Our experiments show that water reacts with Fe-Si alloy at the pressure, temperature, and redox conditions relevant for the Earth’s deep interior. The reaction hydrogenates the Fe liquid while oxidizing Si in the Fe-Si liquid to form silica (Fig. 1). In this reaction, the amount of H2O required for oxidizing Si can be constrained through a redox reaction: e.g., Si0 (metal) + 2H2O → 4H0 (metal) + SiO2. Furthermore, the amount of H alloyed with Fe metal can be estimated.
Studies have suggested that the bottom of early Earth’s magma ocean could have reached pressures up to 60 GPa39. Our experiments up to 60 GPa and 3650 K could therefore be related to the bottom of the deep magma ocean where iron metal diapirs are expected to pond before sinking through the solidified silicate part to merge with the growing core. If water existed in the magma ocean of the early Earth (Text S4), our experiments show that it would have reacted with liquid metallic Fe and oxidized Si to silica. The amount of water estimated to be in the early magma ocean ranges from a few hundred ppm to 1.8 wt% H2O40,41. If 1 wt% H2O existed in the magma ocean and most of the water reacted with Fe-Si liquid, the amount of hydrogen that could be incorporated into the core would be ca. 6.5 x 1021 kg, which is equivalent to ca. 0.34 wt% of hydrogen in the present-day Earth’s core. This value is consistent with an estimation based on partitioning of H between Fe metal and silicate liquid at high pressure7.
Although hydrogen could have been incorporated into the core during early processes, it is also feasible that the subducting slabs have supplied a significant amount of hydrogen (or water) to the CMB region over an extended period (Fig. 4a). Recent high-pressure studies have found some hydrous minerals to be stable in the subducting slabs in the lower mantle, although some water loss at the 660-km discontinuity seems to be inevitable due to storage capacity changes in mineral assemblage42-44. While there are many factors to be considered for the estimation of the amount of water delivered to the CMB (and participated into the reaction), the fraction of the total subducted surface water transported to the CMB and its duration would be important (Fig. 4b). Although these two parameters are uncertain, if 1% of the total subducted water i.e., an annual input of 1012 kgref.12 can be delivered to the core, approximately 1018 kg of hydrogen could have been supplied over the period of modern-style subduction since 2.5 Gya. For a lower efficiency of 0.1% transport to the CMB, the total mass of delivered hydrogen would be 1017 kg. If the water delivered to the CMB is released by dehydration and induces the reaction we observed here, H could be incorporated into the outer core while Si in the outer core could be oxidized (i.e., SiO2) and released to the mantle (Fig. 4a). Silica from the core may not exist as a free phase at the CMB region. Instead, it can react with ferropericlase to form bridgmanite or post-perovskite as observed in our experiments (Fig. 1c).
Because hydrogen has a relatively large effect on the density of iron alloys, the above process could create chemical stratification at the top of the core (Fig. 4a). Flows associated with thermal (or thermochemical) convection in the rest of the outer core are not strong enough to erode such a stable layer, even over geologic time45. While mixing hydrogen throughout the entire core would require a huge amount of energy to counteract its buoyancy, segregating hydrogen in a stable layer does not challenge the bulk energetics of the geodynamo. However, a stable layer could affect the structure of the magnetic field observed at the surface and appear in seismic data. Indeed, several studies support the existence of such a stable layer based on anomalously slow seismic wave speeds at the top of the core22,46 and various geomagnetic observations (e.g., refs.47-49). However, these studies predict different properties for the stable layer (e.g., thicknesses between 60 and 400 km with varying degrees of stability). Other works argue that a stable layer is not required to explain the available observations, but also cannot be excluded (e.g., refs.23,50).
To assess the range of possibilities, we computed the Brunt-Väisälä period (TBV) of a layer with an assumed thickness and extra mass of hydrogen to assess its stability (see Method). Lower values of TBV represent stronger stratification. Although Si is removed from the core, the resulting metallic liquid is still buoyant with respect to the bulk core because H is added51. Transferring ca. 2.7 x 1018 kg of H into the core (and removing the associated amount of Si) could produce a stable layer with a thickness of 400 km and TBV ~ 48 hours, in agreement with ref.49. Even less H (ca. 1.2 x 1018 kg) is required to produce a stable layer with a thickness of ca. 130 km and TBV ~ 24 hours as estimated by ref.52 (Fig. 4c). The amount of hydrogen required to explain these existing models is well within the range we estimated above between 0.4 to 1% of the subducted water (Fig. 4b). Producing the most stable and thick layer (i.e., ca. 300 km thick, TBV ~ 1.6–3.4 hours) compatible with ref.46 with this process alone, however, would need at least ca. 4.0 x 1020 kg of H, which would require a deep water cycle with an unrealistic efficiency of more than ~90%.
Over a few decades, attempts have been made to explain the Eʹ layer with light element enrichment. However, as pointed out by ref.27, no single light elements can explain the low density and low velocity required for the observed stable Eʹ layer at the topmost core simultaneously. While some combination of light element addition and subtraction can provide explanation, it has been unknown what process can drive such an exchange of light elements.
In the case we present here, the core-mantle exchange process would add H while subtracting Si at the topmost outer core. According to ref.53, an increase in H by ca. 0.04 wt% in the Eʹ layer (i.e., a top region of outer core) would increase the velocity by ca. 0.2% at 150 GPa and 4000 K. Using parameters in ref.5, if ca. 0.28 wt% of Si has been removed by 0.04 wt% H addition (see Method, Text S5), the velocity would decrease by ca. 0.72%. Combining these two effects, ca. 0.52% decrease in velocity is expected from the core-mantle chemical exchange we found in our experiments. Such a velocity decrease is within the observed values from seismology, i.e., 0.2–0.7% decrease22,46,54,55. In addition to the exchange process we used in our model, there could be contributions from other processes, such as chemical exchange between oxygen and silicon (e.g., refs.26,56), sub-adiabatic heat flow across the CMB (e.g., ref.57), and lateral variations in the heat flux (e.g., ref.58). Furthermore, input of H from the mantle may induce stratification by immiscible liquids such as Fe-H and Fe-S59. If more than a single factor can contribute, the required amount of H2O for explaining the E’ layer by the chemical exchange could further decrease.
As shown above, the chemical exchange process promoted by subducted water at the CMB could explain the seismic properties and at the same time satisfy the lower density requirement for the dynamic stability of the Eʹ layer. While more studies are desirable, our study reveals the intriguing possibility that the Eʹ layer is the consequence of deep transport of subducted surface water over giga years of deep hydrogen cycle. In addition, our model presented here shows that the mantle and the core systems are not chemically separated over geological time scale, but they may have exchanged some elements even after their differentiations since the early processes.