曲面的切平面和法线
显示表示 z = f ( x , y ) z=f(x,y) z=f(x,y):
-
切平面: z = z 0 + ∂ f ∂ x ( x 0 , y 0 ) ( x − x 0 ) + ∂ f ∂ y ( x 0 , y 0 ) ( y − y 0 ) z=z_0+\frac{\partial f}{\partial x}(x_0,y_0)(x-x_0)+\frac{\partial f}{\partial y}(x_0,y_0)(y-y_0) z=z0+∂x∂f(x0,y0)(x−x0)+∂y∂f(x0,y0)(y−y0)
- z = z 0 + D f ( x 0 , y 0 ) ( x − x 0 , y − y 0 ) z=z_0+Df(x_0,y_0)(x-x_0,y-y_0) z=z0+Df(x0,y0)(x−x0,y−y0)相当于在 ( x 0 , y 0 ) (x_0,y_0) (x0,y0)泰勒展开一阶
-
法向量: n = ( ∂ f ∂ x , ∂ f ∂ y , − 1 ) n=\big(\frac{\partial f}{\partial x},\frac{\partial f}{\partial y},-1\big) n=(∂x∂f,∂y∂f,−1),即上式中 ( x − x 0 ) , ( y − y 0 ) , ( z − z 0 ) (x-x_0),(y-y_0),(z-z_0) (x−x0),(y−y0),(z−z0)前的参数,使得内积为 0 0 0
-
法线: ( x 0 , y 0 , z 0 ) + k n (x_0,y_0,z_0)+kn (x0,y0,z0)+kn,或者写作
x − x 0 ∂ f ∂ x ( x 0 , y 0 ) = y − y 0 ∂ f ∂ y ( x 0 , y 0 ) = z − z 0 − 1 \frac{x-x_0}{\frac{\partial f}{\partial x}(x_0,y_0)}=\frac{y-y_0}{\frac{\partial f}{\partial y}(x_0,y_0)}=\frac{z-z_0}{-1} ∂x∂f(x0,y0)x−x0=∂y∂f(x0,y0)y−y0=−1z−z0
隐式表示 F ( x , y , z ) = 0 F(x,y,z)=0 F(x,y,z)=0
- 切平面:
∂
F
∂
x
(
x
−
x
0
)
+
∂
F
∂
y
(
y
−
y
0
)
+
∂
F
∂
z
(
z
−
z
0
)
=
0
\frac{\partial F}{\partial x}(x-x_0)+\frac{\partial F}{\partial y}(y-y_0)+\frac{\partial F}{\partial z}(z-z_0)=0
∂x∂F(x−x0)+∂y∂F(y−y0)+∂z∂F(z−z0)=0
- D F ( x 0 , y 0 , z 0 ) ( x − x 0 , y − y 0 , z − z 0 ) = 0 DF(x_0,y_0,z_0)(x-x_0,y-y_0,z-z_0)=0 DF(x0,y0,z0)(x−x0,y−y0,z−z0)=0
- 即 ( x 0 , y 0 , z 0 ) (x_0,y_0,z_0) (x0,y0,z0)的微分作用在增量 ( x − x 0 , y − y 0 , z − z 0 ) (x-x_0,y-y_0,z-z_0) (x−x0,y−y0,z−z0)上不会改变 F F F
- 那么在这一平面上上近似于曲面 F = 0 F=0 F=0
- 法向量:
n
=
(
∂
F
∂
x
,
∂
F
∂
y
,
∂
F
∂
z
)
n=\big(\frac{\partial F}{\partial x},\frac{\partial F}{\partial y},\frac{\partial F}{\partial z}\big)
n=(∂x∂F,∂y∂F,∂z∂F)
- 为 F F F的梯度, F F F沿梯度上升最快,因此与 F F F恒等于 0 0 0正交
- 法线:
x − x 0 ∂ f ∂ x ( x 0 , y 0 ) = y − y 0 ∂ f ∂ y ( x 0 , y 0 ) = z − z 0 ∂ f ∂ z ( x 0 , y 0 ) \frac{x-x_0}{\frac{\partial f}{\partial x}(x_0,y_0)}=\frac{y-y_0}{\frac{\partial f}{\partial y}(x_0,y_0)}=\frac{z-z_0}{\frac{\partial f}{\partial z}(x_0,y_0)} ∂x∂f(x0,y0)x−x0=∂y∂f(x0,y0)y−y0=∂z∂f(x0,y0)z−z0
参数表示 ( x , y , z ) = ( x ( u , v ) , y ( u , v ) , z ( u , v ) ) = F ( u , v ) (x,y,z)=(x(u,v),y(u,v),z(u,v))=F(u,v) (x,y,z)=(x(u,v),y(u,v),z(u,v))=F(u,v)
-
切平面参数表示:
x − x 0 = ∂ x ∂ u ( u 0 , v 0 ) ( u − u 0 ) + ∂ x ∂ v ( u 0 , v 0 ) ( v − v 0 ) y − y 0 = ∂ y ∂ u ( u 0 , v 0 ) ( u − u 0 ) + ∂ y ∂ v ( u 0 , v 0 ) ( v − v 0 ) z − z 0 = ∂ z ∂ u ( u 0 , v 0 ) ( u − u 0 ) + ∂ z ∂ v ( u 0 , v 0 ) ( v − v 0 ) x-x_0=\frac{\partial x}{\partial u}(u_0,v_0)(u-u_0)+\frac{\partial x}{\partial v}(u_0,v_0)(v-v_0)\\ y-y_0=\frac{\partial y}{\partial u}(u_0,v_0)(u-u_0)+\frac{\partial y}{\partial v}(u_0,v_0)(v-v_0)\\ z-z_0=\frac{\partial z}{\partial u}(u_0,v_0)(u-u_0)+\frac{\partial z}{\partial v}(u_0,v_0)(v-v_0) x−x0=∂u∂x(u0,v0)(u−u0)+∂v∂x(u0,v0)(v−v0)y−y0=∂u∂y(u0,v0)(u−u0)+∂v∂y(u0,v0)(v−v0)z−z0=∂u∂z(u0,v0)(u−u0)+∂v∂z(u0,v0)(v−v0)- 对 x , y , z x,y,z x,y,z每个方向关于 ( u , v ) (u,v) (u,v)求切线 (一阶泰勒展开)
-
上式对于 X = ( x , y , z ) X=(x,y,z) X=(x,y,z)也可以写成: X − X 0 = ∂ X ∂ u ( u 0 , v 0 ) ∗ k 1 + ∂ X ∂ v ( u 0 , v 0 ) ∗ k 2 X-X_0=\frac{\partial X}{\partial u}(u_0,v_0)*k_1+\frac{\partial X}{\partial v}(u_0,v_0)*k_2 X−X0=∂u∂X(u0,v0)∗k1+∂v∂X(u0,v0)∗k2
-
通过叉乘求出法向量为( i , j , k i,j,k i,j,k分别为 x , y , z x,y,z x,y,z的基,其系数即该坐标分量的大小)
n = ∣ i j k ∂ x ∂ u ∂ y ∂ u ∂ z ∂ u ∂ x ∂ v ∂ y ∂ v ∂ z ∂ v ∣ n= \begin{vmatrix} i & j & k\\ \frac{\partial x}{\partial u} & \frac{\partial y}{\partial u} & \frac{\partial z}{\partial u}\\ \frac{\partial x}{\partial v} & \frac{\partial y}{\partial v} & \frac{\partial z}{\partial v} \end{vmatrix} n=∣∣∣∣∣∣i∂u∂x∂v∂xj∂u∂y∂v∂yk∂u∂z∂v∂z∣∣∣∣∣∣
- 切平面: < n , X > = 0 \left< n,X\right>=0 ⟨n,X⟩=0
- 对于一般的情况,法空间 N ( J F T ( u 0 , v 0 ) ) N(JF^T(u_0,v_0)) N(JFT(u0,v0)),即切空间 R ( J F ( u 0 , v 0 ) ) R(JF(u_0,v_0)) R(JF(u0,v0))的正交补
简单总结:
-
显式表达可以看作隐式表达,隐式表达下:
- 法向量为梯度向量,对于坐标求偏导
- 切平面方程由法向量得到
-
参数表达下:
-
切平面的参数方程,对于参数求偏导
-
根据维数:
-
法向量关于参数 u , v u,v u,v对应的 D u X ( u ) , D v X ( v ) D_uX(u),D_vX(v) DuX(u),DvX(v)求叉乘
-
法平面由切平面求正交补得到(解零空间)
-
-
切平面方程可由法空间得到
-
-
两者之间有明显的对偶关系
极值
极值定义
- 存在小邻域内所有点函数值都小于该点的函数值为极大值(根据小于或小于等于判断是否严格)
- 只能在内点讨论极值
临界点
- ∇ f ( x 0 ) = d f ( x 0 ) = 0 \nabla f(x_0)=df(x_0)=0 ∇f(x0)=df(x0)=0,即一阶微分为0,类似一元函数导数为0
非退化临界点
-
H f ( x 0 ) H_f(x_0) Hf(x0)可逆
-
H f ( x 0 ) H_f(x_0) Hf(x0)正定即极小值,负定即极大值
-
都不是则有正、负特征值,该点为鞍点
-
由泰勒展开得到上述结论: f ( x 0 + v ) = f ( x ) + v T H f ( x 0 ) v + o ( ∥ v ∥ 2 ) f(x_0+v)=f(x)+v^TH_f(x_0)v+o(\Vert v\Vert^2) f(x0+v)=f(x)+vTHf(x0)v+o(∥v∥2)
一般函数求极值:
-
解法一:
- 求出所有临界点
- 如果临界点都是非退化临界点则直接求 H e s s e Hesse Hesse矩阵判断正定性(顺序主子式为正数)
- 对于退化临界点讨论周围的点的 H e s s e Hesse Hesse矩阵
-
解法二:
条件极值
在满足 F k ( x ) = 0 , k = 1 , 2 , . . . , n F_k(x)=0,k=1,2,...,n Fk(x)=0,k=1,2,...,n的所有 x ∈ R m x\in \R ^m x∈Rm中求 f ( x ) f(x) f(x)的极值
若对于 F = ( F 1 , F 2 , . . . , F n ) F=(F_1,F_2,...,F_n) F=(F1,F2,...,Fn), D F ( x ∗ ) DF(x^*) DF(x∗)满行秩则可以根据隐函数定理得到 F ( x ) = 0 F(x)=0 F(x)=0是 m − n m-n m−n维曲面
必要条件
-
∇
f
(
x
∗
)
⊥
ker
D
F
(
x
∗
)
\nabla f(x^*)\perp \ker DF(x^*)
∇f(x∗)⊥kerDF(x∗)
- 即 f ( x ) f(x) f(x)增长方向与 F ( x ) = 0 F(x)=0 F(x)=0曲面的切平面垂直
- 否则可以沿 ∇ f ( x ∗ ) \nabla f(x^*) ∇f(x∗)在切平面的投影方向得到更大值)
- D F ( x ∗ ) DF(x^*) DF(x∗)满行秩则 ∇ F k ( x ∗ ) \nabla F_k(x^*) ∇Fk(x∗)为 ker D F ( x ∗ ) \ker DF(x^*) kerDF(x∗)一组基
- 则 ∇ f ( x ∗ ) = ∑ λ i ∇ F i ( x ∗ ) \nabla f(x^*)=\sum \lambda_i\nabla F_i(x^*) ∇f(x∗)=∑λi∇Fi(x∗)
对于必要条件,可以使用拉格朗日乘子法
-
L ( x , λ 1 , . . . , λ n ) = f ( x ) − λ 1 F 1 ( x ) − . . . − λ n F n ( x ) L(x,\lambda_1,...,\lambda_n)=f(x)-\lambda_1F_1(x)-...-\lambda_nF_n(x) L(x,λ1,...,λn)=f(x)−λ1F1(x)−...−λnFn(x)
-
( x ∗ , λ 1 ∗ , . . . , λ n ∗ ) (x^*,\lambda_1^*,...,\lambda_n^*) (x∗,λ1∗,...,λn∗)是 L L L的临界点:
- 0 = ∇ L ( x ∗ , λ 1 ∗ , . . . , λ n ∗ ) = ( ∇ f ( x ∗ ) − ∑ k = 1 n λ k ∇ F k ( x ∗ ) − F 1 ( x ∗ ) ⋮ − F n ( x ∗ ) ) 0=\nabla L(x^*,\lambda_1^*,...,\lambda_n^*)=\begin{pmatrix}\nabla f(x^*)-\sum_{k=1}^n\lambda_k\nabla F_k(x^*)\\-F_1(x^*)\\ \vdots \\ -F_n(x^*)\end{pmatrix} 0=∇L(x∗,λ1∗,...,λn∗)=⎝⎜⎜⎜⎛∇f(x∗)−∑k=1nλk∇Fk(x∗)−F1(x∗)⋮−Fn(x∗)⎠⎟⎟⎟⎞
- 上一行的列向量中,第一个元素是 m m m维列向量,即对 x x x求偏导在 x ∗ x^* x∗处取,实际上是对 x 1 . . . x m x_1...x_m x1...xm求 m m m个偏导分别为0
- ∂ x L ( x ∗ , λ ) \partial_{x} L(x^*,\lambda) ∂xL(x∗,λ)是 L L L在 ( x ∗ , λ ) (x^*,\lambda) (x∗,λ)处的微分,是 1 ∗ m 1*m 1∗m的 J a c o b i Jacobi Jacobi矩阵,是 ∇ L \nabla L ∇L前 m m m行对应列向量的转置,是 L L L分别对前 m m m个分量求偏导数组合起来的列向量
-
记 H = d 2 f ( x ∗ ) − ( λ 1 , . . . , λ n ) D 2 F ( x ∗ ) H=d^2f(x^*)-(\lambda_1,...,\lambda _n)D^2F(x^*) H=d2f(x∗)−(λ1,...,λn)D2F(x∗)
- 在 F ( x ∗ ) F(x^*) F(x∗)的切平面上 H H H正定为极小值,负定为极大值,
- 同时有正负特征值时不是极值点