数学物理学报, 2021, 41(5): 1283-1295 doi:

论文

变系数广义KdV-Burgers方程的格子Boltzmann模型

张宗宁,, 李春光,, 董建强

北方民族大学数学与信息科学学院 银川 750021

General Propagation Lattice Boltzmann Model for a Variable-Coefficient Compound KdV-Burgers Equation

Zhang Zongning,, Li Chunguang,, Dong jianqiang

School of Mathematics and Information Science, North Minzu University, Yinchuan 750021

通讯作者: 李春光, E-mail: cglizd@hotmail.com

收稿日期: 2020-11-29  

基金资助: 国家自然科学基金.  11761005
宁夏高等教育一流学科建设资助项目.  NXYLXK2017B09
北方民族大学研究生创新项目.  YCX21156
北方民族大学校级一般项目.  2020XYZSX02

Received: 2020-11-29  

Fund supported: the NSFC.  11761005
the First-Class Disciplines Foundation of Ningxia.  NXYLXK2017B09
the Postgraduate Innovation Project of North Minzu University.  YCX21156
the University-level Scientific Research Projects of North Minzu University.  2020XYZSX02

作者简介 About authors

张宗宁,E-mail:zzn5238@163.com , E-mail:zzn5238@163.com

Abstract

This paper studies the numerical calculation method of a kind of general Kdv-Burgers equation with variable coefficients. Firstly, a lattice Boltzmann model of the generalized KdV-Burgers equation with variable coefficients is obtained by selecting the equilibrium distribution function and adding the correction function. The model could accurately recover the KdV-Burgers equation without any assumptions. Secondly, this paper studies the temporal and spatial change trend of the nonlinear high-order derivative term in the equation, and compares it with the analytical solution, and then gives an error analysis. Finally, This paper analyzes the precision of the space and time of the equation. According to the simulation experiment results, the model could reach 2nd order accuracy. Numerical results show that the current lattice Boltzmann model is a satisfactory and efficient algorithm.

Keywords: Lattice Boltzmann equation ; Chapman-Enskog analysis ; KdV-Burgers equation ; Variable coefficient

PDF (816KB) 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

张宗宁, 李春光, 董建强. 变系数广义KdV-Burgers方程的格子Boltzmann模型. 数学物理学报[J], 2021, 41(5): 1283-1295 doi:

Zhang Zongning, Li Chunguang, Dong jianqiang. General Propagation Lattice Boltzmann Model for a Variable-Coefficient Compound KdV-Burgers Equation. Acta Mathematica Scientia[J], 2021, 41(5): 1283-1295 doi:

1 引言

格子Boltzmann方法(lattice Boltzmann method, LBM)已用于模拟流体流动[1], 并扩展到模拟非线性演化方程(nonliner evolution equations, NLEEs), 如各向异性色散方程[2], 变系数对流扩散方程[3], Burgers方程[4], Korteweg-de Vries (KdV)方程[5], 广义Boussinesq方程[6]和变系数Gardner方程[7]等.

与根据时间和空间直接离散控制方程的传统宏观数值方法不同, 格子Boltzmann方法(LBM)是一种介观的数值计算方法[8]. LB方程基于一个基本的离散速度动力学方程, 它包括粒子碰撞和迁移基本过程, 这两个过程可以清楚地还原出各种宏观物理现象. 为了将微观LB输运方程与宏观流体动力学方程联系起来, 文献[9]使用Chapmann-Enskog多尺度展开技术, 把Knudsen数$ \epsilon $(假设平均自由程$ \mathit{l} $与宏观特征长度L的比值很小)作为展开参数, 对LB方程进行逐阶展开. 文献[10]使用直接Taylor展开方法有效地恢复出Navier-Stokes方程和非线性对流扩散方程, 并指出直接Taylor展开是比Chapmann-Enskog多尺度展开更有效的方法.

基于LBM的NLEEs的数值研究一般考虑常系数NLEEs, 在物理环境中, 当考虑到介质的不均匀性和边界的不均一性时, 变系数非线性偏微分方程比常系数偏微分方程更能反映实际情况. 文献[11], 着重于一个具有时变系数的广义Gardner方程, 方程(1.1)被用来模拟弱非线性长波在KdV型介质中的传播, 这种介质的特征是色散和非线性系数会发生变化, 可以描述大气阻塞现象.

$ \begin{equation} u_{t} -6f_{0}(t) uu_{x} +f_{2}(t) u_{xxx} -f_{3}(t) u_{x} +f_{4} (t)(Cu+xu_{x} ) = 0, \end{equation} $

其中, $ u $是关于空间$ x $和时间$ t $的波幅函数, $ C $是非零常数, $ f_{0} (t), f_{1} (t) $$ f_{2} (t) $分别对应于二次非线性和三次非线性和色散系数, $ -f_{3} (t)+f_{4} (t)x $$ f_{4} (t)C $分别表示耗散项和阻尼项.

文献[12]研究了一类变系数的复合Korteweg-de Vries-Burgers(vc-cKdVB)方程, 它是一种在固体材料、等离子体、流体等领域广泛使用的物理模型.

$ \begin{equation} u_{t} +l(t)uu_{x} +m(t)u^{2p}u_{x} +n(t)u_{xx} -r(t)u_{xxx} +s(t)u_{x} = 0, \end{equation} $

其中, 参数$ p $是非负整数, 系数$ l (t), m (t), n (t), r (t) $$ s (t) $是关于时间$ t $的解析函数, 下标表示偏导数. 在方程(1.2)中, 由于非线性项$ u^{2p}u_{x} $项为偶数次幂, 本文考虑更一般的情形, 将非线性项$ u^{2p}u_{x} $推广到$ u^{p}u_{x} $, 利用LBM模拟非线性项为任意次幂的偏微分方程, 即该文研究的变系数KdV-Burgers方程. 方程(1.3)第二项、第三项为平流项, 等号右边的项依次为扩散项、反扩散项或超扩散项.

$ \begin{equation} u_{t} + l(t)uu_{x} +m(t)u^{p} u_{x} = n(t)u_{xx} -r(t)u_{xxx}. \end{equation} $

本文的主要内容如下: 在第2节中, 利用Taylor展开和Chapmann-Enskog多尺度展开技术, 证明了带有修正项的格子Boltzmann模型与变系数KdV-Burgers宏观方程是相容的, 并由宏观方程的系数构造了平衡态分布函数和修正函数. 在第3节中, 利用MATLAB进行数值模拟, 给出不同时刻变系数方程的时空演化图和误差分析图, 并对模型的空间精度和时间精度进行数值分析. 第4节对本文研究的结果进行了总结和展望.

2 非线性系统的格子玻尔兹曼模型

带有修正项的离散分布函数的LB方程$ f_{i} $

$ \begin{equation} f_{i}\left(x+c_{i} \Delta t, t+\Delta t\right)-f_{i}(x, t) = -\frac{1}{\tau}\left[f_{i}(x, t)-f_{i}^{e q}(x, t)\right]+\Delta t h_{i}(x, t), \end{equation} $

其中, $ c_{i} $为离散晶格速度, $ \tau $为松弛时间, $ h_{i} $为修正项, $ f_{i}^{eq} $为平衡态分布函数, $ \left |c\right | $为特征晶格速度, 假设为1.

对(2.1)式进行多元Taylor展开, 并保留到$ O(\epsilon ^{4} ) $项, 可以得到

$ \begin{eqnarray} &&\Delta t\left(\frac{\partial}{\partial t}+c_{i} \frac{\partial}{\partial x}\right) f_{i}+\frac{\Delta t^{2}}{2}\left(\frac{\partial}{\partial t}+c_{i} \frac{\partial}{\partial x}\right)^{2} f_{i}+\frac{\Delta t^{3}}{6}\left(\frac{\partial}{\partial t}+c_{i} \frac{\partial}{\partial x}\right)^{3} f_{i}+O(\epsilon ^{4} )\\& = &-\frac{1}{\tau}\left(f_{i}-f_{i}^{e q}\right)+\Delta t h_{i}. \end{eqnarray} $

利用Chapmann-Enskog多尺度展开技术对时间和空间导数, 分布函数$ f_{i} $, 修正函数$ h_{i} $展开

$ \begin{equation} \frac{\partial}{\partial x} = \epsilon \frac{\partial}{\partial x_{1}}+O\left(\epsilon ^{2}\right), \end{equation} $

$ \begin{equation} \frac{\partial}{\partial t} = \epsilon \frac{\partial}{\partial t_{1}}+\epsilon ^{2} \frac{\partial}{\partial t_{2}}+\epsilon ^{3} \frac{\partial}{\partial t_{3}}+O\left(\epsilon ^{4}\right), \end{equation} $

$ \begin{equation} f_{i} = f_{i}^{(0)}+\epsilon f_{i}^{(1)}+\epsilon ^{2} f_{i}^{(2)}+\epsilon ^{3} f_{i}^{(3)}+O\left(\epsilon ^{4}\right), \end{equation} $

$ \begin{equation} h_{i} = \epsilon h_{i}^{(1)}+O\left(\epsilon ^{2}\right), \end{equation} $

其中, $ \epsilon $是Knudsen数, 定义为$ \epsilon = l/L $, $ l $是平均自由程, $ L $是特征长度, $ \epsilon $可取为时间步长$ \Delta t $[13]. 把(2.3)–(2.6)式代入方程(2.2), 我们可以得到关于$ \epsilon $的0到3阶微分方程

$ \begin{equation} O\left(\epsilon ^{0}\right):-\frac{1}{\tau}\left(f_{i}^{(0)}-f_{i}^{e q}\right) = 0, \quad \mbox{ 即 }\ f_{i}^{(0)} = f_{i}^{e q}. \end{equation} $

$ \begin{equation} O\left(\epsilon ^{1}\right):\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right) f_{i}^{(0)} = -\frac{1}{\tau \Delta t} f_{i}^{(1)}+h_{i}^{(1)}, \end{equation} $

$ \begin{eqnarray} &&O\left(\epsilon ^{2}\right): \frac{\partial}{\partial t_{2}} f_{i}^{(0)}+\Delta t\left(\frac{1}{2}-\tau\right)\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right)^{2} f_{i}^{(0)} \\&&+\Delta t \tau \left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right) h_{\alpha}^{(1)} = -\frac{1}{\tau \Delta t} f_{i}^{(2)}, \end{eqnarray} $

$ \begin{eqnarray} &&O\left(\epsilon ^{3}\right): \frac{\partial}{\partial t_{3}} f_{i}^{(0)}+\Delta t(1-2 \tau) \frac{\partial}{\partial t_{2}}\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right) f_{i}^{(0)} \\&& +\Delta t^{2} \left(\tau^{2}-\tau+\frac{1}{6}\right)\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right)^{3} f_{i}^{(0)}\\&& +\Delta t^{2} \left(\frac{\tau}{2}-\tau^{2}\right)\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right)^{2} h_{i}^{(1)} +\Delta t \tau \frac{\partial}{\partial t_{2}} h_{i}^{(1)} = -\frac{1}{\tau \Delta t} f_{i}^{(3)}. \end{eqnarray} $

与一般的LBM相似, 把宏观物理量$ u $定义为该点的局部粒子分布函数的和, 即

$ \begin{equation} u(x, t) = \sum\limits_{i} f_{i}(x, t), \end{equation} $

由质量守恒定律, $ f_{i} $$ f_{i}^{eq} $满足

$ \begin{equation} u(x, t) = \sum\limits_{i} f_{i}(x, t) = \sum\limits_{i} f_{i}^{(e q)}(x, t), \end{equation} $

由(2.7)式, 有

$ \begin{equation} \sum\limits_{i} f_{i}^{(0)} = u, \quad \sum\limits_{i} f_{i}^{(n)} = 0, \quad n = 1, 2, \cdots. \end{equation} $

由(2.6)式, 有

$ \begin{equation} \sum\limits_{i} h_{i} = \epsilon \sum\limits_{i} h_{i}^{(1)}, \end{equation} $

$ \begin{equation} \sum\limits_{i} c_{i} h_{i} = \epsilon \sum\limits_{i} c_{i} h_{i}^{(1)}, \end{equation} $

$ \begin{equation} \sum\limits_{i} c_{i}^{2} h_{i} = \epsilon \sum\limits_{i} c_{i}^{2}h_{i}^{(1)}. \end{equation} $

为了恢复方程(1.3), 对平衡分布函数和修正函数做如下约束

$ \begin{equation} \sum\limits_{i} c_{i} f_{i}^{(0)} = 0, \end{equation} $

$ \begin{equation} \sum\limits_{i} c_{i}^{2} f_{i}^{(0)} = \lambda_{1}n\left (t \right )u, \end{equation} $

$ \begin{equation} \sum\limits_{i} c_{i}^{3} f_{i}^{(0)} = \lambda_{2}r\left (t \right )u. \end{equation} $

$ \begin{equation} \sum\limits_{i} h_{i} = 0, \end{equation} $

$ \begin{equation} \sum\limits_{i} c_{i} h_{i} = \lambda _{3} l (t) u^{2}+\lambda _{4} m (t)u^{p+1}, \end{equation} $

$ \begin{equation} \sum\limits_{i} c_{i}^{2} h_{i} = 0. \end{equation} $

对方程(2.7)两边关于$ i $求和, 有

$ \begin{equation} \frac{\partial u}{\partial t_{1}}+\frac{\partial}{\partial x_{1}}\left(\sum\limits_{i} c_{i} f_{i}^{(0)}\right) = \sum\limits_{i} h_{i}^{(1)} = 0, \ \mbox{ 即 } \quad \frac{\partial u}{\partial t_{1}} = 0. \end{equation} $

对方程(2.8)两边关于$ i $求和, 有

$ \begin{equation} \frac{\partial u}{\partial t_{2}}+\Delta t\left(\frac{1}{2}-\tau\right) \sum\limits_{i}\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right)^{2} f_{i}^{(0)}+\Delta t \tau \sum\limits_{i}\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right) h_{i}^{(1)} = 0, \end{equation} $

其中

$ \begin{eqnarray} \sum\limits_{i}\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right)^{2} f_{i}^{(0)} & = &\frac{\partial^{2}}{\partial t_{1}^{2}}\left(\sum\limits_{i} f_{i}^{(0)}\right)+2 \frac{\partial^{2}}{\partial t_{1} \partial x_{1}}\left(\sum\limits_{i} c_{i} f_{i}^{(0)}\right)+\frac{\partial^{2}}{\partial x_{1}^{2}}\left(\sum\limits_{i} c_{i}^{2} f_{i}^{(0)}\right) \\& = &\frac{\partial^{2}}{\partial x_{1}^{2}}\left(\sum\limits_{i} c_{i}^{2} f_{i}^{(0)}\right) = \lambda_{1} n(t)\frac{\partial ^{2} u}{\partial x_{1}^{2}}, \end{eqnarray} $

$ \begin{eqnarray} \sum\limits_{i}\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right) h_{i}^{(1)} & = &\frac{\partial}{\partial t_{1}}\left(\sum\limits_{i} h_{i}^{(1)}\right)+\frac{\partial}{\partial x_{1}}\left(\sum\limits_{i} c_{i} h_{i}^{(1)}\right) \\ & = &\frac{\partial}{\partial x_{1}}\left[\lambda _{3} l (t) u^{2}+\lambda _{4}m (t)u^{p+1} \right] \\ & = &2 \lambda _{3} l (t)u\frac{\partial u}{\partial x_{1} } +(p+1)\lambda _{4}m(t) u^{p} \frac{\partial u}{\partial x_{1} }, \end{eqnarray} $

把(2.25)式, (2.26)式代入(2.24)式中

$ \begin{equation} \frac{\partial u}{\partial t_{2}}+\frac{\tau }{\epsilon } [2 \lambda _{3}\Delta t l(t) u \frac{\partial u}{\partial x_{1} }+(p+1)\lambda _{4}\Delta t n (t)u^{p}\frac{\partial u}{\partial x_{1} } ] +\Delta t(\frac{1}{2} -\tau )\lambda _{1} n(t)\frac{\partial^{2} u}{\partial x_{1}^{2} } = 0. \end{equation} $

对方程(2.9)两边关于$ i $求和, 有

$ \begin{eqnarray} &&\frac{\partial u}{\partial t_{3}}+\Delta t^{2}\left(\tau^{2}-\tau+\frac{1}{6}\right) \sum\limits_{i}\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right)^{3} f_{i}^{(0)}\\&& +\Delta t^{2}\left(\frac{\tau}{2}-\tau^{2}\right) \sum\limits_{i}\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right)^{2} h_{i}^{(1)} = 0, \end{eqnarray} $

其中

$ \begin{eqnarray} \sum\limits_{i}\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right)^{3} f_{i}^{(0)} & = &\frac{\partial^{3}}{\partial t_{1}^{3}}\left(\sum\limits_{i} f_{i}^{(0)}\right)+3 \frac{\partial^{3}}{\partial x_{1} \partial t_{1}^{2}}\left(\sum\limits_{i} c_{i} f_{i}^{(0)}\right) \\&&+3 \frac{\partial^{3}}{\partial x_{1}^{2} \partial t_{1}}\left(\sum\limits_{i} c_{i}^{2} f_{i}^{(0)}\right)+\frac{\partial^{3}}{\partial x_{1}^{3}}\left(\sum\limits_{i} c_{i}^{3} f_{i}^{(0)}\right) \\& = &3\lambda _{1} \frac{\partial^{3} }{\partial x_{1}^{2} \partial t_{1} }[n(t) u] +\lambda _{2} r(t)\frac{\partial^{3} u}{\partial x_{1}^{3} }, \end{eqnarray} $

$ \begin{eqnarray} \sum\limits_{i}\left(\frac{\partial}{\partial t_{1}}+c_{i} \frac{\partial}{\partial x_{1}}\right)^{2} h_{i}^{(1)}& = & \frac{\partial^{2}}{\partial t_{1}^{2}}\left(\sum\limits_{i} h_{i}^{(1)}\right)+2 \frac{\partial^{2}}{\partial x_{1} \partial t_{1}}\left(\sum\limits_{i} c_{i} h_{i}^{(2)}\right) +\frac{\partial^{2}}{\partial x_{1}^{2}}\left(\sum\limits_{i} c_{i}^{2} h_{i}^{(1)}\right) \\& = &\frac{2}{\epsilon }\frac{\partial^{2} }{\partial x_{1} \partial t_{1}}[\lambda _{3} l (t)u^{2} +\lambda _{4} m (t)u^{p+1} ], \end{eqnarray} $

把(2.29)式, (2.30)式代入(2.28)式, 得到

$ \begin{eqnarray} && \frac{\partial u}{\partial t_{3}}+\Delta t^{2}(\tau^{2}-\tau+\frac{1}{6})[3\lambda _{1} n(t)\frac{\partial^{3} u}{\partial x_{1}^{2} \partial t_{1} } +\lambda _{2} r(t)\frac{\partial^{3} u}{\partial x_{1}^{3} }]\\&&+\frac{1}{\epsilon }\Delta t^{2}(\tau-2\tau^{2} ) \frac{\partial^{2} }{\partial x_{1} \partial t_{1}}[\lambda _{3} l (t)u^{2} +\lambda _{4} m (t)u^{p+1}] = 0. \end{eqnarray} $

采用LBM文献中所使用的符号, $ DnQm $格子Boltzmann离散速度模型是指在$ n $维空间中具有$ m $个离散速度的模型. 本文采用$ D1Q5 $格子Boltzmann离散速度模型, 其离散速度的集合为:$ c_{i} = \left \{ 0, \pm 1, \pm 2 \right \} $. 由方程$ (2.23)\times \epsilon +(2.27)\times \epsilon^{2} +(2.31)\times \epsilon^{3} $可以导出具有四阶截断误差的宏观方程

$ \begin{eqnarray} &&u_{t}+2 \lambda _{3}\tau \Delta t l(t) u u_{x} +(p+1)\lambda _{4}\tau \Delta t m (t)u^{p}u_{x}+\Delta t(\frac{1}{2} -\tau )\lambda _{1} n(t)u_{xx}\\&&+\Delta t^{2}(\tau^{2}-\tau+\frac{1}{6})[3\lambda _{1}\epsilon u_{xx} \partial t_{1} n(t) +\lambda _{2} r(t)u_{xxx} ]\\& &+\epsilon \Delta t^{2}(\tau-2\tau^{2} ) [2\lambda _{3}uu_{x} \partial t_{1} l (t) +(p+1)\lambda _{4} u^{p}u_{x} \partial t_{1} m (t)]+O(\epsilon ^{4} ) = 0, \end{eqnarray} $

为恢复宏观方程(1.3), 令

$ \begin{equation} \left\{\begin{array}{ll} { }\lambda _{1} = \frac{1}{\Delta t (\tau-0.5)}, \\ { }\lambda _{2} = \frac{1}{\Delta t^{2} (\tau^{2} -\tau+\frac{1}{6} )}, \\ { }\lambda _{3} = \frac{1}{2\Delta t \tau }, \\{ }\lambda _{4} = \frac{1}{(p+1)\Delta t \tau }. \end{array} \right. \end{equation} $

此时方程(2.32)为

$ \begin{equation} u_{t} + l(t) u u_{x} + m (t)u^{p}u_{x}- n(t)u_{xx}+r(t)u_{xxx} = {\cal R}, \end{equation} $

为了推导出局部平衡态分布函数的具体表达式, 我们应该引入一个新的约束条件, 考虑到$ f $的矩都是$ u $的多项式函数, 我们假设平衡分布函数满足以下约束

$ \begin{equation} f_{4}^{\left ( 0 \right ) } = \sigma u, \end{equation} $

其中引入参数$ \sigma $来调节平衡态分布函数

$ \begin{eqnarray} f_{i}^{(0)} = \left\{\begin{array}{lll}{ } {} [1+6\sigma -\frac{n(t)}{c^{2}\Delta t\left ( \tau -0.5 \right ) } +\frac{r(t)}{2c^{3}\Delta t^{2} \left (\tau ^{2} -\tau +\frac{1}{6} \right ) } ]u, \ \ &i = 0, \\{ } {} [-4\sigma +\frac{n(t)}{2 c^{2}\Delta t\left ( \tau -0.5 \right ) } -\frac{r(t)}{2 c^{3}\Delta t^{2} \left (\tau ^{2} -\tau +\frac{1}{6} \right )} ]u, \ \ &i = 1, \\ { } {}[-4\sigma +\frac{n(t)}{2 c^{2}\Delta t\left ( \tau -0.5 \right ) } -\frac{r(t)}{6 c^{3}\Delta t^{2} \left (\tau ^{2} -\tau +\frac{1}{6} \right )} ]u, \ \ &i = 2, \\{ }{} [\sigma +\frac{r(t)}{6 c^{3}\Delta t^{2} \left (\tau ^{2} -\tau +\frac{1}{6} \right )} ]u, \ \ &i = 3, \\ \sigma u, \ \ &i = 4.\end{array}\right. \end{eqnarray} $

耦合方程(2.14)–(2.16). 假设速度1和速度2方向上的修正函数相同, 我们可以推导出补偿函数为

$ \begin{equation} h_{i } = \left\{\begin{array}{ll} { }\frac{1}{2} {\cal H}, &i = 0, \\{ } -\frac{1}{3} {\cal H}, &i = 1, \\{ } -\frac{1}{3} {\cal H}, &i = 2, \\{ } \frac{1}{3} {\cal H}, &i = 3, \\{ }-\frac{1}{6} {\cal H}, &i = 4.\end{array}\right. \end{equation} $

$ \begin{equation} {\cal H} = [\frac{l (t)}{2} u^{2} +\frac{m(t)}{p+1} u^{p+1}]/\left (\Delta t \tau c \right ). \end{equation} $

考虑到方程的系数为时变系数. 因此在具体的数值模拟时, 松弛时间$ \tau $不应该是一个固定常数, 由方程(2.36), 假设$ \tau _{1} $满足

$ \begin{equation} \tau _{1} = \tau ^{2} -\tau +\frac{1}{6}, \end{equation} $

由文献[11], 把参数$ k $设为一个给定的值

$ \begin{equation} k = \frac{r(t)}{c^{3}\Delta t^{2} \tau _{1} }. \end{equation} $

为了确保本文建立的格子Boltzmann模型的稳定性, 因此必须保证松弛时间满足约束条件$ \tau _{1}>0.5 $ (参见文献[14]), 由此可得

$ \begin{equation} \tau (t) = \frac{1}{2} +\sqrt{\frac{1}{12} +\frac{\Delta t}{\Delta x^{3}k } r(t) }, \end{equation} $

$ \begin{equation} \frac{1}{12} +\frac{\Delta t}{\Delta x^{3}k } r(t)\ge 0\Rightarrow k\ge \frac{-12\Delta t r(t)}{\Delta x^{3} }. \end{equation} $

对参数取值总结如下.

(1) 参数$ k $是单松弛时间$ \tau $的调节因子, 一般取值为$ 0.0<k<1.5 $ (参见文献[11]);

(2) 常数$ \sigma $是平衡函数的一个调整因子, 它可以在演化过程中调整模拟结果, 一般$ \sigma $是一个负的微小值, $ -\frac{1}{12} <\sigma <-0.01 $ (参见文献[12]).

3 数值模拟

本文采用均方根误差$ E_{2} $, 最大误差$ E_{\infty } $, 全局相对误差$ GRE $来验证格子Boltzmann模型的有效性, 并采用最小二乘拟合来计算模型的精度.

$ \begin{equation} E_{2} = \frac{1}{N} \sqrt{\sum\limits_{j = 1}^{N}\left [ u\left ( x_{j} , t \right )-u^{*} \left ( x_{j}, t \right ) \right ]^{2} }, \end{equation} $

$ \begin{equation} E_{\infty } = \max\limits_{j = 0, 1, 2\dots, N }\left | u\left ( x_{j} , t \right )-u^{*} \left ( x_{j}, t \right ) \right |, \end{equation} $

$ \begin{equation} GRE = \frac{ { \sum\limits_{j = 1}^{N}}\left | u\left ( x_{j} , t \right )-u^{*} \left ( x_{j}, t \right ) \right | }{ { \sum\limits_{j = 1}^{N} \left | u^{*} \left ( x_{j} , t \right ) \right | } }, \end{equation} $

其中, $ u\left ( x, t \right ) $$ u^{\ast } \left ( x, t \right ) $分别表示数值解和解析解, $ N $是网格节点数. 并通过以下方法计算离散化方法收敛速度.

$ \begin{equation} p \approx\left|\frac{\log E_{\mbox{new}}-\log E_{\mbox{old}}}{\log H_{\mbox{new}}-\log H_{\mbox{old}}}\right|, \end{equation} $

其中$ p $是收敛阶, $ E $代表误差, $ H $代表序列, 下标new和old分别代表新旧状态.

算例3.1   取$ l(t)\! = \!\alpha , m(t)\! = \!n(t)\! = \!0, r(t)\! = \!-\beta $时, 此时方程可以转化成KdV方程[12]

$ \begin{equation} u_{t} +\alpha uu_{x} = -\beta u_{xxx}. \end{equation} $

它被用来描述流体中的波传播和分层内波传播现象, 以及等离子体中的离子声波. 令$ \alpha = 6, $$ \beta = 1 $, KdV方程的双孤子解可以表示为

其中, $ a_{1}, a_{2}, \kappa_{1}, \kappa_{2} $为常数, $ a_{1} = 2, a_{2} = 1, \kappa_{1} = 1.5, \kappa_{2} = 1 $, $ k = 0.11, \sigma = -0.0095 $, $ \Delta x = 0.05, \Delta t = 0.0001, c = \Delta x/\Delta t = 500 $, 计算区域固定在$ I = [-10, 20] $, 图 1给出了数值解和解析解在二维空间中的视觉对比图. 表 1给出了不同时刻数值结果的$ E_{2}, E_{\infty }, GRE $. 图 2给出了数值结果的时空演化图. 图 3左给出了$ \Delta x = \left \{0.01\quad0.02\quad0.03\quad0.04 \right \} $时, $ \Delta t $取0.001时, 不同时刻的空间精度图, $ t = 1 $时数值精度为1.9001, $ t = 2 $时数值精度为2.2003. 图 3右给出了$ \Delta t = \left \{ 0.0001\quad 0.0002\quad 0.0003\quad 0.0004 \right \} $, $ \Delta x $取0.01时, 不同位置的时间精度图, $ x = 0 $时斜率为2.1833, $ x = 5 $时斜率为2.0456. 这与方程(2.34)的精度是相同的.

图 1

图 1   方程(3.5)孤子传输的数值结果(a)和解析解(b)


表 1   不同时刻方程(3.5)数值解与解析解比较分析

Parameterst=0.5t=1.0t=1.5t=2.0
E25.1836e-044.9568e-046.1178e-047.6003e-04
E0.03640.03490.04780.0585
GRE0.03340.03030.03690.0429

新窗口打开| 下载CSV


图 2

图 2   方程(3.5)在不同时刻的数值解和解析解对比图


图 3

图 3   (a) 表示空间数值精度图, (b)表示时间数值精度图


注3.1   数值模拟时, 为区分数值解与解析解, 用matlab中的colormap default表示数值解, colormap jet表示精确解, 下同.

算例3.2   取$ m(t) = \alpha, n(t) = \beta \times t^{\frac{2}{n} -1} $, 方程转化为广义变系数Burgers方程$ B\left ( n, 1 \right ) $ (参见文献[15])

$ \begin{equation} u_{t} +5 \alpha u^{p} u_{x} = \beta t^{\frac{2}{n}-1 } u_{xx}. \end{equation} $

初始条件和边界条件为

其中, 取$ q(t) = \gamma \times t^{-\frac{1}{n} } $, 此时方程的解析解为

上式中, 取$ n = 5, \alpha = 0.5, \beta = 2, \gamma = 0.5 $, $ k = 1.05, \sigma = -0.19, \Delta x = 0.01, \Delta t = 0.0001, $$ c = \Delta x/\Delta t = 100 $, 计算区域为$ I = [-10, 10] $. 图 4给出了数值结果的时空演化图, 图 5给出了数值解和解析解在二维空间中的视觉对比图, 从图中可以看出, 数值结果与解析解吻合良好, 图 6左给出了$ \Delta x = \left \{ 0.01 \quad 0.02 \quad 0.03\quad0.04 \right \} $, $ \Delta t = 0.001 $时, 不同时刻的空间精度图, $ t = 1 $时数值精度为1.9940, $ t = 2 $时数值精度为2.1730. 图 6右给出了$ \Delta t = \left \{ 0.0001\quad 0.0002\quad 0.0003\quad 0.0004 \right \} $, $ \Delta x = 0.01 $, 不同位置的时间精度图, $ x = 0 $时数值精度为2.0288, $ x = 10 $时数值精度为1.9942, 这与方程(2.34)的精度是相同的. 表 2给出了不同时刻数值结果的$ E_{2}, E_{\infty }, GRE $.

图 4

图 4   $ p = 5, \alpha = 0.5, \beta = 2, \gamma = 0.5 $方程(3.6)的数值解(a)和解析解(b)


图 5

图 5   (a) $ t = 1, $ (b) $ t = 2, $ (c) $ t = 3 $数值解和解析解对比图


图 6

图 6   (a) 表示空间数值精度图, (b)表示时间数值精度图


表 2   t = 1.0, 2.0, 3.0不同时刻方程(3.6)数值结果与解析解的比较

Parameterst=1.0t=2.0t=3.0
E21.0660e-047.9380e-057.0678e-05
E0.02300.01510.0125
GRE0.03430.02670.0252

新窗口打开| 下载CSV


算例3.3   考虑如下形式的广义变系数KdV-mKdV方程[15]

$ \begin{equation} u_{t} -15\cos\left ( \pi t/2 \right ) u_{x} -6uu_{x} +0.06u^{p} u_{x} = 0.01u_{xxx}. \end{equation} $

$ p = 2 $时, 方程的解析解可以表示为

上式中, 取$ k = 0.985, \sigma = -0.09, \Delta x = 0.01, \Delta t = 0.0005, c = \Delta x/\Delta t = 500 $, 计算区域固定在$ I = [-20, 10] $. 表 3给出了不同时刻数值结果的$ E_{2}, E_{\infty }, GRE $, 图 7给出了给出了数值结果的时空演化图, 图 8数值解和解析解在二维空间中的视觉对比图, 从图中可以看出, 数值结果与解析解吻合良好. 图 9左给出了$ \Delta x = \left \{ 0.01 \quad 0.02 \quad 0.03\quad0.04 \right \} $时, $ \Delta t $为0.001, 不同时刻的空间精度图, $ t = 1 $时数值精度为1.9532, $ t = 2 $时数值精度为1.9942, 图 9右给出了$ \Delta t = \left \{ 0.0001\quad 0.0002\quad 0.0003\quad 0.0004 \right \} $时, $ \Delta x $取0.01, 不同位置的时间精度图, $ x = 0 $时数值精度为2.0456, $ x = 5 $时数值精度为2.0503, 这与方程(2.34)的数值精度是相同的, 时间和空间均可达到2阶精度.

表 3   t = 1.0, 2.0, 3.0不同时刻方程(3.7)数值解与解析解的对比分析

Parameterst=1.0t=2.0t=3.0
E22.2805e-042.8858e-042.2420e-04
E0.00430.00550.0043
GRE0.05930.07500.0583

新窗口打开| 下载CSV


图 7

图 7   方程(3.7)在不同时刻的数值解和解析解对比图


图 8

图 8   方程(3.7)的数值解(a)和解析解(b)


图 9

图 9   (a) 表示空间数值精度图, (b)表示时间数值精度图


4 结论

本文研究了一类变系数广义KdV-Burgers方程的格子Boltzmann模型. 首先, 通过选择平衡分布函数和适当地增加修正函数建立了方程(1.3). 利用D1Q5格子Boltzmann模型对不同形式的变系数KdV-Burgers方程进行数值模拟试验.其次, 证明了格子Boltzmann模型与宏观方程是相容的. 然后,引入自由参数$ \sigma $便于构造出平衡分布函数$ f_{i}^{eq} $, 引入参数$ \mathit{k} $调整松弛时间, 很好的避免了松弛时间$ \tau $的取值问题. 最后, 对数值算例的时空变化趋势和时间空间精度进行分析, 证明了模型可以从时间上和空间上达到二阶精度. 该模型推广到更高维的变系数Burgers方程, 变系数对流扩散方程, 变系数模糊类波动方程数值求解中.

参考文献

Chen S Y , Doolen G D .

Lattice Boltzmann method for fluid flows

Annual Review of Fluid Mechanics, 1998, 30, 329- 364

DOI:10.1146/annurev.fluid.30.1.329      [本文引用: 1]

Ginzburg I .

Equilibrium-type and link-type lattice Boltzmann models for generic advection and anisotropic-dispersion equation

Advances in Water Resources, 2005, 28 (11): 1171- 1195

DOI:10.1016/j.advwatres.2005.03.004      [本文引用: 1]

Li Q H , Chai Z H , Shi B C .

Lattice Boltzmann model for a class of convection-diffusionequations with variable coefficients

Computers and Mathematics with Applications, 2015, 5 (4): 548- 561

URL     [本文引用: 1]

Zhang J Y , Yan G W .

Lattice Boltzmann method for one and two-dimensional burgers equation

Physica A: Statistical Mechanics and Its Applications, 2008, 387 (19): 4771- 4786

URL     [本文引用: 1]

Yan G W .

A lattice Boltzmann equation for waves

Journal of Computational Physics, 2000, 161 (1): 61- 69

DOI:10.1006/jcph.2000.6486      [本文引用: 1]

Liu F , Shi W P , Wu F F .

A lattice Boltzmann model for the generalized Boussinesq equation

Applied Mathematics and Computation, 2016, 274 (1): 331- 342

URL     [本文引用: 1]

Chai Z H , Shi B H , Zheng L .

A unified lattice Boltzzmann model for some nonlinear partial differential equations

Chaos, Soliton and Fract, 2008, 36 (4): 874- 882

DOI:10.1016/j.chaos.2006.07.023      [本文引用: 1]

何雅玲, 王勇, 李庆. 格子Boltzmann方法的理论及应用. 北京: 科学出版社, 2009

[本文引用: 1]

He Y l , Wang Y , Li Q . The Theory and Application of Lattice Boltzmann Method. Beijing: Science Press, 2009

[本文引用: 1]

默罕默德·阿卜杜勒马吉德. 格子玻尔兹曼方法. 北京: 电子工业出版社, 2015

[本文引用: 1]

Mohamad A A . Lattice Boltzmann Method. Beijing: Electronic Industry Press, 2015

[本文引用: 1]

Chai Z H , Shi B C .

Multiple-relaxation-time lattice Boltzmann method for the Navier-Stokes and nonlinear convection-diffusion equations: modeling, analysis, and elements

Physical Review E, 2020, 102, 023306

URL     [本文引用: 1]

Hu W Q , Gao Y T , Lan Z Z .

Lattice Boltzmann model for a generalized Gardner equation with time-dependent variable coefficient

Applied Mathematical Modelling, 2017, 46, 126- 140

DOI:10.1016/j.apm.2017.01.061      [本文引用: 3]

Lan Z Z , Hu W Q , Gao Y T .

General propagation lattice Boltzmann model for a variable-coefficient compound KdV-Burgers equation

Applied Mathematical Modelling, 2019, 73, 695- 714

DOI:10.1016/j.apm.2019.04.013      [本文引用: 3]

赖惠林, 马昌凤.

非线性偏微分方程的高阶格子BGK模型

中国科学G版, 2009, 39 (7): 913- 922

URL     [本文引用: 1]

Lan H L , Ma C F .

A higer order lattice BGK model for simulating some nonlinear partial differential equations

Sci China Ser G, 2009, 39 (7): 913- 922

URL     [本文引用: 1]

Sterling J D , Chen S Y .

Stability analysis of lattice Boltzmann methods

Journal of Computational Physics, 1996, 123 (1): 196- 206

DOI:10.1006/jcph.1996.0016      [本文引用: 1]

Abd-el-Malek M B , Helal M M .

Group method solution of the generalized forms of Burgers, Burgers-KDV and Kdv equations with time-dependent variable coefficients

Acta Mesh, 2001, 221, 281- 296

URL     [本文引用: 2]

/