基于预处理共轭梯度的大地电磁快速正演  

Magnetotellurics fast forward based on preconditioning conjugate gradient

作　　者：张继锋[1,2] 汤井田[2] 王烨[2] 肖晓[2] 

机构地区：[1]长安大学地质工程与测绘学院,陕西西安710054 [2]中南大学信息物理工程学院,湖南长沙410083

出　　处：《中南大学学报（自然科学版）》2010年第5期1877-1882,共6页Journal of Central South University:Science and Technology

基　　金：国家高技术研究发展计划("863"计划)项目(2007AA06Z134;2006AA06Z105);国家自然科学基金资助项目(40874072)

摘　　要：针对大地电磁法有限元模拟中计算量大的特点,采用按行压缩存储方式的不完全LDLT预处理共轭梯度法快速求解大型复系数方程组。引入不完全LDLT预处理,提出快速求解(LDLT)-1r的方法,以加快预处理共轭梯度法的收敛速度。研究结果表明:当网格节点自由度超过1万时,压缩率达到99.9%,求解方程组时间在1 s以内,为进一步快速反演奠定了基础。Based on the characteristics of large computing cost in finite element modeling about magnetotelluric method,preconditioning conjugate gradient method based on compressed storage was proposed to solve large and complex systems.Incomplete preconditioner LDLT was introduced and fast solving（LDLT）-1r was proposed to accelerate the convergence rate of preconditioning conjugate gradient method.The results show that the compression ratio reaches 99.9%,and it completes in less than 1 s for the equations more than 10 000 degrees of freedom,which further lays the foundation for rapid inverse.

关 键 词：按行压缩存储 预处理 共轭梯度 大地电磁 

分 类 号：P631.3[天文地球—地质矿产勘探]