Dufek and Cooper (2005; hereafter DC) present an interesting and provocative paper on the origins of ^{226}Ra excess in arc lavas. They develop an incongruent continuous melting model to argue that ^{226}Ra excess can be created by partial melting in the lower crust. However, despite the novelty of their model, the conclusions of DC are questionable for three reasons.

Firstly, one of the reasons for developing the continuous melting model is that DC argue that melt can segregate at low melt fractions (1%–10%). At such low melt fractions, accessory phases such as zircon and allanite are saturated in crustal melt because of their relatively low solubility. Although the abundance of these elements may be low, they control U-series geochemistry because their partition coefficients are several orders of magnitude higher than those of major silicate phases (e.g., Berlo et al., 2004). This implies that we would expect U-series disequilibria at small melt fractions to be even higher, exceeding observations.

Secondly, partition coefficients depend on pressure, temperature, and composition, and therefore must be carefully matched to the problem under investigation. Because the partition coefficients of Ra, Th, and U are all very low, small variations can have a large impact on the calculated disequilibria. DC employ a compilation from a large range of systems (amphibolite to lherzolite) and a range in *P-T* conditions (mantle to crust). The problem of partition coefficients is particularly acute in the case of incongruent melting if product phases have vastly different partition coefficients for parent and daughter nuclides. During non-modal melting of amphibolite, cpx is a product phase. In the calculations of DC, cpx is assigned D_{U} and D_{Th} close to unity, while D_{Ra} is very small. Consequently, the transfer of U and Th to the melt is held back relative to Ra, and large Ra excesses develop. This can be anticipated without the development of a complex mathematical model.

For example, using the DC values for D_{Th} and D_{U}, the simple batch model of Berlo et al. (2004) generates ^{226}Ra excesses up to 4 at 20% of melting. DC's choice of D's for cpx comes largely from the study of Barth et al. (2002), who report D_{Th} = 0.88 ± 0.22. This high value for D_{Th} is wholly at odds with what is known about mineral-melt partitioning (Fig. 1). We contend that near-unity D's for U and Th are inappropriate for modeling crustal melting. Appropriate D's for U and Th would be a factor of ~20 lower (e.g., Blundy and Wood, 2003). Such a low value of D_{Th} would not result in significant ^{226}Ra excess, even with the model of DC.

Thirdly, implicit in the paper by DC is the notion that their model is applicable to most arc rocks. As DC note, “amphibolite melts will dominate the trace element budget of mixtures even when they represent a small mass fraction” (Dufek and Cooper, 2005, p. 835). Melts formed in equilibrium with residual garnet carry distinctive trace element signatures that can be resolved from those of primary arc liquids. Log-normalized diagrams can hide much important detail, but inspection of Figure 2A in DC (2005) shows that their modeled lower crustal melts increase ratios such as Sr/Y and La/Yb by factors of ~8 and ~13, respectively, relative to the starting composition. Thus, the mixed melt plotted in DC's Figure 2B has a Sr/Y ratio of ~62 and a La/Yb ratio of ~6.7, in marked contrast to the average arc rock Sr/Y and La/Yb ratios of 19 and 4.3 (calculated using the data compilation from Turner et al. 2003, not filtered for SiO_{2}). Moreover, there is no positive correlation between ^{226}Ra/^{230}Th and either Sr/Y or La/Yb, as would be predicted if the DC model had general application. Indeed, the highest ^{226}Ra excesses have been observed in those rocks with the lowest La/Yb ratios. Rather, elevated Sr/Y and La/Yb ratios are characteristic of quite rare arc rocks commonly referred to as adakites. It has long been argued that adakites are formed by partial melting of amphibolite, either in the lower crust or in the subducting slab. Thus, the DC model may be highly applicable to adakites, but far less so to the vast majority of arc rocks on which hypotheses for rapid melt ascent and differentiation have been based.

In summary, we agree with DC that lower crustal melting may be important in modifying mantle-derived U-series signatures in some arc lavas, especially those erupted through thick continental crust. However, such conclusions are based on the choice of partition coefficients rather than the choice of melting model. For the bulk of island arc lavas, we argue that the effect of interaction with the lower crust will usually be to reduce the extent of mantle derived disequilibria. Thus, the time scales for magma ascent and differentiation inferred from U-series isotope measurements can be treated as maxima.