项目名称: 一类大规模结构线性鞍点问题的高效算法与理论
项目编号: No.11271174
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 数理科学和化学
项目作者: 张国凤
作者单位: 兰州大学
项目金额: 65万元
中文摘要: 摘要:大规模结构化线性方程组广泛地产生于科学计算与工程应用的许多领域,如优化设计、计算流体力学、结构工程、空气动力学、图像处理、集成电路模拟、动力系统模拟及信号处理和控制理论等。因此,研究这类线性代数方程组的高性能计算方法及其理论,就具有重要的理论意义、广泛的应用背景和很高的经济价值。 一类由偏微分方程约束的优化问题,经过合适的有限元离散会产生一类大规模结构线性鞍点问题。本项目主要研究求解这类特殊的线性代数方程组的高质量预处理子和高性能迭代方法的代数构造、数学理论及其具体实现。特别,针对约束偏微分方程分别为泊松方程、对流扩散方程和Stokes方程的具体情形,我们将利用相应矩阵的稀疏结构、子块结构及其代数性质,去构造高质量的预处理子和快速、准确而稳健的计算方法,建立关于这些方法的深刻的数学理论,并编制程序予以具体实现。此外,我们还拟将这些新方法和理论实际应用于大型气动外形的优化设计之中。
中文关键词: 鞍点问题;预处理子;Krylov 子空间方法;收敛性;离散系统
英文摘要: The large scale and structured system of linear equations may arise in many areas of scientific computing and engineering applications, including optimal design, computational fluid mechanics, structural engineering, aerodynamics, image processing, integrated circuits modeling, signal processing, control theory and so on. Hence, to study effective numerical methods and their theory for this class of linear systems is of important theoretical meaning, extensive application background and high economical value. A class of optimization problems constrained by partial differential equations, when discretized by finite element method,will result in a class of large structured linear saddle point problems. In this project, we will mainly study the algebraic constrction, the mathematics theory and the concret implementations of high-quality preconditioners and high-performance iteration methods for solving this class of linear systems. In particular,when the PDE constraints are the Poisson equations, the convection-diffusion equations and the Stokes equations, respectively, by fully utilizing the sparse pattern, the sub-block structure and the algebraic properties of the corresponding coefficient matrices,we will construct high quality preconditioners and fast, accurate and robust computational methods, establish th
英文关键词: saddle point problems;preconditioner;Krylov subspace methods;convergence;discretized systems