这是我之前的一次期末大报告作业,里面的内容也算是从其他文章里搬运的,在此重新誊抄出来。内容如有错误,欢迎各位指正。
[注] 1. 下文提到的 x x x 、 u u u 和 ξ \xi ξ 无论有没有加粗均为向量。
2. 本文已经同步发表在本人的知乎 和CSDN 上,可在其他网站获得同样的阅读效果。
一、 单松弛格子Boltzmann方程
1.1 Boltzmann方程
格子Boltzmann方法基于Boltzmann方程(即(1)式),该方程引入速度分布函数,描述了非热力学平衡状态的热力学系统统计行为:
∂ f ∂ t + ξ ⋅ ∂ f ∂ x + F m ⋅ ∂ f ∂ ξ = Ω ( f ) (1) \frac{\partial f}{\partial t} + \mathbf{\xi} \cdot \frac{\partial f}{\partial \boldsymbol{x}} + \frac{F}{m} \cdot \frac{\partial f}{\partial \mathbf{\xi}}=\Omega(f) \tag{1}
∂ t ∂ f + ξ ⋅ ∂ x ∂ f + m F ⋅ ∂ ξ ∂ f = Ω ( f ) ( 1 )
(1)式中 Ω ( f ) \Omega(f) Ω ( f ) 为碰撞项, ξ = ∂ x ∂ t \mathbf{\xi}=\frac{\partial x}{\partial t} ξ = ∂ t ∂ x 。速度分布函数的定义是:在时刻t t t ,以x x x 为中心的空间微元 d x \mathrm{d}\mathit{x} d x 内,速度在 ξ \xi ξ 和 ( ξ + d ξ ) (\xi+\mathrm{d}\it{\xi}) ( ξ + d ξ ) 之间的分子的数量为 f d x d ξ f \mathrm{d}\it{x} \mathrm{d}\it{\xi} f d x d ξ 。记分子质量为 m m m ,流体宏观速度为u u u ,单位体积内能为e e e ,分子数密度为n n n ,则:
ρ = m n = m ∫ f d ξ \rho = mn = m \int{f} \mathrm{d}\xi
ρ = m n = m ∫ f d ξ
ρ u = m ∫ ξ f d ξ \rho u = m \int{\xi f} \mathrm{d}\xi
ρ u = m ∫ ξ f d ξ
ρ E = ρ e + 1 2 ρ u 2 = m ∫ f ξ 2 2 d ξ \rho E = \rho e + \frac{1}{2} \rho u^2 = m \int{ f \frac{\xi^2}{2} } \mathrm{d} \xi
ρ E = ρ e + 2 1 ρ u 2 = m ∫ f 2 ξ 2 d ξ
碰撞项 描述了分子碰撞对整个系统运动产生的影响,它具有两个特点:①满足质量、动量和能量守恒;②能够反映系统趋于平衡态的趋势。Bhatnagur、Gross和Krook三人在此基础上提出了最为简单且著名的BGK模型,即(2)式。
Ω ( f ) = − 1 τ c ( f − f ( e q ) ) (2) \Omega(f) = -\frac{1}{\tau_c} \left ( f - f^{(eq)} \right ) \tag{2}
Ω ( f ) = − τ c 1 ( f − f ( e q ) ) ( 2 )
其中τ c \tau_c τ c 为松弛时间, f ( e q ) f^{(eq)} f ( e q ) 为平衡态分布函数。 可根据Boltzmann H定理计算得:
f ( e q ) = n ( 2 π R T ) 3 / 2 e x p ( − ( ξ − u ) 2 2 R T ) (3) f^{(eq)} = \frac{n}{(2 \pi RT)^{3/2}} \mathrm{exp}\left( -\frac{(\mathbf{\xi-u})^2}{2RT} \right) \tag{3}
f ( e q ) = ( 2 π R T ) 3 / 2 n e x p ( − 2 R T ( ξ − u ) 2 ) ( 3 )
1.2 格子Boltzmann方程
为了将Boltzmann方程应用于实际的数值模拟,需要将流体视为大量的离散粒子。离散粒子于流体分子不同,它们驻留在网格上,并按一定的规则在网格上发生碰撞和迁移。因此,1.2节将 m f mf m f 记为 f f f ,此时 f f f 为密度分布函数。对密度分布函数 f f f ,不考虑 F F F 的Boltzmann-BGK方程为式。
∂ f ∂ t + ξ ⋅ ∂ f ∂ x = − 1 τ c [ f − f ( e q ) ] (4) \frac{\partial f}{\partial t} + \mathbf{\xi} \cdot \frac{\partial f}{\partial \mathit{x}} = - \frac{1}{\tau_c} \left[ f - f^{(eq)} \right] \tag{4}
∂ t ∂ f + ξ ⋅ ∂ x ∂ f = − τ c 1 [ f − f ( e q ) ] ( 4 )
其宏观量按如下方式确定
ρ = ∫ f d ξ , ρ u = ∫ ξ f d ξ , ρ E = ∫ f ξ 2 2 d ξ \rho = \int { f } \mathrm{d} \xi ,\rho u = \int { \xi f } \mathrm{d} \xi ,\rho E = \int { f \frac{\xi^2}{2} } \mathrm{d} \xi
ρ = ∫ f d ξ , ρ u = ∫ ξ f d ξ , ρ E = ∫ f 2 ξ 2 d ξ
对 f f f 和其对应的 f ( e q ) f^{(eq)} f ( e q ) 可进行如下离散:f i = w i f f_i = w_i f f i = w i f , f i ( e q ) = w i f ( e q ) f_i^{(eq)} = w_i f^{(eq)} f i ( e q ) = w i f ( e q ) 。其中下标 i i i 表示特征方向 c i \boldsymbol{c}_i c i , f i f_i f i 和f i ( e q ) f_i^{(eq)} f i ( e q ) 均沿着c i \boldsymbol{c}_i c i 方向运动。据此,沿着特征方向离散(4)式得到的差分格式为
f i ( x + ξ i Δ t , t + Δ t ) − f i ( x , t ) = − 1 τ [ f i ( x , t ) − f i ( e q ) ( x , t ) ] (5) f_i (x+ξ_i Δt,t+Δt)-f_i (x,t)=-\frac{1}{τ} \left[f_i (x,t)-f_i^{(eq)} (x,t) \right] \tag{5}
f i ( x + ξ i Δ t , t + Δ t ) − f i ( x , t ) = − τ 1 [ f i ( x , t ) − f i ( e q ) ( x , t ) ] ( 5 )
其中 τ = τ c Δ t \tau = \frac{\tau_c}{\Delta t} τ = Δ t τ c 为无量纲松弛时间。由于对应的 f ( e q ) f^{(eq)} f ( e q ) 为:
f ( e q ) = ρ ⋅ e x p ( − ( ξ − u ) 2 2 R T ) (6-1) f^{(eq)} = \rho \cdot \mathrm{exp} ( - \frac{(\xi - \it{u})^2}{2\it{RT}} ) \tag{6-1}
f ( e q ) = ρ ⋅ e x p ( − 2 R T ( ξ − u ) 2 ) ( 6 - 1 )
所以将 f ( e q ) f^{(eq)} f ( e q ) 展开至2阶并离散,可得:
f i ( e q ) = w i ρ [ 1 + ξ i ⋅ u c s 2 + ( ξ i ⋅ u ) 2 2 c s 4 − u ⋅ u 2 c s 2 ] (6-2) f^{(eq)}_i = w_i \rho \left[ 1 + \frac{\xi_i \cdot u}{c_s^2} + \frac{(\xi_i \cdot u)^2}{2 c_s^4} - \frac{u \cdot u}{2 c_s^2} \right] \tag{6-2}
f i ( e q ) = w i ρ [ 1 + c s 2 ξ i ⋅ u + 2 c s 4 ( ξ i ⋅ u ) 2 − 2 c s 2 u ⋅ u ] ( 6 - 2 )
其中 c s = R T c_s = \sqrt{RT} c s = R T 在LBM模拟中一般被称作格子声速。
二、Chapman-Enskog展开分析
将基本碰撞不变量 φ i = 1 , ξ , ξ 2 2 φ_i=1,ξ,\frac{ξ^2}{2} φ i = 1 , ξ , 2 ξ 2 乘以(1)式两端,并对 ξ \xi ξ 积分有:
∂ ρ ∂ t + ∇ ⋅ ( ρ u ) = 0 (7-1) \frac{∂ρ}{\partial t}+∇⋅(ρ \mathbf{u})=0 \tag{7-1}
∂ t ∂ ρ + ∇ ⋅ ( ρ u ) = 0 ( 7 - 1 )
∂ ( ρ u ) ∂ t + ∇ ⋅ ( ρ u u + P ) = ρ a (7-2) \frac{∂(ρu)}{\partial t}+∇⋅(ρ \mathbf{uu}+P) = ρ \mathit{\bf{a}} \tag{7-2}
∂ t ∂ ( ρ u ) + ∇ ⋅ ( ρ u u + P ) = ρ a ( 7 - 2 )
∂ ( ρ E ) ∂ t + ∇ ⋅ ( ρ u E + P ⋅ u + Q ) = ρ a ⋅ u (7-3) \frac{∂(\rho E)}{\partial t} + ∇⋅(\rho \mathbf{u}E + P⋅\mathbf{u} + \mathbf{Q})=\rho \mathbf{a⋅u} \tag{7-3}
∂ t ∂ ( ρ E ) + ∇ ⋅ ( ρ u E + P ⋅ u + Q ) = ρ a ⋅ u ( 7 - 3 )
Chapman-Enskog展开(下文简称C-E展开)是处理Boltzmann方程的一种分析方法,该方法引入小量ε \varepsilon ε 作为展开因子(一般与Knudsen数同阶),将BGK方程在不同的尺度上展开,即:
f = ∑ n = 0 ∞ ε n f ( n ) (8) f = \sum_{n=0}^{\infty} \varepsilon^n f^{(n)} \tag{8}
f = n = 0 ∑ ∞ ε n f ( n ) ( 8 )
并将分布函数、导数、物理量等都按照ε \varepsilon ε 的不同阶次展开。该方法对 f ( n ) f^{(n)} f ( n ) 进行了限制,使其满足:∫ f ( n ) φ i d ξ = 0 \int f^{(n)} φ_i \mathrm{d}ξ=0 ∫ f ( n ) φ i d ξ = 0 , φ i ( i = 1 , . . , 5 ) φ_i (i=1,..,5) φ i ( i = 1 , . . , 5 ) 为基本碰撞不变量。将 f f f 视为 ρ \rho ρ 、u \mathbf{u} u 和 T T T 及其各阶导数的泛函,而时间偏导项则展开为
∂ ∂ t = ∑ n = 0 ∞ ∂ n ∂ t (9) \frac{\partial }{\partial t} = \sum_{n=0}^{\infty } \frac{\partial _n}{\partial t} \tag{9}
∂ t ∂ = n = 0 ∑ ∞ ∂ t ∂ n ( 9 )
将(8)式和(9)式代入到(7)式中。得 O ( ε 0 ) O({\varepsilon}^0) O ( ε 0 ) 方程组为:
∂ 0 ρ ∂ t + ∇ ⋅ ( ρ u ) = 0 (10-1) \frac{∂_0 ρ}{\partial t}+∇⋅(ρ \mathbf{u})=0 \tag{10-1}
∂ t ∂ 0 ρ + ∇ ⋅ ( ρ u ) = 0 ( 1 0 - 1 )
∂ 0 ( ρ u ) ∂ t + ∇ ⋅ ( ρ u u + P ( 0 ) ) = ρ a (10-2) \frac{∂_0 (ρu)}{\partial t}+∇⋅(ρ \mathbf{uu}+P^{(0)}) = ρ \mathbf{a} \tag{10-2}
∂ t ∂ 0 ( ρ u ) + ∇ ⋅ ( ρ u u + P ( 0 ) ) = ρ a ( 1 0 - 2 )
∂ 0 ( ρ E ) ∂ t + ∇ ⋅ ( ρ u E + P ( 0 ) ⋅ u + Q ( 0 ) ) = ρ a ⋅ u (10-3) \frac{∂_0(\rho E)}{\partial t} + ∇⋅(\rho \mathbf{u}E + P^{(0)} ⋅ \mathbf{u} + \mathbf{Q}^{(0)})=\rho \mathbf{a⋅u} \tag{10-3}
∂ t ∂ 0 ( ρ E ) + ∇ ⋅ ( ρ u E + P ( 0 ) ⋅ u + Q ( 0 ) ) = ρ a ⋅ u ( 1 0 - 3 )
以及 O ( ε n ) O({\varepsilon}^n) O ( ε n ) 方程组为 ( n > 0 ) (n>0) ( n > 0 ) :
∂ n ρ ∂ t = 0 (11-1) \frac{∂_n ρ}{\partial t}=0 \tag{11-1}
∂ t ∂ n ρ = 0 ( 1 1 - 1 )
∂ n ( ρ u ) ∂ t + ∇ ⋅ P ( n ) = 0 (11-2) \frac{∂_n (ρu)}{\partial t}+∇ ⋅ P^{(n)} = 0 \tag{11-2}
∂ t ∂ n ( ρ u ) + ∇ ⋅ P ( n ) = 0 ( 1 1 - 2 )
∂ n ( ρ E ) ∂ t + ∇ ⋅ ( P ( n ) ⋅ u + Q ( n ) ) = 0 (11-3) \frac{∂_n (\rho E)}{\partial t} + ∇⋅( P^{(n)} ⋅ \mathbf{u} + \mathbf{Q}^{(n)} ) = 0 \tag{11-3}
∂ t ∂ n ( ρ E ) + ∇ ⋅ ( P ( n ) ⋅ u + Q ( n ) ) = 0 ( 1 1 - 3 )
其中
C = ξ − u , P ( n ) = m ∫ C C f ( n ) d ξ , Q ( n ) = m ∫ C C 2 2 f ( n ) d ξ C=\boldsymbol{ξ-u}, P^{(n)}=m∫CCf^{(n)} \mathrm{d}ξ , Q^{(n)}=m∫C \frac{C^2}{2} f^{(n)} \mathrm{d}ξ
C = ξ − u , P ( n ) = m ∫ C C f ( n ) d ξ , Q ( n ) = m ∫ C 2 C 2 f ( n ) d ξ
只需取精确到O ( ε 1 ) O({\varepsilon}^1) O ( ε 1 ) 的宏观方程组进行分析,即可还原出Naiver-Stokes方程。且具有如下关系:
P = P ( 0 ) + ε P ( 1 ) = p I − 2 ε μ ( S − T r ( S ) 3 I ) P=P^{(0)} + \varepsilon P^{(1)} = p \mathbf{I} - 2 \varepsilon \mu (\mathbf{S} - \frac{\mathrm{Tr}(S)}{3} \mathbf{I})
P = P ( 0 ) + ε P ( 1 ) = p I − 2 ε μ ( S − 3 T r ( S ) I )
Q = Q ( 0 ) + ε Q ( 1 ) = − ε κ ∇ T Q = Q^{(0)} + \varepsilon Q^{(1)} = - \varepsilon \kappa \nabla{T}
Q = Q ( 0 ) + ε Q ( 1 ) = − ε κ ∇ T
其中:κ \kappa κ 为热传导系数, μ \mu μ 为粘性系数。
离散的LBGK方程也可采用类似的方法展开,还原出宏观流体运动方程(见附录A)。展开方式如下:
f i = f i ( 0 ) + ε f i ( 1 ) + ε 2 f i ( 2 ) + . . . f_i = f_i^{(0)} + \varepsilon f_i^{(1)} + \varepsilon^2 f_i^{(2)} + ...
f i = f i ( 0 ) + ε f i ( 1 ) + ε 2 f i ( 2 ) + . . .
∂ ∂ x = ε ∂ ∂ x 1 , x 1 = ε x \frac{\partial}{\partial x} = \varepsilon \frac{\partial}{\partial x_1} , x_1 = \varepsilon x
∂ x ∂ = ε ∂ x 1 ∂ , x 1 = ε x
略去O ( ε 3 ) O({\varepsilon}^3) O ( ε 3 ) 项,并对O ( ε 0 ) O({\varepsilon}^0) O ( ε 0 ) 项、O ( ε 1 ) O({\varepsilon}^1) O ( ε 1 ) 项和O ( ε 2 ) O({\varepsilon}^2) O ( ε 2 ) 项求0阶和1阶速度矩,即可得到不可压缩流体的Navier-Stokes方程。由于在分析过程中略去了O ( ε 3 ) O({\varepsilon}^3) O ( ε 3 ) 项和速度的3阶项,导致结果缺失了部分物理信息。
三、 Grad-13矩方法及其发展
3.1 Grad的矩分析方法
Grad -13矩是一种由数学家Harold Grad提出的分析方法,其基于Hermite多项式进行分析。对于一个长度为 N N N 的向量 x \boldsymbol{x} x ,若定义权函数
ω ( x ) = 1 ( 2 π ) N / 2 e x p ( − x ⋅ x 2 ) (12) \omega( \boldsymbol{x}) = \frac{1}{(2 \pi)^{N/2}} \mathrm{exp} \left(- \frac{ \boldsymbol{x} \cdot \boldsymbol{x}}{2} \right) \tag{12}
ω ( x ) = ( 2 π ) N / 2 1 e x p ( − 2 x ⋅ x ) ( 1 2 )
则其对应的n n n 阶Hermite多项式写作
H ( n ) ( x ) = ( − 1 ) n ω ( x ) ∇ n ω ( x ) (13) \mathcal{H} ^{(n)} ( \boldsymbol{x}) = \frac{(-1)^n}{\omega ( \boldsymbol{x})} \nabla^{n} \omega ( \boldsymbol{x}) \tag{13}
H ( n ) ( x ) = ω ( x ) ( − 1 ) n ∇ n ω ( x ) ( 1 3 )
由Hermite多项式的正交性,可证明得到:
∫ − ∞ + ∞ ω ( x ) H i ( m ) ( x ) H j ( n ) ( x ) d x = δ m n δ i j (14) ∫_{-∞}^{+∞} \omega ( \boldsymbol{x}) H_i^{(m)}( \boldsymbol{x}) H_j^{(n)}( \boldsymbol{x}) \mathrm{d} \boldsymbol{x} = \delta_{mn} \delta_{ij} \tag{14}
∫ − ∞ + ∞ ω ( x ) H i ( m ) ( x ) H j ( n ) ( x ) d x = δ m n δ i j ( 1 4 )
据此,可将任意一个函数 f ( x ) f(x) f ( x ) 使用Hermite多项式进行展开。即:
f ( x ) = ω 1 2 ∑ n = 0 ∞ ∑ i = 1 N a i ( n ) H i ( n ) ( x ) (15) f( \boldsymbol{x})=ω^{\frac{1}{2}} \sum_{n=0}^∞ \sum_{i=1}^N a_i^{(n)} H_i^{(n)}( \boldsymbol{x}) \tag{15}
f ( x ) = ω 2 1 n = 0 ∑ ∞ i = 1 ∑ N a i ( n ) H i ( n ) ( x ) ( 1 5 )
所以
∫ − ∞ − ∞ ω 1 / 2 f ( x ) H j ( m ) ( x ) d x = ∑ i = 1 N a i ( m ) δ i j m = m ! a i m (16) \int_{-\infty}^{-\infty} \omega^{1/2} f(x) \mathcal{H}_j^{(m)} (x) \mathrm{d} \it{x} = \sum_{i=\mathrm{1}}^{N} a_i^{\mathrm{(}m\mathrm{)}} \delta_{ij}^m = m\mathrm{!} a_i^{m} \mathrm{\tag{16}}
∫ − ∞ − ∞ ω 1 / 2 f ( x ) H j ( m ) ( x ) d x = i = 1 ∑ N a i ( m ) δ i j m = m ! a i m ( 1 6 )
其中 a i ( m ) a_i^{(m)} a i ( m ) 可借由Hermite多项式的正交性进行求解。 δ i j m \delta_{ij}^{m} δ i j m 是一个记号,表示 m m m 个 δ i j \delta_{ij} δ i j 的乘积之和, i i i 和 j j j 分别来自下标集 ( i 1 , i 2 , . . . , i m ) (i_1 , i_2 , ... , i_m) ( i 1 , i 2 , . . . , i m ) 和 ( j 1 , j 2 , . . . , j m ) (j_1 , j_2 , ... , j_m) ( j 1 , j 2 , . . . , j m ) 。
在此数学基础之上,Harold Grad将分布函数 f f f 展开为:
f ( x , ξ , t ) = ω ∑ n = 0 ∞ 1 n ! a ( n ) H ( n ) ( ξ ) (17) f(x, \xi, t) = \omega \sum_{n=0}^{\infty} \frac{1}{n!} a^{(n)} H^{(n)}(\xi) \tag{17}
f ( x , ξ , t ) = ω n = 0 ∑ ∞ n ! 1 a ( n ) H ( n ) ( ξ ) ( 1 7 )
根据式(16), a ( n ) a^{(n)} a ( n ) 的表达式为:
a ( n ) = ∫ − ∞ + ∞ f ( x , ξ , t ) H ( n ) ( ξ ) d ξ (18) a^{(n)} = \int_{-\infty}^{+\infty} f(x, \xi, t) H^{(n)}(\xi) d\xi \tag{18}
a ( n ) = ∫ − ∞ + ∞ f ( x , ξ , t ) H ( n ) ( ξ ) d ξ ( 1 8 )
可得 a ( n ) a^{(n)} a ( n ) 的前4项为:
a ( 0 ) = ∫ − ∞ + ∞ f ( x , ξ , t ) d ξ = ρ (19-1) a^{(0)} = \int_{-\infty}^{+\infty} f(x, \xi, t) d\xi= \rho \tag{19-1}
a ( 0 ) = ∫ − ∞ + ∞ f ( x , ξ , t ) d ξ = ρ ( 1 9 - 1 )
a ( 1 ) = ∫ − ∞ + ∞ f ( x , ξ , t ) ξ d ξ = ρ u (19-2) a^{(1)} = \int_{-\infty}^{+\infty} f(x, \xi, t) \xi d\xi = \rho u \tag{19-2}
a ( 1 ) = ∫ − ∞ + ∞ f ( x , ξ , t ) ξ d ξ = ρ u ( 1 9 - 2 )
a ( 2 ) = ∫ − ∞ + ∞ f ( x , ξ , t ) ( ξ 2 − δ ) d ξ = ρ ( u 2 − δ ) + P (19-3) a^{(2)} = \int_{-\infty}^{+\infty} f(x, \xi, t) (\xi^2 - \delta) d\xi = \rho (\mathbf{u}^2 - \mathbf{\delta}) + \mathbf{P} \tag{19-3}
a ( 2 ) = ∫ − ∞ + ∞ f ( x , ξ , t ) ( ξ 2 − δ ) d ξ = ρ ( u 2 − δ ) + P ( 1 9 - 3 )
a ( 3 ) = ∫ − ∞ + ∞ f ( x , ξ , t ) ( ξ 3 − ξ δ ) d ξ = ( D − 1 ) ρ u 3 + u a ( 2 ) + Q (19-4) a^{(3)} = \int_{-\infty}^{+\infty} f(x, \xi, t) (\xi^3 - \xi \delta) d\xi = (D-1) \rho \bf{u}^3 + \bf{u} \it{a}^{\mathrm{(2)}} \mathrm{+} \mathbf{Q} \tag{19-4}
a ( 3 ) = ∫ − ∞ + ∞ f ( x , ξ , t ) ( ξ 3 − ξ δ ) d ξ = ( D − 1 ) ρ u 3 + u a ( 2 ) + Q ( 1 9 - 4 )
注意到从 a ( 0 ) a^{(0)} a ( 0 ) 到 a ( 3 ) a^{(3)} a ( 3 ) 都描述了流场的宏观量,所以通过上述表达式的值可以解出流场宏观量。与C-E分析方法不同的是分析过程中没有舍去高阶量。因此这种展开方法得到的式子合理地引入了更高阶的矩,物理意义更加明确。
3.2 Gauss-Hermite积分的引入和离散化BGK方程
Grad的思路开创了描述的新方式,而单肖文、袁学锋、陈沪东三位学者在此基础上进一步发展,于2006年在 Journal of Fluid Mechanics 上提出了将原思路拓展到离散的Boltzmann-BGK方程中的方法,简化了求解过程。
首先,基于式(16)将 f f f 截断至前 N N N 阶,得:
f ≈ f N = ω ∑ n = 0 N 1 n ! a ( n ) H ( n ) ( ξ ) (20) f \approx f^N = \omega \sum_{n=0}^{N} \frac{1}{n!} a^{(n)} H^{(n)}(\xi) \tag{20}
f ≈ f N = ω n = 0 ∑ N n ! 1 a ( n ) H ( n ) ( ξ ) ( 2 0 )
f N f^N f N 和 f f f 有着完全相同的前 N N N 阶矩。通过截断操作将BGK方程对宏观流动方程的近似保持在动力学层面而不是热力学层面。此时f N f^N f N 可表示为:
f N ( x , ξ , t ) H ( n ) ( ξ ) = ω ( ξ ) p ( x , ξ , t ) (21) f^{N}(x, \xi , t) H^{(n)}(\xi) = \omega(\xi) p(x, \xi, t) \tag{21}
f N ( x , ξ , t ) H ( n ) ( ξ ) = ω ( ξ ) p ( x , ξ , t ) ( 2 1 )
其中 p ( x , ξ , t ) p(x, \xi, t) p ( x , ξ , t ) 为阶数不超过 2 N 2N 2 N 的多项式。使用Gauss-Hermite积分可将 a ( n ) a^{(n)} a ( n ) 精确地展开为 p ( x , ξ , t ) p(x, \xi, t) p ( x , ξ , t ) 的加权和。
a ( n ) = ∫ ω ( ξ ) p ( x , ξ , t ) d ξ = ∑ a = 1 d w a p ( x , ξ a , t ) = ∑ a = 1 d w a ω ( ξ a ) f N ( x , ξ a , t ) H ( n ) ( ξ a ) (22) a^{(n)} =\int \omega(\xi) p(x, \xi, t) d \xi = \sum_{a=1}^{d} w_a p(x, \xi_a, t) = \sum_{a=1}^{d} \frac{w_a}{\omega(\xi_a)} f^{N}(x, \xi_a, t) H^{(n)}(\xi_a) \tag{22}
a ( n ) = ∫ ω ( ξ ) p ( x , ξ , t ) d ξ = a = 1 ∑ d w a p ( x , ξ a , t ) = a = 1 ∑ d ω ( ξ a ) w a f N ( x , ξ a , t ) H ( n ) ( ξ a ) ( 2 2 )
式(22)中 w a w_a w a 和 $ \boldsymbol{\xi}_a$ 是Gauss-Hermite积分的权重和特征向量( a = 1 , 2 , . . . , d a=1,2,...,d a = 1 , 2 , . . . , d )。所以使用 f N ( x , ξ a , t ) f^N (x, \boldsymbol{\xi}_a, t) f N ( x , ξ a , t ) 表示 f N f^N f N 的前 N N N 阶矩在数学上可行。
据此可定义离散化的分布函数 f a f_a f a ( a = 1 , 2 , . . . , d a=1,2,...,d a = 1 , 2 , . . . , d ),其对应的权重和特征方向为 w a w_a w a 和 ξ a \xi_a ξ a :
f a ( x , t ) = w a ω ( ξ a ) f ( x , ξ a , t ) (23) f_a (x,t) = \frac{w_a}{\omega( \boldsymbol{\xi}_a)} f(x, \boldsymbol{\xi}_a, t) \tag{23}
f a ( x , t ) = ω ( ξ a ) w a f ( x , ξ a , t ) ( 2 3 )
则宏观量可表达为:
ρ = ∑ a = 1 d f a , ρ u = ∑ a = 1 d f a ξ a , ρ u u + P = ∑ a = 1 d f a ξ a ξ a (24) \rho = \sum_{a=1}^{d} f_a , \rho \mathbf{u} = \sum_{a=1}^{d} f_a \mathbf{\xi}_a , \rho \mathbf{uu} + \mathbf{P} = \sum_{a=1}^{d} f_a \mathbf{\xi}_a \mathbf{\xi}_a \tag{24}
ρ = a = 1 ∑ d f a , ρ u = a = 1 ∑ d f a ξ a , ρ u u + P = a = 1 ∑ d f a ξ a ξ a ( 2 4 )
下面将(1)式的碰撞项换为(2)式,并把 F m ∂ f ∂ ξ \frac{\mathbf{F}}{m} \frac{\partial f}{\partial \xi} m F ∂ ξ ∂ f 直接记作外力项 − F ( ξ ) -\mathbf{F} (\xi) − F ( ξ ) ,得
∂ f ∂ t + ξ ⋅ ∂ f ∂ x = − 1 τ ( f − f ( e q ) ) + F ( ξ ) (25) \frac{\partial f}{\partial t} + \mathbf{\xi} \cdot \frac{\partial f}{\partial \mathit{x}} = \frac{-1}{\tau} ( f - f^{(eq)} ) + \mathbf{F} (\xi) \tag{25}
∂ t ∂ f + ξ ⋅ ∂ x ∂ f = τ − 1 ( f − f ( e q ) ) + F ( ξ ) ( 2 5 )
对式(25),令 ξ = ξ a \boldsymbol{\xi} = \boldsymbol{\xi}_a ξ = ξ a ,并在两端乘上 w a ω ( ξ a ) \frac{w_a}{\omega( \boldsymbol{\xi}_a)} ω ( ξ a ) w a ,即可得到使用 f a f_a f a 离散的方程( a = 1 , 2 , . . . , d a=1,2,...,d a = 1 , 2 , . . . , d ):
∂ f a ∂ t + ξ a ⋅ ∇ f a = − 1 τ ( f a − f a ( e q ) ) + F a (26-1) \frac{\partial f_a}{\partial t} + \xi_a \cdot \nabla{f_a} = -\frac{1}{\tau} ( f_a - f_a^{(eq)} ) + F_a \tag{26-1}
∂ t ∂ f a + ξ a ⋅ ∇ f a = − τ 1 ( f a − f a ( e q ) ) + F a ( 2 6 - 1 )
f a ( e q ) = w a ω ( ξ a ) f ( e q ) ( ξ a ) (26-2) f_a^{(eq)} = \frac{w_a}{\omega (\xi_a)} f^{(eq)} (\xi_a) \tag{26-2}
f a ( e q ) = ω ( ξ a ) w a f ( e q ) ( ξ a ) ( 2 6 - 2 )
F a = w a ω ( ξ a ) F ( ξ a ) (26-3) F_a = \frac{w_a}{\omega (\xi_a)} \mathbf{F} (\xi_a) \tag{26-3}
F a = ω ( ξ a ) w a F ( ξ a ) ( 2 6 - 3 )
此时只需将 f ( e q ) f^{(eq)} f ( e q ) 和 F F F 进行类似的展开,即可完成对整个系统的离散。
附录A. 使用C-E展开方法将离散的LBGK方程还原为Navier-Stokes方程
LBE方程为(A-1.1)式。源项 F \boldsymbol{F} F 和 f ( e q ) f^{(\mathrm{eq})} f ( e q ) 为由郭照立、郑楚光和施保昌提出的格式,即(A-1.2)式和(A-1.3)式。[注意,这里的 f f f 是密度分布函数]
f i ( x + ξ i Δ t , t + Δ t ) − f i ( x , t ) = − 1 τ [ f i ( x , t ) − f i ( e q ) ( x , t ) ] + Δ t ⋅ F i ( x , t ) (A-1.1) f_i (x+ξ_i Δt,t+Δt)-f_i (x,t)=-\frac{1}{τ} \left[f_i (x,t)-f_i^{(eq)} (x,t) \right]+Δt⋅F_i (x,t) \tag{A-1.1}
f i ( x + ξ i Δ t , t + Δ t ) − f i ( x , t ) = − τ 1 [ f i ( x , t ) − f i ( e q ) ( x , t ) ] + Δ t ⋅ F i ( x , t ) ( A - 1 . 1 )
F i = ( 1 − Δ t 2 τ ) w i [ ξ i − u c s 2 + ( ξ i ⋅ u ) ξ i c s 4 ] ⋅ F (A-1.2) F_i=(1 - \frac{Δt}{2τ}) w_i \left[ \frac{\mathbf{\xi_i-u} }{c_s^2 } + \frac{\mathbf{(\xi_i \cdot u) \xi_i} }{c_s^4} \right] ⋅ F \tag{A-1.2}
F i = ( 1 − 2 τ Δ t ) w i [ c s 2 ξ i − u + c s 4 ( ξ i ⋅ u ) ξ i ] ⋅ F ( A - 1 . 2 )
f i ( e q ) ( x , t ) = ρ w i [ 1 + ξ i ⋅ u c s 2 + u u : ( ξ i ξ i − c s 2 I ) 2 c s 4 ] (A-1.3) f_i^{(eq)} (x,t) = \rho w_i \left[ 1 + \frac{\mathbf{ξ_i⋅u}}{c_s^2} +\frac{ \mathbf{uu} : ( \mathbf{ξ_i ξ_i} - c_s^2 \mathbf{I}) }{2c_s^4} \right] \tag{A-1.3}
f i ( e q ) ( x , t ) = ρ w i [ 1 + c s 2 ξ i ⋅ u + 2 c s 4 u u : ( ξ i ξ i − c s 2 I ) ] ( A - 1 . 3 )
并且对于 f i f_i f i 和 f i ( e q ) f_i^{(eq)} f i ( e q ) 有:
{ ρ = ∑ i f i ρ u = ∑ i ξ i f i + Δ t 2 F \begin{cases} \rho = \sum_i {f_i} \\ \rho \mathbf{u} = \sum_i {\mathbf{\xi_i} f_i} + \frac{\Delta t}{2} F \end{cases} { ρ = ∑ i f i ρ u = ∑ i ξ i f i + 2 Δ t F
和
{ ρ = ∑ i f i ( e q ) ρ u = ∑ i ξ i f i ( e q ) ρ ( u u + c s 2 I ) = ∑ i ξ i ξ i f i ( e q ) \begin{cases} \rho = \sum_i {f_i^{(eq)}} \\ \rho \mathbf{u} = \sum_i {\mathbf{\xi_i} f_i^{(eq)}} \\ \rho (\mathbf{uu} + c_s^2 \mathbf{I}) = \sum_i {\mathbf{\xi_i \xi_i} f_i^{(eq)}} \end{cases}
⎩ ⎪ ⎪ ⎨ ⎪ ⎪ ⎧ ρ = ∑ i f i ( e q ) ρ u = ∑ i ξ i f i ( e q ) ρ ( u u + c s 2 I ) = ∑ i ξ i ξ i f i ( e q )
将(A-1.1)式在 ( x , t ) (\mathbf{x} , t) ( x , t ) 处展开,取至 2 阶项得:
Δ t ∂ f i ∂ t + ξ i Δ t ⋅ ∂ f i ∂ x + 1 2 [ Δ t 2 ∂ 2 f i ∂ t 2 + 2 Δ t ( ξ i Δ t ) ⋅ ∂ 2 f i ∂ t ∂ x + Δ t 2 ( ξ i ξ i ) : ∂ 2 f i ∂ x 2 ] + 1 τ [ f i − f i ( e q ) ] − Δ t F i = 0 (A-2) \begin{aligned}
\Delta t \frac{\partial f_i}{\partial t} + ξ_i \Delta t ⋅ \frac{\partial f_i}{∂x} + &\\
\frac{1}{2} \left[\Delta t^2 \frac{\partial^2 f_i}{\partial t^2} + 2\Delta t(\xi_i \Delta t) ⋅ \frac{\partial^2 f_i}{\partial t∂x} + \Delta t^2 (\xi_i \xi_i): \frac{\partial^2 f_i}{∂x^2} \right] + &\\
\frac{1}{\tau } \left[f_i - f_i^{(eq)} \right] - \Delta t F_i &= 0 \tag{A-2}
\end{aligned}
Δ t ∂ t ∂ f i + ξ i Δ t ⋅ ∂ x ∂ f i + 2 1 [ Δ t 2 ∂ t 2 ∂ 2 f i + 2 Δ t ( ξ i Δ t ) ⋅ ∂ t ∂ x ∂ 2 f i + Δ t 2 ( ξ i ξ i ) : ∂ x 2 ∂ 2 f i ] + τ 1 [ f i − f i ( e q ) ] − Δ t F i = 0 ( A - 2 )
代入多尺度展开等式,并舍去 ε 3 \varepsilon^3 ε 3 项和更高阶的项,可得:
( ε ∂ f i ( 0 ) ∂ t 1 + ε 2 ∂ f i ( 1 ) ∂ t 1 + ε 2 ∂ f i ( 0 ) ∂ t 2 ) + ξ i ⋅ ( ε ∂ f i ( 0 ) ∂ x 1 + ε 2 ∂ f i ( 0 ) ∂ x 1 ) + Δ t 2 [ ε 2 ∂ 2 f i ( 0 ) ∂ t 1 2 + 2 ξ i ε 2 ⋅ ∂ 2 f i ( 0 ) ∂ t 1 ∂ x 1 + ( ξ i ξ i ) : ( ε 2 ∂ 2 f i ( 0 ) ∂ x 2 ) ] + 1 τ Δ t [ f i ( 0 ) + ε f i ( 1 ) + ε 2 f i ( 2 ) − f i ( e q ) ] − ε F i ( 1 ) + O ( ε 3 ) = 0 (A-3) \begin{aligned}
(\varepsilon \frac{\partial f_i^{(0)}}{\partial t_1} + \varepsilon^2 \frac{\partial f_i^{(1)}}{\partial t_1} + \varepsilon^2 \frac{\partial f_i^{(0)}}{\partial t_2} ) + \xi_i ⋅ (\varepsilon \frac{\partial f_i^{(0)}}{\partial x_1} + \varepsilon^2 \frac{\partial f_i^{(0)}}{\partial x_1}) + &\\
\frac{Δt}{2} \left[ \varepsilon^2 \frac{\partial^2 f_i^{(0)}}{\partial t_1^2} + 2 \xi_i \varepsilon^2 ⋅ \frac{\partial^2 f_i^{(0)}}{\partial t_1 \partial x_1} + (\xi_i \xi_i) : (\varepsilon^2 \frac{\partial^2 f_i^{(0)}}{\partial x^2} ) \right] &\\
+ \frac{1}{τΔt} \left[ f_i^{(0)} + \varepsilon f_i^{(1)} + \varepsilon^2 f_i^{(2)} -f_i^{(eq)} \right] - \varepsilon F_i^{(1)} + O(\varepsilon^3 ) &=0 \tag{A-3}
\end{aligned}
( ε ∂ t 1 ∂ f i ( 0 ) + ε 2 ∂ t 1 ∂ f i ( 1 ) + ε 2 ∂ t 2 ∂ f i ( 0 ) ) + ξ i ⋅ ( ε ∂ x 1 ∂ f i ( 0 ) + ε 2 ∂ x 1 ∂ f i ( 0 ) ) + 2 Δ t [ ε 2 ∂ t 1 2 ∂ 2 f i ( 0 ) + 2 ξ i ε 2 ⋅ ∂ t 1 ∂ x 1 ∂ 2 f i ( 0 ) + ( ξ i ξ i ) : ( ε 2 ∂ x 2 ∂ 2 f i ( 0 ) ) ] + τ Δ t 1 [ f i ( 0 ) + ε f i ( 1 ) + ε 2 f i ( 2 ) − f i ( e q ) ] − ε F i ( 1 ) + O ( ε 3 ) = 0 ( A - 3 )
从(A-3)式中提取O ( ε 0 ) O({\varepsilon}^0) O ( ε 0 ) 项、O ( ε 1 ) O({\varepsilon}^1) O ( ε 1 ) 项和O ( ε 2 ) O({\varepsilon}^2) O ( ε 2 ) 项,并令 D k = ∂ ∂ t k + ξ k ⋅ ∂ ∂ x k D_k = \frac{\partial}{\partial t_k} + \xi_k \cdot \frac{\partial}{\partial \mathbf{x}_k} D k = ∂ t k ∂ + ξ k ⋅ ∂ x k ∂ 。(A-3)式对任意 ε \varepsilon ε 都成立,则关于各阶的系数均为0,即:
O ( ε 0 ) = 1 τ Δ t ( f i ( 0 ) − f i ( e q ) ) = 0 (A-4.1) O({\varepsilon}^0) = \frac{1}{\tau \Delta t} (f_i^{(0)} - f_i^{(eq)}) = 0 \tag{A-4.1}
O ( ε 0 ) = τ Δ t 1 ( f i ( 0 ) − f i ( e q ) ) = 0 ( A - 4 . 1 )
O ( ε 1 ) = D 1 f i ( 0 ) + f i ( 1 ) τ Δ t − F i ( 1 ) = 0 (A-4.2) O(\varepsilon^1) = D_1 f_i^{(0)} + \frac{ f_i^{(1)} }{\tau \Delta t} - F_i^{(1)} = 0 \tag{A-4.2}
O ( ε 1 ) = D 1 f i ( 0 ) + τ Δ t f i ( 1 ) − F i ( 1 ) = 0 ( A - 4 . 2 )
O ( ε 2 ) = ∂ f i ( 0 ) ∂ t 2 − ( 1 − 1 2 τ ) D 1 f i ( 1 ) + f i ( 2 ) τ Δ t + Δ t 2 D 1 F i ( 1 ) = 0 (A-4.3) O(\varepsilon^2) = \frac{\partial f_i^{(0)}}{\partial t_2} - (1 - \frac{1}{2\tau}) D_1 f_i^{(1)} + \frac{ f_i^{(2)} }{\tau \Delta t} + \frac{\Delta t}{2} D_1 F_i^{(1)} = 0 \tag{A-4.3}
O ( ε 2 ) = ∂ t 2 ∂ f i ( 0 ) − ( 1 − 2 τ 1 ) D 1 f i ( 1 ) + τ Δ t f i ( 2 ) + 2 Δ t D 1 F i ( 1 ) = 0 ( A - 4 . 3 )
根据多尺度展开的变量关系,可得:
∑ i f i ( 1 ) = ∑ i f i ( 2 ) = 0 \sum_i {f_i^{(1)}} = \sum_i {f_i^{(2)}} = 0 ∑ i f i ( 1 ) = ∑ i f i ( 2 ) = 0 , ∑ i ξ i f i ( 2 ) = 0 \sum_i { \xi_i f_i^{(2)}} = 0 ∑ i ξ i f i ( 2 ) = 0 , ∑ i ξ i f i ( 1 ) = − Δ t 2 F i ( 1 ) \sum_i { \xi_i f_i^{(1)}} = - \frac{\Delta t}{2} F_i^{(1)} ∑ i ξ i f i ( 1 ) = − 2 Δ t F i ( 1 )
{ ∑ i F i = 0 ∑ i ξ i F i = ( 1 − 1 2 τ ) F ∑ i ξ i ξ i F i = ( 1 − 1 2 τ ) 2 u F \left\{\begin{matrix} \sum_{i} {F_i} = 0 \\ \sum_{i} {\xi_i F_i} = (1 - \frac{1}{2 \tau}) F \\ \sum_{i} {\xi_i \xi_i F_i} = (1 - \frac{1}{2 \tau}) 2 u F \end{matrix}\right.
⎩ ⎪ ⎨ ⎪ ⎧ ∑ i F i = 0 ∑ i ξ i F i = ( 1 − 2 τ 1 ) F ∑ i ξ i ξ i F i = ( 1 − 2 τ 1 ) 2 u F
,
{ ∑ i F i ( 1 ) = 0 ∑ i ξ i F i ( 1 ) = ( 1 − 1 2 τ ) F ( 1 ) ∑ i ξ i ξ i F i ( 1 ) = ( 1 − 1 2 τ ) 2 u F ( 1 ) \left\{\begin{matrix} \sum_{i} {F_i^{(1)}} = 0 \\ \sum_{i} {\xi_i F_i^{(1)}} = (1 - \frac{1}{2 \tau}) F^{(1)} \\ \sum_{i} {\xi_i \xi_i F_i^{(1)}} = (1 - \frac{1}{2 \tau}) 2 u F^{(1)} \end{matrix}\right.
⎩ ⎪ ⎪ ⎨ ⎪ ⎪ ⎧ ∑ i F i ( 1 ) = 0 ∑ i ξ i F i ( 1 ) = ( 1 − 2 τ 1 ) F ( 1 ) ∑ i ξ i ξ i F i ( 1 ) = ( 1 − 2 τ 1 ) 2 u F ( 1 )
为方便下文分析,记: Π ( k ) = ∑ i ξ i ξ i f i ( k ) \Pi^{(k)} = \sum_i { \xi_i \xi_i f_i^{(k)} } Π ( k ) = ∑ i ξ i ξ i f i ( k ) , Γ ( k ) = ∑ i ξ i ξ i ξ i f i ( k ) \Gamma^{(k)} = \sum_i { \xi_i \xi_i \xi_i f_i^{(k)} } Γ ( k ) = ∑ i ξ i ξ i ξ i f i ( k ) 。
对 O ( ε 1 ) O({\varepsilon}^1) O ( ε 1 ) 项求1阶和2阶速度矩,得:
∂ ρ ∂ t 1 + ∂ ( ρ u ) ∂ x 1 = 0 (A-5.1) \frac{\partial \rho}{\partial t_1} + \frac{\partial (\rho u)}{\partial x_1} = 0 \tag{A-5.1}
∂ t 1 ∂ ρ + ∂ x 1 ∂ ( ρ u ) = 0 ( A - 5 . 1 )
∂ ( ρ u ) ∂ t 1 + ∂ Π ( 0 ) ∂ x 1 = 0 (A-5.2) \frac{\partial (\rho u)}{\partial t_1} + \frac{\partial \Pi^{(0)}}{\partial x_1} = 0 \tag{A-5.2}
∂ t 1 ∂ ( ρ u ) + ∂ x 1 ∂ Π ( 0 ) = 0 ( A - 5 . 2 )
Π ( 0 ) = ρ ( u u + c s 2 I ) (A-5.3) \Pi^{(0)} = \rho (\it{\bf{uu}} \mathrm{+} c_s^2 \bf{I}) \tag{A-5.3}
Π ( 0 ) = ρ ( u u + c s 2 I ) ( A - 5 . 3 )
对 O ( ε 2 ) O({\varepsilon}^2) O ( ε 2 ) 项求1阶和2阶速度矩,得:
∂ ρ ∂ t 2 = 0 (A-6.1) \frac{\partial {\rho}}{\partial t_{2}} = 0 \tag{A-6.1}
∂ t 2 ∂ ρ = 0 ( A - 6 . 1 )
∂ ( ρ u ) ∂ t 2 + ( 1 − 1 2 τ ) ∂ Π ( 1 ) ∂ x 1 + Δ t ( 1 − 1 2 τ ) ∂ F ( 1 ) ∂ x 1 = 0 (A-6.2) \frac{\partial (\rho \boldsymbol{u})}{\partial t_{2}} + (1 - \frac{1}{2 \tau}) \frac{\partial \Pi^{(1)}}{\partial \boldsymbol{x}_{1}} + \Delta t (1 - \frac{1}{2 \tau}) \frac{\partial \boldsymbol{F}^{(1)}}{\partial \boldsymbol{x}_{1}} = 0 \tag{A-6.2}
∂ t 2 ∂ ( ρ u ) + ( 1 − 2 τ 1 ) ∂ x 1 ∂ Π ( 1 ) + Δ t ( 1 − 2 τ 1 ) ∂ x 1 ∂ F ( 1 ) = 0 ( A - 6 . 2 )
由(A-4.2)式,得: f i ( 1 ) = τ Δ t { D 1 f i ( 0 ) − F i ( 1 ) } f_i^{(1)} = \tau \Delta t \{ D_1 f_i^{(0)} - F_i^{(1)}\} f i ( 1 ) = τ Δ t { D 1 f i ( 0 ) − F i ( 1 ) } ,所以
Π ( 1 ) = ∑ i ξ i ξ i f i ( 1 ) = − τ Δ t [ ∂ ( ρ u u ) ∂ t 1 + c s 2 ∂ ( ρ I ) ∂ t 1 + ∂ Γ ( 0 ) ∂ x 1 − ( 1 − 1 2 τ ) 2 u F ( 1 ) ] (A-7) \Pi^{(1)} = \sum_{i} \boldsymbol{\xi}_{i} \boldsymbol{\xi}_{i} f_{i}^{(1)} = -\tau \Delta t [\frac{\partial (\rho \boldsymbol{u} \boldsymbol{u})}{\partial t_{1}} + c_s^2 \frac{\partial (\rho \boldsymbol{I})}{\partial t_{1}} + \frac{\partial \Gamma^{(0)}}{\partial \boldsymbol{x}_{1}} - (1 - \frac{1}{2 \tau}) 2\boldsymbol{uF^{(1)}} ] \tag{A-7}
Π ( 1 ) = i ∑ ξ i ξ i f i ( 1 ) = − τ Δ t [ ∂ t 1 ∂ ( ρ u u ) + c s 2 ∂ t 1 ∂ ( ρ I ) + ∂ x 1 ∂ Γ ( 0 ) − ( 1 − 2 τ 1 ) 2 u F ( 1 ) ] ( A - 7 )
u F ( 1 ) \boldsymbol{uF^{(1)}} u F ( 1 ) 是Grad的论文中的记号, ( u F ( 1 ) ) l m = 1 2 ( u l F m ( 1 ) + u m F l ( 1 ) ) (\bf{\it{uF}}^{(1)} )_{\it{lm}} = \mathrm{\frac{1}{2}} (\it{ u_l F_m^{(1)} + u_m F_l^{(1)} } \mathrm{)} ( u F ( 1 ) ) l m = 2 1 ( u l F m ( 1 ) + u m F l ( 1 ) ) 。下文为方便分析,取 Π l m ( 1 ) \Pi_{lm}^{(1)} Π l m ( 1 ) 分析。
对第1项,有: ∂ ( ρ u l u m ) ∂ t 1 = ρ u l ∂ u m ∂ t 1 + u m ∂ ( ρ u l ) ∂ t 1 \frac{\partial (\rho u_l u_m)}{\partial t_1} = \rho u_l \frac{\partial u_m}{\partial t_1} + u_m \frac{\partial (\rho u_l)}{\partial t_1} ∂ t 1 ∂ ( ρ u l u m ) = ρ u l ∂ t 1 ∂ u m + u m ∂ t 1 ∂ ( ρ u l ) 。由(A-5.2)式,可得:
ρ ∂ u m ∂ t 1 + u m ∂ ρ ∂ t 1 + ∂ ∂ x 1 n ( ρ u m u n + ρ c s 2 δ m n ) = F m ( 1 ) (A-8.1) \rho \frac{\partial u_m}{\partial t_1} + u_m \frac{\partial \rho}{\partial t_1} + \frac{\partial}{\partial x_{1n}} \left( \rho u_m u_n + \rho c_s^2 \delta_{mn} \right) = F_m^{(1)} \tag{A-8.1}
ρ ∂ t 1 ∂ u m + u m ∂ t 1 ∂ ρ + ∂ x 1 n ∂ ( ρ u m u n + ρ c s 2 δ m n ) = F m ( 1 ) ( A - 8 . 1 )
∂ ( ρ u l ) ∂ t 1 + ∂ ∂ x 1 n ( ρ u l u n + ρ c s 2 δ l n ) = F l ( 1 ) (A-8.2) \frac{\partial (\rho u_l)}{\partial t_1}+ \frac{\partial}{\partial x_{1n}} \left( \rho u_l u_n + \rho c_s^2 \delta_{ln} \right) = F_l^{(1)} \tag{A-8.2}
∂ t 1 ∂ ( ρ u l ) + ∂ x 1 n ∂ ( ρ u l u n + ρ c s 2 δ l n ) = F l ( 1 ) ( A - 8 . 2 )
并且
∂ ( ρ u l u m u n ) ∂ x 1 n = − u l u m ∂ ( ρ u n ) ∂ x 1 n + u l ∂ ( ρ u m u n ) ∂ x 1 n + u m ∂ ( ρ u l u n ) ∂ x 1 n (A-8.3) \frac{\partial (\rho u_l u_m u_n)}{\partial x_{1n}} = - u_l u_m \frac{\partial (\rho u_n)}{\partial x_{1n}} + u_l \frac{\partial (\rho u_m u_n)}{\partial x_{1n}} + u_m \frac{\partial (\rho u_l u_n)}{\partial x_{1n}} \tag{A-8.3}
∂ x 1 n ∂ ( ρ u l u m u n ) = − u l u m ∂ x 1 n ∂ ( ρ u n ) + u l ∂ x 1 n ∂ ( ρ u m u n ) + u m ∂ x 1 n ∂ ( ρ u l u n ) ( A - 8 . 3 )
所以
∂ ( ρ u l u m ) ∂ x 1 n = ( u l F m ( 1 ) + u m F l ( 1 ) ) − ∂ ( ρ u l u m u n ) ∂ x 1 n − c s 2 ( u l ∂ ρ ∂ x 1 m + u m ∂ ρ ∂ x 1 l ) (A-8.4) \frac{\partial (\rho u_l u_m)}{\partial x_{1n}} = (u_l F_m^{(1)} + u_m F_l^{(1)}) - \frac{\partial (\rho u_l u_m u_n)}{\partial x_{1n}} - c_s^2 \left( u_l \frac{\partial \rho}{\partial x_{1m}} + u_m \frac{\partial \rho}{\partial x_{1l}} \right) \tag{A-8.4}
∂ x 1 n ∂ ( ρ u l u m ) = ( u l F m ( 1 ) + u m F l ( 1 ) ) − ∂ x 1 n ∂ ( ρ u l u m u n ) − c s 2 ( u l ∂ x 1 m ∂ ρ + u m ∂ x 1 l ∂ ρ ) ( A - 8 . 4 )
所以第2项的分量为: c s 2 δ l m ∂ ρ ∂ t 1 c_s^2 \delta_{lm} \frac{\partial \rho}{\partial t_1} c s 2 δ l m ∂ t 1 ∂ ρ 。
对第3项的分量,由于
Γ l m n ( 0 ) = ρ c s 2 ( u l δ m n + u n δ l m + u m δ l n ) \Gamma_{lmn}^{(0)} = \rho c_s^2 ( u_l \delta_{mn} + u_n \delta_{lm} + u_m \delta_{ln} )
Γ l m n ( 0 ) = ρ c s 2 ( u l δ m n + u n δ l m + u m δ l n )
,所以:
∂ Γ l m n ( 0 ) ∂ x 1 n = c s 2 ( δ m n ∂ ρ u l ∂ x 1 n + δ l m ∂ ρ u n ∂ x 1 n + δ l n ∂ ρ u m ∂ x 1 n ) (A-8.5) \frac{\partial \Gamma_{lmn}^{(0)}}{\partial x_{1n}} = c_s^2 ( \delta_{mn} \frac{\partial \rho u_l}{\partial x_{1n}} + \delta_{lm} \frac{\partial \rho u_n}{\partial x_{1n}} + \delta_{ln} \frac{\partial \rho u_m}{\partial x_{1n}} ) \tag{A-8.5}
∂ x 1 n ∂ Γ l m n ( 0 ) = c s 2 ( δ m n ∂ x 1 n ∂ ρ u l + δ l m ∂ x 1 n ∂ ρ u n + δ l n ∂ x 1 n ∂ ρ u m ) ( A - 8 . 5 )
所以
Π l m ( 1 ) = − τ Δ t [ − ∂ ( ρ u l u m u n ) ∂ x 1 n + 1 2 τ ( u l F m ( 1 ) + u m F l ( 1 ) ) + ρ c s 2 ( ∂ u l ∂ x 1 m + ∂ u m ∂ x 1 l ) ] (A-8.6) \Pi_{lm}^{(1)} = - \tau \Delta t \left[ - \frac{\partial (\rho u_l u_m u_n)}{\partial x_{1n}} + \frac{1}{2 \tau} (u_l F_m^{(1)} + u_m F_l^{(1)}) + \rho c_s^2 ( \frac{\partial u_l}{\partial x_{1m}} + \frac{\partial u_m}{\partial x_{1l}} ) \right] \tag{A-8.6}
Π l m ( 1 ) = − τ Δ t [ − ∂ x 1 n ∂ ( ρ u l u m u n ) + 2 τ 1 ( u l F m ( 1 ) + u m F l ( 1 ) ) + ρ c s 2 ( ∂ x 1 m ∂ u l + ∂ x 1 l ∂ u m ) ] ( A - 8 . 6 )
略去Π l m ( 1 ) \Pi_{lm}^{(1)} Π l m ( 1 ) 中的速度的3阶项(即 ∂ ∂ x 1 n { ρ u l u m u n } \frac{\partial}{\partial x_{1n}} \{ \rho u_l u_m u_n \} ∂ x 1 n ∂ { ρ u l u m u n } ),有
Π ( 1 ) = − Δ t ⋅ u F ( 1 ) − τ ρ c s 2 Δ t [ ∂ u ∂ x 1 + ( ∂ u ∂ x 1 ) T ] (A-9) \Pi^{(1)} = - \Delta t \cdot \boldsymbol{u F}^{(1)} - \tau \rho c_s^2 \Delta t \left[ \frac{\partial \boldsymbol{u}}{\partial \boldsymbol{x_1}} + \left( \frac{\partial \boldsymbol{u}}{\partial \boldsymbol{x_1}} \right)^{\mathrm{T}} \right] \tag{A-9}
Π ( 1 ) = − Δ t ⋅ u F ( 1 ) − τ ρ c s 2 Δ t [ ∂ x 1 ∂ u + ( ∂ x 1 ∂ u ) T ] ( A - 9 )
回代(A-6.2)式,得等价形式为
∂ ( ρ u ) ∂ t 2 + c s 2 Δ t ( τ − 1 2 ) ∂ ∂ x 1 { ρ [ ∂ u ∂ x 1 + ( ∂ u ∂ x 1 ) T ] } = 0 (A-10) \frac{\partial (\rho \boldsymbol{u})}{\partial t_2} + c_s^2 \Delta t (\tau - \frac{1}{2}) \frac{\partial}{\partial \boldsymbol{x_1}} \left\{ \rho \left[ \frac{\partial \boldsymbol{u}}{\partial \boldsymbol{x_1}} + \left( \frac{\partial \boldsymbol{u}}{\partial \boldsymbol{x_1}} \right)^{\mathrm{T}} \right] \right\} = \boldsymbol{0} \tag{A-10}
∂ t 2 ∂ ( ρ u ) + c s 2 Δ t ( τ − 2 1 ) ∂ x 1 ∂ { ρ [ ∂ x 1 ∂ u + ( ∂ x 1 ∂ u ) T ] } = 0 ( A - 1 0 )
将(A-10)式和(A-5.2)式联立,得
∂ ( ρ u ) ∂ t + ∂ ( ρ u u ) ∂ x = − ∂ ( ρ c s 2 ) ∂ x + c s 2 Δ t ( τ − 1 2 ) ∂ ∂ x { ρ [ ∂ u ∂ x + ( ∂ u ∂ x ) T ] } + F (A-11) \frac{\partial (\rho u)}{\partial t} + \frac{\partial (\rho u u)}{\partial x} = - \frac{\partial (\rho c_s^2)}{\partial x} + c_s^2 \Delta t (\tau - \frac{1}{2}) \frac{\partial}{\partial x}\{\rho [\frac{\partial u}{\partial x} + (\frac{\partial u}{\partial x})^T ]\} + F \tag{A-11}
∂ t ∂ ( ρ u ) + ∂ x ∂ ( ρ u u ) = − ∂ x ∂ ( ρ c s 2 ) + c s 2 Δ t ( τ − 2 1 ) ∂ x ∂ { ρ [ ∂ x ∂ u + ( ∂ x ∂ u ) T ] } + F ( A - 1 1 )
此时令 p = c s 2 ρ p = c_s^2 \rho p = c s 2 ρ , ν = c s 2 Δ t ( τ − 1 2 ) \nu = c_s^2 \Delta t (\tau - \frac{1}{2}) ν = c s 2 Δ t ( τ − 2 1 ) ,即可还原回Navier-Stokes方程
附录B.Gauss-Hermite积分
对于任意一维函数 f ( ξ ) f(\xi) f ( ξ ) ,高斯积分通过寻找积分点 ξ k \xi_k ξ k 和其对应权重 w k w_k w k 实现对 ∫ a b ω ( ξ ) f ( ξ ) d ξ \int_{a}^{b} \omega(\xi) f(\xi) d \xi ∫ a b ω ( ξ ) f ( ξ ) d ξ 的近似,即
∫ a b ω ( ξ ) f ( ξ ) d ξ ≈ ∑ k = 1 n w k ξ k (B-1) \int_{a}^{b} \omega(\xi) f(\xi) d \xi \approx \sum_{k=1}^{n} w_k \xi_k \tag{B-1}
∫ a b ω ( ξ ) f ( ξ ) d ξ ≈ k = 1 ∑ n w k ξ k ( B - 1 )
其中 ω ( ξ ) \omega (\xi) ω ( ξ ) 是任意的权重函数。n 点高斯求积的积分点正是对应的第n个正交多项式的根,此时 ξ k \xi_k ξ k 对应的权重 w k w_k w k 为
w k = < P n − 1 , P n − 1 > P n − 1 ( ξ k ) P n ′ ( ξ k ) (B-2) w_k = \frac{ <P_{n-1}, P_{n-1}> }{ P_{n-1}(\xi_k) P_{n}^{'}(\xi_k) } \tag{B-2}
w k = P n − 1 ( ξ k ) P n ′ ( ξ k ) < P n − 1 , P n − 1 > ( B - 2 )
其中 P n ′ = d P n d x P_{n}^{'} = \frac{d P_n}{d x} P n ′ = d x d P n 。(B-2)式的精度为 2 n − 1 2n-1 2 n − 1 。
所以在一维Gauss-Hermite积分中,取 P n = H ( n ) P_n = \mathcal{H}^{(n)} P n = H ( n ) , H ( n ) \mathcal{H}^{(n)} H ( n ) 的权函数和表达式见(12)式和(13)式(下文同理)。 n点高斯求积的积分点即为 H ( n ) \mathcal{H}^{(n)} H ( n ) 的根。因为 d H ( n ) d ξ = ξ H ( n ) − H ( n + 1 ) = n H ( n − 1 ) \frac{d \mathcal{H}^{(n)}}{d \xi} = \xi \mathcal{H}^{(n)} - \mathcal{H}^{(n+1)} = n \mathcal{H}^{(n-1)} d ξ d H ( n ) = ξ H ( n ) − H ( n + 1 ) = n H ( n − 1 ) ,所以有
w k = n ! / [ n H ( n − 1 ) ( ξ k ) ] 2 (B-3) w_k = n! / [n \mathcal{H}^{(n-1)} (\xi_k)]^2 \tag{B-3}
w k = n ! / [ n H ( n − 1 ) ( ξ k ) ] 2 ( B - 3 )
对于更高维的情况,可做类似构造。分析如下积分
1 ( 2 π ) D / 2 ∫ e x p ( − ξ ⋅ ξ 2 ) p ( ξ ) d ξ (B-4) \frac{1}{(2 \pi)^{D/2}} \int \mathrm{exp}(- \frac{\xi \cdot \xi}{2}) p(\xi) \mathrm{d} \xi \tag{B-4}
( 2 π ) D / 2 1 ∫ e x p ( − 2 ξ ⋅ ξ ) p ( ξ ) d ξ ( B - 4 )
其中 ξ \xi ξ 是长度为D的向量。 p ( ξ ) p(\xi) p ( ξ ) 是D维n次多项式,可写为(B-5)式
p ( ξ ) = ∑ n 1 + n 2 + . . . + n D ≤ n a n 1 n 2 . . . n D ∏ j = 1 D ξ j n j (B-5) p(\xi) = \sum\limits_{n_1 + n_2 + ... +n_D \le n} {a_{n_1 n_2 ... n_D}} \prod\limits_{j=1}\limits^{D}{\xi_{j}^{n_j}} \tag{B-5}
p ( ξ ) = n 1 + n 2 + . . . + n D ≤ n ∑ a n 1 n 2 . . . n D j = 1 ∏ D ξ j n j ( B - 5 )
其中 a n 1 n 2 . . . n D a_{n_1 n_2 ... n_D} a n 1 n 2 . . . n D 为常数。
考虑(B-5)式中的任意一项 ∏ j = 1 D ξ j n j \prod\limits_{j=1}\limits^{D}{\xi_{j}^{n_j}} j = 1 ∏ D ξ j n j (由于a n 1 n 2 . . . n D a_{n_1 n_2 ... n_D} a n 1 n 2 . . . n D 为常数所以可暂不处理,在最后一步被归并到常数项中),回代到式(B-4)中,得
1 ( 2 π ) D / 2 ∫ e x p ( − ξ ⋅ ξ 2 ) p ( ξ ) d ξ = ∏ j = 1 D ( 1 2 π ∫ e x p ( − ξ j 2 2 ) ξ j n j d ξ j ) = ∏ j = 1 D ( ∑ k = 1 n w k ξ k ) (B-6.1) \frac{1}{(2 \pi)^{D/2}} \int \mathrm{exp}(- \frac{\xi \cdot \xi}{2}) p(\xi) \mathrm{d} \xi = \prod\limits_{j=1}\limits^{D} \left( \frac{1}{\sqrt{2 \pi}} \int \mathrm{exp}(- \frac{\xi_{j}^2}{2}) \xi_j^{n_j} \mathrm{d}\xi_j \right) = \prod\limits_{j=1}\limits^{D} ( \sum\limits_{k=1}\limits^{n} w_k \xi_k ) \tag{B-6.1}
( 2 π ) D / 2 1 ∫ e x p ( − 2 ξ ⋅ ξ ) p ( ξ ) d ξ = j = 1 ∏ D ( 2 π 1 ∫ e x p ( − 2 ξ j 2 ) ξ j n j d ξ j ) = j = 1 ∏ D ( k = 1 ∑ n w k ξ k ) ( B - 6 . 1 )
其中 w k w_k w k 和 ξ k \xi_k ξ k 是一维n次多项式高斯积分的权重和积分点。积分结果被转化为多个一维Gauss-Hermite积分的连乘。注意到(B-6.1)式等价于
1 ( 2 π ) D / 2 ∫ e x p ( − ξ ⋅ ξ 2 ) p ( ξ ) d ξ = ∑ k 1 = 1 n . . . ∑ k D = 1 n [ ( w k 1 w k 2 . . . w k D ) ( ξ k 1 n 1 ξ k 2 n 2 . . . ξ k D n D ) ] (B-6.2) \frac{1}{(2 \pi)^{D/2}} \int \mathrm{exp}(- \frac{\xi \cdot \xi}{2}) p(\xi) \mathrm{d} \xi = \sum\limits_{k_1=1}\limits^{n} ... \sum\limits_{k_D=1}\limits^{n} [(w_{k_1} w_{k_2} ... w_{k_D}) (\xi_{k_1}^{n_1} \xi_{k_2}^{n_2} ... \xi_{k_D}^{n_D})] \tag{B-6.2}
( 2 π ) D / 2 1 ∫ e x p ( − 2 ξ ⋅ ξ ) p ( ξ ) d ξ = k 1 = 1 ∑ n . . . k D = 1 ∑ n [ ( w k 1 w k 2 . . . w k D ) ( ξ k 1 n 1 ξ k 2 n 2 . . . ξ k D n D ) ] ( B - 6 . 2 )
如果定义 ξ k 1 . . . k D = ( ξ k 1 , ξ k 2 , . . . , ξ k D ) \xi_{k_1 ... k_D} = (\xi_{k_1}, \xi_{k_2}, ... , \xi_{k_D}) ξ k 1 . . . k D = ( ξ k 1 , ξ k 2 , . . . , ξ k D ) 和 w k 1 . . . k D = w k 1 w k 2 . . . w k D w_{k_1 ... k_D} = w_{k_1} w_{k_2} ... w_{k_D} w k 1 . . . k D = w k 1 w k 2 . . . w k D ,那么(B-4)式等价为
1 ( 2 π ) D / 2 ∫ e x p ( − ξ ⋅ ξ 2 ) p ( ξ ) d ξ = ∑ w k 1 . . . k D p ( ξ k 1 . . . k D ) (B-7) \frac{1}{(2 \pi)^{D/2}} \int \mathrm{exp}(- \frac{\xi \cdot \xi}{2}) p(\xi) \mathrm{d} \xi = \sum w_{k_1 ... k_D} p(\xi_{k_1 ... k_D}) \tag{B-7}
( 2 π ) D / 2 1 ∫ e x p ( − 2 ξ ⋅ ξ ) p ( ξ ) d ξ = ∑ w k 1 . . . k D p ( ξ k 1 . . . k D ) ( B - 7 )
此时则将(B-4)式转换为与一维Gauss-Hermite积分类似的形式,即(B-7)式。
参考