Grain boundary segregation for the Fe-P system: Insights from atomistic modeling and Bayesian inference

In this work we re-assess experimental data for grain boundary (GB) segregation of P in bcc Fe with thermodynamic and statistical methods. The data are based on Auger-Electron-Spectroscopy (AES) measurements which have provided P GB concentrations for various bulk contents and temperatures for ferrite. While in the previous investigations of this system a single-site McLean equation was used to extract segregation enthalpy and entropy, we employ multi-site segregation models as suggested by atomistic simulations. We use a recently introduced methodology based on inter-atomic potentials to calculate the segregation energy spectrum, as well as vibrational entropy and solute-solute interactions of P in polycrystalline Fe, thereby obtaining a three-dimensional distribution of energy, entropy and interactions. Using this trivariate distribution we predict P GB concentrations and compare them to the experimental measurements. Furthermore, we calibrate the parameters of physical models of increasing complexity to the experimental data with an inverse modeling approach based on Bayesian inference. While it is not possible to seamlessly link the results from AES to atomistic modeling, our investigation provides significant new insights for thermodynamic modeling of GB segregation. The vibrational entropy deduced from experiment differs from physics based atomistic simulations even when taking the spectral nature of energy, entropy and interactions into account. However, the correlations of the quantities are in good agreement when considering the trivariate model. Our investigation highlights the need for new and more accurate experimental datasets of GB segregation.