节点梯度光滑有限元配点法
A FINITE ELEMENT COLLOCATION METHOD WITH SMOOTHED NODAL GRADIENTS
-
摘要: 配点法构造简单、计算高效, 但需要用到数值离散形函数的高阶梯度,而传统有限元形函数的梯度在单元边界处通常仅具有C^0连续性,因此无法直接用于配点法分析. 本文通过引入有限元形函数的光滑梯度,提出了节点梯度光滑有限元配点法. 首先基于广义梯度光滑方法,定义了有限元形函数在节点处的一阶光滑梯度值,然后以有限元形函数为核函数构造了有限元形函数的一阶光滑梯度,进而对一阶光滑梯度直接求导并用一阶光滑梯度替换有限元形函数的标准梯度,即完成了有限元形函数二阶光滑梯度的构造.文中以线性有限元形函数为基础的理论分析表明,其光滑梯度不仅满足传统线性有限元形函数梯度对应的一阶一致性条件,而且在均布网格假定下满足更高一阶的二阶一致性条件.因此与传统线性有限元法相比,基于线性形函数的节点梯度光滑有限元法的L_2和H_1误差均具有二次精度,即其H_1误差收敛阶次比传统有限元法高一阶, 呈现超收敛特性.文中通过典型算例验证了节点梯度光滑有限元配点法的精度和收敛性,特别是其H_1或能量误差的精度和收敛率都明显高于传统有限元法.Abstract: The collocation formulation has the salient advantages of simplicity and efficiency, but it requires the employment of high order gradients of shape functions associated with certain discretized strategies. The conventional finite element shape functions are usually C^0 continuous and thus cannot be directly adopted for the collocation analysis. This work presents a finite element collocation method through introducing a set of smoothed gradients of finite element shape functions. In the proposed formulation, the first order nodal smoothed gradients of finite element shape functions are defined with the aid of the general gradient smoothing methodology. Subsequently, the first order smoothed gradients of finite element shape functions are realized by selecting the finite element shape functions as the kernel functions for gradient smoothing. A further differential operation on the first order smoothed gradients then leads to the desired second order smoothed gradients of finite element shape functions, where it is noted that the conventional first order gradients are replaced by the first order smoothed gradients of finite element shape functions. It is theoretically proven that the proposed smoothed gradients of linear finite element shape functions not only meet the first order gradient reproducing conditions that are also satisfied by the conventional gradients of finite element shape functions, but also meet the second order gradient reproducing conditions for uniform meshes that cannot be fulfilled by the conventional finite element formulation. The proposed smoothed gradients of finite element shape functions enable a second order accurate finite element collocation formalism regarding both L_2 and H_1 errors, which is one order higher than the conventional linear finite element method in term of H_1 error, i.e., a superconvergence is achieved by the proposed finite element collocation method with smoothed nodal gradients. Numerical results well demonstrate the convergence and accuracy of the proposed finite element collocation method with smoothed nodal gradients, particularly the superior convergence and accuracy over the conventional finite element method according to the H_1 or energy errors.