∫ a b f ( x ) g ( x ) d x = 0 \int^b_af(x)g(x)dx=0 ∫abf(x)g(x)dx=0表示在[a,b]区间, f ( x ) f(x) f(x)和 g ( x ) g(x) g(x)正交
函数空间的标准正交基是自乘为1,它乘为0的一组函数
最小二乘法通用求解法
目标函数是采样和预测差异最小化,对每个未知参数求偏导,令结果等于0构建方程组,通过解方程组得到每个参数
求目标函数 Φ ( x ) \Phi(x) Φ(x)可以拆成基函数 ϕ i ( x ) \phi_i(x) ϕi(x)的线性组合,可以理解成函数空间中的一个点
它求偏导后,可以构成这样的方程组,
ρ
(
x
i
)
\rho(x_i)
ρ(xi)是权重
a
1
∑
ϕ
1
(
x
i
)
ϕ
1
(
x
i
)
ρ
(
x
i
)
+
ϕ
1
(
x
i
)
ϕ
2
(
x
i
)
ρ
(
x
i
)
+
.
.
.
=
∑
f
(
x
i
)
ϕ
1
(
x
i
)
ρ
(
x
i
)
a_1\sum\phi_1(x_i)\phi_1(x_i)\rho(x_i)+\phi_1(x_i)\phi_2(x_i)\rho(x_i)+...=\sum f(x_i)\phi_1(x_i)\rho(x_i)
a1∑ϕ1(xi)ϕ1(xi)ρ(xi)+ϕ1(xi)ϕ2(xi)ρ(xi)+...=∑f(xi)ϕ1(xi)ρ(xi)
a
1
∑
ϕ
2
(
x
i
)
ϕ
1
(
x
i
)
ρ
(
x
i
)
+
ϕ
2
(
x
i
)
ϕ
2
(
x
i
)
ρ
(
x
i
)
+
.
.
.
=
∑
f
(
x
i
)
ϕ
2
(
x
i
)
ρ
(
x
i
)
a_1\sum\phi_2(x_i)\phi_1(x_i)\rho(x_i)+\phi_2(x_i)\phi_2(x_i)\rho(x_i)+...=\sum f(x_i)\phi_2(x_i)\rho(x_i)
a1∑ϕ2(xi)ϕ1(xi)ρ(xi)+ϕ2(xi)ϕ2(xi)ρ(xi)+...=∑f(xi)ϕ2(xi)ρ(xi)
.
.
.
...
...
通过解方程组求出系数
a
i
a_i
ai
例如拟合 y = a x + b y=ax+b y=ax+b,可以拆成 1 , x 1, x 1,x的线性组合
如果基函数是正交的,那么上面的方程中 ∑ ϕ i ( x i ) ϕ j ( x i ) ρ ( x i ) 当 i ≠ j \sum\phi_i(x_i)\phi_j(x_i)\rho(x_i) 当i≠j ∑ϕi(xi)ϕj(xi)ρ(xi)当i=j时会变成0,会出现大量约简
几种正交基函数的构造
记 ( f ( x ) , g ( x ) ) (f(x),g(x)) (f(x),g(x))是内积符号,即 ∫ f ( x ) g ( x ) \int f(x)g(x) ∫f(x)g(x)
勒让德多项式
L
0
(
x
)
=
1
L_0(x)=1
L0(x)=1
L
1
(
x
)
=
x
L_1(x)=x
L1(x)=x
L
n
+
1
(
x
)
=
2
n
+
1
n
+
1
L
n
(
x
)
−
n
n
+
1
L
n
−
1
(
x
)
L_{n+1}(x)=\frac{2n+1}{n+1}L_n(x)-\frac{n}{n+1}L_{n-1}(x)
Ln+1(x)=n+12n+1Ln(x)−n+1nLn−1(x)
在[-1,1]正交
L n ( x ) = 1 2 n n ! d n ( x 2 − 1 ) n d x n L_n(x)=\frac{1}{2^nn!}\frac{d^n(x^2-1)^n}{dx^n} Ln(x)=2nn!1dxndn(x2−1)n
他有
(
L
n
(
x
)
,
L
m
(
x
)
)
=
{
0
m
≠
n
2
2
n
+
1
n
=
m
(L_n(x),L_m(x))=\begin{cases} 0 & m≠n \\ \frac{2}{2n+1} & n=m\end{cases}
(Ln(x),Lm(x))={02n+12m=nn=m
切比雪夫多项式
T
0
(
x
)
=
1
,
T
1
(
x
)
=
x
T_0(x)=1,T_1(x)=x
T0(x)=1,T1(x)=x
T
n
+
1
(
x
)
=
2
x
T
n
(
x
)
−
T
n
−
1
(
x
)
T_{n+1}(x)=2xT_n(x)-T_{n-1}(x)
Tn+1(x)=2xTn(x)−Tn−1(x)
T
n
(
x
)
=
c
o
s
(
a
r
c
c
o
s
x
)
T_n(x)=cos(arccos\ x)
Tn(x)=cos(arccos x)
在[-1,1],关于权函数
ρ
(
x
)
=
1
1
−
x
2
\rho(x)=\frac{1}{\sqrt{1-x^2}}
ρ(x)=1−x21正交
(
T
n
(
x
)
,
T
m
(
x
)
)
=
{
0
m
≠
n
π
2
n
=
m
=
0
π
m
=
n
≠
0
(T_n(x),T_m(x))=\begin{cases} 0 & m≠n \\ \frac{π}{2}& n=m=0\\ π &m=n≠0 \end{cases}
(Tn(x),Tm(x))=⎩
⎨
⎧02ππm=nn=m=0m=n=0
埃尔米特多项式
H
0
(
x
)
=
1
,
H
1
(
x
)
=
2
x
H_0(x)=1,H_1(x)=2x
H0(x)=1,H1(x)=2x
H
n
+
1
(
x
)
=
2
x
H
n
(
x
)
−
2
n
H
n
−
1
(
x
)
H_{n+1}(x)=2xH_n(x)-2nH_{n-1}(x)
Hn+1(x)=2xHn(x)−2nHn−1(x)
H n ( x ) = ( − 1 ) n e x 2 d n e − x 2 d x n H_n(x)=(-1)^ne^{x^2}\frac{d^ne^{-x^2}}{dx^n} Hn(x)=(−1)nex2dxndne−x2
在正负无穷上,对于权函数 ρ ( x ) = e − x 2 \rho(x)=e^{-x^2} ρ(x)=e−x2正交
( L n ( x ) , L m ( x ) ) = { 0 m ≠ n 2 n ! π n = m (L_n(x),L_m(x))=\begin{cases} 0 & m≠n \\ 2^n!\sqrt{π} & n=m\end{cases} (Ln(x),Lm(x))={02n!πm=nn=m
拉盖尔多项式
U
0
(
x
)
=
1
,
U
1
(
x
)
=
1
−
x
U_0(x)=1, U_1(x)=1-x
U0(x)=1,U1(x)=1−x
U
n
+
1
(
x
)
=
(
1
+
2
n
−
x
)
U
n
(
x
)
−
n
2
U
n
−
1
(
x
)
U_{n+1}(x)=(1+2n-x)U_n(x)-n^2U_{n-1}(x)
Un+1(x)=(1+2n−x)Un(x)−n2Un−1(x)
U n ( x ) = e x d n ( x n e − x ) d x n U_n(x)=e^x\frac{d^n(x^ne^{-x})}{dx^n} Un(x)=exdxndn(xne−x)
在正无穷上,对于权函数 ρ ( x ) = e − x \rho(x)=e^{-x} ρ(x)=e−x正交
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
