It has been shown that about 30% of lithospheric mantle samples are enriched in pyroxene/olivine relative to peridotites residual from partial melting of the primitive mantle (e.g. Boyd, 1989; Kelemen et al., 1992). This is particularly evident for garnet peridotite xenoliths from the Kapvaal craton, which have average normative Opx of about 31 wt% (Boyd, 1989). This average value is much higher than the Opx mode in residues of low pressure melting of primitive mantle, and about the same as the Opx mode in residues derived by partial melting of primitive mantle to the point of Cpx exhaustion at pressures from 25 to 140 kb (e.g. Boyd 1989; Kelemen et al., 1992; Walter et al., 1995; Kinzler, subm.). Thus, about half of all Kapvaal garnet peridotite xenoliths have Opx/Ol higher than in any possible residues of partial melting of primitive peridotite in the upper mantle. Their high Mg#'s, higher than for highly depleted abyssal peridotites, nonetheless suggest a high degree of melting at some point in the history of these lithospheric samples (Boyd, 1989). Boyd (1989) proposed that a high degree of melting at high pressure produced high Mg# residues with about 31% Opx, and metamorphic differentiation created peridotites with higher and lower Opx from these residues. Under these circumstances, there should be no systematic relationship of phase composition with phase proportions in these rocks, especially for samples with Opx > 31%. An alternative is that high Mg# residues with < 25% Opx were created by a high degree of melting at low to moderate pressure, and melt/rock reaction subsequently added SiO2
(+/- other major components) to create high Opx peridotites (Kelemen et al., 1992; Rudnick et al., 1993). Under these circumstances, one would expect systematic variation of phase compositions with Opx mode, extending to the highest Opx proportions.
Using compiled compositions of mantle samples (McDonough and Frey, 1989, plus recent data), mineral norms in weight proportions, and mineral/mineral distribution coefficients (Bodinier et al., 1987), we have calculated [Ni] in olivine in a global database for mantle xenoliths. Even if one assumes that Ni partitioning took place at high temperature, where [Ni]Ol / [Ni]Opx is minimized, we find that [Ni]Ol is positively correlated with the proportion of Opx in samples extending to >50% Opx. Also, [Ni]Ol is not correlated with Cpx proportion. This suggests that the high Opx samples did not form by metamorphic differentiation of a protolith similar to moderate or low Opx samples. Instead, reactions which consume olivine and create Opx, (e.g., Ol1 + liquid1 = liquid2 + Opx +/- Ol2) may have produced high Opx samples and the positive correlation between Ni[Ol] and Opx proportion. This proposed explanation for the composition of lithospheric mantle samples is consistent with the hypothesis that melt/rock reaction in the upper mantle has been intrinsic to the genesis of continental crust (e.g. Kelemen et al., 1993; Kelemen, 1995). In particular, reaction between silicic, small degree melts of subducted eclogite and overlying mantle peridotite could explain global features of both lithospheric mantle (e.g. Rudnick et al., 1993) and continental crust composition, and this is our currently preferred general hypothesis. However, reaction between mantle-derived liquids and conductively cooled upper mantle peridotite certainly has occurred, and could also produce high SiO2 correlated with [Ni]Ol in the continental mantle lithosphere.
Bodinier et al., Am. Min. 72, 902-913, (1987).
Boyd, Earth Planet. Sci. Lett. 96, 15-26 (1989).
Kelemen et al., Nature 358, 635-641 (1992).
Kelemen, Contr. Min. Petr. 120, 1-19 (1995).
Kelemen et al., Earth Planet. Sci. Lett. 120, 111-133 (1993).
Kinzler, J. Geophys. Res., (submitted, 1995).
McDonough & Frey, Rev. Min. 21, 99-145, (1989).
Rudnick et al., Proc 5th Int Kimberlite Conf., (1993).
Walter et al., Earth Planet. Sci. Lett., in press (1995).