大地电磁三维正反演受计算速度、内存空间、精度要求等限制，难于在PC机实现，制约大地电磁法三维解释的发展及应用。本项目拟将计算物理和应用数学领域中解大型微分方程的算法—区域分解法引入到大地电磁三维正演，将三维空间域分解成若干个非重叠子域，通过子域Schur补偿算法和有限差分离散形成各子域、边界点和交叉点的独立线性方程组并求解，然后合成全域解；在反演中，开展数据空间共轭梯度算法，数据空间反演大大减少了反问题的规模，共轭递度技术避免Jacobi矩阵的直接存储有效利用内存资源；同时针对多核PC机，结合区域分解的内在并行性，实现高效并行求解。基于子域算法的正演与数据空间反演技术结合从根本上解决了三维大地电磁正反演对海量计算机内存的需求，另外在子域内求解具有规模小、求解速度快、并行性好等特点，使得大规模大地电磁三维正反演在PC机上实现成为可能。本研究工作的开展将进一步推进大地电磁数据三维解释的发展。

3D modeling and inversion of Magnetotelluric is difficult for implementment on PC by the computational velocity, memory and accuracy, etc.. It seriously hampered the development of MT 3D measure and interpretation. In this project, the domain decomposition is used for modelling problem ,which is successful in solving large-scale problem of differential equation at computational physics and applied mathematics. The global 3D computational domain is split into many smaller subdomains.Solution is gradually accomplished in each subdomains based on staggered finite-defference and Schur complement method.And Data Space Conjugate Gradient method is used for 3D inversion. It would be advoid for storing Jacobi matrix and calculating on larger model space. And the 3D modelling and inversion is accelarated on multicore PC by parallel calculation of domain decomposition to linear equation used on OPENMP.This method is radically resolved huge memory problem with not sorting linear equation of modelling and Jacobi matrix of inversion; and it is small scale,fast calculating speed,high accuracy in subdomain. So the rapid large-scale 3D modelling and inversion algorithms of MT based on domain decomposition is accomplishable for PC.This project would promote the development of the magnetotelluric 3D interpretation techniques.