Deep-Learning Potentials to Simulate Thermo-Mechanical and Physical Behavior of Entropy-Driven Diborides

The thermo-mechanical and physical properties of entropy-driven diborides have been modeled by utilizing the deep-learning (DL) algorithms. The results from ab-initio molecular dynamics (AIMD) simulations and Density Functional Theory (DFT) calculations at the ground state level including forces, energy, and virial data have been fitted into DeepMD-kit to generate the DL potentials to estimate a series of critical materials properties at ultra-high temperatures. The predictions on the properties have been found to be favorably comparable to the theoretical predictions and/or experimental results including the bulk modulus, elastic constants, cohesive and point defect energy of the diboride phases. In addition, the thermal properties including the thermal expansion coefficients and the thermal conductivity at elevated temperature can also be predicted quite well. The support from CMMI Division of NSF (Award No. 1902069) is gratefully acknowledged. We also acknowledge the computational supports from NERSC