Metal ceramics under AI control: A new approach for calculating the mechanical properties of materials

Researchers from Skoltech Materials Center have developed a novel approach to modeling the mechanical properties of complex materials using machine learning and active learning on local chemical configurations. This method allows for accurate calculations of large systems containing tens of thousands of atoms, comparable to direct quantum-mechanical calculations, but with significantly reduced computational requirements.

The study, published in Computational Materials Science, introduces moment tensor potentials (MTPs) which are trained on data obtained from density functional theory (DFT) calculations. During simulations, the system identifies and extracts local atomic environments where MTP energy predictions become unreliable, sending these fragments for additional DFT calculations.

These fragments are then integrated into the training set, progressively expanding the potential's domain of applicability. The approach was applied to WC-Co composites, demonstrating its ability to model large polycrystalline and composite systems, including defects and grain boundaries, with near-DFT accuracy while maintaining quantum-mechanical precision.

This method is applicable to a wide range of heterogeneous materials, providing engineers and materials scientists with a powerful tool for predictive calculation of material properties without the need for expensive experiments.