School of Mathematical Sciences, Yangzhou University, Yangzhou 225002, China School of Hydraulic Science and Engineering, Yangzhou University, Yangzhou 225002, China
Abstract:The phase separation phenomenon between different matters plays an important role in many science fields. And the high order nonlinear Cahn-Hilliard (C-H) equation is often used to describe the phase separation phenomenon between different matters. However, it is difficult to solve the high-order nonlinear C-H equations by the theorical methods and the grid-based methods. Therefore, in this work the meshless methods are addressed, and a local refinement finite pointset method (LR-FPM) is proposed to numerically investigate the high-order nonlinear C-H equations with different boundary conditions. Its constructive process is as follows. 1) The fourth derivative is decomposed into two second derivatives, and then the spatial derivative is discretized by FPM based on the Taylor series expansion and weighted least square method. 2) The local refinement and quintic spline kernel function are employed to improve the numerical accuracy. 3) The Neumann boundary condition with high-order derivatives is accurately imposed when solving the local linear equation sets. The 1D/2D C-H equations with different boundary conditions are first solved to show the ability of the LR-FPM, and the analytical solutions are available for comparison. Meanwhile, we also investigate the numerical error and convergence order of LR-FPM with uniform/non-uniform particle distribution and local refinement. Finally, 1D/2D C-H equation without analytical solution is predicted by using LR-FPM, and compared with the FDM result. The numerical results show that the implement of the boundary value condition is accurate, the LR-FPM indeed has a higher numerical accuracy and convergence order, is more flexible and applicable than the grid-based FDM, and can accurately predict the time evolution of nonlinear diffusive phase separation phenomenon between different materials time. Keywords:pure meshless method/ Cahn-Hilliard equation/ local refinement/ nonlinear diffusion
表2不同时刻下粒子均匀分布与局部加密情况下的L2-范数误差${E_2}$对比 Table2.The L2-norm error ${E_2}$ at different times under the uniform and local refinement particle distributions.
表5粒子均匀分布、局部加密分布与非均匀分布情况下的L2-范数误差${E_2}$对比 Table5.The L2-norm error ${E_2}$ at different times under the uniform, local refinement, and non-uniform particle distributions.
表6t = 0.01 s时不同粒子间距非均匀分布情况下的L2-范数误差${E_2}$和收敛阶 Table6.The L2-norm error ${E_2}$ and convergence rate at t = 0.01 s under non-uniform particle distribution.
该算例为一维无解析解的带Neumann边界C-H方程, 它将描述复杂的相位分离现象, 常被用来证明其保持方程的质量守恒性质和能量耗散性质. 本节运用LR-FPM方法对${\varepsilon ^{\rm{2}}} = 0.3$情况下算例进行了模拟, 并与文献[8]中FDM方法结果进行对比. 图5给出了${\varepsilon ^{\rm{2}}} = 0.3$, 时间步长为${\rm{d}}t = {10^{ - 6}}$, 初始时刻与$ t = 0.2\;{\rm{ s}}, t = 2\;{\rm{ s}}$三个不同时刻两种数值方法的模拟结果. 由图5可知, LR-FPM方法得到的带Neumann边界C-H方程的数值解与FDM结果吻合. 图6给出了${\varepsilon ^{\rm{2}}} = 0.03, t = 0.2\;{\rm{ s}}$时刻下粒子局部加密情况, 在[–1.0, –0.8], [0.8, 1.0]处加密, 每隔0.02的距离布置粒子, 在(–0.8, 0.8)内每隔0.04的距离布置粒子. 由图6可知, 粒子均匀分布与局部加密情况下的数值结果均与FDM结果吻合, 但局部加密情况下的模拟结果更接近于FDM的结果. 这进一步表明提出的LR-FPM粒子方法模拟C-H方程是可靠的. 图 5${\varepsilon ^{\rm{2}}}{\rm{ = 0}}{\rm{.3}}$时不同时刻FDM结果与LR-FPM结果 Figure5. The numerical results obtained using FDM and LR-FPM at different times with ${\varepsilon ^{\rm{2}}}{\rm{ = 0}}{\rm{.3}}$.
图 6${\varepsilon ^2} = 0.03,\; {\rm{ }}t = 0.2\;{\rm{ s}}$时刻下均匀分布与局部加密情况下数值结果对比 Figure6. The present numerical results under uniform and local refinement particle distributions at ${\varepsilon ^2} \!=\! 0.03$, t = 0.2 s.
图 8$t = 0.{\rm{08 \;s}}$时刻本文方法模拟结果与文献[11]中数值等值线分布 (a) 文献[11]中数值结果; (b)?(d) 本文方法在三种粒子分布情况下数值结果 Figure8. The contour results obtained using the present method and the numerical results in ref.[11] at $t = 0.{\rm{08 \;s}}$: (a) Numerical results in [11]; (b)?(d) present numerical results
图9给出了$t = 0.{\rm{08\; s}}$时刻下不同粒子间距均匀分布的情况以及局部加密分布下的数值收敛性. 由图9可知, LR-FPM数值预测二维C-H问题时是收敛的. 因此, LR-FPM能够有效、可靠地模拟C-H方程, 容易实施局部加密或非均匀分布情况下的数值模拟. 图 9$t \!=\! 0.{\rm{08\; s}}$时刻粒子局部加密分布情况下的数值收敛 Figure9. The numerical convergence obtained using the present method under different particle distributions at $t = 0.{\rm{08\; s}}$.