利用全微分运算解决偏导数问题

微分大法好

$df(x) = f’(x)dx$

  1. $d(u \pm v) = du \pm dv$
  2. $d(uv) = udv + vdu$

所以多元函数的微分可以用微分运算法则.

eg.

$f(x, y) = x + y + xy$

$df(x, y) = d(x + y + xy) = dx + dy + d(xy) = dx + dy + ydx + xdy$

$dz = \frac{\partial z}{\partial x}dx + \frac{\partial z}{\partial y}dy$

二元以上函数也同理

其中, $dx$的系数就是$z$对$x$的偏导. 其他自变量同理.

$z = f(\frac{x}{y}, \frac{y}{x})$, 求$\frac{\partial z}{\partial x}$和$\frac{\partial z}{\partial y}$

$$\begin{aligned} dz &= df(\frac{x}{y}, \frac{y}{x}) \\ &= f_1’d(\frac{x}{y}) + f_2’d(\frac{y}{x}) \\ &= f_1’(\frac{1}{y}dx - \frac{x}{y^2}dy) + f_2’(\frac{1}{x}dy - \frac{y}{x^2}dx) \\ &= (\frac{1}{y}f_1’ - \frac{y}{x^2}f_2’)dx + (\frac{1}{x}f_2’ - \frac{x}{y^2}f_1’)dy \end{aligned}$$

所以$\frac{\partial z}{\partial x} = \frac{1}{y}f_1’ - \frac{y}{x^2}f_2’, \frac{\partial z}{\partial y} = \frac{1}{x}f_2’ - \frac{x}{y^2}f_1'$

$x^2 + y^2 + z^2 = 4z$, 求$\frac{\partial ^2 z}{\partial x^2}$

两边微分, 得:

$$\begin{aligned} 2xdx + 2ydy + 2zdz = 4dz \\ \therefore dz = \frac{x}{2-z}dx + \frac{y}{2-z}dy \end{aligned}$$

$$\therefore \frac{\partial z}{\partial x} = \frac{x}{2-z}$$

再两边微分, 得:

$$\begin{aligned} d(\frac{\partial z}{\partial x}) &= \frac{1}{2-z}dx + \frac{x}{(2-z)^2}dz \\ &= \frac{1}{2-z}dx + \frac{x}{(2-z)^2}(\frac{x}{2-z}dx + \frac{y}{2-z}dy) \\ &= \frac{(2-z)^2 + x^2}{(2-z)^3}dx + \frac{(2-z)^2 + xy}{(2-z)^3}dy \end{aligned}$$

所以$\frac{\partial ^2 z}{\partial x^2} = \frac{(2-z)^2 + x^2}{(2-z)^3}$

同时还能得到$\frac{\partial ^2 z}{\partial x \partial y} = \frac{(2-z)^2 + xy}{(2-z)^3}$

$\begin{cases} x = u + v \\ y = u^2 + v^2 \\ z = u^3 + v^3 \end{cases}$ 求 $\frac{\partial u}{\partial x}$ 和 $\frac{\partial u}{\partial y}$

五个变量三个方程, 所以有两个自变量, 根据题意是$x, y$, 把$u, v, z$看成关于$x, y$的函数$u(x, y), v(x, y), z(x, y)$. 于是只要求出$du = \alpha dx + \beta dy$即可

两边微分:

$$\begin{cases} dx = du + dv \\ dy = 2udu + 2vdv \\ dz = 3u^2du + 3v^2dv \end{cases}$$

由于$dv, dz$我们不要, 所以解个方程组把他们干掉就行了.

随便来咯, 可以用初中的方法解方程, 得到$du = \frac{v}{v-u}dx + \frac{1}{2(u-v)}dy$.

也可以用线代的方法:

$$\begin{cases} du + dv + 0 = dx \\ 2udu + 2vdv + 0 = dy \\ 3u^2du + 3v^2dv - dz = 0 \end{cases}$$

$$du = \frac {\left| \begin{array}{ccc} dx & 1 & 0 \\ dy & 2v & 0 \\ 0 & 3v^2 & -1 \end{array} \right|} {\left| \begin{array}{ccc} 1 & 1 & 0 \\ 2u & 2v & 0 \\ 3u^2 & 3v^2 & -1 \end{array} \right|} = \frac{v}{v-u}dx + \frac{1}{2(u-v)}dy$$

同时也可以很方便地求

$$dv = \frac {\left| \begin{array}{ccc} 1 & dx & 0 \\ 2u & dy & 0 \\ 3u^2 & 0 & -1 \end{array} \right|} {\left| \begin{array}{ccc} 1 & 1 & 0 \\ 2u & 2v & 0 \\ 3u^2 & 3v^2 & -1 \end{array} \right|}$$

或者

$$dz = \frac {\left| \begin{array}{ccc} 1 & 1 & dx \\ 2u & 2v & dy \\ 3u^2 & 3v^2 & 0 \end{array} \right|} {\left| \begin{array}{ccc} 1 & 1 & 0 \\ 2u & 2v & 0 \\ 3u^2 & 3v^2 & -1 \end{array} \right|}$$

从而得到 $\frac{\partial v}{\partial x}, \frac{\partial v}{\partial y}, \frac{\partial z}{\partial x}, \frac{\partial z}{\partial y}$

求曲面 $xyz + \sqrt{x^2 + y^2 + z^2} - \sqrt 2 = 0$ 在点(1, 0, -1)处的切平面方程

事实上, 曲面方程可以看成函数$F(x, y, z) = 0$

两边微分, 得

$$dF = xydz + yzdx + zxdy + \frac{xdx + ydy + zdz}{\sqrt{x^2 + y^2 + z^2}}$$

又知道曲面在某点的法向量是$(F_x, F_y, F_z)$, 第一个分量$F_x$就是上述方程中$dx$的系数, $F_y, F_z$同理.

带入点$(1, 0, -1)$, 得

$$dF = \frac{1}{\sqrt 2} (dx - dz) - dy$$

所以法向量$\vec n = (\frac{1}{\sqrt 2}, -1, -\frac{1}{\sqrt 2})$, 而平面方程只需要将$dx$换成$(x - x_0)$, $dy, dz$同理, 然后令$dF = 0$, 即:

$$\frac{1}{\sqrt 2} ((x - 1) - (z + 1)) - y = 0$$

为什么呢? 如果不看成$F(x, y, z) = 0$而是直接对原来的方程两边微分, 就可得到

$$\frac{1}{\sqrt 2} (dx - dz) - dy = 0$$

考虑$dx$的意义, $x = x_0 + dx$, 不就是$dx = x - x_0$吗, 只不过这个$x$距离$x_0$很近, 所以近似成"直线". 而我们进行替换, 则就是"切线"了. 对$dy, dz$也同样处理, 不就是得到"切面"了吗?

或者还可以这样理解, 上述方程可以看成 $(\frac{1}{\sqrt 2}, -1, -\frac{1}{\sqrt 2}) \cdot (dx, dy, dz) = 0$, 而平面就是$(\frac{1}{\sqrt 2}, -1, -\frac{1}{\sqrt 2}) \cdot (x - x_0, y - y_0, z - z_0) = 0$