微分
在 x 0 x_0 x0处微分或导数: ∂ f ( x 0 ) , d f ( x 0 ) \partial f(x_0),df(x_0) ∂f(x0),df(x0)
-
f ( x 0 + v ) = f ( x 0 ) + ∂ f ( x 0 ) ( v ) + o ( v ) f(x_0+v)=f(x_0)+\partial f(x_0)(v)+o(v) f(x0+v)=f(x0)+∂f(x0)(v)+o(v)
-
线性映射
沿向量微分
在 x 0 x_0 x0处沿向量 v v v的导数: ∂ f ∂ v ( x 0 ) , ∂ v f ( x 0 ) \frac{\partial f}{\partial v}(x_0),\partial_vf(x_0) ∂v∂f(x0),∂vf(x0)
- lim t → 0 + f ( x 0 + t v ) − f ( x 0 ) t \lim_{t\to 0^+}\frac{f(x_0+tv)-f(x_0)}{t} limt→0+tf(x0+tv)−f(x0)
- 当 ∥ v ∥ = 1 \Vert v \Vert=1 ∥v∥=1时,称为沿方向 v v v的导数
当 f f f在 x 0 x_0 x0处可微时: ∂ f ( x 0 ) ( v ) = ∂ f ∂ v ( x 0 ) = ∂ v f ( x 0 ) \partial f(x_0)(v)=\frac{\partial f}{\partial v}(x_0)=\partial_vf(x_0) ∂f(x0)(v)=∂v∂f(x0)=∂vf(x0)
- 此时 ∂ f ∂ v ( x 0 ) \frac{\partial f}{\partial v}(x_0) ∂v∂f(x0)关于 v v v是线性的
微分与沿向量微分都可以为值域多维的函数
偏导数
偏导数:对于坐标 ( x 1 , x 2 , . . . , x m ) (x_1,x_2,...,x_m) (x1,x2,...,xm),在 ( a 1 , a 2 , . . . , a m ) (a_1,a_2,...,a_m) (a1,a2,...,am)处,关于坐标 x k x_k xk偏导数 ∂ f ∂ x k ( X ) , f x k ′ ( X ) \frac{\partial f}{\partial x_k}(X),f'_{x_k}(X) ∂xk∂f(X),fxk′(X)
- lim t → 0 f ( a 1 , . . , a k + t , . . a m ) − f ( a 1 , . . . a m ) t \lim_{t\to 0}\frac{f(a_1,..,a_k+t,..a_m)-f(a_1,...a_m)}{t} limt→0tf(a1,..,ak+t,..am)−f(a1,...am)
- 偏导数是 E → R E\to \R E→R的函数,而前面定义的导数既可以是函数也可以是映射
- 有时可以写成: f x k ( X ) , ∂ x k f ( X ) , ∂ k f ( X ) f_{x_k}(X),\partial_{x_k}f(X),\partial_kf(X) fxk(X),∂xkf(X),∂kf(X)
偏导数与沿向量微分的关系
对于 ∂ f ∂ x ( x 0 ) \frac{\partial f}{\partial x}(x_0) ∂x∂f(x0),如果 x x x是一个向量,那么就看作沿向量的导数(可以为值域多元的映射),如果 x x x对应着一个坐标,那么就是偏导数(只能是值域一维的函数)。
-
设 e k e_k ek是坐标 x k x_k xk对应的基底向量,即 x k + 1 x_k+1 xk+1在空间中的向量,则可以将偏导与方向导数联系起来。
-
∂ f ∂ x k ( X ) = ∂ f ∂ e k ( X ) = ∂ f ( X ) ( e k ) = λ ∂ f ( X ) ( e k λ ) \frac{\partial f}{\partial x_k}(X)=\frac{\partial f}{\partial e_k}(X)=\partial f(X)(e_k)=\lambda \partial f(X)(\frac{e_k}{\lambda}) ∂xk∂f(X)=∂ek∂f(X)=∂f(X)(ek)=λ∂f(X)(λek)
-
偏导与范数无关,与坐标系有关(即与基底 e k e_k ek有关)
-
d f ( x 0 ) ( v ) = d f ( x 0 ) ( ∑ ξ i e i ) = ∑ ξ i d f ( x 0 ) ( e i ) = ∑ ξ i ∂ i f ( x 0 ) = ∑ ∂ i f ( x 0 ) d x i ( v ) = ( ∂ 1 f , ∂ 2 f , . . . , ∂ m f ) ( d 1 . . d m ) T ( v ) df(x_0)(v)=df(x_0)(\sum \xi_i e_i)=\sum \xi_idf(x_0)(e_i)=\sum \xi_i \partial_i f(x_0)=\sum \partial_if(x_0)dx_i(v)=(\partial_1f,\partial_2f,...,\partial_mf)(d_1..d_m)^T(v) df(x0)(v)=df(x0)(∑ξiei)=∑ξidf(x0)(ei)=∑ξi∂if(x0)=∑∂if(x0)dxi(v)=(∂1f,∂2f,...,∂mf)(d1..dm)T(v)
-
Jacobi矩阵:对于值域为 n n n维的映射的导数,将上式排成一列即可得到矩阵:
F = ( f 1 , f 2 , . . . f n ) , J F ( x 0 ) = ( ∂ f 1 ∂ x 1 ( x 0 ) . . . ∂ f 1 ∂ x m ( x 0 ) ⋮ ⋱ ⋮ ∂ f n ∂ x 1 ( x 0 ) . . . ∂ f n ∂ x m ( x 0 ) ) F=(f_1,f_2,...f_n),JF(x_0)=\begin{pmatrix} \frac{\partial f_1}{\partial x_1}(x_0) & ... & \frac{\partial f_1}{\partial x_m}(x_0)\\ \vdots & \ddots & \vdots \\ \frac{\partial f_n}{\partial x_1}(x_0) & ... & \frac{\partial f_n}{\partial x_m}(x_0) \end{pmatrix} F=(f1,f2,...fn),JF(x0)=⎝⎜⎛∂x1∂f1(x0)⋮∂x1∂fn(x0)...⋱...∂xm∂f1(x0)⋮∂xm∂fn(x0)⎠⎟⎞
梯度
梯度只对于值域为一维的函数考虑
对于线性函数 L ( v ) : R n → R L(v):\R^n\to \R L(v):Rn→R,定义 ∇ L \nabla L ∇L满足 L ( v ) = < ∇ L , v > L(v)=\left<\nabla L,v\right> L(v)=⟨∇L,v⟩
对于线性函数 d f ( x 0 ) ( v ) df(x_0)(v) df(x0)(v),记 ∇ f ( x 0 ) \nabla f(x_0) ∇f(x0)为在 x 0 x_0 x0处的梯度向量, d f ( x 0 ) ( v ) = < ∇ f ( x 0 ) , v > df(x_0)(v)=\left<\nabla f(x_0),v\right> df(x0)(v)=⟨∇f(x0),v⟩
-
定义 ∇ f ( x 0 ) \nabla f(x_0) ∇f(x0)为 f f f在 x 0 x_0 x0的梯度向量, ∇ f ( x 0 ) = λ v ∗ , ∥ v ∗ ∥ = 1 \nabla f(x_0)=\lambda v^*,\Vert v^*\Vert=1 ∇f(x0)=λv∗,∥v∗∥=1且
λ = d f ( x 0 ) v ∗ = max ∥ v ∥ = 1 d f ( x 0 ) ( v ) \lambda=df(x_0)v^*=\max_{\Vert v\Vert=1}df(x_0)(v) λ=df(x0)v∗=∥v∥=1maxdf(x0)(v) -
形象理解,梯度向量 ∇ f \nabla f ∇f是关于线性函数的在定义域内的向量,对于函数在一点 x 0 x_0 x0的导数就有梯度向量 ∇ f ( x 0 ) \nabla f(x_0) ∇f(x0), ∇ \nabla ∇的方向是 d f df df最大的方向(即 f f f增长最快的方向),且 ∇ \nabla ∇的模长为这个方向上的导数,因此 < ∇ , v > = < λ v ∗ , v > = d f ( x 0 ) ( v ∗ ) ∗ ( v 在 v ∗ 方 向 上 投 影 的 模 长 ) \left<\nabla,v\right>=\left<\lambda v^*,v\right>=df(x_0)(v^*)*(v在v^*方向上投影的模长) ⟨∇,v⟩=⟨λv∗,v⟩=df(x0)(v∗)∗(v在v∗方向上投影的模长)
基于此可以给出在
(
e
1
,
e
2
,
.
.
.
,
e
n
)
(e_1,e_2,...,e_n)
(e1,e2,...,en)为标准正交基的梯度
∇
f
(
x
0
)
:
(
∂
f
∂
x
1
(
x
0
)
,
∂
f
∂
x
2
(
x
0
)
,
.
.
.
∂
f
∂
x
n
(
x
0
)
)
T
实
际
上
:
∇
f
(
x
0
)
=
∑
i
=
1
n
∂
f
∂
x
i
(
x
0
)
e
i
\nabla f(x_0):\bigg(\frac{\partial f}{\partial x_1}(x_0),\frac{\partial f}{\partial x_2}(x_0),...\frac{\partial f}{\partial x_n}(x_0)\bigg)^T\\ 实际上:\\ \nabla f(x_0)=\sum_{i=1}^n\frac{\partial f}{\partial x_i}(x_0)e_i
∇f(x0):(∂x1∂f(x0),∂x2∂f(x0),...∂xn∂f(x0))T实际上:∇f(x0)=i=1∑n∂xi∂f(x0)ei
-
前者由于是坐标所以是列向量
-
< ∇ f ( x 0 ) , v > = ( ∂ f ∂ x 1 ( x 0 ) , ∂ f ∂ x 2 ( x 0 ) , . . . ∂ f ∂ x n ( x 0 ) ) G ( e 1 . . e n ) ( d x 1 ( v ) d x 2 ( v ) ⋮ d x n ( v ) ) \left <\nabla f(x_0),v\right>=\bigg(\frac{\partial f}{\partial x_1}(x_0),\frac{\partial f}{\partial x_2}(x_0),...\frac{\partial f}{\partial x_n}(x_0)\bigg)G(e_1..e_n)\begin{pmatrix}dx_1(v)\\dx_2(v)\\ \vdots \\ dx_n(v)\end{pmatrix} ⟨∇f(x0),v⟩=(∂x1∂f(x0),∂x2∂f(x0),...∂xn∂f(x0))G(e1..en)⎝⎜⎜⎜⎛dx1(v)dx2(v)⋮dxn(v)⎠⎟⎟⎟⎞
-
G G G为 ( e 1 . . e n ) (e_1..e_n) (e1..en)为基的度量矩阵,如果为标准正交基则为 I I I,那么 < ∇ f , v > = d f ( x 0 ) ( v ) \left <\nabla f,v\right>=df(x_0)(v) ⟨∇f,v⟩=df(x0)(v)
-
如果 G ≠ I G\neq I G=I,则 ∇ f ( x 0 ) \nabla f(x_0) ∇f(x0)的坐标为 X = G − 1 ( ∂ f ∂ x 1 ( x 0 ) , ∂ f ∂ x 2 ( x 0 ) , . . . ∂ f ∂ x n ( x 0 ) ) T X=G^{-1}\bigg(\frac{\partial f}{\partial x_1}(x_0),\frac{\partial f}{\partial x_2}(x_0),...\frac{\partial f}{\partial x_n}(x_0)\bigg)^T X=G−1(∂x1∂f(x0),∂x2∂f(x0),...∂xn∂f(x0))T,如此 < ∇ f ( x 0 ) , v > = Y G X = Y G G − 1 X ′ = Y X ′ \left <\nabla f(x_0),v\right>=YGX=YGG^{-1}X'=YX' ⟨∇f(x0),v⟩=YGX=YGG−1X′=YX′
高阶偏导数
对于
i
k
∈
[
1
,
n
]
i_k\in[1,n]
ik∈[1,n],依次对坐标
x
i
1
,
x
i
2
.
.
.
x
i
k
x_{i_1},x_{i_2}...x_{i_k}
xi1,xi2...xik求偏导
∂
k
f
∂
x
i
k
∂
x
i
k
−
1
.
.
.
∂
x
i
1
(
x
0
)
∂
x
i
k
,
x
i
k
−
1
,
.
.
.
,
x
i
1
k
f
∂
i
k
,
i
k
−
1
,
.
.
.
,
x
i
1
k
f
f
x
i
1
,
x
i
2
,
.
.
.
,
x
k
(
k
)
f
i
1
,
i
2
,
.
.
.
,
i
k
(
k
)
\frac{\partial^k f}{\partial x_{i_k}\partial x_{i_{k-1}}...\partial x_{i_1}}(x_0)\\ \partial ^{k}_{x_{i_k},x_{i_{k-1}},...,x_{i_1}}f\\ \partial ^{k}_{i_k,i_{k-1},...,x_{i_1}}f\\ f^{(k)}_{x_{i_1},x_{i_2},...,x_{k}}\\ f^{(k)}_{i_1,i_2,...,i_k}
∂xik∂xik−1...∂xi1∂kf(x0)∂xik,xik−1,...,xi1kf∂ik,ik−1,...,xi1kffxi1,xi2,...,xk(k)fi1,i2,...,ik(k)
- 若 f f f在 x 0 x_0 x0满足 k k k阶连续,那么任意 k k k次偏导,关于求导顺序无关
k
k
k阶微分是关于
k
k
k个向量
v
1
.
.
v
k
v_1..v_k
v1..vk的
k
k
k重线性函数
d
f
k
(
x
)
(
v
1
.
.
v
k
)
=
∑
i
1
,
i
2
.
.
i
k
∂
k
f
∂
x
i
k
∂
x
i
k
−
1
.
.
.
∂
x
i
1
(
x
)
v
1
,
i
1
v
2
,
i
2
.
.
.
v
k
,
i
k
=
∂
k
f
∂
v
k
.
.
∂
v
1
(
x
)
df^k(x)(v_1..v_k)=\sum_{i_1,i_2..i_k}\frac{\partial ^kf}{\partial x_{i_k}\partial x_{i_{k-1}}...\partial x_{i_1}}(x)v_{1,i_1}v_{2,i_2}...v_{k,i_k}=\frac{\partial ^kf}{\partial v_k..\partial v_1}(x)
dfk(x)(v1..vk)=i1,i2..ik∑∂xik∂xik−1...∂xi1∂kf(x)v1,i1v2,i2...vk,ik=∂vk..∂v1∂kf(x)
泰勒展开
一维泰勒公式:
f
(
x
0
+
t
)
=
∑
i
=
0
k
−
1
f
(
i
)
(
x
0
)
i
!
t
i
+
o
(
t
k
−
1
)
=
∑
i
=
0
k
−
1
f
(
i
)
(
x
0
)
i
!
t
i
+
f
(
k
)
(
x
0
+
t
θ
)
k
!
t
k
f(x_0+t)=\sum_{i=0}^{k-1}\frac{f^{(i)}(x_0)}{i!}t^i+o(t^{k-1})=\sum_{i=0}^{k-1}\frac{f^{(i)}(x_0)}{i!}t^i+\frac{f^{(k)}(x_0+t\theta)}{k!}t^k
f(x0+t)=i=0∑k−1i!f(i)(x0)ti+o(tk−1)=i=0∑k−1i!f(i)(x0)ti+k!f(k)(x0+tθ)tk
任意维泰勒公式:
f
(
x
0
+
t
v
)
=
∑
i
=
0
k
−
1
∂
i
f
∂
v
i
(
x
0
)
t
i
i
!
+
o
(
t
k
−
1
)
=
∑
i
=
0
k
−
1
∂
i
f
∂
v
i
(
x
0
)
t
i
i
!
+
∂
k
f
∂
v
k
(
x
0
+
t
θ
)
t
k
k
!
f(x_0+tv)=\sum_{i=0}^{k-1}\frac{\partial^i f}{\partial v^i}(x_0)\frac{t^i}{i!}+o(t^{k-1})=\sum_{i=0}^{k-1}\frac{\partial^i f}{\partial v^i}(x_0)\frac{t^i}{i!}+\frac{\partial^k f}{\partial v^k}(x_0+t\theta)\frac{t^k}{k!}
f(x0+tv)=i=0∑k−1∂vi∂if(x0)i!ti+o(tk−1)=i=0∑k−1∂vi∂if(x0)i!ti+∂vk∂kf(x0+tθ)k!tk
根据高阶偏导数:
f
(
x
0
+
t
v
)
=
∑
i
=
1
k
−
1
∂
i
f
∂
v
i
(
x
0
)
t
i
i
!
+
o
(
t
k
−
1
)
=
∑
i
=
1
k
−
1
∑
α
1
+
.
.
.
+
α
n
=
i
∂
i
f
∂
x
1
α
1
.
.
.
∂
x
n
α
n
(
x
0
)
∏
j
=
1
n
(
t
ξ
j
)
α
j
α
j
!
+
o
(
t
k
−
1
)
令
v
∗
=
t
v
,
∥
v
∥
=
1
,
则
∥
v
0
∥
=
t
f
(
x
0
+
v
∗
)
=
∑
i
=
1
k
−
1
∑
α
1
+
.
.
.
+
α
n
=
i
∂
i
f
∂
x
1
α
1
.
.
.
∂
x
n
α
n
(
x
0
)
∏
j
=
1
n
(
ξ
j
∗
)
α
j
α
j
!
+
o
(
∥
v
∗
∥
k
−
1
)
f(x_0+tv)=\sum_{i=1}^{k-1}\frac{\partial^if}{\partial v^i}(x_0)\frac{t^i}{i!}+o(t^{k-1}) \\=\sum_{i=1}^{k-1}\sum_{\alpha_1+...+\alpha_n=i}\frac{\partial^if}{\partial x_1^{\alpha_1}...\partial x_n^{\alpha_n}}(x_0)\prod_{j=1}^n\frac{(t\xi_j)^{\alpha_j}}{\alpha_j!}+o(t^{k-1}) \\ 令v^*=tv,\Vert v\Vert=1,则\Vert v_0\Vert=t \\ f(x_0+v^*)=\sum_{i=1}^{k-1}\sum_{\alpha_1+...+\alpha_n=i}\frac{\partial^if}{\partial x_1^{\alpha_1}...\partial x_n^{\alpha_n}}(x_0)\prod_{j=1}^n\frac{(\xi^*_j)^{\alpha_j}}{\alpha_j!}+o(\Vert v^*\Vert^{k-1})
f(x0+tv)=i=1∑k−1∂vi∂if(x0)i!ti+o(tk−1)=i=1∑k−1α1+...+αn=i∑∂x1α1...∂xnαn∂if(x0)j=1∏nαj!(tξj)αj+o(tk−1)令v∗=tv,∥v∥=1,则∥v0∥=tf(x0+v∗)=i=1∑k−1α1+...+αn=i∑∂x1α1...∂xnαn∂if(x0)j=1∏nαj!(ξj∗)αj+o(∥v∗∥k−1)
隐函数定理
C
k
C^k
Ck类函数
F
F
F以及向量
X
∈
R
m
,
Y
∈
R
n
X\in \R ^m,Y\in \R ^n
X∈Rm,Y∈Rn,在
(
X
0
,
Y
0
)
(X_0,Y_0)
(X0,Y0)有(由于方程个数即未知数个数,所以
F
F
F和
Y
Y
Y的维数都是
n
n
n)
F
(
X
0
,
Y
0
)
=
(
F
1
(
X
0
,
Y
0
)
,
.
.
.
,
F
n
(
X
0
,
Y
0
)
)
=
0
F(X_0,Y_0)=(F_1(X_0,Y_0),...,F_n(X_0,Y_0))=0
F(X0,Y0)=(F1(X0,Y0),...,Fn(X0,Y0))=0
并且在
(
X
0
,
Y
0
)
(X_0,Y_0)
(X0,Y0)对于
Y
Y
Y到
F
F
F的
J
a
c
o
b
i
Jacobi
Jacobi矩阵有
d
e
t
(
∂
F
∂
Y
)
=
d
e
t
(
∂
(
F
1
,
F
2
,
.
.
.
,
F
n
)
∂
(
y
1
,
y
2
,
.
.
.
,
y
n
)
)
≠
0
det(\frac{\partial F}{\partial Y})=det(\frac{\partial(F_1,F_2,...,F_n)}{\partial(y_1,y_2,...,y_n)})\neq 0
det(∂Y∂F)=det(∂(y1,y2,...,yn)∂(F1,F2,...,Fn))=0
可以得到在
(
X
0
,
Y
0
)
(X_0,Y_0)
(X0,Y0)的小邻域内有
C
k
C^k
Ck类隐函数
Y
=
Y
(
X
)
Y=Y(X)
Y=Y(X),且在
(
X
0
,
Y
0
)
(X_0,Y_0)
(X0,Y0)有
∂
Y
∂
X
=
−
(
∂
F
∂
Y
)
−
1
∂
F
∂
X
\frac{\partial Y}{\partial X}=-(\frac{\partial F}{\partial Y})^{-1}\frac{\partial F}{\partial X}
∂X∂Y=−(∂Y∂F)−1∂X∂F
- 对于二元函数 f ( x , y ) f(x,y) f(x,y),退化为 f ( x 0 , y 0 ) = 0 f(x_0,y_0)=0 f(x0,y0)=0,满足 ∂ f ∂ y ( x 0 , y 0 ) ≠ 0 \frac{\partial f}{\partial y}(x_0,y_0)\neq 0 ∂y∂f(x0,y0)=0则有隐函数 y = y ( x ) y=y(x) y=y(x)且
d y d x ( x 0 , y 0 ) = − ∂ f ∂ x ( x 0 , y 0 ) ∂ f ∂ y ( x 0 , y 0 ) \frac{dy}{dx}(x_0,y_0)=-\frac{\frac{\partial f}{\partial x}(x_0,y_0)}{\frac{\partial f}{\partial y}(x_0,y_0)} dxdy(x0,y0)=−∂y∂f(x0,y0)∂x∂f(x0,y0)
注意 d e t ( ∂ F ∂ Y ) ≠ 0 det(\frac{\partial F}{\partial Y})\neq 0 det(∂Y∂F)=0并不是必要条件而是充分条件
如何快速理解 ∂ Y ∂ X \frac{\partial Y}{\partial X} ∂X∂Y的式子?
-
移项后得到 ∂ F ∂ X + ∂ F ∂ Y ∂ Y ∂ X = 0 \frac{\partial F}{\partial X}+\frac{\partial F }{\partial Y}\frac{\partial Y}{\partial X}=0 ∂X∂F+∂Y∂F∂X∂Y=0
-
这实际上是对 F ( X , Y ( X ) ) = 0 F(X,Y(X))=0 F(X,Y(X))=0,复合函数对于 X X X求偏导数
更加感性但直观的理解?
-
对于 d y d x ( x 0 , y 0 ) = − ∂ f ∂ x ( x 0 , y 0 ) ∂ f ∂ y ( x 0 , y 0 ) \frac{dy}{dx}(x_0,y_0)=-\frac{\frac{\partial f}{\partial x}(x_0,y_0)}{\frac{\partial f}{\partial y}(x_0,y_0)} dxdy(x0,y0)=−∂y∂f(x0,y0)∂x∂f(x0,y0)
-
将左边的 d x dx dx移到右边,将右边的 ∂ f ∂ y \frac{\partial f}{\partial y} ∂y∂f移到左边
-
∂ f ∂ y d y + ∂ f ∂ x d x = 0 \frac{\partial f}{\partial y}dy+\frac{\partial f}{\partial x}dx=0 ∂y∂fdy+∂x∂fdx=0
-
直观来看就是对于 d y / d x dy/dx dy/dx满足这个式子,就有沿着 ( d x , d y ) (dx,dy) (dx,dy)的方向,满足 d f = 0 df=0 df=0,即 f ( x + d x , y + d y ) = f ( x , y ) = 0 f(x+dx,y+dy)=f(x,y)=0 f(x+dx,y+dy)=f(x,y)=0
对于
∂
Y
∂
X
=
−
(
∂
F
∂
Y
)
−
1
∂
F
∂
X
\frac{\partial Y}{\partial X}=-(\frac{\partial F}{\partial Y})^{-1}\frac{\partial F}{\partial X}
∂X∂Y=−(∂Y∂F)−1∂X∂F,如果令
F
(
X
,
Y
)
=
f
(
X
)
−
Y
=
0
F(X,Y)=f(X)-Y=0
F(X,Y)=f(X)−Y=0,那么就有
∂
Y
∂
X
=
−
(
∂
F
∂
Y
)
−
1
∂
F
∂
X
=
(
∂
X
∂
Y
)
−
1
\frac{\partial Y}{\partial X}=-(\frac{\partial F}{\partial Y})^{-1}\frac{\partial F}{\partial X}=(\frac{\partial X}{\partial Y})^{-1}
∂X∂Y=−(∂Y∂F)−1∂X∂F=(∂Y∂X)−1
即对于
Y
=
Y
(
X
)
Y=Y(X)
Y=Y(X)以及
X
=
X
(
Y
)
X=X(Y)
X=X(Y)的
J
a
c
o
b
i
Jacobi
Jacobi矩阵有互为逆的关系