Machine Learning-Aided Exploration of Ultrahard Materials

Ultrahard materials are an essential component in a wide range of industrial applications. In this work, we introduce novel machine learning (ML) features for the prediction of the elastic moduli of materials, from which the Vickers hardness can be calculated. By applying the trained ML models on a space of ∼110,000 materials, these features successfully predict the elastic moduli for a range of materials. This enables the identification of materials with high Vickers hardness, as validated by comparing the predictions against the density functional theory calculations of the moduli. We further explored the predicted moduli by examining several classes of materials with interesting mechanical properties, including binary and ternary alloys, aluminum and magnesium alloys, metal borides, carbides and nitrides, and metal hydrides. Based on our ML models, we identify a number of ultrahard compounds in the B-C and B-C-N chemical spaces and ultrahard ultralight-weight magnesium alloys Mg3Zn and Mg3Cd. We also observe the inverse of the hydrogen embrittlement effect in a number of metal carbides, where the introduction of hydrogen into metal carbides increases their hardness, and find that substitutional doping of Al in transition-metal borides can yield lighter materials without compromising the thermodynamic stability or the hardness of the material.