InfPolyn, a Nonparametric Bayesian Characterization for Composition-Dependent Interdiffusion Coefficients

Composition-dependent interdiffusion coefficients are key parameters in many physical processes. However, finding such coefficients for a system with few components is challenging due to the underdetermination of the governing diffusion equations, the lack of data in practice, and the unknown parame...

Full description

Saved in:

Bibliographic Details
Published in:	Materials 2021-06, Vol.14 (13), p.3635
Main Authors:	Xing, Wei W., Cheng, Ming, Cheng, Kaiming, Zhang, Wei, Wang, Peng
Format:	Article
Language:	eng
Subjects:	Alloy solidification Boltzmann–Matano analysis Coefficients Composition Efficiency Gaussian process Interdiffusion interdiffusion coefficient Mathematical models Methods nonparametric Bayesian Polynomials Process parameters Quaternary systems Simulation Statistical methods
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Composition-dependent interdiffusion coefficients are key parameters in many physical processes. However, finding such coefficients for a system with few components is challenging due to the underdetermination of the governing diffusion equations, the lack of data in practice, and the unknown parametric form of the interdiffusion coefficients. In this work, we propose InfPolyn, Infinite Polynomial, a novel statistical framework to characterize the component-dependent interdiffusion coefficients. Our model is a generalization of the commonly used polynomial fitting method with extended model capacity and flexibility and it is combined with the numerical inversion-based Boltzmann–Matano method for the interdiffusion coefficient estimations. We assess InfPolyn on ternary and quaternary systems with predefined polynomial, exponential, and sinusoidal interdiffusion coefficients. The experiments show that InfPolyn outperforms the competitors, the SOTA numerical inversion-based Boltzmann–Matano methods, with a large margin in terms of relative error (10× more accurate). Its performance is also consistent and stable, whereas the number of samples required remains small.
ISSN:	1996-1944 1996-1944