Development of alternative, screening based, local Cpupled Cluster methods and efficient algorithms for parallel architectures

开发替代的、基于筛选的、局部集群方法和并行架构的高效算法

• 批准号: 56215113
• 负责人: Professor Dr. Alexander A. Auer
• 金额: --
• 依托单位: Max-Planck-Institut für Kohlenforschung
• 依托单位国家: 德国
• 项目类别: Priority Programmes
• 财政年份: 2007
• 资助国家: 德国
• 起止时间: 2006-12-31 至 2010-12-31
• 项目状态: 已结题
• 来源: https://gepris.dfg.de/gepris/projekt/56215113?language=en
• 关键词: Development; alternative; screening; based; local

In the field of Coupled Cluster method development for larger molecular systems a gap has been opening during recent years between the robust and highly accurate conventional methods and the local approximations that suffer from a certain arbitrariness in the truncation scheme based on local orbitals and excitation domains. The dynamical thresholding algorithm developed in our group represents an alternative to the linear scaling methods that focuses on accuracy and error-control rather than on efficiency. In previous work it has been shown that using perturbation theory to prescreen amplitudes leads to a Coupled Cluster approximation with controllable error that does not require any a priori assumptions about the structure of the wavefunction parameters. The project presented in this proposal aims at turning the current pilot implementation into a fully efficient scheme for parallel computer architectures. For this purpose the current screening techniques have to be applied to the intermediate quantities that occur in Coupled Cluster algorithms. In order to reach the break even point of the new, low scaling algorithm to highly efficient conventional Coupled Cluster implementations it is necessary to exploit the computational resources of large scale parallel computer architectures. For the efficient development of a new implementation automatic program generation techniques will be used. Finally the features of the code will be extended to energy gradients and the treatment of higher levels of excitations.

近年来，在大分子体系的耦合簇方法发展领域，传统方法的鲁棒性和高精度与基于局部轨道和激发域的截断方法存在一定的任意性的局部近似方法之间存在着一条鸿沟。我们小组开发的动态阈值算法代表了线性缩放方法的替代方案，该方法侧重于准确性和误差控制，而不是效率。在以前的工作中，它已被证明，使用微扰理论prescreen振幅导致耦合集群近似与可控的错误，不需要任何先验假设的波函数参数的结构。本提案中提出的项目旨在将当前的试点实施转变为并行计算机体系结构的完全有效的方案。为此目的，目前的筛选技术必须应用于耦合聚类算法中出现的中间量。为了达到收支平衡点的新的，低规模的算法，以高效的传统耦合集群实现，它是必要的，利用大规模并行计算机体系结构的计算资源。为了有效地开发一个新的实现，将使用自动程序生成技术。最后，该代码的功能将扩展到能量梯度和更高级别的激励的治疗。