Just as Water changes phase, freezing into ice or evaporating into vapor, forsterite, a key mineral of Earth's mantle, under pressures of hundreds of thousands of atmospheres, reassembles its crystal lattice. This process resembles turning the faces of a Rubik's Cube: atoms shift, forming new combinations, and the minerals wadsleyite and ringwoodite are born — polymorphic modifications with the same chemical formula but different symmetry. It is these rearrangements that drive the slow convection of rocks, which, like a giant conveyor belt, moves continental plates.

The precise conditions of this transformation, the pressure and temperature at which forsterite transitions to wadsleyite or ringwoodite, have long remained elusive. Direct experiments at depths of hundreds of kilometers are impossible, and a full first-principles computer calculation would take centuries even on modern supercomputers. The solution came from machine learning: scientists created a digital twin of forsterite — a neural network model trained on a relatively small set of quantum-mechanical calculations. This digital twin predicts changes in atomic structure almost instantly, and it carefully accounts for the measure of disorder, entropy, which is critical for accurate predictions of phase transitions. This approach resembles a universal set of rules, like the Standard Model in physics, where a limited number of principles describes a wide range of phenomena.

The new phase map, built with the digital twin, showed that forsterite melts at greater depths than classical models suggested, and in different temperature regions of the mantle, its atomic pattern changes differently. This rewrites the story of the planet's thermal evolution, refining our picture of how material mixes in Earth's interior.