(Received 26 May 1988; revised typescript accepted 10 March 1989)

Transcription

1 /90 $3. Reaction Between Ultramafic Rock and Fractionating Basaltic Magma II. Experimental Investigation of Reaction Between Olivine Tholeiite and Harzburgite at Cand5kb by PETER B. KELEMEN 1 *, DAVID B. JOYCE 2, JAMES D. WEBSTER 2!, AND JOHN R. HOLLOWAY 3 1 Geological Sciences AJ-20, University of Washington, Seattle, Washington ^Department of Geology 3 Department of Chemistry, Arizona State University, Tempe, Arizona (Received 26 May 1988; revised typescript accepted 10 March 1989) ABSTRACT We present results of experiments on mixtures of olivine tholeiite and mantle harzburgite, at 5 kb and C, under conditions of controlled hydrogen fugacity. The basalt end-member was Kilauea 1921 olivine tholeiite + 3 wt.% H 2 O, and the harzburgite end-member was a mixture of olivine and orthopyroxene mineral separates made from a mantle-derived lherzolite xenolith. The experiments on mixtures of basalt and harzburgite difl not reach equilibrium in runs ranging from 12 to 2 h duration. Relatively large concentration gradients persisted in both liquid and solid phases in mixed samples, whereas 'control' samples containing only basalt were reasonably homogeneous and were probably close to equilibrium. Compositions of solid phases produced, measured by electron microprobe, show a regular increase in Mg/(Mg + Fe) with increasing proportion of harzburgite at constant temperature, but olivine and clinopyroxene in mixed samples were not in Fe-Mg exchange equilibrium. Modes measured for each sample show that the fraction of liquid relative to the amount of basalt in the sample was constant at constant temperature, and independent of bulk composition: reaction between 1921 basalt and harzburgite does not change the mass of liquid in the system. Average experimental liquid compositions for each sample were obtained by mass balance. Using K d s defined by the 'control' sample for each temperature, and mass balance constraints, phase assemblages (solid- and liquidphase compositions and proportions) were calculated for all mixtures. Whether samples included harzburgite or not, all average experimental liquid compositions, and all predicted liquid compositions, for samples run at 1050 C, are high-alumina basalts by the definition of Kuno (1960). By the criteria of Irvine & Baragar (1971), all but two average experimental liquid compositions in basalt-harzburgite mixtures, and all predicted liquid compositions in basalt-harzburgite mixtures, are calc-alkaline basalts and basaltic andesites, whereas liquids in samples containing only basalt are tholeiitic basalts. Combined crystallization and reaction with harzburgite in the upper mantle will produce calc-alkaline derivative liquids from an olivine tholeiite liquid under conditions of temperature, pressure, water and oxygen fugacity, and initial bulk composition which would produce a tholeiitic liquid line of descent by crystallization in a closed system. Present address: Woods Hole Oceanographic Institution, Woods Hole, Massachusetts t Present address: Grant Institute of Geology, University of Edinburgh, West Mains Road, Edinburgh EH9 3JW, UK [Journal of Petrology, Vol 31, Pan 1, pp , 1990] Q Oxford Univtnily PrtM 1990

2 1 PETER B. KELEMEN ET AL. INTRODUCTION The experiments described in this paper were designed to investigate basalt-peridotite equilibria at temperatures below the solidus of the peridotite end-member and below the liquidus of the basalt. They provide data with direct bearing on the process of combined crystal fractionation of olivine tholeiite and assimilation in the upper mantle and lower crust. Where ascent rates are slow relative to reaction rates, such liquid-solid interaction may be as important as the composition of the parental melt in determining the nature of differentiates, from either 'mantle-derived' or 'slab-derived' primary magmas, which reach the surface or are intruded into the crust. The compositional effects of reaction between basaltic melts and mantle wall rock are dependent upon the stoichiometry of the solid-liquid reactions involved. Some aspects of the reaction stoichiometry can be predicted by application of well-known phase relations in simple systems (Bowen, 1922; Kelemen, 1986, 1990). However, information available at present, including the solution models for multi-component silicate liquids developed in recent years (e.g., review in Ghiorso, 1987), cannot be used to predict accurately the effects of pressure above ~3 kb, or the exact effects of H 2 O contents, on phase relations in partially molten natural silicate systems. Limitations in the solution models reflect the paucity of data on the equilibrium compositions of coexisting silicate liquids and solids at moderate to high pressure and in H 2 O-bearing systems. To begin evaluating some of the complexities of basalt-peridotite interaction in natural systems, and to test general conclusions based on simple and complex system thermodynamic modelling, experiments were conducted on mixtures of olivine tholeiite (Kilauea 1921 basalt) plus harzburgite (natural olivine and orthopyroxene separates). Mixtures with several different proportions of basalt and harzburgite were prepared and run simultaneously at the same temperature and pressure. The mixtures were allowed to react for 12-2 h and were then quenched. The resulting phase compositions and proportions provide a wealth of information on the stoichiometry of basalt-peridotite reaction, and the compositions of the derivative liquids. PREVIOUS STUDIES OF REACTION BETWEEN PERIDOTITE AND NATURAL SILICATE LIQUIDS Recent experimental investigations of reaction between liquid-peridotite pairs which were initially out of equilibrium have been of three types: (1) Equilibration of basalt and mantle harzburgite and lherzolite at temperatures above the solidus of the lherzolite, to determine the composition of partial melts formed in the mantle (Stolper, 1980; Takahashi & Kushiro, 1983; Fujii & Scarfe, 1985; Falloon & Green, 1987). Large masses of lherzolite, relative to basalt, were used to constrain resultant liquids to be saturated in olivine + two pyroxenes + spinel. The main relevance such studies have to the experiments discussed here is that solid-liquid equilibrium was said to have been closely approached in a matter of 2 h (Stolper, 1980) to 60 h (Fujii & Scarfe, 1985). (2) Reaction between highly siliceous liquid and peridotite at temperatures well below the peridotite solidus, to investigate the nature of hybrid solids and liquids formed by siliceous partial melts formed in a Benioff Zone, rising into the mantle wedge at 30 kb. These experiments have been performed by Wyllie and co-workers (granite-peridotite: Sekine & Wyllie, 1982, 1983; Wyllie & Sekine, 1982; tonalite-peridotite: Carroll & WyUie, 1989; Johnston & Wyllie, 1989). In these experiments, high-pressure solid-liquid reactions formed orthopyroxene and liquid (± garnet ±clinopyroxene +quartz).

3 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 101 (3) Fisk (1986) performed 1 atm experiments in which olivine tholeiite was placed in harzburgite crucibles and held near the calculated liquidus temperature of the basalt (12 C) at controlled oxygen fugacity for 2 h, to investigate the effect of reaction between mantle-derived melts and harzburgite in the oceanic lithosphere. Glass adjacent to the peridotite was found to be enriched in silica and alkalis, and depleted in alumina, iron, magnesium, and calcium, relative to glass 'far' from the melt-harzburgite interface. Textures indicative of incongruent melting of orthopyroxene were observed in harzburgite adjacent to glass. Olivine in glass within the 120 /an wide 'reaction zone' was slightly more magnesian (Fo 88 ) than olivine in glass outside the 'reaction zone' (Fo 87 ), and less magnesian than olivine in the crucible (Fo 90 _ 91 ). To produce liquids like those in the 'reaction zone', Fisk calculated that 1 part orthopyroxene + 2 parts liquid (by weight) reacted to form 21 olivine and 0-9 liquid [the ratio of mass assimilated to mass crystallized, (JVf,/Af c ) = O48]. This reaction stoichiometry is not in accord with that predicted for stepwise equilibrium between olivine-saturated magma and orthopyroxene-bearing peridotite (MJM c x 1) or for stepwise equilibrium dissolution of orthopyroxene alone in olivine-saturated magma (MJM C > 1) at constant temperature (Kelemen, 1990). EXPERIMENTAL PROCEDURE Dry starting materials A mixture of 85 wt% olivine +15wt.% orthopyroxene, from hand-picked mineral separates originally prepared and used by Dr. E. Stolper (1980), was used as the harzburgite end-member in this study. The separates were made from a spinel lherzolite nodule (KH77-6) sampled at Kilbourne Hole, Arizona, by Dr. A. J. Irving. They were ground, by Stolper, in an agate mortar for several hours and then passed through a 325 mesh (< 44 /«n) sieve. The orthopyroxene separate contained minor spinel. The basalt used was an olivine tholeiite from the 1921 flow of Kilauea volcano on the island of Hawaii. This basalt (hereafter referred to as 1921 basalt) was collected from the same locality as the material used by Yoder & Tilley (1962). The split used in these experiments is from the same lot as used by Holloway & Burnham (1972) and Helz (1973,1976). The phase relations of compositionally similar basalt from the same locality have also been experimentally investigated by Fudali (1965), Hill & Roeder (1974), Allen et al. (1975) and Allen & Boettcher (1978). The basalt contains phenocrysts of olivine (Fo 84 ) in a groundmass of calcic pyroxene, plagioclase, oxide minerals, and glass. It was reground before these experiments in a SPEX mill, using a tungsten carbide lined container and disk, after which all the material was passed through a 325 mesh (<44 /im) sieve. Both basalt and harzburgite were stored in a drying oven at 110 C. The chemical composition of starting materials are given in Table 1. H 2 O content Approximately 3 wt.% water was added to the basalt. This was done for three reasons: (1) Addition of a few wt.% H 2 O lowers the liquidus and solidus of 1921 basalt such that the basalt is > 60% liquid at temperatures attainable at 5 kb in internally heated gas pressure vessels. (2) Phase relations in many zoned, calc-alkaline, ultramafic to felsic plutonic complexes, of the type described by Kelemen & Ghiorso (1986), indicate that parental basaltic liquids contained several wt.% H 2 O. Evidence for this conclusion includes the commonly observed

4 102 PETER B. KELEMEN ET AL. TABLE 1 Compositions of starting materials used in basalt-harzburgite mixtures (wt.%) Harzburgite Stolper (1980) Kilauea 1921 basalt Holioway <t Burnham (1972) Olivine (85%) Opx (14%) Spinel (/%) Total (1%) 1921H Normalized + 3wt.%H 2 O SiO 2 TiOj A1 2 O 3 Cr 2 O 3 FeO* MnO MgO NiO CaO Na 2 O K 2 O H 2 O Total O07 OOO O OOO OOO OOO 10O OO O6 1O90 2O Mixtures of 1921 (wt.% basalt and harzburgite harzburgite) SiO 2 TiO 2 A1 2 O 3 Cr 2 O 3 FeO* MnO MgO NiO CaO Na 2 O K 2 O H 2 O OOO OOO O O O Oil crystallization sequence olivine-clinopyroxene (± orthopyroxene)-magnetite-hornblende-plagioclase, which is characteristic of the crystallization of H 2 O-bearing olivine tholeiite, as well as more silicic melts, from 3 to 8 kb (Yoder & Tilley, 1962; Eggler, 1972; Eggler & Burnham, 1972; Holioway & Burnham, 1972; Helz, 1973, 1976). This crystallization sequence is also present in xenolith suites from the Aleutians (Conrad et al., 1983; Conrad & Kay, 1984). Furthermore, Anderson (1979, 1982) and Harris & Anderson (1984) reported H 2 O contents of melt inclusions from minerals in 'calc-alkaline' basalts in the range 1^4 wt.%. (3) The presence of H 2 O dissolved in the melt, combined with a fixed partial pressure of H 2 in the gas pressure medium, was used to control oxygen fugacity during the experimental runs (see discussion of oxygen fugacity, below). Water was added to capsules of Pd^-Aggo alloy (1150 and 11 C runs) and Pd 3O -Ag 7O alloy (1075 and 1050 C runs) with a micropipette. Repeated tests demonstrated that (1) evaporation during the loading process did not appreciably decrease the mass of H 2 O in

5 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 103 the capsules and (2) the weight of H 2 O added was reproducible to ±0-01 g. To each sample, 3 wt.% H 2 O relative to basalt, and not to the combined mass of basalt + harzburgite, was added. For a typical mixture the mass of H 2 O was known with a precision of ±5% relative. In a pilot study related to the present experiments, we found that 1921 basalt at 1050 C and 5 kb with 1 and 2 wt.% H 2 O is saturated in plagioclase (±orthopyroxene), as well as olivine, calcic pyroxene, and ilmenite, whereas no plagioclase crystallized in the samples with 3 wt.% H 2 O, for the same run conditions. The bulk composition with 2 wt% H 2 O had about 58 wt.% liquid (34 wt.% H 2 O in the liquid), whereas the composition with 3% H 2 O had 68 wt.% liquid (4-4 wt.% H 2 O in the liquid). Thus a difference of ~1 wt% H 2 O (20-30% relative) in this compositional range makes a very significant difference in phase proportions and compositions. However, comparison of the results of this study with those of Holloway & Burnham (1972; ca. 71 wt.% liquid with ~6 wt.% H 2 O in the melt at 1050 C and 5 kb) at approximately the same temperature and pressure, suggests that increasing H 2 O content in excess of 4 wt.% in the melt leads to only a slight decrease in the proportion of solid phases. The internal consistency of the results of this study, for samples with 3 wt.% H 2 O in the bulk composition, suggests that the potential 5% relative variation in water content due to limitations in measurement did not significantly affect the nature of the run products. The water content in the present experimental systems could have varied appreciably, as hydrogen could combine with oxygen in the capsule to form water if 'free oxygen' were to evolve during the run. Even if all the Fe 2 O 3 in 1921 basalt were converted to FeO, and all the oxygen thus evolved combined with hydrogen to form H 2 O, this would not measurably change the H 2 O concentration in each sample. However, the effect of reduction of FeO/Fe during iron loss to the capsule could have an order of magnitude larger effect on the total H 2 O content of the sample. This problem is discussed further in the subsequent section on iron loss. Sample geometry Basalt and harzburgite powders were mixed in two different geometric configurations (Fig. 1). The first geometric arrangement used was a single 'sandwich', in which upper and APPROX ' 10mm ' Pd30-^g70(1050 C) X basalt harzburgite SArOWCH" basalt > 2.5 mm Pd40-Ag60(1150 C) &Pd30^Ag70(1075 C) harz harz harz harz harz S > I! 5 mm ZEBRA" FIG. 1. Geometry of samples with 1921 basalt-harzburgite mixtures. Rock powder was loaded into Pd-Ag tubing, with welded ends, in parallel bands. The bands of basalt and harzburgite powder maintained their geometry to a surprising degree during runs, forming texturally distinct domains with a sharp boundary. Two configurations, the 'sandwich' and the 'zebra', were used.

6 104 PETER B. KELEMEN ET AL. lower layers of basalt enclosed a single layer of harzburgite (samples ). This is similar to the peridotite-basalt-peridotite 'sandwiches' used by Takahashi & Kushiro (1983) and Fujii & Scarfe (1985). However, it did not work as well, primarily because the lower temperatures used in the present experiments led to slower diffusion of species in the melt phase, and thus a failure to achieve homogeneous equilibrium. A second, 'zebra' configuration was adopted, with several alternating layers of basalt and harzburgite, in an effort to approach equilibrium more closely. Oxygen fugacity The presence of a known quantity of H 2 O dissolved in the melt, combined with a fixed partial pressure of hydrogen in the gas pressure medium, was used to control oxygen fugacity during the experimental runs. Had the amount of melt in each charge (per weight 1921 basalt) varied significantly between different samples run at the same temperature and pressure, then the oxygen fugacity, although fixed for each sample, would have varied between samples. However, as will be seen, this led to differences in oxygen fugacity between samples of ±0-2 log units relative to the nickel-nickel oxide (NNO) oxygen buffer. Hydrogen partial pressure in the system was controlled through the use of an external hydrogen reservoir, with fixed total pressure, communicating with the argon pressure medium through a Pd-Ag alloy membrane permeable to hydrogen but not to argon, oxygen, or H 2 O (Shaw, 1963). At fixed temperature, pressure, and hydrogen fugacity, the oxygen fugacity in the sample varied as a function of water fugacity, which in turn was controlled by fixed bulk composition and the ratio of (anhydrous) solid phases to liquid in each sample. As shown by Whitney (1972) and Webster et al. (1987), the fugacity of oxygen, JOi = / Oi ' a H2O2 where a H20 is the activity of H 2 O in the liquid, and / Oj is the fugacity of oxygen in a hypothetical pure H 2 O fluid phase at P, T, and fixed fugacity of hydrogen. Thus oxygen fugacity is proportional to the square of the activity of H 2 O in the sample. The activity of H 2 O coexisting with the liquid in each sample at run conditions can be calculated as a function of the mole fraction of H 2 O in the melt. For the liquid compositions determined in this study, activities of H 2 O in the melt were calculated via the methods of Burnham (1979) and Stolper (1982). Results using the two methods were essentially identical. Temperature and pressure Samples were run in internally heated gas pressure vessels in the Department of Chemistry at Arizona State University. The pressure vessels are of the type described by Burnham et al. (1969) and Holloway (1971). Temperatures were measured with two or three chrome-alumel thermocouples. The measured temperatures are thought to be accurate to ±2 C. Longitudinal temperature gradients occurred in some runs; these are reported in Table 2, along with other information on run conditions. Pressure measurements, thought to be accurate to + 10 b at 5 kb, were made using Heise 10 psi Bourdon-tube gauges. Slow leaks in the argon lines occasionally resulted in long-term variations in pressure, which are also reported in Table 2. Runs were quenched by turning off power to the furnace windings, resulting in a temperature drop of at least 225 C in the first minute, and 5 C in the first 2 min.

7 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 105 Sample weight Quenched charges were weighed, and their weight was compared with weight before the run. There are several ways, other than by leaks and diffusion of hydrogen through the capsule walls, in which sample weight can change during a run. Chief among these is that adjacent capsules often anneal together slightly. Afterward, one capsule in such a pair may be heavier and the other lighter. Nevertheless, capsule weights were identical to within ±3 g for all samples. Microprobe analysis Analyses were performed using a Jeol JXA-733 electron microprobe. More than 25 wavelength-dispersive analyses for 11 elements were made on crystalline phases in each sample, at an accelerating voltage of 15 kv and a take-off angle of 40, using a sample current corresponding to 35 na on brass. For a few samples, several hundred such analyses were made on two different thin sections. At analytical conditions, a focused electron beam on benitoite caused fluorescence in a volume with a diameter of ~ 5 /zm and a maximum depth of perhaps hah' that diameter (Mathez, pers. comm., 1982). Despite use of a focused beam, and selection of relatively large crystals, many analyses of olivine and clinopyroxene had to be discarded because of the inclusion of glass and/or quench hornblende in the excitation volume. In addition to quantitative analyses of crystalline phases, automated point counts were performed, using analyses of points selected at random along lines perpendicular to the original basalt-harzburgite layering in each sample. Eleven-, five- and four-element, wavelength-dispersive analyses, with a counting time of 10 s per element per point, and eight-element, energy-dispersive measurements at a sample current corresponding to 10 na on brass, with a counting time of 5 s per point, were used. Attempts to obtain quantitatively meaningful microprobe analyses of the glass phase in run products were largely unsuccessful. Inhomogeneity in the glass phase was extreme, because of compositional gradients formed in response to relatively slow diffusion in the liquid phase during experimental runs and as a result of metastable crystallization of olivine, calcic pyroxene, and hornblende during the quench. Analyses of adjacent areas up to 0-01 mm 2 in large glass 'pools', containing apparently similar volumes of quench hornblende crystals, always produced differing results. Data analysis To use automated point counts, some of the 11- and eight-element, wavelength-dispersive analyses of randomly selected points were examined and each analysis was assigned to one of six groups: glass, hornblende, olivine, clinopyroxene, orthopyroxene, and ilmenite. These hand-classified analyses were used to develop multivariate discriminant functions, based on variation in the concentration of Si, Fe, Mg, and Ca, for each temperature. The procedure used is part of the commercially available Statistical Package for the Social Sciences (SPSS) software for IBM PCXT and PCAT computers. The functions were then used to analyze the more numerous, four- and five-element analyses of random points. The discriminant functions correctly grouped >97% of previously classified cases. Results of duplicate calculations of the mode in sample (60 wt.% 1921 basalt, 40 wt.% harzburgite) are given in Table 3. They are well within the variance predicted using statistics developed for point counting (Solomon, 1963).

8 106 PETER B. KELEMEN ET AL. siis 5 HJ3 9, oo I s a; 0- ' 3 1 -H p o-h +1 o -H H i _ <N ii -S ^ 2!> 2 3 ^ 2J 5. o

9 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II < d> d> <±> m 547 m 287 <6 <N <6 (N 264 d> r~ 242 <b 258 ON <N <N ^<? i 1 op O -H -H (N 6 +1 > -H fn fn i> -H m o fn <N +1 ON fn -H I **» o!> 2? < ^ Q»^ 8*2 8' s6 ^ O

10 108 PETER B. KELEMEN ET AL. TABLE 3 Comparison of independent estimates for the mode (vol.%) of two samples run at 1150 C Sample Estimate No. of points Vol.%: Olivine Clinopyroxene Orthopyroxene 'Liquid'^ 1* ±5-6p 7-8 ± ± 1-7(1 51-7±6-2B t t ± 3-2 D 13O±l-9 2-4±O ± ±l-l 11-9 ± ±6-5J ±O ± ±4-8H Hand-classified eight- and 11-element analyses from sample t Discriminant analyses based on functions defined using groups from method *. I Discriminant analysis based on functions defined using groups from hand-classified analyses from both sample (method ) and sample (method ). Hand-classified eight- and 11-element analyses from sample Variance is 1 S.D., calculated using the method of Solomon (1963). % Sum of volumes for phases identified as glass and hornblende is given here as volume per cent 'liquid' in each sample. As quench overgrowths a few microns thick were present on olivine and pyroxenes, this classification procedure would have resulted in an overestimate of the proportion of solid phases in each sample if each point analyzed had been infinitely small. However, analyses of crystals near their edges inevitably included glass (or any other adjacent phase) in the excitation volume. All analyses which appeared to be mixtures of crystals plus glass were classified as glass, and all those which appeared to be mixtures including hornblende were classified as hornblende. For a discussion of the variance of modal estimates, and the variance in liquid compositions calculated by mass balance, see Appendix A. EXPERIMENTAL RESULTS Petrography In most samples, maximum crystal diameter was ~60/im for both olivine and calcic pyroxene (hereafter referred to as clinopyroxene); minimum sizes were < 10 pm, and crystals averaged ~30 pm in diameter. Both olivine and clinopyroxene were euhedral to subhedral in control samples (basalt only), and in former 1921 basalt layers (referred to below as 'basaltic layers') in the mixed basalt-harzburgite samples. Euhedral olivine crystals frequently contained small glass inclusions. In former harzburgite layers ('peridotite layers'), olivine crystals were often rounded. Ilmenite (stable in all charges run at 1050 C) formed smaller crystals (ca. 30 /am in maximum dimension). Orthopyroxene, probably metastable, was observed in a few of the most harzburgite-rich samples, in elongate lathes to 20 //m across and 40 pm long. In all samples, overgrowths of iron-rich quench phases were observed. Olivine typically had the narrowest rims, < 2 /jm in most cases. These were too narrow to permit quantitative microprobe analysis, but were clearly visible in scanning and back-scattered electron images. Similarly, quench overgrowths of iron-rich clinopyroxene on subhedral to euhedral clinopyroxene rarely exceeded a few microns. However, quench hornblende overgrowths on clinopyroxene were common and wide: to 10 fim in places. Small anhedral clinopyroxene crystals, commonly showing ragged, fibrous terminations, varied greatly in composition;

11 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 109 some may have actually nucleated and grown during the quench. These were almost invariably intergrown with hornblende. 'Snowflakes' and 'feathers' of quench hornblende were ubiquitous in all samples, but locally absent in peridotite layers where interstitial glass was found to be relatively rich in silica. This is in accord with the observations of Helz (1973, 1976) and Holloway (unpublished data) that hydrous siliceous liquids produce less quench hornblende than hydrous basaltic liquids quenched at about the same (slow) rate from runs in internally heated pressure vessels. Compositional variation within samples Before a review of compositional variation between samples as a function of varying bulk composition, discussion of compositional variation within samples is required. Olivine, clinopyroxene, and liquid proportions and compositions all varied during the runs as a function of position within the capsule, except for the 'control' samples with no harzburgite. In the peridotite layers, glass is interstitial to olivine (Fig. 2), forming an interconnecting network similar to that attained at textural equilibrium in olivine-basalt systems (Bulau & Waff, 1979). As mentioned above, no quantitative analyses of the glass phase are considered representative of liquid composition during the run. However, one can make the qualitative observation that glass is more silica rich within peridotite layers than within basaltic layers. Olivine is abundant, and often the only crystalline phase, within the peridotite layers, and is rare within the basaltic layers. Olivine is most magnesian near the center of peridotite layers, and most iron-rich along the edges of peridotite layers. Clinopyroxene is abundant in the basaltic layers, and virtually absent in the peridotite layers. However, the largest clinopyroxene crystals in basaltic layers are clustered near peridotite layers. Clinopyroxene is most magnesian near these peridotite layers, and least magnesian near the center of basaltic layers. These data, illustrated in Fig. 3, show that a compositional gradient in the liquid was maintained on a millimeter scale in all samples, throughout the duration of the experiments. Orthopyroxene, found only within peridotite layers, and only in samples with >20 wt.% harzburgite, is probably metastable, persisting only where the compositionally zoned liquid phase became enriched in silica. 'PERIDOTITE' LAYER: sample % harzburgite FIG. 2. Texture within a former harzburgite band in sample (40 wt % harzburgite + 60 wt % basalt). Only two phases are present, glass (black) and olivine (white). Clinopyroxene was not found within any former harzburgite layer. The very close resemblance of this texture to the examples of textural equilibrium between olivine and basaltic liquid presented by Waff& Bulau (1979) and other workers is noteworthy.

12 110 PETER B. KELEMEN ET AL. 8 «7 J. a 0 WT% HARZ 1150 C basalt 3 8 o i 5 4- OUVM aounnmni * DQ 40 WT% HARZ 1150 C I 1 1 distance (mm) dl«tance(mm) FIG. 3. Comparison of the molar ratio of Mg/Fe 2+ in olivine and clinopyroxene within and between samples run at the same temperature and pressure. Phase compositions are plotted against linear distance in the long dimension of the tubular Pd-Ag capsules. The left-hand diagram shows relative compositional homogeneity in a 'control' sample (1150-3) which contained only 1921 basalt plus 3 wt. % H 2 O. The right-hand diagram depicts olivine and clinopyroxene compositions in a 1921 basalt-harzburgite mixture (1150-1, 40 wl % harzburgite), and shows extreme compositional heterogeneity. In the basalt-harzburgite mixture, the approximate positions of former harzburgite layers are indicated by a ruled pattern. Both samples were run for 12 h at 1150 C and 5 kb. It should be noted that, although there is a slight compositional gradient in the 'control' sample (perhaps as a result of rapid iron loss in the narrow end of the capsule; there was no measurable thermal gradient during the run), it shows much less chemical variation than the basalt-harzburgite mixture. Clinopyroxene in the 'control' is consistently more magnesian than coexisting olivine, and the average olivine/clinopyroxene Fe 2+ /Mg K t is consistent with that predicted for equilibrium between the two phases (see text). Although the right-hand diagram, for the basalt-harzburgite mixture, appears confused at first, it actually reveals some important details of the reaction process. The sample originally had three harzburgite bands separated by layers of 1921 basalt The harzburgite bands are preserved as olivine-rich bands perpendicular to the long dimension of the sample capsule. The most magnesian olivine is found within the center of these bands, and the most iron-rich olivine in the sample occurs along the interface between former harzburgite and 1921 basalt layers. Although most of the olivine in the sample is probably relict from the harzburgite, none of the crystals retains the composition of the original olivine (Mg/Fe 2+ >8-5). Clinopyroxene is more magnesian in the two basalt bands between the olivine-rich zones than in the basalt bands at either end of the capsule. Such regular variation is an indication that a diffusion gradient persisted in the liquid throughout the experimental run, on a scale of millimeters. Clinopyroxene crystals are generally unzoned to weakly reverse zoned, except for narrow (< 5 /im), iron-rich rims attributed to growth during the quench. Weak, normal zoning in some large olivine crystals in the sample at 1150 C with no harzburgite is attributed to rapid crystal growth at the beginning of a run, followed by incomplete homogenization by solid diffusion during this short (12 h) run. It is clear that these weak zoning patterns are not due solely to the persistence and partial re-equilibration of relict olivine phenocrysts in the 1921 basalt starting material, because the zoned crystals described here have glass inclusions clearly derived from melt trapped by crystal growth during the run. The glass inclusions do not disrupt the normal zoning profile in such crystals (Fig. 4). In samples containing peridotite layers, little, if any, olivine nucleated during the run. However, no analyzed olivine crystals in any sample retain the composition (Fo 89 ) of olivine in the original harzburgite. All have been modified by diffusion and cation exchange with the liquid. Olivine crystals in high-temperature (1150 Q samples, with a 'zebra' configuration, retain fairly weak zoning profiles (Fig. S\ even though olivine in the center of peridotite layers remains more magnesian than that along the edges of such layers. This suggests that at 1150 C, solid diffusion in olivine (on a scale of tens of microns) was faster than diffusion of Mg- and/or Fe-bearing species in the liquid (on a scale of a few millimeters). In absolute

13 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 111 4S I 2 & 40 0 wt% harz 1150 C distance (microns) 60 FIG. 4. Weak normal zoning in olivine formed by crystal fractionation in the 'control' samples during the experiments, as shown by microprobe line scans across olivine crystals from sample (0 wl % harzburgite, 1150 C, 5 kb, 12 h). For a plot of all analyzed clinopyroxene and olivine compositions from this sample, see Fig. 3. It might be suspected that this crystal was a relict phenocryst from the 1921 basalt, except that it has a glass inclusion in its core. It is clear that the crystal grew during the run, and incorporated trapped melt, because the zoning profile is continuous across the inclusion. terms, diffusion in the liquid must have been less than 1 times faster than diffusion in olivine. In lower-temperature (1050 C) samples, with a single 'sandwich' configuration, some olivine crystals preserve relatively marked compositional zoning. The magnitude of the compositional gradient from core to rim is dependent on crystal size and (for crystals of roughly equal size) on position within the peridotite layer. Zoning profiles from three olivine crystals in sample are illustrated in Fig. 5. Olivine crystals along interfaces between basaltic and peridotite layers are weakly zoned, and are the most iron-rich olivine in the samples. Olivine in the center of peridotite layers is also weakly zoned, and is the most magnesian olivine in the samples. Between these two compositional extremes is a zone, of variable width, in which olivine crystals are relatively strongly zoned and have intermediate compositions. These data strongly suggest that the rate of solid diffusion in olivine, on a lo-^im scale, was nearly equal to the rate of diffusion in the liquid, on a 5-10 mm scale. In absolute terms, diffusion in the liquid was 5-10 times faster than diffusion in olivine at this temperature. Modal analysis Despite the fact that equilibrium was not attained in any of the basalt-harzburgite mixtures, both the mode and composition of the phases vary systematically as a function of harzburgite content in the sample. We will first discuss variation in the mode, illustrated in Fig. 6. Samples show increasing wt.% crystalline phases with increasing wt.% harzburgite in the starting material at constant temperature. The ratio of liquid mass to mass of 1921 basalt

14 112 PETER B. KELEMEN ET AL. vine O c «5 O CD percent weight 40 wt% harz 1150 C I i i :TTTTT77T'TT ^cnrru cnrrn \1 ( u OO. \ distance (microns) 30 wt% harz 1050 C FIG. 5. Left: Microprobe traverses of olivine crystals from sample (40 wt. % harzburgite, 1150 C, 5 kb, 12 h). Compositions of all analyzed olivine and clinopyroxene crystals from this sample are plotted in Fig. 3. Olivine from the center of a former harzburgite band is more magnesian than olivine near the edge of the same band, but none of the crystals shows strong compositional zoning. Diffusion within olivine nearly homogenized these crystals, with a 15-^m radius, despite the fact that an obvious diffusion gradient was maintained in the liquid on a scale of millimeters (see text and caption to Fig. 3). Right: Microprobe traverses actoss olivine crystals in sample (30 wt. % harzburgite, 1050 C, 5 kb, 1 week). This sample was a 'sandwich' type experiment with a single broad harzburgite band flanked by 1921 basalt before reaction. The original harzburgite band was about 3 mm wide along the long dimension of the sample capsule. As in Fig. 5, line scans are shown across olivine from the center of the harzburgite band, from the edge of the band, and from an intermediate location. Despite the week-long run time, compositional zoning within the liquid in sample was clearly more extreme than in sample (Figs. 3 and 5). Olivine from an intermediate position in the former harzburgite band has a much steeper zoning profile than the more magnesian crystal from the center and the more iron-rich crystal from the edge. This is strong evidence for a process of coupled diffusion, in which diffusion rates in the liquid (over millimeters) were approximately equal to those in olivine (over tens of micrometers). Olivine near the edge of the former harzburgite layer reacted with a large reservoir of basaltic liquid, whereas olivine near the center of the layer reacted with liquid which had much lower Fe/Mg. The steep compositional gradient in the transitional olivine crystal probably reflects the development of a moving 'reaction front' in the harzburgite layer. in the starting material is nearly constant at constant temperature, and is independent of bulk composition, consistent with basalt-harzburgite reaction in which MJM C «1. This is also the relationship which would be observed if no reaction at all had occurred. However, it can be seen from Fig. 7 that the compositions of olivine and clinopyroxene vary systematically as a function of the proportion of harzburgite in the starting material, reflecting solid-liquid reaction involving iron enrichment in olivine (originally Fo 89 ), and consequently increasing Mg/(Mg + Fe) in the liquid phase. Clinopyroxene also shows a very slight, but statistically significant, enrichment in Mg/(Mg + Fe) with increasing proportion of harzburgite. Additional evidence for extensive reaction is that jio orthopyroxene was found in samples with <30 wt.% harzburgite. Our pilot study of the phase relations of 1921 basalt with 1, 2, and 3 wt.% H 2 O indicates that orthopyroxene is undersaturated in 1921 basalt with >2wt.% H 2 O at 1050 C. All the orthopyroxene in samples with 10 and 20 wt.% harzburgite dissolved in the hybrid liquid. In summary, extensive solid-liquid reaction can be demonstrated, and the mass liquid/mass 1921 basalt remained essentially constant (within the error of measurement). This result is particularly striking as some portions of the liquid reacted with more harzburgite than other portions of the liquid in the same sample. Figuratively speaking, liquid in the center of broad peridotite layers may have equilibrated within a subsystem

15 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 113 n- 24S 20« J OLIVINE '(batalt) wstght parcant hftburgtta 3M starting ntatl 150 OBSERVED MODE 0PX C rcent harzburglt* Wight p«rc*nt harzburgit«fig. 6. Observed mode in samples run at 1050 C(l week), 1075"C(24h),and 115O C(12 h), and 5 kbfor 1 week, determined by automated point counting with the electron microprobe, as described in the text. The weight fraction of each phase has been divided by the weight fraction of 1921 basalt in the sample. The mass of liquid, relative to the 'initial' mass of liquid in the basalt end-member run at the same conditions, was nearly constant throughout reaction with harzburgite. For comparison, the modes of the starting materials in the bulk composition (40 wt. % harzburgite) at 1050 and 1150 C are shown. Orthopyroxene, in samples with 30 and 40 wt. % harzburgite, was observed only within former harzburgite layers, and probably is metastable.

16 114 PETER B. KELEMEN ET AL 'C 40 2 o C weight percent harzburgite r 10 1 FIG. 7. Molar Mg/(Mg + Fe 2 + ) in average olivine (circles) and clinopyroxene (squares) compositions plotted as a function of changing bulk composition for samples run at 1050X (left) and 1150 C (right). Error bars are for 1 S.D. from the mean, and numbers adjacent to each point indicate the number of analyses used to calculate each average. At both temperatures, both phases show a systematic, statistically significant increase in Mg/(Mg + Fe 2 + ) with increasing proportion of harzburgite in the bulk composition. However, disequilibrium in basalt-harzburgite mixtures is indicated by the compositional variation of both phases within samples, and by crossing trends for clinopyroxene vs. olivine composition. Clearly, Fe-Mg exchange between magnesian olivine in harzburgite and more iron-rich clinopyroxene crystallized from 1921 basalt did not reach equilibrium in any mixture. including 90 wt.% harzburgite and only 10 wt.% 1921 basalt. Liquid in a basaltic layer, on the other hand, may have equilibrated in a subsystem containing only 5 wt.% peridotite. (In fact, the compositions of such subsystems did not lie along a binary mixing line between 1921 basalt and peridotite.) The measured mode of a sample may be thought of as the integrated mode of many subsystems, each with different bulk composition. The mode for all samples indicates solid-liquid reaction stoichiometry with MJM c xl. If the stoichiometry varied TABLE 4a Average olivine compositions, 1050 C Wt.% harz Sample no Si Ti Al Cr Fe Mn Mg Ni Ca mfl-number SiO 2 TiO 2 A1 2 O 3 Cr 2 O 3 FeO MnO MgO NiO CaO n lo03±oo03 0O01±(M) (HXX)±(H) O0±O ±31 O6±O ±32 O6±O1 O8±O1 O743±O ±0265 O055 ±0-7 O0±0O 20 ± ± ± ±1-045 O289±O053 O269±O ±03 OO01±O1 OO±O0 OO02±O ± ±41 O7±01 O6±O ± ±O372 OO40±O036 O0±O0 OO80±O ±1-629 O293±O ±1-444 O324±O041 O210±O ±02 O1 ±0-1 O0±O0 O1±O ±48 05 ± ±50 OO06±O1 05 ±02 O801±0O ± ±26 O0±O ± ± ±1-801 O305±0O28 O195±0O ±02 O0±O0 OO±O0O0 OO01±0O01 O287±O025 OO06±O ±27 O6±OO01 O4±O0 O856±O ±0327 0O27±O013 0O±O0 0O60±0O ±1O ± ±1022 O322±O044 O133±O016 9 O990±OO02 O0±O0 0O±O0O0 0O01±OO ±46 O4±O ±47 0O08±O1 03 ± ±23 4O102±O442 O010±0O16 0O±O0 0O53±O ±2-087 O199±OO ± ±31 O101±O025 15

17 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 115 TABLE 4b Average olivine compositions, 1075 C TABLE 4C Average olivine compositions, 1150 C Wt.% harz Sample no Wt.% harz Sample no Si Ti Al Cr Fe Mn Mg Ni Ca mg-number SiOj TiO 2 AI3O3 Cr 2 O 3 FeO MnO MgO NiO CaO n 1-1 ± O±0O 01 ± 0-413±0O28 OO06±O1 l-560±0o30 O6±0O01 0O07±OO ± ±0270 0O84±0O21 0O±0O O045±0O ± ±25 4O730±lO ± ±O lo±o2 0O±0OO0 0O±0O01 01 ± 0246 ±30 O4±O0 l-738±oo33 0O08±0O01 OO08±OO01 O876±O015 4O307±O234 0O09±0O17 OO05±O ± ±1-355 O180±O023 47O19±H ±75 82 ±27 22 Si Ti Al Cr Fe Mn Mg Ni Ca mg-numbct SiO 2 TiO 2 AI 2 O 3 Cr 2 O 3 FeO MnO MgO NiO CaO n 0999 ±03 oooi±oooo O1 ±01 OOOl ± O344±0O19 O5± ±20 07 ±01 O6±01 O826±O ±0195 O054±OO15 0O44±O ± ±O ± ± ±35 O236±O ±03 O1±0O 0O±0O01 O1±O ±27 05 ± ±28 07 ±01 O6±OO02 O852±O ±0322 O045±0O22 O9±O ±24 14O43± O213±O ± ±43 O207±O ±03 O1±OO 0O±OO O1±OO 0256 ±27 OO04±O ±28 O7 ±01 OO05±O ±13 4O104±O313 O030±O022 O0±OO03 32 ± ± O201±O ±1052 O349±O042 O173±O among different subsystems, then such a consistent result for all the experimental samples would be virtually impossible. Thus, solid-liquid reaction in which MJM C x 1 may be taken as a general result, even for subsystems in which the proportion of harzburgite to 1921 basalt may have exceeded 9:1. Variation in crystalline phases as a function of bulk composition Olivine and clinopyroxene composition at constant temperature, plotted against proportion of harzburgite in the starting material (Fig. 7 and Tables 4 and 5), confirm the points made in the preceding discussion: equilibrium was not attained in any of the 1921 basalt-peridotite mixtures, but Fe/Mg exchange did occur between olivine, liquid, and clinopyroxene. Olivine and, to a much smaller degree, clinopyroxene, show statistically significant increases in Mg/(Mg + Fe 2 + ) with increasing proportion of harzburgite in the sample. In the control samples, with no harzburgite, the olivine/clinopyroxene Fe 2+ /Mg K d s X d =(FeO ol /MgO ol )/(FeO cp 7MgO cpi ) provide a check on the degree to which equilibrium was attained in these samples. The olivine/clinopyroxene K d is 111 at 1150 Cand 1-27 at 1050 C. These are greater than one, as expected from observation of natural olivine-clinopyroxene pairs, and the Fe/Mg fractionation between the two phases increases with decreasing temperature. A low value (0-95) for the K d at 1075 C is clearly in error, probably because of the inclusion of analyses of quench crystals in the average clinopyroxene composition for the 1075 C control sample. In

18 TABLE 5a Average cunopyroxene compositions, 1050 C Wt.% harz Sample no Si Ti Al Cr Fe 1 Fe 2 + Fe 3 + Mn Mg Ni Ca Na my-number SiO 2 TiO 2 A1 2 O 3 Cr 2 O 3 FeO 1 FeO Fe 2 O 3 MnO MgO NiO CaO Na 2 O n l-866±0o16 41 ±05 O177±OO20 08 ± ± ±24 O025±0O12 O5±O ±21 O2±0O ±20 0O24±OO03 O786±O ±0938 l-462±o ±O451 O270±O ±O ± ±0424 O146±O ±O463 O053±O ±O ±35 29 l-858±0o13 0O40±O4 O179±O017 0O08±O2 O232±O025 O187±O026 OO44±O015 0O07±O ±09 O1±O0 O804±0O17 O025±05 O819±O019 50O44± ±0138 4O94±0390 O280±O ± ± ±0539 O208±O ±O195 OO41±O ± ±67 8 l-873±0o12 0O37±O4 O164±0O25 O9±O ±24 O208±0O30 O027±O011 0O06±OO ±24 O2±0O ±23 O021±0O ± ±0604 l-338±o ± ± ±O ±0966 O967±O394 O192±0O ±O ± ±O ± ±26 33 ±08 O147±0O40 0O0»±0O02 O202±O046 O174±0O52 28 ±21 0O07±O ±24 O1±0O01 O806±0O48 26 ± ±38 5O846±O ±O ± ± ± ±1-676 lo03±o752 O207±O ±0435 O023±O ± ±47 7 l-860±o015 O043±0O06 O166±0O24 0O09±O ±25 O195±0O27 O041±O8 07 ± ±33 O2±0O01 O770±0O36 O020±O3 O819±O ± ± ±O539 O3O6±O ±O ±0873 l-510±o295 O212±O038 16O74±O ± ±0944 O280±0O44 17 TABLE 5b Average cunopyroxene compositions, 1075 C TABLE 5C Average cunopyroxene compositions, 1150 C Wt.% harz Sample no Wt.% harz Sample no Si Ti Al Cr Fe' Fe 3 + Fe 2 + Mn Mg Ni Ca Na mg-number SiO 2 TiO 2 A1 2 O 3 Cr 2 O 3 FeO 1 Fe 2 O 3 FeO MnO MgO NiO CaO Na 2 O n 1-869±16 0O45±O9 O164±0O23 O018±0O ±32 0O19±O015 O238±O029 0OO6±0O01 O875±0O17 0O01±0O01 O749±O030 0O22±O4 O786±O022 5O215±O515 l-619±o ±O514 O332±O ± ± ±0934 O198±O ±O316 0O19±0O ± ± ±26 0O30±O5 O146±0O34 0O19±0O09 O199±O030 0O22±0O19 O177±O022 0O06±0O01 O904±0O19 0O03±0O01 O783±O034 0O22±O3 O836±0O ± ± ±O ± ± ± ±O710 O187±0O ±O364 O091±0O ±O880 O306±0O38 13 Si Ti Al Cr Fe Fe 2 + Fe 3 + Mn Mg Ni Ca Na mg-number SiO 2 TiO 2 A1 2 O 3 Cr 2 O 3 FeO 1 Fe 2 O 3 FeO MnO MgO NiO CaO Na 2 O n l-894±0o15 26 ± ±20 0O16±O3 O219±0O13 O180±0O19 0O39±O011 0O07±0O ±23 0O02±O ±23 O027±0O02 O841±0O ±O ± ±0445 O539±O ± ± ±O617 O213±0O ±0442 0O81±O ±0O ±24 16 l-888±0o17 27 ± ±22 0O15±0O05 O212±O018 O177±O020 O035±0O10 0O02±0O ±22 O2±O1 O714±O ±03 O845±O ±O ± ±0497 O519±O ±O ± ±0651 O185±O ±0447 0O74±0O ±O812 O344±0O40 19 l-892±0o13 0O26±0O05 O141±0O18 21 ±07 O192±O015 O165±0O20 0O27±O012 OO02±0O 0969 ±34 0O02±0O - O731±0O40 23 ±03 O854±O ±0435 O950±O ±O413 O733±O ± ±0646 O175±O ± ± ±lO10 O32O±42 24

19 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 117 later calculations using the olivine/clinopyroxene Fe 2+ /Mg K d at 1075 C, an interpolated value of 1-23 has been adopted, and the average clinopyroxene composition adjusted accordingly. These values for K d are greater than those predicted by the expression given by Mori & Green (1978) for the variation of K d with temperature (1-06 at 1150 C; 117 at 1050 C)but their data were for ultramafic bulk compositions at 30-^K) kb, 'dry', and are not directly applicable to phase equilibria in hydrous basaltic liquid at 5 kb. In addition, they may have considered all iron in clinopyroxene to be FeO, which would tend to lower their estimates of K A, and would make their equations applicable only to materials which equilibrated under conditions identical to those of their experiments. Agreement between the results of our experiments and those of Holloway & Burnham (1972) (1-24 at 11 C; 1-32 at 1050 C) is strikingly good, especially considering the differences in H 2 O and O 2 fugacity between the two sets of experiments, and the uncertainty in calculated Fe 2+ /Fe 3+ for clinopyroxene analyzed by electron microprobe (Robinson, 1980). The Fe/Mg K d for olivine-clinopyroxene 'pairs' in basalt-harzburgite mixtures are much less than one, clearly indicative of disequilibrium between the two phases, which can also be seen in the crossing trends for olivine and clinopyroxene composition as a function of bulk composition in Fig. 7. At equilibrium, olivine in basalt-harzburgite mixtures would have been more iron-rich, and clinopyroxene would have been more magnesian. TABLE 6a Compositions of minor phases in samples run at 1050 C Sample no. m.% harz Mineral llmenite Orthopyroxene Average orthopyroxene analysis ' Si Ti Al Cr Fe 1 Mn Mg Ni Ca Na SiO 2 TiO, A1 2 O 3 Cr 2 O 3 FeO 1 MnO MgO NiO CaO Na 2 O Total n OO O089 O I0O O O O O O89 O1073 OO OO O OOOOO O OOO 10O O48 O O29 O0726 O O O O * 9-74* 027 3O40* 3110* I0OO0 3* Adjusted to be in equilibrium with average olivine composition in samples and

20 118 PETER B. KELEMEN ET AL. TABLE 6b Orthopyroxene analysis, sample , 40 wt.% harzburgite Si Ti Al Cr Fe 1 Mn Mg Ni Ca Na SiOj TiOj A1 2 O, Cr 2 O 3 FeO 1 MnO MgO NiO CaO Na 2 O Total n O O O35 O O TABLE 6C Orthopyroxene analyses, sample (40 wt.% harz) Dale Sample 2/24 Opx /17 Opx /17 Opx /17 Opx /17 Opx Avg. of 5 Opx SiOj TiO 2 A1 2 O 3 Cr 2 O 3 FeO MnO MgO NiO CaO N 2 O Total Si Ti Al Cr Fe Mn Mg Ni Ca Na O O0167 O3017 O O21 O1073 0O O O O0120 O O O0134 O O Oil O O0125 O3051 0O OO O OO ± ± ± ± ± ± ± ± ±028 6 ± ±028 l-923±o016 O012±O1 O115±O024 O012±O3 O3±O018 O7±O0 l-536±o031 O3±O0 0O92±10 O4±O0

21 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 119 Liquid compositions Average liquid compositions for each sample have been calculated by mass balance, subtracting the average measured crystalline phase compositions, weighted by their modal proportions, from the bulk composition of the starting material. In addition to olivine and clinopyroxene averages (Tables 4 and 5) analyzed ilmenite and orthopyroxene compositions (Table 6) were used in these calculations. One disadvantage of the pure mass balance method for calculating liquid compositions used in this study is that compositions represent a sort of average, or, more precisely, the sum of mass not in the crystalline phases. As such, they are sensitive to any change in the bulk composition of the system during the run (iron loss to the capsule, hydrogen gain). In addition, average liquid compositions might not correspond to the composition of any liquid which was actually present in inhomogeneous samples. If the calculations are accurate, no change in bulk composition occurred during the run, and all the liquid in the mixed 1921 basalt-harzburgite samples could have somehow been homogenized, then the calculated experimental liquid composition would correspond to the homogenized composition. Another disadvantage of the method used in this study is that all errors in crystallinephase compositions and the mode contribute to the potential inaccuracy of the calculated liquid composition. To assess the magnitude of the uncertainty introduced by all these TABLE 7a Calculated liquid compositions, 1050 C m.% harz Sample no Mode (wt. %) Ol Cpx Ilm Liq Opx Composition (wt. %) SiO 2 TiO 2 A1 2 O 3 Cr 2 O 3 FeO'* MnO MgO NiO CaO Na 2 O K 2 O H 2 O Total Fe 2 O 3 FeO a HlO Log/O 2^ 5-7 ± ±4-7 1O±O ± ± ± ±O5 ll-0±o4 O2±O0 4-2 ± ±O2 3O±O1 O7±O0 4-6 ±O5 10O ± ± ±09 6Ol± ± ± ±1O 8-3±10 01± 2-1 ± 11 1O1±O8 3-1 ±O3 07± ±05 10O O O±5-O 25-5 ± ± ± ± ± ±l-O 8-6 ±1-4 01± 6-6±l ± O ±41 O7±O ± ± 7-4 ± O71O O O61O ± ±10 1 ±2-2 01± 6-4 ± ±O8 2-9±O3 O7±O1 4-3 ± All Fe as FeO. t Fe 2 O 3 and FeO in liquid by the method of Kilinc et al. (1983). % Estimated by the method of Burnham (1979). Estimated by the method of Stolper (1982). t /O 2 estimated as described in text, using the a H2O derived by the method of Stolper (1982).

22 TABLE 7b Calculated liquid compositions, 1075 C Wt. % harz Sample no. Mode (wt. %) Ol Cpx Liq Opx Composition (wt. %) SiO 2 TiOj A1 2 O 3 Cr 2 O 3 FeO* MnO MgO NiO CaO Na 2 O K 2 O H 2 O Total Fe 2 O 3 t FeO "mot fl H,O Log/O 2 H Notes as for Table 7a ± ± ± ±04 2-7± ±04 ll-5±o3 O2±O0 6-5 ±08 OO 9-4 ±02 2-6±01 O6±O0 3-8 ±04 10O ± ± ± ± ± ± ±O8 OO 109±l-5 O2±O0 9-6 ±2-7 8O±O7 06±01 06± ±04 10O O TABLE 7C Calculated liquid compositions, 1150 C Wt. % harz Sample no Mode (wt %) Ol Cpx Liq Opx Composition (wt %) SiO 2 TiO 2 A1 2 O 3 Cr 2 O 3 FeO* MnO MgO NiO CaO Na 2 O K 2 O H 2 O Total Fc 2 O 3 f FeO a Hl ot a Hj o5 Log/OJ 2-6 ± ±3O 8O7±fr ±O4 2-8± ±04 ll-8±o3 02±O0 6-2 ±08 OO 9-7 ±02 2-5±01 O6±O0 37 ±04 11± O6± ± ± ± ±O ±07 OO 1O4±O9 O2±O0 7-8 ± ±O6 2-4 ±02 05± ±04 10OO± O ± ± ± ± ± ± ± ±1-8 O1±O1 3-7 ±31 OO 9-6±O8 3O±O3 07± ± ± Notes as for Table 7a.

23 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 121 factors, a computer program was written to solve the full propagation of error problem for the liquid calculation procedure. The details of this procedure are given in Appendix A. The calculated concentration and variance (1 S.D.) for each element in the liquid in every sample are given in Table 7. The ratio of FeO/Fe 2 O 3 in the liquid was calculated by the method of Kilinc et al. (1983), and then liquid compositions in the control samples were compared with the predicted olivine/liquid Fe 2+ /Mg K d, generally observed to be ~0-3 in experiments on 'dry' basaltic liquid, and varying slightly as a weak function of bulk composition and temperature (Roeder & Emslie, 1970; Roeder, 1974; Longhi et al., 1978; Gee & Sack, 1988). Longhi et al. developed several regression equations relating bulk composition to the value of K d. The equations were calibrated on data from dry experiments at 1 b on high- and low-ti lunar basalts, as well as Roeder and Emslie's data on dry terrestrial basalts, but they are potentially applicable to hydrous liquids at pressure. Using liquid compositions from the controls in our experiments, several of the equations predict K d = at 1050 C, at 1075 C, and at 1150 C, provided one calculates mole fractions with anhydrous components on an eight-oxygen basis, and H 2 O on a one-oxygen basis. The value of K d for observed olivine and calculated liquid in our 'control' experiments was 0-30 at 1050 C C, 0-32 at 1075 C, and 0-24 at 1150 C. Given the total lack of reliable experimental information on the effect of H 2 O on the olivine-liquid Fe 2 + /Mg exchange reaction, K d s from the two lowertemperature 'control' samples were deemed acceptably close to a probable equilibrium value. It seems likely that the 1150 C sample lost iron to the capsule during the run, and thus the calculated liquid composition is more iron-rich than the actual composition during the experiment. This is discussed in a subsequent section. Roeder & Emslie (1970) and Roeder (1974) also calibrated equations relating the olivine/liquid molar distribution coefficients for MgO and FeO, D(Mg) and >(Fe 2 ), in dry systems: Roeder A. Emslie 'Controls', this study T( C) D(M ff ) D(Fr" ) D(Mg) * 1-3* * Approximately adjusted for iron loss as described below. The observed values for D(Mg) and D(Fe 2 + ) are consistently lower than those predicted for 'dry' basalts because the concentration of H 2 O in the liquid has a major effect on the individual element distribution coefficients. This effect can be very easily visualized in the forsterite-fayalite binary system. Addition of H 2 O might not affect the dt/dx slope of the liquidus surface at all, and thus the olivine/liquid Fe 2+ /Mg K d might remain constant. The melting points of both the forsterite and fayalite end-members, on the other hand, are greatly reduced by the presence of H 2 O, and so the D(Mg) and D(Fe) values definitely should be smaller in a hydrous system than in a dry system at a given temperature. Compensation for the effects of iron loss at 1150 C The low olivine/liquid Fe 2 + /Mg K d for the 1150 C control sample is probably indicative of iron loss to the capsule during the experiment Additional evidence for iron loss is the paradoxical trend of Fe concentration in the liquid as a function of bulk composition for the 1150 C series. Calculated liquids show increasing Fe contents, despite obvious loss of iron from the liquid to olivine in the harzburgite reactant, and the constant mass of liquid/1921

24 122 PETER B. KELEMEN ET AL. basalt in the starting material. Use of Pd-Ag capsules with a high Ag content minimizes, but does not eliminate, the problem of iron diffusion into noble metal capsules (Stern & Wyllie, 1975). It is not surprising that the high-temperature samples experienced substantially more iron loss than those at 1050 and 1075 C, both because a more Pd-rich alloy was used in the capsule, and because the diffusion rate of iron into the alloy probably increases exponentially with increasing temperature. Given an estimated olivine/liquid Fe 2+ /Mg K d of 0-3, the measured composition of olivine, and an estimate for Fe 2+ /Fe 3+ in the liquid, a new Fe concentration in the liquid was calculated, and the other element abundances were adjusted accordingly. It was assumed that Fe 2+ /Fe 3+ would be approximately the same in the adjusted liquid composition as in the initially calculated liquid, which makes this otherwise simple problem soluble in closed form. Using the adjusted liquid composition, and the composition of the crystalline phases, a 'corrected' bulk composition for 1921 basalt was constructed. Finally, it was assumed that all 1150 C samples lost the same proportion of iron to the capsules, and the bulk compositions of basalt-harzburgite mixtures were modified accordingly. Adjusted liquid compositions were calculated by mass balance as described above. Adjusted bulk and liquid compositions are given in Table 8. The 'corrected' bulk composition obtained in this manner has 9-4 wt.% FeO t0ul, or 84% of the 111 wt.% FeO tou1 in pristine 1921 basalt + 3wt.% H 2 O. Loss of iron would have produced 'free oxygen', which may have combined with hydrogen diffusing through the capsule to form more H 2 O in the sample. Assuming 16% of the iron in the sample was reduced from Fe 2 + to metallic Fe, the total mass of oxygen liberated would have been 25 g, forming g H 2 O and thus increasing the water content of the sample (~0-2 g) by about 15% relative. This potential addition of H 2 O to the sample has been accounted for in the 'corrected' bulk composition. Predicted phase assemblages If equilibrium had been attained in the 1921 basalt-harzburgite mixtures, continued Fe-Mg exchange between olivine, liquid, and clinopyroxene would have altered the composition and the proportions of all three phases. We have already argued that the proportion of liquid and solid would not be greatly affected by this process, because even very harzburgite-rich subsystems have undergone reactions with MJM C close to one. However, the olivine/clinopyroxene proportions could change significantly, and the composition of all three phases could be altered. Thermodynamic reasoning developed by Kelemen (1990) indicates that, as iron in the liquid would be added to olivine, the amount of olivine in the system should be larger at equilibrium than in the observed experimental mixtures. It is not necessary, however, that the net effect of further Fe-Mg exchange would be to decrease the mass of liquid or to increase Mg/(Mg + Fe) in the liquid. The complementary Fe-Mg exchange reaction involving liquid and chnopyroxene would enrich the clinopyroxene in magnesium, decreasing the mass of clinopytoxene, and enriching the magma in iron relative to magnesium. To summarize, we can predict that the proportion of olivine would increase, the proportion of clinopyroxene would decrease, and the liquid fraction would remain roughly constant as the mixtures approached equilibrium, but it is by no means intuitively obvious how the mg-number [wt.% MgO/(MgO + FeO tou1 )] of the liquid would be affected by the combination of these complementary reactions. Predicted phase assemblages provide a useful complement to the experimental data. Control samples, with homogeneous basaltic starting material, approached equilibrium much more closely than did the basalt-harzburgite mixtures. Observed olivine/liquid and

25 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 123 TABLE 8a Bulk compositions adjusted for iron loss in samples run at 1150 C (see text for further explanation) 1921-A Predicted liq. comp. from olivine New bulk comp. of 1921 basalt (sample ) by mass balance In 1921-A, new Fe/old Fe = Take Fe contents of and and multiply by 0-844, then re-normalize totals Wt. % Fe/Mg Fe/Ma {16% (16% Fe loss by wt.) 20 wt. % harz 40 wt. % harz SiO 2 TiO 2 A1 2 O 3 Cr 2 O 3 FeO 1 MnO MgO NiO CaO Na 2 O K 2 O H 2 O at NNO Fe 2 O 3 FeO OOO O TABLE 8b Liquid compositions adjusted for iron loss, 1150 C Wt. % harz Sample no Composition (wt. %) SiO 2 TiO 2 A1 2 O 3 Cr 2 O 3 FeO* MnO MgO NiO CaO Na 2 O K 2 O H 2 O Total Fe 2 O 3 t FeO a HK> Log/O ±04 2-9±01 15O±O4 9-6±O3 O2±O0 6-4 ± ± ± O6±O0 3-8 ± O ± ± ±07 O1±O1 8-0 ±09 O2±O0 8-2±l ±O6 2-4 ±02 O6±O1 3-5 ± O7± ± ± ± ± ± ±08 3-0±O3 07± ± Notes as for Table 7a.

26 O O O O "0 PI PI 73 OB PI tn 2 PI c- TABLE 9 Calculated equilibrium phase assemblages Sample no. Wt. % harz wt. % SiO 2 TiO 2 A1 2 O 3 Fe 2 O 3 FeO MnO MgO CaO Na 2 O K 2 O H 2 O wt. % mode Ol Cpx Liq Mg/(Mg + Fe 2 + ) (moles) Ol Cpx Ol O 6-9 Ol O O

27 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 125 clinopyroxene/liquid Fe/Mg and Ca/Mg K d s from the control samples, combined with mass balance constraints, may be used to predict phase assemblages for the basalt-harzburgite samples which are also closer to equilibrium than the observed phase assemblages in mixed samples. The calculation procedure, involving solution of a system of linear and non-linear equations, is outlined in Appendix B. Predicted phase assemblages, solutions to these systems of equations, are presented in Table 9. Both MgO and FeO contents in the predicted liquids are lower than in the experimental ones. The net result is that the mg-numbers of experimental and equilibrium liquids are, for the most part, quite similar. The mass of olivine in the predicted assemblage is generally larger than in the experimental samples, and the mass of clinopyroxene is smaller. The combined effect of these two changes is that the mass of the liquid is virtually identical in predicted and experimental assemblages (Fig. 8). This is particularly striking; whereas Mg/(Mg + Fe) in the predicted phase compositions was closely constrained by the K d s taken from the controls at each temperature, the phase proportions were not explicitly constrained in any way. Because the mass of liquid is essentially unchanged, the proportions of species other than SiO 2, FeO, MgO, and CaO in the equilibrium liquids are little changed relative to the experimental liquids. Net increase in the mass of olivine at the expense of clinopyroxene results in an increase in both SiO 2 and CaO concentration in the predicted liquid relative to experimental liquid. Also, predicted olivine is more iron-rich than experimental olivine, and predicted clinopyroxene is more magnesian than experimental clinopyroxene. The fact that both MJM e and liquid Mg/(Mg + Fe) are similar in experimental and predicted assemblages is more than a coincidence. It suggests that local, stepwise reaction of melt and peridotite in small subsystems conformed to constraints similar to those on equilibrium in the larger system. FIG. 8. Predicted phase compositions and proportions were calculated for each basalt-harzburgite mixture, using mass balance constraints and equilibrium constants derived from the control samples run at each temperature as discussed in the text. Here predicted phase proportions (filled circles) are compared with observed phase proportions (open circles with error bars). Error estimates for the observed mode were calculated using the method of Solomon (1963), and represent 1 S.D. Agreement between observed and predicted phase proportions is striking, and suggests that solid-liquid reaction stoichiometry was such that the liquid mass did not vary appreciably, even for subsystems in which the proportion of harzburgite to 1921 basalt may have exceeded 9:1 (see text).

28 126 PETER B. KELEMEN ET AL. Liquids at 1050 C: high-alumina basalt and basaltic andesite The average experimental liquids and the predicted liquids have many features in common, and these will be discussed in the next two sections. All but one (experimental liquid in ) of the liquids in the 1050 C run are high-alumina basalts and basaltic andesites as denned by Kuno (1960) (Fig. 9). The high-alumina liquids so produced contain ~4-4wt.% H 2 O, and are saturated in olivine, clinopyroxene, and ilmenite. Preliminary results of other experiments suggest that high-alumina liquids were also produced at 1050 C and 5 kb in 1921 basalt+ 1 and 2wt.% H 2 O. In these systems, the liquids contained wt.% H 2 O and were saturated with olivine, plagioclase, clinopyroxene, and minor iron-titanium oxides. In the light of these data, it seems probable that crystal fractionation of olivine tholeiite at moderate H 2 O partial pressure and roughly 4-8 k total pressure can produce high-alumina basalt, with or without combined assimilation of mantle wall rock. This observation complements, and does not contradict, statements in the recent literature by Brophy & Marsh (1986), Johnston (1986), and Myers (1988) that high-alumina basalt cannot be produced by fractionation of olivine tholeiite at similar pressures under 'dry' conditions. We should note that Baker & Eggler (1983) investigated the crystallization of a high-alumina basalt (AT-1) at a variety of different H 2 O contents at 2 kb. At no H 2 O fugacity is the liquid saturated with olivine ± clinopyroxene, without plagioclase, unlike our experimental liquids. The liquidus mineral in AT-1 compositions, at any H 2 O content, is plagioclase rather than olivine, suggesting that it cannot be derived by crystal fractionation from an olivine tholeiite parent at a pressure of 2 kb. There are many possible reasons for I ALKALINE & CALC-ALKALINE _ ALKALINE & CALC-ALKALINE SO to 52.5 wt% SiO 2 Kino, NajO+KjO (wt %) 4 O (wt %) FIG. 9. Experimental (open symbols) and predicted (filled symbols) liquid compositions plotted on the discriminant diagram of Kuno (1960). Liquids for 1050 C experiments are shown as circles, results for 1075 C experiments are shown as triangles, and those for 1150 C experiments are shown as squares. All of the experimental and predicted liquids at 1050 C fall well within Kuno's field for high-alumina basalts (HAB), whereas the higher-temperature liquids are transitional between tholeiites and high-alumina basalts. It should be noted that this diagram does not explicitly distinguish calc-alkaline basalts. Kuno (1960, 1968) showed that they may lie in all three fields. The diagram used is for liquids with wt. % SiO 2. Some of the liquids plotted fall outside this range; they are indicated by arrows, with approximate SiO 2 concentration (wl %) given. On the discriminant diagrams appropriate to their silica content, they fall into the samefieldsas they do in this figure. To theright,potential liquid lines of descent for magnesian, tholeiitic liquids (at 5 kb) are shown schematically on the same diagram. The dashed lines indicate isopleths of derivative liquid composition at plagioclase saturation, as a function of H 2 O concentration in the initial liquid, from 0 to 3 wt. %. Even at relatively low water contents, plagioclase crystallization is suppressed, and tholeiitic liquids can evolve by olivine ± clinopyroxene fractionation to higher-alumina compositions. Under dry conditions, plagioclase is an early crystallizing phase, and crystal fractionation depletes the liquid in (aluminous) plagioclase component. Water contents of natural magnesian arc tholeiites probably lie in the range of 1-2 wl % H 2 O.

29 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 127 this difference; our experiments differ from those on AT-1 in pressure, oxygen fugacity, and bulk composition. Pressure is thought to be the most important variable in producing a different order of crystallization in the two experimental systems. As suggested by Baker and Eggler, olivine may be the liquidus phase for AT-1 under 'damp' conditions at pressures >2kb. The H 2 O undersaturated phase relations of basaltic liquids at moderate pressure are very poorly known, and it is difficult to determine whether experimental results from dry systems, AT-1 at 2 kb, or these experiments, are more applicable to the formation of high-alumina basalt containing 1 or 2 wl% H 2 O, such as are observed in nature. This ambiguity has already been noted by Baker & Eggler (1983). We agree that the question of the genesis of high-alumina basalt remains an open one. Variation in liquid composition with temperature and bulk composition Both experimental and predicted liquid compositions in basalt-harzburgite mixtures may have analogues in natural mantle-magma systems. Experimental liquids would correspond to the compositions of variably contaminated basalt, passing fairly rapidly through harzburgite wall rock. Closed-system equilibrium of a fixed bulk composition is unlikely in such a situation, but some portions of the liquid may 'see' a large amount of wall rock. Predicted liquids would be most similar to basalt which equilibrated relatively slowly with wall rock and stoped blocks. The most important result of these experiments is that all the predicted liquids, and all but two of the experimental liquids, in 1921 basalt-harzburgite mixtures are calc-alkaline whereas all the liquids in controls, with no harzburgite, are tholeiitic by the definition of Irvine & Baragar (1971), as can be seen in an AFM diagram (Fig. 10). At constant temperature, the m#-number of the liquids [wt.% MgO/(MgO + FeO toul )] increases as a function of increasing proportion of harzburgite except in the cases of the experimental liquid in samples and Predicted liquid compositions show a slight but continuous increase in SiO 2 with increasing harzburgite, due to dissolution of ortho- and clinopyroxene and crystallization of olivine, maintaining a nearly constant mass of liquid. Liquid lines of descent have been constructed by comparing the bulk composition of 1921 basalt (or the composition corrected for iron loss in the 1150 C run, labelled 1921-A) with the liquids produced in different samples. Liquids in the controls are 'iron-enriched' relative to 1921 basalt at its liquidus, having almost the same Fe concentration but much lower Mg and mg-numbct than the liquidus composition. Liquids in 1921 basalt-harzburgite mixtures have lower Fe and Mg concentrations than 1921 basalt at its liquidus, and have m^-numbers intermediate between the controls and the liquidus composition, i.e., liquids in basalt-harzburgite mixtures show little or no 'iron enrichment' relative to 1921 basalt. We recall that these results apply to two different natural systems, as the bulk compositions of '1921 A' and all 1150 C mixtures were changed by iron loss and hydrogen gain during the experiment. All liquids are enriched in incompatible elements relative to 1921 basalt on its liquidus. In addition, all except the controls and two experimental liquids are enriched in SiO 2 relative to the parental liquid. Predicted liquids in the samples which included harzburgite have more SiO 2 than controls, and so show more SiO 2 enrichment relative to the parent as well. The combination of these effects may be visualized on an AFM diagram (Fig. 10), combined with plots of m#-number vs. Na 2 O + K 2 O and vs. SiO 2 (Fig. 11). It is characteristic of calcalkaline liquid series that they have higher SiO 2 and higher Na 2 O + K 2 O, at a given

30 128 PETER B. KELEMEN ET AL. LIQUID COMPOSITIONS 1050 & 1075C LIQUID COMPOSITIONS 1150'C FIG. 10. Portions of AFM diagrams (A = Na 2 O + K 2 O, F = FeO Fe 2 O 3, M = MgO, in wt. %) showing experimental and predicted liquid compositions produced in basalt-harzburgite mixtures and in control samples containing only basalt. The dark solid curve is the discriminant line of Irvine & Baragar (1971) separating the fields of calc-alkaline magma series (below the curve) from tholeiitic magma series (above the curve). For basalt-harzburgite mixtures, all predicted liquid compositions, and all but two experimental liquid compositions, fall into the calc-alkaline field on this diagram. For the 'control' experiments, with samples containing only basalt, all liquid compositions are tholeiitic. Liquid lines of descent have been constructed from the composition of 1921 basalt (adjusted for iron loss in the case of the 1150 C experiments; see text) to the derivative liquid compositons. The upper diagram shows results of experiments at 1050 C (circles) and 1O75 C (triangles), and the lower diagram shows results for experiments at 1150 C, adjusted for iron loss (squares). Open symbols denote experimental liquid compositions; filled symbols are predicted liquid compositions. The shaded area highlights the range of predicted liquid lines of descent involving combined crystal fractionation and assimilation of harzburgite. Liquid lines of descent for the controls are tholeiitic, whereas liquid lines of descent from 1921 basalt to basalt-harzburgite mixtures show that calc-alkaline series can be produced from an olivine tholeiite parent by reaction with mantle peridotite coupled with crystal fractionation. rngf-number, than tholeiitic liquid series. Both experimental and predicted liquid lines of descent for basalt-harzburgite mixtures form trends with higher SiO 2 and higher combined alkalis, at a given mg-number, than the trends formed by liquid lines of descent for the basalt on its own composition. Along the 'liquid lines of descent', the value of MJM C is no longer equal to one, as decreasing temperature has increased the relative mass crystallized. For instance, when comparing 1921 basalt on its liquidus (at about 12 C with 3 wt.% H 2 O) with the predicted assemblage of sample , with 20 wt.% harzburgite, MJM C over the 150 temperature decrease is ~0-43. If we compare 1921 basalt and the predicted assemblage for sample , with 40 wt.% harzburgite, MJM C is As illustrated by Kelemen (1986, 1990), the larger the M,/M c over a given temperature interval, the higher the m#-number of the derivative liquid. This is the reason for the fan of liquid lines of descent emanating from each parental liquid composition in Figs. 10 and 11.

31 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II C i050 e anhydrous wt% (Na 2O+K 2O) anhydrous wt % (SiOJ 53 FIG. 11. Compositions of experimental and predicted liquid compositions for experiments at 1050 C (circles) and 1150 C, adjusted for iron loss (squares) at 5 kb. The variation of m^-number [wt. % MgOAMgO + FeO""")] with wt. % SiO 2 and Na 2 O + K 2 O is illustrated. Experimental liquids are shown as open symbols; predicted liquid compositions are plotted as closed symbols. Liquid lines of descent have been constructed between the composition of 1921 basalt and the various observed and predicted derivative liquids. The shaded area highlights the range of liquid lines of descent predicted for combined crystallization and reaction with harzburgite. Characteristically, calcalkaline liquid series have higher silica and higher alkali contents, at a given m#-number, than tholeiitic series. All liquid lines of descent for the basalt-harzburgite mixtures defined by predicted liquid compositions, and all but one of the lines of descent defined by experimental liquid compositions, have higher silica and alkali contents at a given m#-number than the liquid line of descent for the crystallizing basalt without harzburgite. DISCUSSION Study of reaction between 1921 basalt and harzburgite has led us to the following conclusions: (1) The ratio of liquid mass to mass of 1921 basalt in the starting material is nearly constant at constant temperature, and is independent of bulk composition, consistent with basalt-harzburgite reaction in which the ratio of mass assimilated to mass crystallized

32 130 PETER B. KELEMEN ET AL. (MJM Q ) is approximately equal to 1. This is true of both experimentally observed and theoretically predicted phase proportions, and is therefore likely for reaction between basalt and peridotite in natural systems, whether equilibrium is attained on a macroscopic scale or not. (2) All the predicted liquids, and all but two of the experimental liquids, in 1921 basalt-harzburgite mixtures are calc-alkaline, whereas all the liquids in controls, with no harzburgite, are tholeiitic. More specifically, liquids in the controls are strongly ironenriched [have much lower Mg/(Mg+Fe)] relative to 1921 basalt at its liquidus, whereas liquids in 1921 basalt-harzburgite mixtures show little or no iron enrichment relative to the liquidus composition. Furthermore, both experimental and predicted liquid lines of descent for basalt-harzburgite mixtures form trends with higher SiO 2 and higher combined alkalis, at a given mg-number, than the trends formed by liquid lines of descent for the basalt on its own composition. Again, similar trends are likely to result from any reaction between tholeiitic liquid and mantle peridotite even where macroscopic equilibrium is not fully attained. (3) An unanticipated result of this investigation is that all predicted and experimental liquids produced at 1050 C from 1921 basalt, with or without harzburgite, were 'highalumina' basalts and basaltic andesites (including > 4 wt.% H 2 O) by the definition of Kuno (1960). Higher total pressure, and slightly lower H 2 O contents, may provide the necessary conditions for producing high-alumina basalts with 1-2 wt.% H 2 O from a mantle-derived, olivine tholeiitic parent by crystal fractionation. (4) Less well constrained, but of general interest, is the observation that rates of diffusion in the liquid and solid phases in these experiments differed by only two to three orders of magnitude, as shown by the persistence of compositional gradients in both liquid, on a millimeter scale, and solids, on a 10-/mi scale, throughout the experiments. Mixtures of basalt and peridotite retained inhomogeneous phase compositions after run times ranging from 11 h at 1150 C to 205 h at 1050 C. It is clear from the results of these experiments, and related thermodynamic modelling (Kelemen, 1986, 1990), that combined assimilation of mantle wall rock and crystallization in cooling basaltic liquids can produce a calc-alkaline liquid line of descent under conditions of temperature, pressure, and initial liquid composition which produce a tholeiitic liquid trend in a closed system. An extended discussion of the general consequences of this conclusion is given in Part I of this series (Kelemen, 1990). The essential characteristics of mantle-magma interaction (near-constant magma mass, decreased iron enrichment, and increased SiO 2 content, in derivative liquids compared with the liquids produced by crystal fractionation alone) would not be affected by changes in the composition of the initial basaltic liquid, nor by changes in the peridotite end-member. Calculations of the effect of combined assimilation of ultramafic rock and crystal fractionation in basaltic magmas (Kelemen, 1986, 1990) were all made in H 2 O-free systems, using a variety of basalt and ultramafic end-members (picrite, MORB, arc tholeiite, high-alumina basalt; olivine, harzburgite, lherzolite) but the results are similar in all important respects to the results of these experiments. Many of the magma types characteristic of subduction-related magmatic arcs can be produced by crystal fractionation from an olivine tholeiite derived by partial melting of the upper mantle. Though none of our results rules out substantial partial melting of the subducting slab, high-alumina basalts may be produced by closed-system fractionation with low, but significant, H 2 O contents in the initial liquid. Calc-alkaline basalt and basaltic andesite can be produced by fractionation combined with reaction of the liquid with mantle wall rock before emplacement of magma in the crust. Both of these processes may produce

33 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 131 only ultramafic cumulates, without plagioclase. If such processes are important in arcs, their crystal products may be left primarily beneath the seismic Moho. Thus, the production of broadly sialic and/or calc-alkaline 'continental' crust in magmatic arcs could occur without recycling of SiO 2 and alumina from the subducting slab, and without large-scale formation of a cumulate lower crust with the average composition of olivine tholeiite. ACKNOWLEDGEMENTS This work forms part of the first author's Ph.D. research, supported by a David A. Johnston Memorial Fellowship, an NSF Graduate Fellowship, an Achievement Reward for College Scientists, and NSF EAR We wish to thank Bernard Evans, Mark Ghiorso, and Stewart McCallum for assistance during the project, and for reviews of various versions of the manuscript. In particular, Mark Ghiorso provided extensive advice on analysis of variance in the estimation of average liquid compositions, and on the leastsquares routine used to calculate predicted phase assemblages. Connie Bertka offered valuable insight regarding sample preparation. Alan Boudreau and Dave McDougall provided technical assistance with sample analysis. Josh Lieberman, Claudia Owen, Rachel Cox, Bill Minarik, and Seema Sonnad assisted in preparing this manuscript. The presentation has been substantially improved as a result of patient and thorough reviews by Jim Beard, Tim Grove, and Dana Johnston. REFERENCES Allen, J. C, & Boettcher, A. L., Amphiboles in andesite and basalt, Part II. Stability as a function of P-T-fR 2 O-fO 2. Am. Miner. 63, Marland, G, Amphiboles in andesite and basalt, Part I: Stability as a function of P-T-JO 2. Ibid. 60, Anderson, A. T., Water in some hypersthenic magmas. J. Geol. 87, Parental basalts in subduction zones. J. Geophys. Res. 87, Baker, D. R., & Eggler, D. H., Fractionation paths of Atka (Aleutians) high-alumina basalts: constraints from phase relations. J. Volcanol. Geotherm. Res. 18, Bowen, N. L., 1922 The behavior of inclusions in igneous magmas. J. Geol. 30, Brophy, J. G., & Marsh, B. D., On the origin of high-alumina arc basalt and the mechanics of melt extraction. J. Petrology 27, Bulau, J. R., & Waff, H. S., Mechanical and thermodynamic constraints on fluid distribution in partial melts. J. Geophys. Res. 84, Burnham, C. W, The importance of volatile constituents. In: Yoder, H. S., Jr. (ed.) The Evolution of the Igneous Rocks, Fiftieth Anniversary Perspectives. Princeton, NJ: Princeton University Press, pp Holloway, J. R., & Davis, N. F., The specific volume of water in the range bars, 2O-9O0 C. Am. J. Sci. 267-A (Schairer Volume), Carroll, M, & Wyllie, P. J., Experimental phase relations in the system tonalite-peridotite-h 2 O at 15 Wb: implications for assimilation and differentiation processes near the crust-mantle boundary. J. Petrology 30, in press Conrad, W. K., & Kay, R. W., Ultramafic and mafic inclusions from Adak Island: crystallization history and implications for the nature of primary magmas and crustal evolution in the Aleutian island arc. Ibid. 25, Kay, S. M., & Kay, R. W., Magma mixing in the Aleutian Arc: evidence from cognate inclusions and composite xenoliths. J. Volcanol. Geotherm. Res. 18, Daly, R. A., Manger, G. E., & Clark, S. P., Jr., Section 4: Density of rocks. In: Clark, S. P. Jr. (ed.) Handbook of Physical Constants, Revised Edition, Geol. Soc. Am. Mem. 97, Eggler, D. H., Water-saturated and undersaturated melting relations in a Paricutin andesite and an estimate of water content in the natural magma. Contr. Miner. Petrol. 34, Burnham, C. W., Crystallization and fractionation trends in the system andesite-hjo-coj-oj at pressures to 10 kb. Geol. Soc. Am. Bull. 84, Falloon, T. J., & Green, D. H., Anhydrous partial melting of MORB pyrolite and other peridotite compositions at 10 kban implications for the origin of primitive MORB glasses. Miner. Petrol. 37, Fisk, M. R., Basalt magma interaction with harzburgite and the formation of high-magnesium andesites. Geophys. Res. Lett. 13,

34 132 PETER B. KELEMEN ET AL. Fudali, R. F., Oxygen fugacities of basaltic and andesitic magmas. Geochim. Cosmochim. Acta 29, Fujii, T., & Scarfe, C. M., Composition of liquids coexisting with spinel lherzolite at 10 kbar and the genesis of MORBs. Contr. Mineral. Petrol. 90, Gee, L. L., & Sack, R. O., Experimental petrology of melilite-nephehnites. J. Petrology 29, Ghiorso, M. S., Modeling magmatic systems: thermodynamic relations. In: Carmichael, I. S. E-, & Eugster, H. P. (cds.) MSA Reviews in Mineralogy, Vol. 17: Thermodynamic Modelling of Geological Materials: Minerals, Fluids and Melts, pp Harris, D. M., & Anderson, A. T., Volatiles, H 2 O, CO 2, and Cl in a subduction related basalt. Contr. Miner. Petrol. 87, Helz, R. T., Phase relations of basalts in their melting range at PH 2 O = 5 kb as a function of oxygen fugacity, Part I: Mafic phases. J. Petrology 14, Phase relations of basalts in their melting range at PH 2O = 5 kb, Part II: Melt compositions. Ibid. 17, Hill, R_, & Roeder, P., The crystallization of spinel from basaltic liquid as a function of oxygen fugacity. J. Geol. 82, Holloway, J. R., Chapter 8: Internally heated pressure vessels. In: Ulmer, G. G. (ed.) Research Techniques for High Pressure and High Temperature. Berlin: Springer-Verlag, pp Burnham, W., Melting relations of basalt with equilibrium water pressure less than total pressure. J. Petrology 13, Irvine, T. N., & Baragar, W. R. A., A guide to the chemical classification of the common volcanic rocks. Can. J. Earth. Sci. 8, Johnston, A. D., Anhydrous P-T phase relations of near-primary high-alumina basalt from the South Sandwich Islands. Contr. Miner. Petrol. 92, Wyllie, P. J, The system tonahte-peridotite-h 2 0 at 30 kbar, with applications to hybridization in subduction zone magmatism. Contr. Miner. Petrol. 102, Kelemen, P. B., Assimilation of ultramafic rock in subduction-related magmatic arcs. J. Geol. 94, Reaction between ultramafic rock and fractionating basaltic magma I. Phase relations, the origin of calcalkaline magma series, and the formation of discordant dunite. J. Petrology 31, Ghiorso, M. S., Assimilation of pendotite in calc-alkaline plutonic complexes: evidence from the Big Jim complex, Washington Cascades. Contr. Miner. Petrol., 94, Kilinc, A., Carmichael, I. S. E., Rivers, M. L., & Sack, R. O, The ferric-ferrous ratio of natural silicate liquids equilibrated in air. Ibid. 83, Kuno, H., High-alumina basalt. J. Petrology 1, 121^ Origin of andesite and its bearing on island arc structure. Bull. Volcanol. 32, Longhi, J., Walker, D., & Hays, J. F., The distribution of Fe and Mg between olivine and lunar basaltic liquids. Geochim. Cosmochim. Acta 42, Mori, T, & Green, D. H., Laboratory duplication of phase equilibria observed in natural garnet lherzolites. J. Geol. 86, Myers, J. D., A critical examination of the possible petrogenetic relations between low- and high-mgo Aleutian basalts. Geol. Soc. Am. Bull., 1, Robie, R. A., & Bethke, P. M., Section 5: Molar volumes and densities of minerals. In: Clark, S. P., Jr. (ed.) Handbook of Physical Constants, Revised Edition, Geol. Soc. Am. Mem. 97, Robinson, P., The composition space of terrestrial pyroxenes: Internal and external limits. In: Prewitt, C. T. (ed.) Reviews in Mineralogy, Vol. 7: Pyroxenes, pp Roeder, P. L, Activity of iron and ohvine solubility in basaltic liquids. Earth Planet. Sci. Lett. 23, Emslie, R. F, Olivine-liquid equilibria. Contr. Miner. Petrol 29, Sekine, T., & Wyllie, P., The system granite-peridotite-h 2 O at 30 kbar with applications to hybridization in subduction zone magmatism. Ibid. 81, Experimental simulation of mantle hybridization in subduction zones. J. Geol. 91, Shaw, H. R., Hydrogen-water vapor mixtures: control of hydrothermal atmospheres by hydrogen osmosis. Science 139, Solomon, M., Counting and sampling errors in modal analysis by point counter. J. Petrology 4, Stern, C. R., & Wyllie, P. J., Effect of iron absorption by noble metal capsules on phase boundaries in rockmelting experiments at 30 kilobars. Am. Miner. 60, Stolper, E., A phase diagram for mid-ocean ridge basalts: preliminary results and implications for petrogenesis. Contr. Miner. Petrol. 74, The speciation of water in silicate melts. Geochim. Cosmochim. Acta 46, Takahashi, E., & Kushiro, I, Melting of a dry peridotite at high pressures and basalt magma genesis. Am. Miner. 68, Waff, H. S, & Bulau, J. R., Equilibrium fluid distribution in an ultramafic partial melt under hydrostatic stress conditions. J. Geophys. Res. 84, Webster, J. D., Holloway, J. R., & Hervig, R. L, Phase equilibria of a Be-, U- and F-enriched vitrophyre from Spor Mountain, Utah. Geochim. Cosmochim. Acta 51, Whitney, J. A., The effect of reduced H 2 O fugacity on the buffering of oxygen fugacity in hydrothermal experiments. Am. Miner

35 ULTRAMAFIC ROCK AND FRACTIONATING BASALTIC MAGMA II 133 Wyllie, P. J., & Sekine, T., The formation of mantle phlogopite in subduction zone hybridization. Contr. Miner. Petrol. 79, Yoder, H. S., & Tilley, C. E., Origin of basaltic magmas: an experimental study of natural and synthetic rock systems. J. Petrology 3, APPENDIX A Variance in the estimate of modal proportions and liquid composition calculated by mass balance Errors in the estimated mode are due to anisotropy of the spatial distribution of phases, limited sample size in point counting, and conversion of volume per cent to weight per cent. Point counts of these samples were made along arbitrary line scans perpendicular to the original basalt-harzburgite layering. Inasmuch as some of the anisotropy in phase distribution was parallel to, rather than perpendicular to, these line scans, systematic error was introduced into the modal estimate. Several parallel line scans of each sample were used in an attempt to minimize this effect. Comparison of individual line scans revealed little variation between them, and we have assumed that the systematic error was negligible. The variance due to random error during point counting can be estimated using the formulae of Solomon (1963). Specifically, provided that the grain radius (averaging ~ 15 pm) is less than half the size of the grid spacing, the squared variance is where p is the estimated volume per cent of a phase, a is the grid spacing, and A is the area of the grid. The area of the grid in this case is the summed length of the line scans on a sample multiplied by their width (5 //m), and the grid spacing is the square root of the ratio of the area to the number of points analyzed. A typical variance for these point counts is s 2 < O05p for 2 points over a cumulative 40 mm of line scans, and s 2 <0-17p for 6 points over a cumulative 120 mm of line scans. Error introduced by converting volume per cent to weight per cent is difficult to estimate. We note that the density of glass, rather than melt density, must be used in the conversion. Measurements of the density of natural basaltic to andesitic glasses at atmospheric pressure and room temperature are scarce. Data tabulated by Daly et al. (1966) were examined, and a value of 2-8 g/cm 3 was chosen as potentially representative of glass± 10 vol.% hornblende. It is hoped that the variance is insignificant compared with the error in the estimates of volume proportions. Silicate mineral compositions were recast in mole per cent of the end-members forsterite, fayalite, clinoenstatite, diopside, and hedenbergite, and densities were then calculated using the data of Robie & Bethke (1966). These estimates are thought to be reasonably accurate. Finally, ilmenite abundance in the 1050 C samples was visually estimated to be ~0-5 vol.% of the basaltic layers; this was considered to be equivalent to ~ 1 wt.% of the proportion of 1921 basalt in each sample. In the mass balance calculation of liquid composition, the concentration of each element in the liquid is given by the equation L = [B-OL- W..-CPX- W cpl -OPX- "V-ILM W..J/D - W ol - W cpl - W opj - W llm ] where L, B, OL, CPX, OPX, and ILM are the concentrations of that element in the liquid, bulk composition, olivine, clinopyroxene, orthopyroxene, and ilmenite, and Jf ol, M^p,, W opx, and W ixm are the weight fractions of olivine, clinopyroxene, orthopyroxene, and ilmenite in the mode. The variance of this estimate (ignoring the unknown degree of covariance introduced during microprobe analyses and point counting) is S 2 = in which SL/8 V s is the partial derivative of L with respect to one of the nine independent variables, and Svf is the variance of Vj. The calculated concentration and variance (1 S.D.) for each element in the liquid in every sample are given in Table 7. APPENDIX B Calculation of predicted phase proportions and compositions for basalt-harzburgite mixtures To calculate a predicted phase assemblage, the following assumptions are necessary: (1) olivine and clinopyroxene may be treated as binary solid solutions, (2) the mass of ilmenite is constant in all

36 134 PETER B. KELEMEN ET AL. samples at 1050 C, (3) all observed orthopyroxene is metastable, and (4) at constant temperature, the distribution of Fe 2+ /Mg between olivine and liquid (Kd FeM,-" u -)> Fe 2+ /Mg between olivinc and clinopyroxene (Kd FlM,-""-)> and Ca/Mg between clinopyroxene and liquid (K dcmm^^ are constant over the entire range of bulk composition in these experiments. Given these postulates, a system of eight linear and non-linear equations in eight variables, using the mass balance constraints and K d s from the control samples, can be constructed and solved using least-squares minimization techniques. Specifically, in addition to the ratios defining the K d s, the equations used were and total MgO = Mg0 ol + MgO cpl + MgO" <1 total FeO=Fe0 0l + FeO cpi + FeO llq total CaO = CaO cpx + CaO'"< total SiO 2 = SiO? + SiO c 2 pi + SiO^ total mass = mass" 1 + mass cpl + mass"" + mass llm (constant) Not all of the variables in these equations are independent, of course, as the 'binary' solid solutions of olivine and clinopyroxene constrain the concentration of FeO (and CaO) in these phases once the concentration of MgO is specified. One independent variable set is: weight fraction forsterite in olivine (1), weight fraction Mg-end-member in clinopyroxene (2), weight fractions of olivine (3) and clinopyroxene (4) in the mode, and weight fractions of four oxides (SiO 2, FeO, MgO, and CaO) in the liquid. The clinopyroxene binary solutions used were Wo 38 (En, Fs) 62 at 1050 C, Wo 40 (En,Fs) 60 at 1075 C, and Wo 42 (En,Fs) 58 at 1150 C, based on the observed composition of clinopyroxene in our samples at each temperature. Solution of this system of equations produced results in which the sum of the squares of the residuals was < 10~ 8 (in units of weight fraction), much smaller than the variance of the input parameters, in all cases. The predicted values for D(Mg) olivine/liquid and clinopyroxene/liquid are very nearly constant. Their small variations (±02 for olivine/liquid, ±03 for clinopyroxene/liquid) are not a systematic function of bulk composition, and probably reflect small inconsistencies in the input parameters (K d s and bulk composition). As these distribution coefficients were predicted to be essentially constant over the range of bulk compositions at constant temperature, equations defining D(Mg) for olivine/liquid and clinopyroxene/liquid were added to the system to better constrain the solution. The sum of the squares of the residuals in this larger system of equations did not exceed 10~ 2 wt%, (10~ 4 weight fraction), and thus was still smaller than the uncertainty in the input parameters. Predicted phase assemblages, solutions to these systems of equations, are presented in Table 9.