Supplementary material of 2007GL029515

The dependence of degree of partial melting F on depth beneath a MOR is of primary importance for esti-mating the residual composition of an oceanic plate. The main parameter controlling the depth distribution of F is the so-called melt productivity function γ=(dF/dP)S [e.g. Asimow et al., 1999]. Many authors have estimated γ resulting from adiabatic decompression of an ascend-ing mantle volume [see Langmuir et al, 1992; Asimow et al., 1999 for a review]. Usually a constat value is chosen, ranging between 10- 20 % per GPa of pressure release. Several lines of evidence indicate that a more representa-tive value for normal MORs would lie closer to the lower bound. Indeed, Klein and Langmuir [1987] and Lang-muir et al. [1992], for example, have shown that 12 % GPa−1 seems to be a proper mean value for γ. However, more recent thermodynamic simulations have shown that the productivity is far from constant along the melting regime, but instead it has at least two or three well de-fined paths with nearly constant γ [Asimow et al., 1999; 2001]. Asimow et al. [1999] highlight that the produc-tivity for reversible adiabatic depressurization melting is not constant, but rather increases as melting proceeds. At pressures & 1.0- 1.5 GPa, F is unlikely to exceed 5 % in both batch and fractional melting regimes. Once the pressure decreases below this value, the productiv-ity increases drastically until the clinopyroxene is totally consumed. Fig. 1 shows a summary of the results presented by Asimow et al [1999] and Asimow et al. [2001] for both batch and fractional melting models. The widely used quasi-linear model (i.e. with constant γ, hereafter QL) is also included in this figure [Langmuir et al., 1992; Tur-cotte and Phipps Morgan, 1992]. There are two impor-tant differences between the non-linear (hereafter NL) and QL models. First, at comparable Po (pressure at which melting begins), the QL gives significantly higher mean fractions of melting F than the NL case, where F i