Recent advances indicate that the amount of carbon released by gradual degassing from the mantle needs to be revised upwards, whereas the carbon supplied by plumes may have been overestimated in the past. Variations in rock types and oxidation state may be very local and exert strong influences on carbon storage and release mechanisms. Deep subduction may be prevented by diapirism in thick sedimentary packages, whereas carbonates in thinner sequences may be subducted. Carbonates stored in the mantle transition zone will melt when they heat up, recognised by coupled stable isotope systems (e.g. Mg, Zn, Ca). There is no single ‘mantle oxygen fugacity’, particularly in the thermal boundary layer (TBL) and lowermost lithosphere where very local mixtures of rock types coexist. Carbonate-rich melts from either subduction or melting of the uppermost asthenosphere trap carbon by redox freezing or as carbonate-rich dykes in this zone. Deeply-derived, reduced melts may form further diamond reservoirs, recognised as polycrystalline diamonds associated with websteritic silicate minerals.
Carbon is released by either edge-driven convection, which tears sections of the TBL and lower lithosphere down so that they melt by a mixture of heating and oxidation, or by lateral advection of solids beneath rifts. Both mechanisms operate at steps in lithosphere thickness and result in carbonate-rich melts, explaining the spatial association of craton edges and carbonate-rich magmatism. High-pressure experiments on individual rock types, and increasingly on reactions between rocks and melts, are fine-tuning our understanding of processes and turning up unexpected results that are not seen in studies of single rocks. Future research should concentrate on elucidating local variations and integrating these with the interpretation of geophysical signals. Global concepts such as average sediment compositions and a uniform mantle oxidation state are not appropriate for small-scale processes; an increased focus on local variations will help to refine carbon budget models.