Systematic compositional variations in the Cape York iron meteorite

Systematic compositional variations in the Cape York iron meteorite

oo16-?037/82/101913-08503.0010 GeochimieaCI Cosmochimicu Aria Vol. 46. pp. 1913-1920 Q Pergamon Press Ltd. 1982.Printed inU.S.A. Systematic composit...

1MB Sizes 0 Downloads 31 Views

oo16-?037/82/101913-08503.0010

GeochimieaCI Cosmochimicu Aria Vol. 46. pp. 1913-1920 Q Pergamon Press Ltd. 1982.Printed inU.S.A.

Systematic compositional variations in the Cape York iron meteorite KIM H. EsBENsEN*t, VAGN F. BuCHWALD *, DAN J. MALVIN’! and JOHN T. WASSON? * Institute of Metallurgy, Technical University of Denmark, DK-2800 Lyngby, Denmark. + Institute of Geophysics and Ptanetary Physics, University of California, Los Angeles, CA 90024, USA. (Received August 6, 1981; accepted in revised form June 25, 1982)

A~~ct-Concentrations of Re, Ir and Au are nearly constant within individual masses of the Cape York IIIAB iron meteorite, but differences between masses can be as large as a factor of 2, the extremes being Savik (5.1 Pg/g Ir) and Agpalilik (2.7 ag/g Ir). The S concentration shows a still larger range from 13 mg/g in Agpalilik to 1.4 mg/g in Savik. A relatively large compositional hiatus between Dog and Agpalilik probably reflects inadequate sampIing of the original material. Concentrations of Ir vary by - 10% and Au by -3% between the ends of an 85-cm section from the Agpalilik mass of Cape York, but other sections through Agpalilik show smaller variations. These concentration ranges are much larger than expected from radial crystallization of a moderately large (radius 10 s of km) core. These variations in the Agpalilik mass may reflect dendritic crystallization, or they may have resulted from the process that produced the large concentration range among the Cape York masses. Large gradients in Re and Ir and smail gradients in Ni and Au were also observed in samples within 2 cm of a large (100 cm3) troilite nodule. These gradients may reflect rapidly changing solid/Iiquid distribution coefficients during the final crystallization of S-rich liquid. The compositional trends among the various masses can either be explained by mixing of disparate end members followed by diffusive homogenization on a scale of m, or by dendritic crystallization on the ceiting of the IIIAB magma chamber. The mixing of a solid similar in composition to Savik with a liquid in equilibrium with this solid yields a good match to the observed trends, in which case Agpalilik consists of a mixture of 64% liquid and 36% solids. The bulk S content of the IIIAB core is calculated to be 14 mg/g on the basis of ihis model. INTRODUCI’ION ELEVENOF the 13 groups of iron meteorites (Kracher ec af., 1980) seem to have formed magmatically by the fractional crystallization of molten cores (Scott, 1972; Wasson, 1972), largely based on the interpretation of systematic variations of all elements within these groups. Particularly striking are the large Ir ranges in the largest magmatic groups--6ooO in IIAB and 2000 in IIIAB. An intriguing question is whether Ir and other elements show measurable ranges within the largest iron meteorites from magmatic groups. The most suitable for study is Cape York (at 57 t, the world’s second largest recovered meteorite), a member of magmatic group IIIAB. The combined mass of the Cape York specimens is only slightly less than that of the IVB Hoba (Namibia) iron meteorite that consists of a single specimen. We have examined samples of all the large Cape York masses (Table 1 ), and also sections obtained from one of the larger slices through the Agpalilik mass (Buchwald, 1975). The Widmanstatten patterns are continuous across all Agpalilik slices, indicating that the precursor y-iron (fee) crystal had linear dimensions of at least 2 m. Cuts through Agpalilik parallel to a 100 piane of the y crystal revealed a large number of elongated troilite (FeS) inclusions oriented parallel to one of the cubic axes. Modal analysis by Buchwald (I 975) yielded a S content f 13 f 1 mg/g) that is far higher than that measured in any other IIIAB iron containing ~90 mg/g Ni; the comparison with irons

having higher Ni contents is not meaningful, since the S content of the liquid increases rapidly during the later stages of crystallization. Buchwald (1975) reports S contents of 5 mg/g in Youanmi (78 mg/ g Ni), 4 mg/g in Angelica (74 mg/g Ni) and 3.5 mg/g in Cumpas (78 mg/g Ni). The Youanmi and Angelica values are highly uncertain (+40%) because of the smalf areas (6100 cm’) of the sections; the Cumpas value is probably good to 220%. One purpose of our study was to try to understand why S showed such large variations uncorrelated with concentrations of other reference elements such as Ni. Chromite and phosphate grains, minerals insoluble in and denser and less dense, respectively, than troilite-metal liquids, are commonly concentrated near opposite ends of the troilite inclusions; this orientation is inferred to indicate the direction of the gravitational field gradient in the Cape York parent body (Buchwald, 1971; Kracher et al., 1977). The dominant IIIAB crystallization vector is radial if, as seems probable, the magma formed a central core. Fractional crystallization requires efficient mixing of the residual liquid; convection is the most plausible mixing process. The convection could result from the basal release of latent heat or from an increased concentration of light elements resulting from their rejection by crystallization at the base of the core. No previous studies of compositional variations across sections of magmatic iron meteorites have been reported. The only extensive survey of a nonmagmatic iron is the study of I5 different specimens of the crater-producing, group IAB Canyon Diablo

1913

1914

E; H. Eshensen Y/ 01

'fable1. Concentrations of rune elenrznts rletermlned b) INAA in severallarge masses from the Cape York iron meteoriteshower. Samplesare arrangedin order of decreasingIr concentration. --~___-_CO Nl cu (,a .2s w Ke Ir .&I mg/g mg/g Pg/g up/g vg/g vgln pg/g ug/g ug/g Savik II Savik I AhnighitoW AhnighitoE WOlMIl

4.98 75.4 185 4.97 74.6 169 4.95 76.3 163 4.94 74.6 167 4.99 76.5 166

18.8 4.11 1.27 580 19.0 4.11 1.30 450 19.2 5.01 1.19 480 19.3 5.38 1.21 500 19.6 5.30 1.10 430

5.13 0.567 5.10 0.565 4.87 0.597 4.86 0.608 4.65 0.682

zpgalilikZl+ 5.03 5.14 82.3 78.9 157 166 19.9 19.6 5.37 7.99 1.13 1.02 430 270 4.42 2.73 0.745 0.97 AgpalilikZo+ 5.12 82.8 178 19.3 7.65 1.02 290 2.69 0.95 Thule 5.15 85.2 196 19.7 8.42 0.9' 270 2.68 1.03 + 95-t conf. limit"

1.3 2.9 1l.h 3.2 6.5 9.T 16.0 1.4 1.1

+ AgpalilikZ is the mean of 12 samples;Qpalilik Zl 1s the mean oP14 samples. * Relative95-t confidencehrmt expressedin "s.

iron (Wasson, 1967, 1968). No variations in Ga or Ge resolvably greater than the f4% experimental uncertainties were found. The Ni range was several times greater than was expected from experimental scatter (*2%) on homogeneous metal, but it was suggested that the variations were correlated with local variations in the amounts of minor phases (especially C, (Fe, Ni)& and (Fe, Ni)3P) in this inclusion-rich iron. The precision of these earlier analyses is considerably less than that now achieved for Ir and Au. SAMPLES We are analyzing three sections through the Agpalilik mass: the Z,-Z’, section discussed briefly in this paper, a parallel section Z&-Z&, and a perpendicular section, Y-Y’. We also analyzed samples from the Thule’, Woman, Dog, Savik I and Savik II masses, and from the extreme ends of the 31 t Ahnighito mass. The linear separation between the east (E) and west (W) samples of Ahnighito is -3.2 m. It is not known whether Ahnighito originally consisted of a single y crystal, nor is the direction of the gravitational vector known. Samples are from the Mineralogical Museum, Copenhagen (Agpalilik, Thule, Savik I and II) and the American Museum of Natural History, New York (Ahnighito, Dog and Woman).

EXPERIMENTAL

TECHNIQUES

AND RESULTS

Our data were obtained by instrumental neutron activation (INAA) using the procedure of Willis and Wasson (1981). An innovation in the present study was to cut each sample to nearly identical shapes of 3 X 5 X 5 mm; this yielded improvement in precision by factors of 1.5-7 for all elements except Cu. Sample standard deviations were calculated for 13 Agpalilik replicates. We list 95% confidence limits on the mean of duplicate determinations (based on Students t distribution with 12 degrees of freedom) in the last line of Table 1. In our study of the Agpalilik sections samples from opposite ends of the sections were packed adjacent in the irradiation vial, thus eliminating the potential for producing spurious gradients as a result of inhomogeneities in the

neutron flux. ’ Although Thule was previously listed as an independent iron, we now consider it likely that this small (50 kg) iron found 80 km to the WNW is an outlier of a shower that covered a very large area.

Our ealier study of the Savik I specimen of c ape York of 5.0 pg/g (Scott et (11, 1973). Thus when our Agpalilik results fell in the range 2.7 c 0.2 kg/g we suspected a standardization error but none could be found. We have subsequently analyzed another piece of Savik 1 and found 5.1 rg/g, in excellent agreement with our earlier result. We report here mean concentrations of Co, Ni, Ga, Cu, As, W, Re, Ir and Au in the Z. and 2, sections; data on individual Agpalilik specimens will be included in a later detailed publication.

yielded an Ir concentration

CONCENTRATIONS, GEOGRAPHY AND THE NUMBER OF METEORITE FALLS Data on the large Cape York masses (Table I) show systematic trends in which elements having solid/liquid partition ratios > 1 (Re, Ir) are negatively correlated with those (Ni, As, Au) having ratios < 1. The signs of these correlations are thus the same as those in the magmatic groups. As shown in Table 1 and Fig. 1, the concentrations also change monotonically with recovery location. The meteoride appears to have moved along a trajectory from WNW towards ESE. The farther along the trajectory, the higher the Ir concentration. There is an appreciable compositional hiatus between Dog (4.4 pg/g Ir) and Agpalilik (2.7 pg/ g). This may be a real hiatus present in the meteoroid, but is also explained if samples of intermediate composition fell but were not recovered. The large compositional range requires either that two very large, very closely related meteorites fell very near each other, or that the compositional variations in a single large (probably 310 m) IIIAB meteoroid span at least a factor of 2. The first possibility seems unlikely. The fail rate of large (2100 ton) iron meteorites can be estimated to be no higher than 0. I yr-’ around the entire world. Terrestrial ages of iron meteorites are almost always 51 Ma, and, given the severe climatic conditions in this region, it is extremely unlikely that that of Cape York exceeds this value. Thus the number of massive irons that has fallen in the last Ma is 6105 over the entire Earth. The probability that another massive iron would fall within 15 km of the first is just d105.rr+ 152 km*/5. 10” km*, i.e., 610%. If allowance is also made for the fact that the second iron is nearly identical to the first, the probability is decreased by at least a factor of one hundred, and if the terrestrial survival age of the “older” iron is ~10~ a, the probability falls linearly with the decrease in age. Since we observe a 10% variation in Ir in Agpalilik alone, it seems almost certain that these remarkably large compositional variations were in fact present within the Cape. York meteoroid. Described later are mixing relationships between the different Cape York masses that also support the interpretation that they are fragments of a single meteoroid. COMPOSITIONAL GRADIENTS AN 85-M SECTION

ACROSS

The Ni, Re, lr and Au variations observed in the Z,-Z’, section of Agpalilik are illustrated in Fig. 2. The section consists of two types of regions: in the microgradient region near a troilite nodule large concentration gradients are observed, whereas in the macrogradient region comprising the remainder of the section concentrations gradually increase (Au, Ni) or decrease (Re, Ir). Because the solubility of S in solid Fe-Ni is very low (43.1 mg/g), we can be sure that the numerous troilite nodules in Cape York did not result from solid-state precipitation, but must have crystallized

1915

Cape York meteorite

FIG. 1. The large Cape York masses were found in a region having an area of -200 km2 about 40 km ENE of Cape York. The Thule iron was found about 80 km WNW of this region. The data suggest a low-angle flight path heading roughly ESE. Composition varies along the path; as shown by the symbols, Ir increases (and, .s.g., Au decreases) with increasing distance along the inferred trajectory.

from a cotectic melt. Relative to the macrogradient trend the concentration of Au near the troilite increases by l-295, Ir increases by 2-3% then falls to a value 13% lower in the sample adjacent to the troilite; Re increases by 35% then falls to a value 44% lower adjacent to the troilite. Variations are more pronounced on the Z’ side of the troilite than on the Z side, perhaps indicating a lack of radial symmetry during deposition of solid Fe-Ni from the melt. The FeS/Fe-Ni mole ratio in a cotectic liquid in equilibrium with Cape York metal is about 6 (Kullerud et al., 1969). The density of FeS is -4.7 g cmF3, that of Fe-Ni -7.9 g cme3. The thickness of the troilite nodule in the X and Z directions is estimated to be about 35 mm, in the Y direction about 100 mm. From these we estimate that a uniform outer layer of metal 0.8 mm thick could have been deposited after the cotectic composition was reached, and FeS joined Fe-Ni as a liquidus phase; if deposition was asymmetric the thickness could have been somewhat greater or smaller. This is consistent with the observation of Buchwald (1975, p. 424) that kamacite swathing Agpalilik troilite is 0.3- 1.5 mm thick. Electron microprobe studies of swathing kamacite (Esbensen, 1981) show small changes in Ni (-5% lower), Co (-30% higher) and P (20 f 20% lower) in the innermost 0.1 mm of swathing kamacite near the troilite, but no resolvable difference in composition between the outer swathing kamacite and the adjacent octahedral kamacite bands. Our samples are 3 mm thick and about 0.5 mm was lost during cutting, thus it is likely that a portion of the samples nearest the troilite was the only material that precipitated from a cotectic liquid, and that an inherent difference in the mode of for-

mation of swathing kamacite and normal kamacite cannot be responsible for the elemental trends in Fig. 2, with the possible but unlikely exception of the samples nearest the troilite. Thus, the trends observed in areas up to 2 cm away from the nodule were mainly

0.30

0.25PlL 50 z4

40

30

20

10

0

to

position (cm)

FIG. 2. Concentrations

20

30 4

of Ni, Re, Ir and Au along the Z,-Z; section of Agpalilik; the data are means of duplicate determinations. Concentrations of Re and Ir decrease by about 10% and that of Au increases by about 3% between the extremes of the section, a distance of 85 cm. Near a troilite nodule Ir and Re increase then decrease as the troilite is approached, whereas Au shows only a marginal increase.

1916

K. H. 1

Ii 50 7

1

L

I

1

40

30

20



8

po%ion

y7

I

L-n,

Esbensen it ~1

------1

greater than a few mm. If diffusion coefficients for Re or Ir in kamacite were available, this observation could be used to estimate cooling rates at tempera-

r

I

I

io

20

30

!

40 7s

F&G.3. ~ancentration of Ir decreases with increasing distance across an 85 cm section of Cape York. We interpret this to indicate that the material on the left crystallized earlier than that on the right, since the equilibrium solid/ liquid ~n~ntration ratio k$, > 1. The liquid was probably stagnant because of trapping by rubble dislodged from the ceiling. The curves show sine wave fits that approximate the distribution expected from diffusional leveling of gradients in material formed by mixing so&d metaf with its parental liquid, or formed by dendritic growth, The solid fine is fit to the Agpalilik 2,-Z‘, data only, the dashed curve is fit to data on all Cape York specimens. The minimum in the curve is somewhat arbitrary; we used values of - 1.4 ag/g Ir for both curves.

produced by crystallization when Fe-Ni was the only liquidus phase. Although the high kx values of ir and Re can account for the final drop in ~n~ntration by crystallization of an Fe-Ni shell inwards into the magma pocket, this does not explain why Ir and especially Re’ show the initial rise before the drop. Continuous fractionation of residual liquid can only account for a decrease for elements having kx > 1. We can make two suggestions: (a) fresh liquid having higher Re and Ir was injected and trapped just before the high Re, Ir samples crystallized, or (b) the rapidly increasing concentration of S resulting from the fact that ks G 0.1 may have caused a “salting out” of Ir and Re, i.e., significant increases in k,, and kRe. The latter explanation is preferred because of its simplicity; Kracher and Wasson (1982) argue that on log Ir-log Ni diagrams of the magmatic groups large variations in slopes reflect the fact that kg, is higher in magmas having higher S con~ntrations. Experimental work by Drake and Jones ( 198 1) provides qualitative confirmation of such a trend. If the sharp microgradients in Re and Ir were produced during c~st~li~tion near the cotectic temperature of - 1250 K (Kullerud et al., 1969), then the preservation of these gradients shows that diffusion at T < 1250 K has been limited; during the intervening period at lower temperatures the mean diffusion length for Re and Ir cannot have been z Taken at face value our Re data suggest that the mean Re content of metal within 1 cm of troitite is higher than that farther away. This is not easily understood, and we tentatively assume that this is a sampling artefact that will disappear when we study the metal near additional nodules.

tures about 2X greater than those recorded in Widmanstltten patterns. The Ir macrogradient across the entire length of the Z,-Z; section is shown in Fig. 3. The ratio of the high extreme to the Iow extreme is - 1.14. Our initial impulse was to interpret this smooth trend of essentially constant slope as a segment of the core-wide IIIAB fractional crystallization trend, but some simple calculations indicate that this idea cannot be correct. The Rayleigh equation X=X.&.(1

-- g )+I

(11

gives the concentration of X in a magma having an initial concentration x after a fraction g of the magma has crystallized. If we insert group-IIIAB fractiona~crystaliization parameters given by Willis (1980) (Ir = 4.07 @g/g, and k,, = 4.6) we can estimate that the maximum and minimum Ir concentrations of 2.83 and 2.49 @g/g, respectively, correspond to fractions crystallized of 0.4082 and 0.4286 of the IIIAB magma. If, as discussed above, we assume that the core crystallized in radial symmetry outwards, these values of g and the linear separation of 85 cm yield a core radius of 70 m. A random section can only provide a lower limit on the Ir gradient, but our preiiminary studies of the Z,-i?& and Y-Y’ sections indicate that the Z,-Z\ gradient is near the maximum present in the slab, and thus that this estimate of the core radius is correct for the model employed. That the 70 m estimate is unrealistically small is shown by comparing it with the lower limit for the IIIAB core radius inferred from the observed fall rate and cosmic-ray-age distribution to be 300 m (Wasson, I974), and with inferred cooling rates that indicate a much larger radius. An alternative possibility is that the normal, lowS IIIAB irons formed by basal crystallization, whereas Cape York formed by crystallization downwards from the ceiling. The much higher S content of Cape York could reflect the fact that, S-rich liquids having density fess than the bulk liquid would occasionally get trapped in cavities in the ceiling, whereas the probability of such trapping in basally crystallized metal is far smaller. FoIlowing this scenario, the general absence of Srich IIIA irons requires that the base/ceiling solidification ratio be rl. Some ceiling crystallization must have occurred and we will take 99 as a reasonable upper limit of the base/~iiing ratio. Geometric calculations based on g values of 0.4082 and 0.4286 and an 85-cm separation yields core radii of 214 m and 12.5 km for base/ceiling ratios of 1 and 99, respectively. The smaller of these is smal’ier than the estimated tower limit of the IIIAB core radius of 300 m; the larger falls within the range of plausible

1917

Cape York meteorite

radii for the IIIAB core, but the required base/ceiling ratio is very high. Another, geometric argument indicates that the compositional gradients are not segments of the continuous gradients across the entire core. As indicated earlier, the crystallization and gravitational vectors should be parallel, yet the Z,-Z’, section is perpendicular to the inferred gravity vector. Our unpublished Ir data on the section parallel to the gravity vector yield no evidence of a systematic variation in composition, the maximum range permitted by the data amounting to -3%. Two possible alternative explanations of the systematic compositional variation along section Zr-Z; are that it is part of the general fractionation trend seen among the Cape York specimens (Table 1) or that it could be an artefact resulting from incomplete sampling of nearly constant concentration levels with superposed fluctuations in the neighborhood of the ubiquitous troilite nodules. Support for the latter possibility is provided by the observation that troilite nodules are much more abundant on the Z’, end of the section where Re and Ir are lowest and Au highest. This can be seen in Fig. 189C of Buchwald (1975, p. 121); the Z’, end of our section (the right side of this photo) appears to have -2X more troilite than the other side. LARGE COMPOSITIONAL RANGE AMONG CAPE YORK MASSES In Fig. 4 the general Au, Ir trend among the Cape York masses is compared to that of IIIAB irons having similar compositions. Whereas the Ahnighito, Woman, Dog and Savik points are in or near the 85% limits of the IIIAB field, Thule and Agpalilik fall well outside the field; Au values of the latter are higher than expected from the observed Ir contents. This suggests that the high-Ir Cape York masses are normal IIIAB irons, whereas the low-Ir masses were formed by (an) atypical process(es). The hiatus between these two clusters is pronounced. The common troilite nodules in Agpalilik testify to the existence of numerous pools of liquid having volumes of lo’-lo4 cm3. These demonstrate that this low-Ir material did not form as a result of the gradual upward movement of an essentially planar solid-liquid interface. The only member of the high-Ir cluster that has been sectioned is Savik I. Boggild ( 1927) prepared several large sections and on the basis of point counts, estimated 0.65 volume % FeS (1.4 mg/g S), 9X less than that in Agpalilik. In fact, the etched section of Savik illustrated by Boggild is remarkably similar to that of Casas Grandes (e.g., Wasson, 1974, p. 55). The Savik section is large enough (1640 cm2) to give reasonable assurance of adequate sampling. These observations support the interpretation of Fig, 4 to indicate that Agpalilik is atypical IIIAB material. Note this additional point: the IIIAB iron Cumpas plots very near

0.5-

1

04t

2

iridium? (p&l

5

6

6

(0

FIG. 4. Whereas IIIAB irons form a linear array on a log Au-log Ir diagram, the Cape York masses form a convex-upwards arc extending from near the center of the IIIAB field at the high-Ir extreme to well above this field at the low-Ir extreme. This array seems best modeled by the mixing of high-Ir, Savik-like solids with their parental liquids. Abbreviations: Agn, Ahnighito; Agp, Agpalilik; Cum, Cumpas; CG, Casas Grandes; Dog, Dog; Sav, Savik; Thu, Thule.

Agpalilik (Fig. 4) and, as noted above, has a S content 2-3X higher than typical IIIAB irons. We will examine two mechanisms that yield highIr materials by normal fractional crystallization processes but could possibly lead to the trapping of liquid in the Agpalilik region: dendritic growth at the top of the magma chamber, or the accumulation on the floor of ceiling blocks disrupted by a tectonic event. MODEL I: TECTONICALLY INDUCED MIXING Any tectonic event that leads to accumulations of ceiling blocks on the floor of the magma chamber would lead to the trapping of liquid in the interstices between the blocks. Although the time the IIIAB core solidified has not been determined by radiometric techniques, considerations of probable heat sources and parent-body dimensions indicate a solidification age of 24 Ga. Thus, the crystallization of the core occurred during a period when there was a high impact rate on all bodies in the inner solar system. As a result impact-induced tectonics may have resulted in the periodic dislodgement of blocks of metal from the outer portion of the core. Tectonic processes of internal origin could have accomplished the same. Since the density of solid metal is significantly greater than that of the liquid, any dislodged blocks would have settled to the base of the core within a short time (hours or days for a 1 m sphere). A ceiling disruption model can simultaneously account for our observed Ir and Au variations, the anomalously high S content of the Agpalilik mass, and the observation that the Agpalilik troilite nodules seem to line up into linear arrays, but these arrays are more-or-less randomly oriented rather than being parallel or perpendicular to the long axes of the individual nodules. The irregular orientations are attributed to the more-or-less random spacing of block

1918

I(.

H. Esbensen fat ul

edges, in contrast to that expected from dendrites 01 a cubic solid that would tend to be parallel or, if secondary, perpendicular to each other. Evidence of various sorts favors the conclusion that IIIAB iron meteorites formed in a relatively small parent body, with radius d 200 km, and probably ~100 km. Thus pressure effects on the liquidus temperature were probably negligible, and simultaneous crystallization at all depths of a slightly undercooIed iron meteorite magma was possible. We have already noted that ceiling crystallization alone does not lead to magma mixing, as required to achieve the observed nearly ideal fractional crystallization in group IIIAB. Release of latent heat by crystallization and rejection of low-density elements into the liquid at the base stirs the magma by convection, and the downward movement of blocks dislodged from the ceiling by tectonic action would also yield stirring. Agpalilik cannot consist of pure trapped liquid in equilibrium with the highest Cape York Ir values (5.1 @g/g in Savik) because Willis’ (1980) lIIAB kt, value of 4.6 implies that the mean Ir content of the trapped interstitial liquid was only I. 1 &g/g. Correction for the volume of Agpalilik occupied by troilite would only raise the Ir content of the metal to - 1.2 fig/g, still much less than the Agpalilik value. We noted that the arrangement of Agpalilik troilite nodules into approximatety linear arrays could reflect the presence of trapped Iiquid along the edges of irregularly shaped blocks. To pursue this line of reasoning, we have included on Fig. 4 the loci of compositions produced by mixing Savik-like solids with the coexisting liquid; Scott (1977) modeled the main-group pallasites with a similar calculation. We estimated the composition of the liquid by assuming that Savik metal formed from it under equilibrium conditions. The mean Savik composition from Table 1 must be corrected for the presence of a small amount of trapped liquid which we estimate below on the basis of its S content to be about 1%. For k,, we take Willis’ ( 1980) value of 4.6; kAu we calculated from the slope of the reference line in Fig. 4 to be 0.35 (slope = (kAu - l)/(k,, -. 1)). From these data we can calculate that the Au,, = 0.49 &g/g, I&l = 5.4 fig/g, Auliq = 1.4 pg/g, IS, = f .2 fig/g. The calculated mixing curve gives a satisfactory fit to the data on the various Cape York masses. The worst discrepancy is Agpalilik, whose Au is about 10% lower than the calculated trajectory. This could indicate the need for a slightly more complicated model involving both earlier and later formed solids and a later liquid. If we assume that the simple 2component model is valid, then Agpafilik and Thule are mixtures of -64% liquid and -34% solid (hatchings along the mixing curve show the liquid fraction in 10% increments), whereas Dog, the mass having next lowest Ir content, corresponds to a liquid fraction of only 26%. One could also construct a mixing model whose low-Ir end member corresponds to a later solid. But

whereas the solid-liquid model requires the mixing of phases that must have coexisted, a solid-solid model would require the mixing of materials formed at widely separated times and the exclusion of solids formed during the intervening period. And such a model would also offer no explanation of the anomalously high S content of Agpaliiik. MODEL

2: DENDRITIC

SOLIDIFICATION

Dendritic growth is common in the laboratory freezing of binary metallic systems, but does not occur if the product of the solidification rate and the compositional gradient is too small (Flemings, 1974). Dendritic growth models for the cores of magmatic iron-meteorite groups have been discussed qualitatively in the recent literature (e.g., Narayan and Goldstein, 1982); two of us (Esbensen and Buchwald, 1982) offered a dendritic model to explain the macrogradient trend in Agpalilik. Unfortunately, there seems to be no thorough theoretical discussion of dendritic growth in the asteroidal size range (1 km < radius < 100 km). We infer from the discussion in Flemings (1974) that whether or not dendrites grow at the base of a large convecting magma body depends on the thickness of a boundary layer that does not take part in the convection, and the magnitude of the compositional gradient across this boundary Layer. Dendritic growth may occur within this boundary layer as a result of “compositional undercooling”, i.e., because the liquidus temperature increases more rapidly than the actual melt temperature with distance from the growth front. If these two temperature gradients are nearly the same, minimization of surface energy leads to planar front growth. If the layer is thin and the rate of crystallization is low, diffusion will reduce the compositional gradient and thus the liquidus gradient to a negligible value. If dendritic growth were possible in a thin (say,
1919

Cape York meteorite

by the statement on p. 223 of Wager and Brown (1967) that observations of “adcumulus growth without the production of any conspicuous crescumulate structure ” “is nearly always the case in Skaergaard, Bushveld and many other intrusions.” Most of the materials in layered intrusions seem to have been deposited as flat sheets although some unconformities are attributed to slumping of wall deposits or the dislodgement and settling of ceiling blocks. In the group IIIAB magma the very high degree of fractionation of Re, OS and Ir (by factors of 2000) can only be understood in terms of fractional crystallization that approaches ideal conditions, i.e., complete mixing of the liquid, no mixing of the solid. If growth in a stagnant region between dendrites is responsible for the large amount of trapped liquid in Cape York, then such regions must have been exceptional in the IIIAB core. If the Cape York suite did form as dendrites, we can use our results to estimate their spacing. Flemings (1974) notes that composition profiles produced by dendritic growth followed by diffusion can be approximated with sine waves. We fitted sine functions of varying amplitude to the Ir profile along section Z,-Z’, for samples well away from the troilite nodule. The squares of the residuals are low and nearly constant for an Ir maximum near 3 pg/g and minima over a range of O-l.4 pg/g; these yield wavelengths (i.e., dendrite spacings) of 6-7 m. One of the better fits is shown as the solid curve in Fig. 3. A second scenario is that the Z,-Z’, fractionation is part of an overall fractionation that included It values ranging up to 5.1 pg/g; this yields a larger dendrite spacing. The dashed curve on Fig. 3 shows the best fit obtained with maximum of 5.1 pg/g and minimum of 1.4 rg/ g Ir; the resulting dendrite spacing is -33 m. We emphasize that these values are calculated for illustration only; IIIAB irons such as Agpalilik that are rich in trapped liquid are so exceptional that there is no reason to think that such a model would have general applicability to group IIIAB even if essentially correct for Agpalilik. A more quantitative basis for rejecting a dendritic growth model is that this does not in simple fashion predict the trajectory of the Cape York specimens on Fig. 4. Dendritic growth under various conditions is discussed by Flemings (1974). If we define the X direction to be the growth direction parallel to the axis of the dendrite and r to be the direction to be an outward radius perpendicular to the axis, the Rayleigh equation holds for the unrealistic case where there is no mixing in the solid nor in the liquid in the X direction, but complete mixing in the liquid in the r direction. For the more realistic case where some diffusive mixing in the liquid in the X-direction is allowed, Flemings gives the equation X=kxX[&+(l

-S)(l

-#x-I]

(2)

where the meaning of the symbols is the same as in equation (1) except that d is the mean composition of the liquid at the start of dendrite growth and ax = _z

D.(; 1 -R-X

where D is the diffusion coefficient of X in the liquid, G is the thermal gradient in the liquid adjacent to the solid-liquid interface, 1 is the slope of the gradient in the liquidus temperature at the interface, and R is the rate of movement of the interface. The constant -a is the relative difference between the concentration of X at the growing interface (Xi,,) and that in the well-mixed liquid: -a = (Xi”, - X)/X. It is difficult to estimate these factors for the IIIAB core, but we can examine whether there are values of aAu and air that would yield the array of Cape York points on Fig. 4. We have tried various empirical fits of Equation 2 to the Cape York array using the liquid composition estimated in the previous section, with the imposed constraint that the aAu and aI, values cannot differ by more than a factor of 5. This constraint seems reasonable, since z N Ir, and, D should not differ for Au and Ir by more than a small factor, probably ~5. We find we can roughly fit the Cape York array, but only if we use relatively large values of a (e.g., aI, = 0.1, aAu = 0.5). These values cannot give the entire factor of 2000 IIIAB Ir fractionation, since the maximum fractionation achievable is

kIra,r = 1.3aI,. _

-.

kr

-

1

Because of this and because only a small subset of possible a values are consistent with the Cape York array, we doubt that the dendritic model can be correct for the entire IIIAB core. However, as Esbensen and Buchwald (1982) point out, it may be able to explain the macrogradient in the Z,-Z’, section of Agpalilik. CLOSING REMARKS: THE S CONTENT OF THE IIIAB CORE

In contrast to siderophile elements, it has never been possible to determine the S content of a magmatic core. The problem is as follows. Laboratory experiments indicate that ks is very low, 60.01. The S/( Fe + Ni) mass ratio is about 0.3 in CI chondrites, and probably -0.05 in the IIIAB parent body based on the low abundance of other moderately volatile elements such as Ge or As (Willis, 1980). Thus the equilibrium S concentration in a high-Ir IIIAB iron is expected to be 60.5 mg/g, and that for a IIIAB iron corresponding to 50% crystallization (-0.5 pg/ g Ir) should still be below 2 mg/g. Observed S contents commonly exceed these values, a fact generally interpreted to indicate that the observed S contents reflect small (or in Agpalilik, large) amounts of trapped liquids. Further, the absence of a tendency

for S to increase appreciably with increasing IIIAB Ni content has been interpreted b! Kracher and Wasson (1982) to indicate that random variations in the amount of trapped l~qmd obscure evidence regarding the increase of S that must be occurring in the liquid. Since there has been no independent method to estimate the amount of trapped liquid, the S content of the IIIAB core (or that of any other iron meteorite group) could not be estimated from the Rayleigh equation. If our mixing model (64% liquid, 36% solid) of Agpalilik is correct, and if its S content (13 mg/g) is representative of the volume within which subsolidus metal homogenization occurred. then, if the metal contained -0.5 mg/g S, the S content of the trapped liquid S can be obtained from the relationship 13 = S * 0.64 + 0.5 - 0.36. This yields a liquid S content of 20 mg/g. We can use the Rayleigh for. malism to calculate that the initial S of the IIIAB core was 14 mg/g based on an estimate of the fraction crystallized when Savik formed (0.3) and the assumption that a negligible fraction of the S entered the crystallizing core. This yields a IIIAB/CI-chondrite Ni-normalized S abundance ratio of 0.03. The corresponding ratio for Ge, which is somewhat less volatile than S, is 0.22, thus the calculated S content is in generai agreement with observations for chondrites (Wasson and Chou, 1974) and iron meteorite groups (Wasson and Wai, 1976) that group/C1 abundance ratios decrease with increasing elemental volatility. However. the decrease between Ge and S is substantially larger than observed between adjacent siderophiles (e.g., factor of 1.2 between Sb and Ge). A small part of this difference could result from assumptions or approximations in our calculations of the S content of the liquid, but a drop in abundance ratio by a factor of several between Ge and S seems certain, and requires an explanation in terms of plausible nebular or planetary processes Acknowledgements-We are indebted to J. Willis for teaching INAA techniques to two of us and to G. W. Kallemeyn, G. No, J. Pai and A. Young for technical assistance. We thank M. Prinz for samples of Woman, Dog and Ahnighito E and W. Neutron irradiations were capably handled by T. Zane and C. Ashbaugh at the UCLA reactor. K. E. thanks the Danish State Natural Science Research Council for support, This research was supported in part by NASA grant NGR 05-007-329.

REFERENCES 0. B. (1927) The meteoritic iron from Savik near Cane York, North Greenland. Meddelelser om Gronland 74, 1 l-30. Buchwald V. F. (197 1) The Cape York shower, a typical group IIIA iron meteorite, formed by directional solidification in a gravity field (abstract). Meteorftics 6, 252253. Buchwald V. F. (1975) Handbook of Iron Meteorites. University of California Press, 1418 pp.

Boggild

Drake M. J. and Jones J. t lY8l) Experimental

trace etement partitioning applied to iron meteorites (abstract ). Meteorirics 16, 309 Esbensen K. H ( I98 I 1Chemicui and petro!ogical srudies in iron meteoritic. Ph D Thesis, Technical liniversity of Denmark. 286 pp Esbensen K. H. and Buchwald V. I. (1982) PIanet(otd) core crystallization and fractionation-evidence from the Agpalilik mass of the Cape York iron meteorite shower. Proc NATO Conj. Early Evolution of ihe Terrestrial Planers and rheir Atmospheres (in press). Flemings M. C. ( 1974) Solidification Processing. McGrawHill, 364 pp. Kracher A., Kurat G. and Buchwald V. F. (1977) Cape York: The extraordinary mineralogy of an ordinary iron meteorite and its implication for the genesis of IIIAB irons. Geochem. J. 11, 207-217. Kracher A. and Wasson J. T. (1982) The role 01 S in the evolution of the parental cores of the iron meteorites. Georhim. Cosmochim. Acfu (submitted). Kracher A., Willis J. and Wasson J. T. (1980) Chemical classification of iron meteorites-IX. A new group (IIF). revision of IAB and IIICD, and data on 57 additional irons. Geochim. Cosmochim. Acra 44, 773-787. Kullerud G.. Yund R. A. and Moh G. H. (1969) Phase Relations in the Cu-Fe-S. Cu-Ni-S, and Fe-Ni-S Systems. In Magmatic Ore Deposits (ed. H. D. B Wilson). pp. 323.-343. Econ. Geol. Monographs. Loper D. E. and Roberts P. H. ( 1981) Study of conditions at the inner core boundary of the Earth. Phys. Earth Planer. Inr. 24, 302-307 Narayan C. and Goldstein J. I. (1982) A dendritic solidification model to explain Ge-Ni variations in iron meteorite chemical groups Geochim. Cosmochim. ilcra 46, 259-268. Scott E. R. D. (1972) Chemical fractionation in iron meieorites and its interpretation. Geochim. Cosmochim. Acta 36, I205 1236. Scott E. R. D. (I 977) Geochemical relationships between some pallasites and iron meteorites. Mineral. Mag. 41, 265- 272. Scott E. R. D., Wasson J. T. and Buchwald V. F. (1973) The chemical classification of iron meteorites-VII. A reinvestigation of irons with Ge concentrations between 25 and 80 ppm. Geochim. Cosmochim. Acra 37. 1957. 1983. Wager L. R. and Brown G. M. (1967) Layered Igneous Rocks. W. H. Freeman and Co., 588 pp. Wasson J. T. (1967) Concentrations of Ni, Ga and Ge in a series of Canyon Diablo and Odessa meteorite specimens. J. Geophys. Res. 72, 721-730. Wasson J. T. (1968) Concentrations of nickel, gallium, and iridium in Canyon Diablo and other germanium, Arizona octahedrites. J. Geophys. Res. 73, 3207-3211. Wasson J. T. (1972) Parent-body models for the formation of iron meteorites. Proc. 24th Intern. Geol. Cong. 1% 161. 168. Wasson J. T. (1974) Mereorires. Springer-Verlag, 316 pp. Wasson J. T. and Chou C.-L. (1974) Fractionation of moderately volatile elements in ordinary chondrites. Mefeoritics 9, 69-84. Wasson J. T. and Wai C. M. (1976) Explanation for the very low Ga and Ge concentrations in some iron meteorite groups. Nature 261, 114-l 16. Willis J. (1980) The mean compositions of iron meteorite parenr bodies. Ph.D. Thesis, University of California, LOS Angeles, 208 pp. Willis J. and Wasson J. T. (1981) Instrumental neutron activation analysis of iron meteorites. Radiochim. Acta 29, 45551.