Machine Learning Directed Search for Ultraincompressible, Superhard Materials

Currently, superhard materials with widespread commercial application (i.e. diamond, c-BN) require extreme temperatures and pressures to produce. Of interest is a separate class of superhard materials which combine transition metals with light main group elements and can be synthesized via common high-temperature metallurgical techniques. To expedite the discovery process, a machine-learning (ML) model is developed to predict bulk and shear moduli; mechanical properties which scale with hardness. From the model, a rhenium tungsten carbide and molybdenum tungsten borocarbide are selected and synthesized at ambient pressure via arc melting. Bulk modulus of each compound is determined experimentally through high-pressure diamond anvil cell measurements, supporting the ML predictions with less than 10% error. Vickers hardness is measured, indicating each composition surpasses the superhard threshold of H_V = 40 GPa at low loads (0.49 N). Furthermore, DFT calculations are employed on compositions of intermediate predicted hardness to corroborate the ML model. These results demonstrate the promise of machine-learning techniques for the identification of novel materials with desirable mechanical properties.