AF: Small: Parallel Methods for Large, Atomic-scale Quantitative Analysis of Materials

材料科学的实验进展已经使材料的近原子尺度成像成为可能，使原子排列与宏观材料特性相关，并使智能材料设计成为可能。由于材料科学界面临着快速采用这种新技术的需求，拟议研究的目标是为原子尺度数据的定量分析开发全面的算法基础，从解释原子探针断层扫描图像到分析和特征提取。并行算法和高性能的实现将被强调，由于大数据集的大小，可以达到数亿到十亿个原子和超越。研究人员将开发并行算法，以1）处理原子采样中的固有限制和错误，2）确定晶体学方向，3）执行基于自相关的聚类分析，以及4）通过线性和非线性流形学习提取相干结构和内在几何特征。 这项研究将由一个跨学科的PI团队进行，该团队具有并行算法，高性能计算，计算和微分几何以及材料科学方面的专业知识。该研究旨在推动材料科学从原子尺度固态现象的可视化和模拟到直接观察，定量分析和解释。

Parallel Methods for Large, Atomic-scale Quantitative Analysis of MaterialsProject SummaryExperimental advances in materials science have enabled near atomic scale imaging of materials, making it possible to relate atomic arrangements to macroscopic material properties and enabling smart materials design. Driven by the needs of the materials science community facing a rapid adoption of this new technology, the goal of the proposed research is to develop comprehensive algorithmic foundations for quantitative analysis of atomic-scale data from interpreting atom probe tomography images to their analysis and feature extraction. Parallel algorithms and high performance implementations will be emphasized due to the large data set sizes which can reach several hundred million to a billion atoms and beyond. The investigators will develop parallel algorithms to 1) deal with inherent limitations and errors in atomic sampling, 2) determine crystallographic orientation, 3) perform autocorrelation based clustering analysis, and 4) extract coherent structures and intrinsic geometric features through linear and non-linear manifold learning. The research will be carried out by an interdisciplinary team of PIs with expertise in parallel algorithms, high performance computing, computational and differential geometry, and materials science. The research is intended to advance materials science from visualization and simulation of atomistic scale solid state phenomena to its direct observation, quantitative analysis and interpretation.