I will present new findings from a multidisciplinary study that combines experiments and geodynamical modelling. Experiments were conducted in the laser-heated diamond anvil cell to determine the melt relations and trace element partitioning between minerals and melts at lower mantle pressures, up to core-mantle boundary conditions (50 to 130 GPa). Phase relations are obtained, and the compositional evolution of the crystallising melt along that of the forming solids are determined at all mantle pressures. In a pyrolitic magma ocean, the first mineral to crystallise in the deep mantle (down to CMB depths) is iron-depleted calcium-bearing bridgmanite. Residual melts are strongly iron-enriched as crystallisation proceeds, making them denser than any of the coexisting solids at deep mantle conditions.

We measured the partitioning of major (Ca, Al, Fe) and trace (Sm, Nd, Lu) elements between bridgmanite and the melt, and find that Brg incorporates some calcium, but that D < 1 so that the melt is enriched in Ca and CaPv does form upon crystallisation. Nd, Sm, and Lu are more compatible than at lower pressure (25 GPa) with D values of 0.25, 0.45 and 1 respectively. The Nd/Sm fractionation increases from 0.3 (25 GPa) to 0.56 (100 GPa), indicating a flattening of the fractionation trends at high pressures.

These partition coefficients are an order of magnitude larger than previously thought, and this has important implications on the 142Nd anomaly in primitive Bridgmanite, which is expected to have very large positive values, and that we will present and discuss. Conversely, CaPv and the residual melt have large negative values. Both these end-members could be mantle sources for both primitive and modern (if they were preserved) lavas and mantle rocks.