求三角形标准化法向量对三个顶点的导数

我们有一个三角形有三个顶点: $x_i$ , $x_j$ , 和 $x_k$ .

标准化的法向量为:
$$\overrightarrow{n}=\frac{(\overrightarrow{x_j}-\overrightarrow{x_i})\times (x_k-\overrightarrow{x_i})}{||(\overrightarrow{x_j}-\overrightarrow{x_i})\times (\overrightarrow{x_k}-\overrightarrow{x_i})||}$$
为了求偏导, 即, $\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{x_i}}$,$\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{x_j}}$, and $\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{x_k}}$,它们均为3x3矩阵, 使用链式法则得到.
变量替换$\overrightarrow{a}=\overrightarrow{x_j}-\overrightarrow{x_i}$ and $\overrightarrow{b}=\overrightarrow{x_k}-\overrightarrow{x_i}$, 使用链式法则后:
$$\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{x_{i,j,k}}}=\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{a}}\frac{\partial \overrightarrow{a}}{\partial \overrightarrow{x_{i,j,k}}}+\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{b}}\frac{\partial \overrightarrow{b}}{\partial \overrightarrow{x_{i,j,k}}}$$
这些偏导可以很容易求得:
$$\{\begin{array}{l}\frac{\partial \overrightarrow{a}}{\partial \overrightarrow{x_i}}=-{\text{I}}_{3\times 3}\\ \frac{\partial \overrightarrow{a}}{\partial \overrightarrow{x_j}}={\text{I}}_{3\times 3}\\ \frac{\partial \overrightarrow{a}}{\partial \overrightarrow{x_k}}={\text{0}}_{3\times 3}\end{array}$$
$$\{\begin{array}{l}\frac{\partial \overrightarrow{b}}{\partial \overrightarrow{x_i}}=-{\text{I}}_{3\times 3}\\ \frac{\partial \overrightarrow{b}}{\partial \overrightarrow{x_j}}={\text{0}}_{3\times 3}\\ \frac{\partial \overrightarrow{b}}{\partial \overrightarrow{x_k}}={\text{I}}_{3\times 3}\end{array}$$
现在我们只需要求 $\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{a}}$ 和 $\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{b}}$:
$\text{引理}$:
$$\frac{\partial (\overrightarrow{a}\times \overrightarrow{b})}{\partial \overrightarrow{a}}=-[B],\frac{\partial (\overrightarrow{a}\times \overrightarrow{b})}{\partial \overrightarrow{b}}=[A],\frac{\partial ||\overrightarrow{n}||}{\partial \overrightarrow{n}}=\frac{\overrightarrow{n}}{||\overrightarrow{n}||}$$
其中$[A]$ 和 $[B]$ 是向量$\overrightarrow{a}$ 和$\overrightarrow{b}$的反对称矩阵（skew matrix）
因此:
$$\begin{array}{rl}\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{a}}& =\frac{\partial }{\partial \overrightarrow{a}}(\frac{\overrightarrow{a}\times \overrightarrow{b}}{||\overrightarrow{a}\times \overrightarrow{b}||})\\ & =\frac{\frac{\partial (\overrightarrow{a}\times \overrightarrow{b})}{\partial \overrightarrow{a}}||\overrightarrow{a}\times \overrightarrow{b}||-(\overrightarrow{a}\times \overrightarrow{b})\otimes \frac{\partial ||\overrightarrow{a}\times \overrightarrow{b}||}{\partial \overrightarrow{a}}}{||\overrightarrow{a}\times \overrightarrow{b}|{|}^2}\\ & =\frac{\frac{\partial (\overrightarrow{a}\times \overrightarrow{b})}{\partial \overrightarrow{a}}||\overrightarrow{a}\times \overrightarrow{b}||-(\overrightarrow{a}\times \overrightarrow{b})\otimes \frac{(\overrightarrow{a}\times \overrightarrow{b})}{||\overrightarrow{a}\times \overrightarrow{b}||}\frac{\partial (\overrightarrow{a}\times \overrightarrow{b})}{\partial \overrightarrow{a}}}{||\overrightarrow{a}\times \overrightarrow{b}|{|}^2}\\ & =\frac{\frac{\partial (\overrightarrow{a}\times \overrightarrow{b})}{\partial \overrightarrow{a}}-\frac{(\overrightarrow{a}\times \overrightarrow{b})}{||\overrightarrow{a}\times \overrightarrow{b}||}\otimes \frac{(\overrightarrow{a}\times \overrightarrow{b})}{||\overrightarrow{a}\times \overrightarrow{b}||}\frac{\partial (\overrightarrow{a}\times \overrightarrow{b})}{\partial \overrightarrow{a}}}{||\overrightarrow{a}\times \overrightarrow{b}||}\\ & =\frac{\frac{\partial (\overrightarrow{a}\times \overrightarrow{b})}{\partial \overrightarrow{a}}-\overrightarrow{n}\otimes \overrightarrow{n}\frac{\partial (\overrightarrow{a}\times \overrightarrow{b})}{\partial \overrightarrow{a}}}{||\overrightarrow{a}\times \overrightarrow{b}||}\\ & =\frac{{\text{I}}_{3\times 3}-\overrightarrow{n}\otimes \overrightarrow{n}}{||\overrightarrow{a}\times \overrightarrow{b}||}\frac{\partial (\overrightarrow{a}\times \overrightarrow{b})}{\partial \overrightarrow{a}}\end{array}$$
同样的:
$$\begin{array}{rl}\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{b}}& =\frac{{\text{I}}_{3\times 3}-\overrightarrow{n}\otimes \overrightarrow{n}}{||\overrightarrow{a}\times \overrightarrow{b}||}\frac{\partial (\overrightarrow{a}\times \overrightarrow{b})}{\partial \overrightarrow{b}}\end{array}$$
替换$T=\frac{{\text{I}}_{3\times 3}-\overrightarrow{n}\otimes \overrightarrow{n}}{||\overrightarrow{a}\times \overrightarrow{b}||}$
$$\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{a}}=-T[B],\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{b}}=T[A]$$
带入上面的链式法则，我们最终得到:
$$\{\begin{array}{l}\frac{\partial \overrightarrow{n}}{\partial \overrightarrow{x_i}}=T([B]-[A])\\ \frac{\partial \overrightarrow{n}}{\partial \overrightarrow{x_j}}=-T[B]\\ \frac{\partial \overrightarrow{n}}{\partial \overrightarrow{x_k}}=T[A]\end{array}$$

参考：
1）https://math.stackexchange.com/questions/2614305/derivative-of-cross-product-of-vectors-with-respect-to-vector
2）https://math.stackexchange.com/questions/1985897/can-matrices-do-cross-product-with-vectors