移动最小二乘法导数近似讨论
2018-03-21杨建军杨子乐黄旺文丕华
杨建军 杨子乐 黄旺 文丕华


摘要:在移动最小二乘法(moving least squares method, MLS)构造无网格形函数的数值方法中,通常采用无单元伽辽金法(element-free Galerkin method, EFG)的建议,将系数向量a参与导数运算。为探讨这种导数近似算法在更一般无网格法中的适用性和合理性,针对系数向量a是否应参与运算的问题进行讨论和数值检验。结果表明:单纯从近似意义上讲,这种将系数向量代入导数运算的算法并不具有优势;从数值方法的应用意义上讲,这种导数近似算法对数值求解,特别是强式无网格法,会带来一系列潜在不稳定的问题。建议在MLS导数近似中,系数向量a不应当参与导数运算,并提出采用一种由核基函数代替普通基函数的核近似法。
关键词:无网格法; 移动最小二乘法; 导数近似; 系数向量; 核近似
中图分类号:O241.3,O242.2,O302
文獻标志码:A
文章编号:1006-0871(2018)01-0028-07
Abstract: In the numerical method of meshless shape function constructing with moving least squares method(MLS), the element-free Galerkin method(EFG) is adopted customarily, and the coefficient vector a usually take part in the derivative operation. To investigate the applicability and rationality of this derivative approximation algorithm applying to more general meshless method, the discussion and numerical tests are carried out which focus on whether the coefficient vector a should take part in computing or not. The results show that on approximate means only, there is no advantage if the coefficient vector takes part in computing; on the application means of the numerical method, there is a series of potential instability problems using the derivative approximation algorithm on numerical calculation, especially for the strong meshless method. It is suggested that the coefficient vector a should not be included in the derivative operation. A core approximation method is proposed in which the common basis function is replaced by core primary function.
Key words: meshless method; moving least squares method; derivative approximation; coefficient vector; core approximation
0 引 言
移动最小二乘 (moving least squares, MLS)法最初用于地理信息绘制和表面拟合。[1-3]随着无网格法研究的兴起,MLS法被广泛用于构造无网格形函数。[4-9]
BELYTSCHKO等[10]提出无单元伽辽金法(element-free Galerkin methods, EFG)时,主张系数向量a(x)不应当被假定为常数向量,而应参与求导运算。此后,这一观点被广为接受,在MLS法的导数近似中得到普遍应用。[11-18]
MLS法在执行导数近似时,系数向量a(x)是否该参与导数运算是一个由来已久的问题。鉴于该问题涉及导数近似中基本概念的争议,并且应用范围非常广泛,因此有必要重新审视。
1 移动最小二乘法
设问题域Ω上的场函数为u(x),这个场函数的值只在一些域内的散乱点上已知或可测量。……
