AI Sheds New Light on Earth’s Inner Core

ICTP scientists Zhi Li and Sandro Scandolo have applied AI methods to study the microscopic structure of the Earth’s inner core, investigating the role played by silicon—one of the main light elements in the core’s iron alloy.  Their results show that silicon profoundly affects the way atoms arrange under extreme pressure and temperature conditions. According to their study, silicon stabilizes the alloy into a cubic, rather than hexagonal structure, a finding that could explain why seismic waves travel only very slowly through the inner core.

Located more than 5,000 kilometers beneath the Earth's surface, where temperatures exceed 6,000 degrees Celsius and pressure is more than three million times the standard atmospheric pressure, the planet's inner core cannot be observed directly and numerical simulations have been key to access its secrets. Most of the studies done until now have only focused on pure iron, showing that in such extreme conditions it crystallizes into a hexagonal structure.

Observations, however, have revealed that seismic waves propagate through the Earth's core much more slowly than they would in a hexagonal lattice. Li and Scandolo’s results could finally tell us why. “Many hypotheses have been made so far,” says Scandolo, “Our results predict that the seismic wave velocity in the iron-silicon alloy closely matches observed values, strongly suggesting that the atoms in the inner core form a cubic structure.”  

The study, which was published in Nature Communications, is the first to consider the role played by light elements such as silicon—which is about half as heavy as iron—in the inner core, something that has only been made possible by AI. This is because assessing the role played by silicon atoms in the alloy requires considering all the many random arrangements in which they can be found. Such a large number of configurations is simply impossible to simulate with traditional methods. The AI algorithms used by Li and Scandolo, instead, are trained to efficiently compute the forces between each pair of atoms, and can therefore deal with larger and more complex systems, over longer time scales.

The study, which required large computational resources to train the AI algorithms and run the simulations, was made possible by a special early access to Leonardo, a recently established GPU-based facility at CINECA, which at the time, in 2023, was the fourth fastest computer in the world. It is one of the projects funded by the ICSC - Italian Research Center on High Performance Computing, Big Data and Quantum Computing, one of the five Italian national centres established by the Italian National Recovery and Resilience Plan (NRRP).