Yang Sun and colleagues post ECAM, a model of Earth's core alloys nearly 100 times faster than its parent

A team led by Yang Sun has posted an arXiv preprint introducing the Earth's Core Atomistic Model, or ECAM, for alloys of iron, nickel, sulfur, silicon, oxygen, carbon, and hydrogen — the mix used to represent Earth's core. The group sampled structures with a large foundational atomistic model, then labeled a compact subset, drawn from nearly half a million configurations, with high-accuracy first-principles calculations and iterative exploration. The lightweight model adds thermal electronic contributions, the electron heat capacity that ordinary interatomic potentials often leave out at core temperatures.

On the tests described, ECAM corrects artifacts in the parent model's treatment of core materials and matches earlier ab initio results for density, seismic velocity, Gibbs free energy, and phase diagrams of iron and binary alloys. The practical number is speed: nearly two orders of magnitude faster than the foundational model that proposed the candidate structures. This is a computational methods paper: the training pool is nearly 500,000 configurations labeled by first-principles calculations, and the checks reported are against prior theory for iron and binaries, not a laboratory sample of core alloy.

Earth's magnetic field is sustained by convection in molten iron alloy, and inner-core freezing rejects much of the light-element load into the melt. Seismologists record the result as density and wave speed; dynamo models need buoyancy and the melting point. Those quantities are usually computed one composition at a time, because electronic-structure methods do not scale to long runs of a seven-element melt. A faster domain-specific potential lets modelers test which mix of sulfur, silicon, oxygen, carbon, and hydrogen can meet those constraints at once. That is a geophysics question, not a new observation of the core.

The work is a preprint, so the claims have not yet passed journal peer review, and matching earlier calculations means ECAM can inherit their bias versus shock-wave or diamond-anvil data. What would strengthen the evidence is an independent equation-of-state check, plus a partitioning run whose inner-core density and shear velocity sit beside seismic reference models. If speed and binary-alloy accuracy both hold in the full seven-element liquid, core-dynamics groups could drop pure-iron proxies for multicomponent crystallization on routine computers within a few years of adoption.