The Earth’s magnetic field has existed for at least 3.5 billion years, according to the paleomagnetic rock record. But the mechanism of the early field has been the subject of hot debate in the Earth science community. Judging by the high thermal conductivity of the Earth's core, the temperature of the core must have been too high in the past to sustain permanent magnetism, according to recent findings.

Therefore, another mechanism must exist within the Earth's core that continually generates the geomagnetic field in addition to convection in the Earth's fluid outer core, which surrounds its solid inner core. The convection in the outer core, we now know, is driven by both thermal and compositional buoyancy sources at the inner core boundary. These are produced as the Earth slowly cools and iron in the iron-rich fluid alloy solidifies onto the inner core, giving off latent heat and the light constituent of the alloy. These buoyancy forces cause fluid to rise, inducing convection.

This previously accepted scenario as the origin of the magnetic field of the Earth has recently been challenged by (1) the upward revision of the thermal conductivity of the core, which implies an inner-core nucleation younger than 1 billion years and (2) a total secular cooling of the core in excess of 1,000 K to drive thermal convection before about 1 billion years ago, which is required to explain palaeomagnetic intensity measurements.

The Earth’s core is about 10% less dense than pure iron, which suggests that it contains light elements as well as iron. Modeling of core formation at high pressure (around 40–60 gigapascals) and high temperature (about 3,500 kelvin) in a deep magma ocean predicts that both silicon and oxygen are among the impurities in the liquid outer core. However, only the binary systems Fe-Si and Fe-O have been studied in detail at high pressures, and little is known about the compositional evolution of the Fe–Si–O ternary alloy under core conditions.

A team at the Earth-Life Science Institute of the Tokyo Institute of Technology, led by Prof. Kei Hirose, performed melting experiments on liquid Fe–Si–O alloy at core pressures in a laser-heated diamond-anvil cell. Their results demonstrated that the liquidus field of silicon dioxide is unexpectedly wide at the iron-rich portion of the Fe–Si–O ternary, such that an initial Fe–Si–O core crystallizes SiO2 as it cools. If crystallization proceeds on top of the core, the buoyancy released should have been more than sufficient to power core convection and a dynamo, in spite of high thermal conductivity, from as early as the Hadean eon, more than 4 billion years ago, when the Earth first started to form. The finding also provides insights into the makeup of the inside of the Earth – the SiO2 saturation sets limits on silicon and oxygen concentrations in the present-day outer core.

Recent core-formation models propose that the present outer core contains 2 wt% Si and 5 wt% O by assuming relatively oxidizing conditions, or 8-9 wt% Si and 2-4 wt% O by assuming relatively reducing conditions. Hirose’s team examined the fate of a cooling liquid Fe–Si–O alloy spanning a similar range of compositions using a combination of high-pressure, high-temperature experiments in a laser-heated diamond-anvil cell (DAC) and textural/chemical characterizations of recovered samples. Beginning with an ultrafine-grained homogeneous mixture, after several seconds of heating they obtained a pool of quenched molten iron alloy in contact with solid SiO2 at the cooler portion in all runs. Owing to an intrinsic temperature gradient in the DAC, both melting and crystallization occurred simultaneously in the laser-heated sample – melting in its hotter portion and crystallization in its cooler portion.

Disequilibrium caused by the Soret effect (by which different particle types exhibit different responses to the force of a temperature gradient) is often observed in high-temperature experiments on silicates, and typically enriches silicon and iron in hot and cool regions, respectively. However, the team’s observations show the opposite distribution, while the homogeneity of the melted portion — as expected from the fast diffusion of Si and O in liquid Fe — and its physical contact with the solid SiO2 imply that conditions were not far from equilibrium. From these observations, the researchers inferred that solid SiO2 crystallizes from an Fe–Si–O ternary alloy, and in the Earth’s core this may be a gradual process coupled to secular cooling that exerts an important influence on core composition, dynamics and the geodynamo.

Four of the experiments, with three different starting compositions, were performed at uppermost-core pressures (133-145 GPa) and temperatures of about 4,000 K. The composition of molten iron found in run 1 was Fe–0.6 wt% Si–1.3 wt% O (starting composition Fe–3.8 wt% Si–4.4 wt% O), indicating that liquid Fe–Si–O alloy crystallized SiO2 oxide until most of the silicon and oxygen were consumed from the liquid. The temperature at the liquid/solid boundary was estimated to be 3,990 ± 120 K at 142 GPa, close to the melting temperature of pure iron. In run 4, they obtained liquid Fe–5.8 wt% Si–0.5 wt% O from silicon-rich Fe–Si–O starting material, showing that silicon-rich liquid Fe–Si–O crystallized SiO2 oxide until the supply of oxygen in the metallic melt was exhausted. Similarly in run 7, oxygen-rich Fe–Si–O liquid evolved to nearly silicon-free liquid Fe–0.4 wt % Si–4.4 wt % O via crystallization of SiO2. These results are consistent with crystallization of SiO2 from an Fe–Si–O metallic liquid until one or both of silicon and oxygen is nearly depleted.

The liquidus field of SiO2 (the compositional range in which SiO2 crystallizes first) occupies a large portion of the Fe–Si–O ternary system at uppermost core pressures (Fig. 2a). This is consistent with the phase diagram in the FeO–SiO2 binary system, which exhibits a eutectic very close to the FeO endmember (>99.4 wt% FeO). These results are also consistent with earlier reports on metal-silicate partitioning of Si and O. Using this data, they developed a thermodynamical model of SiO2 saturation in liquid Fe. Depending on predictions for the original core composition, SiO2 crystallization near core-mantle boundary (CMB) pressure would begin after cooling to 4,500–5,200 K (note that the temperature of core metal increases during descent, for example from 3,400 K at 54 GPa to 4,725–5,724 K at 135 GPa with 10%–25% gravitational potential energy release).

Thus, if the present-day CMB temperature is at 4,000 K, the current outer core has been subjected to SiO2 crystallization and is likely to be depleted in either Si (<0.7 wt%) or O (<1.0 wt%). Both Si and O could be depleted if the initial core had a molar ratio of Si:O close to 1:2 (that is, stoichiometric SiO2).

New constraints

These results impose new constraints upon the composition of the Earth’s core. In particular, the outer core cannot include both Si and O at concentrations higher than is permitted by SiO2 saturation. Some recent calculations suggest that the density and seismic velocity of the outer core are best explained by liquid Fe-1.9 wt% Si-3.7 wt% O, but this is ruled out by the team’s experiments, unless the CMB temperature is higher than 4,400 K. Other proposed Si-rich initial core compositions are unlikely to be consistent with inner-core properties because they evolve to the Si-rich side of the Fe-FeSi eutectic via the crystallization of SiO2 and then form CsCl-type Fe–Si alloy.

Because their experiments were conducted at shallow core conditions, the uncertainties are too large to constrain whether SiO2 saturation increases or decreases along adiabats at pressures deep inside the core. So they cannot reliably predict whether SiO2 crystallization from an initially well-mixed isentropic liquid Fe–Si–O core would begin from the top or the center. The slopes of adiabatic saturation curves suggest that crystallization from the center might be preferred. Equilibrium crystallization of SiO2 driven by secular cooling that begins from the center would produce an outwardly expanding (Si + O)-depleted region, as buoyant SiO2 is expelled to shallower depths where it may dissolve into under-saturated fluid.

However, if the residual liquid follows SiO2 saturation, then it will be progressively more depleted in Si + O (and hence more dense) with increasing depth. Unless this residual gravitationally stable density stratification were to be mixed away (for which there is at present no proposed mechanism), the advance of the SiO2 crystallization front to the CMB would produce an outer core that is not well mixed and isentropic. This is contrary to geomagnetic and seismic inferences.

This model also permits an alternative scenario, in which crystallization proceeds at shallower depths first, with buoyant SiO2 accumulating on the top of the core and producing a denser (Si + O)-depleted liquid that would sink into the deeper core, thus helping to drive core convection. In this scenario, SiO2 crystallization would continue today at the CMB, while the progressively (Si + O)-depleted outer-core liquid would crystallize solid Fe-rich metal at the inner-core boundary.

Before the growth of the inner core (which provides a compositional buoyancy source to drive core convection at present), the team estimates that crystallization of SiO2 on top of the core allows for a secular core cooling rate 3.88 times smaller than is required for unassisted thermal convection alone, and the requisite CMB heat flow to sustain the dynamo is 2.75 times smaller. Even if we assume a relatively high core thermal conductivity (exceeding 90 W m−1 K−1), as proposed in recent studies, a geodynamo with ohmic dissipation of 1 TW could be maintained by core convection while crystallizing SiO2 and cooling at a modest rate of about 100 K Gyr−1.

Using their thermodynamical model, the team estimates that secular core cooling of around 100 K at saturation would crystallize about 0.4 wt% SiO to produce a pure layer of SiO2 that is about 2km thick at the CMB. The fate of this SiO2 depends on the nature of the base of the mantle. If the mantle is extensively molten, then SiO₂ cumulates would react with the magma and promote crystallization of (Mg,Fe)SiO3-bridgmanite, because it is the expected liquidus phase over a wide range of pressures and melt compositions in the deep mantle. If there is no basal melt and SiO2 underplates the base of the solid mantle, it could initially form a dense mush containing liquid iron in the solid interstices, similar to a previously proposed scenario. Such a mixture would undergo viscous compaction as the liquid is expelled and may form a high electrical conductivity layer at the CMB if it retains sufficient interstitial metal to form an interconnected network. In this latter case, the SiO2 mushy cumulates would be subjected to viscous coupling with solid basal mantle flows, and swept toward convergent upwelling zones, possibly consistent with seismically observed ultralow-velocity zones, depending on the thickness and residual iron liquid fraction.

The ELSI team’s experiments show that there is no need to invoke singular events to resolve the enigma of powering a long-running dynamo. Rather than requiring core formation at extremely high temperatures (exceeding 5,000 K) to introduce a certain amount of Mg into the core, sufficient Si + O can be incorporated via the metal-segregation processes in a deep magma ocean at moderate temperatures. Thus, the model is potentially as applicable to planetary dynamos elsewhere in the Solar System as to the dynamo operating in the Earth.