Pressure-induced changes in properties of multicomponent silicate melts in magma oceans controlled chemical differentiation of the silicate earth and the composition of partial melts that might have formed hidden reservoirs. Although melt properties show complex pressure dependences, the melt structures at high pressure and the atomistic origins of these changes are largely unknown because of their complex pressure-composition dependence, intrinsic to multicomponent magmatic melts. Chemical

more »

... traints such as the nonbridging oxygen (NBO) content at 1 atm, rather than the structural parameters for melt polymerization, are commonly used to account for pressure-induced changes in the melt properties. Here, we show that the pressure-induced NBO fraction in diverse silicate melts show a simple and general trend where all the reported experimental NBO fractions at high pressure converge into a single decaying function. The pressure-induced changes in the NBO fraction account for and predict the silica content, nonlinear variations in entropy, and the transport properties of silicate melts in Earth's mantle. The melt properties at high pressure are largely different from what can be predicted for silicate melts with a fixed NBO fraction at 1 atm. The current results with simplicity in melt polymerization at high pressure provide a molecular link to the chemical differentiation, possibly missing Si content in primary mantle through formation of hidden Si-rich mantle reservoirs. E arly in Earth history, during the magma-ocean phase, the chemical differentiation of the silicate earth, formation of core, and evolution of atmosphere were controlled by the properties of silicate melts at high pressure (1-6). Pioneering experimental studies show that these thermodynamic and transport properties relevant to the chemical evolution of the Earth vary nonlinearly with changes in pressure (7-9). For instance, the solubility of Ar into melts increases with increasing pressure and then decreases drastically with a further increase in pressure, with data suggesting a strong composition dependence (7, 10, 11). Similarly, complex behaviors were reported for the diffusivity and viscosity of silicate melts at high pressure (8). Although the trend in the silica activity in the melts at high pressure is not known, phase relations of mantle melts and minerals imply varying activity coefficients of the oxide in silicate melts with changes in pressure (12-14). Changes of up to two orders of magnitude in the element partitioning coefficient between melts and crystal/coexisting phases have been reported stemming mostly from the effect of the melt composition, constraining the fate of radioactive nuclides in the Earth's interior (15) (16) (17) (18) . The key to understanding these nonlinear changes in melt properties with pressure is the melt structure at high-pressure in a short-(e.g., coordination number) to medium-range scale (19) (20) (21) . While recent progress in mineral physics provides the link between the macroscopic properties and the structures of the crystals, the nature of changes in the melt structure at high pressures, such as those deep within the magma-ocean, remain poorly constrained as detailed knowledge about the structure of melts cannot be determined based on their compositions alone. Even more challenging is to unveil the structure of "multicomponent," and hence, natural silicate melts in the earth's interior (22). Most of the previous studies focused on the pressure-induced bonding transition in simple model melts (e.g., from single component, to ternary) (e.g., refs. 3, 23-26) and references therein). NMR spectra of simple melt compositions are subjected to less inhomogeneous broadening due to a relatively small number of melt structural components: For the quaternary oxide glasses, the expected number of binary correlations is up to 16; inhomogeneous broadening associated with such complexity obscures the otherwise useful structural information such as coordination number and degree of melt polymerizations. Although the degree of polymerization in melts originally describes melt structures, the mole fraction of nonbridging oxygen [X NBO , NBO∕ðNBO þ bridging oxygen;BOÞ] at 1 atm can be calculated from the chemical composition of melts. Therefore, X NBO is often regarded as a chemical constraint from which other properties of melt structure are predicted. However, the X NBO at high pressure varies with pressure with composition dependence (the Si∕Al ratio, fractions of alkali content, and types of networkmodifying cations). The systematic relation between X NBO at high pressure and melt composition has not been available, limiting its usefulness in modeling the melt properties at high pressure. The simple predictive NBO fraction in the melt, if available, could be useful to yield the microscopic origins of melt properties. The advent of high-resolution, element-specific, multinuclear NMR techniques such as triple quantum magic angle spinning (3QMAS) NMR allows us to obtain previously unknown details of the pressure-induced structural changes in multicomponent melts at high pressure (23, (27) (28) (29) . Quaternary Ca-Na aluminosilicate (CNAS) melts is a model system for slab-driven magmatic melts and midocean ridge basalts (MORBs) composition melts in the Earth's interior (30, 31) and provides insights into the structure of complex primordial melts and magmatic reservoirs. Results and Discussion The remarkable resolution among atomic configurations in the quaternary CNAS glasses quenched from melts at high pressure are shown in the multicomponent (Al-27, Na-23, O-17), twodimensional 3QMAS NMR spectra (Fig. 1) . The Al-27 NMR spectra for the CNAS melts (Fig. 1, Left) show ½4 Al, ½5 Al, and ½6 Al at 8 GPa, whereas ½4 Al is dominant (approximately 100%) at 1 atm (32). The fractions for ½4 Al, ½5 Al, and ½6 Al are approximately 76.1%, 16.7%, and 7.1%, respectively (see SI Appendix). The peak width of the ½4 Al for 8-GPa glasses in the MAS dimension is larger than that for 1 atm, suggesting an increase in the topological disorder due to the Al-O bond length distribution with pressure.