无网格法精度分析及在电磁法二维正演中的应用（英文）

Precision of meshfree methods and application to forward modeling of two-dimensional electromagnetic sources

无网格法形函数构造不依赖预定义的单元,具有计算精度高、处理复杂模型便利等优点。本文介绍了无单元Galerkin法(EFGM)、点插值法(PIM)与径向基点插值法(RPIM)三种全域弱式无网格法的近似原理及特点;以二维泊松方程为例研究了支持域无量纲尺寸、场节点与背景网格设置对无网格法计算精度的影响。将RPIM与EFGM应用于频率域线源二维正演,给出了RPIM形状参数的推荐值;分析了均匀介质模型大地电磁(MT)二维正演无网格法边界条件直接加载与罚函数法加载的精度差异,结合PIM与RPIM边界条件加载便利及EFGM计算复杂模型精度高的优势,提出了EFG-PIM及EFGRPIM耦合算法,数值计算结果验证了耦合算法的有效性。研究发现:无网格法及其耦合方法适用于电磁法数值模拟;支持域无量纲尺寸取1.0时无网格法精度与效率高,场节点与背景网格重合时计算效果佳;泊松方程求解PIM及RPIM精度较EFGM低,计算均匀介质MT响应精度较EFGM高;RPIM改善了PIM计算涉及的奇异性问题,对应支持域无量纲尺寸选择空间大。

Meshfree method offers high accuracy and computational capability and constructs the shape function without relying on predefined elements. We comparatively analyze the global weak form meshfree methods, such as element-free Galerkin method （EFGM）, the point interpolation method （PIM）, and the radial point interpolation method （RPIM）. Taking two dimensional Poisson equation as an example, we discuss the support-domain dimensionless size, the field nodes, and background element settings with respect to their effect on calculation accuracy of the meshfree method. RPIM and EFGM are applied to controlled- source two-dimensional electromagnetic modeling with fixed shape parameters. The accuracy of boundary conditions imposed directly and by a penalty function are discussed in the case of forward modeling of two-dimensional magnetotellurics in a homogeneous medium model. The coupling algorithm of EFG-PIM and EFG-RPIM are generated by integrating the PIM or RPIM and EFGM. The results of the numerical modeling suggest the following. First, the proposed meshfree method and corresponding coupled methods are well-suited for electromagnetic numerical modeling. The accuracy of the algorithm is the highest when the support-domain dimensionless size is 1.0 and the distribution of field nodes is consistent with the nodes of background elements. Second, the accuracy of PIM and RPIM are lower than that of EFGM for the Poisson equation but higher than EFGM for the homogeneous medium MT response. Third, RPIM overcomes the matrix inversion problem of PIM and has a wider selection of support-domain dimensionless sizes as compared to RPIM.