Efficient sampling and representation of the grain boundary geometry and composition space: atomistic simulation meets statistical methodology

现代优化材料性能的尝试越来越多地探索基于数据的研究方法。因此，例如，在多组分体系中，需要大量关于晶界结构和能量的数据，以用于微结构的开发。金属和合金显微组织中的晶界对材料的力学性能和功能性能有着重要的影响。它们可以通过溶质偏析来操纵，从而导致晶界能、流动性、结构和内聚力的变化。在这个项目中，我们通过建立一个选择性和高效的高通量计算框架，通过原子模拟来确定这样的晶界能，从而推进基于数据的材料科学。我们将开发必要的算法，利用“实验设计”的方法和最近的发展，在建模晶界分布函数。与传统的高通量计算相比，新的实验技术设计与非参数估计相结合，可以识别和评估晶界参数空间最关键的区域，并通过一系列超球谐函数表达晶界能量。由此产生的计划可以用来提供晶界能量作为一个功能的几何形状和组成作为输入的热力学和动力学，以及微观力学建模。此外，这些见解将有助于更有效地规划模拟和实验，以研究其他多维非线性关系。

Modern attempts to optimise materials’ properties increasingly explore data based research approaches. As a consequence, there is for instance a demand of extensive data on structure and energies of grain boundaries in multicomponent systems for microstructure development. Grain boundaries in microstructures of metals and alloys have a significant influence on the material’s mechanical and functional properties. They can be manipulated via solute segregation, leading to changes in grain boundary energy, mobility, structure, and cohesion. In this project we advance data based materials science by establishing a selective and efficient high-throughput computational framework to determine such grain boundary energies via atomistic simulations. We will develop the necessary algorithms by making use of “design of experiment” approaches and recent developments in modelling grain boundary distribution functions. In contrast to conventional high-throughput calculations, the new design of experiment techniques in combination with nonparametric estimation leads to an identification and assessment of the most critical regions of the grain boundary parameter space and an expression of the grain boundary energy by a series of hyperspherical harmonics. The resulting scheme can be used to provide grain boundary energies as a function geometry and composition as input for thermodynamic and kinetic, as well as micromechanical modelling. Furthermore, the insights will contribute to a more efficient planning of simulations and experiments for investigations of other multi-dimensional, non-linear relationships as well.