous inclusions, mineral±melt±bubble inclusions, open cavities around spinel, and faceted sub-microscopic pits, the last of which can be seen only at very high magnification. Cavernous and ellipsoidal inclusions A tholeiite (sample MF-C4), contains numerous olivines with large cavernous and ellipsoidal cavities. These are especially evident in reflected light (Fig. 7) and occur in both rounded (Fig. 7) and euhedral faceted grains (Fig. 4b). Some of the cavities have angular surfaces, but most are ellipsoidal, with their long axes parallel to the long dimension of the mineral, which is the crystallographic c-axis (Fig. 8a). Some of the cavities clearly acted as stress guides for later contraction fractures, one set of which in Fig. 7 is prominently nearly concentric to the outer surface of the mineral. Two spinel inclusions in the olivine also contain such cavities. One linear array of microinclusions diagonally cuts the lower third of the mineral (double-headed arrow). Scanning electron micrographs reveal the tendency of the larger, angular cavities in the olivine from this sample to have faceted morphologies (Fig. 8b±d). 431

12 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 Fig. 7. Composite photomicrograph in reflected light of an olivine phenocryst with large, ellipsoidal, cavities (black), and which also encloses several Cr-spinels (white). Light gray spots are cavities below the polished surface of the grain revealed by internal reflections. The linear inclusion train to the right of the arrowed line should be noted. Sample MF-C4, Alexander Selkirk. Some smaller cavities contain euhedral crystals of plagioclase and pyroxene (Fig. 8c and d). Another sample, MF-C3, has olivine with faceted inclusions partly filled with devitrified melt (Fig. 8e). A spinel crystal once occupied the sharply curving central cavity in this inclusion. Its space is bounded at the top by a bubble. Other inclusions in olivine from this sample, although exquisitely faceted, are empty (Fig. 8f ). Ellipsoidal inclusions in sample PIN-12 are teardrop shaped (Fig. 8g). Orientation along a common fracture is suggested. The inclusion evidently was once larger, being surrounded now by a zone of secondary olivine growth, as identified by dispersed X-ray emission analysis. Mineral±melt±bubble inclusions The most easily seen variety of inclusion in Juan Fernandez olivine is the large, usually isolated cavity containing devitrified melt, crystals of silicate or spinel, and one or more bubbles. In two samples, PIN-12 and MF-S1, these inclusions also contain sulfide. Such multiphase inclusions range in size down to the densely swarmed dust-sized inclusions in the olivine of Fig. 5d. Reflected light reveals dendritic herringbone clinopyroxene (Fig. 9a and b), sulfide and spinel. In Fig. 9a, the open cavity or bubble is in the spinel; in Fig. 9c, the bubble, with a sulfide at its rim, is in the devitrified melt (dark gray). In Fig. 9a and c, the mineral that resembles a stack of blocks is dendritic clinopyroxene. The dendritic morphology of the pyroxene is developed spectacularly in Fig. 9b, where its appearance is heightened in relief by alteration of the surrounding glass to soft clay minerals. Sulfides show up prominently in multiphase inclusions of sample PIN-12 (Fig. 9a). Some cluster on the surface of the large spinel at the top. Oil immersion allows optical resolution of sulfides both on the polished surface of the section and, by nearly matching the index of basaltic glass, to some extent within it, by internal reflection off surfaces of sulfide globules. These grow fainter with depth over the distance of a few microns. The effect is better illustrated by a direct comparison of the same inclusion, provided in Fig. 9d and e. Both photomicrographs are in reflected light, but the larger of the two, using oil immersion, clearly 432

13 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS Fig. 8. Features of large cavities in olivines. (a) Close-up of some cavities from the olivine of Fig. 6, reflected light. (b) Scanning electron micrograph of a cavernous faceted cavity in another olivine of sample MF-C4. (c) Small faceted cavity with intergrown crystals of pyroxene and plagioclase. Sample MF-C4, Alexander Selkirk. (d) Faceted cavity in an olivine phenocryst containing glass, the rounded mold of a Cr-spinel, and a bubble at top right. Sample MF-3. (e) Small rounded cavity with a large tabular crystal of plagioclase and cellular clinopyroxene tending to branch. Sample MF-C4. (f) Regularly faceted sawtooth cavity. Sample MF-3. (g) Ellipsoidal cavity along a fracture. Sample PIN-12, Robinson Crusoe. The smooth material lining the cavity is olivine, as determined by peak heights using energydispersive X-ray spectrometry. highlights the tiny sulfide droplets that are dispersed in the glass, and diminishes the reflectivity contrast between silicate minerals and glass. The `scorpion's hook' here is clinopyroxene. There is a small bubble in this inclusion, at the upper left in Fig. 9d. Most multiphase inclusions contain glass. In some cases they have a great deal of glass, even when host olivine is in fairly coarse-grained rocks. Clinopyroxene in the inclusions is spherulitic to cellular±dendritic: morphologies typical of this mineral in rapidly cooled basalts. With glass present, the morphologies resemble those commonly seen within 1±3 cm of the glass rims of submarine pillow basalts (see Bryan, 1972; Natland, 1979). They suggest undercoolings of tens of degrees to perhaps more than 100 (Kirkpatrick, 1979). The other consequence of the presence of glass is that the bubbles in them are actually trapped vesicles that were quenched in place, just as they are in pillow basalts. They are not bubbles that nucleated and grew in situ, as a consequence either of pressure reduction during eruption or of crystal growth within the inclusion during slower cooling of the host groundmass. Then there would have been no glass. These inclusions demonstrate that olivine occluded spinel, and that inclusions also often contain partly devitrified glass and a vapor phase plus sulfide in a minority of samples. Considered individually, many inclusions contain disproportionately high percentages of spinel or spinel and sulfide. As first noted by Roedder (1965), olivine phenocrysts are clearly the preferential surfaces of nucleation at least for spinels and bubbles, often together. The tendency was 433

14 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 Fig. 9. Photomicrographs of multiphase inclusions. (a) Inclusion containing dark gray glass, light gray dendritic clinopyroxene, a large white spinel, and dozens of tiny, bright sulfides. Oil-immersion reflected light. Sample MF-S1. (b) Spectacular pyroxene dendrites, with relief enhanced by alteration of surrounding glass, in an inclusion. Sample MF-C4. (c) Several small multiphase inclusions, two with dendritic clinopyroxene and spinel, the larger with a bubble, in normal vertically illuminated reflected light. Sample PIN-8. (d) and (e) are two views of the same multiphase inclusion, with an unusual hook-like faceted and partly cellular aggregate of clinopyroxene, two marginal bubbles, and numerous sulfides. Sample MF-SI. Many more sulfides show up under oil-immersion reflected light (e) than under standard vertical illumination (c). (f ) Skeletal spinel at the margin of an olivine phenocryst. Sample MF-C4. Standard vertical illumination. 434

15 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS undiminished by high cooling rates experienced near the point of eruption, when dendritic outer zones of olivines grew; those zones still incorporated skeletal spinel (Fig. 9f ). The disproportionate share of the volume of inclusions belonging to spinel and sulfides is best seen in smaller inclusions, whether they formed along curvilinear arrays or in dense swarms like those of Fig. 5e. Figure 10a shows a pitted spinel occupying almost all the space of one inclusion, but there is a small amount of devitrified melt, and a bubble, at the upper left. Figure 10b±d shows spherical sulfide inclusions, some with a meniscus separating the sulfide, which was an immiscible liquid phase, from a vapor phase. The olivine grew completely around the globular sulfides and their adjacent volatile fluids. One inclusion in Fig. 10c is almost completely occupied by two phases, sulfide and spinel, which divide it about equally. The inclusion to the upper left of this one is about two-thirds filled with devitrified glass, but about one-quarter of that is occupied by a sulfide. A bubble comprises about one-third of the area of the inclusion. The menisci in adjacent inclusions in Fig. 9b have different orientations (top, bottom, side) suggesting either that the olivine was still moving and that the bubbles were present when sulfide crystallization (at 700 C), took place, or that the effects of surface tension and viscosity within the inclusions prevented bubbles adjacent to menisci from responding to gravity. In the extreme, small inclusions can be occupied exclusively by spinel and a bubble cavity, or sulfide separated from a bubble cavity by a meniscus. Portions of curvilinear arrays sometimes have only spinel (Fig. 9e and f ); others contain only sulfide and bubbles (Fig. 9d). Some linear arrays with spinel are crossed by other arrays of tinier inclusions containing sulfide but no spinel. In most cases the odd somewhat larger inclusion in the array will contain some devitrified melt with very fine-grained dendritic pyroxene. In these rocks there evidently were effects of physical fractionation, first of molten silicate away from refractory spinel, then of molten sulfide and volatile fluids away from silicate melt, as injected material squirted its way along fractures propagating through the olivine. In the olivine of Fig. 5d, which swarms with small inclusions, the distribution of spinel is not uniform; they are especially concentrated just adjacent to the outer, inclusion-free zoned rim (Fig. 10g). Cavities adjacent to spinel Although empty spaces can be seen around many included spinels in reflected light, these features are best revealed on the broken surfaces of the olivines by SEM. Perfectly euhedral small spinel (10 mm) usually has small, curving cavities next to one or more crystal faces (Fig. 11a±c). The cavities exist even where the spinel protrudes into larger rounded cavities (Fig. 11b). Adjacent olivine also usually has stellate arrays of short fractures, each tending to originate from a spinel crystal interface (Fig. 11a±c). In cross-section, these typically form flat, curving trajectories around the spinel (Fig. 11d). A view of one half of a cavity, with the spinel removed (Fig. 11e), shows that the spinel was attached to one flat surface, and was contained in an otherwise spherical space now criss-crossed by fractures. Larger spinel is usually more thoroughly embedded in host olivines, but even these crystals are partly surrounded by empty space (Fig. 11f ). Sense of scale is important. Under a microscope, the small spinel crystals of Fig. 11a±e would be barely visible in the individual olivine grains shown in Fig. 5. The empty spaces next to them would not be visible at all. Faceted, sub-microscopic pits A considerable surprise during examinination by SEM was the discovery of myriads of extremely small, faceted pits on the surfaces of some olivine. In Fig. 12a, the tiny cavities bear some relationship to larger, cavernous cavities, as they decorate one face of a crystal termination, perhaps 021 or 010, whereas the several largest cavities are on the opposing and otherwise undecorated crystal face, and some intermediate-sized cavities are on the crystal edge where the two surfaces meet. At the lower left, the field of small cavities abruptly terminates at a curving string of intermediatesized cavities just above another intersection of planar crystal surfaces. The curving string of cavities may be a zonation boundary. The small cavities in this cross-section are of the order of 1 mm wide or less, and 1±7 mm long, having cross-sections of individual or connected parallelograms with faces oriented in the direction of opposing planar surfaces in Fig. 12a. In Fig. 12b and c, they are closely and regularly spaced. They would be impossible to see in transmitted light. Only the largest might be visible in reflected light, but their shapes would not be evident. Similar cavities viewed in a different orientation are shown in Fig. 12d±f. Here, the alignment and regular orientation of the inclusions again are evident, although a false curvature is imparted by the conchoidally fractured surface of the olivine being examined. The field shown in Fig. 12b is a continuation downward on the same curving surface as in Fig. 12e. The cavities are more coalesced and widely spaced. Figure 12f is a larger magnification of a portion of the field in Fig. 12d. The faceted aspect of the cavities is spectacularly expressed (compare with 435

16 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 Fig. 10. Spinel and sulfides in multiphase inclusions. (a) A rounded Cr-spinel, itself containing small inclusions recessed by polishing, occupies most of a larger inclusion, other phases of which, including a bubble, can be seen at the upper left. Shadowing of the small inclusions in the spinel, using differential interference contrast reflected light, reveals that they also contain bubbles. (b) Sulfides in small round inclusions along a curvilinear array. Several of the inclusions also have menisci in various orientations. Sample PIN-12. (c) Three small inclusions along a linear array in an olivine, each with different proportions of melt plus dendritic pyroxene, bubbles, spinel (gray) and sulfide (bright). Differential interference-contrast reflected light. Sample MF-C4. (d) High-magnification detail of sulfide inclusions, in oil-immersion reflected light, showing the presence of a meniscus and two sulfide phases, pentlandite (bright) and pyrrhotite (light gray) separated by an irregular boundary. Sample PIN-12. (e) A curving array of inclusions within an olivine. Sample MF-C2, Alexander Selkirk. The inclusions are filled mainly with spinel. The train of inclusions near the arrowed line is blown up in (f), in reflected light. The cavities near some of the spinel, and two small isolated cavities at upper right, should be noted. (g) A swirl of tiny spinel grains near the edge of the inclusion-rich olivine of Fig. 4f±h. Sample MF-S1, Alexander Selkirk, in differential interference-contrast reflected light. 436

17 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS Fig. 11. Scanning electron micrographs of spinel in olivine from sample PIN-12. (a)±(c) show relationships between small spinel euhedra, adjacent cavities, and fractures. The growth pattern of olivine next to the rounded cavity surface in (a) suggests inward growth or reaction of the olivine toward the spinel. (d) Face-on view of a fracture around a spinel grain. (e) A cavity like those of (a)±(d) from which the spinel has been removed. The spinel adhered to the olivine on the flat surface at lower left. The rest of the cavity surrounding the cavity, however, was nearly spherical. It is crossed by a number of fractures that formed by contraction of the olivine around the spinel. (f) Edge of a large spinel with irregular space between it and surrounding olivine. The smoothly crystalline olivine next to the spinel may have resulted from reaction or partial re-equilibration of the spinel with melt in the inclusion, and the host olivine. Fig. 8b, e and f ). The coalesced cavities are the inverse shape of so-called hopper crystals (Donaldson, 1976), with widened central portions and spiky projections from cavity terminations along crystal planes. To the extent that these examples reveal, the faceted pits are empty; they contain no spinel, no sulfide, and no trapped glass. The olivine of Fig. 5d and e, with its swarms of tiny multiphase inclusions, is as close to an exception as may exist, but many of those inclusions are larger, and they are not faceted. Whether the tiny faceted cavities are common or fairly rare is difficult to say without a more systematic SEM survey. They may be confined to or at least more prevalent on crystal terminations than in the interiors of individual grains. 437

18 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 Fig. 12. Scanning electron micrographs of tiny faceted cavities in olivines, sample KF-19. (a) Tiny pits along one crystal surface, concentrated both along one crystal edge and against an apparent external zone of the crystal devoid of cavities at lower left. Several much larger faceted cavities are on the opposing crystal face. (b) Detail of small cavities in (a), showing their regular spacing and the tendency of faceted surfaces to be oblique to the overall pattern of spacing. (c) Another view of closely spaced and regularly spaced tiny cavities showing a similar obliquity of faceted surfaces to the alignment of the cavities. (d)±(g) show a series of micrographs of a single array of tiny, faceted cavities on a broken and curving olivine surface. (d) shows the overall spacing and arrangement of cavities. (e) shows that the pattern of spacing and coalescence of cavities changes on the mineral surface below the area of (d). (f) A close-up view of a portion of (d). In this view, the faceted cavities are more symmetrically disposed along the general orientation of the inclusions. The contrast with (b) and (c) may be a consequence of a less oblique angle of the mineral surface to the cavities. 438

19 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS The illustrations here suggest a total porosity of about 5±10%. If submicroscopic pits are common, they may represent a significant proportion of the volume of volatiles trapped by olivines. In most samples, however, they are probably subordinate in volume to cavernous cavities or those adjacent to spinel. Tiny cavities termed micropores have been observed in scanning electron micrographs of alkali feldspars in granitic rocks (Walker et al., 1995). These are produced during the subsolidus transition between orthoclase and microcline, apparently by dissolution into late magmatic and hydrothermal fluids. The majority of them appear to form at temperatures 5400 C as a consequence of penetrative alteration of their plutonic hosts. Although there is usually a clear crystallographic control on their orientation, they are not arranged in rows and are rarely faceted. Usually, they are at least partially rounded and occur as spaces between subgrains in microperthites. The only parallel between them and the tiny, faceted cavities in Juan Fernandez olivine is that both formed in the presence of fluids. The latter, however, resulted from crystal±fluid interaction at the magmatic stage; there is no indication that the olivine surrounding the faceted cavities is altered at all. ORIGIN OF INCLUSIONS DURING CRYSTAL GROWTH A somewhat surprising impression is that olivine in these basalts is full of holes. The occasional low total for an olivine electron-microprobe analysis may be quite valid. Variations in Cr 2 O 3 contents beyond detection limits may have nothing to do with the olivine crystal structure; instead, the microprobe beam probably encountered some very tiny spinel. Concentrations of structural Cr in olivine may be beyond the detection limit of the electron microprobe. Most significantly, there is ample room in the various cavities combined for a large quantity of volatiles. Roedder (1965, 1984) identified two types of inclusions in olivine: `primary', which were incorporated directly into growing crystal faces, and `secondary', which were introduced into the mineral grains along straight or curvilinear fractures that are now annealed. The latter are clearly important in the Juan Fernandez olivine. However, the olivine also contains faceted cavernous inclusions, ellipsoidal inclusions, rounded multiphase inclusions, tiny single-phase or multiphase inclusions with preferential crystallographic orientation, cavities adjacent to spinel, and tiny faceted pits. These are all forms of primary inclusions. They are so diverse that this single term does not do them descriptive justice, although it does connote the common aspect of their origin. Drever & Johnston (1957) emphasized mechanisms of crystal growth, rather than resorption, as the most important factor in producing irregular morphologies among olivine phenocrysts. Experimental studies (Donaldson, 1976) extended this conception to the scale of tiny olivine with elongate, strikingly faceted or highly decorated morphologies that grew at exceedingly high undercooling, as for example in lunar samples or the margins of pillow basalts. The experiments showed that undercooling is systematically related to the diverse morphologies olivine exhibits from the interior of flows to their quenched margins. The same range of faceted morphologies exists here, but in reciprocal form, in the various primary cavities that surround inclusions, whether they are empty or filled with glass and minerals, among the Juan Fernandez olivine crystals. The tiniest ones are like the most elaborate hopper olivine in pillow margins. Somewhat larger ones have serrated boundaries of the type typical of skeletal or coarsely dendritic overgrowths. The largest ones may either be faceted or they are rounded, as are the large re-entrants in the coarsely skeletal olivine phenocrysts illustrated by Drever & Johnston (1957), and the similar and even larger `harrisitic' olivines, which are coarsely skeletal crystals orthogonal to layering in gabbroic intrusions, as discussed by Donaldson (1982). The only real distinction between skeletal re-entrants and the large cavities is that in most cases the cavities are empty or only partly filled. I consequently believe that all the primary inclusions described here have morphologies indicating crystal growth around them at a range of undercooling. Despite this, as all cavities occur in otherwise euhedral or tabular olivine, the general characteristics of crystal growth had to be at small undercooling, at rates of perhaps 02±04 mm/s (Donaldson, 1975), which would produce a typical euhedral phenocryst in a few days. Thus each re-entrant or faceted cavity represents a local spot on the crystal that experienced heightened undercooling. The key to this is the presence of bubbles. Theory predicts that when the surface energy between vapor and crystal is less than that between vapor and melt, then the activation energy for nucleation on the crystal surface is less than in the melt, thus the crystal surface is where bubbles will preferentially attach (Sigbee, 1969). Bubble nucleation consequently is dominantly heterogeneous, and favored by both the presence of crystals in the melt where bubbles attach, and the roughness of the crystal surfaces (Bagdassarov & Dingwell, 1993). The roughness reduces the activation energy needed for nucleation even further than the 439

20 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 simple presence of crystals. Observations on silicic melts also indicate that bubbles may also nucleate preferentially on certain minerals, in particular the Fe±Ti oxides (Hurwitz & Navon, 1994; Navon & Lyakhovsky, 1998). In Juan Fernandez tholeiitic picrites, another oxide mineral, namely spinel included in olivine, is almost invariably associated with a bubble, but whether this has to do with the spinel adding roughness to the surface of the growing olivine or a predilection for bubbles to nucleate on the spinel is difficult to say. Whether or not spinel is present, bubbles that nucleate on olivine surfaces would in the first instance be local sites without molten silicate to supply the growing olivine. The olivine would tend to grow around them. In the second place, the bubbles represent places where local liquidus conditions are modified. Lipman et al. (1985) and Lipman & Banks (1987) argued that increases in microphenocryst content during the eruption at Mauna Loa in 1984, but without a change in lava temperature, resulted from undercooling of 20±30 below the liquidus as a result of vesiculation and volatile release that occurred as the lava flowed down the mountain (see Cashman & Mangan, 1994). Release of volatiles simply elevates liquidus boundaries, driving melts from conditions that produce tabular crystals to those that produce skeletal or dendritic crystals. On the surfaces of growing phenocrysts, bubble nucleation would also modify melt viscosities, thus changing diffusion gradients in the layer of melt immediately around the crystals. Growing crystals also produce their own supercooling effect when there is material in the melt that is more soluble than in the crystal. This is called constitutional supercooling, and it results in the concentration of the impurity being greatest near the crystal (e.g. Kirkpatrick, 1975). Both the liquidus temperature and the undercooling increase away from the crystal; with the result that protuberances on the crystal grow faster the further they are from the main crystal interface. The tendency to faceted morphologies is accentuated. Regular spacing of inclusions and cavities is important. Kirkpatrick (1975, 1981) emphasized that there is as yet no adequate theory to account for planar interface stability during crystal growth, despite the fact that euhedral crystals are common. Existing theory predicts that, with increasing undercooling, interface instability should set in at some point controlled by the crystal growth rate Y and the rate of diffusion D of the rate-controlling component of the melt (Cahn, 1967). When D/Y is small, large euhedral crystals can form. As D/Y increases, planar interfaces are no longer stable. The first instabilities to appear have long wavelengths, resulting in faceted crystals. As D/Y increases still further, the spacing of instabilities decreases; morphologies are dendritic or skeletal. At very high undercooling, and greatest D/Y, morphologies are branching (spherulitic). Relationships between undercooling and morphologies are borne out by a number of petrographic and experimental studies (e.g. Kirkpatrick, 1979; Lofgren, 1983). Implicitly, the spacing of instabilities, and thence of protuberances on planar interfaces, is regular at a given undercooling (Keith & Padden, 1963). Thus the regular spacing of dendrites on olivine phenocrysts (Fig. 5e and f ), of ellipsoidal cavities (Fig. 7a), of inclusion dendrite spacings (Fig. 7h), and of rows of tiny faceted cavities (Fig. 11) are probably aspects of development of planar instabilities during the heightened undercooling occasioned by sudden cooling or vesiculation. In the case of tiny faceted cavities, the instabilities developed at a characteristic spacing of 3±5 mm, and determined where tiny bubbles nucleated. Given the otherwise euhedral morphology of the olivine enclosing these cavities, this may represent only incipient breakdown of interface stability, at small undercooling that would not be apparent at all except for the bubbles that formed. If this is the case, then the spacing may represent the critical distance for the onset of interface instability of olivine in these materials (Cahn, 1967). The presence of non-uniform boundary layers around the growing crystals is also important. At somewhat elevated undercooling promoted by vesiculation, spots near where bubbles are attached to an olivine would not experience as efficient an exchange of components into and away from the growing crystal as when nearer equilibrium; constitutional supercooling would not be uniform. Conversely, because of very local viscosity contrasts in the boundary layer, such exchange would be more efficient on portions of crystal surfaces away from bubbles. The surface energy of the interface varies, and this controls the shape of surface instabilities (Cahn, 1967). The bubbles therefore inhibit diffusion parallel to the crystal interface, this tending to promote crystal growth instabilities. In general, melt species that are particularly incompatible with the olivine structure should tend to build up irregularly in the boundary layer, being more concentrated nearer the bubbles. These would include alkalis, S, Cr, and volatiles, and would accentuate any tendency for irregularities on the olivine surfaces to act as sites for spinel and sulfide nucleation. This may account for the unusual concentration of Cr-spinel and, in two samples, sulfides, even in large multiphase inclusions. In crystalline basalt, there is no way to determine whether such boundary layers existed. Glass next to quenched olivine in pillow rims might have a somewhat different composition than glass further away, 440

21 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS Fig. 13. Tiny sulfides in boundary layers near olivine quench crystals and phenocrysts in the glassy margin of sample SD-7C, a picritic tholeiite from Siqueiros Fracture Zone, eastern Pacific (Natland, 1989). The photomicrographs are in oil-immersion reflected light. The olivine in (a) is a typical hopper crystal; that in (b) is more equant, but it has small dendritic prongs. The tiny bright specks are minute sulfides, concentrated in boundary layers adjacent to the olivines. The sulfides in (c) are in clear glass (black) between a spinel (light gray) and an olivine (dark gray). The spinel, sulfides, and glass are in a skeletal embayment in a large olivine phenocryst. In (d), the tiny sulfides are concentrated near the straight edge of a large, faceted, olivine phenocryst. but no one has as yet attempted to prove the existence of boundary layers in natural glasses using an electron microprobe, although some of the potential inhomogeneities perhaps have shown up in analyses of glass inclusions. For example, Dixon et al. (1991) reported S contents among Puna Ridge glass inclusions in olivines, other minerals, and gabbro xenoliths that are in excess of concentrations necessary for sulfide saturation. These occur even in basalts with quenched glasses not saturated in sulfide. Supersaturation in S can also be inferred using data from inclusions of the 1959 Kilauea summit eruption (Anderson & Brown, 1993). Dixon et al. (1991) speculated that the excess results from post-entrapment growth of the surrounding minerals. However, they did not consider the possibility of boundary-layer buildup during skeletal crystal growth, and did not report whether there are any very tiny sulfide globules in their inclusions like those in Fig. 8. Degassed basalts erupted subaerially or in shallow water do not have such high concentrations of sulfur, and thus are not likely to provide evidence for a boundary layer enriched in S or sulfide next to a crystal surface of a quenched olivine. Indeed, in most Juan Fernandez tholeiites, sulfide is not present even in inclusions in olivine phenocrysts, suggesting that S was almost completely lost to degassing before their crystallization. Nevertheless, as sulfide is present in inclusions of two samples, I have examined quenched glass next to olivine in the pillow rim of a primitive, sulfide-saturated picritic abyssal tholeiite dredged from Siqueiros Fracture Zone, eastern Pacific, using highmagnification (1600) oil-immersion reflected light. The specimen has been described in more detail elsewhere (Natland, 1980, 1989). This glass, indeed, has myriads of tiny sulfide globules in boundary layers next to hopper olivine (Fig. 13a and b) and faceted olivine phenocrysts (Fig. 13c and d). Most of the sulfides in 441

22 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 these photomicrographs show up because of internal reflection, and the grains are so tiny that they produce a noticeable asterism at this high magnification. Nevertheless, they are clearly much more reflective than either the olivine or the quench clinopyroxene dendrites in Fig. 13a and in transmitted light they appear as tiny black specks. The sulfide in Fig. 13c is in clear glass between an olivine phenocryst and a spinel grain contained within a skeletal embayment in the olivine. This confirms that boundary-layer segregation of sulfide can occur at the high undercooling characteristic of pillow margins near growing olivine, and thus need not result from closed-system post-entrapment crystallization of host olivine, although it does not prove that it occurred in sulfide-saturated basalts from either Juan Fernandez or Kilauea. Heightened undercooling resulting from bubble formation also explains why many melt inclusions retain glass, or have clinopyroxenes with elaborate dendritic morphologies, even in lavas with a holocrystalline groundmass. Such morphologies are at odds with Roedder's (1965) inference that bubbles within melt inclusions were not there originally, but that they nucleated and grew as the inclusions slowly crystallized. Instead, the bubbles, like the glass and its elaborate silicate minerals, were quenched in place. Occurrence in a thin section of a population of olivine phenocrysts with variable proportions of inclusions is a reflection of uneven distribution of volatiles exsolving from a melt in a magma body at large (e.g. Kennedy, 1955; see also Farley & Craig, 1994, for discussion pertaining to Ar concentrations), and from processes acting at crystallization interfaces. Indeed, crystal surfaces of olivine may not have been the exclusive sites of bubble nucleation. Bubbles may instead have nucleated in the melt, perhaps in response to magma stirring or mixing, and simply attached to olivine crystals as they streamed past. This would have produced uneven bubble distribution on the olivine, and variable concentrations of inclusions in different olivine crystals of the same composition, now present in a single lava sample. Large bubbles would not have remained attached, their buoyancy being too great for the effects of surface tension at the bubble±crystal interface to overcome. Large discontinuities in mineral surfaces, such as intersections of crystal faces, would probably have aggregated larger bubbles from smaller ones moving across the crystal surfaces, and been principal loci of escape of buoyant bubbles (Fig. 12a). The attached bubbles might have been sufficient to overcome the density contrast between olivine and liquid, and thus to retard or eliminate the tendency for olivine to sink. Attached (as opposed to directly nucleated) bubbles would also have physically penetrated any boundary-layer melt, perturbing not only the growth of the crystal, but the chemistry of the boundary layer as well. Inclusions along annealed secondary curvilinear fracture surfaces contain the same phases as primary multiphase inclusions, although particular phasesð spinel, sulfide, and volatilesðtend to be concentrated along different portions of them. As mentioned earlier, this suggests a mechanism of physical separation of these phases during flow along the microfractures. Sulfides and volatiles appear to have been exceptionally mobile and easy to separate from silicate melt, the former in the form of an immiscible liquid, the latter as either a two-phase or a supercritical fluid. The unusual concentrations of spinel, sulfide, and volatiles (in the form of empty cavities) along these healed microfractures point to incorporation of the same boundary layer material that was entrapped during normal crystal growth by larger multiphase inclusions. The fractures formed at the same time that vesiculation introduced local perturbations in the properties of the boundary layer melts. Differential stresses on the mineral surface resulting from contact with melt at different undercooling, or from contrasting effects of surface tension at bubble±melt contacts, may have produced some of the fractures. However, most of the fractures are probably simple contraction features resulting from sudden cooling of the growing olivine phenocrysts during magma mixing. Additional indications of mixing and some of the conditions of vesiculation are provided by mineral compositions, considered next. MINERAL COMPOSITIONS Tables 1 and 2 give compositions of olivine and spinel analyzed in eight Juan Fernandez olivine tholeiites. Olivine cores and rims together range in composition from Fo 88 to Fo 58, and have an asymmetric distribution about a single prominent mode at Fo 82 (Fig. 14a). All values are sufficiently iron rich for the olivine to be phenocrystic. Spinel ranges from magnesiochromite to titanian chromite in composition. It is strongly bimodal in composition (Fig. 14b), and its Mg-number correlates with the Fo content of host olivine (Fig. 14c). The bimodality in spinel defines two sample groups intergrown with olivine, Group 1 with more, and Group 2 with less, magnesian olivine and spinel (Fig. 14a±c). Some even more iron-rich spinel is not intergrown with olivine. Within each rock, there is a range of mineral compositions, but no sample has both spinel and olivine bridging between Groups 1 and 2. Most spinel is more chromian, but less magnesian, and all is more titanian, than spinel in abyssal tholeiites (Fig. 14d and e), in abyssal and alpine peridotites (e.g. 442

23 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS Table 1: Electron microprobe analyses of olivine and their estimated equilibrium liquid compositions and temperatures Sample Run-analysis* Typey SiO 2 FeO MnO MgO CaO NiO Total Fo Mg L T ( C) KIL z Alexander Selkirk (Mas Afuera) MF C PH MF C PH MF C PH MF C PH MF C PH MF C PH 6R MF C PH 6C MF C GM MF C R MF C R MF C GM MF C PH MF C PH MF C PH MF C PH MF C PH MF C PH MF C PH MF C SK 1C MF C SK 1R MF C PH MF C PH MF C PH MF PH MF D MF w pl MF PH MF SK MF PH 3C MF PH 3R MF GM(eu) MF PH MF PH MF PH MF GM MF PH MF PH MF PH MF PH MF GM MF GM MF GM MF GM MF GM MF S M1 G

24 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 Table 1: Continued Sample Run-analysis* Typey SiO 2 FeO MnO MgO CaO NiO Total Fo Mg L T ( C) KIL z MF S M1 G MF S M2 G MF S M2 G MF S M2 G MF S M3 G MF S M3 G MF S M4 G MF S M4 G MF S PH 1C MF S PH 1R MF S MF S GM MF S GM MF S PH MF S PH MF S GM MF S PH 5C MF S PH MF S PH MF S H MF S GM MF S PH MF S GM MF S GM 4R Robinson Crusoe (Mas a Tierra) PIN GM PIN GM PIN GM PIN MPH PIN MPH PIN MPH PIN MPH PIN MPH PIN GM 2R PIN MPH PIN MPH PIN PH PIN PH PIN PH PIN PH PIN GM PIN GM PIN GM *Run-analysis: electron microprobe runs on different dates and analysis number on that date; run given to distinguish analyses with the same number. ytype: PH, phenocryst; MPH, microphenocryst; GM, groundmass; C, core; R, rim; M, grain mount; G, grain; Ð, spot; E, euhedral; D, dendrite; SK, skeletal grain. Number after PH, MPH, GM or M identifies different grains. zt ( C) KIL is crystallization temperature calculated using modified Kilauea geothermometer as described in the text. 444

25 Table 2: Electron microprobe analyses of spinel, compositional parameters, and estimated equilibrium liquid compositions and temperatures compared with those of adjacent intergrown olivine and its Fo content Sample Run-analysis* TiO 2 Al 2 O 3 Cr 2 O 3 Fe 2 O 3 y FeOy FeO T y MnO MgO NiO Total Mg-no.z Cr-no.z Fe-no.z Fo Mg L (ol)x T ( C) Ol x Mg L (Sp){ T ( C) Sp { 445 Alexander Selkirk (Mas Afuera) MF C PH MF C MF C PH MF C PH MF MF MF C PH MF C PH MF C PH MF C PH MF C PH MF C PH MF C PH MF C PH MF C GM MF C PH MF C MF C SK1R MF C GM MF C PH MF C PH MF C PH MF C SK MF PH MF PH MF PH MF MF PH MF S M1G MF S M1G MF S M2G MF S M4G NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS

26 Table 2: Continued Sample Run-analysis* TiO 2 Al 2 O 3 Cr 2 O 3 Fe 2 O 3 y FeOy FeO T y MnO MgO NiO Total Mg-no.z Cr-no.z Fe-no.z Fo Mg L (ol)x T ( C) Ol x Mg L (Sp){ T ( C) Sp { 446 MF S PH1C MF S PH MF S PH MF S PH MF S PH MF S PH Robinson Crusoe (Mas a Tierra) PF PF PF PF PIN MPH PIN MPH PIN MPH PIN MPH PIN MPH PIN MPH *Run-analysis: electron microprobe runs on different dates and analysis number on that date; run given to distinguish analyses with the same number. Remaining notation identifies intergrown olivine phenocryst and microphenocryst as in the second footnote to Table 1. yfe 2 O 3 and FeO are calculated from total iron as FeO T by stoichiometry. zmg-number ˆ Mg/(Mg Fe 2 ); Cr-number ˆ Cr/(Cr Al); Fe-number ˆ Fe 3 /(Fe 3 Cr Al) from structural formulae. xmg L (ol) is calculated using K D values as discussed in the text. T ( C) Ol is calculated using Mg L and the modified Kilauea geothermometer of Helz & Thornber (1987) as discussed in the text. {Mg L (Sp) and T ( C) Sp are calculated using the procedure of Allan et al. (1988, 1989). JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003

27 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS Dick & Bullen, 1984), and in residual Pacific mantle xenoliths (e.g. Dieu, 1995). This and the phenocrystic olivine is sufficient to support the initial thesis that mantle helium was trapped in phenocrysts in the cavities described, rather than in minerals extracted from mantle wall rocks. On the other hand, the Juan Fernandez spinel is similar to spinel associated with olivine phenocrysts in Puna Ridge tholeiites (Clague et al., (1995), although the most magnesian is not quite as chromian. In both mineral suites, as Mg-number decreases, the Cr 2 O 3 content decreases and both TiO 2 and Fe 2 O 3 increase even as the Cr-number increases and may decrease within individual samples (Fig. 14d). These internal trends demonstrate that some samples have two or more populations of spinel within each, an indication of magma mixing. The populations of olivine and spinel within individual samples are not directly related to the bulk compositions of the rocks. Two samples apiece containing Group 1 and Group 2 phenocrysts, identified in Fig. 14c by being enclosed in boxes in the lists of sample names beneath the data clusters, are among the group circled in Fig. 3e with the arrow labeled `mixing' passing through it. Olivine and spinel characteristic of different stages of differentiation thus mixed with even more strongly differentiated magma. The strongest contrast between a primitive (Group 1) phenocryst assemblage and both the composition and proportion of the differentiated magma into which the phenocrysts were mixed is sample PF-05 (* in Fig. 14c; identified in Fig. 3e). The compositions of olivine and spinel provide important information on the conditions in which volatiles were trapped within them. The compositions of either mineral, for example, can be used to estimate the Mg L [ ˆ Mg/(Mg Fe 2 )] of the liquids from which they crystallized (Roeder & Emslie, 1970; Allan et al., 1988; Allan, 1992). Figure 14f compares the results of such calculations for samples from Juan Fernandez and Kilauea±Puna Ridge using all samples for which coexisting olivine and spinel compositions have been determined. At both places, the spinel that crystallized from the most iron-rich liquids is titanian chromite, with 47% TiO 2. Ideally, the two calculations should provide nearly the same estimated Mg L, and indeed a regression through the data for Kilauea and Puna Ridges falls very nearly on a 1:1 trend. This is not the case for Juan Fernandez: there is a divergence with decreasing Mg L, such that spinel appears to have crystallized from more iron-rich melts than immediately intergrown olivine as differentiation proceeded. In detail, a number of samples from Kilauea and Puna Ridge show the same divergence, but these are not sufficient to shift the regression very much. To understand this, some of the assumptions built into the calculations need to be considered. In the first place, to determine Mg L, the value of the Fe±Mg partition coefficient for olivine, K D, must be assumed, even for spinel, as the calculated exchange free energy for spinel, G EX, depends on it (Sack, 1982; Sack & Ghiorsio, 1991). It is often taken to be 030, as originally determined by Roeder & Emslie (1970), but it can also be calculated from glass compositions using an algorithm of Carmichael & Ghiorsio (1990). For Kilauea±Puna Ridge glasses with Mg L 563, this value indeed is appropriate. I use the same assumption for all Juan Fernandez olivine except the three most magnesian in Fig. 14f. Above Mg L ˆ 63, the algorithm of Sack & Ghiorsio (1991) indicates that K D should increase to 034 among the most magnesian glasses from Kilauea and Puna Ridge (see Wilkinson & Hensel, 1988), and I have taken this into account for those glasses. For the three samples with most magnesian olivine from Juan Fernandez, for which no glass compositions exist, I interpolated using a linear regression between olivine compositions and Puna Ridge glasses more magnesian than Mg L ˆ 63. Regardless of complications for primitive compositions, however, Juan Fernandez samples diverge from those of Kilauea± Puna Ridge most strongly below Mg L 5 60, and among these, K D ˆ 030 is appropriate for both places. A second assumption concerns the Mg L of the basaltic liquid from which the minerals crystallized. It is related to the oxidation state of that liquid, and will vary depending on the ratio Fe 2 /(Fe 2 Fe 3 ) of the melt. For glasses analyzed by electron microprobe, this ratio has to be assumed. The data plotted in Fig. 14f are based on Fe 2 /(Fe 2 Fe 3 ) ˆ 086, which is a common enough assumption, for example, in calculating CIPW norms. This ratio, however, is somewhat higher than has been determined for many abyssal tholeiites, for which an average value of at least 090 is recommended (Christie et al., 1987). It is also not likely to be appropriate across the board for either Kilauea±Puna Ridge (Carmichael & Ghiorsio, 1990) or Juan Fernandez tholeiites. The assumption of a constant Fe 2 /(Fe 2 Fe 3 ), however, when applied to estimation of Mg L using coexisting olivine and spinel, can indicate differences in oxidation state of host liquids. Higher Fe 2 /(Fe 2 Fe 3 ) than 086, for example, will shift Mg L calculated from olivine to lower values, more in accord with those calculated from spinel. Indeed, Juan Fernandez spinel compositions have systematically lower Fe-number [ ˆ Fe 3 / [(Fe 3 Cr Al)] at any given Mg-number than spinel from Kilauea±Puna Ridge. On this basis, the Juan Fernandez tholeiites were not as oxidized during crystallization of their phenocrysts, especially those with more differentiated compositions. 447

28 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH

29 NATLAND HELIUM CAPTURE BY OLIVINE PHENOCRYSTS Departures from an ideal 1:1 correspondence between Mg L values calculated for olivine and spinel might be a consequence of partial re-equilibration of spinel with melt before entrainment in current host lavas. Scowen et al. (1991) showed that an initial population of spinel at Kilauea Iki's lava lake became richer in Fe 2, Fe 3, and Ti, and poorer in Mg, Al, and Cr, than the liquidus chromite as the lake cooled over 22 years. The transformation resulted both from cooling and a shift in f(o 2 ) as a result of interaction with the atmosphere and rainwater percolating down through cracks in the crust in the lava lake. With a few exceptions, Puna Ridge spinel, which Clague et al. (1995) viewed to have become entrained together with associated olivine in more differentiated, iron-rich magmas, did not experience similar changes in composition. However, although compositions of coexisting olivine and spinel provided by Scowen et al. (1991) show a shift toward re-equilibration with more iron-rich liquids through time (Fig. 14f ), there was not much of a shift in the relative oxidation state (with respect to, say, the nickel±nickel oxide buffer) that could have induced a marked trend away from the Kilauea±Puna Ridge regression and the ideal 1:1 estimate for melt Mg L values. This explanation therefore appears to be inadequate to explain such a shift among the Juan Fernandez picrites, leaving change in oxidation state as the best explanation for these relationships. To explain a 5 mol % difference in Mg L calculated from olivine and spinel for a differentiated Kilauea tholeiite, about half the iron usually present as Fe 2 O 3 should be converted to FeO, producing a 35±4 mol % increase in the Fo content of olivine, and a reduction of about 15 log units in f(o 2 ). The matter is complicated because usually the olivine and spinel did not crystallize from a melt having the composition of the host basalt, and because the oxidation state may have continued to vary after the minerals crystallized. A mechanism for reduction in f(o 2 ) was outlined by Anderson & Wright (1972), who showed that the composition of coexisting ilmenite and magnetite in the groundmasses of some Kilauea tholeiites is a consequence of reduction in f(o 2 ) after `effervescence' of the magma during transport and flow as lava. Of particular importance is the loss of sulfur by the reaction FeS 3Fe 2 O 3 ˆ 7FeO SO 2. Loss of 500 ppm of S to degassing of SO 2 results in reduction of 072% Fe 2 O 3 to FeO. None of the Juan Fernandez samples with Group 2 assemblages of olivine and spinel contains sulfide within inclusions either in olivine or in the groundmass, and it is present in inclusions only in two samples with Group 1 phenocrysts. Degassing is therefore implicated as a potential cause for the systematic divergence during magmatic differentiation of Mg L of Juan Fernandez tholeiitic liquids as estimated using olivine and spinel. If this is the case, then the vesiculation took place before or at least during crystallization of olivine and spinel phenocrysts, and therefore before final mixing of porphyritic magmas with the more extreme differentiates that resulted in the present bulk compositions of the rocks. Loss of S by degassing at Kilauea occurs at pressures mainly below 100 bars (Moore, 1965; Killingley & Muenow, 1975; Kyser & O'Neil, 1984), thus much of the history of crystallization and mixing involving olivine and spinel among Juan Fernandez tholeiites, and consequently of incorporation of volatiles including helium into phenocrysts, took place in the shallow probably subaerial crust. There, magmas may well have interacted with altered rock or groundwater before crystallization of olivine. This might explain the high proportion of atmospheric argon in volatiles extracted from olivine in Juan Fernandez samples (Farley & Craig, 1994). Olivine compositions can also be used to estimate crystallization temperatures and, of more interest, differences in temperature of magmas involved in mixing. For Juan Fernandez tholeiites, I use a modification of the geothermometer of Helz & Thornber (1987) for Kilauea Iki's lava lake, which is based on temperature measurements as the lake cooled that varied linearly with the MgO content of glasses quenched in cores. Clague et al. (1995) applied this to Puna Ridge glasses. Their temperatures can consequently be related to glass Mg L, and this in turn to the values of Mg L calculated for olivine and spinel. For Puna Ridge glasses, the relationship between Mg L and temperature breaks down into two linear trends with different slopes Fig. 14 (opposite). Summary of compositions of olivine and spinel in Juan Fernandez olivine tholeiites. Sample groups are indicated by number, and different shading or symbol color. (a) Histogram of olivine compositions with a principal peak at Fo 82. (b) Histogram of spinel Mg-numbers [ ˆ Mg/[Mg Fe 2 )]. The distribution is bimodal, defining sample groups 1 and 2, comprising the ranges of spinel enclosed in olivine (black and dark gray, respectively), and (light gray) spinel with lower Mg-number in the groundmasses of the rocks. (c) Spinel Mgnumber vs Fo contents of immediately adjacent intergrown olivine. This figure provides the basis for the groupings in (a) and (b). (d) Spinel Mg-number vs Cr-number. Filled circles, Juan Fernandez tholeiites, shaded as in (a) and (b); open diamonds and triangles, Kilauea±Puna Ridge tholeiites and xenocrysts, respectively (D. Clague, unpublished data). The shaded field for spinel in Pacific MORB and seamounts is based on data of Natland et al. (1983), Natland (1989) and Natland (unpublished), plus Allan et al. (1988, 1989). (e) Mg-number vs Ti for spinel. Symbols and eastern Pacific array are as in (d), except that a light gray shaded field represents data for Kilauea and Puna Ridge. (f) Mg L calculated from olivine based on the procedure of Roeder & Emslie (1970) vs Mg L calculated from coexisting spinel, based on the procedure of Allan et al. (1988). (See text for details.) Symbols are as in (d), with addition of bold open squares representing re-equilibrated spinel from Kilauea Iki lava lake (Scowen et al. (1991). Trends JF and KP are respective linear regressions through the data for Juan Fernandez and Kilauea±Puna Ridge. 449

30 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 3 MARCH 2003 Fig. 15. The Kilauea geothermometer of Helz & Thornber (1987) modified to use with glass Mg L by calibrating to temperatures calculated for glasses from Puna Ridge by Clague et al. (1995). Two linear regressions shown in the figure can be used, corresponding to olivinecontrolled liquids with Mg L 4 60, and Mg L 60, where Mg L is calculated using Fe 2 /(Fe 2 Fe 3 ) ˆ 086. (Fig. 15), the less steep one being for glasses with Mg L 562, corresponding to the transition between olivinecontrolled and multiply saturated glasses. Other geothermometers, such as that of Montierth et al. (1995) for Mauna Loa, give different temperatures by about 20±40 but the same differences in temperatures of contrasting magmas involved in mixing. Ranges and averages of crystallization temperatures of olivine phenocrysts, olivine rims and groundmass, and spinel, in Juan Fernandez tholeiites are presented in Fig. 16. Samples are listed downward in order of decreasing average olivine phenocryst temperature (or Fo content). The bar plot to the right shows the difference in temperatures calculated for average phenocryst and average rim or groundmass olivine, and the graph at the bottom gives both Fo content and Mg L as functions of temperature, with the breaks in slope corresponding to the transition between olivine-controlled and cotectic crystallization of olivine, clinopyroxene, and plagioclase. This took place at 1150 C. The ranges of temperatures calculated for the spinel populations shown in Fig. 16 are similar to the ranges found for rim or groundmass olivine, not large olivine phenocrysts. However, as discussed above, this apparently is the result of too high an estimate for the temperatures of olivine crystallization on the assumption of constant Fe 2 /(Fe 2 Fe 3 ). If this ratio increased by degassing of SO 2, then temperatures of olivine crystallization were lower, and those of spinel and olivine phenocrysts would correspond. The actual temperatures of crystallization of olivine phenocrysts in sample Fig. 16. Calculated average temperatures and temperature ranges of crystallization for olivine and spinel populations in Juan Fernandez olivine tholeiites and alkalic olivine basalts, arranged in order of decreasing temperatures. Dashed lines give ranges to individual extreme estimated temperatures. The bar graph to the right gives differences between average temperatures calculated for olivine phenocrysts and olivine rims plus groundmass crystals. Actual olivine compositions and calculated equilibrium Mg L values can be interpolated from the curves at the bottom. PIN-12, for example, were probably less by 50, and thus closer to 1200±1150 C. Groundmass olivine crystallized below 1150 C. Temperature contrasts between average phenocrystic and rim or groundmass olivine range between 20 and 60 C. These represent the differences in average temperature of crystallization of the two populations of olivine in each sample, and record the fundamental information about mixing of magmas. The cooler magmas involved in mixing in each case were at or below 1150 C. In four cases based on spinel the hotter magmas were just above this temperature, and in three cases they were at or below it. These estimates support the earlier inference from bulk compositions that mixing took place between primitive, olivine-charged magma and cooler, significantly more differentiated multiply saturated magma. However, apart from very rare individual skeletal plagioclase phenocrysts, there is no indication in any of the samples that silicate minerals other than olivine were crystallizing at the 450