数学物理学报, 2021, 41(2): 451-467 doi:

论文

向列相液晶流的模块grad-div稳定化有限元方法

李婷,, 黄鹏展,

A Modular grad-div Stabilized Finite Element Method for Nematic Liquid Crystal Flow

Li Ting,, Huang Pengzhan,

通讯作者: 黄鹏展, E-mail: hpzh007@yahoo.com

收稿日期: 2020-02-25  

基金资助: 国家自然科学基金.  11861067

Received: 2020-02-25  

Fund supported: the NSFC.  11861067

作者简介 About authors

李婷,E-mail:lt1003887017@163.com , E-mail:lt1003887017@163.com

Abstract

In this paper, we presents a modular grad-div stabilized finite element method for nematic liquid crystal flow, which adds to the backward Euler scheme a post precessing step. This method can penalize for lack of mass conservation but it does not increase computational time for increasing stabilized parameters. Moreover, error estimates for velocity and molecular orientation of the nematic liquid crystal flow are shown. Finally, the theoretical findings and numerical efficiency are verified by numerical experiments.

Keywords: Nematic liquid crystal model ; Modular grad-div stabilized method ; Error estimates ; Finite element method ; Backward Euler scheme

PDF (439KB) 元数据 多维度评价 相关文章 导出 EndNote| Ris| Bibtex  收藏本文

本文引用格式

李婷, 黄鹏展. 向列相液晶流的模块grad-div稳定化有限元方法. 数学物理学报[J], 2021, 41(2): 451-467 doi:

Li Ting, Huang Pengzhan. A Modular grad-div Stabilized Finite Element Method for Nematic Liquid Crystal Flow. Acta Mathematica Scientia[J], 2021, 41(2): 451-467 doi:

1 引言

本文考虑了最简单的向列相液晶流的模块grad-div稳定化有限元方法. 由Ericksen等人[11-12]和Leslie等人[20]最初建立的Ericksen-Leslie模型可以模拟流体动力学向列相液晶的流动. 该模型是对流速和分子方向的宏观连续性描述. Lin等人对该问题进行了深入的研究[21-24], 构造了一个简化的Ericksen-Leslie模型, 参见文献[21], 这对原有Ericksen-Leslie模型[4]进行理论和数值分析而言是一个不错的开端.

在有界凸区域$ \Omega\subset {{{\Bbb R}} }^{2} $中, 我们考虑简化的Ericksen-Leslie模型(参见文献[3, 21])

$ \begin{equation} \textbf{u}_t-\nu\Delta \textbf{u}+(\textbf{u}\cdot\nabla)\textbf{u}+\nabla p+\lambda \nabla\cdot(\nabla{\textbf{d}}\odot\nabla{\textbf{d}}) = \textbf{f}, \end{equation} $

$ \begin{equation} \textbf{d}_t-\gamma\Delta \textbf{d}+(\textbf{u}\cdot\nabla)\textbf{d} = \gamma|\nabla \textbf{d}|^2\textbf{d}, \end{equation} $

$ \begin{equation} \nabla\cdot\textbf{u} = 0, \; \; \; |\textbf{d}| = 1, \end{equation} $

$ ({\bf x}, t)\in Q_{T}, $其中$ Q_{T} = \Omega\times(0, T), $$ T\in(0, \infty). $$ {\bf u}({\bf x}, t):Q_{T}\rightarrow {{{\Bbb R}} }^{2} $$ p({\bf x}, t):Q_{T}\rightarrow {{\Bbb R}} $分别表示速度场和压力. $ {\bf d}({\bf x}, t):Q_{T}\rightarrow {\mathbb S}\subset {{{\Bbb R}} }^{2} $是向列相液晶材料分子取向场的指向矢. $ {\bf f}({\bf x}, t):Q_{T}\rightarrow {{{\Bbb R}} }^{2} $代表外力. 参数$ \nu $表示粘性系数, $ \lambda $表示动能与势能之间的竞争, $ \gamma $表示分子取向场的微观弹性弛豫时间. $ |\nabla\bf d| $$ |\bf d| $代表$ \nabla\bf d $$ \bf d $的欧几里德范式. 另外, $ \nabla{{\textbf{d}}}\odot\nabla{\textbf{d}} $$ 2\times 2 $矩阵, 矩阵中的元素$ (i, j) $可写为$ (\sum\limits_{k = 1}^2\frac{\partial d_k}{\partial {x}_i}\frac{\partial d_k}{\partial {x}_j})_{i, j} $.

方程(1.1)–(1.3) 的初始条件和边界条件由下式给出[3]

$ \begin{eqnarray} & {\bf u}({\bf x}, 0) = {\bf u}_{0}({\bf x}), \quad{\bf d}({\bf x}, 0) = {\bf d}_{0}({\bf x}), \quad \mathrm{in}\ \Omega, \end{eqnarray} $

$ \begin{eqnarray} & {\bf u}|_{S_T} = 0, \quad \partial_{{\bf n}}{\bf d}|_{S_T} = 0, \end{eqnarray} $

$ \nabla\cdot{\bf u}_{0} = 0 $$ |{\bf d}_{0}| = 1, $其中$ S_T = \partial\Omega\times(0, T), $$ {\bf n} $$ \partial\Omega $的单位外法线.

由于简化的Ericksen-Leslie模型的方程(1.1)–(1.5)不仅包括强非线性, 不可压缩性和非凸约束$ |{\bf d}| = 1 $, 而且还包括谐波映射流与运动的流体方程之间的耦合, 我们想要有效地求解这些方程并不容易. 因此, 许多研究者已投入大量精力来开发一些有效的数值方法来研究该模型, 参见文献[5, 8-10, 15-17, 34]. 但是, 对于这种简化的Ericksen-Leslie模型, 大多数方法通常都忽略或没有严格执行质量守恒定律. 如果数值算法满足质量守恒性, 那么所得的数值解将获得更大的物理精度. 当我们意识到质量守恒定律的重要性的时候, 自然而然地就会研究可以保持该定律的数值方法.

grad-div稳定化有限元方法[14], 可以通过减少压力对速度误差的影响来弥补缺乏质量守恒的缺点, 并提高求解精度. 目前, 与这种方法相关的文章有很多, 参见文献[1, 19, 25, 27, 29]. 但是, grad-div稳定化有限元方法取决于稳定化参数, 具有大的稳定化参数的grad-div稳定化有限元方法(大参数是不可避免的, 参见文献[13]), 可能会使相应的线性代数系统条件数变差, 并且该方法由于稀疏性降低和耦合性增加而需要大量的计算时间. 为了解决这些问题, 研究者提出了一种标准grad-div稳定化有限元方法的有效变形, 称为模块grad-div稳定化有限元方法[13], 该方法用于计算Navier-Stokes方程的解. 研究者发现该方法不受稳定化参数变化的影响, 而标准方法的计算成本随着稳定化参数的增长而迅速增加. 后来, 秦等人将该方法应用于Stokes/Darcy方程[30], Boussinesq方程[2]和磁流体动力学方程[26]. Rong等已将这种模块grad-div稳定化有限元方法从一阶时间精度提高到了二阶时间精度[31]. Akbas等将第二种方法扩展到了磁流体动力学方程[2].

本文的目的是将文献[13]中的方法推广到向列相液晶流问题(1.1)–(1.5)式. 方程(1.1)–(1.5)的数值方法包括两个步骤. 第一步是将简化的Ericksen-Leslie模型与向后欧拉时间混合有限元空间离散化近似. 第二步是后处理步骤, 我们引入了解耦的一阶grad-div稳定步骤(即模块grad-div稳定化方法). 本文的其余部分安排如下, 在第2节, 本文将介绍一些数学预备知识, 并为简化的Ericksen-Leslie模型提出模块grad-div稳定化有限元方法. 第3节得到了该方法的误差估计. 第4节进行了一些数值实验, 以验证本文提出算法的理论分析的有效性.

2 简化Ericksen-Leslie模型的模块grad-div稳定化方法

为了引入简化的Ericksen-Leslie模型方程(1.1)–(1.5)的模块grad-div稳定化有限元离散格式, 我们将给出一些记号.

首先, 我们记Lebesgue空间为$ L^{p}(\Omega) $和Sobolve空间为$ W_{p}^{m}(\Omega), $$ m\in{\mathbb N^{+}}, $$ 1\leq p\leq \infty, $它们的范数分别由$ \|\cdot\|_{L^{p}(\Omega)} $$ \|\cdot\|_{W^{m}_{p}(\Omega)} $表示. 另外, $ H^{m}(\Omega) $用于表示特定的Sobolev空间$ W^{m}_{2}(\Omega) $$ \|\cdot\|_{m} $则表示$ H^{m}(\Omega) $中的范数. $ \|\cdot\|_0 $$ (\cdot, \cdot) $表示$ L^2(\Omega) $的范数和内积. 对于$ \Omega $中的函数空间$ X $, $ L^p(0, T;X) $表示$ Q_T $上定义的所有函数的空间, 其范数

是有限的.

此外, 对于简化Ericksen-Leslie方程(1.1)–(1.5), 我们引入以下函数空间

然后, 根据上述函数空间的上述定义, 我们给出方程(1.1)–(1.5)的变分形式对所有的$ ({\bf v}, q, \phi)\in {\bf V}\times M\times {\bf W} $$ ({\bf u}, p, {\bf d})\in L^{2}(0, T;{\bf V})\times L^{2}(0, T;M)\times L^{2}(0, T;{\bf W}) $

$ \begin{equation} ({\bf u}_{t}, {\bf v})+\nu(\nabla {\bf u}, \nabla{\bf v})-(p, \nabla\cdot{\bf v}) +(\nabla\cdot{\bf u}, q)+(({\bf u}\cdot\nabla){\bf u}, {\bf v})\\ -\lambda(\nabla {\bf d}\odot\nabla{\bf d}, \nabla{\bf v}) = ({\bf f}, {\bf v}), \end{equation} $

$ \begin{equation} ({\bf d}_{t}, \phi)+\gamma (\nabla{\bf d}, \nabla{\phi}) -\gamma(|\nabla {\bf d}|^{2}{\bf d}, \phi)+(({\bf u}\cdot\nabla){\bf d}, {\phi}) = 0. \end{equation} $

我们令$ N $为大于0的整数, $ \{t_m\}_{m = 0}^{N} $为区间$ [0, T] $的均匀剖分, 时间步长$ \Delta t = \frac{T}{N}, $$ t_m = m\Delta t, $并对区域$ \Omega $做一致剖分$ K_h, $该网格由三角形单元$ K $组成, 网格步长$ h = \max_{K\in K_h}h_K, $其中$ h_K $是单元$ K $的直径. 接下来, 我们在$ K_h $上定义以下有限元子空间

其中$ P_i(K) $$ (i = 0, 1, 2) $表示$ K\in K_{h} $上阶数最多为$ i $次的多项式空间. 需要注意的是, 最低阶协调有限元对$ {\bf V}_{h}\times M_{h} $不满足离散的inf-sup条件. 因此, 本文为了满足该条件, 使用稳定的双线性项[6, 35]

其中$ \Pi $是从$ L^{2}(\Omega) $$ X_h $的投影算子[6, 32-33].

另外, 本文引入了广义双线性形式, 对所有的$ ({\bf u}_{h}, p_{h}), ({\bf v}_{h}, q_{h})\in {\bf V}_{h}\times M_{h} $

以及反对称三线性形式, 对所有$ {\bf u}_h\in{\bf V}_h, $$ {\bf v}_h, \ {\bf w}_h\in {\bf V}_h $$ {\bf W}_h $

$ \begin{eqnarray} b({\bf u}_h, {\bf v}_h, {\bf w}_h)& = &(({\bf u}_h\cdot\nabla){\bf v}_h, {\bf w}_h) +\frac{1}{2}((\nabla\cdot{\bf u}_h){\bf v}_h, {\bf w}_h){}\\ & = &\frac{1}{2}(({\bf u}_h\cdot\nabla){\bf v}_h, {\bf w}_h)-\frac{1}{2}(({\bf u}_h\cdot\nabla){\bf w}_h, {\bf v}_h), \end{eqnarray} $

该三线性项满足以下性质[28]

$ \begin{equation} |b({\bf u}_h, {\bf v}, {\bf w}_h)|\leq C\|{\bf u}_h\|_{0}\|{\bf v}\|_{2}\|\nabla{\bf w}_h\|_{0}, \end{equation} $

对所有的$ {\bf u}_h, {\bf w}_h\in {\bf V}_h, $$ {\bf W}_h $$ {\bf v}\in H^2(\Omega)^2 $都成立. 我们设$ C $为一个正常数, 该常数与网格大小$ h $和时间步长$ \Delta t $无关, 并且在不同情况下可以表示不同的值. 本文应用了逆不等式[7], 它适用于$ \textbf{v}_h\in{\bf V}_h $$ {\bf W}_h, $

$ \begin{equation} \|\textbf{v}_h\|_{W^{l}_{p}(\Omega)^2}\leq Ch^{s-l+2\min\{0, \frac{1}{p}-\frac{1}{q}\}}\|\textbf{v}_h\|_{W^s_{q}(\Omega)^2}, \quad 1\leq p, q\leq\infty, \; \; 0\leq s\leq l. \end{equation} $

在本节的其余部分, 我们用$ ({\bf u}_{h}^{m}, p_{h}^{m}, {\bf d}_{h}^{m}) $完全离散逼近$ ({\bf u}(t_m), p(t_m), {\bf d}(t_m)). $接下来我们准备为简化的Ericksen-Leslie模型提供模块grad-div稳定化有限元方法, 如下所示.

算法 2.1  第一步  我们给定$ {\bf u}_{h}^{m}\in {\bf V}_{h} $$ {\bf d}_{h}^{m}\in{\bf W}_{h}, $对所有的$ ({\bf v}_h, q_h, \phi_{h})\in {\bf V}_{h}\times M_{h}\times{\bf W}_{h} $找到$ ({\bf \hat{u}}_{h}^{m+1}, p_{h}^{m+1}, {\bf d}_{h}^{m+1})\in {\bf V}_{h}\times M_{h}\times{\bf W}_{h}, $

$ \begin{eqnarray} && \left(\frac{{\bf d}_{h}^{m+1}-{\bf d}_{h}^{m}}{\Delta t}, \phi_{h}\right)+\gamma (\nabla{\bf d}_h^{m+1}, \nabla{\phi}_{h})-\gamma(|\nabla {\bf d}_{h}^{m}|^{2}{\bf d}_{h}^{m}, \phi_{h}) +b({\bf u}_{h}^{m}, {\bf d}_{h}^{m+1}, \phi_{h}) = 0, \end{eqnarray} $

$ \begin{eqnarray} && \left(\frac{{\bf \hat{u}}_{h}^{m+1}-{\bf u}_{h}^{m}}{\Delta t}, {\bf v}_{h}\right)+B({\bf \hat{u}}_{h}^{m+1}, p_{h}^{m+1};{\bf v}_{h}, q_{h}) +b({\bf u}_{h}^{m}, {\bf \hat{u}}_{h}^{m+1}, {\bf v}_{h}){}\\ &&+\lambda(\nabla {\bf d}_{h}^{m+1}\odot \nabla {\bf d}_{h}^{m}, \nabla{\bf v}_{h}) = ({\bf f}^{m+1}, {\bf v}_{h}). \end{eqnarray} $

第二步  根据从(2.6)式和(2.7)式得到的$ {\bf \hat{u}}_{h}^{m+1} $, 找到$ {\bf u}_{h}^{m+1}\in{\bf V}_{h}, $使得对于所有的$ {\bf v}_h\in {\bf V}_{h} $

$ \begin{equation} ({\bf u}_{h}^{m+1}, {\bf v}_{h})+(\beta+\chi\Delta t)(\nabla\cdot{\bf u}_{h}^{m+1}, \nabla\cdot{\bf v}_{h}) = ({\bf \hat{u}}_{h}^{m+1}, {\bf v}_{h})+\beta(\nabla\cdot{\bf u}_{h}^{m}, \nabla\cdot{\bf v}_{h}), \end{equation} $

其中$ \beta>0 $$ \chi>0 $是两个稳定化参数.

注 2.1  算法2.1中变量的计算顺序如下. 首先该算法由$ {\bf d}_h^{m} $求出$ {\bf d}_h^{m+1}, $然后根据$ {\bf u}_h^{m}, $$ {\bf d}_h^{m} $求出$ {\bf \hat{u}}_{h}^{m+1} $$ p^{m+1}_h, $再由方程(2.6)求出$ {\bf d}_h^{m+1} $, 最后根据$ {\bf u}_h^{m} $求出$ {\bf u}_h^{m+1} $和方程(2.7)求出$ {\bf \hat{u}}_{h}^{m+1} $. 需要注意的是算法2.1的第一步是参见文献[3]中提出的算法. 为了保持质量守恒性并提高求解精度, 并且不会出现由大稳定化参数而引起的计算时间过长的问题, 本文引入了第二步.

3 误差分析

在本节中, 我们证明了该算法的数值解收敛于真解. 本文首先介绍了下面的投影[3, 7], $ ({\bf R}_{h}, Q_{h}): {\bf V}\times M\longrightarrow{\bf V}_{h}\times M_{h} $

$ \begin{equation} {B}({\bf R}_{h}{\bf v}, Q_{h}q;{\bf v}_{h}, q_{h}) = {B}({\bf v}, q;{\bf v}_{h}, q_{h})-G(q, q_h), \end{equation} $

对于$ ({\bf v}, q)\in{\bf V}\times M $$ ({\bf v}_{h}, q_{h})\in{\bf V}_{h}\times M_{h}, $上述投影满足以下性质[3, 7, 18]

$ \begin{equation} \begin{array}{l} \|{\bf v}(t_m)-{\bf R}_{h}{\bf v}(t_m)\|_{0} +h\|\nabla({\bf v}(t_m)-{\bf R}_{h}{\bf v}(t_m))\|_0\leq Ch^{2}\|{\bf v}(t_m)\|_{2}, \\ \|q(t_m)-{Q}_{h}q(t_m)\|_0\leq Ch\|q(t_m)\|_1, \end{array} \end{equation} $

其中$ ({\bf v}, q)\in{\bf V}\cap H^2(\Omega)^2\times M\cap H^1(\Omega) $$ 0\leq m\leq N. $此外, 本文还定义了投影算子[3, 7]$ {\bf P}_h: {\bf W}\rightarrow {\bf W}_{h} $

$ \begin{equation} (\nabla({\bf d}(t_{m})-{\bf P}_h{\bf d}(t_{m})), \nabla{\phi}_h) = 0, \quad\forall{\bf d}\in{\bf W}, \; \phi_h\in{\bf W}_h. \end{equation} $

上述投影算子满足以下性质[3, 7, 18]

$ \begin{equation} \|{\bf d}(t_{m})-{\bf P}_{h}{\bf d}(t_{m})\|_{0} +h\|\nabla({\bf d}(t_{m})-{\bf P}_{h}{\bf d}(t_{m}))\|_{0}\leq Ch^{3}\|{\bf d}(t_{m})\|_{3}, \end{equation} $

其中$ {\bf d}\in H^3(\Omega)^2\cap {\bf W}. $

本文标记$ (\tilde{{\bf u}}^m, \tilde{p}^m, \tilde{{\bf d}}^{m}) = ({\bf R}_h{\bf u}(t_m), Q_hp(t_m), {\bf P}_h{\bf d}(t_{m})) $, 然后拆分数值误差如下

假设$ {\bf e}^{0} = 0, \ {\bm \epsilon}^{0} = 0. $

为了获得误差方程, 我们在方程(2.1)中设$ ({\bf v}, q, \phi) = ({\bf v}_h, q_h, \phi_h) $, 则由(3.1)式和(3.3)式可以得到

$ \begin{eqnarray} &&\left(\frac{\tilde{{\bf d}}^{m+1}-\tilde{{\bf d}}^{m}}{\Delta t}, \phi_{h}\right)+\gamma (\nabla\tilde{{\bf d}}^{m+1}, \nabla{\phi}_{h})-\gamma(|\nabla {\bf d}(t_{m+1})|^{2}{\bf d}(t_{m+1}), \phi_{h}){}\\ & &+b({\bf u}(t_{m+1}), {\bf d}(t_{m+1}), \phi_{h}) = (\mathit{\pmb{\omega}}_{d}^{m+1}, \phi_{h}), \end{eqnarray} $

$ \begin{eqnarray} &&\left(\frac{\tilde{{\bf u}}^{m+1}-\tilde{{\bf u}}^{m}}{\Delta t}, {\bf v}_{h}\right)+B(\tilde{{\bf u}}^{m+1}, \tilde{p}^{m+1};{\bf v}_{h}, q_{h}) +b({\bf u}(t_{m+1}), {\bf u}(t_{m+1}), {\bf v}_{h}){}\\ &&-\lambda(\nabla{\bf d}(t_{m+1})\odot \nabla {\bf d}(t_{m+1}), \nabla{\bf v}_{h}) = (\mathit{\pmb{\omega}}_{u}^{m+1}, {\bf v}_{h})+({\bf f}^{m+1}, {\bf v}_{h}), \end{eqnarray} $

其中

$ \begin{equation} \mathit{\pmb{\omega}}_{u}^{m+1} = \frac{\tilde{{\bf u}}^{m+1}-\tilde{{\bf u}}^{m}}{\Delta t}-{\bf u}_{t}(t_{m+1}), \quad \mathit{\pmb{\omega}}_{d}^{m+1} = \frac{\tilde{{\bf d}}^{m+1} -\tilde{{\bf d}}^{m}}{\Delta t}-{\bf d}_{t}(t_{m+1}). \end{equation} $

用(3.5)式减去(2.6)式得到

$ \begin{eqnarray} &&\left(\frac{{\bm \epsilon}^{m+1}-{\bm \epsilon}^{m}}{\Delta t}, \phi_{h}\right) +\gamma (\nabla{\bm \epsilon}^{m+1}, \nabla{\phi}_{h})-\gamma(|\nabla {\bf d}(t_{m+1})|^{2}{\bf d}(t_{m+1}) -|\nabla {\bf d}_{h}^{m}|^{2}{\bf d}_{h}^{m}, \phi_{h}){}\\ &&+b({\bf u}(t_{m+1}), {\bf d}(t_{m+1}), \phi_{h}) -b({\bf u}_{h}^{m}, {\bf d}_{h}^{m+1}, \phi_{h}) = (\mathit{\pmb{\omega}}_{d}^{m+1}, \phi_{h}). \end{eqnarray} $

用(3.6)式减去(2.7)式得到

$ \begin{eqnarray} & &\left(\frac{{\bf \hat{e}}^{m+1}-{\bf e}^{m}}{\Delta t}, {\bf v}_{h}\right)+B({\bf \hat{e}}^{m+1}, \eta^{m+1};{\bf v}_{h}, q_{h}) +b({\bf u}(t_{m+1}), {\bf u}(t_{m+1}), {\bf v}_{h}) -b({\bf u}_{h}^{m}, {\bf \hat{u}}_{h}^{m+1}, {\bf v}_{h}){}\\ &&-\lambda(\nabla{\bf d}(t_{m+1})\odot \nabla {\bf d}(t_{m+1})-\nabla {\bf d}_{h}^{m+1}\odot \nabla {\bf d}_{h}^{m}, \nabla{\bf v}_{h}) = (\mathit{\pmb{\omega}}_{u}^{m+1}, {\bf v}_{h}). \end{eqnarray} $

算法2.1的第二步, 我们在误差分析中引用了文献[13, 引理10]中的一个重要引理.

引理 3.1[13]  考虑算法2.1的第二步, 我们可以得到

引理 3.2  考虑算法2.1的第一步, 我们可以得到

  在(3.8)式中取$ {\phi}_{h} = {\bm \epsilon}^{m+1}, $我们可以得到

$ \begin{eqnarray} &&\frac{1}{2\Delta t}\|{\bm \epsilon}^{m+1}\|_{0}^{2} +\frac{1}{2\Delta t}\|{\bm \epsilon}^{m+1}-{\bm \epsilon}^{m}\|_{0}^{2} -\frac{1}{2\Delta t}\|{\bm \epsilon}^{m}\|_{0}^{2} +\gamma\|\nabla{\bm \epsilon}^{m+1}\|_{0}^{2}{}\\ &\leq& |(\mathit{\pmb{\omega}}_{d}^{m+1}, {\bm \epsilon}^{m+1})| +|b({\bf u}_h^{m}, {\bf d}_{h}^{m+1}, {\bm \epsilon}^{m+1}) -b({\bf u}(t_{m+1}), {\bf d}(t_{m+1}), {\bm \epsilon}^{m+1})|{}\\ &&+\gamma|(|\nabla {\bf d}(t_{m+1})|^{2} {\bf d}(t_{m+1})- |\nabla {\bf d}_{h}^{m}|^{2}{\bf d}_{h}^{m}, {\bm \epsilon}^{m+1})| = :I_{1}+I_{2}+I_{3}. \end{eqnarray} $

我们首先应用Cauchy-Schwarz和Young不等式估计$ I_1 $如下

$ \begin{equation} I_{1}\leq \frac{1}{2}(\|\mathit{\pmb{\omega}}_{d}^{m+1}\|_{0}^{2} +\|{\bm \epsilon}^{m+1}\|_{0}^{2}), \end{equation} $

其次, 通过简单的计算使得

$ \begin{eqnarray} I_{2}& \leq&|b({\bf e}^m, {\bf d}(t_{m+1}), {\bm \epsilon}^{m+1})| +|b({\bf e}_c^m, {\bf d}(t_{m+1}), {\bm \epsilon}^{m+1})| {}\\ &&+|b({\bf u}(t_m)-{\bf u}(t_{m+1}), {\bf d}(t_{m+1}), {\bm \epsilon}^{m+1})| +|b({\bf e}^m, {\bm \epsilon}_c^{m+1}, {\bm \epsilon}^{m+1})| {}\\ &&+|b({\bf e}_c^m, {\bm \epsilon}_c^{m+1}, {\bm \epsilon}^{m+1})| +|b({\bf u}(t_m), {\bm \epsilon}_c^{m+1}, {\bm \epsilon}^{m+1})| = :\sum\limits_{i = 1}^6I_{2}^{i}, \end{eqnarray} $

然后, 由(2.4)式, 投影性质(3.2)式, (3.4)式和Young不等式可以得出

我们将上述估计值与正则性假设相结合可以得到

$ \begin{equation} I_2\leq\frac{\gamma}{4}\|\nabla{\bm \epsilon}^{m+1}\|_0^2 +\frac{C}{\gamma}(1+h^4)\|{\bf e}^m\|_0^2 +\frac{C\Delta t}{\gamma}\|{\bf u}_{t}(t)\|_{L^2(t_{m}, t_{m+1};L^2(\Omega)^2)}^2 +Ch^4. \end{equation} $

为了估计$ I_3, $我们把$ |\nabla {\bf d}(t_{m+1})|^{2} {\bf d}(t_{m+1})- |\nabla {\bf d}_{h}^{m}|^{2}{\bf d}_{h}^{m} $重写为

接下来, 由Hölder和Young不等式, 我们得到

$ \begin{eqnarray} I_{3}&\leq& \gamma \|\nabla{\bf d}(t_{m+1})+\nabla{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2} \|{\bf d}(t_{m+1})\|_{L^{\infty}(\Omega)^2}\|\nabla{\bf d}(t_{m+1})-\nabla{\bf d}(t_{m})\|_{0}\|{\bm \epsilon}^{m+1}\|_{0}{}\\ &&+\gamma \|\nabla{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}^{2}\|{\bf d}(t_{m+1})-{\bf d}(t_{m})\|_{0}\|{\bm \epsilon}^{m+1}\|_{0}{}\\ &&+\gamma\|\nabla{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}^{2}\|{\bm \epsilon}^{m} +{\bm \epsilon}_{c}^{m}\|_{0}\|{\bm \epsilon}^{m+1}\|_{0}{}\\ &&+2\gamma\|\nabla{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}\|{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}\|\nabla{\bm \epsilon}^{m} +\nabla{\bm \epsilon}_{c}^{m}\|_{0}\|{\bm \epsilon}^{m+1}\|_{0}{}\\ &&+2\gamma\|\nabla{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}\|\nabla{\bm \epsilon}^{m} +\nabla{\bm \epsilon}_{c}^{m}\|_{0}\|{\bm \epsilon}^{m} +{\bm \epsilon}_{c}^{m}\|_{L^{\infty}(\Omega)^2}\|{\bm \epsilon}^{m+1}\|_{0}{}\\ &&+\gamma\|{\bm \epsilon}^{m}+{\bm \epsilon}_{c}^{m}\|_{L^{\infty}(\Omega)^2}\|\nabla{\bm \epsilon}^{m} +\nabla{\bm \epsilon}_{c}^{m}\|_{0}\|\nabla{\bm \epsilon}^{m} +\nabla{\bm \epsilon}_{c}^{m}\|_{L^{3}(\Omega)^2}\|{\bm \epsilon}^{m+1}\|_{L^{6}(\Omega)^2}{}\\ &&+\gamma\|{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}\|\nabla{\bm \epsilon}^{m} +\nabla{\bm \epsilon}_{c}^{m}\|_{0}\|\nabla{\bm \epsilon}^{m} +\nabla{\bm \epsilon}_{c}^{m_{k}}\|_{L^{3}(\Omega)^2}\|{\bm \epsilon}^{m+1}\|_{L^{6}(\Omega)^2}{}\\ & = &:\sum\limits_{i = 1}^7I_{3}^{i}. \end{eqnarray} $

为了估计(3.14)式中的每个$ I_3^i $, 本文应用Cauchy-Schwarz和Young不等式和(3.4)式可以得到

以及

再综合上述估计可以得到

$ \begin{eqnarray} I_{3}&\leq & \frac{\gamma}{4}\|\nabla{\bm \epsilon}^{m+1}\|_{0}^{2}+\frac{C\Delta t}{\gamma} \|{\bf d}_{t}(t)\|^2_{L^{2}(t_{m}, t_{m+1};H^1(\Omega)^2)} +C h^{4}+C\gamma\|\nabla{\bm \epsilon}^{m}\|_{0}^2 {}\\ &&+C\gamma\|\nabla{\bm \epsilon}^{m}\|_{0}^4 +C\gamma\|\nabla{\bm \epsilon}^{m}\|_{0}^6. \end{eqnarray} $

最后, 我们把(3.11)式, (3.13)式和(3.15)式与(3.10)式结合起来, 将不等式乘以$ 2\Delta t $并求和$ m = 0, 1, \cdots, N-1, $得到

$ \begin{eqnarray} &&\|{\bm \epsilon}^{N}\|_{0}^{2}+\gamma\Delta t \sum\limits_{m = 0}^{N-1}\|\nabla{\bm \epsilon}^{m+1}\|_{0}^{2} \leq \Delta t\sum\limits_{m = 0}^{N-1}\|\mathit{\pmb{\omega}}_{d}^{m+1}\|_{0}^{2} +\Delta t\sum\limits_{m = 0}^{N-1}\|{\bm \epsilon}^{m+1}\|_{0}^{2} +\frac{C\Delta t}{\gamma}\sum\limits_{m = 0}^{N-1}\|{\bf e}^m\|_0^2{}\\ &&+Ch^{4}+\frac{C\Delta t^2}{\gamma}\sum\limits_{m = 0}^{N-1}\|{\bf u}_{t}(t)\|_{L^2(t_{m}, t_{m+1};L^2(\Omega)^2)}^2 +\frac{C\Delta t^2}{\gamma}\sum\limits_{m = 0}^{N-1} \|{\bf d}_{t}(t)\|^2_{L^{2}(t_{m}, t_{m+1};H^1(\Omega)^2)}{}\\ & &+C\gamma\Delta t\sum\limits_{m = 0}^{N-1}(\|\nabla{\bm \epsilon}^{m}\|_{0}^2 +\|\nabla{\bm \epsilon}^{m}\|_{0}^4 +\|\nabla{\bm \epsilon}^{m}\|_{0}^6). \end{eqnarray} $

引理3.2证毕.

引理 3.3  考虑算法2.1的第一步, 我们可以得到

  我们在(3.9)式中设$ ({\bf v}_{h}, q_{h}) = ({\bf \hat{e}}^{m+1}, \eta^{m+1}) $使得

$ \begin{eqnarray} &&\frac{1}{2\Delta t}\|{\bf \hat{e}}^{m+1}\|_{0}^{2}+\frac{1}{2\Delta t} \|{\bf \hat{e}}^{m+1}-{\bf e}^{m}\|_{0}^{2}-\frac{1}{2\Delta t}\|{\bf e}^{m}\|_{0}^{2}+\nu\|\nabla {\bf \hat{e}}^{m+1}\|_{0}^{2}+ \|\eta^{m+1}-\Pi\eta^{m+1}\|_{0}^{2}{}\\ &\leq&|(\mathit{\pmb{\omega}}_{{\bf u}}^{m+1}, {\bf \hat{e}}^{m+1}) +|b({\bf u}(t_{m+1}), {\bf u}(t_{m+1}), {\bf \hat{e}}^{m+1}) -b({\bf u}_{h}^{m}, {\bf \hat{u}}_{h}^{m+1}, {\bf \hat{e}}^{m+1})|{}\\ &&+\lambda|(\nabla{\bf d}(t_{m+1})\odot \nabla {\bf d}(t_{m+1})-\nabla {\bf d}_{h}^{m+1}\odot \nabla {\bf d}_{h}^{m}, \nabla {\bf \hat{e}}^{m+1})| = : I_{4}+I_{5}+I_{6}. \end{eqnarray} $

考虑到Cauchy-Schwarz和Young不等式, 本文把$ I_{4} $估计为

$ \begin{equation} I_{4} \leq \frac{C}{\nu}\|\mathit{\pmb{\omega}}_{u}^{m+1}\|_{0}^{2}+ \frac{\nu}{6}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}^{2}. \end{equation} $

$ I_5 $加上和减去一些三线性项可以得到

$ \begin{eqnarray} I_5&\leq&\left|b({\bf u}_h^m-{\bf u}(t_{m+1}), {\bf \hat{u}}_h^{m+1}, {\bf \hat{e}}^{m+1})\right| +\left|b({\bf u}(t_{m+1}), {\bf \hat{u}}_h^{m+1}-{\bf u}(t_{m+1}), {\bf \hat{e}}^{m+1})\right|{}\\ &\leq&\left|b({\bf u}_h^m-\tilde{{\bf u}}^m, {\bf \hat{u}}_h^{m+1}, {\bf \hat{e}}^{m+1})\right| +\left|b(\tilde{{\bf u}}^m-{\bf u}(t_{m}), {\bf \hat{u}}_h^{m+1}, {\bf \hat{e}}^{m+1})\right|{}\\ &&+\left|b({\bf u}(t_{m})-{\bf u}(t_{m+1}), {\bf \hat{u}}_h^{m+1}, {\bf \hat{e}}^{m+1})\right| +\left|b({\bf u}(t_{m+1}), {\bf \hat{u}}_h^{m+1}-{\bf u}(t_{m+1}), {\bf \hat{e}}^{m+1})\right|{}\\ &\leq& \left|b({\bf e}^m, {\bf e}_c^{m+1}, {\bf \hat{e}}^{m+1})\right| +\left|b({\bf e}^m, {\bf u}(t_{m+1}), {\bf \hat{e}}^{m+1})\right|+\left|b({\bf e}_c^m, {\bf e}_c^{m+1}, {\bf \hat{e}}^{m+1})\right|{}\\ && +\left|b({\bf e}_c^m, {\bf u}(t_{m+1}), {\bf \hat{e}}^{m+1})\right|+\left|b({\bf u}(t_{m})-{\bf u}(t_{m+1}), {\bf u}(t_{m+1}), {\bf \hat{e}}^{m+1})\right|{}\\ &&+\left|b({\bf u}(t_{m+1}), {\bf e}_c^{m+1}, {\bf \hat{e}}^{m+1})\right| +\left|b({\bf u}(t_{m})-{\bf u}(t_{m+1}), {\bf e}_c^{m+1}, {\bf \hat{e}}^{m+1})\right| = :\sum\limits_{i = 1}^7I_{5}^{i}. \end{eqnarray} $

结合(2.4)式与Young不等式, 我们推出

以及

我们把这些估计值与(3.19)式结合在一起可以得到

$ \begin{equation} I_5\leq \frac{C}{\nu}\left(\|{\bf e}^{m}\|_0^2+h^4 +\Delta t\|{\bf u}_{t}(t)\|^2_{L^2(t_{m}, t_{m+1};L^2(\Omega)^2)}\right) +\frac{\nu}{6}\|\nabla {\bf \hat{e}}^{m+1}\|_0^2. \end{equation} $

现在我们估计(3.17)式的最后一项

$ \begin{eqnarray} I_{6}&\leq&\lambda|(\nabla({\bm \epsilon}^{m+1}+{\bm \epsilon}_c^{m+1})\odot\nabla {\bf d}(t_{m+1}), \nabla{\bf \hat{e}}^{m+1})| {}\\ &&+\lambda|(\nabla({\bm \epsilon}^{m+1}+{\bm \epsilon}_c^{m+1}) \odot\nabla({\bf d}(t_{m})-{\bf d}(t_{m+1})), \nabla{\bf \hat{e}}^{m+1})|{}\\ &&+\lambda|(\nabla {\bf d}(t_{m+1})\odot\nabla({\bf d}(t_{m+1})-{\bf d}(t_{m})), \nabla{\bf \hat{e}}^{m+1})|{}\\ &&+\lambda|(\nabla(-{\bm \epsilon}^{m+1}-{\bm \epsilon}_c^{m+1})\odot \nabla({\bm \epsilon}^{m}+{\bm \epsilon}_c^{m}), \nabla{\bf \hat{e}}^{m+1})|{}\\ &&+\lambda|(\nabla {\bf d}(t_{m+1})\odot\nabla({\bm \epsilon}^{m}+{\bm \epsilon}_c^{m}), \nabla{\bf \hat{e}}^{m+1})| = :\sum\limits_{i = 1}^5I_{6}^{i}. \end{eqnarray} $

再次应用Hölder和Young不等式, 我们可以得到

$ \begin{eqnarray} I_{6}^{1}&\leq&\lambda\|\nabla{\bm \epsilon}^{m+1} +\nabla{\bm \epsilon}_c^{m+1}\|_{0}\|\nabla{\bf d}(t_{m+1})\|_{L^{\infty}(\Omega)^2} \|\nabla{\bf \hat{e}}^{m+1}\|_{0}{}\\ &\leq&\frac{C{\lambda}^{2}}{\nu}(\|\nabla{\bm \epsilon}^{m+1}\|^2 +h^4\|{\bf d}(t_{m+1})\|_3^2)\|\nabla{\bf d}(t_{m+1})\|_{L^{\infty}(\Omega)^2}^2 +\frac{\nu}{30}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}^{2}, \end{eqnarray} $

$ \begin{eqnarray} I_{6}^{5}&\leq&\lambda\|\nabla{\bf d}(t_{m+1})\|_{L^{\infty}(\Omega)^2} \|\nabla{\bm \epsilon}^{m}+\nabla{\bm \epsilon}_c^{m}\|_{0}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}{}\\ &\leq&\frac{C{\lambda}^{2}}{\nu}(\|\nabla{\bm \epsilon}^{m}\|_{0}^2 +h^4\|{\bf d}(t_m)\|_3^2)\|\nabla{\bf d}(t_{m+1})\|^2_{L^{\infty}(\Omega)^2} +\frac{\nu}{30}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}^{2}, \end{eqnarray} $

以及

$ \begin{eqnarray} I_{6}^{3}&\leq&\lambda\|\nabla{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}\|\nabla {\bf d}(t_{m+1})-\nabla{\bf d}(t_{m})\|_{0}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}{}\\ &\leq&\frac{C{\lambda}^{2}{\Delta t}}{\nu}\|\nabla{\bf d}(t_{m})\|^2_{L^{\infty}(\Omega)^2}\|{\bf d}_t\|^2_{L^{2}(t_m, t_{m+1};H^1(\Omega)^2)} +\frac{\nu}{30}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}^{2}. \end{eqnarray} $

由逆不等式(2.5), $ I_{6}^2 $$ I_{6}^4 $可以被估计为

$ \begin{eqnarray} I_{6}^{2}&\leq&\lambda\|\nabla{\bm \epsilon}^{m+1} +\nabla{\bm \epsilon}_c^{m+1}\|_{L^3(\Omega)^2}\|\nabla {\bf d}(t_{m+1})-\nabla{\bf d}(t_{m})\|_{L^6(\Omega)^2} \|\nabla{\bf \hat{e}}^{m+1}\|_{0}{}\\ &\leq&\frac{C{\lambda}^{2}h^{-2}}{\nu}\|\nabla{\bm \epsilon}^{m+1} +\nabla{\bm \epsilon}_c^{m+1}\|_0^2\|\nabla{\bf d}(t_{m+1})-\nabla{\bf d}(t_{m})\|_0^2 +\frac{\nu}{30}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}^{2}{}\\ &\leq&\frac{C{\lambda}^{2}\Delta th^{-2}}{\nu}(\|\nabla{\bm \epsilon}^{m+1}\|_0^2+h^4\|{\bf d}(t_m)\|_3^2) \|{\bf d}_t\|^2_{L^{2}(t_m, t_{m+1};H^1(\Omega)^2)} +\frac{\nu}{30}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}^{2}, \end{eqnarray} $

$ \begin{eqnarray} I_{6}^{4}&\leq&\lambda\|\nabla{\bm \epsilon}^{m+1}+\nabla{\bm \epsilon}_c^{m+1}\|_{L^3(\Omega)^2} \|\nabla{\bm \epsilon}^{m}+\nabla{\bm \epsilon}_c^{m}\|_{L^6(\Omega)^2}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}{}\\ &\leq&\frac{C\lambda^2h^{-2}}{\nu} \|\nabla{\bm \epsilon}^{m+1}+\nabla{\bm \epsilon}_c^{m+1}\|_{0}^2\|\nabla{\bm \epsilon}^{m} +\nabla{\bm \epsilon}_c^{m}\|_{0}^2+\frac{\nu}{15}\|\nabla{\bf e}^{m+1}\|_0^2{}\\ &\leq&\frac{C\lambda^2h^{-2}}{\nu}(\|\nabla{\bm \epsilon}^{m+1}\|_0^2 +h^4\|{\bf d}(t_{m+1})\|_3^2)(\|\nabla{\bm \epsilon}^{m}\|_0^2+h^4\|{\bf d}(t_m)\|_3^2) +\frac{\nu}{30}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}^{2}.{\qquad} \end{eqnarray} $

因此, 我们将(3.22)–(3.26)式代入(3.21)式可以得到$ I_{6} $

$ \begin{eqnarray} I_{6}&\leq&\frac{\nu}{6}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}^{2} +\frac{C{\lambda}^{2}}{\nu}(\|\nabla{\bm \epsilon}^{m+1}\|^2 +h^4)+\frac{C{\lambda}^{2}}{\nu}(\|\nabla{\bm \epsilon}^{m}\|_{0}^2 +h^4){}\\ &&+\frac{C{\lambda}^{2}\Delta th^{-2}}{\nu}(\|\nabla{\bm \epsilon}^{m+1}\|_0^2 +h^4)\|{\bf d}_t\|^2_{L^{2}(t_m, t_{m+1};H^1(\Omega)^2)} +\frac{C{\lambda}^{2}{\Delta t}}{\nu}\|{\bf d}_t\|^2_{L^{2}(t_m, t_{m+1};H^1(\Omega)^2)} {}\\ &&+\frac{C\lambda^2h^{-2}}{\nu}(\|\nabla{\bm \epsilon}^{m+1}\|_0^2 +h^4)(\|\nabla{\bm \epsilon}^{m}\|_0^2+h^4). \end{eqnarray} $

接下来, 我们把(3.18)式, (3.20)式和(3.27)式与(3.17)式相结合, 乘以$ 2\Delta t $, 可以得到

$ \begin{eqnarray} &&\|{\bf \hat{e}}^{m+1}\|_{0}^{2} +\|{\bf \hat{e}}^{m+1}-{\bf e}^{m}\|_{0}^{2}-\|{\bf e}^{m}\|_{0}^{2}+\nu\Delta t\|\nabla {\bf \hat{e}}^{m+1}\|_{0}^{2}{}\\ &\leq &\frac{C\Delta t}{\nu}\|\mathit{\pmb{\omega}}_{u}^{m+1}\|_{0}^{2} +\frac{C\Delta t}{\nu}\|{\bf e}^{m}\|_0^2+C\Delta th^4+C\Delta t^2\|{\bf u}_{t}(t)\|^2_{L^2(t_{m}, t_{m+1};L^2(\Omega)^2)}{}\\ &&+\frac{C{\lambda}^{2}\Delta t}{\nu}(1+\Delta t h^{-2})\|\nabla{\bm \epsilon}^{m+1}\|_{0}^2+C\Delta t^2h^2 +\frac{C\lambda^2\Delta t^2}{\nu}\|{\bf d}_{t}(t)\|^2_{L^2(t_{m}, t_{m+1};H^1(\Omega)^2)}{}\\ &&+\frac{C\lambda^2\Delta th^{-2}}{\nu}(\|\nabla{\bm \epsilon}^{m+1}\|_0^2+h^4) (\|\nabla{\bm \epsilon}^{m}\|_0^2+h^4)+\frac{C{\lambda}^{2}\Delta t}{\nu}\|\nabla{\bm \epsilon}^{m}\|_{0}^2. \end{eqnarray} $

引理3.3证毕.

引理 3.4  考虑算法2.1的第一步有

  我们在(3.8)式中令$ \phi_{h} = \frac{1}{\Delta t}({\bm \epsilon}^{m+1}-{\bm \epsilon}^{m}) = :d_{t}{\bm \epsilon}^{m+1}, $可以得到

$ \begin{eqnarray} &&\|d_t{\bm \epsilon}^{m+1}\|_0^2+\frac{\gamma}{2\Delta t} \|\nabla{\bm \epsilon}^{m+1}\|_0^2-\frac{\gamma}{2\Delta t} \|\nabla{\bm \epsilon}^{m}\|_0^2 +\frac{\gamma}{2\Delta t}\|\nabla{\bm \epsilon}^{m+1}-\nabla{\bm \epsilon}^{m}\|_0^2{}\\ &\leq& |(\mathit{\pmb{\omega}}_{d}^{m+1}, d_t{\bm \epsilon}^{m+1})| +|b({\bf u}_h^{m}, {\bf d}_{h}^{m+1}, d_t{\bm \epsilon}^{m+1}) -b({\bf u}(t_{m+1}), {\bf d}(t_{m+1}), d_t{\bm \epsilon}^{m+1})|{}\\ &&+\gamma|(|\nabla {\bf d}(t_{m+1})|^{2} {\bf d}(t_{m+1})- |\nabla {\bf d}_{h}^{m}|^{2}{\bf d}_{h}^{m}, d_t{\bm \epsilon}^{m+1})| = :I_7+I_8+I_9. \end{eqnarray} $

我们由Cauchy-Schwarz和Young不等式可以推出

$ \begin{equation} I_{7}\leq C\|\mathit{\pmb{\omega}}_{d}^{m+1}\|_{0}^{2} +\frac{1}{6}\|d_t{\bm \epsilon}^{m+1}\|_{0}^{2}, \end{equation} $

通过使用与(3.12)式同样的方法, 能够得到

$ I_{8} $可以分别被估计为

综合上述分析我们可以得出

$ \begin{eqnarray} I_8&\leq&\frac{1}{6}\|d_t{\bm \epsilon}^{m+1}\|_0^2+C\|{\bf e}^m\|_0^2 +C\Delta t\|{\bf u}_{t}(t)\|_{L^2(t_{m}, t_{m+1};H^1(\Omega)^2)}^2{}\\ &&+ Ch^{-2}\|{\bf e}^m\|_0^2\|\nabla{\bm \epsilon}^{m}\|_0^2 +C\|\nabla{\bm \epsilon}^{m+1}\|_0^2+Ch^4. \end{eqnarray} $

此外, $ I_{9} $使用与(3.14)式相同的方法可以有

$ \begin{eqnarray} I_{9} &\leq& \gamma \|\nabla{\bf d}(t_{m+1})+\nabla{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2} \|{\bf d}(t_{m+1})\|_{L^{\infty}(\Omega)^2}\|\nabla{\bf d}(t_{m+1})-\nabla{\bf d}(t_{m})\|_0\|d_t{\bm \epsilon}^{m+1}\|_{0}{}\\ & &+\gamma\|\nabla{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}^{2}\|{\bf d}(t_{m+1})-{\bf d}(t_{m})\|_{0}\|d_t{\bm \epsilon}^{m+1}\|_{0}{}\\ &&+\gamma\|\nabla{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}^{2}\|{\bm \epsilon}^{m}+{\bm \epsilon}_{c}^{m}\|_{0}\|d_t{\bm \epsilon}^{m+1}\|_{0}{}\\ &&+2\gamma\|\nabla{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}\|{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2} \|\nabla{\bm \epsilon}^{m}+\nabla{\bm \epsilon}_{c}^{m}\|_{0}\|d_t{\bm \epsilon}^{m+1}\|_{0}{}\\ &&+2\gamma\|\nabla{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}\|\nabla{\bm \epsilon}^{m} +\nabla{\bm \epsilon}_{c}^{m}\|_{0}\|{\bm \epsilon}^{m} +{\bm \epsilon}_{c}^{m}\|_{L^{\infty}(\Omega)^2}\|d_t{\bm \epsilon}^{m+1}\|_{0}{}\\ & &+\gamma\|{\bm \epsilon}^{m}+{\bm \epsilon}_{c}^{m}\|_{L^{\infty}(\Omega)^2} \|\nabla{\bm \epsilon}^{m}+\nabla{\bm \epsilon}_{c}^{m}\|_{0} \|\nabla{\bm \epsilon}^{m}+\nabla{\bm \epsilon}_{c}^{m}\|_{L^{3}(\Omega)^2}\|d_t{\bm \epsilon}^{m+1}\|_{L^{6}(\Omega)^2}{}\\ & &+\gamma\|{\bf d}(t_{m})\|_{L^{\infty}(\Omega)^2}\|\nabla{\bm \epsilon}^{m} +\nabla{\bm \epsilon}_{c}^{m}\|_{0}\|\nabla{\bm \epsilon}^{m} +\nabla{\bm \epsilon}_{c}^{m}\|_{L^{3}(\Omega)^2}\|d_t{\bm \epsilon}^{m+1}\|_{L^{6}(\Omega)^2}{}\\ & = &:\sum\limits_{i = 1}^7I_{9}^{i}. \end{eqnarray} $

为了估计$ I_9, $本文应用Cauchy-Schwarz和Young不等式, 以及(2.5)式, 可以得到

$ \begin{eqnarray} I_{9}^{1}+I_{9}^{2}&\leq &C\gamma^2\Delta t\|{\bf d}_{t}(t)\|^2_{L^{2}(t_{m}, t_{m+1};H^1(\Omega)^2)} +\frac{1}{30}\|d_{t}{\bm \epsilon}^{m+1}\|_{0}^2, {}\\ I_{9}^{3}+I_{9}^{4}&\leq& C\gamma^2(h^{4}\|{\bf d}(t_{m})\|^2_3+\|\nabla{\bm \epsilon}^{m}\|_{0}^{2}) +\frac{1}{30}\|d_{t}{\bm \epsilon}^{m+1}\|_{0}^{2}, {}\\ I_{9}^{5}&\leq & C \gamma^2(h^{8}\|{\bf d}(t_{m})\|^4_3+\|\nabla{\bm \epsilon}^{m}\|_{0}^4) +\frac{1}{30}\|d_{t}{\bm \epsilon}^{m+1}\|_{0}, {}\\ I_{9}^{6}&\leq& C\gamma h^{-1} \|\nabla{\bm \epsilon}^{m}+\nabla{\bm \epsilon}_{c}^{m}\|_{0}^3\|d_{t}{\bm \epsilon}^{m+1}\|_{0}\\ &\leq& C\gamma^2 h^{-2} (h^{12}\|{\bf d}(t_{m})\|^6_3+\|\nabla{\bm \epsilon}^{m}\|_{0}^6) +\frac{1}{30} \|d_{t}{\bm \epsilon}^{m+1}\|_{0}^{2}, {}\\ I_{9}^{7}&\leq& C\gamma h^{-1}\|\nabla{\bm \epsilon}^{m}+\nabla{\bm \epsilon}_{c}^{m}\|^2_{0}\|d_{t}{\bm \epsilon}^{m+1}\|_{0}{}\\ &\leq & C\gamma^2 h^{-2} (h^{8}\|{\bf d}(t_{m})\|^4_3+\|\nabla{\bm \epsilon}^{m}\|_{0}^4)+\frac{1}{30} \|d_{t}{\bm \epsilon}^{m_+1}\|_{0}^{2}. {} \end{eqnarray} $

那么$ I_9 $可以被估计为

$ \begin{eqnarray} I_{9}&\leq & \frac{1}{6}\|d_t{\bm \epsilon}^{m+1}\|_{0}^{2}+C\gamma^2\Delta t\|{\bf d}_{t}(t)\|^2_{L^{2}(t_{m}, t_{m+1};H^1(\Omega)^2)}+C h^{4}{}\\ &&+C\gamma^2(\|\nabla{\bm \epsilon}^{m}\|_{0}^2 +\|\nabla{\bm \epsilon}^{m}\|_{0}^4+h^{-2}\|\nabla{\bm \epsilon}^{m}\|_{0}^4 +h^{-2}\|\nabla{\bm \epsilon}^{m}\|_{0}^6). \end{eqnarray} $

我们将(3.30)式, (3.31)式, (3.34)式与(3.29)式相结合, 乘以$ 2\Delta t $并求和$ m = 0, 1, \cdots, N-1, $可以得出

$ \begin{eqnarray} &&\gamma \|\nabla{\bm \epsilon}^{N}\|_0^2+\Delta t\sum\limits_{m = 0}^{N-1}\|d_t{\bm \epsilon}^{m+1}\|_0^2{}\\ &\leq& C\Delta t\sum\limits_{m = 0}^{N-1}\|\mathit{\pmb{\omega}}_{d}^{m+1}\|_{0}^{2} +C\Delta t\sum\limits_{m = 0}^{N-1}\|{\bf e}^m\|_0^2+Ch^4+C\Delta t^2{}\\ &&+C\gamma^2\Delta t\sum\limits_{m = 0}^{N-1}(\|\nabla{\bm \epsilon}^{m}\|_{0}^2 +\|\nabla{\bm \epsilon}^{m}\|_{0}^4 +h^{-2}\|\nabla{\bm \epsilon}^{m}\|_{0}^4 +h^{-2}\|\nabla{\bm \epsilon}^{m}\|_{0}^6){}\\ &&+ Ch^{-2}\Delta t\sum\limits_{m = 0}^{N-1}\|{\bf e}^m\|_0^2\|\nabla{\bm \epsilon}^{m}\|_0^2 +C\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla{\bm \epsilon}^{m+1}\|_0^2. \end{eqnarray} $

引理3.4证毕.

下面给出了所提出算法的误差估计.

定理 3.1  假设真解是光滑的. 如果时间步长$ \Delta t $和空间步长$ h $满足$ \Delta t\leq Ch^2, $则存在一个正常数$ h_0, $使得当$ h \geq h_0 $时以下估计成立

$ \begin{eqnarray} &&\|{\bf e}^{N}\|_{0}^{2}+\gamma \|\nabla{\bm \epsilon}^N\|_0^2+\|{\bm \epsilon}^{N}\|_{0}^{2} +\beta\|\nabla\cdot{\bf e}^{N}\|_0^2+\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla{\bf \hat{e}}^{m+1}\|_{0}^{2}{}\\ & &+\gamma\Delta t \sum\limits_{m = 0}^{N-1}\|\nabla{\bm \epsilon}^{m+1}\|_{0}^{2} +\chi\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla\cdot{\bf e}^{m+1}\|_0^2\leq C({\Delta t}^{2}+h^{4}). \end{eqnarray} $

  我们使用(3.1)式并将(3.28)式从0到$ N-1 $求和, 可以得到

$ \begin{eqnarray} &&\|{\bf e}^{N}\|_0^2+\beta\|\nabla\cdot{\bf e}^{N}\|_0^2 +\sum\limits_{m = 0}^{N-1}(\|{\bf \hat{e}}^{m+1}-{\bf e}^{m+1}\|_0^2 +\|{\bf \hat{e}}^{m+1}-{\bf e}^{m}\|_0^2 +\frac{\beta}{2}\|\nabla\cdot({\bf e}^{m+1}-{\bf e}^{m})\|_0^2){}\\ &&+\nu\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla {\bf \hat{e}}^{m+1}\|_{0}^{2}+\chi\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla\cdot{\bf e}^{m+1}\|_0^2{}\\ &\leq &C\Delta t^2+Ch^4+C\Delta th^2+\frac{C\Delta t}{\nu}\sum\limits_{m = 0}^{N-1}\|\mathit{\pmb{\omega}}_{u}^{m+1}\|_{0}^{2} +\frac{C\Delta t}{\nu}\sum\limits_{m = 0}^{N-1}\|{\bf e}^{m}\|_0^2 +\beta\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla\cdot{\bf e}^{m}\|_0^2{}\\ &&+C\sum\limits_{m = 0}^{N-1}(\chi\Delta t\|\nabla{\bf e}_c^{m+1}\|_0^2 +\beta(\Delta t+1)\|\nabla{\bf e}_{c, t}^{m+1}\|^2_{L^2(t_m, t_{m+1};L^2(\Omega)^2)}){}\\ &&+\frac{C{\lambda}^{2}\Delta t}{\nu}\sum\limits_{m = 0}^{N-1}\|\nabla{\bm \epsilon}^{m}\|_{0}^2 +\|{\bf e}^0\|_0^2 +\beta\|\nabla\cdot{\bf e}^0\|_0^2+\frac{C{\lambda}^{2}\Delta t}{\nu} (1+\Delta t h^{-2})\sum\limits_{m = 0}^{N-1}\|\nabla{\bm \epsilon}^{m+1}\|_{0}^2{}\\ &&+\frac{C\lambda^2\Delta th^{-2}}{\nu}\sum\limits_{m = 0}^{N-1}(\|\nabla{\bm \epsilon}^{m+1}\|_0^2 +h^4)(\|\nabla{\bm \epsilon}^{m}\|_0^2+h^4). \end{eqnarray} $

我们把(3.16)式, (3.35)式和(3.37)式加起来, 可以得到

$ \begin{eqnarray} &&\|{\bf e}^{N}\|_0^2+\beta\|\nabla\cdot{\bf e}^{N}\|_0^2 +\|{\bm \epsilon}^{N}\|_{0}^{2} +\gamma \|\nabla{\bm \epsilon}^{N}\|_0^2{}\\ &&+\sum\limits_{m = 0}^{N-1}(\|{\bf \hat{e}}^{m+1}-{\bf e}^{m+1}\|_0^2 +\|{\bf \hat{e}}^{m+1}-{\bf e}^{m}\|_0^2 +\frac{\beta}{2}\|\nabla\cdot({\bf e}^{m+1}-{\bf e}^{m})\|_0^2) {}\\ &&+\nu\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla {\bf \hat{e}}^{m+1}\|_{0}^{2}+\chi\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla\cdot{\bf e}^{m+1}\|_0^2 +\gamma\Delta t \sum\limits_{m = 0}^{N-1}\|\nabla{\bm \epsilon}^{m+1}\|_{0}^{2} +\Delta t\sum\limits_{m = 0}^{N-1}\|d_t{\bm \epsilon}^{m+1}\|_0^2{}\\ &\leq& C\Delta t^2+Ch^4+C\Delta t\sum\limits_{m = 0}^{N-1}\|\mathit{\pmb{\omega}}_{d}^{m+1}\|_{0}^{2} +\frac{C\Delta t}{\nu}\sum\limits_{m = 0}^{N-1}\|\mathit{\pmb{\omega}}_{u}^{m+1}\|_{0}^{2} +\Delta t\sum\limits_{m = 0}^{N-1}\|{\bm \epsilon}^{m+1}\|_{0}^{2}{}\\ &&+C\gamma\Delta t\sum\limits_{m = 0}^{N-1}(\|\nabla{\bm \epsilon}^{m}\|_{0}^2 +\|\nabla{\bm \epsilon}^{m}\|_{0}^4 +\|\nabla{\bm \epsilon}^{m}\|_{0}^6) +\frac{C\Delta t}{\gamma}\sum\limits_{m = 0}^{N-1}\|{\bf e}^m\|_0^2 +\frac{C\Delta t}{\nu}\sum\limits_{m = 0}^{N-1}\|{\bf e}^{m}\|_0^2{}\\ &&+C\Delta t\sum\limits_{m = 0}^{N-1}\|{\bf e}^m\|_0^2+C\sum\limits_{m = 0}^{N-1} (\chi\Delta t\|\nabla{\bf e}_c^{m+1}\|_0^2+\beta(\Delta t+1)\|\nabla{\bf e}_{c, t}^{m+1}\|^2_{L^2(t_m, t_{m+1};L^2(\Omega)^2)}){}\\ &&+C\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla{\bm \epsilon}^{m+1}\|_{0}^2 +\beta\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla\cdot {\bf e}^m\|_0^2+\frac{C\lambda^2\Delta t}{\nu}(1+\Delta t h^{-2}) \sum\limits_{m = 0}^{N-1}\|\nabla{\bm \epsilon}^{m+1}\|_0^2{}\\ &&+\frac{C\lambda^2\Delta th^{-2}}{\nu}\sum\limits_{m = 0}^{N-1}(\|\nabla{\bm \epsilon}^{m+1}\|_0^2 +h^4)(\|\nabla{\bm \epsilon}^{m}\|_0^2 +h^4)+Ch^{-2}\Delta t\sum^{N-1}_{m = 0}\|{\bf e}^m\|_0^2\|\nabla{\bm \epsilon}^{m}\|_{0}^2{}\\ &&+C\gamma^2\Delta t\sum\limits_{m = 0}^{N-1}(\|\nabla{\bm \epsilon}^{m}\|_{0}^2 +\|\nabla{\bm \epsilon}^{m}\|_{0}^4 +h^{-2}\|\nabla{\bm \epsilon}^{m}\|_{0}^4 +h^{-2}\|\nabla{\bm \epsilon}^{m}\|_{0}^6). \end{eqnarray} $

根据Cauchy-Schwarz不等式可以有

$ \begin{eqnarray} \|\mathit{\pmb{\omega}}_{u}^{m+1}\|_0&\leq&\left\|\frac{\tilde{{\bf u}}^{m+1} -\tilde{{\bf u}}^{m}}{\Delta t}-\frac{{\bf u}(t_{m+1})-{\bf u}(t_{m})}{\Delta t}\right\|_0+ \left\|\frac{{\bf u}(t_{m+1})-{\bf u}(t_{m})}{\Delta t}-{\bf u}_{t}(t_{m+1})\right\|_0{}\\ &\leq&\frac{1}{\Delta t^{1/2}}\left(\int_{t_{m}}^{t_{m+1}}\|\tilde{{\bf u}}_{t}-{\bf u}_{t}\|_{0}^{2} \mbox{d}t\right)^{1/2} +\Delta t^{1/2}\left(\int_{t_{m}}^{t_{m+1}}\|{\bf u}_{tt}\|_{0}^{2}\mbox{d}t\right)^{1/2} \end{eqnarray} $

$ \begin{eqnarray} \|\mathit{\pmb{\omega}}_{d}^{m+1}\|_0&\leq&\left\|\frac{\tilde{{\bf d}}^{m+1} -\tilde{{\bf d}}^{m}}{\Delta t}-\frac{{\bf d}(t_{m+1})-{\bf d}(t_{m})}{\Delta t}\right\|_0+ \left\|\frac{{\bf d}(t_{m+1})-{\bf d}(t_{m})}{\Delta t}-{\bf d}_{t}(t_{m+1})\right\|_0{}\\ &\leq&\frac{1}{\Delta t^{1/2}}\left(\int_{t_{m}}^{t_{m+1}}\|\tilde{{\bf d}}_{t}-{\bf d}_{t}\|_{0}^{2} \mbox{d}t\right)^{1/2} +\Delta t^{1/2}\left(\int_{t_{m}}^{t_{m+1}}\|{\bf d}_{tt}\|_{0}^{2} \mbox{d}t\right)^{1/2}. \end{eqnarray} $

使用(3.39)式, (3.40)式和投影算子的近似性质, 我们可以得到以下估计

$ \begin{equation} \Delta t\sum\limits_{m = 0}^{N-1}\|\mathit{\pmb{\omega}}_{u}^{m+1}\|_{0}^{2}+ \Delta t\sum\limits_{m = 0}^{N-1}\|\mathit{\pmb{\omega}}_{d}^{m+1}\|_{0}^{2} \leq C({\Delta t}^{2}+h^{4}). \end{equation} $

接下来本文使用数学归纳法证明当$ 0\leq m\leq N $$ \|\nabla{\bm \epsilon}^{m}\|_0\leq h $成立. 当$ m = 0 $时这个不等式显然成立. 如果我们假设当$ 0< m \leq N-1 $$ \Delta t \leq Ch^2 $成立, 那么应用(3.41)式可以把(3.38)式重写为

$ \begin{eqnarray} &&\|{\bf e}^N\|_{0}^{2}+\beta\|\nabla \cdot{\bf e}^N\|_0^2+\|{\bm \epsilon}^N\|_{0}^{2}+\gamma \|\nabla{\bm \epsilon}^N\|_0^2+\nu\Delta t\sum\limits_{m = 0}^{N-1} \|\nabla {\bf \hat{e}}^{m+1}\|_{0}^{2}+\gamma\Delta t \sum\limits_{m = 0}^{N-1}\|\nabla{\bm \epsilon}^{m+1}\|_{0}^{2}{}\\ &\leq& C\Delta t^2+Ch^4+C\Delta t\sum\limits_{m = 0}^{N-1}\|{\bf e}^{m+1}\|_{0}^{2} +C\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla{\bm \epsilon}^{m+1}\|_0^2{}\\ &&+\beta\Delta t\sum\limits_{m = 0}^{N-1}\|\nabla\cdot {\bf e}^m\|_0^2 +\Delta t\sum\limits_{k = 0}^{N-1}\|{\bm \epsilon}^{m+1}\|_{0}^{2}. \end{eqnarray} $

因此, 通过应用Gronwall引理我们可以得到

$ \begin{equation} \|\nabla{\bm \epsilon}^N\|_{0}^{2}\leq Ch^4\leq h^2, \quad \mbox{if}\quad h\leq \frac{1}{\sqrt{C}} = h_0, \end{equation} $

最后,再次结合Gronwall的引理和(3.43)式就完成了归纳法证明.

4 数值实验

本节将提供一些数值算例来检验前面各节中得到的数值理论. 在数值算例中, 初始值取为$ {\bf u}_{0} = 0, \ {\bf f} = 0, \ {\bf d}_{0} = ({\sin(a)}, {\cos(a)}), \ a = \pi(x^{2}+y^{2})^{2}. $

第一个数值算例主要是验证算法的收敛阶.由于该问题的精确解是未知的, 本文通过在非常细的网格($ h $ = 1/150) 上计算标准Galerkin方法, 把得出的数值解作为精确解, 同时把参数设为$ \lambda = \mu = \gamma = 1. $

为了验证收敛阶, 这个算例取网格步长为$ h $ = 1/20, 1/30和1/40, 设$ \beta = 0.2, \ \chi = 1000. $数值结果展示在表 12中, 从表中可以看到该算法运行良好, 并且像理论分析一样保持了收敛阶.

表 1   t = hT = 0.1处的数值误差和收敛阶

h$\|\nabla\cdot({\bf u}(t_m)-{\bf u}_h^m)\|_{0} $收敛阶$\|\nabla({\bf u}(t_m)- {\bf u}_h^m)\|_{0}$收敛阶
1/200.02547070.0301272
1/300.01711520.98050800.02076590.9177351
1/400.01292940.97490670.01351011.4942699

新窗口打开| 下载CSV


表 2   t = h2T = 0.1处的数值误差和收敛阶

h$\|\nabla({\bf u}(t_m)-{\bf u}_h^m)\|_{0} $收敛阶$\|\nabla({\bf d}(t_m)-{\bf d}_h^m)\|_{0}$收敛阶
1/300.02135870.0119277
1/400.01484251.26516210.00620362.2724003
1/500.01171681.05972570.00385392.1333538

新窗口打开| 下载CSV


第二个数值算例是模块grad-div稳定化有限元算法与An的算法(无grad-div稳定的标准算法)作对比, 在该数值实验中, 本文设置$ h = 1/100 $$ \Delta t = 1/40. $然后, 在$ 1.0e-4< \nu \leq 1.0 $范围内, 改变$ \nu $的值.

表 3中列出了通过这些算法在$ T = 0.1 $处得到的数值结果. 事实证明, 当粘度值减小时, An算法的数值误差变大. 但是, 本文提出的算法仍然可以正常计算.

表 3   随着υ减小的数值误差

υ$\|\nabla\cdot ({\bf u}(t_m)- {\bf u}_h^m)\|_{0} $
An的算法[3]
算法2.1$\|\nabla({\bf u}(t_m)- {\bf u}_h^m) \|_{0}$
An的算法[3]
算法2.1
1.00.1772910.0163640.1553930.015200
1.0e-10.4117600.0171991.7977900.049710
1.0e-20.0986200.0212415.6446000.122195
1.0e-31.1475600.0229557.0073300.143346
1.0e-41.1664600.0231737.1771500.145903

新窗口打开| 下载CSV


为了测试计算时间, 本文设$ \chi = 0.2, \ \Delta t = h = 1/30, $并且改变grad-div参数$ \beta $$ 0\leq\beta\leq 8000. $在此数值算例中, GMRES用于算法2.1的第一步, UMFPACK用于第二步. 如果GMRES迭代失败, 我们用'F' 表示结果. 此外, 我们还展示了标准grad-div稳定化有限元算法相对于模块grad-div稳定化有限元算法的时间百分比增长, 数值结果列于表 4.

表 4   随着β增加, 两种grad-div稳定化方法的计算时间

β标准grad-div模块grad-div增加(%)
0.18.8117.57816.271
0.29.8777.46132.381
0.410.4187.48839.129
0.813.1857.43777.289
8F7.566-
80F7.413-
800F7.458-
8000F7.312-

新窗口打开| 下载CSV


参考文献

Akbas M , Linke A , Rebholz L G , Schroeder P W .

The analogue of grad-div stabilization in DG methods for incompressible flows: Limiting behavior and extension to tensor-product meshes

Comput Methods Appl Mech Engrg, 2018, 341, 917- 938

DOI:10.1016/j.cma.2018.07.019      [本文引用: 1]

Akbas M, Rebholz L G. Modular grad-div stabilization for multiphysics flow problems. 2020, arXiv: 2001. 10100

[本文引用: 2]

An R , Su J .

Optimal error estimates of semi-implicit Galerkin method for time dependent nematic liquid crystal flows

J Sci Comput, 2018, 74, 979- 1008

DOI:10.1007/s10915-017-0479-7      [本文引用: 9]

Badia S , Guillén-Gónzalez F , Gutiérrez-Santacreu J V .

An overview on numerical analyses of nematic liquid crystal flows

Arch Comput Methods Eng, 2011, 18, 285- 313

DOI:10.1007/s11831-011-9061-x      [本文引用: 1]

Becker R , Feng X B , Prohl A .

Finite element approximations of the Ericksen-Leslie model for nematic liquid crystal flow

SIAM J Numer Anal, 2008, 46, 1704- 1731

DOI:10.1137/07068254X      [本文引用: 1]

Bochev P , Dohrmann C , Gunzburger M .

Stabilization of low-order mixed finite element for the Stokes equations

SIAM J Numer Anal, 2006, 44, 82- 101

DOI:10.1137/S0036142905444482      [本文引用: 2]

Brenner S , Scott L . The Mathematical Theory of Finite Element Methods. Berlin: Springer, 1994

[本文引用: 5]

Cabrales R C , Guillén-González F , Gutiérrez-Santacreu J V .

A time-splitting finite element stable approximation for the Ericksen-Leslie equations

SIAM J Sci Comput, 2015, 37, B261- B282

DOI:10.1137/140960979      [本文引用: 1]

Cabrales R C , Guillén-González F , Gutiérrez-Santacreu J V .

A projection-based time-splitting algorithm for approximating nematic liquid crystal flows with stretching

Z Angew Math Mech, 2017, 97, 1204- 1219

DOI:10.1002/zamm.201600247     

Du Q , Guo B , Shen J .

Fourier spectral approximation to a dissipative system modeling the flow of liquid crystals

SIAM J Numer Anal, 2001, 39, 735- 762

DOI:10.1137/S0036142900373737      [本文引用: 1]

Ericksen J .

Conservation laws for liquid crystals

Trans Soc Rheol, 1961, 5, 22- 34

[本文引用: 1]

Ericksen J .

Continuum theory of nematic liquid crystals

Res Mech, 1987, 21, 381- 392

URL     [本文引用: 1]

Fiordilino J A , Layton W , Rong Y .

An efficient and modular grad-div stabilization

Comput Methods Appl Mech Engrg, 2018, 335, 917- 938

[本文引用: 5]

Franca L P , Hughes T J .

Two classes of mixed finite element methods

Comput Methods Appl Mech Engrg, 1988, 69, 89- 129

DOI:10.1016/0045-7825(88)90168-5      [本文引用: 1]

Girault V , Guillén-González F .

Mixed formulation, approximation and decoupling algorithm for a penalized nematic liquid crystals model

Math Comp, 2011, 80, 781- 819

DOI:10.1090/S0025-5718-2010-02429-9      [本文引用: 1]

Guillén-González F , Gutiérrez-Santacreu J V .

A linear mixed finite element scheme for a nematic Ericksen-Leslie liquid crystal model

ESAIM: Math Model Numer Anal, 2013, 47, 1433- 1464

DOI:10.1051/m2an/2013076     

Guillén-González F , Koko J .

A splitting in time scheme and augmented lagrangian method for a nematic liquid crystal problem

J Sci Comput, 2015, 65, 1129- 1144

DOI:10.1007/s10915-015-0002-y      [本文引用: 1]

He Y N , Wang A W , Mei L Q .

Stabilized finite-element method for the stationary Navier-Stokes equations

J Engrg Math, 2005, 51, 367- 380

DOI:10.1007/s10665-004-3718-5      [本文引用: 2]

Jenkins E W , John V , Linke A , Rebholz L G .

On the parameter choice in grad-div stabilization for stokes equations

Adv Comput Math, 2014, 40, 491- 516

DOI:10.1007/s10444-013-9316-1      [本文引用: 1]

Leslie F .

Some constitutive equations for liquid crystals

Arch Ration Mech, 1987, 21, 381- 392

[本文引用: 1]

Lin F H .

Nonlinear theory of defects in nematics liquid crystals: Phase transitation and flow phenomena

Commun Pure Appl Math, 1989, 42, 789- 814

DOI:10.1002/cpa.3160420605      [本文引用: 3]

Lin F H , Liu C .

Existence of solutions for the Ericksen-Leslie system

Arch Ration Mech Anal, 2000, 154, 135- 156

DOI:10.1007/s002050000102     

Lin F H , Lin J , Wang C .

Liquid crystal flows in two dimensions

Arch Ration Mech Anal, 2010, 197, 297- 336

DOI:10.1007/s00205-009-0278-x     

Lin F H , Wang C .

On the uniqueness of heat flow of harmonic maps and hydrodynamic flow of nematic liquid crystals

Chin Ann Math Ser B, 2010, 31, 921- 938

DOI:10.1007/s11401-010-0612-5      [本文引用: 1]

Linke A , Rebholz L G .

On a reduced sparsity stabilization of grad-div type for incompressible flow problems

Comput Methods Appl Mech Engrg, 2013, 261/262, 142- 153

DOI:10.1016/j.cma.2013.04.005      [本文引用: 1]

Lu X , Huang P .

A modular grad-div stabilization for the 2D/3D nonstationary incompressible magnetohydrodynamic equations

J Sci Comput, 2020, 82, 3

DOI:10.1007/s10915-019-01114-x      [本文引用: 1]

Minev P , Vabishchevich P N .

Spliting schemes for unsteady problems involving the grad-div operator

Appl Numer Math, 2018, 124, 130- 139

DOI:10.1016/j.apnum.2017.10.005      [本文引用: 1]

Nochetto R , Pyo J H .

A finite element gauge-Uzawa method. Part I: The Navier-Stokes equations

SIAM J Numer Anal, 2005, 43, 1043- 1068

DOI:10.1137/040609756      [本文引用: 1]

Olshanskii M , Reusken A .

Grad-div stabilization for Stokes equations

Math Comp, 2004, 73, 1699- 1718

[本文引用: 1]

Qin Y , Hou Y , Huang P , Wang Y .

Numerical analysis of two grad-div stabilization methods for the time-dependent Stokes/Darcy model

Comput Math Appl, 2020, 79, 817- 832

DOI:10.1016/j.camwa.2019.07.032      [本文引用: 1]

Rong Y, Fiordilino J A. Numerical analysis of a BDF2 modular grad-div Stabilization method for the Navier-Stokes equations. 2018, arXiv: 1806.10750

[本文引用: 1]

Song L , Hou Y , Cai Z .

Recovery-based error estimator for stabilized finite element methods for the Stokes equation

Comput Meth Appl Mech Engrg, 2014, 272, 1- 16

DOI:10.1016/j.cma.2014.01.004      [本文引用: 1]

Song L , Su H , Feng X .

Recovery-based error estimator for stabilized finite element method for the stationary Navier-Stokes problem

SIAM J Sci Comput, 2016, 38, A3758- A3772

DOI:10.1137/15M1015261      [本文引用: 1]

Zhang S , Liu C , Zhang H .

Numerical simulations of hydrodynamics of nematic liquid crystals: Effects of kinematic transports

Commun Comput Phys, 2011, 9, 974- 993

DOI:10.4208/cicp.160110.290610a      [本文引用: 1]

Zheng H , Hou Y , Shi F .

A posteriori error estimates of stabilization of low-order mixed finite elements for incompressible flow

SIAM J Sci Comput, 2010, 32, 1346- 1360

DOI:10.1137/090771508      [本文引用: 1]

/