Journal of Petrology Advance Access originally published online on July 25, 2007
Journal of Petrology 2007 48(9):1725-1760; doi:10.1093/petrology/egm036
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Crystallization and Breakdown of Metasomatic Phases in Graphite-bearing Peridotite Xenoliths from Marsabit (Kenya)
1Institut De Géologie Et DHydrogéologie, Université De Neuchâtel, Rue Emile-Argand 11, CH-2009 Neuchâtel, Switzerland
2Institute of Geological Sciences, University of Bern, Baltzerstrasse 1, CH-3012, Bern, Switzerland
RECEIVED AUGUST 4, 2006; ACCEPTED JUNE 15, 2007
| ABSTRACT |
|---|
Mantle-derived xenoliths from the Marsabit shield volcano (eastern flank of the Kenya rift) include porphyroclastic spinel peridotites characterized by variable styles of metasomatism. The petrography of the xenoliths indicates a transition from primary clinopyroxene-bearing cryptically metasomatized harzburgite (light rare earth element, U, and Th enrichment in clinopyroxene) to modally metasomatized clinopyroxene-free harzburgite and dunite. The metasomatic phases include amphibole (low-Ti Mg-katophorite), Na-rich phlogopite, apatite, graphite and metasomatic low-Al orthopyroxene. Transitional samples show that metasomatism led to replacement of clinopyroxene by amphibole. In all modally metasomatized xenoliths melt pockets (silicate glass containing silicate and oxide micro-phenocrysts, carbonates and empty vugs) occur in close textural relationship with the earlier metasomatic phases. The petrography, major and trace element data, together with constraints from thermobarometry and fO2 calculations, indicate that the cryptic and modal metasomatism are the result of a single event of interaction between peridotite and an orthopyroxene-saturated volatile-rich silicate melt. The unusual style of metasomatism (composition of amphibole, presence of graphite, formation of orthopyroxene) reflects low P –T conditions (
850–1000°C at < 1·5 GPa) in the wall-rocks during impregnation and locally low oxygen fugacities. The latter allowed the precipitation of graphite from CO2. The inferred melt was possibly derived from alkaline basic melts by melt–rock reaction during the development of the Tertiary–Quaternary Kenya rift. Glass-bearing melt pockets formed at the expense of the early phases, mainly through incongruent melting of amphibole and orthopyroxene, triggered by infiltration of a CO2-rich fluid and heating related to the magmatic activity that ultimately sampled and transported the xenoliths to the surface. KEY WORDS: graphite; peridotite xenoliths; Kenya Rift; modal metasomatism; silicate glass
| INTRODUCTION |
|---|
During recent decades numerous studies have described different styles of metasomatism in the Earth's upper mantle. These studies, in combination with experimental work, show that the agents responsible for mantle metasomatism include mafic silicate melts, carbonate melts or C–O–H-rich fluids (e.g. reviews by Menzies & Hawkesworth, 1987
We present a study of metasomatized, graphite-bearing xenoliths from Marsabit (northern Kenya), to investigate melt–rock reaction and fractionation processes. The products of metasomatism are texturally and compositionally variable and unusual. An early Na–Si-rich assemblage comprises volatile-bearing phases (amphibole, phlogopite, apatite) and metasomatic orthopyroxene, as well as graphite. Associated with these phases is a second metasomatic assemblage of carbonate-bearing silicate glass patches (melt pockets), similar to those described from many other xenolith suites (e.g. Frey & Green, 1974
; Dautria et al., 1992
; Ionov et al., 1993
, 1994
; Zinngrebe & Foley, 1995
; Chazot et al., 1996a
; Draper & Green, 1997
; Yaxley et al., 1997
; Coltorti et al., 2000
; Laurora et al., 2001
; Bali et al., 2002
; Ban et al., 2005
). Our approach is based on detailed investigation of trace element variation in minerals and integration of textural constraints. It emphasizes that the exotic nature of the metasomatic products reflects a particular pre-existing physical and chemical environment—i.e. low ambient temperatures (and probably pressures) and relatively low oxygen fugacities—rather than an exotic initial metasomatic agent. We will show that metasomatism in this case does not necessitate multiple metasomatic events of different nature, but simply reflects the product of melt–rock reaction, possibly initiated by common, continental rift-related basanites. We further present mass-balance calculations, which show that the melt pockets were formed by in situ partial melting of the early assemblage (mainly amphibole and orthopyroxene), triggered by infiltration of a CO2-rich fluid.
| SAMPLE CONTEXT |
|---|
The investigated mantle xenoliths were collected from scoriae of basanitic to alkali-basaltic Quaternary cinder cones of the Marsabit shield volcano (Volker, 1990
|
In our previous study (Kaeser et al., 2006
|
Thermobarometry calculations revealed that all the porphyroclastic rocks experienced decompression and cooling from high pressures and temperatures (mainly preserved in formerly garnet-bearing Group I lherzolites; see Fig. 2) to low mantle P–T conditions of <900°C at <1·5 GPa (Kaeser et al., 2006
|
| ANALYTICAL METHODS |
|---|
Modal compositions (Table 1) were determined from digitized images of thin sections and rock slice scans (for details, see Kaeser et al., 2006
10–15 µm) and a lower beam current (15 nA) to avoid Na loss during analysis. Natural and synthetic silicates and oxides were used as standards. Carbonates were analysed with a beam current of 10 nA and a beam size of 10 µm, using natural carbonates and sulfates as standards. The concentrations of CO2 (in carbonates) and Fe3+ (in spinel) were calculated based on stoichiometry. The raw data of all microprobe analyses were corrected using routine PAP software.
Trace element contents in minerals were analysed in-situ on polished thin sections (40–50 µm thick), using a laser ablation (LA) instrument equipped with a 193 nm ArF excimer laser (Lambda Physik, Germany) coupled to an ELAN 6100 (Perkin Elmer, Canada) quadrupole inductively coupled plasma mass spectrometer (ICPMS) at the Institut für Isotopengeologie und Mineralische Rohstoffe, ETH Zürich [see Pettke et al. (2004
) and references therein for the instrumental setup, capabilities and operating conditions]. Raw data were reduced using the LAMTRACE program. Laser pit diameters were between 14 and 110 µm, depending on grain size and the absence or presence of mineral, fluid or melt inclusions, and of exsolution lamellae.
| PETROGRAPHY |
|---|
Group III harzburgites and dunites display two texturally different types of modal metasomatic modification. The first type of metasomatism formed clusters and veinlets of volatile-bearing minerals (amphibole, phlogopite, ± graphite, ± apatite; Fig. 3a–e), associated with metasomatic orthopyroxene. The second type is characterized by the occurrence of melt pockets (e.g. Frey & Green, 1974
|
Based on petrographic investigations, we subdivided Group III peridotites into four types (Type III-a to -d). Textural and geochemical features show that the transition from Type III-a to Type III-c (+d) mainly corresponds to the transformation of modally non-metasomatized spinel harzburgite to pervasively overprinted amphibole harzburgite–dunite. Twenty out of 27 Group III samples investigated petrographically show evidence for modal metasomatism (i.e. they belong to Type III-b or -c).
Type III-a
Type III-a is classified as porphyroclastic, cpx-I-bearing spl harzburgite, devoid of volatile-bearing phases, but showing geochemical evidence of cryptic metasomatism (Kaeser et al., 2006
, and this study). This rock type was originally garnet-bearing and was subjected to decompression, as is indicated by the presence of rare spl–opx–cpx symplectites (Table 1 and Kaeser et al., 2006
).
Type III-b
Xenoliths of Type III-b are cpx-I-bearing spinel harzburgites; however, they contain metasomatic volatile-bearing phases (Table 1). One xenolith (Ke 1965/15) contains veinlets of pale green amphibole, phlogopite, graphite ± apatite (Figs 3e and 4b). Graphite occurs as macroscopic flakes (up to 500 µm long; Fig. 3e), similar to those described from cratonic, kimberlite-hosted peridotite xenoliths (Pearson et al., 1994
). Cpx-I in the vicinity of veinlets is strongly resorbed and develops patches or lamellae of Ti-poor amphibole, indicating metasomatic replacement (Fig. 4b). In all Type III-b samples amphibole and clinopyroxene are partially replaced by the late glass-bearing assemblage (e.g. Fig. 4a). Cpx-I is overgrown by a secondary, green Na-rich Cr diopside (Dim) occasionally containing fluid, glass and/or carbonate inclusions (Fig. 2c). Dim further occurs as micro-phenocrysts in melt pockets, sometimes enclosing (resorbed?) orthopyroxene.
|
Type III-c
Type III-c samples are harzburgites and dunites devoid of cpx-I. Samples transitional between Type III-b and -c still contain relics of cpx-I. Metasomatic products in Type III-c xenoliths generally occur as clusters of amphibole, phlogopite, graphite and rare apatite. The primary assemblage is dominated by ol-I. Green, Ti-poor amphibole and phlogopite typically form aggregates around relic, skeletal spl-I (Fig. 3c and d). Dunite xenolith Ke 1965/25 is unusual as it contains centimetre-scale clusters of large, euhedral twinned amphibole crystals (Fig. 3a) with small phlogopite inclusions. Graphite is associated with amphibole and phlogopite (sometimes found as inclusions in the latter; Fig. 3d) or orthopyroxene (Fig. 3c). Straight contacts between amphibole, phlogopite and ol-I indicate textural equilibrium. Phlogopite also occurs as euhedral, randomly dispersed flakes in orthopyroxene (Fig. 3b). This, together with compositional arguments (e.g. very low Al contents; see below), suggests that orthopyroxene is related to metasomatism (Opxm). Sample Ke 1965/3 also contains early metasomatic interstitial apatite.
Late metasomatic assemblages are ubiquitous in Type III-c xenoliths. They form melt pockets, interconnected by glass films along olivine grain boundaries. Euhedral apple-green Na-rich Cr diopside micro-phenocrysts (Dim; up to 0·3 mm long) in individual melt pockets commonly have the same crystallographic orientation (Fig. 4a), suggesting growth at the expense of an earlier phase. Back-scattered electron (BSE) imaging reveals patchy zoning (Figs 4e and 5a). Melt pockets in several samples enclose relic, resorbed amphibole grains (Fig. 4d). Phlogopite (e.g. in Ke 785*) appears to be less resorbed (Fig. 4d). Early amphibole is rimmed by a kaersutitic selvage (Fig. 4d), indicating that the early amphibole is not in chemical equilibrium with the silicate glass. When in contact with glass, Opxm is also strongly resorbed. Several features suggest incongruent breakdown by a process similar to the reaction opx + melt
ol + cpx + modified melt (e.g. Shaw et al., 1998
): thin glass films are present on orthopyroxene cleavage planes, late euhedral Olm + Dim ± Chrm nucleated within orthopyroxene and Olm overgrows orthopyroxene (Fig. 3b; development of strongly concave opx grain boundaries).
|
Glass in Type III-b and -c samples contains considerable amounts of rounded carbonates (e.g. Figs 4e and 5a), texturally similar to interstitial, glass-related carbonate globules from other xenolith suites (e.g. Ionov et al., 1993
Sample Ke 1959/27 (Type III-c) is unusual in several aspects. It contains kaersutite phenocrysts hosted in the glass. Further, Dim micro-phenocrysts are overgrown by Ti-augite rims (Fig. 4f), and the glass is of an untypical dark brown colour. This xenolith records interaction with a melt similar to the host basanite during ascent to the surface.
Type III-d
Type III-d is a unique sample (Ke 1970/6) of metasomatic coarse-grained spinel harzburgite (grain size
4 mm; Fig. 3f). Olivine and orthopyroxene are almost strain free. Spinel occurs as rounded, black grains (chromite) surrounded by a cluster of apple-green Na-rich Cr diopside (Fig. 3f) containing glass inclusions. Volatile-bearing phases are absent. Several orthopyroxene grains show numerous melt inclusions forming continuous films or trails along cleavage planes. These remarkable textural features can be explained by almost complete annealing of a former Type III-c spinel harzburgite containing silicate glass + Dim micro-phenocrysts, which subsequently recrystallized to form the clinopyroxene clusters around spinel and orthopyroxene with melt inclusions.
| MINERAL COMPOSITION |
|---|
Compositional data (major and trace elements) for non-metasomatic primary phases from all peridotite types, as well as for Ti-pargasite and phlogopite from Group I samples have been given by Kaeser et al. (2006
Early metasomatic assemblages
Clinopyroxene
Cryptically metasomatized cpx-I (Type III-a and -b) are diopsides (Table 2) with high Mg-numbers and low Ti contents typical for mantle harzburgite (e.g. Pearson et al., 2003
). The major element compositions of unmetasomatized cpx-I from Type III-a and -b xenoliths are similar. However, Type III-b samples occasionally contain clinopyroxene that is texturally identical to cpx-I but contains significantly higher amounts of Na2O (up to 2·54 wt%; see Fig. 6a), pointing to cryptic metasomatism. Trace element analyses (Table 2) show enrichment of U, Th, Sr, and LREE relative to middle REE (MREE), variable depletions in heavy REE (HREE), and marked negative anomalies for high field strength elements (HFSE; Nb, Ta, Zr, Hf, Ti; Fig. 7a). Trace element enrichment is strongest in the high-Na cpx-I (Fig. 7a).
|
|
|
Orthopyroxene
Opx-I from Type III-a and -b xenoliths is enstatite-rich (Mg-number of 91·2–92·5). It is LREE-depleted and slightly enriched in U and Th, indicating cryptic metasomatism (Table 3). Opxm (Types III-c and -d) has higher Mg-numbers than opx-I (up to 93·1) and is extremely Al-poor (0·2–1· 0 wt% Al2O3). LREE and MREE are slightly enriched relative to HFSE, which is unusual for mantle-derived orthopyroxene (Rampone et al., 1991
|
Amphibole
Green amphibole (Type III-b and -c) is mostly Ti-poor (Table 4) with very high Si and Na contents and high Mg-numbers (Fig. 8a and b). Low-Ti amphibole in Type III-c is Mg-katophorite (Fig. 8a; according to Leake et al., 1997
|
|
|
|
Phlogopite
Phlogopites (Table 5) have very high Na2O contents (up to 2·88 wt%). Ti concentrations are mostly very low, and Mg-numbers are high (>93). Fluorine and Cl were not detected by electron microprobe. Generally, Na-rich phlogopites have higher U and Th concentrations and lower HFSE contents than phlogopite from the Group I (grt)–spl lherzolites (Kaeser et al., 2006
|
Apatite
Considerable differences were found between the two textural types of apatite (early in sample Ke 1965/3 and phosphate globules associated with glass in Ke 1959/15). Microprobe analyses only poorly satisfy apatite stoichiometry and show fairly low totals (Table 5). This is probably due to the abundant inclusions, which are inevitably included in the analysis. Early apatite is F-rich (1·89–2·53 wt% F) with variable Cl contents (0·92–1·57 wt%). It has extremely high REE and SrO contents (3·38–5·24 wt%; see Table 5). The MgO, Na2O and FeOtot contents are typical for upper mantle apatites (OReilly & Griffin, 2000
In contrast, apatite globules in silicate glass are Cl-rich (2·96–3·14 wt% Cl) and have lower SrO and Na2O, but slightly higher FeOtot contents (Table 5). Both types of apatite are strongly LREE enriched [(La/Yb)N up to 783] and are the major host for U and Th (up to 33 500 times primitive mantle values), whereas HFSE concentrations are very low (Fig. 11b).
|
Late (glass-bearing) assemblages
Silicate glass
Representative results of microprobe and LA-ICPMS analyses of silicate glass are given in Table 6 and selected compositional plots are displayed in Fig. 12a–f. Generally, glass is siliceous (up to 66 wt% SiO2) and peraluminous (12·45–25·07 wt% Al2O3) with characteristics (very low MgO and FeO, high alkalis) similar to those in many other xenolith suites (e.g. Ionov et al., 1994
70; Fig. 12b). Considerable compositional heterogeneity can occur on a very small scale within individual samples. In sample 785*, for examples, glass in the vicinity of Opxm contains up to 65 wt% SiO2, whereas a nearby melt pocket (at less than 5 mm distance) contains glass with
58 wt% SiO2.
|
|
With respect to trace elements, glass is always LREE enriched [(La/Yb)N up to 77] at variable absolute REE concentrations (Fig. 13; YbN from 1· 4 to 91· 0). Zr, Hf and Ti are depleted relative to the REE, and Nb and Ta relative to U and Th. Trace element compositions always mirror the composition of the associated earlier phases (i.e. amphibole in most cases but also clinopyroxene, orthopyroxene, phlogopite and apatite).
|
Glass in Group III xenoliths is clearly different compared with Ti-rich subalkaline glass occasionally found in Group I lherzolites from Marsabit (related to melting of Ti-pargasite; see Kaeser, 2006
Clinopyroxene micro-phenocrysts
Clinopyroxene micro-phenocrysts (Dim) are Na-rich Cr diopsides (up to 3·23 wt% Na2O and 6·45 wt% Cr2O3; Fig. 6a and b) with low TiO2 contents (<0·5 wt%) and high Mg-numbers (up to 94·9). In all samples, Dim are strongly zoned with Cr and Na contents generally decreasing from cores to rims (Fig. 6b). Dim in contact with basanitic glass in the kaersutite-bearing dunite (Ke 1959/27) has resorbed rims, overgrown by Ti-augite (Fig. 4f) with significantly lower Na2O (<1 wt%) contents and Mg-numbers (
85; see Table 2).
Trace element patterns of Dim are similar in shape to the associated glass (except LILE: Cs to U; Fig. 13). Negative HFSE anomalies are mostly due to higher REE abundances; that is, the absolute concentrations of HFSE are similar to those of cpx-I from Type III-a and -b (see Fig. 7a; no real HFSE depletion). Texturally equilibrated clinopyroxene in the Type III-d spl harzburgite (Ke 1970/6; Fig. 3f) is a Na-rich Cr diopside with similar major (Fig. 6a) and trace element characteristics (Fig. 7b) similar to those of the glass-hosted micro-phenocryst cores.
Olivine, spinel, kaersutite
Compared with ol-I, glass-related Olm is more magnesian (Table 3; up to Fo94·6) and enriched in Ca. Ol-I in contact with glass develops strong compositional gradients approaching the composition of Olm at the rims. Spinel micro-phenocrysts (Chrm) are Mg-rich chromites (Cr-number up to 66·5), enriched in (calculated) Fe2O3 compared with spl-I.
Kaersutite occurs as tiny rims on amphibole in Type III-b and -c, and as micro-phenocrysts in Type III-c sample Ke 1959/27. Reaction rims are characterized by higher Mg-numbers than the original amphibole. They are too narrow to be analysed for trace elements. Kaersutite micro-phenocrysts in sample Ke 1959/27 have low Mg-number (about 82) and high K2O contents (Table 4). Major element characteristics as well as trace element patterns (convex-upward REE patterns, high Nb–Ta, low U–Th; Fig. 9d) are similar to those of hornblendites and amphibole megacrysts from alkaline magmas (e.g. Shaw & Eyzaguirre, 2000
).
Carbonates
Most glass-related carbonate globules contain Mg-calcite (Table 7) with an average Mg-number of 96·5 and Ca/(Ca + Mg) between 0·87 and 0·98. In sample Ke 1965/3, MgO correlates positively with SrO (up to 5·35 wt% MgO and 0·89 wt% SrO). Other elements show very low concentrations (Table 7). One sample (Type III-c Ke 1959/15) contains zoned crystals ranging from dolomitic to calcitic in composition (Fig. 5a). Carbonates have low REE contents with flat to slightly LREE-enriched patterns (Fig. 14). Extended trace element plots exhibit a very broad range of concentrations in different samples (e.g. Pb; Fig. 14). Large positive peaks for U, Pb and Sr are typical. In addition, U and Nb are strongly enriched compared with Th and Ta, respectively. Hf, Zr and Ti show very low concentrations. Overall, major and trace element characteristics of the carbonates are similar to those of carbonates found in other peridotite xenoliths (e.g. Ionov, 1998
; Lee et al., 2000
; Laurora et al., 2001
; Ionov & Harmer, 2002
; see Fig. 14).
|
|
| FORMATION OF THE EARLY ASSEMBLAGES |
|---|
Despite large compositional variations, numerous common textural and compositional features indicate that the early metasomatic assemblages in all Group III samples formed during a single metasomatic event. In the discussion below, we argue that during a first metasomatic stage, the primary (pre-metasomatic) phase assemblage in Group III spinel harzburgites was partially replaced by metasomatic minerals, finally resulting in clinopyroxene-free amphibole- and phlogopite-bearing dunite and harzburgite (Type III-c). The initial metasomatic agent was probably an evolved CO2–H2O-rich melt, and the transition from peridotites of Type III-a to -c can be explained by progressive melt–rock reaction.
Transition from cryptic (Type III-a) to modal (Type III-b and -c) metasomatism
Formation of new phases requires (1) crystallization from a melt or fluid phase (decreasing melt or fluid mass) and/or (2) melt–solid or fluid–solid reaction. Textural features such as spinel being replaced by amphibole and phlogopite (e.g. Fig. 3c) indicate melt(fluid)–rock reaction. The geochemical signatures observed in the different peridotite types thus probably reflect pre-existing compositional heterogeneities coupled with metasomatic modification. This is illustrated best in sample Ke 1965/1 (Type III-a), the least metasomatized xenolith in this study. It can be subdivided into three parts containing clinopyroxene with different REE signatures (Zones A, B and C in Fig. 15). The LREE are increasingly enriched from Zone A to C, indicating increasing cryptic metasomatism. HREE abundances, on the other hand, are very heterogeneous in the least metasomatized Zone A, moderately depleted in Zone B, and homogeneously low in the most metasomatized Zone C (containing graphite; Fig. 15), similar to REE patterns in clinopyroxene from the amphibole- and phlogopite-bearing Type III-b sample (Ke 1965/15; Fig. 7a).
|
The decoupling of strongly (LREE, Th, U) from moderately incompatible elements (HREE) can be explained by chromatographic fractionation during percolation of small amounts of melts or fluids through an initially non-metasomatized peridotite (Navon & Stolper, 1987
Similar fractionation processes can explain the modal and compositional differences between Type III-b and -c. Compared with Type III-b, higher abundances of metasomatic phases in Type III-c suggests that higher amounts of fluid or melt crystallized in, or passed through, these rocks. The amphibole-rich dunite (Ke 1965/25), for example, contains low-Ti Mg-katophorite exhibiting the lowest (La/Yb)N and (La/Ce)N ratios, and high (Nb/La)N and (Nb/Th)N ratios (Fig. 10a and c), indicating a moderately fractionated metasomatic agent. Mg-katophorite from the phlogopite-rich harzburgite (Ke 785*; Type III-c), and amphiboles from the more cpx-rich harzburgites (Type III-b) show progressively more fractionated patterns (e.g. Fig. 10c) and decreasing abundances in amphibole-compatible elements (e.g. Nb; Fig. 9a). Thus, the transition from Type III-b to -c could reflect decreasing distance between the site at which the xenolith was sampled and the source of the initial metasomatizing agent (e.g. Zanetti et al., 1996
; Ionov et al., 2002
; Bodinier et al., 2004
). Judging from the abundance of modal metasomatic phases, the amphibole dunite (Ke 1965/25) was probably closest to the source of the metasomatizing agent. The phlogopite-rich nature of Ke 785*, which indicates higher K and LILE (Cs, Rb, Ba) contents in the remaining agent, is in agreement with a more fractionated melt or fluid, at somewhat greater distance from the metasomatic source.
The Nb abundances in amphibole also agree with this model as well. For example, amphibole from Type III-c Ke 1965/25 has comparatively high Nb concentrations (Fig. 9c), in line with relatively high DNbamph/melt values constrained by different studies (e.g. Ionov & Hofmann, 1995
; LaTourrette et al., 1995
). In Ti-poor systems, such as those of this study, Nb uptake by amphibole is even enhanced as a result of increased DNb values (Tiepolo et al., 2001
). Consequently, in an event during which amphibole crystallizes continuously, the remaining fluid or melt will become progressively depleted in Nb (Ionov et al., 2002
), resulting in formation of Nb-depleted amphiboles further away from the source (Zanetti et al., 1996
; Ionov et al., 2002
); such Nb-depleted amphibole is found in sample Ke 785*, as well as in Type III-b xenoliths (e.g. Ke 1959/15; Fig. 9a). Amphibole signatures from xenolith suites modelled successfully in the context of percolative trace element fractionation (e.g. Spitsbergen; Ionov et al., 2002
) show similar ranges (e.g. in Nb/Th ratios; Fig. 10a) to those documented for our samples. This further supports the hypothesis that similar percolation mechanisms acted in the mantle beneath Marsabit.
The role of pre-existing heterogeneities and exotic minerals
Although a model based on trace element fractionation during melt percolation is supported by textures and can account for many compositional features, some compositional peculiarities remain unexplained. First, clinopyroxene and amphibole in several samples (Ke 785*, 1965/1 and 1965/15) are characterized by extremely low HREE abundances, more than one order of magnitude lower than in amphiboles from the strongly metasomatized amphibole dunite Ke 1965/25 (Fig. 9a–c). The main phases that crystallize during metasomatism are low-Ti Mg-katophorite and phlogopite. However, DHREEamph/melt data (e.g. LaTourrette et al., 1995
) indicate that amphibole crystallization alone is unlikely to deplete the metasomatic agent to such a high degree in HREE. The same probably accounts for the extreme Nb depletion in amphiboles from Ke 785*, relative to the other Type III-c xenolith (Ke 1965/25; see also Ionov et al., 2002
).
Further, clinopyroxene, phlogopite and amphibole in the Type III-b sample Ke 1965/15, on one hand, all show strongly fractionated REE patterns (e.g. high La/Ce in amphibole; Fig. 10c) but, on the other hand, are characterized by a positive Ti anomaly, decoupled from other HFSE (e.g. Hf and Zr; Fig. 9b). Trace element fractionation during amphibole crystallization (i.e. formation of Type III-c amphibole dunites), however, should lower Ti contents in the remaining fluid or melt (e.g. Moine et al., 2001
).
Based on textures, mineralogy and dominant common compositional features, the hypothesis that Type III-b and -c xenoliths were metasomatized by completely independent metasomatic agents can be ruled out. Therefore, two other possibilities remain to account for the inconsistent observations described above: (1) inference of other phases, reactants or products, that are able to fractionate specific trace elements, present in the mantle beneath Marsabit but not present in the few investigated samples; (2) pre-metasomatic compositional heterogeneities, similar to the inherited HREE variation in Type III-a clinopyroxene (see above).
Garnet would be an important sink for HREE, but metasomatism occurred in the spinel field, after decompression, with maximum pressures of <1·5 GPa (Kaeser et al., 2006
), so garnet had already broken down to symplectites (Fig. 15). Percolation of melt through garnet-bearing websterites, prior to metasomatism in Ke 1965/1 and 1965/15, could be an alternative. Websterites have been interpreted to occur in close spatial relationship with porphyroclastic peridotites in the shallow mantle beneath Marsabit (Kaeser et al., 2006
), and they occasionally contain rutile and ilmenite, which could further account for the strong Nb and Ta depletion in some samples (e.g. Ke 785*). The available compositional data from the websterites, however, do not indicate metasomatic modification of websteritic garnet (Olker, 2001
; B. Kaeser, unpublished data) nor compositional similarities to Group III peridotites.
Careful study by SEM rules out the presence of exotic phases such as zircon or Ti-oxides, described from other xenolith suites (e.g. Bodinier et al., 1996
; Rudnick et al., 1999
). Nevertheless, it cannot be excluded that the metasomatizing agent percolated through mantle domains containing these phases before infiltrating the studied rocks.
The preservation of pre-metasomatic compositional heterogeneities is another possibility that must be considered. That such rock signals partly overlap with the metasomatic signal is indicated by the HREE variation in Type III-a clinopyroxene (see above) and also by other compositional parameters. For example, both Mg-katophorite and phlogopite growing at the expense of spinel have very high Cr2O3 contents, whereas euhedral Mg-katophorite in the amphibole dunite (Ke 1965/25) or in phlogopite inclusions in Opxm is comparably Cr2O3-poor (Fig. 8b; Tables 4 and 5). Pre-existing heterogeneities would essentially be preserved through elements that are relatively compatible in a given mineral (i.e. that are not mobilized during metasomatism). This would be in agreement with the preserved HREE variations in Type III-a clinopyroxene, high Cr2O3 in amphiboles and phlogopite, and Ti spikes in amphibole from Ke 1965/15.
Further, very high Mg-numbers in replacive amphiboles support melt (fluid)–rock interaction at low melt (fluid)/rock ratios, where Mg and Fe contents are buffered by the ambient peridotite (Kelemen et al., 1995
; Tiepolo et al., 2001
). The slightly lower Mg-number of euhedral amphiboles in Ke 1965/25, combined with their lower Cr2O3 contents (Fig. 8b), suggests a higher melt (fluid)/rock ratio, which is further supported by considerably lower Mg-number in associated ol-I (
Fo88; Table 3).
The limited number of investigated samples does not allow us to decide which of the processes described above acted, and to what extent. However, given the heterogeneity observed in the numerous non-metasomatized xenoliths from Marsabit (Kaeser et al., 2006
), pre-metasomatic features certainly affect the final product of the melt–rock reaction process to a considerable degree.
The nature of the metasomatizing agent(s)
Graphite stability and oxygen fugacity considerations
The Marsabit xenoliths are unusual in that they occur in an off-cratonic setting yet bear macroscopic graphite. The latter usually either occurs in kimberlite-hosted peridotites (on-craton) or in pyroxenites from alpine-type peridotite bodies (e.g. Pearson et al., 1994
). In the phlogopite peridotites from Finero, graphite has recently been reported as nanoscopic layers in phlogopite (Ferraris et al., 2004
). In xenoliths from Marsabit, graphite occurs in the same textural relationship as in kimberlite-hosted peridotite xenoliths (i.e. millimetre-sized flakes, typica














