The Deep Rubik's Cube: How Minerals Rearrange Under Pressure ✎ express

Original: "Machine learning determines the Mg2SiO4 P-T phase diagram"
arXiv:2602.01730 · 2026-02-02 · arXiv non-exclusive · independent analysis · 1 min
A digital Rubik's Cube solver instantly calculates how the mineral forsterite changes its pattern under pressure and when it melts.

📄 Showing the "Simple" version — "Advanced" is not ready yet. Add it to favorites to help prioritize it.

Just as water freezes into ice or evaporates, forsterite in Earth's mantle, under pressures of thousands of atmospheres, reassembles its atomic pattern — like a Rubik's Cube whose faces rearrange into new combinations. The minerals wadsleyite and ringwoodite are born, and these rearrangements fuel the slow churning of rocks that drives the continents.

The precise conditions for this transformation have long remained elusive: deep-earth experiments are impossible, and a direct computer simulation would take centuries. The solution is a digital twin of forsterite, trained on a modest dataset and capable of predicting changes instantly. Crucially, it carefully accounts for the measure of disorder — entropy — critical for accuracy. This approach resembles a universal rulebook, akin to the Standard Model in physics.

Back in the 1960s, Lev Landau explained how substances abruptly change symmetry — like a Rubik's Cube where each turn creates a new pattern.

The new phase map revealed that forsterite melts deeper than previously thought, and in different regions of the mantle, its pattern changes differently. This rewrites the story of our planet's evolution.

The most unexpected pattern — ringwoodite — was found trapped inside a diamond in 2014, like a secret design on the face of a giant cube.

In its pure form, forsterite is the gemstone peridot. Its deep-earth siblings, ringwoodite and wadsleyite, hide inside diamonds like secret patterns.

Fictional journeys to the center of the Earth now have a scientific map: we know what strange mineral patterns will be encountered along the way.

Scientists
Ludwig BoltzmannRudolf ClausiusJacob BekensteinStephen HawkingJames Clerk MaxwellEdward Witten
Concepts
olivine Machine Learning machine learned interatomic potential training Michaelis-Menten equation transfer learning transparency mantle Reinforcement Learning neural network physics modeling