对于给定一个关于 $x, y$ 的函数式:
$$ f(x, y) $$
求它的全微分,
$$ df(x, y) = \sum_{i=0}^{n}\frac{\partial f_i}{\partial x_i}dx_i + \sum_{i=0}^{n}\frac{\partial f_i}{\partial y_i}dy_i\tag{1} $$
我们假设,存在 $g(x, y)$,其自变量 $u$ 存在有,
$$ u_i = \frac{\partial f_i}{\partial x_i} $$
代入 $(1)$ 则有,
$$ df(x,y) = \sum_{i=0}^{n}u_idx_i + \sum_{i=0}^{n}\frac{\partial f_i}{\partial y_i}dy_i\tag{2} $$
又由,
$$ d(u_ix_i) = x_idu_i + u_idx_i \tag{3} $$
$(2)$ 可化为,
$$ df(x,y) = \sum{d(u_ix_i)} - \sum{x_idu_i} + \sum\frac{\partial f_i}{\partial y_i}dy_i $$ 即 $$ d(\sum{du_ix_i} - f) = \sum{x_idu_i} - \sum{\frac{\partial f_i}{\partial y_i}dy_i}\tag{4} $$
我们假设,
$$ g(u,y) = \sum{du_ix_i - f} \tag{5} $$
所以 $(4)$ 与 $g(u,y)$ 的全微分可以建立关系式:
$$\begin{cases}
dg(u,y) = \sum{x_idu_i} - \sum{\frac{\partial f_i}{\partial y_i}dy_i}
dg(u,y) = \sum{\frac{\partial g_i}{\partial u_i}}du_i-\sum{\frac{\partial g_i}{\partial y_i}}dy_i
\end{cases}\tag{6}
$$
从 $(6)$ 中得到以下结论:
- 新函数对新自变量的偏微分等于原函数的自变量
$$ \frac{\partial g_i}{\partial u_i} = x_i $$
- 原函数对原自变量的偏微分等于新函数自变量
$$ \frac{\partial f_i}{\partial x_i} = ui $$
- 原函数对不变量的偏微分等于新函数对不变量的偏微分
$$ \frac{\partial f_i}{\partial y_i} = \frac{\partial g_i}{\partial y_i} $$