Speeding up GW quasiparticle calculations to meet the challenge of fast and accurate materials prediction

加速 GW 准粒子计算，应对快速准确材料预测的挑战

• 批准号: 1506669
• 负责人: Peihong Zhang
• 金额: $ 32万
• 依托单位: SUNY at Buffalo
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2015
• 资助国家: 美国
• 起止时间: 2015-06-01 至 2020-05-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1506669&HistoricalAwards=false
• 关键词: Speeding; up; GW; quasiparticle; calculations

NON-TECHNICAL SUMMARYDesigning novel materials with desired properties is of vital importance for solving some of the most pressing challenges facing our society in energy and environmental issues. Computational materials design relies on theoretical and computational methods that can reliably predict materials properties within a reasonable time frame for both their lowest-energy state and excited states. Electronic structure methods based on Density Functional Theory are capable of predicting many important lowest-energy state properties, and the so-called "GW method" is by far the most successful and theoretically sound method for predicting excited-state properties of materials. Unfortunately, despite celebrated advances, accurate and efficient predictions of excited-state properties of complex solid systems remain a major challenge. This project supports theoretical and computational research and education that will significantly reduce the computational time and required memory of GW calculations for complex systems such as nanostructures, complex compounds, surfaces, interfaces, and materials containing spatially localized electrons. This will enable fast and accurate predictions for excited-state properties of a significantly wide range of scientifically and technologically important materials.Research will be performed in collaboration with both domestic and international groups. The developed code will first be made available to interested research groups, and then will be released to public after validation and optimization. This project will provide interdisciplinary training in physics, computational materials science, and high performance computing for both graduate and undergraduate students. Such integrated training will greatly broaden students' knowledge base and skill sets in preparation for their future career.TECHNICAL SUMMARYAccurate predictions of excited states properties are critical for computational screening and design of materials for energy and electronics applications. Unfortunately, despite much research effort and celebrated advances, notably the development of first-principles GW methods, accurate and efficient predictions of excited-state properties of solids remain a major challenge. This is particularly true for systems with large unit cells, such as nanostructures, complex multinary compounds, surfaces, interfaces, and materials containing localized electrons due to the unfavorable scaling of the computational cost of GW calculations with respect to the system size.This project supports theoretical and computational research and education which involves the development of several new techniques that will dramatically reduce the computational cost of GW calculations for large complex systems. These new approaches include a) A Fourier filtering technique for drastically reducing the storage and computation cost associated with high-energy states; b) An energy-integration approach for alleviating the burden of explicit band-by-band summation in conventional GW calculations; c) A novel strategy for reducing the computational and memory requirement of the dielectric matrix; and d) Implementation of diagonalization methods that calculate only those eigenstates at or near pre-determined energies for GW calculations. These new approaches, once fully developed and integrated, are expected to result in over two orders of magnitude reduction in computational time and memory requirement for GW calculations on large systems.Research will be performed in collaboration with both domestic and international groups. The developed code will first be made available to interested research groups, and then will be released to public after validation and optimization. This project will provide interdisciplinary training in physics, computational materials science, and high performance computing for both graduate and undergraduate students. Such integrated training will greatly broaden students' knowledge base and skill sets in preparation for their future career.

设计具有理想性能的新材料对于解决我们社会在能源和环境问题上面临的一些最紧迫的挑战至关重要。计算材料设计依赖于理论和计算方法，这些方法可以在合理的时间范围内可靠地预测材料的最低能态和激发态的性质。基于密度泛函理论的电子结构方法能够预测许多重要的最低能态性质，而所谓的“GW方法”是迄今为止预测材料激发态性质最成功和理论上最合理的方法。不幸的是，尽管取得了令人瞩目的进展，但准确有效地预测复杂固体系统的激发态性质仍然是一个重大挑战。该项目支持理论和计算研究和教育，将显著减少复杂系统（如纳米结构、复杂化合物、表面、界面和包含空间定域电子的材料）的计算时间和GW计算所需的内存。这将能够快速准确地预测各种科学和技术上重要材料的激发态特性。研究将与国内和国际团体合作进行。开发的代码将首先提供给感兴趣的研究小组，然后在验证和优化后向公众发布。该项目将为研究生和本科生提供物理学、计算材料科学和高性能计算方面的跨学科培训。这种综合训练将大大拓宽学生的知识基础和技能，为他们未来的职业生涯做准备。技术概述：激发态特性的准确预测对于能源和电子应用材料的计算筛选和设计至关重要。不幸的是，尽管许多研究工作和著名的进展，特别是第一原理GW方法的发展，准确和有效地预测固体的激发态性质仍然是一个主要的挑战。这对于具有大晶胞的系统尤其如此，例如纳米结构、复杂的多化合物、表面、界面和包含局域电子的材料，这是由于GW计算的计算成本相对于系统尺寸的不利缩放。该项目支持理论和计算研究和教育，涉及几种新技术的开发，这些技术将大大降低大型复杂系统的GW计算成本。这些新方法包括a)大幅度降低与高能态相关的存储和计算成本的傅立叶滤波技术；b)一种能量集成方法，以减轻传统GW计算中显式逐带求和的负担；c)一种降低介电矩阵计算和存储需求的新策略；d)实现对角化方法，仅计算GW计算中处于或接近预定能量的特征态。这些新方法一旦完全开发和集成，预计将使大型系统上GW计算的计算时间和内存需求减少两个数量级以上。研究将与国内和国际团体合作进行。开发的代码将首先提供给感兴趣的研究小组，然后在验证和优化后向公众发布。该项目将为研究生和本科生提供物理学、计算材料科学和高性能计算方面的跨学科培训。这种综合训练将大大拓宽学生的知识基础和技能，为他们未来的职业生涯做准备。