ECCS-CDS&E: High-performance computational framework for real-time phase retrieval X-Ray imaging

现代科学技术越来越依赖于利用纳米级组成的独特性质的材料和器件。能够表征这种复杂的纳米级结构是至关重要的。表征纳米材料和器件的独特工具是x射线成像，它可以提供具有高空间和时间分辨率的结构图像。在典型的x射线成像装置中，x射线源产生从结构散射的x射线束。散射的x射线束由相机捕获，并对得到的图像进行处理以重建实际的结构构型。由于波长短，x射线可以穿透结构的体积，因此不仅可以进行二维成像，还可以进行三维成像。x射线成像通常需要大量的计算工作来进行图像重建，并且计算时间通常比执行实验的时间长。该项目旨在通过创建用于实时x射线成像的高性能计算工具来缩小这一差距。创建的软件是为了服务于同步加速器科学家的广泛社区。纳米级成像技术的发展旨在应用于从凝聚态物理到单分子生物学和化学等领域。计算研究人员和x射线散射和材料物理专家之间密切的跨学科研究工作将使在这些领域工作的研究生和博士后受益，并可能产生新的和意想不到的发现。本研究项目的目标是创建一个高性能的计算框架，用于在x射线散射设施中进行实时x射线成像。x射线成像将使用一种新兴的三维x射线成像方法，称为相干x射线衍射成像（CXDI），该方法基于x射线衍射图样的计算相位检索。提出的研究包括算法和代码实现、实际应用和教育部分。介绍了一组相位检索x射线成像的算法，包括图像表示和使用相关约束的方法，实空间到互反空间之间计算转换的方法，以及在大规模并行计算系统上的实现。该研究还引入了一个虚拟同步加速器框架，利用计算方法来模拟实验图像重建，并提供所需的约束，以磁性纳米结构为测试平台。开发的计算框架将应用于纳米结构材料和器件的实时研究，例如纳米级磁性器件。该计划通过加强计算物理、微磁学和电磁学的本科和研究生课程，以及提供研究生和本科生培训，促进多样性，将研究和教育结合起来。

Modern science and technology increasingly rely on materials and devices utilizing unique properties of nanometer-scale composition. It is critical to be able to characterize such complex nanoscale configurations. A unique tool for characterization of nanoscale materials and devices is X-ray imaging, which can provide structural images with a high spatial and time resolution. In a typical X-ray imaging setup an X-ray source generates an X-ray beam that is scattered from the structure. The scattered X-ray beam is captured by a camera, and the resulting picture is processed to reconstruct the actual structure configuration. Due to their short wavelength X-rays can penetrate the structure's volume thus enabling not only two-dimensional but also three-dimensional imaging. X-ray imaging typically requires a significant computational effort for image reconstruction and often the computational time is longer than the time for performing the experiment. This project aims at closing this gap by creating high-performance computational tools for real-time X-ray imaging. The created software is to serve a broad community of synchrotron scientists. The developments in nanoscale imaging are intended to have applications in areas ranging from condensed matter physics to biology and chemistry of single molecules. A close interdisciplinary research effort between a computational researcher and an expert in X-ray scattering and material physics will benefit graduate students and postdocs working in these fields, and is likely to result in new and unexpected discoveries. The goal of this research project is to create a high-performance computational framework for real-time X-ray imaging to be performed at X-ray scattering facilities. The X-ray imaging is to be accomplished using a new emerging three-dimensional X-ray imaging methodology called Coherent X-ray Diffractive Imaging (CXDI), which is based on computational phase-retrieval of the X-ray diffraction patterns. The proposed research includes algorithm and code implementation, practical application, and educational components. A set of algorithms for phase retrieval X-ray imaging are to be introduced, including methods for image representation and use of relevant constraints, methods for computing transformations between real to reciprocal spaces, and implementations on massively parallel computing systems. The research effort also introduces a virtual synchrotron framework utilizing computational methods to simulate experimental image reconstruction and provide required constraints, with magnetic nanostructures as a testbed. The developed computational framework is to be applied to real-time study of nanostructured materials and devices, such as nanoscale magnetic devices. The program integrates research and education by enhancing undergraduate and graduate courses on computational physics, micromagnetics and electromagnetics as well as providing graduate and undergraduate student training and contributing to diversity.