论文标题
使用递归投影方法对异质材料的有效响应
Effective Response of Heterogeneous Materials using the Recursive Projection Method
论文作者
论文摘要
本文将递归投影方法(RPM)应用于找到周期性异质固体的有效机械响应的问题。以前的工作将快速傅立叶变换(FFT)与各种定点方法结合使用,以解决周期性单位单元格上的问题。在从基于图像的建模到位错可塑性的一系列问题中,这些问题已被证明非常强大。但是,如果弹性特性具有较高的对比度,例如在空隙的情况下,固定点迭代会收敛非常缓慢,或者根本不会收敛。本文从各个角度来看,研究了缓慢或缺乏融合的原因。特别是,当材料包含刚度为零或非常小的区域时,缺乏独特性,并且能量景观具有平坦或浅的方向。因此,在这项工作中,固定点迭代被RPM迭代取代。 RPM使用固定点迭代来自适应地识别在哪些固定点迭代不稳定的子空间,并且仅在不稳定的子空间上执行牛顿迭代,而在互补稳定的子空间上进行固定点迭代。在可能的情况下,这种有效的固定点迭代的结合以及需要昂贵但稳定的牛顿迭代的结合,可显示该方法的稳健有效收敛。特别是,对于参考介质的各种选择,RPM-FFT收敛良好,而通常的固定点迭代通常对此选择敏感。
This paper applies the Recursive Projection Method (RPM) to the problem of finding the effective mechanical response of a periodic heterogeneous solid. Previous works apply the Fast Fourier Transform (FFT) in combination with various fixed-point methods to solve the problem on the periodic unit cell. These have proven extremely powerful in a range of problems ranging from image-based modeling to dislocation plasticity. However, the fixed-point iterations can converge very slowly, or not at all, if the elastic properties have high contrast, such as in the case of voids. The paper examines the reasons for slow, or lack of convergence, in terms of a variational perspective. In particular, when the material contains regions with zero or very small stiffness, there is lack of uniqueness, and the energy landscape has flat or shallow directions. Therefore, in this work, the fixed-point iteration is replaced by the RPM iteration. The RPM uses the fixed-point iteration to adaptively identify the subspace on which fixed-point iterations are unstable, and performs Newton iterations only on the unstable subspace, while fixed-point iterations are performed on the complementary stable subspace. This combination of efficient fixed-point iterations where possible, and expensive but well-convergent Newton iterations where required, is shown to lead to robust and efficient convergence of the method. In particular, RPM-FFT converges well for a wide range of choices of the reference medium, while usual fixed-point iterations are usually sensitive to this choice.