上节课主要介绍了非线性方程的几种数值解法,其中包括交叉法(二分法、线性插值法)和开放法(牛顿法、割线法、固定点法)。本节课主要介绍线性方程组的数值求解方法,主要分为直接法和迭代法两类。直接法包括高斯消去法(Gauss elimination)、高斯约当法(Gauss-Jordan)以及LU分解法,迭代法包括Jacobi法和高斯塞德尔法(Gauss-Seidel) 。
1. 高斯消去法(Gauss Elimination Method)
高斯消去法是通过将线性方程组转化为上三角的形式,然后一步一步回代(back substitution)求解的方法。
例:\[ \begin{cases}x_1+2x_2=5 \\ 3x_1+4x_2=6\end{cases} \]
- 系统法:\[ \begin{cases}x_1+2x_2=5 \\ 3x_1+4x_2=6\end{cases} \rightarrow \begin{cases}x_1+2x_2=5 \\ 0-2x_2=-9\end{cases} \rightarrow \begin{cases}x_1+2x_2=5 \\ x_2=4.5\end{cases} \rightarrow \begin{cases}x_1=-4 \\ x_2=4.5\end{cases}\]
- 系统法的增广矩阵形式:将方程写为矩阵的形式 \[\begin{bmatrix}1 & 2 \\3 & 4\end{bmatrix} \begin{bmatrix}x_1 \\ x_2 \end{bmatrix}=\begin{bmatrix}5\\6\end{bmatrix}\] 系统法的求解过程可以写成如下增广矩阵的变换形式 \[\begin{bmatrix}1 & 2 & 5 \\3 & 4 & 6\end{bmatrix} \rightarrow \begin{bmatrix}1 & 2 & 5 \\0 & -2 & -9\end{bmatrix} \rightarrow \begin{bmatrix}1 & 2 & 5 \\0 & 1 & 4.5\end{bmatrix} \rightarrow \begin{bmatrix}1 & 0 & -4 \\0 & 1 & 4.5\end{bmatrix} \]
例:
\[\begin{cases}2x_1+x_2-x_3=1\\x_1+2x_2+x_3=8\\-x_1+x_2-x_3=-5\end{cases}\]
matrix:
\[\begin{bmatrix}2 & 1 & -1 & 1\\ 1 &2 & 1 & 8\\ -1 & 1 & -1 & -5 \end{bmatrix} \rightarrow \begin{bmatrix}2 & 1 & -1 & 1\\ 0 &1.5 & 1.5 & 7.5\\ 0 & 1.5 & -1.5 & -4.5 \end{bmatrix}\]
\[\rightarrow \begin{bmatrix}2 & 1 & -1 & 1\\ 0 &1.5 & 1.5 & 7.5\\ 0 & 0 & -3 & -12\end{bmatrix} \rightarrow \begin{cases}-3x_2 = -12\\x_3=4\end{cases} \rightarrow \begin{cases}x_1 =2 \\x_2=1\\x_3=4\end{cases}\]
类似的,\(n \times n\) 形式的方程组: \[ \begin{cases}a_{11}x_1+a_{12}x_2+\cdots+a_{1n}x_n=b_1 \\ a_{21}x_1+a_{22}x_2+\cdots+a_{2n}x_n=b_2\\ \vdots\\ a_{n1}x_1+a_{n2}x_2+\cdots+a_{nn}x_n=b_n\\ \end{cases}\]
可以写作如下矩阵形式:\[A=\begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1n}\\ a_{21} & a_{22} & \cdots & a_{2n}\\ \vdots\\ a_{n1} & a_{n2} & \cdots & a_{nn} \end{bmatrix}, x=\begin{bmatrix}x_1 \cdots x_n\end{bmatrix}, b=\begin{bmatrix}b_1 \cdots b_n\end{bmatrix}, Ax=b\]
高斯消去法的核心:
- 方程组写为矩阵形式
- 矩阵行变换:将一行中所有元素乘上一个标量,减去(加上)另外一行,该操作不会改变方程的解
- 逐行变换,直至将矩阵转换为便于求解的形式
一般而言,如下三种矩阵形式方便求解:
- 对角形式: \(\begin{bmatrix}1 & & 0\\ & \ddots & \\0 & & 1 \end{bmatrix}\), \(x_n=b_n\)
- 上三角形式: \(\begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1n}\\ & a_{22} & \cdots & a_{2n}\\ & & \ddots&\vdots\\ & & & a_{nn}\\ \end{bmatrix}\),使用回代法求解(back substitution): \[a_{nn}x_n=b_n \rightarrow x_n = \frac{b_n}{a_{nn}}\] \[a_{n-1,n-1}x_{n-1}+ a_{n-1,n}x_n=b_{n-1}\rightarrow x_{n-1}=\frac{b_{n-1}-a_{n-1,n}x_n}{a_{n-1,n-1}} \] 重复下去 \[x_i=\frac{b_i-\sum^n_{j=i+1}a_{ij}x_j}{a_{ii}}\]
- 下三角形式 :\(\begin{bmatrix} a_{11} & & &\\ a_{21} & a_{22} & &\\ \vdots & \vdots & \ddots&\\ a_{n1}& a_{n2}& \cdots & a_{nn}\\ \end{bmatrix}\) 使用前向替换法(forward substitution): \[a_{11}x_1=b_1 \rightarrow x_1 = \frac{b_1}{a_{11}}\] \[a_{2,1}x_{1}+ a_{2,2}x_2=b_{2}\rightarrow x_{2}=\frac{b_{2}-a_{2,1}x_1}{a_{2,2}}\] 重复下去 \[x_i=\frac{b_i-\sum^{i-1}_{j=1}a_{ij}x_j}{a_{ii}}\]
Matlab实现(求解\(Ax=b\), 使用左除\(x=A\backslash b\)):
>> A = [1 2;3 4]
A =
1 2
3 4
>> b = [5 6]'
b =
5
6
>> x = A\b
x =
-4.0000
4.5000
>> format long
>> x = A\b
x =
-3.999999999999999
4.499999999999999
MATLAB实现(高斯消去法):
function x=Gauss(a,b)
ab=[a,b];
[R,C]=size(ab);
for j=1:R-1
for i=j+1:R
ab(i,j:C)=ab(i,j:C)-ab(i,j)/ab(j,j)*ab(j,j:C);
end
end
x=zeros(R,1);
x(R)=ab(R,C)/ab(R,R);
for i=R-1:-1:1
x(i)=(ab(i,C)-ab(i,i+1:R)*x(i+1:R))/ab(i,i);
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>> A = [1 2;3 4]
A =
1 2
3 4
>> b = [5 6]'
b =
5
6
>> format long
>> Gauss(A,b)
ans =
-4.000000000000000
4.500000000000000
MATLAB实现(回代法与前向替换法)
function y = BackwardSub(a,b)
% The function solves a system of linear equations ax=b
% where a is upper triangular by using backward substitution.
% Input variables:
% a The matrix of coefficients.
% b A column vector of constants.
% Output variable:
% y A colum vector with the solution.
n = length(b);
y(n,1) = b(n)/a(n,n);
for i = n-1:-1:1
y(i,1)=(b(i)-a(i,i+1:n)*y(i+1:n,1))./a(i,i);
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function y = ForwardSub(a,b)
% The function solves a system of linear equations ax=b
% where a is lower triangular by using forward substitution.
% Input variables:
% a The matrix of coefficients.
% b A column vector of constants.
% Output variable:
% y A colum vector with the solution.
n = length(b);
y(1,1) = b(1)/a(1,1);
for i = 2:n
y(i,1)=(b(i)-a(i,1:i-1)*y(1:i-1,1))./a(i,i);
end
2. 高斯主元消去(Gauss Pivoting)
可以发现,当主元为\(0\)或非常小时,使用高斯消去法会失去精度,因此,在必要时可以交换行的顺序: \[\begin{bmatrix} 0 & 2 & 5\\ 3 & 4& 5 \end{bmatrix}, \] 高斯主元法就是通过交换行的顺序,来保证主元不为0,并且尽可能取绝对值较大的值,下面通过一个案例,说明主元很小时,高斯消去法会造成计算精度上出现较大误差,而高斯主元法的计算结果会更为精确。
例: \[ \begin{cases}0.0003x_1 + 12.34x_2=12.343 \\0.4321 x_1+x_2=5.321\end{cases}\] 使用高斯消去:
- \(\frac{0.4321}{0.0003}=1440\) (round-off error, since \(0.0003*1440 = 0.4320\))
- \[ \begin{cases}0.0003x_1 + 12.34x_2=12.343 \\ 0.0001 x_1+17770x_2=-17760\end{cases} \rightarrow x_2 = -\frac{17760}{17770}=0.9994 \rightarrow x_1 = \frac{12.343-12.34*0.9994}{0.0003}=33.33\]
MATLAB运算结果:
>> A = [0.0003 12.34;0.4321 1]
A =
0.000300000000000 12.340000000000000
0.432100000000000 1.000000000000000
>> b = [12.343 5.321]'
b =
12.343000000000000
5.321000000000000
>> A\b
ans =
10.000000000000000
1.000000000000000
若交换行序(高斯主元消去):
- \(\frac{0.0003}{0.4321}=0.0006943\)(round-off error, since \(0.4321*0.0006943 = 0.0003\))
- \[ \begin{cases}0.4321 x_1+x_2=5.321 \\ 0x_1 + 12.34x_2=12.34\end{cases} \rightarrow x_2 = -\frac{12.34}{12.34}=1\rightarrow x_1 = \frac{5.321-1}{0.4321}=10\]
MATLAB实现(高斯主元消去)
function x=GaussPivot(a,b)
ab=[a,b];
[R,C]=size(ab);
for j=1:R-1
if ab(j,j)==0
for k=j+1:R
if ab(k,j)~=0
abTemp=ab(j,:);
ab(j,:)=ab(k,:);
ab(k,:)=abTemp;
break
end
end
end
for i=j+1:R
ab(i,j:C)=ab(i,j:C)-ab(i,j)/ab(j,j)*ab(j,j:C);
end
end
x=zeros(R,1);
x(R)=ab(R,C)/ab(R,R);
for i=R-1:-1:1
x(i)=(ab(i,C)-ab(i,i+1:R)*x(i+1:R))/ab(i,i);
end
end
3. 高斯约当法(Gauss-Jordan Method, 同时适用于求矩阵的逆)
高斯约当法一般采用高斯主元消去法,不同的是,高斯约当法每次会将主元规范化为\(1\),且最终将原矩阵转化为单位矩阵。
例:\[\begin{bmatrix}4 & 1 & 2 & 21\\ 2 & -2 & 2 & 8\\ 1 & -2 & 4 & 16\end{bmatrix} \rightarrow \begin{bmatrix}1 & -2 & 4 & 16\\ 2 & -2 & 2 & 8 \\4 & 1 & 2 & 21 \end{bmatrix} \rightarrow \begin{bmatrix}1 & -2 & 4 & 16\\ 0 & 2 & -6 & -24 \\0 & 9 & -14 & -43 \end{bmatrix}\] \[ \rightarrow \begin{bmatrix}1 & -2 & 4 & 16\\ 0 & 1 & -3 & -12 \\0 & 9 & -14 & -43 \end{bmatrix} \rightarrow \begin{bmatrix}1 & 0 & -2 & -8\\ 0 & 1 & -3 & -12 \\0 & 0 & 13 & 65 \end{bmatrix} \rightarrow \begin{bmatrix}1 & 0 & -2 & -8\\ 0 & 1 & -3 & -12 \\0 & 0 & 1 & 5 \end{bmatrix}\] \[ \rightarrow \begin{bmatrix}1 & 0 & 0 & 2\\ 0 & 1 & 0 & 3 \\0 & 0 & 1 & 5 \end{bmatrix}\]
可以看到,高斯约当法可以很方便地用来求解一个矩阵的逆:\(AB=I\), \(A\begin{bmatrix}b_1 & b_2 & b_3 \end{bmatrix} \rightarrow \begin{bmatrix}a_{11} & a_{12} & a_{13} & 1 & 0 & 0\\ a_{21} & a_{22} & a_{23} & 0 & 1 & 0\\ a_{31} & a_{32} & a_{33} & 0 & 0 & 1\end{bmatrix} \rightarrow \begin{bmatrix}1 & 0 & 0 &c_{11} & c_{12} & c_{13} & \\ 0 & 1 & 0 & c_{21} & c_{22} & c_{23} &\\ 0 & 0 & 1& c_{31} & c_{32} & c_{33} \end{bmatrix}\)
4. LU分解法(LU-decomposition Method)
LU分解是将一个矩阵分解为一个下三角矩阵和上三角矩阵的乘积:\(A=LU\), 其中\(L\) 表示下三角矩阵, \(U\)表示上三角矩阵。LU分解之后,求解线性方程\(Ax=b\) 就等价于求解如下两个方程:\[Ux=y,Ly=b\]
LU分解主要有两种方法来实现:
- 高斯消去(Gauss Elimination);
- Crout's Method
4.1 Crout's Method
\[\begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1n}\\ a_{21} & a_{22} & \cdots & a_{2n}\\ \vdots\\ a_{n1} & a_{n2} & \cdots & a_{nn} \end{bmatrix}=\begin{bmatrix}L_{11} & & & & \\ L_{21} & L_{22} & & & \\ \vdots & \vdots & \ddots & \\ L_{n1} & L_{n2} & \cdots & L_{nn} \end{bmatrix} \begin{bmatrix}1 & U_{12} & U_{13} & \cdots & U_{1n} \\ & 1 & U_{23} & \cdots &U_{2n}\\ & & \ddots & & \vdots \\ & & & & 1\end{bmatrix} \]
例:
\[\begin{bmatrix} a_{11} & a_{12} & a_{13}\\ a_{21} & a_{22} & a_{23}\\ a_{31} & a_{32} & a_{33} \end{bmatrix}=\begin{bmatrix} L_{11} & 0 & 0\\ L_{21} & L_{22} & 0\\ L_{31} & L_{32} & L_{33} \end{bmatrix}\begin{bmatrix} 1 & U_{12} & U_{13}\\ 0 &1 & U_{23}\\ 0 & 0 & 1 \end{bmatrix}=\begin{bmatrix} L_{11} & L_{11}U_{12} & L_{11}U_{13}\\ L_{21} & L_{21}U_{12}+L_{22} & L_{21}U_{13}+ L_{22}U_{23}\\ L_{31} & L_{31}U_{12}+L_{32} & L_{33}+ L_{31}U_{13}+ L_{32}U_{23} \end{bmatrix}\]
采用待定系数法求解:\[L_{11}=a_{11},L_{21}=a_{21},L_{31}=a_{31}\] \[U_{12} = \frac{a_{12}}{L_{11}}, U_{13}=\frac{a_{13}}{L_{11}}\] \[L_{22}=a_{22}-L_{21}U_{12},L_{32}=a_{32}-L_{31}U_{12}\] \[U_{23}=\frac{a_{23}-L_{21}U_{13}}{L_{22}}\] \[L_{33}=a_{33}-L_{31}U_{13}-L_{32}U_{23}\]
结合计算过程,可以推导出,对于\(n\times n\)矩阵,使用Crout's Method进行LU分解的通用形式为:
- \(L\)矩阵的第一列 \[L_{i1}=a_{i1},i=1,2,...,n\]
- \(U\)矩阵的对角元素 \[U_{ii}=1,i=1,2,...,n\]
- \(U\)矩阵的第\(1\)行 \[U_{1j} = \frac{a_{1j}}{L_{11}},j=2,...,n\]
- \(L\)矩阵的其他元素 \[L_{ij}=a_{ij}-\sum^{j-1}_{k=1}L_{ik}U_{kj},i=2,3,...,n,j=2,3,...,i\]
- \(U\)矩阵的其他元素 \[U_{ij} = \frac{a_{ij}-\sum^{j-1}_{k=1}L_{ik}U_{kj}}{L_{ii}},i=2,3,...,n,j=i+1,i+2,...,n\]
MATLAB实现(LU分解,Crout's Method)
function [L, U] = LUdecompCrout(A)
% The function decomposes the matrix A into a lower triangular matrix L
% and an upper triangular matrix U, using Crout's method such that A=LU.
% Input variables:
% A The matrix of coefficients.
% Output variable:
% L Lower triangular matrix.
% U Upper triangular matrix.
[R, C] = size(A);
for i = 1:R
L(i,1) = A(i,1);
U(i,i) = 1;
end
for j = 2:R
U(1,j)= A(1,j)/L(1,1);
end
for i = 2:R
for j = 2:i
L(i,j)=A(i,j)-L(i,1:j-1)*U(1:j-1,j);
end
for j = i+1:R
U(i,j)=(A(i,j)-L(i,1:i-1)*U(1:i-1,j))/L(i,i);
end
end
4.2 高斯消去
\[\begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1n}\\ a_{21} & a_{22} & \cdots & a_{2n}\\ \vdots\\ a_{n1} & a_{n2} & \cdots & a_{nn} \end{bmatrix}=\begin{bmatrix}1 & & & & \\ m_{21} & 1 & & & \\ m_{31} & m_{32} & 1 & &\\ \vdots & \vdots & \vdots & \ddots & \\ m_{n1} & m_{n2} & \cdots & m_{n,n-1} & 1\end{bmatrix} \begin{bmatrix}a'_{11} & a'_{12} & \cdots & a'_{1n} \\ & a'_{22} & \cdots & a'_{2n}\\ & & \ddots & \vdots\\ & & & a'_{nn}\end{bmatrix}\]
general formulation: \(m_{ij}=\frac{a'_{ij}}{a'_{jj}}\)
例: \(\begin{bmatrix}1 & 2\\3 & 4\end{bmatrix}=\begin{bmatrix}1 & 0\\3 & 1\end{bmatrix}\begin{bmatrix}1 & 2\\0 & -2\end{bmatrix}\)
高斯消去法进行LU分解也可以通过待定系数法求解出矩阵参数:
- \(U\)矩阵第一行:\[U_{1j}=a_{1j},j=1,2,...,n\]
- \(L\)矩阵对角元素:\[L_{ii}=1,i=1,2,...,n\]
- \(L\)矩阵第一列:\[L_{i1}=\frac{a_{i1}}{U_{11}}\]
- \(U\)矩阵其他元素:\[U_{ij}=a_{ij}-\sum^{i-1}_{k=1}L_{ik}U_{kj},j=2,...,n,i=2,...,j\]
- \(L\)矩阵其他元素:\[L_{ij}=\frac{a_{i,j}-\sum^{j-1}_{k=1}L_{ik}U_{kj}}{L_{jj}},j=2,...,n,i=j+1,...,n\]
MATLAB实现(LU分解,高斯法,待定系数)
function [L,U] = LUdecopGauss(A)
[R,C] = size(A);
for i = 1:C
L(i,i) = 1;
U(1,i) = A(1,i);
end
for i = 2:R
L(i,1) = A(i,1)/U(1,1);
end
for j = 2:C
for i = 2:j
U(i,j) = A(i,j)-L(i,1:i-1)*U(1:i-1,j);
end
for i= (j+1):R
L(i,j) = (A(i,j)-L(i,1:j-1)*U(1:j-1,j))/U(j,j);
end
end
end
高斯消去的待定系数法与Crout's 方法的待定系数法相比,要难以理解,可读性差,主要在于通解形式与常规的求解顺序不一致。为便于理解,还有一种拉格朗日形式来实现高斯消去法:
回顾高斯消去的过程:
- 第一步运算 \[A = \begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1n}\\ a_{21} & a_{22} & \cdots & a_{2n}\\ \vdots\\ a_{n1} & a_{n2} & \cdots & a_{nn} \end{bmatrix}\rightarrow A'=\begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1n}\\ 0 & a'_{22} & \cdots & a'_{2n}\\ \vdots\\ 0 & a'_{n2} & \cdots & a'_{nn} \end{bmatrix}=L_1A,L_1=\begin{bmatrix} 1 & 0 & \cdots & 0\\ -\frac{a_{21}}{a_{11}} & 1 & \cdots & 0\\ \vdots\\ -\frac{a_{n1}}{a_{11}} & 0 & \cdots & 1 \end{bmatrix}\] 这是一步行变换
- 第二步运算 \[A'=\begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1n}\\ 0 & a'_{22} & \cdots & a'_{2n}\\ \vdots\\ 0 & a'_{n2} & \cdots & a'_{nn} \end{bmatrix}\rightarrow A''=\begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1n}\\ 0 & a'_{22} & \cdots & a'_{2n}\\ \vdots\\ 0 & 0 & \cdots & a''_{nn} \end{bmatrix}=L_2A‘,L_2=\begin{bmatrix} 1 & 0 & \cdots & 0\\ 0 & 1 & \cdots & 0\\ \vdots\\ 0 & -\frac{a'_{n2}}{a'_{22}} & \cdots & 1 \end{bmatrix}\]
- 第\(n-1\)步运算后 \[\begin{bmatrix}U_{11} & U_{12} & \cdots & U_{1n} \\ & U_{22} & \cdots & U_{2n}\\ & & \ddots & \vdots\\ & & & U_{nn}\end{bmatrix}=L_{n-1}L_{n-2}...L_1A\] 此时即为LU分解中的U矩阵,而\(L=L^{-1}_{1}...L^{-1}_{n-2}L^{-1}_{n-1}=\begin{bmatrix} 1 & 0 & \cdots & 0\\ \frac{a_{21}}{a_{11}} & 1 & \cdots & 0\\ \vdots\\ \frac{a_{n1}}{a_{11}} & 0 & \cdots & 1 \end{bmatrix}\begin{bmatrix} 1 & 0 & \cdots & 0\\ 0 & 1 & \cdots & 0\\ \vdots\\ 0 & \frac{a'_{n2}}{a'_{22}} & \cdots & 1 \end{bmatrix}...=\begin{bmatrix} 1 & 0 & \cdots & 0\\ \frac{a_{21}}{a_{11}} & 1 & \cdots & 0\\ \vdots\\ \frac{a_{n1}}{a_{11}} & \frac{a'_{n2}}{a'_{22}} & \cdots & 1 \end{bmatrix}\)可以看到 \(L_{ij}=\frac{a_{ij}}{a'_{jj}}\)
MATLAB实现(LU分解,高斯消去,拉格朗日形式)
function [L,U]=LUGauss2(A)
[R,C]=size(A);
L = zeros(size(A));
for i = 1:C
L(i,i) = 1;
end
for j=1:R-1
for i=j+1:R
L(i,j) = A(i,j)/A(j,j);
A(i,j:C)=A(i,j:C)-A(i,j)/A(j,j)*A(j,j:C);
end
end
U = A;
end
5. 总结
本节课主要介绍了求解线性方程组的几种直接法,包括高斯消去、高斯主元消去、高斯约当、LU分解法等,其中高斯约当法可以用来求矩阵的逆。 矩阵的LU分解又可以通过高斯消去和Crout 方法来实现,对参数矩阵进行LU分解后,很容易可以求解方程。下节课介绍求解线性方程组的迭代法。