Mineral and whole-rock chemical data for peridotite xenoliths in basaltic lavas on Spitsbergen are examined to reassess mechanisms of melt–fluid interaction with peridotites and their relative role versus melt composition in mantle metasomatism. The enrichment patterns in the xenoliths on primitive mantle-normalized diagrams range from Th–La–Ce ‘inflections’ in weakly metasomatized samples (normally without amphibole) to a continuous increase in abundances from Ho to Ce typical for amphibole-bearing xenoliths. Numerical modelling of interaction between depleted peridotites and enriched melts indicates that these patterns do not result from simple mixing of the two end-members but can be explained by chromatographic fractionation during reactive porous melt flow, which produces a variety of enrichment patterns in a single event. Many metasomatized xenoliths have negative high field strength element and Pb anomalies and Sr spikes relative to rare earth elements of similar compatibility, and highly fractionated Nb/Ta and Zr/Hf. Although amphibole precipitation can produce Nb–Ta anomalies, some of these features cannot be attributed to percolation-related fractionation alone and have to be a signature of the initial melt (possibly carbonate rich). In general, chemical and mineralogical fingerprints of a metasomatic medium are strongest near its source (e.g. a vein) whereas element patterns farther in the metasomatic ‘column’ are increasingly controlled by fractionation mechanisms.

Knowledge of the partitioning behavior of trace elements between solid and liquid is a prerequisite for modern igneous and mantle petrology. Most of the mathematical models simulating melt generation, migration and evolution within the mantle and/or the crust require the availability of reliable solid/liquid partition coefficients for the mineral phases involved in the process.

Spinel-facies mantle xenoliths occur in a diatreme cutting through the Neogene Southern Patagonia Plateau at Gobernador Gregores (Santa Cruz Province, Argentina). This plateau is in a back-arc position with respect to the Chile trench. Xenoliths differ in their whole-rock composition from other South America occurrences, having higher CaO/Al₂O₃ ratios and, in some samples, TiO₂ enrichment, whereas the Na₂O/Al₂O₃ variation range is similar. Three assemblages can be distinguished. Assemblage 1, in anhydrous protogranular lherzolites and harzburgites, contains clinopyroxene with a depleted major and trace element composition, indicating pre-metasomatic depletion processes. This assemblage fully recrystallized to Assemblage 2 (amphibole ± phlogopite ± Cl-apatite-bearing) during a metasomatic episode. This causes clinopyroxene to acquire geochemical characteristics often attributed to carbonate-melt metasomatism. Noticeably, amphibole is markedly enriched in Nb (up to 298 ppm), especially when depleted in Ti. A further event, related to decompression during xenolith uplift to the surface, induces closed-system (perhaps with the exception of CO₂ addition) disequilibrium melting of Assemblage 2, dominantly of amphibole. It is found in pockets (where amphibole is a residual phase) consisting of Na–Si-rich glass and carbonate (Mg-rich calcite) drops, and in veins originating from the pockets (Assemblage 3). Euhedral olivine, clinopyroxene and spinel crystallize only in the silicate glass. So do new, euhedral apatite crystals when glass is in contact with previous Assemblage 2 apatite. Textural evidence and comparison with experimental work suggest that silicate glass and carbonates are the result of unmixing of a former homogeneous melt. Because of the different flow rates of carbonate and silicate melt, the xenoliths become enriched in carbonate, which is found in the veins during their migration. Thus, the high CaO/Al₂O₃ ratio of whole rocks provides inconclusive evidence of carbonatite metasomatism. This factor, and other minor deviations from the expected results of carbonatite metasomatism, lead us to hypothesize an aqueous, Cl-rich fluid, possibly slab derived, as an alternative agent. Amphibole, resulting from reactive porous flow of this agent in the mantle, could fully explain the observed geochemical features, as indicated by estimates of its partition coefficients.