Figure 3

Evidence of excess water storage in grain boundaries. (a) Interface density versus H₂O content in grain boundaries, such as estimated from the measured amount of H₂O (black points) with error bar (red solid lines) minus the 150 ppm stored in the olivine crystal structure²⁶. Using 42 SIMS analyses (located in Fig. 1b), the linear best-fit correlation of our data points (dotted black line) indicates a coefficient (R²) of 0.3. For comparison, we give calculations of the maximum amount of H₂O that grain boundaries can contain within a volume of 10 × 10 × 4 µm³, and considering boundaries between 0.8 and 1 nm thick (grey area). An estimation is also given for a boundary thickness of 2 nm (dotted grey line). These estimations are based on grain boundary densities calculated on 20 × 20 µm² EBSD maps (see Supplementary Fig. S5), and then extrapolated in three dimensions (3-D) through multiplication by 4/π²⁷. The green dotted rectangle shows the range of density and H₂O content documented in the high-strain zone. (b) H₂O content in grain boundaries versus strain, here depicted as the orthogonal distance from the high-strain zone. In this case, the best-fit correlation (dotted line) indicates a R² of 0.59. We also give the bulk H₂O content, which corresponds to the amount of water added before experiment (1330 wt. ppm) minus the loss of H₂O during the experiment (550 wt. ppm deduced from a hot-pressing sample; see Supplementary Fig. S1) and H₂O stored in crystals structure (150 wt. ppm)²⁶. This highlights a deficit of water in the low-strain zone and excess in the high-strain one (see text). (c) Strain (i.e., orthogonal distance from the high-strain zone) versus H₂O content per unit of grain boundary, as estimated through normalization of the values in (a) by their respective 3-D interface density. The best-fit correlation (dotted line) indicates that H₂O increases with strain with a coefficient R² of 0.58. This coefficient excludes a few anomalies (3) located far above the first-order trend.