A robust and intelligent computational framework for assessing diffusion coefficients in alloys

Abstract

In order to determine the diffusion coefficients for alloys accurately, high-quality thermodynamic and kinetic databases are of essential prerequisites. Under the premise of ensuring the evaluation accuracy, the fewer the parameters being optimized and stored in databases, the higher the quality of the databases for avoiding over-fitting and achieving reliable extrapolations. So far, no general methods and criteria to obtain high-quality kinetic databases with the fewest parameters to be optimized have been proposed. To remedy this situation, we proposed a robust computational framework including Levenberg-Marquardt algorithm and χ2 test to obtain accurate diffusion coefficients with the fewest mobility parameters for single-phase alloys. Benchmark tests for Cu-Mn, Ag-Mg, Ni-Cu, Ni-Ti and Cu-Ti binary fcc phases as well as Ni-Cr-V, Cu-Ni-Sn and Ni-Ti-Cu ternary fcc phases were conducted. According to the tests, general criteria for selecting mobility parameters of both binary and ternary interactions to be stored in kinetic databases were proposed for the first time. The present calculations indicate that this computational framework can work well for both binary and ternary systems. We also compared the diffusion coefficients evaluated in the present work with those through the widely utilized DICTRA software and recently developed Z-Z-Z model. It was found that this framework can better reproduce the diffusion coefficients even with fewer mobility parameters, demonstrating higher accuracy and efficiency. This computational framework is incorporated in our recently developed CALTPP (CALculation of ThermoPhysical Properties) program. We guarantee that the present computational framework will contribute to establish not only kinetic databases but also other property databases with a high quality for the sake of material design.