解线性代数方程组的迭代法
向量序列和矩阵序列的极限
见矩阵论部分
迭代法的基本理论
简单迭代及其收敛性
A x = b ⟺ x = B x + g ⟹ x ( k + 1 ) = B x ( k ) + g \begin{aligned} Ax = b &\iff x = Bx+g \\ &\Longrightarrow x^{(k+1)} = Bx^{(k)} + g \end{aligned} Ax=b⟺x=Bx+g⟹x(k+1)=Bx(k)+g
- 若 { x ( k ) } k = 0 ∞ → x ‾ \{x^{(k)}\}_{k=0}^\infty \rightarrow \overline x {x(k)}k=0∞→x，则 x ‾ \overline x x是方程的解
- ∀ x ( 0 ) , x ( k ) → x ‾ ⟺ ρ ( B ) < 1 \forall x^{(0)},x^{(k)} \rightarrow \overline x \iff \rho(B) <1 ∀x(0),x(k)→x⟺ρ(B)<1
- 收敛速度：
η
=
−
ln
ρ
(
B
)
\eta = -\ln \rho(B)
η=−lnρ(B)
∥ x ( k ) − x ‾ ∥ ≤ ∥ B ∥ 1 − ∥ B ∥ ∥ x ( k ) − x ( k − 1 ) ∥ ∥ x ( k ) − x ‾ ∥ ≤ ∥ B ∥ k 1 − ∥ B ∥ ∥ x ( 1 ) − x ( 0 ) ∥ \begin{aligned} &\parallel x^{(k)} - \overline x \parallel \leq \frac{\parallel B \parallel}{1- \parallel B \parallel } \parallel x^{(k)} - x^{(k-1)} \parallel \\ &\parallel x^{(k)} - \overline x \parallel \leq \frac{\parallel B \parallel^k}{1- \parallel B \parallel } \parallel x^{(1)} - x^{(0)} \parallel \\ \end{aligned} ∥x(k)−x∥≤1−∥B∥∥B∥∥x(k)−x(k−1)∥∥x(k)−x∥≤1−∥B∥∥B∥k∥x(1)−x(0)∥
Gauss-Seidel迭代及其收敛性
x ( k + 1 ) = B x ( k ) + g ⟺ x ( k + 1 ) = ( B 1 + B 2 ) x ( k ) + g ⟹ x ( k + 1 ) = B 1 x ( k + 1 ) + B 2 x ( k ) + g ⟺ x ( k + 1 ) = ( I − B 1 ) − 1 B 2 x ( k ) + ( I − B 1 ) − 1 g \begin{aligned} x^{(k+1)} = B x^{(k)} + g &\iff x^{(k+1)} = (B_1+B_2)x^{(k)} + g \\ &\Longrightarrow x^{(k+1)} = B_1x^{(k+1)} + B_2x^{(k)} + g \\ &\iff x^{(k+1)} = (I-B_1)^{-1}B_2 x^{(k)} + (I-B_1)^{-1}g \end{aligned} x(k+1)=Bx(k)+g⟺x(k+1)=(B1+B2)x(k)+g⟹x(k+1)=B1x(k+1)+B2x(k)+g⟺x(k+1)=(I−B1)−1B2x(k)+(I−B1)−1g
- B 1 B_1 B1是严格下三角， B 2 B_2 B2是上三角
- 计算 x i ( k + 1 ) x_i^{(k+1)} xi(k+1)时，前 i − 1 i-1 i−1个分量用新值，后面的用旧值
- ∥ B ∥ 1 < 1 or ∥ B ∥ ∞ < 1 \parallel B \parallel_1 <1 \text{ or } \parallel B \parallel_\infty <1 ∥B∥1<1 or ∥B∥∞<1， Gauss-Seidel \text{Gauss-Seidel} Gauss-Seidel收敛
几种常用的迭代法
Jacobi迭代
A
x
=
b
,
A
=
L
+
D
+
U
⟺
x
=
−
D
−
1
(
L
+
U
)
x
+
D
−
1
b
⟹
x
(
k
+
1
)
=
B
J
x
(
k
)
+
g
,
B
J
=
−
D
−
1
(
L
+
U
)
,
g
=
D
−
1
b
\begin{aligned} Ax = b,A = L+D+U &\iff x = -D^{-1}(L+U)x+D^{-1}b \\ &\Longrightarrow x^{(k+1)} = B_Jx^{(k)} +g,B_J = -D^{-1}(L+U),g = D^{-1}b \end{aligned}
Ax=b,A=L+D+U⟺x=−D−1(L+U)x+D−1b⟹x(k+1)=BJx(k)+g,BJ=−D−1(L+U),g=D−1b
x
i
(
k
+
1
)
=
1
a
i
i
(
b
i
−
∑
j
=
1
,
j
≠
i
n
a
i
j
x
j
(
k
)
)
x_i^{(k+1)} = \frac{1}{a_{ii}}(b_i-\sum_{j=1,j\neq i}^n a_{ij}x_j^{(k)})
xi(k+1)=aii1(bi−j=1,j=i∑naijxj(k))
收敛的充分条件
- ρ ( B J ) < 1 \rho(B_J)<1 ρ(BJ)<1
- 严格对角占优
基于Jacobi迭代的Gauss-Seidel迭代
x
(
k
+
1
)
=
−
D
−
1
(
L
+
U
)
x
(
k
)
+
D
−
1
b
⟺
x
(
k
+
1
)
=
−
D
−
1
L
x
(
k
)
−
D
−
1
U
x
(
k
)
+
D
−
1
b
⟹
x
(
k
+
1
)
=
−
D
−
1
L
x
(
k
+
1
)
−
D
−
1
U
x
(
k
)
+
D
−
1
b
⟺
x
(
k
+
1
)
=
−
(
L
+
D
)
−
1
U
x
(
k
)
+
(
L
+
D
)
−
1
b
⟹
B
s
=
−
(
L
+
D
)
−
1
U
\begin{aligned} &x^{(k+1)} = - D^{-1}(L+U)x^{(k)} +D^{-1}b \\ \iff &x^{(k+1)} = -D^{-1}Lx^{(k)} - D^{-1}Ux^{(k)} +D^{-1}b \\ \Longrightarrow &x^{(k+1)} = -D^{-1}Lx^{(k+1)} - D^{-1}Ux^{(k)} +D^{-1}b\\ \iff & x^{(k+1)} = -(L+D)^{-1}Ux^{(k)} + (L+D)^{-1}b \\ \Longrightarrow &B_s = -(L+D)^{-1}U \end{aligned}
⟺⟹⟺⟹x(k+1)=−D−1(L+U)x(k)+D−1bx(k+1)=−D−1Lx(k)−D−1Ux(k)+D−1bx(k+1)=−D−1Lx(k+1)−D−1Ux(k)+D−1bx(k+1)=−(L+D)−1Ux(k)+(L+D)−1bBs=−(L+D)−1U
x
i
(
k
+
1
)
=
1
a
i
i
(
b
i
−
∑
j
=
1
i
−
1
a
i
j
x
j
(
k
+
1
)
−
∑
j
=
i
n
a
i
j
x
j
(
k
)
)
x_i^{(k+1)} = \frac{1}{a_{ii}}(b_i-\sum_{j=1}^{i-1} a_{ij}x_j^{(k+1)}-\sum_{j=i}^{n} a_{ij}x_j^{(k)})
xi(k+1)=aii1(bi−j=1∑i−1aijxj(k+1)−j=i∑naijxj(k))
收敛充分条件：
- ρ ( B s ) < 1 \rho(B_s) < 1 ρ(Bs)<1
- A A A严格对角占优
- A A A对称正定
逐次超松弛迭代
SOR(Successive Over-Relaxation)
\text{SOR(Successive Over-Relaxation)}
SOR(Successive Over-Relaxation)
A
x
=
b
,
x
(
k
)
⟺
r
(
k
)
=
b
−
A
x
(
k
)
⟹
x
(
k
+
1
)
=
x
(
k
)
+
w
D
−
1
r
(
k
)
⟺
x
i
(
k
+
1
)
=
x
i
(
k
)
+
w
a
i
i
(
b
i
−
∑
j
=
1
i
−
1
a
i
j
x
j
(
k
)
−
∑
j
=
i
n
a
i
j
x
j
(
k
)
)
⟺
x
i
(
k
+
1
)
=
x
i
(
k
)
+
w
a
i
i
(
b
i
−
∑
j
=
1
i
−
1
a
i
j
x
j
(
k
+
1
)
−
∑
j
=
i
n
a
i
j
x
j
(
k
)
)
⟺
x
i
(
k
+
1
)
=
(
1
−
w
)
x
i
(
k
)
+
w
a
i
i
(
b
i
−
∑
j
=
1
i
−
1
a
i
j
x
j
(
k
)
−
∑
j
=
i
+
1
n
a
i
j
x
j
(
k
)
)
⟺
x
(
k
+
1
)
=
(
1
−
w
)
x
(
k
)
+
w
D
−
1
(
b
−
L
x
(
k
+
1
)
−
U
x
(
k
)
)
⟺
x
(
k
+
1
)
=
B
w
x
(
k
)
+
w
(
D
+
w
L
)
−
1
b
B
w
=
(
D
+
w
L
)
−
1
[
(
1
−
w
)
D
−
w
U
]
\begin{aligned} Ax = b,x^{(k)} &\iff r^{(k)} = b-Ax^{(k)} \\ &\Longrightarrow x^{(k+1)} =x^{(k)} + w D^{-1}r^{(k)} \\ &\iff x_i^{(k+1)} = x_i^{(k)} + \frac{w}{a_{ii}}(b_i-\sum_{j=1}^{i-1}a_{ij}x_j^{(k)} -\sum_{j=i}^na_{ij}x_j^{(k)}) \\ &\iff x_i^{(k+1)} = x_i^{(k)} + \frac{w}{a_{ii}}(b_i-\sum_{j=1}^{i-1}a_{ij}x_j^{(k+1)} -\sum_{j=i}^na_{ij}x_j^{(k)}) \\ &\iff x_i^{(k+1)} = (1-w)x_i^{(k)} + \frac{w}{a_{ii}}(b_i-\sum_{j=1}^{i-1}a_{ij}x_j^{(k)} -\sum_{j=i+1}^na_{ij}x_j^{(k)}) \\ &\iff x^{(k+1)} = (1-w)x^{(k)} +wD^{-1}(b-Lx^{(k+1)} - Ux^{(k)})\\ &\iff x^{(k+1)} = B_wx^{(k)} +w(D+wL)^{-1}b\\ & \ \ \ \ \ \ \ \ \ \ B_w = (D+wL)^{-1}[(1-w)D-wU] \end{aligned}
Ax=b,x(k)⟺r(k)=b−Ax(k)⟹x(k+1)=x(k)+wD−1r(k)⟺xi(k+1)=xi(k)+aiiw(bi−j=1∑i−1aijxj(k)−j=i∑naijxj(k))⟺xi(k+1)=xi(k)+aiiw(bi−j=1∑i−1aijxj(k+1)−j=i∑naijxj(k))⟺xi(k+1)=(1−w)xi(k)+aiiw(bi−j=1∑i−1aijxj(k)−j=i+1∑naijxj(k))⟺x(k+1)=(1−w)x(k)+wD−1(b−Lx(k+1)−Ux(k))⟺x(k+1)=Bwx(k)+w(D+wL)−1b Bw=(D+wL)−1[(1−w)D−wU]
- ∀ x ( 0 ) \forall x^{(0)} ∀x(0)收敛 ⟹ 0 < w < 2 \Longrightarrow 0 <w<2 ⟹0<w<2
- ρ ( B ω ) < 1 \rho(B_\omega) < 1 ρ(Bω)<1
- A A A严格对角占优， 0 < ω ≤ 1 ⟹ ∀ x ( 0 ) 0< \omega \leq 1 \Longrightarrow \forall x^{(0)} 0<ω≤1⟹∀x(0)收敛
- A A A对称正定， 0 < w < 2 ⟹ ∀ x ( 0 ) 0<w<2 \Longrightarrow \forall x^{(0)} 0<w<2⟹∀x(0)收敛
函数插值
插值问题与插值多项式
插值问题
已知被插函数
f
(
x
)
f(x)
f(x)在插值区间
[
a
,
b
]
[a,b]
[a,b]内
n
+
1
n+1
n+1 个插值节点
{
x
i
}
i
=
0
n
\{x_i\}_{i=0}^n
{xi}i=0n处的函数值
{
f
(
x
i
)
}
i
=
0
n
\{f(x_i)\}_{i=0}^n
{f(xi)}i=0n。如果
φ
(
x
)
\varphi(x)
φ(x)满足以下插值条件
∀
x
i
,
φ
(
x
i
)
=
f
(
x
i
)
\forall x_i ,\varphi(x_i) = f(x_i)
∀xi,φ(xi)=f(xi)
则称其为插值函数，这类问题叫做插值问题。
R
n
(
x
)
=
f
(
x
)
−
φ
(
x
)
R_n(x) = f(x) -\varphi(x)
Rn(x)=f(x)−φ(x)
被称为截断误差/插值余项。
插值多项式
取插值函数集合
Φ
\Phi
Φ为不超过
n
n
n次的多项式空间
P
n
=
s
p
a
n
{
1
,
x
,
x
2
,
⋯
,
x
n
}
P^n=span \{1,x,x^2,\cdots,x^n\}
Pn=span{1,x,x2,⋯,xn}，
{
1
,
x
,
x
2
,
⋯
,
x
n
}
\{1,x,x^2,\cdots,x^n\}
{1,x,x2,⋯,xn}是最基本的一种插值基函数。不同插值方法对应着不同的插值基函数，插值基函数对应的不同的权值。
满足插值条件的不超过
n
n
n次的插值多项式存在且唯一。
R
n
(
x
)
=
f
(
x
)
−
φ
(
x
)
=
f
(
n
+
1
)
(
ξ
)
(
n
+
1
)
!
ω
n
+
1
(
x
)
≤
M
n
+
1
(
n
+
1
)
!
∣
ω
n
+
1
(
x
)
∣
(插值余项)
R_n(x) = f(x) - \varphi(x) = \frac{f^{(n+1)}(\xi)}{(n+1)!}\omega_{n+1}(x) \leq \frac{M_{n+1}}{(n+1)!} \mid \omega_{n+1}(x) \mid \tag{插值余项}
Rn(x)=f(x)−φ(x)=(n+1)!f(n+1)(ξ)ωn+1(x)≤(n+1)!Mn+1∣ωn+1(x)∣(插值余项)
其中
ω
n
+
1
(
x
)
=
∏
i
=
0
n
(
x
−
x
i
)
\omega_{n+1}(x) = \prod_{i=0}^n (x-x_i)
ωn+1(x)=i=0∏n(x−xi)
Lagrange插值
l i ( x i ) = 1 , l i ( x j ) = 0 ⟹ l i ( x ) = ω n + 1 ( x ) ( x − x i ) ω n + 1 ′ ( x i ) ⟹ L n ( x ) = ∑ i = 0 n f ( x i ) l i ( x ) \begin{aligned} l_i(x_i)=1,l_i(x_j) =0 &\Longrightarrow l_i(x) = \frac{\omega_{n+1}(x)}{(x-x_i)\omega^\prime_{n+1}(x_i)}\\ &\Longrightarrow L_n(x) = \sum_{i=0}^nf(x_i)l_i(x) \\ \end{aligned} li(xi)=1,li(xj)=0⟹li(x)=(x−xi)ωn+1′(xi)ωn+1(x)⟹Ln(x)=i=0∑nf(xi)li(x)
- ∑ i = 0 n l i ( x ) = 1 \sum_{i=0}^nl_i(x) = 1 ∑i=0nli(x)=1
- n + 1 n+1 n+1个 L a g r a n g e \rm Lagrange Lagrange插值基函数 l i ( x ) l_i(x) li(x)线性无关，构成线性空间上的一组基。
Newton插值
差商及其性质
-
零阶差商
f [ x k ] = f ( x k ) f[x_k] = f(x_k) f[xk]=f(xk) -
一阶差商
f [ x j , x k ] = f ( x k ) − f ( x j ) x k − x j f[x_j,x_k] = \frac{f(x_k)-f(x_j)}{x_k-x_j} f[xj,xk]=xk−xjf(xk)−f(xj) -
二阶差商
f [ x i , x j , x k ] = f [ x j , x k ] − f [ x i , x j ] x k − x i f[x_i,x_j,x_k] = \frac{f[x_j,x_k] - f[x_i,x_j]}{x_k-x_i} f[xi,xj,xk]=xk−xif[xj,xk]−f[xi,xj] -
k k k阶差商
f [ x 0 , x 1 , ⋯ , x k ] = f [ x 1 , x 2 , ⋯ , x k ] − f [ x 0 , x 1 , ⋯ , x k − 1 ] x k − x 0 f[x_0,x_1,\cdots,x_k] = \frac{f[x_1,x_2,\cdots,x_k] - f[x_0,x_1,\cdots,x_{k-1}]}{x_k-x_0} f[x0,x1,⋯,xk]=xk−x0f[x1,x2,⋯,xk]−f[x0,x1,⋯,xk−1]
差商的性质
-
差商的函数值表示
f [ x 0 , x 1 , ⋯ , x k ] = ∑ i = 0 k f ( x i ) ω k + 1 ( x i ) f[x_0,x_1,\cdots,x_k] = \sum_{i=0}^k \frac{f(x_i)}{\omega_{k+1}(x_i)} f[x0,x1,⋯,xk]=i=0∑kωk+1(xi)f(xi) -
差商的对称性
f [ x 0 , x 1 , ⋯ , x k ] = f [ x i 0 , x i 1 , ⋯ , x i k ] f[x_0,x_1,\cdots,x_k] = f[x_{i0},x_{i1},\cdots,x_{ik}] f[x0,x1,⋯,xk]=f[xi0,xi1,⋯,xik] -
差商与导数的关系
f [ x 0 , x 1 , ⋯ , x n ] = f ( n ) ( η ) n ! f[x_0,x_1,\cdots,x_n] = \frac{f^{(n)}(\eta)}{n!} f[x0,x1,⋯,xn]=n!f(n)(η)
差商表
Newton插值公式
N n ( x ) = f ( x 0 ) + f [ x 0 , x 1 ] ω 1 ( x ) + f [ x 0 , x 1 , x 2 ] ω 2 ( x ) + ⋯ + f [ x 0 , x 1 , ⋯ , x n ] ω n ( x ) N_n(x) = f(x_0)+f[x_0,x_1]\omega_1(x) + f[x_0,x_1,x_2]\omega_2(x) + \cdots +f[x_0,x_1,\cdots,x_n]\omega_n(x) Nn(x)=f(x0)+f[x0,x1]ω1(x)+f[x0,x1,x2]ω2(x)+⋯+f[x0,x1,⋯,xn]ωn(x)
- N n + 1 ( x ) = N n ( x ) + f [ x 0 , x 1 , ⋯ , x n , x n + 1 ] ω n + 1 ( x ) N_{n+1}(x) = N_n(x)+ f[x_0,x_1,\cdots,x_n,x_{n+1}]\omega_{n+1}(x) Nn+1(x)=Nn(x)+f[x0,x1,⋯,xn,xn+1]ωn+1(x)
- R n ( x ) = f ( x ) − N n ( x ) = f [ x , x 0 , x 1 , ⋯ , x n ] w n + 1 ( x ) = f ( n + 1 ) ( ξ ) ( n + 1 ) ! w n + 1 ( x ) R_n(x) = f(x)-N_n(x) = f[x,x_0,x_1,\cdots,x_n]w_{n+1}(x) = \cfrac{f^{(n+1)}(\xi)}{(n+1)!} w_{n+1}(x) Rn(x)=f(x)−Nn(x)=f[x,x0,x1,⋯,xn]wn+1(x)=(n+1)!f(n+1)(ξ)wn+1(x)
等距节点插值
差分算子及其性质
y = f ( x ) , x i = a + i h , f ( x i ) , h = b − a n y = f(x),x_i = a+ih,f(x_i) ,h = \frac{b-a}{n} y=f(x),xi=a+ih,f(xi),h=nb−a
- 向前差分算子 Δ \Delta Δ
{ Δ f ( x ) = f ( x + h ) − f ( x ) Δ m f ( x ) = Δ ( Δ m − 1 f ( x ) ) \begin{cases} \Delta f(x) = f(x+h) - f(x) \\ \Delta^m f(x) = \Delta(\Delta ^{m-1}f(x)) \end{cases} {Δf(x)=f(x+h)−f(x)Δmf(x)=Δ(Δm−1f(x))
-
向后差分算子 ∇ \nabla ∇
{ ∇ f ( x ) = f ( x ) − f ( x − h ) ∇ m f ( x ) = ∇ ( ∇ m − 1 f ( x ) ) \begin{cases} \nabla f(x) = f(x) - f(x-h) \\ \nabla^mf(x) = \nabla(\nabla^{m-1}f(x)) \end{cases} {∇f(x)=f(x)−f(x−h)∇mf(x)=∇(∇m−1f(x)) -
恒等算子 I I I
{ I f ( x ) = f ( x ) I m f ( x ) = I ( I m − 1 f ( x ) ) \begin{cases} If(x) = f(x) \\ I^mf(x) = I(I^{m-1}f(x)) \end{cases} {If(x)=f(x)Imf(x)=I(Im−1f(x)) -
移位算子 E E E
{ E f ( x ) = f ( x + h ) E s f ( x ) = f ( x + s h ) \begin{cases} Ef(x) = f(x+h) \\ E^s f(x) = f(x+sh) \end{cases} {Ef(x)=f(x+h)Esf(x)=f(x+sh)
向前差分表
等距节点插值公式
-
Newton \text{Newton} Newton向前插值公式
N n ( a + t h ) = f ( x 0 ) + ∑ k = 1 n Δ k f ( x 0 ) [ t k ] N_n(a+th) = f(x_0) + \sum_{k=1}^n \Delta^kf(x_0) \begin{bmatrix} t \\ k\end{bmatrix} Nn(a+th)=f(x0)+k=1∑nΔkf(x0)[tk] -
Newton \text{Newton} Newton向后插值公式
N n ( x ) = f ( x n ) + ∑ k = 1 n ∇ k f ( x n ) [ t + k − 1 k ] N_n(x) = f(x_n)+ \sum_{k=1}^n \nabla^kf(x_n) \begin{bmatrix} t+k-1 \\ k \end{bmatrix} Nn(x)=f(xn)+k=1∑n∇kf(xn)[t+k−1k]
Hermite插值
插值条件中含导数的代数插值问题
- 构建重节点差商表【有导数节点处扩充】
- 先求出只满足 f ( x i ) f(x_i) f(xi)的 H ( x ) H(x) H(x)，再通过待定系数法求解
- 存在且唯一
R 2 n + 1 ( x ) = f ( 2 n + 2 ) ( ξ ) ( 2 n + 2 ) ! w n + 1 2 ( x ) R_{2n+1}(x) =\frac{f^{(2n+2)}(\xi)}{(2n+2)!}w^2_{n+1}(x) R2n+1(x)=(2n+2)!f(2n+2)(ξ)wn+12(x)
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
