对于求解超定线性代数方程组的随机坐标下降 (Randomized Coordinate Descent or RCD) 方法,
为选取每步迭代所需调用的系数矩阵的目标列,我们提出了一个新的有效的概率准则,
并由此构造出了贪婪随机坐标下降 (Greedy Randomized Coordinate Descent or GRCD) 方法。
当系数矩阵满秩时,我们证明了GRCD方法收敛到相应的最小二乘解;理论分析与数值实验均表明,
GRCD方法的收敛速率比RCD方法更快。对于不相容的超定线性代数方程组,
我们提出了部分随机扩展Kaczmarz (Partially Randomized Extended Kaczmarz or PREK) 方法,
对其收敛性和收敛速率的上界分析以及相关数值实验均表明,在一定条件下PREK方法
可以比随机扩展Kaczmarz方法更为有效。
对于稳定化的鞍点问题,我们推广了正则化的Hermitian和反Hermitian分裂 (RHSS) 迭代方法,
并证明了该方法无条件收敛到问题的唯一解;同时,我们也证明了由这类RHSS方法所产生的
预处理矩阵的特征值具有很好的聚集性质。数值实验证实,无论作为迭代法还是作为预处理子,
这类RHSS方法均比Hermitian和反Hermitian分裂 (HSS) 方法更为有效。
特别,不精确RHSS预处理Krylov子空间方法比相应的精确方法具有更好的数值性质
和更高的计算效率。
2019年正式发表的论文如下:
- On partially randomized extended Kaczmarz method
for solving large sparse overdetermined inconsistent
linear systems,
Linear Algebra and its Applications,
578(2019), 225-250.
(With Wen-Ting Wu)
- Regularized HSS iteration methods for
stabilized saddle-point problems,
IMA Journal of Numerical Analysis,
39:4(2019), 1888-1923.
- On greedy randomized coordinate descent methods
for solving large linear least-squares problems,
Numerical Linear Algebra with Applications,
26:4(2019), e2237:1-15.
(With Wen-Ting Wu)
- On multistep Rayleigh quotient iterations
for Hermitian eigenvalue problems,
Computers and Mathematics with Applications,
77:9(2019), 2396-2406.
(With Cun-Qiang Miao and Shuai Jian)
- On banded M-splitting iteration methods
for solving discretized spatial fractional
diffusion equations,
BIT Numerical Mathematics,
59:1(2019), 1-33.
(With Kang-Ya Lu)
- Computing eigenpairs of Hermitian matrices
in perfect Krylov subspaces,
Numerical Algorithms,
82:4(2019), 1251-1277.
(With Cun-Qiang Miao)