项目名称: 矩阵秩与惯量函数最优化问题与应用的研究
项目编号: No.11271384
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: 田永革
作者单位: 中央财经大学
项目金额: 39万元
中文摘要: 本项目试图建立一套以矩阵秩和惯量为目标函数的优化理论,并开展这种优化理论在统计学中应用的研究.内容分为矩阵优化理论和统计应用两部分:(1)利用矩阵的联合分解,矩阵的对合变换和矩阵的广义逆等方法,建立一批常用的矩阵秩和惯量解析计算公式;推导出一批线性和非线性秩和惯量目函数最大最小值问题的解析解;给出这些目标函数取得这些最大最小值的计算方法和实用程序。(2)用上述方法研究回归分析中的模型和统计量的最优性质,包括各种线性、半线性、非线性回归分析模型的相容性;回归模型中参数的隐性约束;多余观测与缺失数据;参数的可估性;参数估计的唯一性;回归模型中参数不同估计量之间的等价性;正确及错误回归模型中估计量之间的关系;回归模型与其各种子模型之间的关系;回归模型中参数估计量的各种可加性分解与分块分解;开发相应的计算方法和应用程序。
中文关键词: 矩阵;秩;惯性指数;最优化;解析解
英文摘要: The proposed research program of the applicant tries to establish a complete theory on optimizations of rank and inertia functions of several fundamental matrix-valued functions. The task includes two parts in both matrix optimzation theory and statistical applications: (1) derive various analytical expansion formulas for calculating ranks and inertias of matrices by using simultaneous decompositions, congruence transformations and generalized inverses; utilize the formulas obtained to solve rank and inertia maximization and minimization problems of some ordinary linear and quadratic matrix-valued functions and provide the corresponding computational methods and practical procedures. (2) Use the matrix optimization theory established to characterize cretain optmial behaviors of regression models in statistics, such as, admissability and consistency of linear, semi-linear, and non-parametric regression models, existence and deduction of implicit restrictions, estimability of unknown parameters, equivalence of different estimations, relations of full models and their sub-models, additive decompositions and block decomposition of estimations. The computational methods for finishing these tasks will be developed,which we believe will be greatly enrich the theoy of regression analysis.
英文关键词: matrix;rank;inertia;optimization;analytical solution