image: Schematic illustration showing the integration of five key thermodynamic parameters into AI4Min-PE for real-time calculation and visualization
Credit: ©Science China Press
The Earth’s deep interior, made up of the mantle and the core, acts as the planet’s “engine,” powering the magnetic field, plate tectonics, volcanic activity, and the cycling of key elements. It also plays a vital role in shaping Earth’s resources and the evolution of life. Yet conditions inside Earth are extreme, with temperatures reaching up to 6000 K and pressures up to 350 GPa, making it extraordinarily difficult to probe how minerals and elements behave in such environments. Traditional experiments and simulations, while important, are costly and limited in scope. This has created an urgent need for faster and more efficient tools to predict chemical behavior under extreme conditions.
A Chinese joint team from Peking University and Beihang University has now developed a new artificial intelligence tool designed to tackle this challenge. Their model, called ShapKAN, combines advanced deep learning with Shapley feature selection and has been integrated into a cloud-based platform named AI4Min-PE (http://pe.ai4mineral.com). The platform allows researchers to instantly predict and visualize five critical thermodynamic parameters—including electronegativity, work function, formation energy, chemical affinity, and relative bond energy—across pressures ranging from the Earth’s surface to conditions found in the core (up to 500 GPa). Each prediction takes just milliseconds, offering speed and accuracy far beyond conventional approaches.
Beyond performance, AI4Min-PE is designed for openness and accessibility. It features an intuitive web interface and step-by-step guides, enabling users worldwide to calculate, visualize, and download results with ease.
Initial applications have shown the platform’s power for scientific discovery. For example, in subduction zone studies, the model revealed that high pressure can significantly enhance the reducing strength of transition metals like iron and manganese, helping convert CO2 and H2O into compounds such as methane and hydrogen. At the boundary between Earth’s mantle and core, the platform’s predictions suggest that light elements including sulfur, carbon, hydrogen, nitrogen, and oxygen are more likely to enter the metallic core, while silicon tends to remain in the mantle. Such findings provide new insights into the exchange of matter between Earth’s interior layers.
By combining artificial intelligence with Earth science, AI4Min-PE offers a powerful new tool for exploring the chemistry of materials under extreme conditions. It promises to accelerate discoveries not only in deep Earth research, but also in planetary evolution and the design of high-pressure materials.
Journal
Science Bulletin